lustre/lfsck/lfsck_layout.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) 2014, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_layout.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 #ifndef EXPORT_SYMTAB
  32 # define EXPORT_SYMTAB
  33 #endif
  34 #define DEBUG_SUBSYSTEM S_LFSCK
  35
  36 #include <linux/bitops.h>
  37 #include <linux/rbtree.h>
  38
  39 #include <lustre/lustre_idl.h>
  40 #include <lu_object.h>
  41 #include <dt_object.h>
  42 #include <lustre_fid.h>
  43 #include <lustre_lib.h>
  44 #include <lustre_net.h>
  45 #include <lustre/lustre_user.h>
  46 #include <md_object.h>
  47 #include <obd_class.h>
  48
  49 #include "lfsck_internal.h"
  50
  51 #define LFSCK_LAYOUT_MAGIC_V1           0xB173AE14
  52 #define LFSCK_LAYOUT_MAGIC_V2           0xB1734D76
  53
  54 #define LFSCK_LAYOUT_MAGIC              LFSCK_LAYOUT_MAGIC_V2
  55
  56 struct lfsck_layout_seq {
  57         struct list_head         lls_list;
  58         __u64                    lls_seq;
  59         __u64                    lls_lastid;
  60         __u64                    lls_lastid_known;
  61         struct dt_object        *lls_lastid_obj;
  62         unsigned int             lls_dirty:1;
  63 };
  64
  65 struct lfsck_layout_slave_target {
  66         /* link into lfsck_layout_slave_data::llsd_master_list. */
  67         struct list_head        llst_list;
  68         /* The position for next record in the rbtree for iteration. */
  69         struct lu_fid           llst_fid;
  70         /* Dummy hash for iteration against the rbtree. */
  71         __u64                   llst_hash;
  72         __u64                   llst_gen;
  73         atomic_t                llst_ref;
  74         __u32                   llst_index;
  75 };
  76
  77 struct lfsck_layout_slave_data {
  78         /* list for lfsck_layout_seq */
  79         struct list_head         llsd_seq_list;
  80
  81         /* list for the masters involve layout verification. */
  82         struct list_head         llsd_master_list;
  83         spinlock_t               llsd_lock;
  84         __u64                    llsd_touch_gen;
  85         struct dt_object        *llsd_rb_obj;
  86         struct rb_root           llsd_rb_root;
  87         rwlock_t                 llsd_rb_lock;
  88         unsigned int             llsd_rbtree_valid:1;
  89 };
  90
  91 struct lfsck_layout_object {
  92         struct lu_attr           llo_attr;
  93         atomic_t                 llo_ref;
  94         __u64                    llo_cookie;
  95 };
  96
  97 struct lfsck_layout_req {
  98         struct lfsck_assistant_req       llr_lar;
  99         struct lfsck_layout_object      *llr_parent;
 100         struct dt_object                *llr_child;
 101         __u32                            llr_ost_idx;
 102         __u32                            llr_lov_idx; /* offset in LOV EA */
 103 };
 104
 105 struct lfsck_layout_slave_async_args {
 106         struct obd_export                *llsaa_exp;
 107         struct lfsck_component           *llsaa_com;
 108         struct lfsck_layout_slave_target *llsaa_llst;
 109 };
 110
 111 static struct lfsck_layout_object *
 112 lfsck_layout_object_init(const struct lu_env *env, struct dt_object *obj,
 113                          __u64 cookie)
 114 {
 115         struct lfsck_layout_object *llo;
 116         int                         rc;
 117
 118         OBD_ALLOC_PTR(llo);
 119         if (llo == NULL)
 120                 return ERR_PTR(-ENOMEM);
 121
 122         rc = dt_attr_get(env, obj, &llo->llo_attr, BYPASS_CAPA);
 123         if (rc != 0) {
 124                 OBD_FREE_PTR(llo);
 125
 126                 return ERR_PTR(rc);
 127         }
 128
 129         llo->llo_cookie = cookie;
 130         atomic_set(&llo->llo_ref, 1);
 131
 132         return llo;
 133 }
 134
 135 static inline void
 136 lfsck_layout_llst_put(struct lfsck_layout_slave_target *llst)
 137 {
 138         if (atomic_dec_and_test(&llst->llst_ref)) {
 139                 LASSERT(list_empty(&llst->llst_list));
 140
 141                 OBD_FREE_PTR(llst);
 142         }
 143 }
 144
 145 static inline int
 146 lfsck_layout_llst_add(struct lfsck_layout_slave_data *llsd, __u32 index)
 147 {
 148         struct lfsck_layout_slave_target *llst;
 149         struct lfsck_layout_slave_target *tmp;
 150         int                               rc   = 0;
 151
 152         OBD_ALLOC_PTR(llst);
 153         if (llst == NULL)
 154                 return -ENOMEM;
 155
 156         INIT_LIST_HEAD(&llst->llst_list);
 157         llst->llst_gen = 0;
 158         llst->llst_index = index;
 159         atomic_set(&llst->llst_ref, 1);
 160
 161         spin_lock(&llsd->llsd_lock);
 162         list_for_each_entry(tmp, &llsd->llsd_master_list, llst_list) {
 163                 if (tmp->llst_index == index) {
 164                         rc = -EALREADY;
 165                         break;
 166                 }
 167         }
 168         if (rc == 0)
 169                 list_add_tail(&llst->llst_list, &llsd->llsd_master_list);
 170         spin_unlock(&llsd->llsd_lock);
 171
 172         if (rc != 0)
 173                 OBD_FREE_PTR(llst);
 174
 175         return rc;
 176 }
 177
 178 static inline void
 179 lfsck_layout_llst_del(struct lfsck_layout_slave_data *llsd,
 180                       struct lfsck_layout_slave_target *llst)
 181 {
 182         bool del = false;
 183
 184         spin_lock(&llsd->llsd_lock);
 185         if (!list_empty(&llst->llst_list)) {
 186                 list_del_init(&llst->llst_list);
 187                 del = true;
 188         }
 189         spin_unlock(&llsd->llsd_lock);
 190
 191         if (del)
 192                 lfsck_layout_llst_put(llst);
 193 }
 194
 195 static inline struct lfsck_layout_slave_target *
 196 lfsck_layout_llst_find_and_del(struct lfsck_layout_slave_data *llsd,
 197                                __u32 index, bool unlink)
 198 {
 199         struct lfsck_layout_slave_target *llst;
 200
 201         spin_lock(&llsd->llsd_lock);
 202         list_for_each_entry(llst, &llsd->llsd_master_list, llst_list) {
 203                 if (llst->llst_index == index) {
 204                         if (unlink)
 205                                 list_del_init(&llst->llst_list);
 206                         else
 207                                 atomic_inc(&llst->llst_ref);
 208                         spin_unlock(&llsd->llsd_lock);
 209
 210                         return llst;
 211                 }
 212         }
 213         spin_unlock(&llsd->llsd_lock);
 214
 215         return NULL;
 216 }
 217
 218 static inline void lfsck_layout_object_put(const struct lu_env *env,
 219                                            struct lfsck_layout_object *llo)
 220 {
 221         if (atomic_dec_and_test(&llo->llo_ref))
 222                 OBD_FREE_PTR(llo);
 223 }
 224
 225 static struct lfsck_layout_req *
 226 lfsck_layout_assistant_req_init(struct lfsck_layout_object *parent,
 227                                 const struct lu_fid *pfid,
 228                                 struct dt_object *child, __u32 ost_idx,
 229                                 __u32 lov_idx)
 230 {
 231         struct lfsck_layout_req *llr;
 232
 233         OBD_ALLOC_PTR(llr);
 234         if (llr == NULL)
 235                 return ERR_PTR(-ENOMEM);
 236
 237         INIT_LIST_HEAD(&llr->llr_lar.lar_list);
 238         llr->llr_lar.lar_fid = *pfid;
 239
 240         atomic_inc(&parent->llo_ref);
 241         llr->llr_parent = parent;
 242         llr->llr_child = child;
 243         llr->llr_ost_idx = ost_idx;
 244         llr->llr_lov_idx = lov_idx;
 245
 246         return llr;
 247 }
 248
 249 static void lfsck_layout_assistant_req_fini(const struct lu_env *env,
 250                                             struct lfsck_assistant_req *lar)
 251 {
 252         struct lfsck_layout_req *llr =
 253                         container_of0(lar, struct lfsck_layout_req, llr_lar);
 254
 255         lfsck_object_put(env, llr->llr_child);
 256         lfsck_layout_object_put(env, llr->llr_parent);
 257         OBD_FREE_PTR(llr);
 258 }
 259
 260 static int
 261 lfsck_layout_assistant_sync_failures_interpret(const struct lu_env *env,
 262                                                struct ptlrpc_request *req,
 263                                                void *args, int rc)
 264 {
 265         if (rc == 0) {
 266                 struct lfsck_async_interpret_args *laia = args;
 267                 struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
 268
 269                 ltd->ltd_synced_failures = 1;
 270                 atomic_dec(laia->laia_count);
 271         }
 272
 273         return 0;
 274 }
 275
 276 /**
 277  * Notify remote LFSCK instances about former failures.
 278  *
 279  * The local LFSCK instance has recorded which OSTs have ever failed to respond
 280  * some LFSCK verification requests (maybe because of network issues or the OST
 281  * itself trouble). During the respond gap, the OST may missed some OST-objects
 282  * verification, then the OST cannot know whether related OST-objects have been
 283  * referenced by related MDT-objects or not, then in the second-stage scanning,
 284  * these OST-objects will be regarded as orphan, if the OST-object contains bad
 285  * parent FID for back reference, then it will misguide the LFSCK to make wrong
 286  * fixing for the fake orphan.
 287  *
 288  * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
 289  * it will scan the bitmap for the ever failed OSTs, and notify them that they
 290  * have ever missed some OST-object verification and should skip the handling
 291  * for orphan OST-objects on all MDTs that are in the layout LFSCK.
 292  *
 293  * \param[in] env       pointer to the thread context
 294  * \param[in] com       pointer to the lfsck component
 295  * \param[in] lr        pointer to the lfsck request
 296  */
 297 static void lfsck_layout_assistant_sync_failures(const struct lu_env *env,
 298                                                  struct lfsck_component *com,
 299                                                  struct lfsck_request *lr)
 300 {
 301         struct lfsck_async_interpret_args *laia  =
 302                                 &lfsck_env_info(env)->lti_laia2;
 303         struct lfsck_assistant_data       *lad   = com->lc_data;
 304         struct lfsck_layout               *lo    = com->lc_file_ram;
 305         struct lfsck_instance             *lfsck = com->lc_lfsck;
 306         struct lfsck_tgt_descs            *ltds  = &lfsck->li_ost_descs;
 307         struct lfsck_tgt_desc             *ltd;
 308         struct ptlrpc_request_set         *set;
 309         atomic_t                           count;
 310         __u32                              idx;
 311         int                                rc    = 0;
 312         ENTRY;
 313
 314         if (!lad->lad_incomplete || lo->ll_flags & LF_INCOMPLETE)
 315                 RETURN_EXIT;
 316
 317         /* If the MDT has ever failed to verfiy some OST-objects,
 318          * then sync failures with them firstly. */
 319         lr->lr_flags2 = lo->ll_flags | LF_INCOMPLETE;
 320
 321         atomic_set(&count, 0);
 322         memset(laia, 0, sizeof(*laia));
 323         laia->laia_count = &count;
 324         set = ptlrpc_prep_set();
 325         if (set == NULL)
 326                 GOTO(out, rc = -ENOMEM);
 327
 328         down_read(&ltds->ltd_rw_sem);
 329         cfs_foreach_bit(lad->lad_bitmap, idx) {
 330                 ltd = LTD_TGT(ltds, idx);
 331                 LASSERT(ltd != NULL);
 332
 333                 laia->laia_ltd = ltd;
 334                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
 335                                 lfsck_layout_assistant_sync_failures_interpret,
 336                                 laia, LFSCK_NOTIFY);
 337                 if (rc != 0) {
 338                         CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
 339                                "notify target %x for %s phase1 done: "
 340                                "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
 341                                ltd->ltd_index, lad->lad_name, rc);
 342
 343                         break;
 344                 }
 345
 346                 atomic_inc(&count);
 347         }
 348         up_read(&ltds->ltd_rw_sem);
 349
 350         if (rc == 0 && atomic_read(&count) > 0)
 351                 rc = ptlrpc_set_wait(set);
 352
 353         ptlrpc_set_destroy(set);
 354
 355         if (rc == 0 && atomic_read(&count) > 0)
 356                 rc = -EINVAL;
 357
 358         GOTO(out, rc);
 359
 360 out:
 361         if (rc != 0)
 362                 /* If failed to sync failures with the OSTs, then have to
 363                  * mark the whole LFSCK as LF_INCOMPLETE to skip the whole
 364                  * subsequent orphan OST-object handling. */
 365                 lo->ll_flags |= LF_INCOMPLETE;
 366
 367         lr->lr_flags2 = lo->ll_flags;
 368 }
 369
 370 static int lfsck_layout_get_lovea(const struct lu_env *env,
 371                                   struct dt_object *obj, struct lu_buf *buf)
 372 {
 373         int rc;
 374
 375 again:
 376         rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LOV, BYPASS_CAPA);
 377         if (rc == -ERANGE) {
 378                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV,
 379                                   BYPASS_CAPA);
 380                 if (rc <= 0)
 381                         return rc;
 382
 383                 lu_buf_realloc(buf, rc);
 384                 if (buf->lb_buf == NULL)
 385                         return -ENOMEM;
 386
 387                 goto again;
 388         }
 389
 390         if (rc == -ENODATA)
 391                 rc = 0;
 392
 393         if (rc <= 0)
 394                 return rc;
 395
 396         if (unlikely(buf->lb_buf == NULL)) {
 397                 lu_buf_alloc(buf, rc);
 398                 if (buf->lb_buf == NULL)
 399                         return -ENOMEM;
 400
 401                 goto again;
 402         }
 403
 404         return rc;
 405 }
 406
 407 static int lfsck_layout_verify_header(struct lov_mds_md_v1 *lmm)
 408 {
 409         __u32 magic;
 410         __u32 pattern;
 411
 412         magic = le32_to_cpu(lmm->lmm_magic);
 413         /* If magic crashed, keep it there. Sometime later, during OST-object
 414          * orphan handling, if some OST-object(s) back-point to it, it can be
 415          * verified and repaired. */
 416         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) {
 417                 struct ost_id   oi;
 418                 int             rc;
 419
 420                 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
 421                 if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
 422                         rc = -EOPNOTSUPP;
 423                 else
 424                         rc = -EINVAL;
 425
 426                 CDEBUG(D_LFSCK, "%s LOV EA magic %u on "DOSTID"\n",
 427                        rc == -EINVAL ? "Unknown" : "Unsupported",
 428                        magic, POSTID(&oi));
 429
 430                 return rc;
 431         }
 432
 433         pattern = le32_to_cpu(lmm->lmm_pattern);
 434         /* XXX: currently, we only support LOV_PATTERN_RAID0. */
 435         if (lov_pattern(pattern) != LOV_PATTERN_RAID0) {
 436                 struct ost_id oi;
 437
 438                 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
 439                 CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u on "DOSTID"\n",
 440                        pattern, POSTID(&oi));
 441
 442                 return -EOPNOTSUPP;
 443         }
 444
 445         return 0;
 446 }
 447
 448 #define LFSCK_RBTREE_BITMAP_SIZE        PAGE_CACHE_SIZE
 449 #define LFSCK_RBTREE_BITMAP_WIDTH       (LFSCK_RBTREE_BITMAP_SIZE << 3)
 450 #define LFSCK_RBTREE_BITMAP_MASK        (LFSCK_RBTREE_BITMAP_WIDTH - 1)
 451
 452 struct lfsck_rbtree_node {
 453         struct rb_node   lrn_node;
 454         __u64            lrn_seq;
 455         __u32            lrn_first_oid;
 456         atomic_t         lrn_known_count;
 457         atomic_t         lrn_accessed_count;
 458         void            *lrn_known_bitmap;
 459         void            *lrn_accessed_bitmap;
 460 };
 461
 462 static inline int lfsck_rbtree_cmp(struct lfsck_rbtree_node *lrn,
 463                                    __u64 seq, __u32 oid)
 464 {
 465         if (seq < lrn->lrn_seq)
 466                 return -1;
 467
 468         if (seq > lrn->lrn_seq)
 469                 return 1;
 470
 471         if (oid < lrn->lrn_first_oid)
 472                 return -1;
 473
 474         if (oid - lrn->lrn_first_oid >= LFSCK_RBTREE_BITMAP_WIDTH)
 475                 return 1;
 476
 477         return 0;
 478 }
 479
 480 /* The caller should hold llsd->llsd_rb_lock. */
 481 static struct lfsck_rbtree_node *
 482 lfsck_rbtree_search(struct lfsck_layout_slave_data *llsd,
 483                     const struct lu_fid *fid, bool *exact)
 484 {
 485         struct rb_node           *node  = llsd->llsd_rb_root.rb_node;
 486         struct rb_node           *prev  = NULL;
 487         struct lfsck_rbtree_node *lrn   = NULL;
 488         int                       rc    = 0;
 489
 490         if (exact != NULL)
 491                 *exact = true;
 492
 493         while (node != NULL) {
 494                 prev = node;
 495                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 496                 rc = lfsck_rbtree_cmp(lrn, fid_seq(fid), fid_oid(fid));
 497                 if (rc < 0)
 498                         node = node->rb_left;
 499                 else if (rc > 0)
 500                         node = node->rb_right;
 501                 else
 502                         return lrn;
 503         }
 504
 505         if (exact == NULL)
 506                 return NULL;
 507
 508         /* If there is no exactly matched one, then to the next valid one. */
 509         *exact = false;
 510
 511         /* The rbtree is empty. */
 512         if (rc == 0)
 513                 return NULL;
 514
 515         if (rc < 0)
 516                 return lrn;
 517
 518         node = rb_next(prev);
 519
 520         /* The end of the rbtree. */
 521         if (node == NULL)
 522                 return NULL;
 523
 524         lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 525
 526         return lrn;
 527 }
 528
 529 static struct lfsck_rbtree_node *lfsck_rbtree_new(const struct lu_env *env,
 530                                                   const struct lu_fid *fid)
 531 {
 532         struct lfsck_rbtree_node *lrn;
 533
 534         OBD_ALLOC_PTR(lrn);
 535         if (lrn == NULL)
 536                 return ERR_PTR(-ENOMEM);
 537
 538         OBD_ALLOC(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 539         if (lrn->lrn_known_bitmap == NULL) {
 540                 OBD_FREE_PTR(lrn);
 541
 542                 return ERR_PTR(-ENOMEM);
 543         }
 544
 545         OBD_ALLOC(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 546         if (lrn->lrn_accessed_bitmap == NULL) {
 547                 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 548                 OBD_FREE_PTR(lrn);
 549
 550                 return ERR_PTR(-ENOMEM);
 551         }
 552
 553         RB_CLEAR_NODE(&lrn->lrn_node);
 554         lrn->lrn_seq = fid_seq(fid);
 555         lrn->lrn_first_oid = fid_oid(fid) & ~LFSCK_RBTREE_BITMAP_MASK;
 556         atomic_set(&lrn->lrn_known_count, 0);
 557         atomic_set(&lrn->lrn_accessed_count, 0);
 558
 559         return lrn;
 560 }
 561
 562 static void lfsck_rbtree_free(struct lfsck_rbtree_node *lrn)
 563 {
 564         OBD_FREE(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 565         OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 566         OBD_FREE_PTR(lrn);
 567 }
 568
 569 /* The caller should hold lock. */
 570 static struct lfsck_rbtree_node *
 571 lfsck_rbtree_insert(struct lfsck_layout_slave_data *llsd,
 572                     struct lfsck_rbtree_node *lrn)
 573 {
 574         struct rb_node           **pos    = &llsd->llsd_rb_root.rb_node;
 575         struct rb_node            *parent = NULL;
 576         struct lfsck_rbtree_node  *tmp;
 577         int                        rc;
 578
 579         while (*pos != NULL) {
 580                 parent = *pos;
 581                 tmp = rb_entry(parent, struct lfsck_rbtree_node, lrn_node);
 582                 rc = lfsck_rbtree_cmp(tmp, lrn->lrn_seq, lrn->lrn_first_oid);
 583                 if (rc < 0)
 584                         pos = &(*pos)->rb_left;
 585                 else if (rc > 0)
 586                         pos = &(*pos)->rb_right;
 587                 else
 588                         return tmp;
 589         }
 590
 591         rb_link_node(&lrn->lrn_node, parent, pos);
 592         rb_insert_color(&lrn->lrn_node, &llsd->llsd_rb_root);
 593
 594         return lrn;
 595 }
 596
 597 extern const struct dt_index_operations lfsck_orphan_index_ops;
 598
 599 static int lfsck_rbtree_setup(const struct lu_env *env,
 600                               struct lfsck_component *com)
 601 {
 602         struct lu_fid                   *fid    = &lfsck_env_info(env)->lti_fid;
 603         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 604         struct dt_device                *dev    = lfsck->li_bottom;
 605         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
 606         struct dt_object                *obj;
 607
 608         fid->f_seq = FID_SEQ_LAYOUT_RBTREE;
 609         fid->f_oid = lfsck_dev_idx(lfsck);
 610         fid->f_ver = 0;
 611         obj = dt_locate(env, dev, fid);
 612         if (IS_ERR(obj))
 613                 RETURN(PTR_ERR(obj));
 614
 615         /* Generate an in-RAM object to stand for the layout rbtree.
 616          * Scanning the layout rbtree will be via the iteration over
 617          * the object. In the future, the rbtree may be written onto
 618          * disk with the object.
 619          *
 620          * Mark the object to be as exist. */
 621         obj->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 622         obj->do_index_ops = &lfsck_orphan_index_ops;
 623         llsd->llsd_rb_obj = obj;
 624         llsd->llsd_rbtree_valid = 1;
 625         dev->dd_record_fid_accessed = 1;
 626
 627         CDEBUG(D_LFSCK, "%s: layout LFSCK init OST-objects accessing bitmap\n",
 628                lfsck_lfsck2name(lfsck));
 629
 630         return 0;
 631 }
 632
 633 static void lfsck_rbtree_cleanup(const struct lu_env *env,
 634                                  struct lfsck_component *com)
 635 {
 636         struct lfsck_instance           *lfsck = com->lc_lfsck;
 637         struct lfsck_layout_slave_data  *llsd  = com->lc_data;
 638         struct rb_node                  *node  = rb_first(&llsd->llsd_rb_root);
 639         struct rb_node                  *next;
 640         struct lfsck_rbtree_node        *lrn;
 641
 642         lfsck->li_bottom->dd_record_fid_accessed = 0;
 643         /* Invalid the rbtree, then no others will use it. */
 644         write_lock(&llsd->llsd_rb_lock);
 645         llsd->llsd_rbtree_valid = 0;
 646         write_unlock(&llsd->llsd_rb_lock);
 647
 648         while (node != NULL) {
 649                 next = rb_next(node);
 650                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 651                 rb_erase(node, &llsd->llsd_rb_root);
 652                 lfsck_rbtree_free(lrn);
 653                 node = next;
 654         }
 655
 656         if (llsd->llsd_rb_obj != NULL) {
 657                 lfsck_object_put(env, llsd->llsd_rb_obj);
 658                 llsd->llsd_rb_obj = NULL;
 659         }
 660
 661         CDEBUG(D_LFSCK, "%s: layout LFSCK fini OST-objects accessing bitmap\n",
 662                lfsck_lfsck2name(lfsck));
 663 }
 664
 665 static void lfsck_rbtree_update_bitmap(const struct lu_env *env,
 666                                        struct lfsck_component *com,
 667                                        const struct lu_fid *fid,
 668                                        bool accessed)
 669 {
 670         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
 671         struct lfsck_rbtree_node        *lrn;
 672         bool                             insert = false;
 673         int                              idx;
 674         int                              rc     = 0;
 675         ENTRY;
 676
 677         if (unlikely(!fid_is_sane(fid) || fid_is_last_id(fid)))
 678                 RETURN_EXIT;
 679
 680         if (!fid_is_idif(fid) && !fid_is_norm(fid))
 681                 RETURN_EXIT;
 682
 683         read_lock(&llsd->llsd_rb_lock);
 684         if (!llsd->llsd_rbtree_valid)
 685                 GOTO(unlock, rc = 0);
 686
 687         lrn = lfsck_rbtree_search(llsd, fid, NULL);
 688         if (lrn == NULL) {
 689                 struct lfsck_rbtree_node *tmp;
 690
 691                 LASSERT(!insert);
 692
 693                 read_unlock(&llsd->llsd_rb_lock);
 694                 tmp = lfsck_rbtree_new(env, fid);
 695                 if (IS_ERR(tmp))
 696                         GOTO(out, rc = PTR_ERR(tmp));
 697
 698                 insert = true;
 699                 write_lock(&llsd->llsd_rb_lock);
 700                 if (!llsd->llsd_rbtree_valid) {
 701                         lfsck_rbtree_free(tmp);
 702                         GOTO(unlock, rc = 0);
 703                 }
 704
 705                 lrn = lfsck_rbtree_insert(llsd, tmp);
 706                 if (lrn != tmp)
 707                         lfsck_rbtree_free(tmp);
 708         }
 709
 710         idx = fid_oid(fid) & LFSCK_RBTREE_BITMAP_MASK;
 711         /* Any accessed object must be a known object. */
 712         if (!test_and_set_bit(idx, lrn->lrn_known_bitmap))
 713                 atomic_inc(&lrn->lrn_known_count);
 714         if (accessed && !test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
 715                 atomic_inc(&lrn->lrn_accessed_count);
 716
 717         GOTO(unlock, rc = 0);
 718
 719 unlock:
 720         if (insert)
 721                 write_unlock(&llsd->llsd_rb_lock);
 722         else
 723                 read_unlock(&llsd->llsd_rb_lock);
 724 out:
 725         if (rc != 0 && accessed) {
 726                 struct lfsck_layout *lo = com->lc_file_ram;
 727
 728                 CDEBUG(D_LFSCK, "%s: fail to update OST-objects accessing "
 729                        "bitmap, and will cause incorrect LFSCK OST-object "
 730                        "handling, so disable it to cancel orphan handling "
 731                        "for related device. rc = %d\n",
 732                        lfsck_lfsck2name(com->lc_lfsck), rc);
 733
 734                 lo->ll_flags |= LF_INCOMPLETE;
 735                 lfsck_rbtree_cleanup(env, com);
 736         }
 737 }
 738
 739 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
 740                                    const struct lfsck_layout *src)
 741 {
 742         int i;
 743
 744         des->ll_magic = le32_to_cpu(src->ll_magic);
 745         des->ll_status = le32_to_cpu(src->ll_status);
 746         des->ll_flags = le32_to_cpu(src->ll_flags);
 747         des->ll_success_count = le32_to_cpu(src->ll_success_count);
 748         des->ll_run_time_phase1 = le32_to_cpu(src->ll_run_time_phase1);
 749         des->ll_run_time_phase2 = le32_to_cpu(src->ll_run_time_phase2);
 750         des->ll_time_last_complete = le64_to_cpu(src->ll_time_last_complete);
 751         des->ll_time_latest_start = le64_to_cpu(src->ll_time_latest_start);
 752         des->ll_time_last_checkpoint =
 753                                 le64_to_cpu(src->ll_time_last_checkpoint);
 754         des->ll_pos_latest_start = le64_to_cpu(src->ll_pos_latest_start);
 755         des->ll_pos_last_checkpoint = le64_to_cpu(src->ll_pos_last_checkpoint);
 756         des->ll_pos_first_inconsistent =
 757                         le64_to_cpu(src->ll_pos_first_inconsistent);
 758         des->ll_objs_checked_phase1 = le64_to_cpu(src->ll_objs_checked_phase1);
 759         des->ll_objs_failed_phase1 = le64_to_cpu(src->ll_objs_failed_phase1);
 760         des->ll_objs_checked_phase2 = le64_to_cpu(src->ll_objs_checked_phase2);
 761         des->ll_objs_failed_phase2 = le64_to_cpu(src->ll_objs_failed_phase2);
 762         for (i = 0; i < LLIT_MAX; i++)
 763                 des->ll_objs_repaired[i] =
 764                                 le64_to_cpu(src->ll_objs_repaired[i]);
 765         des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
 766         des->ll_bitmap_size = le32_to_cpu(src->ll_bitmap_size);
 767 }
 768
 769 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
 770                                    const struct lfsck_layout *src)
 771 {
 772         int i;
 773
 774         des->ll_magic = cpu_to_le32(src->ll_magic);
 775         des->ll_status = cpu_to_le32(src->ll_status);
 776         des->ll_flags = cpu_to_le32(src->ll_flags);
 777         des->ll_success_count = cpu_to_le32(src->ll_success_count);
 778         des->ll_run_time_phase1 = cpu_to_le32(src->ll_run_time_phase1);
 779         des->ll_run_time_phase2 = cpu_to_le32(src->ll_run_time_phase2);
 780         des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
 781         des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
 782         des->ll_time_last_checkpoint =
 783                                 cpu_to_le64(src->ll_time_last_checkpoint);
 784         des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
 785         des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
 786         des->ll_pos_first_inconsistent =
 787                         cpu_to_le64(src->ll_pos_first_inconsistent);
 788         des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
 789         des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
 790         des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
 791         des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
 792         for (i = 0; i < LLIT_MAX; i++)
 793                 des->ll_objs_repaired[i] =
 794                                 cpu_to_le64(src->ll_objs_repaired[i]);
 795         des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
 796         des->ll_bitmap_size = cpu_to_le32(src->ll_bitmap_size);
 797 }
 798
 799 /**
 800  * Load the OST bitmap from the lfsck_layout trace file.
