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         /* How many times we have failed to get the master status. */
  76         int                     llst_failures;
  77 };
  78
  79 struct lfsck_layout_slave_data {
  80         /* list for lfsck_layout_seq */
  81         struct list_head         llsd_seq_list;
  82
  83         /* list for the masters involve layout verification. */
  84         struct list_head         llsd_master_list;
  85         spinlock_t               llsd_lock;
  86         __u64                    llsd_touch_gen;
  87         struct dt_object        *llsd_rb_obj;
  88         struct rb_root           llsd_rb_root;
  89         rwlock_t                 llsd_rb_lock;
  90         unsigned int             llsd_rbtree_valid:1;
  91 };
  92
  93 struct lfsck_layout_slave_async_args {
  94         struct obd_export                *llsaa_exp;
  95         struct lfsck_component           *llsaa_com;
  96         struct lfsck_layout_slave_target *llsaa_llst;
  97 };
  98
  99 static inline void
 100 lfsck_layout_llst_put(struct lfsck_layout_slave_target *llst)
 101 {
 102         if (atomic_dec_and_test(&llst->llst_ref)) {
 103                 LASSERT(list_empty(&llst->llst_list));
 104
 105                 OBD_FREE_PTR(llst);
 106         }
 107 }
 108
 109 static inline int
 110 lfsck_layout_llst_add(struct lfsck_layout_slave_data *llsd, __u32 index)
 111 {
 112         struct lfsck_layout_slave_target *llst;
 113         struct lfsck_layout_slave_target *tmp;
 114         int                               rc   = 0;
 115
 116         OBD_ALLOC_PTR(llst);
 117         if (llst == NULL)
 118                 return -ENOMEM;
 119
 120         INIT_LIST_HEAD(&llst->llst_list);
 121         llst->llst_gen = 0;
 122         llst->llst_index = index;
 123         atomic_set(&llst->llst_ref, 1);
 124
 125         spin_lock(&llsd->llsd_lock);
 126         list_for_each_entry(tmp, &llsd->llsd_master_list, llst_list) {
 127                 if (tmp->llst_index == index) {
 128                         rc = -EALREADY;
 129                         break;
 130                 }
 131         }
 132         if (rc == 0)
 133                 list_add_tail(&llst->llst_list, &llsd->llsd_master_list);
 134         spin_unlock(&llsd->llsd_lock);
 135
 136         if (rc != 0)
 137                 OBD_FREE_PTR(llst);
 138
 139         return rc;
 140 }
 141
 142 static inline void
 143 lfsck_layout_llst_del(struct lfsck_layout_slave_data *llsd,
 144                       struct lfsck_layout_slave_target *llst)
 145 {
 146         bool del = false;
 147
 148         spin_lock(&llsd->llsd_lock);
 149         if (!list_empty(&llst->llst_list)) {
 150                 list_del_init(&llst->llst_list);
 151                 del = true;
 152         }
 153         spin_unlock(&llsd->llsd_lock);
 154
 155         if (del)
 156                 lfsck_layout_llst_put(llst);
 157 }
 158
 159 static inline struct lfsck_layout_slave_target *
 160 lfsck_layout_llst_find_and_del(struct lfsck_layout_slave_data *llsd,
 161                                __u32 index, bool unlink)
 162 {
 163         struct lfsck_layout_slave_target *llst;
 164
 165         spin_lock(&llsd->llsd_lock);
 166         list_for_each_entry(llst, &llsd->llsd_master_list, llst_list) {
 167                 if (llst->llst_index == index) {
 168                         if (unlink)
 169                                 list_del_init(&llst->llst_list);
 170                         else
 171                                 atomic_inc(&llst->llst_ref);
 172                         spin_unlock(&llsd->llsd_lock);
 173
 174                         return llst;
 175                 }
 176         }
 177         spin_unlock(&llsd->llsd_lock);
 178
 179         return NULL;
 180 }
 181
 182 static struct lfsck_layout_req *
 183 lfsck_layout_assistant_req_init(struct lfsck_assistant_object *lso,
 184                                 struct dt_object *child, __u32 ost_idx,
 185                                 __u32 lov_idx)
 186 {
 187         struct lfsck_layout_req *llr;
 188
 189         OBD_ALLOC_PTR(llr);
 190         if (llr == NULL)
 191                 return ERR_PTR(-ENOMEM);
 192
 193         INIT_LIST_HEAD(&llr->llr_lar.lar_list);
 194         llr->llr_lar.lar_parent = lfsck_assistant_object_get(lso);
 195         llr->llr_child = child;
 196         llr->llr_ost_idx = ost_idx;
 197         llr->llr_lov_idx = lov_idx;
 198
 199         return llr;
 200 }
 201
 202 static void lfsck_layout_assistant_req_fini(const struct lu_env *env,
 203                                             struct lfsck_assistant_req *lar)
 204 {
 205         struct lfsck_layout_req *llr =
 206                         container_of0(lar, struct lfsck_layout_req, llr_lar);
 207
 208         lfsck_object_put(env, llr->llr_child);
 209         lfsck_assistant_object_put(env, lar->lar_parent);
 210         OBD_FREE_PTR(llr);
 211 }
 212
 213 static int
 214 lfsck_layout_assistant_sync_failures_interpret(const struct lu_env *env,
 215                                                struct ptlrpc_request *req,
 216                                                void *args, int rc)
 217 {
 218         if (rc == 0) {
 219                 struct lfsck_async_interpret_args *laia = args;
 220                 struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
 221
 222                 ltd->ltd_synced_failures = 1;
 223                 atomic_dec(laia->laia_count);
 224         }
 225
 226         return 0;
 227 }
 228
 229 /**
 230  * Notify remote LFSCK instances about former failures.