 801  *
 802  * \param[in] env       pointer to the thread context
 803  * \param[in] com       pointer to the lfsck component
 804  *
 805  * \retval              0 for success
 806  * \retval              negative error number on failure or data corruption
 807  */
 808 static int lfsck_layout_load_bitmap(const struct lu_env *env,
 809                                     struct lfsck_component *com)
 810 {
 811         struct dt_object                *obj    = com->lc_obj;
 812         struct lfsck_assistant_data     *lad    = com->lc_data;
 813         struct lfsck_layout             *lo     = com->lc_file_ram;
 814         cfs_bitmap_t                    *bitmap = lad->lad_bitmap;
 815         loff_t                           pos    = com->lc_file_size;
 816         ssize_t                          size;
 817         __u32                            nbits;
 818         int                              rc;
 819         ENTRY;
 820
 821         if (com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size >
 822             lo->ll_bitmap_size)
 823                 nbits = com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size;
 824         else
 825                 nbits = lo->ll_bitmap_size;
 826
 827         if (unlikely(nbits < BITS_PER_LONG))
 828                 nbits = BITS_PER_LONG;
 829
 830         if (nbits > bitmap->size) {
 831                 __u32 new_bits = bitmap->size;
 832                 cfs_bitmap_t *new_bitmap;
 833
 834                 while (new_bits < nbits)
 835                         new_bits <<= 1;
 836
 837                 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
 838                 if (new_bitmap == NULL)
 839                         RETURN(-ENOMEM);
 840
 841                 lad->lad_bitmap = new_bitmap;
 842                 CFS_FREE_BITMAP(bitmap);
 843                 bitmap = new_bitmap;
 844         }
 845
 846         if (lo->ll_bitmap_size == 0) {
 847                 lad->lad_incomplete = 0;
 848                 CFS_RESET_BITMAP(bitmap);
 849
 850                 RETURN(0);
 851         }
 852
 853         size = (lo->ll_bitmap_size + 7) >> 3;
 854         rc = dt_read(env, obj, lfsck_buf_get(env, bitmap->data, size), &pos);
 855         if (rc != size)
 856                 RETURN(rc >= 0 ? -EINVAL : rc);
 857
 858         if (cfs_bitmap_check_empty(bitmap))
 859                 lad->lad_incomplete = 0;
 860         else
 861                 lad->lad_incomplete = 1;
 862
 863         RETURN(0);
 864 }
 865
 866 /**
 867  * Load the layout LFSCK trace file from disk.
 868  *
 869  * The layout LFSCK trace file records the layout LFSCK status information
 870  * and other statistics, such as how many objects have been scanned, and how
 871  * many objects have been repaired, and etc. It also contains the bitmap for
 872  * failed OSTs during the layout LFSCK. All these information will be loaded
 873  * from disk to RAM when the layout LFSCK component setup.
 874  *
 875  * \param[in] env       pointer to the thread context
 876  * \param[in] com       pointer to the lfsck component
 877  *
 878  * \retval              positive number for file data corruption, the caller
 879  *                      should reset the layout LFSCK trace file
 880  * \retval              0 for success
 881  * \retval              negative error number on failure
 882  */
 883 static int lfsck_layout_load(const struct lu_env *env,
 884                              struct lfsck_component *com)
 885 {
 886         struct lfsck_layout             *lo     = com->lc_file_ram;
 887         ssize_t                          size   = com->lc_file_size;
 888         loff_t                           pos    = 0;
 889         int                              rc;
 890
 891         rc = dt_read(env, com->lc_obj,
 892                      lfsck_buf_get(env, com->lc_file_disk, size), &pos);
 893         if (rc == 0) {
 894                 return -ENOENT;
 895         } else if (rc < 0) {
 896                 CDEBUG(D_LFSCK, "%s: failed to load lfsck_layout: rc = %d\n",
 897                        lfsck_lfsck2name(com->lc_lfsck), rc);
 898                 return rc;
 899         } else if (rc != size) {
 900                 CDEBUG(D_LFSCK, "%s: lfsck_layout size %u != %u; reset it\n",
 901                        lfsck_lfsck2name(com->lc_lfsck), rc, (unsigned int)size);
 902                 return 1;
 903         }
 904
 905         lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
 906         if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
 907                 CDEBUG(D_LFSCK, "%s: invalid lfsck_layout magic %#x != %#x, "
 908                        "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
 909                        lo->ll_magic, LFSCK_LAYOUT_MAGIC);
 910                 return 1;
 911         }
 912
 913         return 0;
 914 }
 915
 916 /**
 917  * Store the layout LFSCK trace file on disk.
 918  *
 919  * The layout LFSCK trace file records the layout LFSCK status information
 920  * and other statistics, such as how many objects have been scanned, and how
 921  * many objects have been repaired, and etc. It also contains the bitmap for
 922  * failed OSTs during the layout LFSCK. All these information will be synced
 923  * from RAM to disk periodically.
 924  *
 925  * \param[in] env       pointer to the thread context
 926  * \param[in] com       pointer to the lfsck component
 927  *
 928  * \retval              0 for success
 929  * \retval              negative error number on failure
 930  */
 931 static int lfsck_layout_store(const struct lu_env *env,
 932                               struct lfsck_component *com)
 933 {
 934         struct dt_object        *obj    = com->lc_obj;
 935         struct lfsck_instance   *lfsck  = com->lc_lfsck;
 936         struct lfsck_layout     *lo_ram = com->lc_file_ram;
 937         struct lfsck_layout     *lo     = com->lc_file_disk;
 938         struct thandle          *th;
 939         struct dt_device        *dev    = lfsck_obj2dev(obj);
 940         cfs_bitmap_t            *bitmap = NULL;
 941         loff_t                   pos;
 942         ssize_t                  size   = com->lc_file_size;
 943         __u32                    nbits  = 0;
 944         int                      rc;
 945         ENTRY;
 946
 947         if (lfsck->li_master) {
 948                 struct lfsck_assistant_data *lad = com->lc_data;
 949
 950                 bitmap = lad->lad_bitmap;
 951                 nbits = bitmap->size;
 952
 953                 LASSERT(nbits > 0);
 954                 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
 955         }
 956
 957         lo_ram->ll_bitmap_size = nbits;
 958         lfsck_layout_cpu_to_le(lo, lo_ram);
 959         th = dt_trans_create(env, dev);
 960         if (IS_ERR(th))
 961                 GOTO(log, rc = PTR_ERR(th));
 962
 963         rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
 964                                      (loff_t)0, th);
 965         if (rc != 0)
 966                 GOTO(out, rc);
 967
 968         if (bitmap != NULL) {
 969                 rc = dt_declare_record_write(env, obj,
 970                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
 971                                 (loff_t)size, th);
 972                 if (rc != 0)
 973                         GOTO(out, rc);
 974         }
 975
 976         rc = dt_trans_start_local(env, dev, th);
 977         if (rc != 0)
 978                 GOTO(out, rc);
 979
 980         pos = 0;
 981         rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos, th);
 982         if (rc != 0)
 983                 GOTO(out, rc);
 984
 985         if (bitmap != NULL) {
 986                 pos = size;
 987                 rc = dt_record_write(env, obj,
 988                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
 989                                 &pos, th);
 990         }
 991
 992         GOTO(out, rc);
 993
 994 out:
 995         dt_trans_stop(env, dev, th);
 996
 997 log:
 998         if (rc != 0)
 999                 CDEBUG(D_LFSCK, "%s: fail to store lfsck_layout: rc = %d\n",
1000                        lfsck_lfsck2name(lfsck), rc);
1001
1002         return rc;
1003 }
1004
1005 static int lfsck_layout_init(const struct lu_env *env,
1006                              struct lfsck_component *com)
1007 {
1008         struct lfsck_layout *lo = com->lc_file_ram;
1009         int rc;
1010
1011         memset(lo, 0, com->lc_file_size);
1012         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
1013         lo->ll_status = LS_INIT;
1014         down_write(&com->lc_sem);
1015         rc = lfsck_layout_store(env, com);
1016         up_write(&com->lc_sem);
1017
1018         return rc;
1019 }
1020
1021 static int fid_is_for_ostobj(const struct lu_env *env,
1022                              struct lfsck_instance *lfsck,
1023                              struct dt_object *obj, const struct lu_fid *fid)
1024 {
1025         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
1026         struct lu_seq_range     *range  = &lfsck_env_info(env)->lti_range;
1027         struct lustre_mdt_attrs *lma;
1028         int                      rc;
1029
1030         fld_range_set_any(range);
1031         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
1032         if (rc == 0) {
1033                 if (fld_range_is_ost(range))
1034                         return 1;
1035
1036                 return 0;
1037         }
1038
1039         lma = &lfsck_env_info(env)->lti_lma;
1040         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, lma, sizeof(*lma)),
1041                           XATTR_NAME_LMA, BYPASS_CAPA);
1042         if (rc == sizeof(*lma)) {
1043                 lustre_lma_swab(lma);
1044
1045                 return lma->lma_compat & LMAC_FID_ON_OST ? 1 : 0;
1046         }
1047
1048         rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID, BYPASS_CAPA);
1049
1050         return rc > 0;
1051 }
1052
1053 static struct lfsck_layout_seq *
1054 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
1055 {
1056         struct lfsck_layout_seq *lls;
1057
1058         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1059                 if (lls->lls_seq == seq)
1060                         return lls;
1061
1062                 if (lls->lls_seq > seq)
1063                         return NULL;
1064         }
1065
1066         return NULL;
1067 }
1068
1069 static void
1070 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
1071                         struct lfsck_layout_seq *lls)
1072 {
1073         struct lfsck_layout_seq *tmp;
1074         struct list_head        *pos = &llsd->llsd_seq_list;
1075
1076         list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
1077                 if (lls->lls_seq < tmp->lls_seq) {
1078                         pos = &tmp->lls_list;
1079                         break;
1080                 }
1081         }
1082         list_add_tail(&lls->lls_list, pos);
1083 }
1084
1085 static int
1086 lfsck_layout_lastid_create(const struct lu_env *env,
1087                            struct lfsck_instance *lfsck,
1088                            struct dt_object *obj)
1089 {
1090         struct lfsck_thread_info *info   = lfsck_env_info(env);
1091         struct lu_attr           *la     = &info->lti_la;
1092         struct dt_object_format  *dof    = &info->lti_dof;
1093         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
1094         struct dt_device         *dt     = lfsck_obj2dev(obj);
1095         struct thandle           *th;
1096         __u64                     lastid = 0;
1097         loff_t                    pos    = 0;
1098         int                       rc;
1099         ENTRY;
1100
1101         if (bk->lb_param & LPF_DRYRUN)
1102                 return 0;
1103
1104         memset(la, 0, sizeof(*la));
1105         la->la_mode = S_IFREG |  S_IRUGO | S_IWUSR;
1106         la->la_valid = LA_MODE | LA_UID | LA_GID;
1107         memset(dof, 0, sizeof(*dof));
1108         dof->dof_type = dt_mode_to_dft(S_IFREG);
1109
1110         th = dt_trans_create(env, dt);
1111         if (IS_ERR(th))
1112                 GOTO(log, rc = PTR_ERR(th));
1113
1114         rc = dt_declare_create(env, obj, la, NULL, dof, th);
1115         if (rc != 0)
1116                 GOTO(stop, rc);
1117
1118         rc = dt_declare_record_write(env, obj,
1119                                      lfsck_buf_get(env, &lastid,
1120                                                    sizeof(lastid)),
1121                                      pos, th);
1122         if (rc != 0)
1123                 GOTO(stop, rc);
1124
1125         rc = dt_trans_start_local(env, dt, th);
1126         if (rc != 0)
1127                 GOTO(stop, rc);
1128
1129         dt_write_lock(env, obj, 0);
1130         if (likely(dt_object_exists(obj) == 0)) {
1131                 rc = dt_create(env, obj, la, NULL, dof, th);
1132                 if (rc == 0)
1133                         rc = dt_record_write(env, obj,
1134                                 lfsck_buf_get(env, &lastid, sizeof(lastid)),
1135                                 &pos, th);
1136         }
1137         dt_write_unlock(env, obj);
1138
1139         GOTO(stop, rc);
1140
1141 stop:
1142         dt_trans_stop(env, dt, th);
1143
1144 log:
1145         CDEBUG(D_LFSCK, "%s: layout LFSCK will create LAST_ID for <seq> "
1146                LPX64": rc = %d\n",
1147                lfsck_lfsck2name(lfsck), fid_seq(lfsck_dto2fid(obj)), rc);
1148
1149         return rc;
1150 }
1151
1152 static int
1153 lfsck_layout_lastid_reload(const struct lu_env *env,
1154                            struct lfsck_component *com,
1155                            struct lfsck_layout_seq *lls)
1156 {
1157         __u64   lastid;
1158         loff_t  pos     = 0;
1159         int     rc;
1160
1161         dt_read_lock(env, lls->lls_lastid_obj, 0);
1162         rc = dt_record_read(env, lls->lls_lastid_obj,
1163                             lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
1164         dt_read_unlock(env, lls->lls_lastid_obj);
1165         if (unlikely(rc != 0))
1166                 return rc;
1167
1168         lastid = le64_to_cpu(lastid);
1169         if (lastid < lls->lls_lastid_known) {
1170                 struct lfsck_instance   *lfsck  = com->lc_lfsck;
1171                 struct lfsck_layout     *lo     = com->lc_file_ram;
1172
1173                 lls->lls_lastid = lls->lls_lastid_known;
1174                 lls->lls_dirty = 1;
1175                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1176                         LASSERT(lfsck->li_out_notify != NULL);
1177
1178                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1179                                              LE_LASTID_REBUILDING);
1180                         lo->ll_flags |= LF_CRASHED_LASTID;
1181
1182                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
1183                                "LAST_ID file (1) for the sequence "LPX64
1184                                ", old value "LPU64", known value "LPU64"\n",
1185                                lfsck_lfsck2name(lfsck), lls->lls_seq,
1186                                lastid, lls->lls_lastid);
1187                 }
1188         } else if (lastid >= lls->lls_lastid) {
1189                 lls->lls_lastid = lastid;
1190                 lls->lls_dirty = 0;
1191         }
1192
1193         return 0;
1194 }
1195
1196 static int
1197 lfsck_layout_lastid_store(const struct lu_env *env,
1198                           struct lfsck_component *com)
1199 {
1200         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1201         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
1202         struct dt_device                *dt     = lfsck->li_bottom;
1203         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
1204         struct lfsck_layout_seq         *lls;
1205         struct thandle                  *th;
1206         __u64                            lastid;
1207         int                              rc     = 0;
1208         int                              rc1    = 0;
1209
1210         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1211                 loff_t pos = 0;
1212
1213                 if (!lls->lls_dirty)
1214                         continue;
1215
1216                 CDEBUG(D_LFSCK, "%s: layout LFSCK will sync the LAST_ID for "
1217                        "<seq> "LPX64" as <oid> "LPU64"\n",
1218                        lfsck_lfsck2name(lfsck), lls->lls_seq, lls->lls_lastid);
1219
1220                 if (bk->lb_param & LPF_DRYRUN) {
1221                         lls->lls_dirty = 0;
1222                         continue;
1223                 }
1224
1225                 th = dt_trans_create(env, dt);
1226                 if (IS_ERR(th)) {
1227                         rc1 = PTR_ERR(th);
1228                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1229                                "the LAST_ID for <seq> "LPX64"(1): rc = %d\n",
1230                                lfsck_lfsck2name(com->lc_lfsck),
1231                                lls->lls_seq, rc1);
1232                         continue;
1233                 }
1234
1235                 lastid = cpu_to_le64(lls->lls_lastid);
1236                 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1237                                              lfsck_buf_get(env, &lastid,
1238                                                            sizeof(lastid)),
1239                                              pos, th);
1240                 if (rc != 0)
1241                         goto stop;
1242
1243                 rc = dt_trans_start_local(env, dt, th);
1244                 if (rc != 0)
1245                         goto stop;
1246
1247                 dt_write_lock(env, lls->lls_lastid_obj, 0);
1248                 rc = dt_record_write(env, lls->lls_lastid_obj,
1249                                      lfsck_buf_get(env, &lastid,
1250                                      sizeof(lastid)), &pos, th);
1251                 dt_write_unlock(env, lls->lls_lastid_obj);
1252                 if (rc == 0)
1253                         lls->lls_dirty = 0;
1254
1255 stop:
1256                 dt_trans_stop(env, dt, th);
1257                 if (rc != 0) {
1258                         rc1 = rc;
1259                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1260                                "the LAST_ID for <seq> "LPX64"(2): rc = %d\n",
1261                                lfsck_lfsck2name(com->lc_lfsck),
1262                                lls->lls_seq, rc1);
1263                 }
1264         }
1265
1266         return rc1;
1267 }
1268
1269 static int
1270 lfsck_layout_lastid_load(const struct lu_env *env,
1271                          struct lfsck_component *com,
1272                          struct lfsck_layout_seq *lls)
1273 {
1274         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1275         struct lfsck_layout     *lo     = com->lc_file_ram;
1276         struct lu_fid           *fid    = &lfsck_env_info(env)->lti_fid;
1277         struct dt_object        *obj;
1278         loff_t                   pos    = 0;
1279         int                      rc;
1280         ENTRY;
1281
1282         lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck));
1283         obj = dt_locate(env, lfsck->li_bottom, fid);
1284         if (IS_ERR(obj))
1285                 RETURN(PTR_ERR(obj));
1286
1287         /* LAST_ID crashed, to be rebuilt */
1288         if (dt_object_exists(obj) == 0) {
1289                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1290                         LASSERT(lfsck->li_out_notify != NULL);
1291
1292                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1293                                              LE_LASTID_REBUILDING);
1294                         lo->ll_flags |= LF_CRASHED_LASTID;
1295
1296                         CDEBUG(D_LFSCK, "%s: layout LFSCK cannot find the "
1297                                "LAST_ID file for sequence "LPX64"\n",
1298                                lfsck_lfsck2name(lfsck), lls->lls_seq);
1299
1300                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1301                             cfs_fail_val > 0) {
1302                                 struct l_wait_info lwi = LWI_TIMEOUT(
1303                                                 cfs_time_seconds(cfs_fail_val),
1304                                                 NULL, NULL);
1305
1306                                 /* Some others may changed the cfs_fail_val
1307                                  * as zero after above check, re-check it for
1308                                  * sure to avoid falling into wait for ever. */
1309                                 if (likely(lwi.lwi_timeout > 0)) {
1310                                         struct ptlrpc_thread *thread =
1311                                                 &lfsck->li_thread;
1312
1313                                         up_write(&com->lc_sem);
1314                                         l_wait_event(thread->t_ctl_waitq,
1315                                                      !thread_is_running(thread),
1316                                                      &lwi);
1317                                         down_write(&com->lc_sem);
1318                                 }
1319                         }
1320                 }
1321
1322                 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1323         } else {
1324                 dt_read_lock(env, obj, 0);
1325                 rc = dt_read(env, obj,
1326                         lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1327                         &pos);
1328                 dt_read_unlock(env, obj);
1329                 if (rc != 0 && rc != sizeof(__u64))
1330                         GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1331
1332                 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1333                         LASSERT(lfsck->li_out_notify != NULL);
1334
1335                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1336                                              LE_LASTID_REBUILDING);
1337                         lo->ll_flags |= LF_CRASHED_LASTID;
1338
1339                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds invalid "
1340                                "LAST_ID file for the sequence "LPX64
1341                                ": rc = %d\n",
1342                                lfsck_lfsck2name(lfsck), lls->lls_seq, rc);
1343                 }
1344
1345                 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1346                 rc = 0;
1347         }
1348
1349         GOTO(out, rc);
1350
1351 out:
1352         if (rc != 0)
1353                 lfsck_object_put(env, obj);
1354         else
1355                 lls->lls_lastid_obj = obj;
1356
1357         return rc;
1358 }
1359
1360 static void lfsck_layout_record_failure(const struct lu_env *env,
1361                                         struct lfsck_instance *lfsck,
1362                                         struct lfsck_layout *lo)
1363 {
1364         __u64 cookie;
1365
1366         lo->ll_objs_failed_phase1++;
1367         cookie = lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
1368                                                         lfsck->li_di_oit);
1369         if (lo->ll_pos_first_inconsistent == 0 ||
1370             lo->ll_pos_first_inconsistent < cookie) {
1371                 lo->ll_pos_first_inconsistent = cookie;
1372
1373                 CDEBUG(D_LFSCK, "%s: layout LFSCK hit first non-repaired "
1374                        "inconsistency at the pos ["LPU64"]\n",
1375                        lfsck_lfsck2name(lfsck),
1376                        lo->ll_pos_first_inconsistent);
1377         }
1378 }
1379
1380 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1381                                            struct lfsck_component *com,
1382                                            int rc)
1383 {
1384         struct lfsck_instance   *lfsck = com->lc_lfsck;
1385         struct lfsck_layout     *lo    = com->lc_file_ram;
1386
1387         down_write(&com->lc_sem);
1388         lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1389                                 HALF_SEC - lfsck->li_time_last_checkpoint);
1390         lo->ll_time_last_checkpoint = cfs_time_current_sec();
1391         lo->ll_objs_checked_phase2 += com->lc_new_checked;
1392
1393         if (rc > 0) {
1394                 if (lo->ll_flags & LF_INCOMPLETE) {
1395                         lo->ll_status = LS_PARTIAL;
1396                 } else {
1397                         if (lfsck->li_master) {
1398                                 struct lfsck_assistant_data *lad = com->lc_data;
1399
1400                                 if (lad->lad_incomplete)
1401                                         lo->ll_status = LS_PARTIAL;
1402                                 else
1403                                         lo->ll_status = LS_COMPLETED;
1404                         } else {
1405                                 lo->ll_status = LS_COMPLETED;
1406                         }
1407                 }
1408                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
1409                         lo->ll_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
1410                 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1411                 lo->ll_success_count++;
1412         } else if (rc == 0) {
1413                 if (lfsck->li_status != 0)
1414                         lo->ll_status = lfsck->li_status;
1415                 else
1416                         lo->ll_status = LS_STOPPED;
1417         } else {
1418                 lo->ll_status = LS_FAILED;
1419         }
1420
1421         rc = lfsck_layout_store(env, com);
1422         up_write(&com->lc_sem);
1423
1424         return rc;
1425 }
1426
1427 static int lfsck_layout_trans_stop(const struct lu_env *env,
1428                                    struct dt_device *dev,
1429                                    struct thandle *handle, int result)
1430 {
1431         int rc;
1432
1433         handle->th_result = result;
1434         rc = dt_trans_stop(env, dev, handle);
1435         if (rc > 0)
1436                 rc = 0;
1437         else if (rc == 0)
1438                 rc = 1;
1439
1440         return rc;
1441 }
1442
1443 /**
1444  * Get the system default stripe size.
1445  *
1446  * \param[in] env       pointer to the thread context
1447  * \param[in] lfsck     pointer to the lfsck instance
1448  * \param[out] size     pointer to the default stripe size
1449  *
1450  * \retval              0 for success
1451  * \retval              negative error number on failure
1452  */
1453 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1454                                            struct lfsck_instance *lfsck,
1455                                            __u32 *size)
1456 {
1457         struct lov_user_md      *lum = &lfsck_env_info(env)->lti_lum;
1458         struct dt_object        *root;
1459         int                      rc;
1460
1461         root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1462         if (IS_ERR(root))
1463                 return PTR_ERR(root);
1464
1465         /* Get the default stripe size via xattr_get on the backend root. */
1466         rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1467                           XATTR_NAME_LOV, BYPASS_CAPA);
1468         if (rc > 0) {
1469                 /* The lum->lmm_stripe_size is LE mode. The *size also
1470                  * should be LE mode. So it is unnecessary to convert. */
1471                 *size = lum->lmm_stripe_size;
1472                 rc = 0;
1473         } else if (unlikely(rc == 0)) {
1474                 rc = -EINVAL;
1475         }
1476
1477         lfsck_object_put(env, root);
1478
1479         return rc;
1480 }
1481
1482 /**
1483  * \retval       +1: repaired
1484  * \retval        0: did nothing
1485  * \retval      -ve: on error
1486  */
1487 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1488                                      struct thandle *handle,
1489                                      struct dt_object *parent,
1490                                      struct lu_fid *cfid,
1491                                      struct lu_buf *buf,
1492                                      struct lov_ost_data_v1 *slot,
1493                                      int fl, __u32 ost_idx)
1494 {
1495         struct ost_id           *oi     = &lfsck_env_info(env)->lti_oi;
1496         struct lov_mds_md_v1    *lmm    = buf->lb_buf;
1497         struct lu_buf            ea_buf;
1498         int                      rc;
1499         __u32                    magic;
1500         __u16                    count;
1501         ENTRY;
1502
1503         magic = le32_to_cpu(lmm->lmm_magic);
1504         count = le16_to_cpu(lmm->lmm_stripe_count);
1505
1506         fid_to_ostid(cfid, oi);
1507         ostid_cpu_to_le(oi, &slot->l_ost_oi);
1508         slot->l_ost_gen = cpu_to_le32(0);
1509         slot->l_ost_idx = cpu_to_le32(ost_idx);
1510
1511         if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE) {
1512                 struct lov_ost_data_v1 *objs;
1513                 int                     i;
1514
1515                 if (magic == LOV_MAGIC_V1)
1516                         objs = &lmm->lmm_objects[0];
1517                 else
1518                         objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1519                 for (i = 0; i < count; i++, objs++) {
1520                         if (objs != slot && lovea_slot_is_dummy(objs))
1521                                 break;
1522                 }
1523
1524                 /* If the @slot is the last dummy slot to be refilled,
1525                  * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1526                 if (i == count)
1527                         lmm->lmm_pattern &= ~cpu_to_le32(LOV_PATTERN_F_HOLE);
1528         }
1529
1530         lfsck_buf_init(&ea_buf, lmm, lov_mds_md_size(count, magic));
1531         rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle,
1532                           BYPASS_CAPA);
1533         if (rc == 0)
1534                 rc = 1;
1535
1536         RETURN(rc);
1537 }
1538
1539 /**
1540  * \retval       +1: repaired
1541  * \retval        0: did nothing
1542  * \retval      -ve: on error
1543  */
1544 static int lfsck_layout_extend_lovea(const struct lu_env *env,
1545                                      struct lfsck_instance *lfsck,
1546                                      struct thandle *handle,
1547                                      struct dt_object *parent,
1548                                      struct lu_fid *cfid,
1549                                      struct lu_buf *buf, int fl,
1550                                      __u32 ost_idx, __u32 ea_off, bool reset)
1551 {
1552         struct lov_mds_md_v1    *lmm    = buf->lb_buf;
1553         struct lov_ost_data_v1  *objs;
1554         int                      rc;
1555         __u16                    count;
1556         bool                     hole   = false;
1557         ENTRY;
1558
1559         if (fl == LU_XATTR_CREATE || reset) {
1560                 __u32 pattern = LOV_PATTERN_RAID0;
1561
1562                 count = ea_off + 1;
1563                 LASSERT(buf->lb_len >= lov_mds_md_size(count, LOV_MAGIC_V1));
1564
1565                 if (ea_off != 0 || reset) {
1566                         pattern |= LOV_PATTERN_F_HOLE;
1567                         hole = true;
1568                 }
1569
1570                 memset(lmm, 0, buf->lb_len);
1571                 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1572                 lmm->lmm_pattern = cpu_to_le32(pattern);
1573                 fid_to_lmm_oi(lfsck_dto2fid(parent), &lmm->lmm_oi);
1574                 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1575
1576                 rc = lfsck_layout_get_def_stripesize(env, lfsck,
1577                                                      &lmm->lmm_stripe_size);
1578                 if (rc != 0)
1579                         RETURN(rc);
1580
1581                 objs = &lmm->lmm_objects[ea_off];
1582         } else {
1583                 __u32   magic = le32_to_cpu(lmm->lmm_magic);
1584                 int     gap;
1585
1586                 count = le16_to_cpu(lmm->lmm_stripe_count);
1587                 if (magic == LOV_MAGIC_V1)
1588                         objs = &lmm->lmm_objects[count];
1589                 else
1590                         objs = &((struct lov_mds_md_v3 *)lmm)->
1591                                                         lmm_objects[count];
1592
1593                 gap = ea_off - count;
1594                 if (gap >= 0)
1595                         count = ea_off + 1;
1596                 LASSERT(buf->lb_len >= lov_mds_md_size(count, magic));
1597
1598                 if (gap > 0) {
1599                         memset(objs, 0, gap * sizeof(*objs));
1600                         lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
1601                         hole = true;
1602                 }
1603
1604                 lmm->lmm_layout_gen =
1605                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
1606                 objs += gap;
1607         }
1608
1609         lmm->lmm_stripe_count = cpu_to_le16(count);
1610         rc = lfsck_layout_refill_lovea(env, handle, parent, cfid, buf, objs,
1611                                        fl, ost_idx);
1612
1613         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
1614                DFID": parent "DFID", OST-index %u, stripe-index %u, fl %d, "
1615                "reset %s, %s LOV EA hole: rc = %d\n",
1616                lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1617                ost_idx, ea_off, fl, reset ? "yes" : "no",
1618                hole ? "with" : "without", rc);
1619
1620         RETURN(rc);
1621 }
1622
1623 static int __lfsck_layout_update_pfid(const struct lu_env *env,
1624                                       struct dt_object *child,
1625                                       const struct lu_fid *pfid, __u32 offset)
1626 {
1627         struct dt_device        *dev    = lfsck_obj2dev(child);
1628         struct filter_fid       *ff     = &lfsck_env_info(env)->lti_new_pfid;
1629         struct thandle          *handle;
1630         struct lu_buf            buf    = { NULL };
1631         int                      rc;
1632
1633         ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
1634         ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
1635         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
1636          * MDT-object's FID::f_ver, instead it is the OST-object index in its
1637          * parent MDT-object's layout EA. */
1638         ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
1639         lfsck_buf_init(&buf, ff, sizeof(struct filter_fid));
1640
1641         handle = dt_trans_create(env, dev);
1642         if (IS_ERR(handle))
1643                 RETURN(PTR_ERR(handle));
1644
1645         rc = dt_declare_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
1646         if (rc != 0)
1647                 GOTO(stop, rc);
1648
1649         rc = dt_trans_start_local(env, dev, handle);
1650         if (rc != 0)
1651                 GOTO(stop, rc);
1652
1653         rc = dt_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle,
1654                           BYPASS_CAPA);
1655
1656         GOTO(stop, rc);
1657
1658 stop:
1659         dt_trans_stop(env, dev, handle);
1660
1661         return rc;
1662 }
1663
1664 /**
1665  * \retval       +1: repaired
1666  * \retval        0: did nothing
1667  * \retval      -ve: on error
1668  */
1669 static int lfsck_layout_update_pfid(const struct lu_env *env,
1670                                     struct lfsck_component *com,
1671                                     struct dt_object *parent,
1672                                     struct lu_fid *cfid,
1673                                     struct dt_device *cdev, __u32 ea_off)
1674 {
1675         struct dt_object        *child;
1676         int                      rc     = 0;
1677         ENTRY;
1678
1679         child = lfsck_object_find_by_dev(env, cdev, cfid);
1680         if (IS_ERR(child))
1681                 RETURN(PTR_ERR(child));
1682
1683         rc = __lfsck_layout_update_pfid(env, child,
1684                                         lu_object_fid(&parent->do_lu), ea_off);
1685         lfsck_object_put(env, child);
1686
1687         RETURN(rc == 0 ? 1 : rc);
1688 }
1689
1690 /**
1691  * This function will create the MDT-object with the given (partial) LOV EA.
1692  *
1693  * Under some data corruption cases, the MDT-object of the file may be lost,
1694  * but its OST-objects, or some of them are there. The layout LFSCK needs to
1695  * re-create the MDT-object with the orphan OST-object(s) information.
1696  *
1697  * On the other hand, the LFSCK may has created some OST-object for repairing
1698  * dangling LOV EA reference, but as the LFSCK processing, it may find that
1699  * the old OST-object is there and should replace the former new created OST
1700  * object. Unfortunately, some others have modified such newly created object.
1701  * To keep the data (both new and old), the LFSCK will create MDT-object with
1702  * new FID to reference the original OST-object.
1703  *
1704  * \param[in] env       pointer to the thread context
1705  * \param[in] com       pointer to the lfsck component
1706  * \param[in] ltd       pointer to target device descriptor
1707  * \param[in] rec       pointer to the record for the orphan OST-object
1708  * \param[in] cfid      pointer to FID for the orphan OST-object
1709  * \param[in] infix     additional information, such as the FID for original
1710  *                      MDT-object and the stripe offset in the LOV EA
1711  * \param[in] type      the type for describing why the orphan MDT-object is
1712  *                      created. The rules are as following:
1713  *
1714  *  type "C":           Multiple OST-objects claim the same MDT-object and the
1715  *                      same slot in the layout EA. Then the LFSCK will create
1716  *                      new MDT-object(s) to hold the conflict OST-object(s).
1717  *
1718  *  type "N":           The orphan OST-object does not know which one was the
1719  *                      real parent MDT-object, so the LFSCK uses new FID for
1720  *                      its parent MDT-object.
1721  *
1722  *  type "R":           The orphan OST-object knows its parent MDT-object FID,
1723  *                      but does not know the position (the file name) in the
1724  *                      layout.
1725  *
1726  *  type "D":           The MDT-object is a directory, it may knows its parent
1727  *                      but because there is no valid linkEA, the LFSCK cannot
1728  *                      know where to put it back to the namespace.
1729  *  type "O":           The MDT-object has no linkEA, and there is no name
1730  *                      entry that references the MDT-object.
1731  *
1732  *  type "P":           The orphan object to be created was a parent directory
1733  *                      of some MDT-object which linkEA shows that the @orphan
1734  *                      object is missing.
1735  *
1736  * The orphan name will be like:
1737  * ${FID}-${infix}-${type}-${conflict_version}
1738  *
1739  * \param[in] ea_off    the stripe offset in the LOV EA
1740  *
1741  * \retval              positive on repaired something
1742  * \retval              0 if needs to repair nothing
1743  * \retval              negative error number on failure
1744  */
1745 static int lfsck_layout_recreate_parent(const struct lu_env *env,
1746                                         struct lfsck_component *com,
1747                                         struct lfsck_tgt_desc *ltd,
1748                                         struct lu_orphan_rec *rec,
1749                                         struct lu_fid *cfid,
1750                                         const char *infix,
1751                                         const char *type,
1752                                         __u32 ea_off)
1753 {
1754         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1755         struct dt_insert_rec            *dtrec  = &info->lti_dt_rec;
1756         char                            *name   = info->lti_key;
1757         struct lu_attr                  *la     = &info->lti_la;
1758         struct dt_object_format         *dof    = &info->lti_dof;
1759         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1760         struct lu_fid                   *pfid   = &rec->lor_fid;
1761         struct lu_fid                   *tfid   = &info->lti_fid3;
1762         struct dt_device                *dev    = lfsck->li_bottom;
1763         struct dt_object                *lpf    = lfsck->li_lpf_obj;
1764         struct dt_object                *pobj   = NULL;
1765         struct dt_object                *cobj   = NULL;
1766         struct thandle                  *th     = NULL;
1767         struct lu_buf                   *ea_buf = &info->lti_big_buf;
1768         struct lu_buf                    lov_buf;
1769         struct lustre_handle             lh     = { 0 };
1770         struct linkea_data               ldata  = { NULL };
1771         struct lu_buf                    linkea_buf;
1772         const struct lu_name            *pname;
1773         int                              size   = 0;
1774         int                              idx    = 0;
1775         int                              rc     = 0;
1776         ENTRY;
1777
1778         if (unlikely(lpf == NULL))
1779                 GOTO(log, rc = -ENXIO);
1780
1781         /* We use two separated transactions to repair the inconsistency.
1782          *
1783          * 1) create the MDT-object locally.
1784          * 2) update the OST-object's PFID EA if necessary.
1785          *
1786          * If 1) succeed, but 2) failed, then the OST-object's PFID EA will be
1787          * updated when the layout LFSCK run next time.
1788          *
1789          * If 1) failed, but 2) succeed, then such MDT-object will be re-created
1790          * when the layout LFSCK run next time. */
1791
1792         if (fid_is_zero(pfid)) {
1793                 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
1794                 if (rc != 0)
1795                         GOTO(log, rc);
1796
1797                 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
1798                 if (IS_ERR(cobj))
1799                         GOTO(log, rc = PTR_ERR(cobj));
1800         }
1801
1802         pobj = lfsck_object_find_by_dev(env, dev, pfid);
1803         if (IS_ERR(pobj))
1804                 GOTO(log, rc = PTR_ERR(pobj));
1805
1806         LASSERT(infix != NULL);
1807         LASSERT(type != NULL);
1808
1809         do {
1810                 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
1811                          type, idx++);
1812                 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1813                                (const struct dt_key *)name, BYPASS_CAPA);
1814                 if (rc != 0 && rc != -ENOENT)
1815                         GOTO(log, rc);
1816         } while (rc == 0);
1817
1818         rc = linkea_data_new(&ldata,
1819                              &lfsck_env_info(env)->lti_linkea_buf);
1820         if (rc != 0)
1821                 GOTO(log, rc);
1822
1823         pname = lfsck_name_get_const(env, name, strlen(name));
1824         rc = linkea_add_buf(&ldata, pname, lfsck_dto2fid(lfsck->li_lpf_obj));
1825         if (rc != 0)
1826                 GOTO(log, rc);
1827
1828         memset(la, 0, sizeof(*la));
1829         la->la_uid = rec->lor_uid;
1830         la->la_gid = rec->lor_gid;
1831         la->la_mode = S_IFREG | S_IRUSR;
1832         la->la_valid = LA_MODE | LA_UID | LA_GID;
1833
1834         memset(dof, 0, sizeof(*dof));
1835         dof->dof_type = dt_mode_to_dft(S_IFREG);
1836         /* Because the dof->dof_reg.striped = 0, the LOD will not create
1837          * the stripe(s). The LFSCK will specify the LOV EA via
1838          * lfsck_layout_extend_lovea(). */
1839
1840         size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
1841         if (ea_buf->lb_len < size) {
1842                 lu_buf_realloc(ea_buf, size);
1843                 if (ea_buf->lb_buf == NULL)
1844                         GOTO(log, rc = -ENOMEM);
1845         }
1846
1847         /* Hold update lock on the .lustre/lost+found/MDTxxxx/.
1848          *
1849          * XXX: Currently, we do not grab the PDO lock as normal create cases,
1850          *      because creating MDT-object for orphan OST-object is rare, we
1851          *      do not much care about the performance. It can be improved in
1852          *      the future when needed. */
1853         rc = lfsck_ibits_lock(env, lfsck, lpf, &lh,
1854                               MDS_INODELOCK_UPDATE, LCK_EX);
1855         if (rc != 0)
1856                 GOTO(log, rc);
1857
1858         /* The 1st transaction. */
1859         th = dt_trans_create(env, dev);
1860         if (IS_ERR(th))
1861                 GOTO(unlock, rc = PTR_ERR(th));
1862
1863         rc = dt_declare_create(env, pobj, la, NULL, dof, th);
1864         if (rc != 0)
1865                 GOTO(stop, rc);
1866
1867         lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
1868         rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
1869                                   LU_XATTR_REPLACE, th);
1870         if (rc != 0)
1871                 GOTO(stop, rc);
1872
1873         dtrec->rec_fid = pfid;
1874         dtrec->rec_type = S_IFREG;
1875         rc = dt_declare_insert(env, lpf,
1876                                (const struct dt_rec *)dtrec,
1877                                (const struct dt_key *)name, th);
1878         if (rc != 0)
1879                 GOTO(stop, rc);
1880
1881         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1882                        ldata.ld_leh->leh_len);
1883         rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
1884                                   XATTR_NAME_LINK, 0, th);
1885         if (rc != 0)
1886                 GOTO(stop, rc);
1887
1888         rc = dt_trans_start_local(env, dev, th);
1889         if (rc != 0)
1890                 GOTO(stop, rc);
1891
1892         dt_write_lock(env, pobj, 0);
1893         rc = dt_create(env, pobj, la, NULL, dof, th);
1894         if (rc == 0)
1895                 rc = lfsck_layout_extend_lovea(env, lfsck, th, pobj, cfid,
1896                                 &lov_buf, 0, ltd->ltd_index, ea_off, true);
1897         dt_write_unlock(env, pobj);
1898         if (rc < 0)
1899                 GOTO(stop, rc);
1900
1901         rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
1902                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1903         if (rc != 0)
1904                 GOTO(stop, rc);
1905
1906         rc = dt_xattr_set(env, pobj, &linkea_buf,
1907                           XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1908         if (rc == 0 && cobj != NULL) {
1909                 dt_trans_stop(env, dev, th);
1910                 th = NULL;
1911
1912                 /* The 2nd transaction. */
1913                 rc = __lfsck_layout_update_pfid(env, cobj, pfid, ea_off);
1914         }
1915
1916         GOTO(stop, rc);
1917
1918 stop:
1919         if (th != NULL)
1920                 dt_trans_stop(env, dev, th);
1921
1922 unlock:
1923         lfsck_ibits_unlock(&lh, LCK_EX);
1924
1925 log:
1926         if (cobj != NULL && !IS_ERR(cobj))
1927                 lfsck_object_put(env, cobj);
1928         if (pobj != NULL && !IS_ERR(pobj))
1929                 lfsck_object_put(env, pobj);
1930
1931         if (rc < 0)
1932                 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
1933                        "recreate the lost MDT-object: parent "DFID
1934                        ", child "DFID", OST-index %u, stripe-index %u, "
1935                        "infix %s, type %s: rc = %d\n",
1936                        lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
1937                        ltd->ltd_index, ea_off, infix, type, rc);
1938
1939         return rc >= 0 ? 1 : rc;
1940 }
1941
1942 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
1943                                                    struct lfsck_component *com,
1944                                                    const struct lu_fid *fid,
1945                                                    __u32 index)
1946 {
1947         struct lfsck_thread_info *info  = lfsck_env_info(env);
1948         struct lfsck_request     *lr    = &info->lti_lr;
1949         struct lfsck_instance    *lfsck = com->lc_lfsck;
1950         struct lfsck_tgt_desc    *ltd;
1951         struct ptlrpc_request    *req;
1952         struct lfsck_request     *tmp;
1953         struct obd_export        *exp;
1954         int                       rc    = 0;
1955         ENTRY;
1956
1957         ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
1958         if (unlikely(ltd == NULL))
1959                 RETURN(-ENXIO);
1960
1961         exp = ltd->ltd_exp;
1962         if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
1963                 GOTO(put, rc = -EOPNOTSUPP);
1964
1965         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
1966         if (req == NULL)
1967                 GOTO(put, rc = -ENOMEM);
1968
1969         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
1970         if (rc != 0) {
1971                 ptlrpc_request_free(req);
1972
1973                 GOTO(put, rc);
1974         }
1975
1976         memset(lr, 0, sizeof(*lr));
1977         lr->lr_event = LE_CONDITIONAL_DESTROY;
1978         lr->lr_active = LFSCK_TYPE_LAYOUT;
1979         lr->lr_fid = *fid;
1980
1981         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
1982         *tmp = *lr;
1983         ptlrpc_request_set_replen(req);
1984
1985         rc = ptlrpc_queue_wait(req);
1986         ptlrpc_req_finished(req);
1987
1988         GOTO(put, rc);
1989
1990 put:
1991         lfsck_tgt_put(ltd);
1992
1993         return rc;
1994 }
1995
1996 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
1997                                                   struct lfsck_component *com,
1998                                                   struct lfsck_request *lr)
1999 {
2000         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2001         struct lu_attr                  *la     = &info->lti_la;
2002         ldlm_policy_data_t              *policy = &info->lti_policy;
2003         struct ldlm_res_id              *resid  = &info->lti_resid;
2004         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2005         struct dt_device                *dev    = lfsck->li_bottom;
2006         struct lu_fid                   *fid    = &lr->lr_fid;
2007         struct dt_object                *obj;
2008         struct thandle                  *th     = NULL;
2009         struct lustre_handle             lh     = { 0 };
2010         __u64                            flags  = 0;
2011         int                              rc     = 0;
2012         ENTRY;
2013
2014         obj = lfsck_object_find_by_dev(env, dev, fid);
2015         if (IS_ERR(obj))
2016                 RETURN(PTR_ERR(obj));
2017
2018         dt_read_lock(env, obj, 0);
2019         if (dt_object_exists(obj) == 0 ||
2020             lfsck_is_dead_obj(obj)) {
2021                 dt_read_unlock(env, obj);
2022
2023                 GOTO(put, rc = -ENOENT);
2024         }
2025
2026         /* Get obj's attr without lock firstly. */
2027         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
2028         dt_read_unlock(env, obj);
2029         if (rc != 0)
2030                 GOTO(put, rc);
2031
2032         if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
2033                 GOTO(put, rc = -ETXTBSY);
2034
2035         /* Acquire extent lock on [0, EOF] to sync with all possible written. */
2036         LASSERT(lfsck->li_namespace != NULL);
2037
2038         memset(policy, 0, sizeof(*policy));
2039         policy->l_extent.end = OBD_OBJECT_EOF;
2040         ost_fid_build_resid(fid, resid);
2041         rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
2042                                     policy, LCK_EX, &flags, ldlm_blocking_ast,
2043                                     ldlm_completion_ast, NULL, NULL, 0,
2044                                     LVB_T_NONE, NULL, &lh);
2045         if (rc != ELDLM_OK)
2046                 GOTO(put, rc = -EIO);
2047
2048         dt_write_lock(env, obj, 0);
2049         /* Get obj's attr within lock again. */
2050         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
2051         if (rc != 0)
2052                 GOTO(unlock, rc);
2053
2054         if (la->la_ctime != 0)
2055                 GOTO(unlock, rc = -ETXTBSY);
2056
2057         th = dt_trans_create(env, dev);
2058         if (IS_ERR(th))
2059                 GOTO(unlock, rc = PTR_ERR(th));
2060
2061         rc = dt_declare_ref_del(env, obj, th);
2062         if (rc != 0)
2063                 GOTO(stop, rc);
2064
2065         rc = dt_declare_destroy(env, obj, th);
2066         if (rc != 0)
2067                 GOTO(stop, rc);
2068
2069         rc = dt_trans_start_local(env, dev, th);
2070         if (rc != 0)
2071                 GOTO(stop, rc);
2072
2073         rc = dt_ref_del(env, obj, th);
2074         if (rc != 0)
2075                 GOTO(stop, rc);
2076
2077         rc = dt_destroy(env, obj, th);
2078         if (rc == 0)
2079                 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2080                        "OST-object "DFID" that was created for reparing "
2081                        "dangling referenced case. But the original missing "
2082                        "OST-object is found now.\n",
2083                        lfsck_lfsck2name(lfsck), PFID(fid));
2084
2085         GOTO(stop, rc);
2086
2087 stop:
2088         dt_trans_stop(env, dev, th);
2089
2090 unlock:
2091         dt_write_unlock(env, obj);
2092         ldlm_lock_decref(&lh, LCK_EX);
2093
2094 put:
2095         lfsck_object_put(env, obj);
2096
2097         return rc;
2098 }
2099
2100 /**
2101  * Some OST-object has occupied the specified layout EA slot.
2102  * Such OST-object may be generated by the LFSCK when repair
2103  * dangling referenced MDT-object, which can be indicated by
2104  * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2105  * is true and such OST-object has not been modified yet, we
2106  * will replace it with the orphan OST-object; otherwise the
2107  * LFSCK will create new MDT-object to reference the orphan.
2108  *
2109  * \retval       +1: repaired
2110  * \retval        0: did nothing
2111  * \retval      -ve: on error
2112  */
2113 static int lfsck_layout_conflict_create(const struct lu_env *env,
2114                                         struct lfsck_component *com,
2115                                         struct lfsck_tgt_desc *ltd,
2116                                         struct lu_orphan_rec *rec,
2117                                         struct dt_object *parent,
2118                                         struct lu_fid *cfid,
2119                                         struct lu_buf *ea_buf,
2120                                         struct lov_ost_data_v1 *slot,
2121                                         __u32 ea_off)
2122 {
2123         struct lfsck_thread_info *info          = lfsck_env_info(env);
2124         struct lu_fid            *cfid2         = &info->lti_fid2;
2125         struct ost_id            *oi            = &info->lti_oi;
2126         struct lov_mds_md_v1     *lmm           = ea_buf->lb_buf;
2127         struct dt_device         *dev           = lfsck_obj2dev(parent);
2128         struct thandle           *th            = NULL;
2129         struct lustre_handle      lh            = { 0 };
2130         __u32                     ost_idx2      = le32_to_cpu(slot->l_ost_idx);
2131         int                       rc            = 0;
2132         ENTRY;
2133
2134         ostid_le_to_cpu(&slot->l_ost_oi, oi);
2135         rc = ostid_to_fid(cfid2, oi, ost_idx2);
2136         if (rc != 0)
2137                 GOTO(out, rc);
2138
2139         /* Hold layout lock on the parent to prevent others to access. */
2140         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2141                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2142                               LCK_EX);
2143         if (rc != 0)
2144                 GOTO(out, rc);
2145
2146         rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2147
2148         /* If the conflict OST-obejct is not created for fixing dangling
2149          * referenced MDT-object in former LFSCK check/repair, or it has
2150          * been modified by others, then we cannot destroy it. Re-create
2151          * a new MDT-object for the orphan OST-object. */
2152         if (rc == -ETXTBSY) {
2153                 /* No need the layout lock on the original parent. */
2154                 lfsck_ibits_unlock(&lh, LCK_EX);
2155
2156                 fid_zero(&rec->lor_fid);
2157                 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2158                          "-"DFID"-%x", PFID(lu_object_fid(&parent->do_lu)),
2159                          ea_off);
2160                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2161                                                 info->lti_tmpbuf, "C", ea_off);
2162
2163                 RETURN(rc);
2164         }
2165
2166         if (rc != 0 && rc != -ENOENT)
2167                 GOTO(unlock, rc);
2168
2169         th = dt_trans_create(env, dev);
2170         if (IS_ERR(th))
2171                 GOTO(unlock, rc = PTR_ERR(th));
2172
2173         rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2174                                   LU_XATTR_REPLACE, th);
2175         if (rc != 0)
2176                 GOTO(stop, rc);
2177
2178         rc = dt_trans_start_local(env, dev, th);
2179         if (rc != 0)
2180                 GOTO(stop, rc);
2181
2182         dt_write_lock(env, parent, 0);
2183         lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2184         rc = lfsck_layout_refill_lovea(env, th, parent, cfid, ea_buf, slot,
2185                                        LU_XATTR_REPLACE, ltd->ltd_index);
2186         dt_write_unlock(env, parent);
2187
2188         GOTO(stop, rc);
2189
2190 stop:
2191         dt_trans_stop(env, dev, th);
2192
2193 unlock:
2194         lfsck_ibits_unlock(&lh, LCK_EX);
2195
2196 out:
2197         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2198                "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2199                "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2200                lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2201                PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2202                ea_off, rc);
2203
2204         return rc >= 0 ? 1 : rc;
2205 }
2206
2207 /**
2208  * \retval       +1: repaired
2209  * \retval        0: did nothing
2210  * \retval      -ve: on error
2211  */
2212 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2213                                        struct lfsck_component *com,
2214                                        struct lfsck_tgt_desc *ltd,
2215                                        struct lu_orphan_rec *rec,
2216                                        struct dt_object *parent,
2217                                        struct lu_fid *cfid,
2218                                        __u32 ost_idx, __u32 ea_off)
2219 {
2220         struct lfsck_thread_info *info          = lfsck_env_info(env);
2221         struct lu_buf            *buf           = &info->lti_big_buf;
2222         struct lu_fid            *fid           = &info->lti_fid2;
2223         struct ost_id            *oi            = &info->lti_oi;
2224         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2225         struct dt_device         *dt            = lfsck_obj2dev(parent);
2226         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2227         struct thandle            *handle       = NULL;
2228         size_t                    lovea_size;
2229         struct lov_mds_md_v1     *lmm;
2230         struct lov_ost_data_v1   *objs;
2231         struct lustre_handle      lh            = { 0 };
2232         __u32                     magic;
2233         int                       fl            = 0;
2234         int                       rc            = 0;
2235         int                       rc1;
2236         int                       i;
2237         __u16                     count;
2238         bool                      locked        = false;
2239         ENTRY;
2240
2241         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2242                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2243                               LCK_EX);
2244         if (rc != 0) {
2245                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2246                        "LOV EA for "DFID": parent "DFID", OST-index %u, "
2247                        "stripe-index %u: rc = %d\n",
2248                        lfsck_lfsck2name(lfsck), PFID(cfid),
2249                        PFID(lfsck_dto2fid(parent)), ost_idx, ea_off, rc);
2250
2251                 RETURN(rc);
2252         }
2253
2254 again:
2255         if (locked) {
2256                 dt_write_unlock(env, parent);
2257                 locked = false;
2258         }
2259
2260         if (handle != NULL) {
2261                 dt_trans_stop(env, dt, handle);
2262                 handle = NULL;
2263         }
2264
2265         if (rc < 0)
2266                 GOTO(unlock_layout, rc);
2267
2268         lovea_size = rc;
2269         if (buf->lb_len < lovea_size) {
2270                 lu_buf_realloc(buf, lovea_size);
2271                 if (buf->lb_buf == NULL)
2272                         GOTO(unlock_layout, rc = -ENOMEM);
2273         }
2274
2275         if (!(bk->lb_param & LPF_DRYRUN)) {
2276                 handle = dt_trans_create(env, dt);
2277                 if (IS_ERR(handle))
2278                         GOTO(unlock_layout, rc = PTR_ERR(handle));
2279
2280                 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2281                                           fl, handle);
2282                 if (rc != 0)
2283                         GOTO(stop, rc);
2284
2285                 rc = dt_trans_start_local(env, dt, handle);
2286                 if (rc != 0)
2287                         GOTO(stop, rc);
2288         }
2289
2290         dt_write_lock(env, parent, 0);
2291         locked = true;
2292         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV, BYPASS_CAPA);
2293         if (rc == -ERANGE) {
2294                 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV,
2295                                   BYPASS_CAPA);
2296                 LASSERT(rc != 0);
2297                 goto again;
2298         } else if (rc == -ENODATA || rc == 0) {
2299                 lovea_size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2300                 /* If the declared is not big enough, re-try. */
2301                 if (buf->lb_len < lovea_size) {
2302                         rc = lovea_size;
2303                         goto again;
2304                 }
2305                 fl = LU_XATTR_CREATE;
2306         } else if (rc < 0) {
2307                 GOTO(unlock_parent, rc);
2308         } else if (unlikely(buf->lb_len == 0)) {
2309                 goto again;
2310         } else {
2311                 fl = LU_XATTR_REPLACE;
2312                 lovea_size = rc;
2313         }
2314
2315         if (fl == LU_XATTR_CREATE) {
2316                 if (bk->lb_param & LPF_DRYRUN)
2317                         GOTO(unlock_parent, rc = 1);
2318
2319                 LASSERT(buf->lb_len >= lovea_size);
2320
2321                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2322                                                buf, fl, ost_idx, ea_off, false);
2323
2324                 GOTO(unlock_parent, rc);
2325         }
2326
2327         lmm = buf->lb_buf;
2328         rc1 = lfsck_layout_verify_header(lmm);
2329
2330         /* If the LOV EA crashed, the rebuild it. */
2331         if (rc1 == -EINVAL) {
2332                 if (bk->lb_param & LPF_DRYRUN)
2333                         GOTO(unlock_parent, rc = 1);
2334
2335                 LASSERT(buf->lb_len >= lovea_size);
2336
2337                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2338                                                buf, fl, ost_idx, ea_off, true);
2339
2340                 GOTO(unlock_parent, rc);
2341         }
2342
2343         /* For other unknown magic/pattern, keep the current LOV EA. */
2344         if (rc1 != 0)
2345                 GOTO(unlock_parent, rc = rc1);
2346
2347         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2348          * been verified in lfsck_layout_verify_header() already. If some
2349          * new magic introduced in the future, then layout LFSCK needs to
2350          * be updated also. */
2351         magic = le32_to_cpu(lmm->lmm_magic);
2352         if (magic == LOV_MAGIC_V1) {
2353                 objs = &lmm->lmm_objects[0];
2354         } else {
2355                 LASSERT(magic == LOV_MAGIC_V3);
2356                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2357         }
2358
2359         count = le16_to_cpu(lmm->lmm_stripe_count);
2360         if (count == 0)
2361                 GOTO(unlock_parent, rc = -EINVAL);
2362         LASSERT(count > 0);
2363
2364         /* Exceed the current end of MDT-object layout EA. Then extend it. */
2365         if (count <= ea_off) {
2366                 if (bk->lb_param & LPF_DRYRUN)
2367                         GOTO(unlock_parent, rc = 1);
2368
2369                 lovea_size = lov_mds_md_size(ea_off + 1, magic);
2370                 /* If the declared is not big enough, re-try. */
2371                 if (buf->lb_len < lovea_size) {
2372                         rc = lovea_size;
2373                         goto again;
2374                 }
2375
2376                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2377                                                buf, fl, ost_idx, ea_off, false);
2378
2379                 GOTO(unlock_parent, rc);
2380         }
2381
2382         LASSERTF(rc > 0, "invalid rc = %d\n", rc);
2383
2384         for (i = 0; i < count; i++, objs++) {
2385                 /* The MDT-object was created via lfsck_layout_recover_create()
2386                  * by others before, and we fill the dummy layout EA. */
2387                 if (lovea_slot_is_dummy(objs)) {
2388                         if (i != ea_off)
2389                                 continue;
2390
2391                         if (bk->lb_param & LPF_DRYRUN)
2392                                 GOTO(unlock_parent, rc = 1);
2393
2394                         lmm->lmm_layout_gen =
2395                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2396                         rc = lfsck_layout_refill_lovea(env, handle, parent,
2397                                                        cfid, buf, objs, fl,
2398                                                        ost_idx);
2399
2400                         CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
2401                                "dummy layout slot for "DFID": parent "DFID
2402                                ", OST-index %u, stripe-index %u: rc = %d\n",
2403                                lfsck_lfsck2name(lfsck), PFID(cfid),
2404                                PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
2405
2406                         GOTO(unlock_parent, rc);
2407                 }
2408
2409                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2410                 rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
2411                 if (rc != 0) {
2412                         CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
2413                                "invalid layout EA at the slot %d, index %u\n",
2414                                lfsck_lfsck2name(lfsck),
2415                                PFID(lfsck_dto2fid(parent)), i,
2416                                le32_to_cpu(objs->l_ost_idx));
2417
2418                         GOTO(unlock_parent, rc);
2419                 }
2420
2421                 /* It should be rare case, the slot is there, but the LFSCK
2422                  * does not handle it during the first-phase cycle scanning. */
2423                 if (unlikely(lu_fid_eq(fid, cfid))) {
2424                         if (i == ea_off) {
2425                                 GOTO(unlock_parent, rc = 0);
2426                         } else {
2427                                 /* Rare case that the OST-object index
2428                                  * does not match the parent MDT-object
2429                                  * layout EA. We trust the later one. */
2430                                 if (bk->lb_param & LPF_DRYRUN)
2431                                         GOTO(unlock_parent, rc = 1);
2432
2433                                 dt_write_unlock(env, parent);
2434                                 if (handle != NULL)
2435                                         dt_trans_stop(env, dt, handle);
2436                                 lfsck_ibits_unlock(&lh, LCK_EX);
2437                                 rc = lfsck_layout_update_pfid(env, com, parent,
2438                                                         cfid, ltd->ltd_tgt, i);
2439
2440                                 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
2441                                        "updated OST-object's pfid for "DFID
2442                                        ": parent "DFID", OST-index %u, "
2443                                        "stripe-index %u: rc = %d\n",
2444                                        lfsck_lfsck2name(lfsck), PFID(cfid),
2445                                        PFID(lfsck_dto2fid(parent)),
2446                                        ltd->ltd_index, i, rc);
2447
2448                                 RETURN(rc);
2449                         }
2450                 }
2451         }
2452
2453         /* The MDT-object exists, but related layout EA slot is occupied
2454          * by others. */
2455         if (bk->lb_param & LPF_DRYRUN)
2456                 GOTO(unlock_parent, rc = 1);
2457
2458         dt_write_unlock(env, parent);
2459         if (handle != NULL)
2460                 dt_trans_stop(env, dt, handle);
2461         lfsck_ibits_unlock(&lh, LCK_EX);
2462         if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
2463                 objs = &lmm->lmm_objects[ea_off];
2464         else
2465                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
2466         rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
2467                                           buf, objs, ea_off);
2468
2469         RETURN(rc);
2470
2471 unlock_parent:
2472         if (locked)
2473                 dt_write_unlock(env, parent);
2474
2475 stop:
2476         if (handle != NULL)
2477                 dt_trans_stop(env, dt, handle);
2478
2479 unlock_layout:
2480         lfsck_ibits_unlock(&lh, LCK_EX);
2481
2482         return rc;
2483 }
2484
2485 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
2486                                         struct lfsck_component *com,
2487                                         struct lfsck_tgt_desc *ltd,
2488                                         struct lu_orphan_rec *rec,
2489                                         struct lu_fid *cfid)
2490 {
2491         struct lfsck_layout     *lo     = com->lc_file_ram;
2492         struct lu_fid           *pfid   = &rec->lor_fid;
2493         struct dt_object        *parent = NULL;
2494         __u32                    ea_off = pfid->f_stripe_idx;
2495         int                      rc     = 0;
2496         ENTRY;
2497
2498         if (!fid_is_sane(cfid))
2499                 GOTO(out, rc = -EINVAL);
2500
2501         if (fid_is_zero(pfid)) {
2502                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2503                                                   "", "N", ea_off);
2504                 GOTO(out, rc);
2505         }
2506
2507         pfid->f_ver = 0;
2508         if (!fid_is_sane(pfid))
2509                 GOTO(out, rc = -EINVAL);
2510
2511         parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
2512         if (IS_ERR(parent))
2513                 GOTO(out, rc = PTR_ERR(parent));
2514
2515         if (unlikely(dt_object_remote(parent) != 0))
2516                 GOTO(put, rc = -EXDEV);
2517
2518         if (dt_object_exists(parent) == 0) {
2519                 lfsck_object_put(env, parent);
2520                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2521                                                   "", "R", ea_off);
2522                 GOTO(out, rc);
2523         }
2524
2525         if (!S_ISREG(lu_object_attr(&parent->do_lu)))
2526                 GOTO(put, rc = -EISDIR);
2527
2528         rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
2529                                          ltd->ltd_index, ea_off);
2530
2531         GOTO(put, rc);
2532
2533 put:
2534         if (rc <= 0)
2535                 lfsck_object_put(env, parent);
2536         else
2537                 /* The layout EA is changed, need to be reloaded next time. */
2538                 lu_object_put_nocache(env, &parent->do_lu);
2539
2540 out:
2541         down_write(&com->lc_sem);
2542         com->lc_new_scanned++;
2543         com->lc_new_checked++;
2544         if (rc > 0) {
2545                 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
2546                 rc = 0;
2547         } else if (rc < 0) {
2548                 lo->ll_objs_failed_phase2++;
2549         }
2550         up_write(&com->lc_sem);
2551
2552         return rc;
2553 }
2554
2555 static int lfsck_layout_scan_orphan(const struct lu_env *env,
2556                                     struct lfsck_component *com,
2557                                     struct lfsck_tgt_desc *ltd)
2558 {
2559         struct lfsck_assistant_data     *lad    = com->lc_data;
2560         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2561         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
2562         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2563         struct ost_id                   *oi     = &info->lti_oi;
2564         struct lu_fid                   *fid    = &info->lti_fid;
2565         struct dt_object                *obj;
2566         const struct dt_it_ops          *iops;
2567         struct dt_it                    *di;
2568         int                              rc     = 0;
2569         ENTRY;
2570
2571         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
2572                "scanning for OST%04x\n",
2573                lfsck_lfsck2name(lfsck), ltd->ltd_index);
2574
2575         if (cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
2576                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
2577                        "scanning for OST%04x\n",
2578                        lfsck_lfsck2name(lfsck), ltd->ltd_index);
2579
2580                 RETURN(0);
2581         }
2582
2583         ostid_set_seq(oi, FID_SEQ_IDIF);
2584         ostid_set_id(oi, 0);
2585         rc = ostid_to_fid(fid, oi, ltd->ltd_index);
2586         if (rc != 0)
2587                 GOTO(log, rc);
2588
2589         obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
2590         if (unlikely(IS_ERR(obj)))
2591                 GOTO(log, rc = PTR_ERR(obj));
2592
2593         rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_orphan_features);
2594         if (rc != 0)
2595                 GOTO(put, rc);
2596
2597         iops = &obj->do_index_ops->dio_it;
2598         di = iops->init(env, obj, 0, BYPASS_CAPA);
2599         if (IS_ERR(di))
2600                 GOTO(put, rc = PTR_ERR(di));
2601
2602         rc = iops->load(env, di, 0);
2603         if (rc == -ESRCH) {
2604                 /* -ESRCH means that the orphan OST-objects rbtree has been
2605                  * cleanup because of the OSS server restart or other errors. */
2606                 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
2607                 GOTO(fini, rc);
2608         }
2609
2610         if (rc == 0)
2611                 rc = iops->next(env, di);
2612         else if (rc > 0)
2613                 rc = 0;
2614
2615         if (rc < 0)
2616                 GOTO(fini, rc);
2617
2618         if (rc > 0)
2619                 GOTO(fini, rc = 0);
2620
2621         do {
2622                 struct dt_key           *key;
2623                 struct lu_orphan_rec    *rec = &info->lti_rec;
2624
2625                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
2626                     unlikely(!thread_is_running(&lfsck->li_thread)))
2627                         break;
2628
2629                 key = iops->key(env, di);
2630                 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
2631                 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
2632                 if (rc == 0)
2633                         rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
2634                                         &com->lc_fid_latest_scanned_phase2);
2635                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2636                         GOTO(fini, rc);
2637
2638                 lfsck_control_speed_by_self(com);
2639                 do {
2640                         rc = iops->next(env, di);
2641                 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
2642         } while (rc == 0);
2643
2644         GOTO(fini, rc);
2645
2646 fini:
2647         iops->put(env, di);
2648         iops->fini(env, di);
2649 put:
2650         lfsck_object_put(env, obj);
2651
2652 log:
2653         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
2654                "scanning for OST%04x: rc = %d\n",
2655                lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
2656
2657         return rc > 0 ? 0 : rc;
2658 }
2659
2660 /* For the MDT-object with dangling reference, we need to repare the
2661  * inconsistency according to the LFSCK sponsor's requirement:
2662  *
2663  * 1) Keep the inconsistency there and report the inconsistency case,
2664  *    then give the chance to the application to find related issues,
2665  *    and the users can make the decision about how to handle it with
2666  *    more human knownledge. (by default)
2667  *
2668  * 2) Re-create the missing OST-object with the FID/owner information. */
2669 static int lfsck_layout_repair_dangling(const struct lu_env *env,
2670                                         struct lfsck_component *com,
2671                                         struct dt_object *parent,
2672                                         struct lfsck_layout_req *llr,
2673                                         const struct lu_attr *pla)
2674 {
2675         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2676         struct filter_fid               *pfid   = &info->lti_new_pfid;
2677         struct dt_object_format         *dof    = &info->lti_dof;
2678         struct lu_attr                  *cla    = &info->lti_la2;
2679         struct dt_object                *child  = llr->llr_child;
2680         struct dt_device                *dev    = lfsck_obj2dev(child);
2681         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
2682         struct thandle                  *handle;
2683         struct lu_buf                   *buf;
2684         struct lustre_handle             lh     = { 0 };
2685         int                              rc;
2686         bool                             create;
2687         ENTRY;
2688
2689         if (com->lc_lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ)
2690                 create = true;
2691         else
2692                 create = false;
2693
2694         if (!create)
2695                 GOTO(log, rc = 1);
2696
2697         memset(cla, 0, sizeof(*cla));
2698         cla->la_uid = pla->la_uid;
2699         cla->la_gid = pla->la_gid;
2700         cla->la_mode = S_IFREG | 0666;
2701         cla->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
2702                         LA_ATIME | LA_MTIME | LA_CTIME;
2703         memset(dof, 0, sizeof(*dof));
2704
2705         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2706                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2707                               LCK_EX);
2708         if (rc != 0)
2709                 GOTO(log, rc);
2710
2711         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
2712         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
2713         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2714          * MDT-object's FID::f_ver, instead it is the OST-object index in its
2715          * parent MDT-object's layout EA. */
2716         pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
2717         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
2718
2719         handle = dt_trans_create(env, dev);
2720         if (IS_ERR(handle))
2721                 GOTO(unlock1, rc = PTR_ERR(handle));
2722
2723         rc = dt_declare_create(env, child, cla, NULL, dof, handle);
2724         if (rc != 0)
2725                 GOTO(stop, rc);
2726
2727         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
2728                                   LU_XATTR_CREATE, handle);
2729         if (rc != 0)
2730                 GOTO(stop, rc);
2731
2732         rc = dt_trans_start_local(env, dev, handle);
2733         if (rc != 0)
2734                 GOTO(stop, rc);
2735
2736         dt_read_lock(env, parent, 0);
2737         if (unlikely(lfsck_is_dead_obj(parent)))
2738                 GOTO(unlock2, rc = 1);
2739
2740         rc = dt_create(env, child, cla, NULL, dof, handle);
2741         if (rc != 0)
2742                 GOTO(unlock2, rc);
2743
2744         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
2745                           handle, BYPASS_CAPA);
2746
2747         GOTO(unlock2, rc);
2748
2749 unlock2:
2750         dt_read_unlock(env, parent);
2751
2752 stop:
2753         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2754
2755 unlock1:
2756         lfsck_ibits_unlock(&lh, LCK_EX);
2757
2758 log:
2759         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found dangling "
2760                "reference for: parent "DFID", child "DFID", OST-index %u, "
2761                "stripe-index %u, owner %u/%u. %s: rc = %d\n",
2762                lfsck_lfsck2name(com->lc_lfsck), PFID(lfsck_dto2fid(parent)),
2763                PFID(lfsck_dto2fid(child)), llr->llr_ost_idx,
2764                llr->llr_lov_idx, pla->la_uid, pla->la_gid,
2765                create ? "Create the lost OST-object as required" :
2766                         "Keep the MDT-object there by default", rc);
2767
2768         return rc;
2769 }
2770
2771 /* If the OST-object does not recognize the MDT-object as its parent, and
2772  * there is no other MDT-object claims as its parent, then just trust the
2773  * given MDT-object as its parent. So update the OST-object filter_fid. */
2774 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
2775                                               struct lfsck_component *com,
2776                                               struct dt_object *parent,
2777                                               struct lfsck_layout_req *llr,
2778                                               const struct lu_attr *pla)
2779 {
2780         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2781         struct filter_fid               *pfid   = &info->lti_new_pfid;
2782         struct lu_attr                  *tla    = &info->lti_la3;
2783         struct dt_object                *child  = llr->llr_child;
2784         struct dt_device                *dev    = lfsck_obj2dev(child);
2785         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
2786         struct thandle                  *handle;
2787         struct lu_buf                   *buf;
2788         struct lustre_handle             lh     = { 0 };
2789         int                              rc;
2790         ENTRY;
2791
2792         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2793                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2794                               LCK_EX);
2795         if (rc != 0)
2796                 GOTO(log, rc);
2797
2798         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
2799         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
2800         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2801          * MDT-object's FID::f_ver, instead it is the OST-object index in its
2802          * parent MDT-object's layout EA. */
2803         pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
2804         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
2805
2806         handle = dt_trans_create(env, dev);
2807         if (IS_ERR(handle))
2808                 GOTO(unlock1, rc = PTR_ERR(handle));
2809
2810         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
2811         if (rc != 0)
2812                 GOTO(stop, rc);
2813
2814         tla->la_valid = LA_UID | LA_GID;
2815         tla->la_uid = pla->la_uid;
2816         tla->la_gid = pla->la_gid;
2817         rc = dt_declare_attr_set(env, child, tla, handle);
2818         if (rc != 0)
2819                 GOTO(stop, rc);
2820
2821         rc = dt_trans_start_local(env, dev, handle);
2822         if (rc != 0)
2823                 GOTO(stop, rc);
2824
2825         dt_write_lock(env, parent, 0);
2826         if (unlikely(lfsck_is_dead_obj(parent)))
2827                 GOTO(unlock2, rc = 1);
2828
2829         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle,
2830                           BYPASS_CAPA);
2831         if (rc != 0)
2832                 GOTO(unlock2, rc);
2833
2834         /* Get the latest parent's owner. */
2835         rc = dt_attr_get(env, parent, tla, BYPASS_CAPA);
2836         if (rc != 0)
2837                 GOTO(unlock2, rc);
2838
2839         tla->la_valid = LA_UID | LA_GID;
2840         rc = dt_attr_set(env, child, tla, handle, BYPASS_CAPA);
2841
2842         GOTO(unlock2, rc);
2843
2844 unlock2:
2845         dt_write_unlock(env, parent);
2846
2847 stop:
2848         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2849
2850 unlock1:
2851         lfsck_ibits_unlock(&lh, LCK_EX);
2852
2853 log:
2854         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired unmatched "
2855                "MDT-OST pair for: parent "DFID", child "DFID", OST-index %u, "
2856                "stripe-index %u, owner %u/%u: rc = %d\n",
2857                lfsck_lfsck2name(com->lc_lfsck), PFID(lfsck_dto2fid(parent)),
2858                PFID(lfsck_dto2fid(child)), llr->llr_ost_idx, llr->llr_lov_idx,
2859                pla->la_uid, pla->la_gid, rc);
2860
2861         return rc;
2862 }
2863
2864 /* If there are more than one MDT-objects claim as the OST-object's parent,
2865  * and the OST-object only recognizes one of them, then we need to generate
2866  * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
2867 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
2868                                                    struct lfsck_component *com,
2869                                                    struct dt_object *parent,
2870                                                    struct lfsck_layout_req *llr,
2871                                                    struct lu_attr *la,
2872                                                    struct lu_buf *buf)
2873 {
2874         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2875         struct dt_allocation_hint       *hint   = &info->lti_hint;
2876         struct dt_object_format         *dof    = &info->lti_dof;
2877         struct ost_id                   *oi     = &info->lti_oi;
2878         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2879         struct dt_device                *dev    = lfsck->li_bottom;
2880         struct lu_device                *d      =
2881                                 &lfsck_obj2dev(llr->llr_child)->dd_lu_dev;
2882         struct lu_object                *o;
2883         struct lu_object                *n;
2884         struct dt_object                *child  = NULL;
2885         struct thandle                  *handle = NULL;
2886         struct lov_mds_md_v1            *lmm;
2887         struct lov_ost_data_v1          *objs;
2888         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
2889         struct lu_fid                    tfid;
2890         struct lustre_handle             lh     = { 0 };
2891         struct lu_buf                    ea_buf;
2892         __u32                            magic;
2893         __u32                            index;
2894         int                              rc;
2895         ENTRY;
2896
2897         /* We use two separated transactions to repair the inconsistency.
2898          *
2899          * 1) create the child (OST-object).
2900          * 2) update the parent LOV EA according to the child's FID.
2901          *
2902          * If 1) succeed, but 2) failed or aborted, then such OST-object will be
2903          * handled as orphan when the layout LFSCK run next time.
2904          *
2905          * If 1) failed, but 2) succeed, then such OST-object will be re-created
2906          * as dangling referened case when the layout LFSCK run next time. */
2907
2908         /* The 1st transaction. */
2909         o = lu_object_anon(env, d, NULL);
2910         if (IS_ERR(o))
2911                 GOTO(log, rc = PTR_ERR(o));
2912
2913         n = lu_object_locate(o->lo_header, d->ld_type);
2914         if (unlikely(n == NULL)) {
2915                 lu_object_put_nocache(env, o);
2916
2917                 GOTO(log, rc = -EINVAL);
2918         }
2919
2920         child = container_of(n, struct dt_object, do_lu);
2921         memset(hint, 0, sizeof(*hint));
2922         la->la_valid = LA_UID | LA_GID;
2923         memset(dof, 0, sizeof(*dof));
2924
2925         handle = dt_trans_create(env, dev);
2926         if (IS_ERR(handle))
2927                 GOTO(log, rc = PTR_ERR(handle));
2928
2929         rc = dt_declare_create(env, child, la, hint, dof, handle);
2930         if (rc != 0)
2931                 GOTO(stop, rc);
2932
2933         rc = dt_trans_start_local(env, dev, handle);
2934         if (rc != 0)
2935                 GOTO(stop, rc);
2936
2937         rc = dt_create(env, child, la, hint, dof, handle);
2938         dt_trans_stop(env, dev, handle);
2939         handle = NULL;
2940         if (rc != 0)
2941                 GOTO(log, rc);
2942
2943         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2944                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2945                               LCK_EX);
2946         if (rc != 0)
2947                 GOTO(log, rc);
2948
2949         /* The 2nd transaction. */
2950
2951         /* XXX: Generally, we should use bottom device (OSD) to update parent
2952          *      LOV EA. But because the LOD-object still references the wrong
2953          *      OSP-object that should be detached after the parent's LOV EA
2954          *      refreshed. Unfortunately, there is no suitable API for that.
2955          *      So we have to make the LOD to re-load the OSP-object(s) via
2956          *      replacing the LOV EA against the LOD-object.
2957          *
2958          *      Once the DNE2 patches have been landed, we can replace the
2959          *      LOD device with the OSD device. LU-6230. */
2960
2961         dev = lfsck->li_next;
2962         parent = lfsck_object_locate(dev, parent);
2963         if (IS_ERR(parent))
2964                 GOTO(log, rc = PTR_ERR(parent));
2965
2966         handle = dt_trans_create(env, dev);
2967         if (IS_ERR(handle))
2968                 GOTO(log, rc = PTR_ERR(handle));
2969
2970         rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2971                                   LU_XATTR_REPLACE, handle);
2972         if (rc != 0)
2973                 GOTO(stop, rc);
2974
2975         rc = dt_trans_start_local(env, dev, handle);
2976         if (rc != 0)
2977                 GOTO(stop, rc);
2978
2979         dt_write_lock(env, parent, 0);
2980         if (unlikely(lfsck_is_dead_obj(parent)))
2981                 GOTO(unlock, rc = 0);
2982
2983         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV, BYPASS_CAPA);
2984         if (unlikely(rc == 0 || rc == -ENODATA || rc == -ERANGE))
2985                 GOTO(unlock, rc = 0);
2986
2987         lmm = buf->lb_buf;
2988         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2989          * been verified in lfsck_layout_verify_header() already. If some
2990          * new magic introduced in the future, then layout LFSCK needs to
2991          * be updated also. */
2992         magic = le32_to_cpu(lmm->lmm_magic);
2993         if (magic == LOV_MAGIC_V1) {
2994                 objs = &lmm->lmm_objects[llr->llr_lov_idx];
2995         } else {
2996                 LASSERT(magic == LOV_MAGIC_V3);
2997                 objs =
2998                 &((struct lov_mds_md_v3 *)lmm)->lmm_objects[llr->llr_lov_idx];
2999         }
3000
3001         ostid_le_to_cpu(&objs->l_ost_oi, oi);
3002         index = le32_to_cpu(objs->l_ost_idx);
3003         rc = ostid_to_fid(&tfid, oi, index);
3004         /* Someone changed layout during the LFSCK, no need to repair then. */
3005         if (rc == 0 && !lu_fid_eq(&tfid, lu_object_fid(&llr->llr_child->do_lu)))
3006                 GOTO(unlock, rc = 0);
3007
3008         lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
3009         fid_to_ostid(lu_object_fid(&child->do_lu), oi);
3010         ostid_cpu_to_le(oi, &objs->l_ost_oi);
3011         objs->l_ost_gen = cpu_to_le32(0);
3012         objs->l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
3013         lfsck_buf_init(&ea_buf, lmm,
3014                        lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3015                                        magic));
3016         rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV,
3017                           LU_XATTR_REPLACE, handle, BYPASS_CAPA);
3018
3019         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
3020
3021 unlock:
3022         dt_write_unlock(env, parent);
3023
3024 stop:
3025         if (handle != NULL)
3026                 dt_trans_stop(env, dev, handle);
3027
3028 log:
3029         lfsck_ibits_unlock(&lh, LCK_EX);
3030         if (child != NULL)
3031                 lfsck_object_put(env, child);
3032
3033         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired multiple "
3034                "references for: parent "DFID", OST-index %u, stripe-index %u, "
3035                "owner %u/%u: rc = %d\n", lfsck_lfsck2name(lfsck), PFID(pfid),
3036                llr->llr_ost_idx, llr->llr_lov_idx, la->la_uid, la->la_gid, rc);
3037
3038         return rc;
3039 }
3040
3041 /* If the MDT-object and the OST-object have different owner information,
3042  * then trust the MDT-object, because the normal chown/chgrp handle order
3043  * is from MDT to OST, and it is possible that some chown/chgrp operation
3044  * is partly done. */
3045 static int lfsck_layout_repair_owner(const struct lu_env *env,
3046                                      struct lfsck_component *com,
3047                                      struct dt_object *parent,
3048                                      struct lfsck_layout_req *llr,
3049                                      struct lu_attr *pla)
3050 {
3051         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3052         struct lu_attr                  *tla    = &info->lti_la3;
3053         struct dt_object                *child  = llr->llr_child;
3054         struct dt_device                *dev    = lfsck_obj2dev(child);
3055         struct thandle                  *handle;
3056         int                              rc;
3057         ENTRY;
3058
3059         handle = dt_trans_create(env, dev);
3060         if (IS_ERR(handle))
3061                 GOTO(log, rc = PTR_ERR(handle));
3062
3063         tla->la_uid = pla->la_uid;
3064         tla->la_gid = pla->la_gid;
3065         tla->la_valid = LA_UID | LA_GID;
3066         rc = dt_declare_attr_set(env, child, tla, handle);
3067         if (rc != 0)
3068                 GOTO(stop, rc);
3069
3070         rc = dt_trans_start_local(env, dev, handle);
3071         if (rc != 0)
3072                 GOTO(stop, rc);
3073
3074         /* Use the dt_object lock to serialize with destroy and attr_set. */
3075         dt_read_lock(env, parent, 0);
3076         if (unlikely(lfsck_is_dead_obj(parent)))
3077                 GOTO(unlock, rc = 1);
3078
3079         /* Get the latest parent's owner. */
3080         rc = dt_attr_get(env, parent, tla, BYPASS_CAPA);
3081         if (rc != 0)
3082                 GOTO(unlock, rc);
3083
3084         /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
3085         if (unlikely(tla->la_uid != pla->la_uid ||
3086                      tla->la_gid != pla->la_gid))
3087                 GOTO(unlock, rc = 1);
3088
3089         tla->la_valid = LA_UID | LA_GID;
3090         rc = dt_attr_set(env, child, tla, handle, BYPASS_CAPA);
3091
3092         GOTO(unlock, rc);
3093
3094 unlock:
3095         dt_read_unlock(env, parent);
3096
3097 stop:
3098         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3099
3100 log:
3101         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired inconsistent "
3102                "file owner for: parent "DFID", child "DFID", OST-index %u, "
3103                "stripe-index %u, owner %u/%u: rc = %d\n",
3104                lfsck_lfsck2name(com->lc_lfsck), PFID(lfsck_dto2fid(parent)),
3105                PFID(lfsck_dto2fid(child)), llr->llr_ost_idx, llr->llr_lov_idx,
3106                pla->la_uid, pla->la_gid, rc);
3107
3108         return rc;
3109 }
3110
3111 /* Check whether the OST-object correctly back points to the
3112  * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
3113 static int lfsck_layout_check_parent(const struct lu_env *env,
3114                                      struct lfsck_component *com,
3115                                      struct dt_object *parent,
3116                                      const struct lu_fid *pfid,
3117                                      const struct lu_fid *cfid,
3118                                      const struct lu_attr *pla,
3119                                      const struct lu_attr *cla,
3120                                      struct lfsck_layout_req *llr,
3121                                      struct lu_buf *lov_ea, __u32 idx)
3122 {
3123         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3124         struct lu_buf                   *buf    = &info->lti_big_buf;
3125         struct dt_object                *tobj;
3126         struct lov_mds_md_v1            *lmm;
3127         struct lov_ost_data_v1          *objs;
3128         struct lustre_handle             lh     = { 0 };
3129         int                              rc;
3130         int                              i;
3131         __u32                            magic;
3132         __u16                            count;
3133         ENTRY;
3134
3135         if (fid_is_zero(pfid)) {
3136                 /* client never wrote. */
3137                 if (cla->la_size == 0 && cla->la_blocks == 0) {
3138                         if (unlikely(cla->la_uid != pla->la_uid ||
3139                                      cla->la_gid != pla->la_gid))
3140                                 RETURN (LLIT_INCONSISTENT_OWNER);
3141
3142                         RETURN(0);
3143                 }
3144
3145                 RETURN(LLIT_UNMATCHED_PAIR);
3146         }
3147
3148         if (unlikely(!fid_is_sane(pfid)))
3149                 RETURN(LLIT_UNMATCHED_PAIR);
3150
3151         if (lu_fid_eq(pfid, lu_object_fid(&parent->do_lu))) {
3152                 if (llr->llr_lov_idx == idx)
3153                         RETURN(0);
3154
3155                 RETURN(LLIT_UNMATCHED_PAIR);
3156         }
3157
3158         tobj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3159         if (IS_ERR(tobj))
3160                 RETURN(PTR_ERR(tobj));
3161
3162         if (dt_object_exists(tobj) == 0 ||
3163             lfsck_is_dead_obj(tobj))
3164                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3165
3166         if (!S_ISREG(lfsck_object_type(tobj)))
3167                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3168
3169         /* Load the tobj's layout EA, in spite of it is a local MDT-object or
3170          * remote one on another MDT. Then check whether the given OST-object
3171          * is in such layout. If yes, it is multiple referenced, otherwise it
3172          * is unmatched referenced case. */
3173         rc = lfsck_layout_get_lovea(env, tobj, buf);
3174         if (rc == 0 || rc == -ENOENT)
3175                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3176
3177         if (rc < 0)
3178                 GOTO(out, rc);
3179
3180         lmm = buf->lb_buf;
3181         magic = le32_to_cpu(lmm->lmm_magic);
3182         if (magic == LOV_MAGIC_V1) {
3183                 objs = &lmm->lmm_objects[0];
3184         } else {
3185                 LASSERT(magic == LOV_MAGIC_V3);
3186                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3187         }
3188
3189         count = le16_to_cpu(lmm->lmm_stripe_count);
3190         for (i = 0; i < count; i++, objs++) {
3191                 struct lu_fid           *tfid   = &info->lti_fid2;
3192                 struct ost_id           *oi     = &info->lti_oi;
3193                 __u32                    idx2;
3194
3195                 if (lovea_slot_is_dummy(objs))
3196                         continue;
3197
3198                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3199                 idx2 = le32_to_cpu(objs->l_ost_idx);
3200                 rc = ostid_to_fid(tfid, oi, idx2);
3201                 if (rc != 0) {
3202                         CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
3203                                "invalid layout EA at the slot %d, index %u\n",
3204                                lfsck_lfsck2name(com->lc_lfsck),
3205                                PFID(pfid), i, idx2);
3206
3207                         GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3208                 }
3209
3210                 if (lu_fid_eq(cfid, tfid)) {
3211                         rc = lfsck_ibits_lock(env, com->lc_lfsck, tobj, &lh,
3212                                               MDS_INODELOCK_UPDATE |
3213                                               MDS_INODELOCK_LAYOUT |
3214                                               MDS_INODELOCK_XATTR,
3215                                               LCK_EX);
3216                         if (rc != 0)
3217                                 GOTO(out, rc);
3218
3219                         dt_read_lock(env, tobj, 0);
3220
3221                         /* For local MDT-object, re-check existence
3222                          * after taken the lock. */
3223                         if (!dt_object_remote(tobj)) {
3224                                 if (dt_object_exists(tobj) == 0 ||
3225                                     lfsck_is_dead_obj(tobj)) {
3226                                         rc = LLIT_UNMATCHED_PAIR;
3227                                 } else {
3228                                         *lov_ea = *buf;
3229                                         rc = LLIT_MULTIPLE_REFERENCED;
3230                                 }
3231
3232                                 GOTO(unlock, rc);
3233                         }
3234
3235                         /* For migration case, the new MDT-object and old
3236                          * MDT-object may reference the same OST-object at
3237                          * some migration internal time.
3238                          *
3239                          * For remote MDT-object, the local MDT may not know
3240                          * whether it has been removed or not.  Try checking
3241                          * for a non-existent xattr to check if this object
3242                          * has been been removed or not. */
3243                         rc = dt_xattr_get(env, tobj, &LU_BUF_NULL,
3244                                           XATTR_NAME_DUMMY, BYPASS_CAPA);
3245                         if (unlikely(rc == -ENOENT || rc >= 0)) {
3246                                 rc = LLIT_UNMATCHED_PAIR;
3247                         } else if (rc == -ENODATA) {
3248                                 *lov_ea = *buf;
3249                                 rc = LLIT_MULTIPLE_REFERENCED;
3250                         }
3251
3252                         GOTO(unlock, rc);
3253                 }
3254         }
3255
3256         GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3257
3258 unlock:
3259         if (lustre_handle_is_used(&lh)) {
3260                 dt_read_unlock(env, tobj);
3261                 lfsck_ibits_unlock(&lh, LCK_EX);
3262         }
3263
3264 out:
3265         lfsck_object_put(env, tobj);
3266
3267         return rc;
3268 }
3269
3270 static int lfsck_layout_assistant_handler_p1(const struct lu_env *env,
3271                                              struct lfsck_component *com,
3272                                              struct lfsck_assistant_req *lar)
3273 {
3274         struct lfsck_layout_req              *llr    =
3275                         container_of0(lar, struct lfsck_layout_req, llr_lar);
3276         struct lfsck_layout                  *lo     = com->lc_file_ram;
3277         struct lfsck_thread_info             *info   = lfsck_env_info(env);
3278         struct filter_fid_old                *pea    = &info->lti_old_pfid;
3279         struct lu_fid                        *pfid   = &info->lti_fid;
3280         struct lu_buf                         buf    = { NULL };
3281         struct dt_object                     *parent;
3282         struct dt_object                     *child  = llr->llr_child;
3283         struct lu_attr                       *pla    = &info->lti_la;
3284         struct lu_attr                       *cla    = &info->lti_la2;
3285         struct lfsck_instance                *lfsck  = com->lc_lfsck;
3286         struct lfsck_bookmark                *bk     = &lfsck->li_bookmark_ram;
3287         enum lfsck_layout_inconsistency_type  type   = LLIT_NONE;
3288         __u32                                 idx    = 0;
3289         int                                   rc;
3290         ENTRY;
3291
3292         parent = lfsck_object_find_bottom(env, lfsck, &lar->lar_fid);
3293         if (IS_ERR(parent))
3294                 RETURN(PTR_ERR(parent));
3295
3296         if (unlikely(lfsck_is_dead_obj(parent)))
3297                 GOTO(put_parent, rc = 0);
3298
3299         rc = dt_attr_get(env, parent, pla, BYPASS_CAPA);
3300         if (rc != 0)
3301                 GOTO(out, rc);
3302
3303         rc = dt_attr_get(env, child, cla, BYPASS_CAPA);
3304         if (rc == -ENOENT) {
3305                 if (unlikely(lfsck_is_dead_obj(parent)))
3306                         GOTO(put_parent, rc = 0);
3307
3308                 type = LLIT_DANGLING;
3309                 goto repair;
3310         }
3311
3312         if (rc != 0)
3313                 GOTO(out, rc);
3314
3315         lfsck_buf_init(&buf, pea, sizeof(struct filter_fid_old));
3316         rc = dt_xattr_get(env, child, &buf, XATTR_NAME_FID, BYPASS_CAPA);
3317         if (unlikely(rc >= 0 && rc != sizeof(struct filter_fid_old) &&
3318                      rc != sizeof(struct filter_fid))) {
3319                 type = LLIT_UNMATCHED_PAIR;
3320                 goto repair;
3321         }
3322
3323         if (rc < 0 && rc != -ENODATA)
3324                 GOTO(out, rc);
3325
3326         if (rc == -ENODATA) {
3327                 fid_zero(pfid);
3328         } else {
3329                 fid_le_to_cpu(pfid, &pea->ff_parent);
3330                 /* Currently, the filter_fid::ff_parent::f_ver is not the
3331                  * real parent MDT-object's FID::f_ver, instead it is the
3332                  * OST-object index in its parent MDT-object's layout EA. */
3333                 idx = pfid->f_stripe_idx;
3334                 pfid->f_ver = 0;
3335         }
3336
3337         rc = lfsck_layout_check_parent(env, com, parent, pfid,
3338                                        lu_object_fid(&child->do_lu),
3339                                        pla, cla, llr, &buf, idx);
3340         if (rc > 0) {
3341                 type = rc;
3342                 goto repair;
3343         }
3344
3345         if (rc < 0)
3346                 GOTO(out, rc);
3347
3348         if (unlikely(cla->la_uid != pla->la_uid ||
3349                      cla->la_gid != pla->la_gid)) {
3350                 type = LLIT_INCONSISTENT_OWNER;
3351                 goto repair;
3352         }
3353
3354 repair:
3355         if (bk->lb_param & LPF_DRYRUN) {
3356                 if (type != LLIT_NONE)
3357                         GOTO(out, rc = 1);
3358                 else
3359                         GOTO(out, rc = 0);
3360         }
3361
3362         switch (type) {
3363         case LLIT_DANGLING:
3364                 rc = lfsck_layout_repair_dangling(env, com, parent, llr, pla);
3365                 break;
3366         case LLIT_UNMATCHED_PAIR:
3367                 rc = lfsck_layout_repair_unmatched_pair(env, com, parent,
3368                                                         llr, pla);
3369                 break;
3370         case LLIT_MULTIPLE_REFERENCED:
3371                 rc = lfsck_layout_repair_multiple_references(env, com, parent,
3372                                                              llr, pla, &buf);
3373                 break;
3374         case LLIT_INCONSISTENT_OWNER:
3375                 rc = lfsck_layout_repair_owner(env, com, parent, llr, pla);
3376                 break;
3377         default:
3378                 rc = 0;
3379                 break;
3380         }
3381
3382         GOTO(out, rc);
3383
3384 out:
3385         down_write(&com->lc_sem);
3386         if (rc < 0) {
3387                 struct lfsck_assistant_data *lad = com->lc_data;
3388
3389                 if (unlikely(lad->lad_exit)) {
3390                         rc = 0;
3391                 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
3392                            rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
3393                            rc == -EHOSTUNREACH) {
3394                         /* If cannot touch the target server,
3395                          * mark the LFSCK as INCOMPLETE. */
3396                         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
3397                                "talk with OST %x: rc = %d\n",
3398                                lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
3399                         lfsck_lad_set_bitmap(env, com, llr->llr_ost_idx);
3400                         lo->ll_objs_skipped++;
3401                         rc = 0;
3402                 } else {
3403                         lfsck_layout_record_failure(env, lfsck, lo);
3404                 }
3405         } else if (rc > 0) {
3406                 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
3407                          "unknown type = %d\n", type);
3408
3409                 lo->ll_objs_repaired[type - 1]++;
3410                 if (bk->lb_param & LPF_DRYRUN &&
3411                     unlikely(lo->ll_pos_first_inconsistent == 0))
3412                         lo->ll_pos_first_inconsistent =
3413                         lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
3414                                                         lfsck->li_di_oit);
3415         }
3416         up_write(&com->lc_sem);
3417
3418 put_parent:
3419         lfsck_object_put(env, parent);
3420
3421         return rc;
3422 }
3423
3424 static int lfsck_layout_assistant_handler_p2(const struct lu_env *env,
3425                                              struct lfsck_component *com)
3426 {
3427         struct lfsck_assistant_data     *lad    = com->lc_data;
3428         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3429         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
3430         struct lfsck_tgt_descs          *ltds   = &lfsck->li_ost_descs;
3431         struct lfsck_tgt_desc           *ltd;
3432         int                              rc     = 0;
3433         ENTRY;
3434
3435         CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan start\n",
3436                lfsck_lfsck2name(lfsck));
3437
3438         spin_lock(&ltds->ltd_lock);
3439         while (!list_empty(&lad->lad_ost_phase2_list)) {
3440                 ltd = list_entry(lad->lad_ost_phase2_list.next,
3441                                  struct lfsck_tgt_desc,
3442                                  ltd_layout_phase_list);
3443                 list_del_init(&ltd->ltd_layout_phase_list);
3444                 if (bk->lb_param & LPF_ALL_TGT) {
3445                         spin_unlock(&ltds->ltd_lock);
3446                         rc = lfsck_layout_scan_orphan(env, com, ltd);
3447                         if (rc != 0 && bk->lb_param & LPF_FAILOUT)
3448                                 RETURN(rc);
3449
3450                         if (unlikely(lad->lad_exit ||
3451                                      !thread_is_running(&lfsck->li_thread)))
3452                                 RETURN(0);
3453                         spin_lock(&ltds->ltd_lock);
3454                 }
3455         }
3456
3457         if (list_empty(&lad->lad_ost_phase1_list))
3458                 rc = 1;
3459         else
3460                 rc = 0;
3461         spin_unlock(&ltds->ltd_lock);
3462
3463         CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan stop: rc = %d\n",
3464                lfsck_lfsck2name(lfsck), rc);
3465
3466         RETURN(rc);
3467 }
3468
3469 static int
3470 lfsck_layout_slave_async_interpret(const struct lu_env *env,
3471                                    struct ptlrpc_request *req,
3472                                    void *args, int rc)
3473 {
3474         struct lfsck_layout_slave_async_args *llsaa = args;
3475         struct obd_export                    *exp   = llsaa->llsaa_exp;
3476         struct lfsck_component               *com   = llsaa->llsaa_com;
3477         struct lfsck_layout_slave_target     *llst  = llsaa->llsaa_llst;
3478         struct lfsck_layout_slave_data       *llsd  = com->lc_data;
3479         struct lfsck_reply                   *lr    = NULL;
3480         bool                                  done  = false;
3481
3482         if (rc != 0) {
3483                 /* It is quite probably caused by target crash,
3484                  * to make the LFSCK can go ahead, assume that
3485                  * the target finished the LFSCK prcoessing. */
3486                 done = true;
3487         } else {
3488                 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
3489                 if (lr->lr_status != LS_SCANNING_PHASE1 &&
3490                     lr->lr_status != LS_SCANNING_PHASE2)
3491                         done = true;
3492         }
3493
3494         if (done) {
3495                 CDEBUG(D_LFSCK, "%s: layout LFSCK slave gets the MDT %x "
3496                        "status %d\n", lfsck_lfsck2name(com->lc_lfsck),
3497                        llst->llst_index, lr != NULL ? lr->lr_status : rc);
3498
3499                 lfsck_layout_llst_del(llsd, llst);
3500         }
3501
3502         lfsck_layout_llst_put(llst);
3503         lfsck_component_put(env, com);
3504         class_export_put(exp);
3505
3506         return 0;
3507 }
3508
3509 static int lfsck_layout_async_query(const struct lu_env *env,
3510                                     struct lfsck_component *com,
3511                                     struct obd_export *exp,
3512                                     struct lfsck_layout_slave_target *llst,
3513                                     struct lfsck_request *lr,
3514                                     struct ptlrpc_request_set *set)
3515 {
3516         struct lfsck_layout_slave_async_args *llsaa;
3517         struct ptlrpc_request                *req;
3518         struct lfsck_request                 *tmp;
3519         int                                   rc;
3520         ENTRY;
3521
3522         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
3523         if (req == NULL)
3524                 RETURN(-ENOMEM);
3525
3526         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
3527         if (rc != 0) {
3528                 ptlrpc_request_free(req);
3529                 RETURN(rc);
3530         }
3531
3532         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3533         *tmp = *lr;
3534         ptlrpc_request_set_replen(req);
3535
3536         llsaa = ptlrpc_req_async_args(req);
3537         llsaa->llsaa_exp = exp;
3538         llsaa->llsaa_com = lfsck_component_get(com);
3539         llsaa->llsaa_llst = llst;
3540         req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
3541         ptlrpc_set_add_req(set, req);
3542
3543         RETURN(0);
3544 }
3545
3546 static int lfsck_layout_async_notify(const struct lu_env *env,
3547                                      struct obd_export *exp,
3548                                      struct lfsck_request *lr,
3549                                      struct ptlrpc_request_set *set)
3550 {
3551         struct ptlrpc_request   *req;
3552         struct lfsck_request    *tmp;
3553         int                      rc;
3554         ENTRY;
3555
3556         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
3557         if (req == NULL)
3558                 RETURN(-ENOMEM);
3559
3560         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
3561         if (rc != 0) {
3562                 ptlrpc_request_free(req);
3563                 RETURN(rc);
3564         }
3565
3566         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3567         *tmp = *lr;
3568         ptlrpc_request_set_replen(req);
3569         ptlrpc_set_add_req(set, req);
3570
3571         RETURN(0);
3572 }
3573
3574 static int
3575 lfsck_layout_slave_query_master(const struct lu_env *env,
3576                                 struct lfsck_component *com)
3577 {
3578         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
3579         struct lfsck_instance            *lfsck = com->lc_lfsck;
3580         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3581         struct lfsck_layout_slave_target *llst;
3582         struct obd_export                *exp;
3583         struct ptlrpc_request_set        *set;
3584         int                               rc    = 0;
3585         int                               rc1   = 0;
3586         ENTRY;
3587
3588         set = ptlrpc_prep_set();
3589         if (set == NULL)
3590                 GOTO(log, rc = -ENOMEM);
3591
3592         memset(lr, 0, sizeof(*lr));
3593         lr->lr_event = LE_QUERY;
3594         lr->lr_active = LFSCK_TYPE_LAYOUT;
3595
3596         llsd->llsd_touch_gen++;
3597         spin_lock(&llsd->llsd_lock);
3598         while (!list_empty(&llsd->llsd_master_list)) {
3599                 llst = list_entry(llsd->llsd_master_list.next,
3600                                   struct lfsck_layout_slave_target,
3601                                   llst_list);
3602                 if (llst->llst_gen == llsd->llsd_touch_gen)
3603                         break;
3604
3605                 llst->llst_gen = llsd->llsd_touch_gen;
3606                 list_move_tail(&llst->llst_list,
3607                                &llsd->llsd_master_list);
3608                 atomic_inc(&llst->llst_ref);
3609                 spin_unlock(&llsd->llsd_lock);
3610
3611                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
3612                                                llst->llst_index);
3613                 if (exp == NULL) {
3614                         lfsck_layout_llst_del(llsd, llst);
3615                         lfsck_layout_llst_put(llst);
3616                         spin_lock(&llsd->llsd_lock);
3617                         continue;
3618                 }
3619
3620                 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
3621                 if (rc != 0) {
3622                         CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
3623                                "query %s for layout: rc = %d\n",
3624                                lfsck_lfsck2name(lfsck),
3625                                exp->exp_obd->obd_name, rc);
3626
3627                         rc1 = rc;
3628                         lfsck_layout_llst_put(llst);
3629                         class_export_put(exp);
3630                 }
3631                 spin_lock(&llsd->llsd_lock);
3632         }
3633         spin_unlock(&llsd->llsd_lock);
3634
3635         rc = ptlrpc_set_wait(set);
3636         ptlrpc_set_destroy(set);
3637
3638         GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
3639
3640 log:
3641         CDEBUG(D_LFSCK, "%s: layout LFSCK slave queries master: rc = %d\n",
3642                lfsck_lfsck2name(com->lc_lfsck), rc);
3643
3644         return rc;
3645 }
3646
3647 static void
3648 lfsck_layout_slave_notify_master(const struct lu_env *env,
3649                                  struct lfsck_component *com,
3650                                  enum lfsck_events event, int result)
3651 {
3652         struct lfsck_layout              *lo    = com->lc_file_ram;
3653         struct lfsck_instance            *lfsck = com->lc_lfsck;
3654         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3655         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
3656         struct lfsck_layout_slave_target *llst;
3657         struct obd_export                *exp;
3658         struct ptlrpc_request_set        *set;
3659         int                               rc;
3660         ENTRY;
3661
3662         CDEBUG(D_LFSCK, "%s: layout LFSCK slave notifies master\n",
3663                lfsck_lfsck2name(com->lc_lfsck));
3664
3665         set = ptlrpc_prep_set();
3666         if (set == NULL)
3667                 RETURN_EXIT;
3668
3669         memset(lr, 0, sizeof(*lr));
3670         lr->lr_event = event;
3671         lr->lr_flags = LEF_FROM_OST;
3672         lr->lr_status = result;
3673         lr->lr_index = lfsck_dev_idx(lfsck);
3674         lr->lr_active = LFSCK_TYPE_LAYOUT;
3675         lr->lr_flags2 = lo->ll_flags;
3676         llsd->llsd_touch_gen++;
3677         spin_lock(&llsd->llsd_lock);
3678         while (!list_empty(&llsd->llsd_master_list)) {
3679                 llst = list_entry(llsd->llsd_master_list.next,
3680                                   struct lfsck_layout_slave_target,
3681                                   llst_list);
3682                 if (llst->llst_gen == llsd->llsd_touch_gen)
3683                         break;
3684
3685                 llst->llst_gen = llsd->llsd_touch_gen;
3686                 list_move_tail(&llst->llst_list,
3687                                &llsd->llsd_master_list);
3688                 atomic_inc(&llst->llst_ref);
3689                 spin_unlock(&llsd->llsd_lock);
3690
3691                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
3692                                                llst->llst_index);
3693                 if (exp == NULL) {
3694                         lfsck_layout_llst_del(llsd, llst);
3695                         lfsck_layout_llst_put(llst);
3696                         spin_lock(&llsd->llsd_lock);
3697                         continue;
3698                 }
3699
3700                 rc = lfsck_layout_async_notify(env, exp, lr, set);
3701                 if (rc != 0)
3702                         CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
3703                                "notify %s for layout: rc = %d\n",
3704                                lfsck_lfsck2name(lfsck),
3705                                exp->exp_obd->obd_name, rc);
3706
3707                 lfsck_layout_llst_put(llst);
3708                 class_export_put(exp);
3709                 spin_lock(&llsd->llsd_lock);
3710         }
3711         spin_unlock(&llsd->llsd_lock);
3712
3713         ptlrpc_set_wait(set);
3714         ptlrpc_set_destroy(set);
3715
3716         RETURN_EXIT;
3717 }
3718
3719 /*
3720  * \ret -ENODATA: unrecognized stripe
3721  * \ret = 0     : recognized stripe
3722  * \ret < 0     : other failures
3723  */
3724 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
3725                                            struct lfsck_component *com,
3726                                            struct lu_fid *cfid,
3727                                            struct lu_fid *pfid)
3728 {
3729         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3730         struct lu_buf                   *buf    = &info->lti_big_buf;
3731         struct ost_id                   *oi     = &info->lti_oi;
3732         struct dt_object                *obj;
3733         struct lov_mds_md_v1            *lmm;
3734         struct lov_ost_data_v1          *objs;
3735         __u32                            idx    = pfid->f_stripe_idx;
3736         __u32                            magic;
3737         int                              rc     = 0;
3738         int                              i;
3739         __u16                            count;
3740         ENTRY;
3741
3742         pfid->f_ver = 0;
3743         obj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3744         if (IS_ERR(obj))
3745                 RETURN(PTR_ERR(obj));
3746
3747         dt_read_lock(env, obj, 0);
3748         if (unlikely(dt_object_exists(obj) == 0 ||
3749                      lfsck_is_dead_obj(obj)))
3750                 GOTO(unlock, rc = -ENOENT);
3751
3752         if (!S_ISREG(lfsck_object_type(obj)))
3753                 GOTO(unlock, rc = -ENODATA);
3754
3755         rc = lfsck_layout_get_lovea(env, obj, buf);
3756         if (rc < 0)
3757                 GOTO(unlock, rc);
3758
3759         if (rc == 0)
3760                 GOTO(unlock, rc = -ENODATA);
3761
3762         lmm = buf->lb_buf;
3763         rc = lfsck_layout_verify_header(lmm);
3764         if (rc != 0)
3765                 GOTO(unlock, rc);
3766
3767         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
3768          * been verified in lfsck_layout_verify_header() already. If some
3769          * new magic introduced in the future, then layout LFSCK needs to
3770          * be updated also. */
3771         magic = le32_to_cpu(lmm->lmm_magic);
3772         if (magic == LOV_MAGIC_V1) {
3773                 objs = &lmm->lmm_objects[0];
3774         } else {
3775                 LASSERT(magic == LOV_MAGIC_V3);
3776                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3777         }
3778
3779         fid_to_ostid(cfid, oi);
3780         count = le16_to_cpu(lmm->lmm_stripe_count);
3781         for (i = 0; i < count; i++, objs++) {
3782                 struct ost_id oi2;
3783
3784                 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
3785                 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
3786                         GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
3787         }
3788
3789         GOTO(unlock, rc = -ENODATA);
3790
3791 unlock:
3792         dt_read_unlock(env, obj);
3793         lfsck_object_put(env, obj);
3794
3795         return rc;
3796 }
3797
3798 /*
3799  * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
3800  * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
3801  * layout EA from MDT to OST. On one hand, the OST no need to understand
3802  * the layout EA structure; on the other hand, it may cause trouble when
3803  * transfer large layout EA from MDT to OST via normal OUT RPC.
3804  *
3805  * \ret > 0: unrecognized stripe
3806  * \ret = 0: recognized stripe
3807  * \ret < 0: other failures
3808  */
3809 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
3810                                           struct lfsck_component *com,
3811                                           struct lu_fid *cfid,
3812                                           struct lu_fid *pfid)
3813 {
3814         struct lfsck_instance    *lfsck  = com->lc_lfsck;
3815         struct obd_device        *obd    = lfsck->li_obd;
3816         struct seq_server_site   *ss     = lfsck_dev_site(lfsck);
3817         struct obd_export        *exp    = NULL;
3818         struct ptlrpc_request    *req    = NULL;
3819         struct lfsck_request     *lr;
3820         struct lu_seq_range      *range  = &lfsck_env_info(env)->lti_range;
3821         int                       rc     = 0;
3822         ENTRY;
3823
3824         if (unlikely(fid_is_idif(pfid)))
3825                 RETURN(1);
3826
3827         fld_range_set_any(range);
3828         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), range);
3829         if (rc != 0)
3830                 RETURN(rc == -ENOENT ? 1 : rc);
3831
3832         if (unlikely(!fld_range_is_mdt(range)))
3833                 RETURN(1);
3834
3835         exp = lustre_find_lwp_by_index(obd->obd_name, range->lsr_index);
3836         if (unlikely(exp == NULL))
3837                 RETURN(1);
3838
3839         if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
3840                 GOTO(out, rc = -EOPNOTSUPP);
3841
3842         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
3843         if (req == NULL)
3844                 GOTO(out, rc = -ENOMEM);
3845
3846         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
3847         if (rc != 0) {
3848                 ptlrpc_request_free(req);
3849
3850                 GOTO(out, rc);
3851         }
3852
3853         lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3854         memset(lr, 0, sizeof(*lr));
3855         lr->lr_event = LE_PAIRS_VERIFY;
3856         lr->lr_active = LFSCK_TYPE_LAYOUT;
3857         lr->lr_fid = *cfid; /* OST-object itself FID. */
3858         lr->lr_fid2 = *pfid; /* The claimed parent FID. */
3859
3860         ptlrpc_request_set_replen(req);
3861         rc = ptlrpc_queue_wait(req);
3862         ptlrpc_req_finished(req);
3863
3864         if (rc == -ENOENT || rc == -ENODATA)
3865                 rc = 1;
3866
3867         GOTO(out, rc);
3868
3869 out:
3870         if (exp != NULL)
3871                 class_export_put(exp);
3872
3873         return rc;
3874 }
3875
3876 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
3877                                           struct lfsck_component *com,
3878                                           struct lfsck_request *lr)
3879 {
3880         struct dt_object        *obj;
3881         int                      rc     = 0;
3882         ENTRY;
3883
3884         obj = lfsck_object_find_bottom(env, com->lc_lfsck, &lr->lr_fid);
3885         if (IS_ERR(obj))
3886                 GOTO(log, rc = PTR_ERR(obj));
3887
3888         dt_write_lock(env, obj, 0);
3889         if (unlikely(dt_object_exists(obj) == 0 ||
3890                      lfsck_is_dead_obj(obj)))
3891                 GOTO(unlock, rc = 0);
3892
3893         rc = __lfsck_layout_update_pfid(env, obj, &lr->lr_fid2,
3894                                         lr->lr_fid2.f_ver);
3895
3896         GOTO(unlock, rc);
3897
3898 unlock:
3899         dt_write_unlock(env, obj);
3900         lfsck_object_put(env, obj);
3901
3902 log:
3903         CDEBUG(D_LFSCK, "%s: layout LFSCK slave repaired pfid for "DFID
3904                ", parent "DFID": rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
3905                PFID(&lr->lr_fid), PFID(&lr->lr_fid2), rc);
3906
3907         return rc;
3908 }
3909
3910 /* layout APIs */
3911
3912 static void lfsck_layout_slave_quit(const struct lu_env *env,
3913                                     struct lfsck_component *com);
3914
3915 static int lfsck_layout_reset(const struct lu_env *env,
3916                               struct lfsck_component *com, bool init)
3917 {
3918         struct lfsck_layout     *lo    = com->lc_file_ram;
3919         int                      rc;
3920
3921         down_write(&com->lc_sem);
3922         if (init) {
3923                 memset(lo, 0, com->lc_file_size);
3924         } else {
3925                 __u32 count = lo->ll_success_count;
3926                 __u64 last_time = lo->ll_time_last_complete;
3927
3928                 memset(lo, 0, com->lc_file_size);
3929                 lo->ll_success_count = count;
3930                 lo->ll_time_last_complete = last_time;
3931         }
3932
3933         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
3934         lo->ll_status = LS_INIT;
3935
3936         if (com->lc_lfsck->li_master) {
3937                 struct lfsck_assistant_data *lad = com->lc_data;
3938
3939                 lad->lad_incomplete = 0;
3940                 CFS_RESET_BITMAP(lad->lad_bitmap);
3941         }
3942
3943         rc = lfsck_layout_store(env, com);
3944         up_write(&com->lc_sem);
3945
3946         CDEBUG(D_LFSCK, "%s: layout LFSCK reset: rc = %d\n",
3947                lfsck_lfsck2name(com->lc_lfsck), rc);
3948
3949         return rc;
3950 }
3951
3952 static void lfsck_layout_fail(const struct lu_env *env,
3953                               struct lfsck_component *com, bool new_checked)
3954 {
3955         struct lfsck_layout *lo = com->lc_file_ram;
3956
3957         down_write(&com->lc_sem);
3958         if (new_checked)
3959                 com->lc_new_checked++;
3960         lfsck_layout_record_failure(env, com->lc_lfsck, lo);
3961         up_write(&com->lc_sem);
3962 }
3963
3964 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
3965                                           struct lfsck_component *com, bool init)
3966 {
3967         struct lfsck_instance   *lfsck   = com->lc_lfsck;
3968         struct lfsck_layout     *lo      = com->lc_file_ram;
3969         int                      rc;
3970
3971         if (!init) {
3972                 rc = lfsck_checkpoint_generic(env, com);
3973                 if (rc != 0)
3974                         return rc > 0 ? 0 : rc;
3975         }
3976
3977         down_write(&com->lc_sem);
3978         if (init) {
3979                 lo->ll_pos_latest_start =
3980                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
3981         } else {
3982                 lo->ll_pos_last_checkpoint =
3983                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
3984                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3985                                 HALF_SEC - lfsck->li_time_last_checkpoint);
3986                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
3987                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
3988                 com->lc_new_checked = 0;
3989         }
3990
3991         rc = lfsck_layout_store(env, com);
3992         up_write(&com->lc_sem);
3993
3994         CDEBUG(D_LFSCK, "%s: layout LFSCK master checkpoint at the pos ["
3995                LPU64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
3996                lfsck->li_pos_current.lp_oit_cookie, rc);
3997
3998         return rc;
3999 }
4000
4001 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
4002                                          struct lfsck_component *com, bool init)
4003 {
4004         struct lfsck_instance   *lfsck = com->lc_lfsck;
4005         struct lfsck_layout     *lo    = com->lc_file_ram;
4006         int                      rc;
4007
4008         if (com->lc_new_checked == 0 && !init)
4009                 return 0;
4010
4011         down_write(&com->lc_sem);
4012         if (init) {
4013                 lo->ll_pos_latest_start =
4014                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
4015         } else {
4016                 lo->ll_pos_last_checkpoint =
4017                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
4018                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4019                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4020                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4021                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4022                 com->lc_new_checked = 0;
4023         }
4024
4025         rc = lfsck_layout_store(env, com);
4026         up_write(&com->lc_sem);
4027
4028         CDEBUG(D_LFSCK, "%s: layout LFSCK slave checkpoint at the pos ["
4029                LPU64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
4030                lfsck->li_pos_current.lp_oit_cookie, rc);
4031
4032         return rc;
4033 }
4034
4035 static int lfsck_layout_prep(const struct lu_env *env,
4036                              struct lfsck_component *com,
4037                              struct lfsck_start *start)
4038 {
4039         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4040         struct lfsck_layout     *lo     = com->lc_file_ram;
4041         struct lfsck_position   *pos    = &com->lc_pos_start;
4042
4043         fid_zero(&pos->lp_dir_parent);
4044         pos->lp_dir_cookie = 0;
4045         if (lo->ll_status == LS_COMPLETED ||
4046             lo->ll_status == LS_PARTIAL ||
4047             /* To handle orphan, must scan from the beginning. */
4048             (start != NULL && start->ls_flags & LPF_OST_ORPHAN)) {
4049                 int rc;
4050
4051                 rc = lfsck_layout_reset(env, com, false);
4052                 if (rc == 0)
4053                         rc = lfsck_set_param(env, lfsck, start, true);
4054
4055                 if (rc != 0) {
4056                         CDEBUG(D_LFSCK, "%s: layout LFSCK prep failed: "
4057                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4058
4059                         return rc;
4060                 }
4061         }
4062
4063         down_write(&com->lc_sem);
4064         lo->ll_time_latest_start = cfs_time_current_sec();
4065         spin_lock(&lfsck->li_lock);
4066         if (lo->ll_flags & LF_SCANNED_ONCE) {
4067                 if (!lfsck->li_drop_dryrun ||
4068                     lo->ll_pos_first_inconsistent == 0) {
4069                         lo->ll_status = LS_SCANNING_PHASE2;
4070                         list_move_tail(&com->lc_link,
4071                                        &lfsck->li_list_double_scan);
4072                         pos->lp_oit_cookie = 0;
4073                 } else {
4074                         int i;
4075
4076                         lo->ll_status = LS_SCANNING_PHASE1;
4077                         lo->ll_run_time_phase1 = 0;
4078                         lo->ll_run_time_phase2 = 0;
4079                         lo->ll_objs_checked_phase1 = 0;
4080                         lo->ll_objs_checked_phase2 = 0;
4081                         lo->ll_objs_failed_phase1 = 0;
4082                         lo->ll_objs_failed_phase2 = 0;
4083                         for (i = 0; i < LLIT_MAX; i++)
4084                                 lo->ll_objs_repaired[i] = 0;
4085
4086                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4087                         fid_zero(&com->lc_fid_latest_scanned_phase2);
4088                 }
4089         } else {
4090                 lo->ll_status = LS_SCANNING_PHASE1;
4091                 if (!lfsck->li_drop_dryrun ||
4092                     lo->ll_pos_first_inconsistent == 0)
4093                         pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
4094                 else
4095                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4096         }
4097         spin_unlock(&lfsck->li_lock);
4098         up_write(&com->lc_sem);
4099
4100         return 0;
4101 }
4102
4103 static int lfsck_layout_slave_prep(const struct lu_env *env,
4104                                    struct lfsck_component *com,
4105                                    struct lfsck_start_param *lsp)
4106 {
4107         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4108         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4109         struct lfsck_layout             *lo     = com->lc_file_ram;
4110         struct lfsck_start              *start  = lsp->lsp_start;
4111         int                              rc;
4112
4113         rc = lfsck_layout_prep(env, com, start);
4114         if (rc != 0)
4115                 return rc;
4116
4117         if (lo->ll_flags & LF_CRASHED_LASTID &&
4118             list_empty(&llsd->llsd_master_list)) {
4119                 LASSERT(lfsck->li_out_notify != NULL);
4120
4121                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4122                                      LE_LASTID_REBUILDING);
4123         }
4124
4125         if (!lsp->lsp_index_valid)
4126                 return 0;
4127
4128         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
4129         if (rc == 0 && start != NULL && start->ls_flags & LPF_OST_ORPHAN) {
4130                 LASSERT(!llsd->llsd_rbtree_valid);
4131
4132                 write_lock(&llsd->llsd_rb_lock);
4133                 rc = lfsck_rbtree_setup(env, com);
4134                 write_unlock(&llsd->llsd_rb_lock);
4135         }
4136
4137         CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
4138                LPU64"]\n", lfsck_lfsck2name(lfsck),
4139                com->lc_pos_start.lp_oit_cookie);
4140
4141         return rc;
4142 }
4143
4144 static int lfsck_layout_master_prep(const struct lu_env *env,
4145                                     struct lfsck_component *com,
4146                                     struct lfsck_start_param *lsp)
4147 {
4148         int rc;
4149         ENTRY;
4150
4151         rc = lfsck_layout_load_bitmap(env, com);
4152         if (rc != 0) {
4153                 rc = lfsck_layout_reset(env, com, false);
4154                 if (rc == 0)
4155                         rc = lfsck_set_param(env, com->lc_lfsck,
4156                                              lsp->lsp_start, true);
4157
4158                 if (rc != 0)
4159                         GOTO(log, rc);
4160         }
4161
4162         rc = lfsck_layout_prep(env, com, lsp->lsp_start);
4163         if (rc != 0)
4164                 RETURN(rc);
4165
4166         rc = lfsck_start_assistant(env, com, lsp);
4167
4168         GOTO(log, rc);
4169
4170 log:
4171         CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
4172                LPU64"]\n", lfsck_lfsck2name(com->lc_lfsck),
4173                com->lc_pos_start.lp_oit_cookie);
4174
4175         return 0;
4176 }
4177
4178 /* Pre-fetch the attribute for each stripe in the given layout EA. */
4179 static int lfsck_layout_scan_stripes(const struct lu_env *env,
4180                                      struct lfsck_component *com,
4181                                      struct dt_object *parent,
4182                                      struct lov_mds_md_v1 *lmm)
4183 {
4184         struct lfsck_thread_info        *info    = lfsck_env_info(env);
4185         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4186         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
4187         struct lfsck_layout             *lo      = com->lc_file_ram;
4188         struct lfsck_assistant_data     *lad     = com->lc_data;
4189         struct lfsck_layout_object      *llo     = NULL;
4190         struct lov_ost_data_v1          *objs;
4191         struct lfsck_tgt_descs          *ltds    = &lfsck->li_ost_descs;
4192         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4193         struct ptlrpc_thread            *athread = &lad->lad_thread;
4194         struct l_wait_info               lwi     = { 0 };
4195         struct lu_buf                    buf;
4196         int                              rc      = 0;
4197         int                              i;
4198         __u32                            magic;
4199         __u16                            count;
4200         ENTRY;
4201
4202         lfsck_buf_init(&buf, &info->lti_old_pfid,
4203                        sizeof(struct filter_fid_old));
4204         count = le16_to_cpu(lmm->lmm_stripe_count);
4205         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
4206          * been verified in lfsck_layout_verify_header() already. If some
4207          * new magic introduced in the future, then layout LFSCK needs to
4208          * be updated also. */
4209         magic = le32_to_cpu(lmm->lmm_magic);
4210         if (magic == LOV_MAGIC_V1) {
4211                 objs = &lmm->lmm_objects[0];
4212         } else {
4213                 LASSERT(magic == LOV_MAGIC_V3);
4214                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4215         }
4216
4217         for (i = 0; i < count; i++, objs++) {
4218                 struct lu_fid           *fid    = &info->lti_fid;
4219                 struct ost_id           *oi     = &info->lti_oi;
4220                 struct lfsck_layout_req *llr;
4221                 struct lfsck_tgt_desc   *tgt    = NULL;
4222                 struct dt_object        *cobj   = NULL;
4223                 __u32                    index;
4224                 bool                     wakeup = false;
4225
4226                 if (unlikely(lovea_slot_is_dummy(objs)))
4227                         continue;
4228
4229                 l_wait_event(mthread->t_ctl_waitq,
4230                              lad->lad_prefetched < bk->lb_async_windows ||
4231                              !thread_is_running(mthread) ||
4232                              thread_is_stopped(athread),
4233                              &lwi);
4234
4235                 if (unlikely(!thread_is_running(mthread)) ||
4236                              thread_is_stopped(athread))
4237                         GOTO(out, rc = 0);
4238
4239                 if (unlikely(lfsck_is_dead_obj(parent)))
4240                         GOTO(out, rc = 0);
4241
4242                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4243                 index = le32_to_cpu(objs->l_ost_idx);
4244                 rc = ostid_to_fid(fid, oi, index);
4245                 if (rc != 0) {
4246                         CDEBUG(D_LFSCK, "%s: get invalid layout EA for "DFID
4247                                ": "DOSTID", idx:%u\n", lfsck_lfsck2name(lfsck),
4248                                PFID(lfsck_dto2fid(parent)), POSTID(oi), index);
4249                         goto next;
4250                 }
4251
4252                 tgt = lfsck_tgt_get(ltds, index);
4253                 if (unlikely(tgt == NULL)) {
4254                         CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
4255                                "did not join the layout LFSCK\n",
4256                                lfsck_lfsck2name(lfsck), index);
4257                         lfsck_lad_set_bitmap(env, com, index);
4258                         goto next;
4259                 }
4260
4261                 /* There is potential deadlock race condition between object
4262                  * destroy and layout LFSCK. Consider the following scenario:
4263                  *
4264                  * 1) The LFSCK thread obtained the parent object firstly, at
4265                  *    that time, the parent object has not been destroyed yet.
4266                  *
4267                  * 2) One RPC service thread destroyed the parent and all its
4268                  *    children objects. Because the LFSCK is referencing the
4269                  *    parent object, then the parent object will be marked as
4270                  *    dying in RAM. On the other hand, the parent object is
4271                  *    referencing all its children objects, then all children
4272                  *    objects will be marked as dying in RAM also.
4273                  *
4274                  * 3) The LFSCK thread tries to find some child object with
4275                  *    the parent object referenced. Then it will find that the
4276                  *    child object is dying. According to the object visibility
4277                  *    rules: the object with dying flag cannot be returned to
4278                  *    others. So the LFSCK thread has to wait until the dying
4279                  *    object has been purged from RAM, then it can allocate a
4280                  *    new object (with the same FID) in RAM. Unfortunately, the
4281                  *    LFSCK thread itself is referencing the parent object, and
4282                  *    cause the parent object cannot be purged, then cause the
4283                  *    child object cannot be purged also. So the LFSCK thread
4284                  *    will fall into deadlock.
4285                  *
4286                  * We introduce non-blocked version lu_object_find() to allow
4287                  * the LFSCK thread to return failure immediately (instead of
4288                  * wait) when it finds dying (child) object, then the LFSCK
4289                  * thread can check whether the parent object is dying or not.
4290                  * So avoid above deadlock. LU-5395 */
4291                 cobj = lfsck_object_find_by_dev_nowait(env, tgt->ltd_tgt, fid);
4292                 if (IS_ERR(cobj)) {
4293                         if (lfsck_is_dead_obj(parent)) {
4294                                 lfsck_tgt_put(tgt);
4295
4296                                 GOTO(out, rc = 0);
4297                         }
4298
4299                         rc = PTR_ERR(cobj);
4300                         goto next;
4301                 }
4302
4303                 rc = dt_declare_attr_get(env, cobj, BYPASS_CAPA);
4304                 if (rc != 0)
4305                         goto next;
4306
4307                 rc = dt_declare_xattr_get(env, cobj, &buf, XATTR_NAME_FID,
4308                                           BYPASS_CAPA);
4309                 if (rc != 0)
4310                         goto next;
4311
4312                 if (llo == NULL) {
4313                         llo = lfsck_layout_object_init(env, parent,
4314                                 lfsck->li_pos_current.lp_oit_cookie);
4315                         if (IS_ERR(llo)) {
4316                                 rc = PTR_ERR(llo);
4317                                 goto next;
4318                         }
4319                 }
4320
4321                 llr = lfsck_layout_assistant_req_init(llo,
4322                                                       lfsck_dto2fid(parent),
4323                                                       cobj, index, i);
4324                 if (IS_ERR(llr)) {
4325                         rc = PTR_ERR(llr);
4326                         goto next;
4327                 }
4328
4329                 cobj = NULL;
4330                 spin_lock(&lad->lad_lock);
4331                 if (lad->lad_assistant_status < 0) {
4332                         spin_unlock(&lad->lad_lock);
4333                         lfsck_layout_assistant_req_fini(env, &llr->llr_lar);
4334                         lfsck_tgt_put(tgt);
4335                         RETURN(lad->lad_assistant_status);
4336                 }
4337
4338                 list_add_tail(&llr->llr_lar.lar_list, &lad->lad_req_list);
4339                 if (lad->lad_prefetched == 0)
4340                         wakeup = true;
4341
4342                 lad->lad_prefetched++;
4343                 spin_unlock(&lad->lad_lock);
4344                 if (wakeup)
4345                         wake_up_all(&athread->t_ctl_waitq);
4346
4347 next:
4348                 down_write(&com->lc_sem);
4349                 com->lc_new_checked++;
4350                 if (rc < 0)
4351                         lfsck_layout_record_failure(env, lfsck, lo);
4352                 up_write(&com->lc_sem);
4353
4354                 if (cobj != NULL && !IS_ERR(cobj))
4355                         lfsck_object_put(env, cobj);
4356
4357                 if (likely(tgt != NULL))
4358                         lfsck_tgt_put(tgt);
4359
4360                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4361                         GOTO(out, rc);
4362         }
4363
4364         GOTO(out, rc = 0);
4365
4366 out:
4367         if (llo != NULL && !IS_ERR(llo))
4368                 lfsck_layout_object_put(env, llo);
4369
4370         return rc;
4371 }
4372
4373 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
4374  * the OST-object's attribute and generate an structure lfsck_layout_req on the
4375  * list ::lad_req_list.
4376  *
4377  * For each request on above list, the lfsck_layout_assistant thread compares
4378  * the OST side attribute with local attribute, if inconsistent, then repair it.
4379  *
4380  * All above processing is async mode with pipeline. */
4381 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
4382                                         struct lfsck_component *com,
4383                                         struct dt_object *obj)
4384 {
4385         struct lfsck_thread_info        *info   = lfsck_env_info(env);
4386         struct ost_id                   *oi     = &info->lti_oi;
4387         struct lfsck_layout             *lo     = com->lc_file_ram;
4388         struct lfsck_assistant_data     *lad    = com->lc_data;
4389         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4390         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
4391         struct thandle                  *handle = NULL;
4392         struct lu_buf                   *buf    = &info->lti_big_buf;
4393         struct lov_mds_md_v1            *lmm    = NULL;
4394         struct dt_device                *dev    = lfsck_obj2dev(obj);
4395         struct lustre_handle             lh     = { 0 };
4396         struct lu_buf                    ea_buf = { NULL };
4397         int                              rc     = 0;
4398         int                              size   = 0;
4399         bool                             locked = false;
4400         bool                             stripe = false;
4401         bool                             bad_oi = false;
4402         ENTRY;
4403
4404         if (!S_ISREG(lfsck_object_type(obj)))
4405                 GOTO(out, rc = 0);
4406
4407         if (lad->lad_assistant_status < 0)
4408                 GOTO(out, rc = -ESRCH);
4409
4410         fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
4411         lmm_oi_cpu_to_le(oi, oi);
4412         dt_read_lock(env, obj, 0);
4413         locked = true;
4414
4415 again:
4416         if (dt_object_exists(obj) == 0 ||
4417             lfsck_is_dead_obj(obj))
4418                 GOTO(out, rc = 0);
4419
4420         rc = lfsck_layout_get_lovea(env, obj, buf);
4421         if (rc <= 0)
4422                 GOTO(out, rc);
4423
4424         size = rc;
4425         lmm = buf->lb_buf;
4426         rc = lfsck_layout_verify_header(lmm);
4427         /* If the LOV EA crashed, then it is possible to be rebuilt later
4428          * when handle orphan OST-objects. */
4429         if (rc != 0)
4430                 GOTO(out, rc);
4431
4432         if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0)
4433                 GOTO(out, stripe = true);
4434
4435         /* Inconsistent lmm_oi, should be repaired. */
4436         bad_oi = true;
4437         lmm->lmm_oi = *oi;
4438
4439         if (bk->lb_param & LPF_DRYRUN) {
4440                 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4441
4442                 GOTO(out, stripe = true);
4443         }
4444
4445         if (!lustre_handle_is_used(&lh)) {
4446                 dt_read_unlock(env, obj);
4447                 locked = false;
4448                 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4449                                       MDS_INODELOCK_LAYOUT |
4450                                       MDS_INODELOCK_XATTR, LCK_EX);
4451                 if (rc != 0)
4452                         GOTO(out, rc);
4453
4454                 handle = dt_trans_create(env, dev);
4455                 if (IS_ERR(handle))
4456                         GOTO(out, rc = PTR_ERR(handle));
4457
4458                 lfsck_buf_init(&ea_buf, lmm, size);
4459                 rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
4460                                           LU_XATTR_REPLACE, handle);
4461                 if (rc != 0)
4462                         GOTO(out, rc);
4463
4464                 rc = dt_trans_start_local(env, dev, handle);
4465                 if (rc != 0)
4466                         GOTO(out, rc);
4467
4468                 dt_write_lock(env, obj, 0);
4469                 locked = true;
4470
4471                 goto again;
4472         }
4473
4474         rc = dt_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
4475                           LU_XATTR_REPLACE, handle, BYPASS_CAPA);
4476         if (rc != 0)
4477                 GOTO(out, rc);
4478
4479         lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4480
4481         GOTO(out, stripe = true);
4482
4483 out:
4484         if (locked) {
4485                 if (lustre_handle_is_used(&lh))
4486                         dt_write_unlock(env, obj);
4487                 else
4488                         dt_read_unlock(env, obj);
4489         }
4490
4491         if (handle != NULL && !IS_ERR(handle))
4492                 dt_trans_stop(env, dev, handle);
4493
4494         lfsck_ibits_unlock(&lh, LCK_EX);
4495
4496         if (bad_oi)
4497                 CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
4498                        DFID": rc = %d\n", lfsck_lfsck2name(lfsck),
4499                        bk->lb_param & LPF_DRYRUN ? "found" : "repaired",
4500                        PFID(lfsck_dto2fid(obj)), rc);
4501
4502         if (stripe) {
4503                 rc = lfsck_layout_scan_stripes(env, com, obj, lmm);
4504         } else {
4505                 down_write(&com->lc_sem);
4506                 com->lc_new_checked++;
4507                 if (rc < 0)
4508                         lfsck_layout_record_failure(env, lfsck, lo);
4509                 up_write(&com->lc_sem);
4510         }
4511
4512         return rc;
4513 }
4514
4515 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
4516                                        struct lfsck_component *com,
4517                                        struct dt_object *obj)
4518 {
4519         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4520         struct lfsck_layout             *lo     = com->lc_file_ram;
4521         const struct lu_fid             *fid    = lfsck_dto2fid(obj);
4522         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4523         struct lfsck_layout_seq         *lls;
4524         __u64                            seq;
4525         __u64                            oid;
4526         int                              rc;
4527         ENTRY;
4528
4529         LASSERT(llsd != NULL);
4530
4531         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
4532             cfs_fail_val == lfsck_dev_idx(lfsck)) {
4533                 struct l_wait_info       lwi = LWI_TIMEOUT(cfs_time_seconds(1),
4534                                                            NULL, NULL);
4535                 struct ptlrpc_thread    *thread = &lfsck->li_thread;
4536
4537                 l_wait_event(thread->t_ctl_waitq,
4538                              !thread_is_running(thread),
4539                              &lwi);
4540         }
4541
4542         lfsck_rbtree_update_bitmap(env, com, fid, false);
4543
4544         down_write(&com->lc_sem);
4545         if (fid_is_idif(fid))
4546                 seq = 0;
4547         else if (!fid_is_norm(fid) ||
4548                  !fid_is_for_ostobj(env, lfsck, obj, fid))
4549                 GOTO(unlock, rc = 0);
4550         else
4551                 seq = fid_seq(fid);
4552         com->lc_new_checked++;
4553
4554         lls = lfsck_layout_seq_lookup(llsd, seq);
4555         if (lls == NULL) {
4556                 OBD_ALLOC_PTR(lls);
4557                 if (unlikely(lls == NULL))
4558                         GOTO(unlock, rc = -ENOMEM);
4559
4560                 INIT_LIST_HEAD(&lls->lls_list);
4561                 lls->lls_seq = seq;
4562                 rc = lfsck_layout_lastid_load(env, com, lls);
4563                 if (rc != 0) {
4564                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
4565                               "load LAST_ID for "LPX64": rc = %d\n",
4566                               lfsck_lfsck2name(com->lc_lfsck), seq, rc);
4567                         lo->ll_objs_failed_phase1++;
4568                         OBD_FREE_PTR(lls);
4569                         GOTO(unlock, rc);
4570                 }
4571
4572                 lfsck_layout_seq_insert(llsd, lls);
4573         }
4574
4575         if (unlikely(fid_is_last_id(fid)))
4576                 GOTO(unlock, rc = 0);
4577
4578         if (fid_is_idif(fid))
4579                 oid = fid_idif_id(fid_seq(fid), fid_oid(fid), fid_ver(fid));
4580         else
4581                 oid = fid_oid(fid);
4582
4583         if (oid > lls->lls_lastid_known)
4584                 lls->lls_lastid_known = oid;
4585
4586         if (oid > lls->lls_lastid) {
4587                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
4588                         /* OFD may create new objects during LFSCK scanning. */
4589                         rc = lfsck_layout_lastid_reload(env, com, lls);
4590                         if (unlikely(rc != 0)) {
4591                                 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
4592                                       "reload LAST_ID for "LPX64": rc = %d\n",
4593                                       lfsck_lfsck2name(com->lc_lfsck),
4594                                       lls->lls_seq, rc);
4595
4596                                 GOTO(unlock, rc);
4597                         }
4598
4599                         if (oid <= lls->lls_lastid ||
4600                             lo->ll_flags & LF_CRASHED_LASTID)
4601                                 GOTO(unlock, rc = 0);
4602
4603                         LASSERT(lfsck->li_out_notify != NULL);
4604
4605                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4606                                              LE_LASTID_REBUILDING);
4607                         lo->ll_flags |= LF_CRASHED_LASTID;
4608
4609                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
4610                                "LAST_ID file (2) for the sequence "LPX64
4611                                ", old value "LPU64", known value "LPU64"\n",
4612                                lfsck_lfsck2name(lfsck), lls->lls_seq,
4613                                lls->lls_lastid, oid);
4614                 }
4615
4616                 lls->lls_lastid = oid;
4617                 lls->lls_dirty = 1;
4618         }
4619
4620         GOTO(unlock, rc = 0);
4621
4622 unlock:
4623         up_write(&com->lc_sem);
4624
4625         return rc;
4626 }
4627
4628 static int lfsck_layout_exec_dir(const struct lu_env *env,
4629                                  struct lfsck_component *com,
4630                                  struct lu_dirent *ent, __u16 type)
4631 {
4632         return 0;
4633 }
4634
4635 static int lfsck_layout_master_post(const struct lu_env *env,
4636                                     struct lfsck_component *com,
4637                                     int result, bool init)
4638 {
4639         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4640         struct lfsck_layout     *lo     = com->lc_file_ram;
4641         int                      rc;
4642         ENTRY;
4643
4644         lfsck_post_generic(env, com, &result);
4645
4646         down_write(&com->lc_sem);
4647         spin_lock(&lfsck->li_lock);
4648         if (!init)
4649                 lo->ll_pos_last_checkpoint =
4650                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
4651
4652         if (result > 0) {
4653                 if (lo->ll_flags & LF_INCOMPLETE)
4654                         lo->ll_status = LS_PARTIAL;
4655                 else
4656                         lo->ll_status = LS_SCANNING_PHASE2;
4657                 lo->ll_flags |= LF_SCANNED_ONCE;
4658                 lo->ll_flags &= ~LF_UPGRADE;
4659                 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4660         } else if (result == 0) {
4661                 if (lfsck->li_status != 0)
4662                         lo->ll_status = lfsck->li_status;
4663                 else
4664                         lo->ll_status = LS_STOPPED;
4665                 if (lo->ll_status != LS_PAUSED)
4666                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4667         } else {
4668                 lo->ll_status = LS_FAILED;
4669                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4670         }
4671         spin_unlock(&lfsck->li_lock);
4672
4673         if (!init) {
4674                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4675                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4676                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4677                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4678                 com->lc_new_checked = 0;
4679         }
4680
4681         rc = lfsck_layout_store(env, com);
4682         up_write(&com->lc_sem);
4683
4684         CDEBUG(D_LFSCK, "%s: layout LFSCK master post done: rc = %d\n",
4685                lfsck_lfsck2name(lfsck), rc);
4686
4687         RETURN(rc);
4688 }
4689
4690 static int lfsck_layout_slave_post(const struct lu_env *env,
4691                                    struct lfsck_component *com,
4692                                    int result, bool init)
4693 {
4694         struct lfsck_instance   *lfsck = com->lc_lfsck;
4695         struct lfsck_layout     *lo    = com->lc_file_ram;
4696         int                      rc;
4697         bool                     done  = false;
4698
4699         rc = lfsck_layout_lastid_store(env, com);
4700         if (rc != 0)
4701                 result = rc;
4702
4703         LASSERT(lfsck->li_out_notify != NULL);
4704
4705         down_write(&com->lc_sem);
4706         spin_lock(&lfsck->li_lock);
4707         if (!init)
4708                 lo->ll_pos_last_checkpoint =
4709                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
4710
4711         if (result > 0) {
4712                 lo->ll_status = LS_SCANNING_PHASE2;
4713                 lo->ll_flags |= LF_SCANNED_ONCE;
4714                 if (lo->ll_flags & LF_CRASHED_LASTID) {
4715                         done = true;
4716                         lo->ll_flags &= ~LF_CRASHED_LASTID;
4717
4718                         CDEBUG(D_LFSCK, "%s: layout LFSCK has rebuilt "
4719                                "crashed LAST_ID files successfully\n",
4720                                lfsck_lfsck2name(lfsck));
4721                 }
4722                 lo->ll_flags &= ~LF_UPGRADE;
4723                 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4724         } else if (result == 0) {
4725                 if (lfsck->li_status != 0)
4726                         lo->ll_status = lfsck->li_status;
4727                 else
4728                         lo->ll_status = LS_STOPPED;
4729                 if (lo->ll_status != LS_PAUSED)
4730                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4731         } else {
4732                 lo->ll_status = LS_FAILED;
4733                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4734         }
4735         spin_unlock(&lfsck->li_lock);
4736
4737         if (done)
4738                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4739                                      LE_LASTID_REBUILT);
4740
4741         if (!init) {
4742                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4743                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4744                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4745                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4746                 com->lc_new_checked = 0;
4747         }
4748
4749         rc = lfsck_layout_store(env, com);
4750         up_write(&com->lc_sem);
4751
4752         lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
4753
4754         CDEBUG(D_LFSCK, "%s: layout LFSCK slave post done: rc = %d\n",
4755                lfsck_lfsck2name(lfsck), rc);
4756
4757         return rc;
4758 }
4759
4760 static int lfsck_layout_dump(const struct lu_env *env,
4761                              struct lfsck_component *com, struct seq_file *m)
4762 {
4763         struct lfsck_instance   *lfsck = com->lc_lfsck;
4764         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
4765         struct lfsck_layout     *lo    = com->lc_file_ram;
4766         int                      rc;
4767
4768         down_read(&com->lc_sem);
4769         seq_printf(m, "name: lfsck_layout\n"
4770                       "magic: %#x\n"
4771                       "version: %d\n"
4772                       "status: %s\n",
4773                       lo->ll_magic,
4774                       bk->lb_version,
4775                       lfsck_status2names(lo->ll_status));
4776
4777         rc = lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
4778         if (rc < 0)
4779                 goto out;
4780
4781         rc = lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4782         if (rc < 0)
4783                 goto out;
4784
4785         rc = lfsck_time_dump(m, lo->ll_time_last_complete,
4786                              "time_since_last_completed");
4787         if (rc < 0)
4788                 goto out;
4789
4790         rc = lfsck_time_dump(m, lo->ll_time_latest_start,
4791                              "time_since_latest_start");
4792         if (rc < 0)
4793                 goto out;
4794
4795         rc = lfsck_time_dump(m, lo->ll_time_last_checkpoint,
4796                              "time_since_last_checkpoint");
4797         if (rc < 0)
4798                 goto out;
4799
4800         seq_printf(m, "latest_start_position: "LPU64"\n"
4801                       "last_checkpoint_position: "LPU64"\n"
4802                       "first_failure_position: "LPU64"\n",
4803                       lo->ll_pos_latest_start,
4804                       lo->ll_pos_last_checkpoint,
4805                       lo->ll_pos_first_inconsistent);
4806
4807         seq_printf(m, "success_count: %u\n"
4808                       "repaired_dangling: "LPU64"\n"
4809                       "repaired_unmatched_pair: "LPU64"\n"
4810                       "repaired_multiple_referenced: "LPU64"\n"
4811                       "repaired_orphan: "LPU64"\n"
4812                       "repaired_inconsistent_owner: "LPU64"\n"
4813                       "repaired_others: "LPU64"\n"
4814                       "skipped: "LPU64"\n"
4815                       "failed_phase1: "LPU64"\n"
4816                       "failed_phase2: "LPU64"\n",
4817                       lo->ll_success_count,
4818                       lo->ll_objs_repaired[LLIT_DANGLING - 1],
4819                       lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
4820                       lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
4821                       lo->ll_objs_repaired[LLIT_ORPHAN - 1],
4822                       lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
4823                       lo->ll_objs_repaired[LLIT_OTHERS - 1],
4824                       lo->ll_objs_skipped,
4825                       lo->ll_objs_failed_phase1,
4826                       lo->ll_objs_failed_phase2);
4827
4828         if (lo->ll_status == LS_SCANNING_PHASE1) {
4829                 __u64 pos;
4830                 const struct dt_it_ops *iops;
4831                 cfs_duration_t duration = cfs_time_current() -
4832                                           lfsck->li_time_last_checkpoint;
4833                 __u64 checked = lo->ll_objs_checked_phase1 +
4834                                 com->lc_new_checked;
4835                 __u64 speed = checked;
4836                 __u64 new_checked = com->lc_new_checked *
4837                                     msecs_to_jiffies(MSEC_PER_SEC);
4838                 __u32 rtime = lo->ll_run_time_phase1 +
4839                               cfs_duration_sec(duration + HALF_SEC);
4840
4841                 if (duration != 0)
4842                         do_div(new_checked, duration);
4843                 if (rtime != 0)
4844                         do_div(speed, rtime);
4845                 seq_printf(m, "checked_phase1: "LPU64"\n"
4846                               "checked_phase2: "LPU64"\n"
4847                               "run_time_phase1: %u seconds\n"
4848                               "run_time_phase2: %u seconds\n"
4849                               "average_speed_phase1: "LPU64" items/sec\n"
4850                               "average_speed_phase2: N/A\n"
4851                               "real-time_speed_phase1: "LPU64" items/sec\n"
4852                               "real-time_speed_phase2: N/A\n",
4853                               checked,
4854                               lo->ll_objs_checked_phase2,
4855                               rtime,
4856                               lo->ll_run_time_phase2,
4857                               speed,
4858                               new_checked);
4859
4860                 LASSERT(lfsck->li_di_oit != NULL);
4861
4862                 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
4863
4864                 /* The low layer otable-based iteration position may NOT
4865                  * exactly match the layout-based directory traversal
4866                  * cookie. Generally, it is not a serious issue. But the
4867                  * caller should NOT make assumption on that. */
4868                 pos = iops->store(env, lfsck->li_di_oit);
4869                 if (!lfsck->li_current_oit_processed)
4870                         pos--;
4871                 seq_printf(m, "current_position: "LPU64"\n", pos);
4872
4873         } else if (lo->ll_status == LS_SCANNING_PHASE2) {
4874                 cfs_duration_t duration = cfs_time_current() -
4875                                           lfsck->li_time_last_checkpoint;
4876                 __u64 checked = lo->ll_objs_checked_phase2 +
4877                                 com->lc_new_checked;
4878                 __u64 speed1 = lo->ll_objs_checked_phase1;
4879                 __u64 speed2 = checked;
4880                 __u64 new_checked = com->lc_new_checked *
4881                                     msecs_to_jiffies(MSEC_PER_SEC);
4882                 __u32 rtime = lo->ll_run_time_phase2 +
4883                               cfs_duration_sec(duration + HALF_SEC);
4884
4885                 if (duration != 0)
4886                         do_div(new_checked, duration);
4887                 if (lo->ll_run_time_phase1 != 0)
4888                         do_div(speed1, lo->ll_run_time_phase1);
4889                 if (rtime != 0)
4890                         do_div(speed2, rtime);
4891                 rc = seq_printf(m, "checked_phase1: "LPU64"\n"
4892                                 "checked_phase2: "LPU64"\n"
4893                                 "run_time_phase1: %u seconds\n"
4894                                 "run_time_phase2: %u seconds\n"
4895                                 "average_speed_phase1: "LPU64" items/sec\n"
4896                                 "average_speed_phase2: "LPU64" items/sec\n"
4897                                 "real-time_speed_phase1: N/A\n"
4898                                 "real-time_speed_phase2: "LPU64" items/sec\n"
4899                                 "current_position: "DFID"\n",
4900                                 lo->ll_objs_checked_phase1,
4901                                 checked,
4902                                 lo->ll_run_time_phase1,
4903                                 rtime,
4904                                 speed1,
4905                                 speed2,
4906                                 new_checked,
4907                                 PFID(&com->lc_fid_latest_scanned_phase2));
4908                 if (rc <= 0)
4909                         goto out;
4910
4911         } else {
4912                 __u64 speed1 = lo->ll_objs_checked_phase1;
4913                 __u64 speed2 = lo->ll_objs_checked_phase2;
4914
4915                 if (lo->ll_run_time_phase1 != 0)
4916                         do_div(speed1, lo->ll_run_time_phase1);
4917                 if (lo->ll_run_time_phase2 != 0)
4918                         do_div(speed2, lo->ll_run_time_phase2);
4919                 seq_printf(m, "checked_phase1: "LPU64"\n"
4920                            "checked_phase2: "LPU64"\n"
4921                            "run_time_phase1: %u seconds\n"
4922                            "run_time_phase2: %u seconds\n"
4923                            "average_speed_phase1: "LPU64" items/sec\n"
4924                            "average_speed_phase2: "LPU64" objs/sec\n"
4925                            "real-time_speed_phase1: N/A\n"
4926                            "real-time_speed_phase2: N/A\n"
4927                            "current_position: N/A\n",
4928                            lo->ll_objs_checked_phase1,
4929                            lo->ll_objs_checked_phase2,
4930                            lo->ll_run_time_phase1,
4931                            lo->ll_run_time_phase2,
4932                            speed1,
4933                            speed2);
4934         }
4935 out:
4936         up_read(&com->lc_sem);
4937
4938         return rc;
4939 }
4940
4941 static int lfsck_layout_master_double_scan(const struct lu_env *env,
4942                                            struct lfsck_component *com)
4943 {
4944         struct lfsck_layout             *lo     = com->lc_file_ram;
4945         struct lfsck_assistant_data     *lad    = com->lc_data;
4946         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4947         struct lfsck_tgt_descs          *ltds;
4948         struct lfsck_tgt_desc           *ltd;
4949         struct lfsck_tgt_desc           *next;
4950         int                              rc;
4951
4952         rc = lfsck_double_scan_generic(env, com, lo->ll_status);
4953
4954         if (thread_is_stopped(&lad->lad_thread)) {
4955                 LASSERT(list_empty(&lad->lad_req_list));
4956                 LASSERT(list_empty(&lad->lad_ost_phase1_list));
4957                 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4958
4959                 ltds = &lfsck->li_ost_descs;
4960                 spin_lock(&ltds->ltd_lock);
4961                 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
4962                                          ltd_layout_phase_list) {
4963                         list_del_init(&ltd->ltd_layout_phase_list);
4964                 }
4965                 spin_unlock(&ltds->ltd_lock);
4966
4967                 ltds = &lfsck->li_mdt_descs;
4968                 spin_lock(&ltds->ltd_lock);
4969                 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4970                                          ltd_layout_phase_list) {
4971                         list_del_init(&ltd->ltd_layout_phase_list);
4972                 }
4973                 spin_unlock(&ltds->ltd_lock);
4974         }
4975
4976         return rc;
4977 }
4978
4979 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
4980                                           struct lfsck_component *com)
4981 {
4982         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4983         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4984         struct lfsck_layout             *lo     = com->lc_file_ram;
4985         struct ptlrpc_thread            *thread = &lfsck->li_thread;
4986         int                              rc;
4987         ENTRY;
4988
4989         CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan start\n",
4990                lfsck_lfsck2name(lfsck));
4991
4992         atomic_inc(&lfsck->li_double_scan_count);
4993
4994         if (lo->ll_flags & LF_INCOMPLETE)
4995                 GOTO(done, rc = 1);
4996
4997         com->lc_new_checked = 0;
4998         com->lc_new_scanned = 0;
4999         com->lc_time_last_checkpoint = cfs_time_current();
5000         com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
5001                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
5002
5003         while (1) {
5004                 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(30),
5005                                                      NULL, NULL);
5006
5007                 rc = lfsck_layout_slave_query_master(env, com);
5008                 if (list_empty(&llsd->llsd_master_list)) {
5009                         if (unlikely(!thread_is_running(thread)))
5010                                 rc = 0;
5011                         else
5012                                 rc = 1;
5013
5014                         GOTO(done, rc);
5015                 }
5016
5017                 if (rc < 0)
5018                         GOTO(done, rc);
5019
5020                 rc = l_wait_event(thread->t_ctl_waitq,
5021                                   !thread_is_running(thread) ||
5022                                   lo->ll_flags & LF_INCOMPLETE ||
5023                                   list_empty(&llsd->llsd_master_list),
5024                                   &lwi);
5025                 if (unlikely(!thread_is_running(thread)))
5026                         GOTO(done, rc = 0);
5027
5028                 if (lo->ll_flags & LF_INCOMPLETE)
5029                         GOTO(done, rc = 1);
5030
5031                 if (rc == -ETIMEDOUT)
5032                         continue;
5033
5034                 GOTO(done, rc = (rc < 0 ? rc : 1));
5035         }
5036
5037 done:
5038         rc = lfsck_layout_double_scan_result(env, com, rc);
5039         lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE,
5040                         (rc > 0 && lo->ll_flags & LF_INCOMPLETE) ? 0 : rc);
5041         lfsck_layout_slave_quit(env, com);
5042         if (atomic_dec_and_test(&lfsck->li_double_scan_count))
5043                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5044
5045         CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan finished, "
5046                "status %d: rc = %d\n",
5047                lfsck_lfsck2name(lfsck), lo->ll_status, rc);
5048
5049         return rc;
5050 }
5051
5052 static void lfsck_layout_master_data_release(const struct lu_env *env,
5053                                              struct lfsck_component *com)
5054 {
5055         struct lfsck_assistant_data     *lad    = com->lc_data;
5056         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5057         struct lfsck_tgt_descs          *ltds;
5058         struct lfsck_tgt_desc           *ltd;
5059         struct lfsck_tgt_desc           *next;
5060
5061         LASSERT(lad != NULL);
5062         LASSERT(thread_is_init(&lad->lad_thread) ||
5063                 thread_is_stopped(&lad->lad_thread));
5064         LASSERT(list_empty(&lad->lad_req_list));
5065
5066         com->lc_data = NULL;
5067
5068         ltds = &lfsck->li_ost_descs;
5069         spin_lock(&ltds->ltd_lock);
5070         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
5071                                  ltd_layout_phase_list) {
5072                 list_del_init(&ltd->ltd_layout_phase_list);
5073         }
5074         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5075                                  ltd_layout_phase_list) {
5076                 list_del_init(&ltd->ltd_layout_phase_list);
5077         }
5078         list_for_each_entry_safe(ltd, next, &lad->lad_ost_list,
5079                                  ltd_layout_list) {
5080                 list_del_init(&ltd->ltd_layout_list);
5081         }
5082         spin_unlock(&ltds->ltd_lock);
5083
5084         ltds = &lfsck->li_mdt_descs;
5085         spin_lock(&ltds->ltd_lock);
5086         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5087                                  ltd_layout_phase_list) {
5088                 list_del_init(&ltd->ltd_layout_phase_list);
5089         }
5090         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5091                                  ltd_layout_phase_list) {
5092                 list_del_init(&ltd->ltd_layout_phase_list);
5093         }
5094         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
5095                                  ltd_layout_list) {
5096                 list_del_init(&ltd->ltd_layout_list);
5097         }
5098         spin_unlock(&ltds->ltd_lock);
5099
5100         if (likely(lad->lad_bitmap != NULL))
5101                 CFS_FREE_BITMAP(lad->lad_bitmap);
5102
5103         OBD_FREE_PTR(lad);
5104 }
5105
5106 static void lfsck_layout_slave_data_release(const struct lu_env *env,
5107                                             struct lfsck_component *com)
5108 {
5109         struct lfsck_layout_slave_data *llsd = com->lc_data;
5110
5111         lfsck_layout_slave_quit(env, com);
5112         com->lc_data = NULL;
5113         OBD_FREE_PTR(llsd);
5114 }
5115
5116 static void lfsck_layout_master_quit(const struct lu_env *env,
5117                                      struct lfsck_component *com)
5118 {
5119         struct lfsck_assistant_data     *lad    = com->lc_data;
5120         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5121         struct lfsck_tgt_descs          *ltds;
5122         struct lfsck_tgt_desc           *ltd;
5123         struct lfsck_tgt_desc           *next;
5124
5125         LASSERT(lad != NULL);
5126
5127         lfsck_quit_generic(env, com);
5128
5129         LASSERT(thread_is_init(&lad->lad_thread) ||
5130                 thread_is_stopped(&lad->lad_thread));
5131         LASSERT(list_empty(&lad->lad_req_list));
5132
5133         ltds = &lfsck->li_ost_descs;
5134         spin_lock(&ltds->ltd_lock);
5135         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
5136                                  ltd_layout_phase_list) {
5137                 list_del_init(&ltd->ltd_layout_phase_list);
5138         }
5139         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5140                                  ltd_layout_phase_list) {
5141                 list_del_init(&ltd->ltd_layout_phase_list);
5142         }
5143         spin_unlock(&ltds->ltd_lock);
5144
5145         ltds = &lfsck->li_mdt_descs;
5146         spin_lock(&ltds->ltd_lock);
5147         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5148                                  ltd_layout_phase_list) {
5149                 list_del_init(&ltd->ltd_layout_phase_list);
5150         }
5151         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5152                                  ltd_layout_phase_list) {
5153                 list_del_init(&ltd->ltd_layout_phase_list);
5154         }
5155         spin_unlock(&ltds->ltd_lock);
5156 }
5157
5158 static void lfsck_layout_slave_quit(const struct lu_env *env,
5159                                     struct lfsck_component *com)
5160 {
5161         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5162         struct lfsck_layout_seq          *lls;
5163         struct lfsck_layout_seq          *next;
5164         struct lfsck_layout_slave_target *llst;
5165
5166         LASSERT(llsd != NULL);
5167
5168         list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
5169                                  lls_list) {
5170                 list_del_init(&lls->lls_list);
5171                 lfsck_object_put(env, lls->lls_lastid_obj);
5172                 OBD_FREE_PTR(lls);
5173         }
5174
5175         spin_lock(&llsd->llsd_lock);
5176         while (!list_empty(&llsd->llsd_master_list)) {
5177                 llst = list_entry(llsd->llsd_master_list.next,
5178                                   struct lfsck_layout_slave_target, llst_list);
5179                 list_del_init(&llst->llst_list);
5180                 spin_unlock(&llsd->llsd_lock);
5181                 lfsck_layout_llst_put(llst);
5182                 spin_lock(&llsd->llsd_lock);
5183         }
5184         spin_unlock(&llsd->llsd_lock);
5185
5186         lfsck_rbtree_cleanup(env, com);
5187 }
5188
5189 static int lfsck_layout_master_in_notify(const struct lu_env *env,
5190                                          struct lfsck_component *com,
5191                                          struct lfsck_request *lr,
5192                                          struct thandle *th)
5193 {
5194         struct lfsck_instance           *lfsck = com->lc_lfsck;
5195         struct lfsck_layout             *lo    = com->lc_file_ram;
5196         struct lfsck_assistant_data     *lad   = com->lc_data;
5197         struct lfsck_tgt_descs          *ltds;
5198         struct lfsck_tgt_desc           *ltd;
5199         bool                             fail  = false;
5200         ENTRY;
5201
5202         if (lr->lr_event == LE_PAIRS_VERIFY) {
5203                 int rc;
5204
5205                 rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
5206                                                      &lr->lr_fid2);
5207
5208                 RETURN(rc);
5209         }
5210
5211         CDEBUG(D_LFSCK, "%s: layout LFSCK master handles notify %u "
5212                "from %s %x, status %d, flags %x, flags2 %x\n",
5213                lfsck_lfsck2name(lfsck), lr->lr_event,
5214                (lr->lr_flags & LEF_FROM_OST) ? "OST" : "MDT",
5215                lr->lr_index, lr->lr_status, lr->lr_flags, lr->lr_flags2);
5216
5217         if (lr->lr_event != LE_PHASE1_DONE &&
5218             lr->lr_event != LE_PHASE2_DONE &&
5219             lr->lr_event != LE_PEER_EXIT)
5220                 RETURN(-EINVAL);
5221
5222         if (lr->lr_flags & LEF_FROM_OST)
5223                 ltds = &lfsck->li_ost_descs;
5224         else
5225                 ltds = &lfsck->li_mdt_descs;
5226         spin_lock(&ltds->ltd_lock);
5227         ltd = LTD_TGT(ltds, lr->lr_index);
5228         if (ltd == NULL) {
5229                 spin_unlock(&ltds->ltd_lock);
5230
5231                 RETURN(-ENXIO);
5232         }
5233
5234         list_del_init(&ltd->ltd_layout_phase_list);
5235         switch (lr->lr_event) {
5236         case LE_PHASE1_DONE:
5237                 if (lr->lr_status <= 0 || lr->lr_flags2 & LF_INCOMPLETE) {
5238                         if (lr->lr_flags2 & LF_INCOMPLETE) {
5239                                 if (lr->lr_flags & LEF_FROM_OST)
5240                                         lfsck_lad_set_bitmap(env, com,
5241                                                              ltd->ltd_index);
5242                                 else
5243                                         lo->ll_flags |= LF_INCOMPLETE;
5244                         }
5245                         ltd->ltd_layout_done = 1;
5246                         list_del_init(&ltd->ltd_layout_list);
5247                         fail = true;
5248                         break;
5249                 }
5250
5251                 if (lr->lr_flags & LEF_FROM_OST) {
5252                         if (list_empty(&ltd->ltd_layout_list))
5253                                 list_add_tail(&ltd->ltd_layout_list,
5254                                               &lad->lad_ost_list);
5255                         list_add_tail(&ltd->ltd_layout_phase_list,
5256                                       &lad->lad_ost_phase2_list);
5257                 } else {
5258                         if (list_empty(&ltd->ltd_layout_list))
5259                                 list_add_tail(&ltd->ltd_layout_list,
5260                                               &lad->lad_mdt_list);
5261                         list_add_tail(&ltd->ltd_layout_phase_list,
5262                                       &lad->lad_mdt_phase2_list);
5263                 }
5264                 break;
5265         case LE_PHASE2_DONE:
5266                 ltd->ltd_layout_done = 1;
5267                 if (!list_empty(&ltd->ltd_layout_list)) {
5268                         list_del_init(&ltd->ltd_layout_list);
5269                         if (lr->lr_flags2 & LF_INCOMPLETE) {
5270                                 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
5271                                 fail = true;
5272                         }
5273                 }
5274
5275                 break;
5276         case LE_PEER_EXIT:
5277                 fail = true;
5278                 ltd->ltd_layout_done = 1;
5279                 list_del_init(&ltd->ltd_layout_list);
5280                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) &&
5281                     !(lr->lr_flags & LEF_FROM_OST))
5282                                 lo->ll_flags |= LF_INCOMPLETE;
5283                 break;
5284         default:
5285                 break;
5286         }
5287         spin_unlock(&ltds->ltd_lock);
5288
5289         if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
5290                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5291
5292                 memset(stop, 0, sizeof(*stop));
5293                 stop->ls_status = lr->lr_status;
5294                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5295                 lfsck_stop(env, lfsck->li_bottom, stop);
5296         } else if (lfsck_phase2_next_ready(lad)) {
5297                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
5298         }
5299
5300         RETURN(0);
5301 }
5302
5303 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
5304                                         struct lfsck_component *com,
5305                                         struct lfsck_request *lr,
5306                                         struct thandle *th)
5307 {
5308         struct lfsck_instance            *lfsck = com->lc_lfsck;
5309         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5310         struct lfsck_layout_slave_target *llst;
5311         int                               rc;
5312         ENTRY;
5313
5314         switch (lr->lr_event) {
5315         case LE_FID_ACCESSED:
5316                 lfsck_rbtree_update_bitmap(env, com, &lr->lr_fid, true);
5317                 RETURN(0);
5318         case LE_CONDITIONAL_DESTROY:
5319                 rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
5320                 RETURN(rc);
5321         case LE_PAIRS_VERIFY: {
5322                 lr->lr_status = LPVS_INIT;
5323                 /* Firstly, if the MDT-object which is claimed via OST-object
5324                  * local stored PFID xattr recognizes the OST-object, then it
5325                  * must be that the client given PFID is wrong. */
5326                 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5327                                                     &lr->lr_fid3);
5328                 if (rc <= 0)
5329                         RETURN(0);
5330
5331                 lr->lr_status = LPVS_INCONSISTENT;
5332                 /* The OST-object local stored PFID xattr is stale. We need to
5333                  * check whether the MDT-object that is claimed via the client
5334                  * given PFID information recognizes the OST-object or not. If
5335                  * matches, then need to update the OST-object's PFID xattr. */
5336                 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5337                                                     &lr->lr_fid2);
5338                 /* For rc < 0 case:
5339                  * We are not sure whether the client given PFID information
5340                  * is correct or not, do nothing to avoid improper fixing.
5341                  *
5342                  * For rc > 0 case:
5343                  * The client given PFID information is also invalid, we can
5344                  * NOT fix the OST-object inconsistency.
5345                  */
5346                 if (rc != 0)
5347                         RETURN(rc);
5348
5349                 lr->lr_status = LPVS_INCONSISTENT_TOFIX;
5350                 rc = lfsck_layout_slave_repair_pfid(env, com, lr);
5351
5352                 RETURN(rc);
5353         }
5354         case LE_PHASE1_DONE: {
5355                 if (lr->lr_flags2 & LF_INCOMPLETE) {
5356                         struct lfsck_layout *lo = com->lc_file_ram;
5357
5358                         lo->ll_flags |= LF_INCOMPLETE;
5359                         llst = lfsck_layout_llst_find_and_del(llsd,
5360                                                               lr->lr_index,
5361                                                               true);
5362                         if (llst != NULL) {
5363                                 lfsck_layout_llst_put(llst);
5364                                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5365                         }
5366                 }
5367
5368                 RETURN(0);
5369         }
5370         case LE_PHASE2_DONE:
5371         case LE_PEER_EXIT:
5372                 CDEBUG(D_LFSCK, "%s: layout LFSCK slave handle notify %u "
5373                        "from MDT %x, status %d\n", lfsck_lfsck2name(lfsck),
5374                        lr->lr_event, lr->lr_index, lr->lr_status);
5375                 break;
5376         default:
5377                 RETURN(-EINVAL);
5378         }
5379
5380         llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
5381         if (llst == NULL)
5382                 RETURN(0);
5383
5384         lfsck_layout_llst_put(llst);
5385         if (list_empty(&llsd->llsd_master_list))
5386                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5387
5388         if (lr->lr_event == LE_PEER_EXIT &&
5389             (lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT ||
5390              (list_empty(&llsd->llsd_master_list) &&
5391               (lr->lr_status == LS_STOPPED ||
5392                lr->lr_status == LS_CO_STOPPED)))) {
5393                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5394
5395                 memset(stop, 0, sizeof(*stop));
5396                 stop->ls_status = lr->lr_status;
5397                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5398                 lfsck_stop(env, lfsck->li_bottom, stop);
5399         }
5400
5401         RETURN(0);
5402 }
5403
5404 static int lfsck_layout_query(const struct lu_env *env,
5405                               struct lfsck_component *com)
5406 {
5407         struct lfsck_layout *lo = com->lc_file_ram;
5408
5409         return lo->ll_status;
5410 }
5411
5412 /* with lfsck::li_lock held */
5413 static int lfsck_layout_slave_join(const struct lu_env *env,
5414                                    struct lfsck_component *com,
5415                                    struct lfsck_start_param *lsp)
5416 {
5417         struct lfsck_instance            *lfsck = com->lc_lfsck;
5418         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5419         struct lfsck_layout_slave_target *llst;
5420         struct lfsck_start               *start = lsp->lsp_start;
5421         int                               rc    = 0;
5422         ENTRY;
5423
5424         if (start == NULL || !(start->ls_flags & LPF_OST_ORPHAN))
5425                 RETURN(0);
5426
5427         if (!lsp->lsp_index_valid)
5428                 RETURN(-EINVAL);
5429
5430         /* If someone is running the LFSCK without orphan handling,
5431          * it will not maintain the object accessing rbtree. So we
5432          * cannot join it for orphan handling. */
5433         if (!llsd->llsd_rbtree_valid)
5434                 RETURN(-EBUSY);
5435
5436         spin_unlock(&lfsck->li_lock);
5437         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5438         spin_lock(&lfsck->li_lock);
5439         if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
5440                 spin_unlock(&lfsck->li_lock);
5441                 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
5442                                                       true);
5443                 if (llst != NULL)
5444                         lfsck_layout_llst_put(llst);
5445                 spin_lock(&lfsck->li_lock);
5446                 rc = -EAGAIN;
5447         }
5448
5449         RETURN(rc);
5450 }
5451
5452 static struct lfsck_operations lfsck_layout_master_ops = {
5453         .lfsck_reset            = lfsck_layout_reset,
5454         .lfsck_fail             = lfsck_layout_fail,
5455         .lfsck_checkpoint       = lfsck_layout_master_checkpoint,
5456         .lfsck_prep             = lfsck_layout_master_prep,
5457         .lfsck_exec_oit         = lfsck_layout_master_exec_oit,
5458         .lfsck_exec_dir         = lfsck_layout_exec_dir,
5459         .lfsck_post             = lfsck_layout_master_post,
5460         .lfsck_dump             = lfsck_layout_dump,
5461         .lfsck_double_scan      = lfsck_layout_master_double_scan,
5462         .lfsck_data_release     = lfsck_layout_master_data_release,
5463         .lfsck_quit             = lfsck_layout_master_quit,
5464         .lfsck_in_notify        = lfsck_layout_master_in_notify,
5465         .lfsck_query            = lfsck_layout_query,
5466 };
5467
5468 static struct lfsck_operations lfsck_layout_slave_ops = {
5469         .lfsck_reset            = lfsck_layout_reset,
5470         .lfsck_fail             = lfsck_layout_fail,
5471         .lfsck_checkpoint       = lfsck_layout_slave_checkpoint,
5472         .lfsck_prep             = lfsck_layout_slave_prep,
5473         .lfsck_exec_oit         = lfsck_layout_slave_exec_oit,
5474         .lfsck_exec_dir         = lfsck_layout_exec_dir,
5475         .lfsck_post             = lfsck_layout_slave_post,
5476         .lfsck_dump             = lfsck_layout_dump,
5477         .lfsck_double_scan      = lfsck_layout_slave_double_scan,
5478         .lfsck_data_release     = lfsck_layout_slave_data_release,
5479         .lfsck_quit             = lfsck_layout_slave_quit,
5480         .lfsck_in_notify        = lfsck_layout_slave_in_notify,
5481         .lfsck_query            = lfsck_layout_query,
5482         .lfsck_join             = lfsck_layout_slave_join,
5483 };
5484
5485 static void lfsck_layout_assistant_fill_pos(const struct lu_env *env,
5486                                             struct lfsck_component *com,
5487                                             struct lfsck_position *pos)
5488 {
5489         struct lfsck_assistant_data     *lad = com->lc_data;
5490         struct lfsck_layout_req         *llr;
5491
5492         if (list_empty(&lad->lad_req_list))
5493                 return;
5494
5495         llr = list_entry(lad->lad_req_list.next,
5496                          struct lfsck_layout_req,
5497                          llr_lar.lar_list);
5498         pos->lp_oit_cookie = llr->llr_parent->llo_cookie - 1;
5499 }
5500
5501 struct lfsck_assistant_operations lfsck_layout_assistant_ops = {
5502         .la_handler_p1          = lfsck_layout_assistant_handler_p1,
5503         .la_handler_p2          = lfsck_layout_assistant_handler_p2,
5504         .la_fill_pos            = lfsck_layout_assistant_fill_pos,
5505         .la_double_scan_result  = lfsck_layout_double_scan_result,
5506         .la_req_fini            = lfsck_layout_assistant_req_fini,
5507         .la_sync_failures       = lfsck_layout_assistant_sync_failures,
5508 };
5509
5510 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
5511 {
5512         struct lfsck_component  *com;
5513         struct lfsck_layout     *lo;
5514         struct dt_object        *root = NULL;
5515         struct dt_object        *obj;
5516         int                      rc;
5517         ENTRY;
5518
5519         OBD_ALLOC_PTR(com);
5520         if (com == NULL)
5521                 RETURN(-ENOMEM);
5522
5523         INIT_LIST_HEAD(&com->lc_link);
5524         INIT_LIST_HEAD(&com->lc_link_dir);
5525         init_rwsem(&com->lc_sem);
5526         atomic_set(&com->lc_ref, 1);
5527         com->lc_lfsck = lfsck;
5528         com->lc_type = LFSCK_TYPE_LAYOUT;
5529         if (lfsck->li_master) {
5530                 com->lc_ops = &lfsck_layout_master_ops;
5531                 com->lc_data = lfsck_assistant_data_init(
5532                                 &lfsck_layout_assistant_ops,
5533                                 LFSCK_LAYOUT);
5534                 if (com->lc_data == NULL)
5535                         GOTO(out, rc = -ENOMEM);
5536         } else {
5537                 struct lfsck_layout_slave_data *llsd;
5538
5539                 com->lc_ops = &lfsck_layout_slave_ops;
5540                 OBD_ALLOC_PTR(llsd);
5541                 if (llsd == NULL)
5542                         GOTO(out, rc = -ENOMEM);
5543
5544                 INIT_LIST_HEAD(&llsd->llsd_seq_list);
5545                 INIT_LIST_HEAD(&llsd->llsd_master_list);
5546                 spin_lock_init(&llsd->llsd_lock);
5547                 llsd->llsd_rb_root = RB_ROOT;
5548                 rwlock_init(&llsd->llsd_rb_lock);
5549                 com->lc_data = llsd;
5550         }
5551         com->lc_file_size = sizeof(*lo);
5552         OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
5553         if (com->lc_file_ram == NULL)
5554                 GOTO(out, rc = -ENOMEM);
5555
5556         OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
5557         if (com->lc_file_disk == NULL)
5558                 GOTO(out, rc = -ENOMEM);
5559
5560         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
5561         if (IS_ERR(root))
5562                 GOTO(out, rc = PTR_ERR(root));
5563
5564         if (unlikely(!dt_try_as_dir(env, root)))
5565                 GOTO(out, rc = -ENOTDIR);
5566
5567         obj = local_file_find_or_create(env, lfsck->li_los, root,
5568                                         LFSCK_LAYOUT,
5569                                         S_IFREG | S_IRUGO | S_IWUSR);
5570         if (IS_ERR(obj))
5571                 GOTO(out, rc = PTR_ERR(obj));
5572
5573         com->lc_obj = obj;
5574         rc = lfsck_layout_load(env, com);
5575         if (rc > 0)
5576                 rc = lfsck_layout_reset(env, com, true);
5577         else if (rc == -ENOENT)
5578                 rc = lfsck_layout_init(env, com);
5579
5580         if (rc != 0)
5581                 GOTO(out, rc);
5582
5583         lo = com->lc_file_ram;
5584         switch (lo->ll_status) {
5585         case LS_INIT:
5586         case LS_COMPLETED:
5587         case LS_FAILED:
5588         case LS_STOPPED:
5589         case LS_PARTIAL:
5590                 spin_lock(&lfsck->li_lock);
5591                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
5592                 spin_unlock(&lfsck->li_lock);
5593                 break;
5594         default:
5595                 CERROR("%s: unknown lfsck_layout status %d\n",
5596                        lfsck_lfsck2name(lfsck), lo->ll_status);
5597                 /* fall through */
5598         case LS_SCANNING_PHASE1:
5599         case LS_SCANNING_PHASE2:
5600                 /* No need to store the status to disk right now.
5601                  * If the system crashed before the status stored,
5602                  * it will be loaded back when next time. */
5603                 lo->ll_status = LS_CRASHED;
5604                 if (!lfsck->li_master)
5605                         lo->ll_flags |= LF_INCOMPLETE;
5606                 /* fall through */
5607         case LS_PAUSED:
5608         case LS_CRASHED:
5609         case LS_CO_FAILED:
5610         case LS_CO_STOPPED:
5611         case LS_CO_PAUSED:
5612                 spin_lock(&lfsck->li_lock);
5613                 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
5614                 spin_unlock(&lfsck->li_lock);
5615                 break;
5616         }
5617
5618         if (lo->ll_flags & LF_CRASHED_LASTID) {
5619                 LASSERT(lfsck->li_out_notify != NULL);
5620
5621                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5622                                      LE_LASTID_REBUILDING);
5623         }
5624
5625         GOTO(out, rc = 0);
5626
5627 out:
5628         if (root != NULL && !IS_ERR(root))
5629                 lfsck_object_put(env, root);
5630
5631         if (rc != 0) {
5632                 lfsck_component_cleanup(env, com);
5633                 CERROR("%s: fail to init layout LFSCK component: rc = %d\n",
5634                        lfsck_lfsck2name(lfsck), rc);
5635         }
5636
5637         return rc;
5638 }
5639
5640 struct lfsck_orphan_it {
5641         struct lfsck_component           *loi_com;
5642         struct lfsck_rbtree_node         *loi_lrn;
5643         struct lfsck_layout_slave_target *loi_llst;
5644         struct lu_fid                     loi_key;
5645         struct lu_orphan_rec              loi_rec;
5646         __u64                             loi_hash;
5647         unsigned int                      loi_over:1;
5648 };
5649
5650 static int lfsck_fid_match_idx(const struct lu_env *env,
5651                                struct lfsck_instance *lfsck,
5652                                const struct lu_fid *fid, int idx)
5653 {
5654         struct seq_server_site  *ss;
5655         struct lu_server_fld    *sf;
5656         struct lu_seq_range     *range = &lfsck_env_info(env)->lti_range;
5657         int                      rc;
5658
5659         /* All abnormal cases will be returned to MDT0. */
5660         if (!fid_is_norm(fid)) {
5661                 if (idx == 0)
5662                         return 1;
5663
5664                 return 0;
5665         }
5666
5667         ss = lfsck_dev_site(lfsck);
5668         if (unlikely(ss == NULL))
5669                 return -ENOTCONN;
5670
5671         sf = ss->ss_server_fld;
5672         LASSERT(sf != NULL);
5673
5674         fld_range_set_any(range);
5675         rc = fld_server_lookup(env, sf, fid_seq(fid), range);
5676         if (rc != 0)
5677                 return rc;
5678
5679         if (!fld_range_is_mdt(range))
5680                 return -EINVAL;
5681
5682         if (range->lsr_index == idx)
5683                 return 1;
5684
5685         return 0;
5686 }
5687
5688 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
5689                                         struct dt_object *obj)
5690 {
5691         struct dt_device        *dev    = lfsck_obj2dev(obj);
5692         struct thandle          *handle;
5693         int                      rc;
5694         ENTRY;
5695
5696         handle = dt_trans_create(env, dev);
5697         if (IS_ERR(handle))
5698                 RETURN_EXIT;
5699
5700         rc = dt_declare_ref_del(env, obj, handle);
5701         if (rc != 0)
5702                 GOTO(stop, rc);
5703
5704         rc = dt_declare_destroy(env, obj, handle);
5705         if (rc != 0)
5706                 GOTO(stop, rc);
5707
5708         rc = dt_trans_start_local(env, dev, handle);
5709         if (rc != 0)
5710                 GOTO(stop, rc);
5711
5712         dt_write_lock(env, obj, 0);
5713         rc = dt_ref_del(env, obj, handle);
5714         if (rc == 0)
5715                 rc = dt_destroy(env, obj, handle);
5716         dt_write_unlock(env, obj);
5717
5718         GOTO(stop, rc);
5719
5720 stop:
5721         dt_trans_stop(env, dev, handle);
5722
5723         CDEBUG(D_LFSCK, "destroy orphan OST-object "DFID": rc = %d\n",
5724                PFID(lfsck_dto2fid(obj)), rc);
5725
5726         RETURN_EXIT;
5727 }
5728
5729 static int lfsck_orphan_index_lookup(const struct lu_env *env,
5730                                      struct dt_object *dt,
5731                                      struct dt_rec *rec,
5732                                      const struct dt_key *key,
5733                                      struct lustre_capa *capa)
5734 {
5735         return -EOPNOTSUPP;
5736 }
5737
5738 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
5739                                              struct dt_object *dt,
5740                                              const struct dt_rec *rec,
5741                                              const struct dt_key *key,
5742                                              struct thandle *handle)
5743 {
5744         return -EOPNOTSUPP;
5745 }
5746
5747 static int lfsck_orphan_index_insert(const struct lu_env *env,
5748                                      struct dt_object *dt,
5749                                      const struct dt_rec *rec,
5750                                      const struct dt_key *key,
5751                                      struct thandle *handle,
5752                                      struct lustre_capa *capa,
5753                                      int ignore_quota)
5754 {
5755         return -EOPNOTSUPP;
5756 }
5757
5758 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
5759                                              struct dt_object *dt,
5760                                              const struct dt_key *key,
5761                                              struct thandle *handle)
5762 {
5763         return -EOPNOTSUPP;
5764 }
5765
5766 static int lfsck_orphan_index_delete(const struct lu_env *env,
5767                                      struct dt_object *dt,
5768                                      const struct dt_key *key,
5769                                      struct thandle *handle,
5770                                      struct lustre_capa *capa)
5771 {
5772         return -EOPNOTSUPP;
5773 }
5774
5775 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
5776                                           struct dt_object *dt,
5777                                           __u32 attr,
5778                                           struct lustre_capa *capa)
5779 {
5780         struct dt_device                *dev    = lu2dt_dev(dt->do_lu.lo_dev);
5781         struct lfsck_instance           *lfsck;
5782         struct lfsck_component          *com    = NULL;
5783         struct lfsck_layout_slave_data  *llsd;
5784         struct lfsck_orphan_it          *it     = NULL;
5785         struct lfsck_layout             *lo;
5786         int                              rc     = 0;
5787         ENTRY;
5788
5789         lfsck = lfsck_instance_find(dev, true, false);
5790         if (unlikely(lfsck == NULL))
5791                 RETURN(ERR_PTR(-ENXIO));
5792
5793         com = lfsck_component_find(lfsck, LFSCK_TYPE_LAYOUT);
5794         if (unlikely(com == NULL))
5795                 GOTO(out, rc = -ENOENT);
5796
5797         lo = com->lc_file_ram;
5798         if (lo->ll_flags & LF_INCOMPLETE)
5799                 GOTO(out, rc = -ESRCH);
5800
5801         llsd = com->lc_data;
5802         if (!llsd->llsd_rbtree_valid)
5803                 GOTO(out, rc = -ESRCH);
5804
5805         OBD_ALLOC_PTR(it);
5806         if (it == NULL)
5807                 GOTO(out, rc = -ENOMEM);
5808
5809         it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
5810         if (it->loi_llst == NULL)
5811                 GOTO(out, rc = -ENXIO);
5812
5813         if (dev->dd_record_fid_accessed) {
5814                 /* The first iteration against the rbtree, scan the whole rbtree
5815                  * to remove the nodes which do NOT need to be handled. */
5816                 write_lock(&llsd->llsd_rb_lock);
5817                 if (dev->dd_record_fid_accessed) {
5818                         struct rb_node                  *node;
5819                         struct rb_node                  *next;
5820                         struct lfsck_rbtree_node        *lrn;
5821
5822                         /* No need to record the fid accessing anymore. */
5823                         dev->dd_record_fid_accessed = 0;
5824
5825                         node = rb_first(&llsd->llsd_rb_root);
5826                         while (node != NULL) {
5827                                 next = rb_next(node);
5828                                 lrn = rb_entry(node, struct lfsck_rbtree_node,
5829                                                lrn_node);
5830                                 if (atomic_read(&lrn->lrn_known_count) <=
5831                                     atomic_read(&lrn->lrn_accessed_count)) {
5832                                         rb_erase(node, &llsd->llsd_rb_root);
5833                                         lfsck_rbtree_free(lrn);
5834                                 }
5835                                 node = next;
5836                         }
5837                 }
5838                 write_unlock(&llsd->llsd_rb_lock);
5839         }
5840
5841         /* read lock the rbtree when init, and unlock when fini */
5842         read_lock(&llsd->llsd_rb_lock);
5843         it->loi_com = com;
5844         com = NULL;
5845
5846         GOTO(out, rc = 0);
5847
5848 out:
5849         if (com != NULL)
5850                 lfsck_component_put(env, com);
5851
5852         CDEBUG(D_LFSCK, "%s: init the orphan iteration: rc = %d\n",
5853                lfsck_lfsck2name(lfsck), rc);
5854
5855         lfsck_instance_put(env, lfsck);
5856         if (rc != 0) {
5857                 if (it != NULL)
5858                         OBD_FREE_PTR(it);
5859
5860                 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
5861         }
5862
5863         return (struct dt_it *)it;
5864 }
5865
5866 static void lfsck_orphan_it_fini(const struct lu_env *env,
5867                                  struct dt_it *di)
5868 {
5869         struct lfsck_orphan_it           *it    = (struct lfsck_orphan_it *)di;
5870         struct lfsck_component           *com   = it->loi_com;
5871         struct lfsck_layout_slave_data   *llsd;
5872         struct lfsck_layout_slave_target *llst;
5873
5874         if (com != NULL) {
5875                 CDEBUG(D_LFSCK, "%s: fini the orphan iteration\n",
5876                        lfsck_lfsck2name(com->lc_lfsck));
5877
5878                 llsd = com->lc_data;
5879                 read_unlock(&llsd->llsd_rb_lock);
5880                 llst = it->loi_llst;
5881                 LASSERT(llst != NULL);
5882
5883                 /* Save the key and hash for iterate next. */
5884                 llst->llst_fid = it->loi_key;
5885                 llst->llst_hash = it->loi_hash;
5886                 lfsck_layout_llst_put(llst);
5887                 lfsck_component_put(env, com);
5888         }
5889         OBD_FREE_PTR(it);
5890 }
5891
5892 /**
5893  * \retval       +1: the iteration finished
5894  * \retval        0: on success, not finished
5895  * \retval      -ve: on error
5896  */
5897 static int lfsck_orphan_it_next(const struct lu_env *env,
5898                                 struct dt_it *di)
5899 {
5900         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5901         struct filter_fid_old           *pfid   = &info->lti_old_pfid;
5902         struct lu_attr                  *la     = &info->lti_la;
5903         struct lfsck_orphan_it          *it     = (struct lfsck_orphan_it *)di;
5904         struct lu_fid                   *key    = &it->loi_key;
5905         struct lu_orphan_rec            *rec    = &it->loi_rec;
5906         struct lfsck_component          *com    = it->loi_com;
5907         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5908         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
5909         struct dt_object                *obj;
5910         struct lfsck_rbtree_node        *lrn;
5911         int                              pos;
5912         int                              rc;
5913         __u32                            save;
5914         __u32                            idx    = it->loi_llst->llst_index;
5915         bool                             exact  = false;
5916         ENTRY;
5917
5918         if (it->loi_over)
5919                 RETURN(1);
5920
5921 again0:
5922         lrn = it->loi_lrn;
5923         if (lrn == NULL) {
5924                 lrn = lfsck_rbtree_search(llsd, key, &exact);
5925                 if (lrn == NULL) {
5926                         it->loi_over = 1;
5927                         RETURN(1);
5928                 }
5929
5930                 it->loi_lrn = lrn;
5931                 if (!exact) {
5932                         key->f_seq = lrn->lrn_seq;
5933                         key->f_oid = lrn->lrn_first_oid;
5934                         key->f_ver = 0;
5935                 }
5936         } else {
5937                 key->f_oid++;
5938                 if (unlikely(key->f_oid == 0)) {
5939                         key->f_seq++;
5940                         it->loi_lrn = NULL;
5941                         goto again0;
5942                 }
5943
5944                 if (key->f_oid >=
5945                     lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
5946                         it->loi_lrn = NULL;
5947                         goto again0;
5948                 }
5949         }
5950
5951         if (unlikely(atomic_read(&lrn->lrn_known_count) <=
5952                      atomic_read(&lrn->lrn_accessed_count))) {
5953                 struct rb_node *next = rb_next(&lrn->lrn_node);
5954
5955                 while (next != NULL) {
5956                         lrn = rb_entry(next, struct lfsck_rbtree_node,
5957                                        lrn_node);
5958                         if (atomic_read(&lrn->lrn_known_count) >
5959                             atomic_read(&lrn->lrn_accessed_count))
5960                                 break;
5961                         next = rb_next(next);
5962                 }
5963
5964                 if (next == NULL) {
5965                         it->loi_over = 1;
5966                         RETURN(1);
5967                 }
5968
5969                 it->loi_lrn = lrn;
5970                 key->f_seq = lrn->lrn_seq;
5971                 key->f_oid = lrn->lrn_first_oid;
5972                 key->f_ver = 0;
5973         }
5974
5975         pos = key->f_oid - lrn->lrn_first_oid;
5976
5977 again1:
5978         pos = find_next_bit(lrn->lrn_known_bitmap,
5979                             LFSCK_RBTREE_BITMAP_WIDTH, pos);
5980         if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
5981                 key->f_oid = lrn->lrn_first_oid + pos;
5982                 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
5983                         key->f_seq++;
5984                         key->f_oid = 0;
5985                 }
5986                 it->loi_lrn = NULL;
5987                 goto again0;
5988         }
5989
5990         if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
5991                 pos++;
5992                 goto again1;
5993         }
5994
5995         key->f_oid = lrn->lrn_first_oid + pos;
5996         obj = lfsck_object_find_bottom(env, lfsck, key);
5997         if (IS_ERR(obj)) {
5998                 rc = PTR_ERR(obj);
5999                 if (rc == -ENOENT) {
6000                         pos++;
6001                         goto again1;
6002                 }
6003                 RETURN(rc);
6004         }
6005
6006         dt_read_lock(env, obj, 0);
6007         if (dt_object_exists(obj) == 0 ||
6008             lfsck_is_dead_obj(obj)) {
6009                 dt_read_unlock(env, obj);
6010                 lfsck_object_put(env, obj);
6011                 pos++;
6012                 goto again1;
6013         }
6014
6015         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
6016         if (rc != 0)
6017                 GOTO(out, rc);
6018
6019         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, pfid, sizeof(*pfid)),
6020                           XATTR_NAME_FID, BYPASS_CAPA);
6021         if (rc == -ENODATA) {
6022                 /* For the pre-created OST-object, update the bitmap to avoid
6023                  * others LFSCK (second phase) iteration to touch it again. */
6024                 if (la->la_ctime == 0) {
6025                         if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
6026                                 atomic_inc(&lrn->lrn_accessed_count);
6027
6028                         /* For the race between repairing dangling referenced
6029                          * MDT-object and unlink the file, it may left orphan
6030                          * OST-object there. Destroy it now! */
6031                         if (unlikely(!(la->la_mode & S_ISUID))) {
6032                                 dt_read_unlock(env, obj);
6033                                 lfsck_layout_destroy_orphan(env, obj);
6034                                 lfsck_object_put(env, obj);
6035                                 pos++;
6036                                 goto again1;
6037                         }
6038                 } else if (idx == 0) {
6039                         /* If the orphan OST-object has no parent information,
6040                          * regard it as referenced by the MDT-object on MDT0. */
6041                         fid_zero(&rec->lor_fid);
6042                         rec->lor_uid = la->la_uid;
6043                         rec->lor_gid = la->la_gid;
6044                         GOTO(out, rc = 0);
6045                 }
6046
6047                 dt_read_unlock(env, obj);
6048                 lfsck_object_put(env, obj);
6049                 pos++;
6050                 goto again1;
6051         }
6052
6053         if (rc < 0)
6054                 GOTO(out, rc);
6055
6056         if (rc != sizeof(struct filter_fid) &&
6057             rc != sizeof(struct filter_fid_old))
6058                 GOTO(out, rc = -EINVAL);
6059
6060         fid_le_to_cpu(&rec->lor_fid, &pfid->ff_parent);
6061         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
6062          * MDT-object's FID::f_ver, instead it is the OST-object index in its
6063          * parent MDT-object's layout EA. */
6064         save = rec->lor_fid.f_stripe_idx;
6065         rec->lor_fid.f_ver = 0;
6066         rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_fid, idx);
6067         /* If the orphan OST-object does not claim the MDT, then next.
6068          *
6069          * If we do not know whether it matches or not, then return it
6070          * to the MDT for further check. */
6071         if (rc == 0) {
6072                 dt_read_unlock(env, obj);
6073                 lfsck_object_put(env, obj);
6074                 pos++;
6075                 goto again1;
6076         }
6077
6078         rec->lor_fid.f_stripe_idx = save;
6079         rec->lor_uid = la->la_uid;
6080         rec->lor_gid = la->la_gid;
6081
6082         CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u\n",
6083                lfsck_lfsck2name(com->lc_lfsck), PFID(key), PFID(&rec->lor_fid),
6084                rec->lor_uid, rec->lor_gid);
6085
6086         GOTO(out, rc = 0);
6087
6088 out:
6089         dt_read_unlock(env, obj);
6090         lfsck_object_put(env, obj);
6091         if (rc == 0)
6092                 it->loi_hash++;
6093
6094         return rc;
6095 }
6096
6097 /**
6098  * \retval       +1: locate to the exactly position
6099  * \retval        0: cannot locate to the exactly position,
6100  *                   call next() to move to a valid position.
6101  * \retval      -ve: on error
6102  */
6103 static int lfsck_orphan_it_get(const struct lu_env *env,
6104                                struct dt_it *di,
6105                                const struct dt_key *key)
6106 {
6107         struct lfsck_orphan_it  *it   = (struct lfsck_orphan_it *)di;
6108         int                      rc;
6109
6110         it->loi_key = *(struct lu_fid *)key;
6111         rc = lfsck_orphan_it_next(env, di);
6112         if (rc == 1)
6113                 return 0;
6114
6115         if (rc == 0)
6116                 return 1;
6117
6118         return rc;
6119 }
6120
6121 static void lfsck_orphan_it_put(const struct lu_env *env,
6122                                 struct dt_it *di)
6123 {
6124 }
6125
6126 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
6127                                           const struct dt_it *di)
6128 {
6129         struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6130
6131         return (struct dt_key *)&it->loi_key;
6132 }
6133
6134 static int lfsck_orphan_it_key_size(const struct lu_env *env,
6135                                     const struct dt_it *di)
6136 {
6137         return sizeof(struct lu_fid);
6138 }
6139
6140 static int lfsck_orphan_it_rec(const struct lu_env *env,
6141                                const struct dt_it *di,
6142                                struct dt_rec *rec,
6143                                __u32 attr)
6144 {
6145         struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6146
6147         *(struct lu_orphan_rec *)rec = it->loi_rec;
6148
6149         return 0;
6150 }
6151
6152 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
6153                                    const struct dt_it *di)
6154 {
6155         struct lfsck_orphan_it  *it   = (struct lfsck_orphan_it *)di;
6156
6157         return it->loi_hash;
6158 }
6159
6160 /**
6161  * \retval       +1: locate to the exactly position
6162  * \retval        0: cannot locate to the exactly position,
6163  *                   call next() to move to a valid position.
6164  * \retval      -ve: on error
6165  */
6166 static int lfsck_orphan_it_load(const struct lu_env *env,
6167                                 const struct dt_it *di,
6168                                 __u64 hash)
6169 {
6170         struct lfsck_orphan_it           *it   = (struct lfsck_orphan_it *)di;
6171         struct lfsck_layout_slave_target *llst = it->loi_llst;
6172         int                               rc;
6173
6174         LASSERT(llst != NULL);
6175
6176         if (hash != llst->llst_hash) {
6177                 CDEBUG(D_LFSCK, "%s: the given hash "LPU64" for orphan "
6178                        "iteration does not match the one when fini "
6179                        LPU64", to be reset.\n",
6180                        lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
6181                        llst->llst_hash);
6182                 fid_zero(&llst->llst_fid);
6183                 llst->llst_hash = 0;
6184         }
6185
6186         it->loi_key = llst->llst_fid;
6187         it->loi_hash = llst->llst_hash;
6188         rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
6189         if (rc == 1)
6190                 return 0;
6191
6192         if (rc == 0)
6193                 return 1;
6194
6195         return rc;
6196 }
6197
6198 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
6199                                    const struct dt_it *di,
6200                                    void *key_rec)
6201 {
6202         return 0;
6203 }
6204
6205 const struct dt_index_operations lfsck_orphan_index_ops = {
6206         .dio_lookup             = lfsck_orphan_index_lookup,
6207         .dio_declare_insert     = lfsck_orphan_index_declare_insert,
6208         .dio_insert             = lfsck_orphan_index_insert,
6209         .dio_declare_delete     = lfsck_orphan_index_declare_delete,
6210         .dio_delete             = lfsck_orphan_index_delete,
6211         .dio_it = {
6212                 .init           = lfsck_orphan_it_init,
6213                 .fini           = lfsck_orphan_it_fini,
6214                 .get            = lfsck_orphan_it_get,
6215                 .put            = lfsck_orphan_it_put,
6216                 .next           = lfsck_orphan_it_next,
6217                 .key            = lfsck_orphan_it_key,
6218                 .key_size       = lfsck_orphan_it_key_size,
6219                 .rec            = lfsck_orphan_it_rec,
6220                 .store          = lfsck_orphan_it_store,
6221                 .load           = lfsck_orphan_it_load,
6222                 .key_rec        = lfsck_orphan_it_key_rec,
6223         }
6224 };