 231  *
 232  * The local LFSCK instance has recorded which OSTs have ever failed to respond
 233  * some LFSCK verification requests (maybe because of network issues or the OST
 234  * itself trouble). During the respond gap, the OST may missed some OST-objects
 235  * verification, then the OST cannot know whether related OST-objects have been
 236  * referenced by related MDT-objects or not, then in the second-stage scanning,
 237  * these OST-objects will be regarded as orphan, if the OST-object contains bad
 238  * parent FID for back reference, then it will misguide the LFSCK to make wrong
 239  * fixing for the fake orphan.
 240  *
 241  * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
 242  * it will scan the bitmap for the ever failed OSTs, and notify them that they
 243  * have ever missed some OST-object verification and should skip the handling
 244  * for orphan OST-objects on all MDTs that are in the layout LFSCK.
 245  *
 246  * \param[in] env       pointer to the thread context
 247  * \param[in] com       pointer to the lfsck component
 248  * \param[in] lr        pointer to the lfsck request
 249  */
 250 static void lfsck_layout_assistant_sync_failures(const struct lu_env *env,
 251                                                  struct lfsck_component *com,
 252                                                  struct lfsck_request *lr)
 253 {
 254         struct lfsck_async_interpret_args *laia  =
 255                                 &lfsck_env_info(env)->lti_laia2;
 256         struct lfsck_assistant_data       *lad   = com->lc_data;
 257         struct lfsck_layout               *lo    = com->lc_file_ram;
 258         struct lfsck_instance             *lfsck = com->lc_lfsck;
 259         struct lfsck_tgt_descs            *ltds  = &lfsck->li_ost_descs;
 260         struct lfsck_tgt_desc             *ltd;
 261         struct ptlrpc_request_set         *set;
 262         atomic_t                           count;
 263         __u32                              idx;
 264         int                                rc    = 0;
 265         ENTRY;
 266
 267         if (!lad->lad_incomplete || lo->ll_flags & LF_INCOMPLETE)
 268                 RETURN_EXIT;
 269
 270         /* If the MDT has ever failed to verfiy some OST-objects,
 271          * then sync failures with them firstly. */
 272         lr->lr_flags2 = lo->ll_flags | LF_INCOMPLETE;
 273
 274         atomic_set(&count, 0);
 275         memset(laia, 0, sizeof(*laia));
 276         laia->laia_count = &count;
 277         set = ptlrpc_prep_set();
 278         if (set == NULL)
 279                 GOTO(out, rc = -ENOMEM);
 280
 281         down_read(&ltds->ltd_rw_sem);
 282         cfs_foreach_bit(lad->lad_bitmap, idx) {
 283                 ltd = lfsck_ltd2tgt(ltds, idx);
 284                 LASSERT(ltd != NULL);
 285
 286                 laia->laia_ltd = ltd;
 287                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
 288                                 lfsck_layout_assistant_sync_failures_interpret,
 289                                 laia, LFSCK_NOTIFY);
 290                 if (rc != 0) {
 291                         CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
 292                                "notify target %x for %s phase1 done: "
 293                                "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
 294                                ltd->ltd_index, lad->lad_name, rc);
 295
 296                         break;
 297                 }
 298
 299                 atomic_inc(&count);
 300         }
 301         up_read(&ltds->ltd_rw_sem);
 302
 303         if (rc == 0 && atomic_read(&count) > 0)
 304                 rc = ptlrpc_set_wait(set);
 305
 306         ptlrpc_set_destroy(set);
 307
 308         if (rc == 0 && atomic_read(&count) > 0)
 309                 rc = -EINVAL;
 310
 311         GOTO(out, rc);
 312
 313 out:
 314         if (rc != 0)
 315                 /* If failed to sync failures with the OSTs, then have to
 316                  * mark the whole LFSCK as LF_INCOMPLETE to skip the whole
 317                  * subsequent orphan OST-object handling. */
 318                 lo->ll_flags |= LF_INCOMPLETE;
 319
 320         lr->lr_flags2 = lo->ll_flags;
 321 }
 322
 323 static int lfsck_layout_get_lovea(const struct lu_env *env,
 324                                   struct dt_object *obj, struct lu_buf *buf)
 325 {
 326         int rc;
 327
 328 again:
 329         rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LOV);
 330         if (rc == -ERANGE) {
 331                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV);
 332                 if (rc <= 0)
 333                         return rc;
 334
 335                 lu_buf_realloc(buf, rc);
 336                 if (buf->lb_buf == NULL)
 337                         return -ENOMEM;
 338
 339                 goto again;
 340         }
 341
 342         if (rc == -ENODATA)
 343                 rc = 0;
 344
 345         if (rc <= 0)
 346                 return rc;
 347
 348         if (unlikely(buf->lb_buf == NULL)) {
 349                 lu_buf_alloc(buf, rc);
 350                 if (buf->lb_buf == NULL)
 351                         return -ENOMEM;
 352
 353                 goto again;
 354         }
 355
 356         return rc;
 357 }
 358
 359 static int lfsck_layout_verify_header(struct lov_mds_md_v1 *lmm)
 360 {
 361         __u32 magic;
 362         __u32 pattern;
 363
 364         magic = le32_to_cpu(lmm->lmm_magic);
 365         /* If magic crashed, keep it there. Sometime later, during OST-object
 366          * orphan handling, if some OST-object(s) back-point to it, it can be
 367          * verified and repaired. */
 368         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) {
 369                 struct ost_id   oi;
 370                 int             rc;
 371
 372                 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
 373                 if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
 374                         rc = -EOPNOTSUPP;
 375                 else
 376                         rc = -EINVAL;
 377
 378                 CDEBUG(D_LFSCK, "%s LOV EA magic %u on "DOSTID"\n",
 379                        rc == -EINVAL ? "Unknown" : "Unsupported",
 380                        magic, POSTID(&oi));
 381
 382                 return rc;
 383         }
 384
 385         pattern = le32_to_cpu(lmm->lmm_pattern);
 386         /* XXX: currently, we only support LOV_PATTERN_RAID0. */
 387         if (lov_pattern(pattern) != LOV_PATTERN_RAID0) {
 388                 struct ost_id oi;
 389
 390                 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
 391                 CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u on "DOSTID"\n",
 392                        pattern, POSTID(&oi));
 393
 394                 return -EOPNOTSUPP;
 395         }
 396
 397         return 0;
 398 }
 399
 400 #define LFSCK_RBTREE_BITMAP_SIZE        PAGE_CACHE_SIZE
 401 #define LFSCK_RBTREE_BITMAP_WIDTH       (LFSCK_RBTREE_BITMAP_SIZE << 3)
 402 #define LFSCK_RBTREE_BITMAP_MASK        (LFSCK_RBTREE_BITMAP_WIDTH - 1)
 403
 404 struct lfsck_rbtree_node {
 405         struct rb_node   lrn_node;
 406         __u64            lrn_seq;
 407         __u32            lrn_first_oid;
 408         atomic_t         lrn_known_count;
 409         atomic_t         lrn_accessed_count;
 410         void            *lrn_known_bitmap;
 411         void            *lrn_accessed_bitmap;
 412 };
 413
 414 static inline int lfsck_rbtree_cmp(struct lfsck_rbtree_node *lrn,
 415                                    __u64 seq, __u32 oid)
 416 {
 417         if (seq < lrn->lrn_seq)
 418                 return -1;
 419
 420         if (seq > lrn->lrn_seq)
 421                 return 1;
 422
 423         if (oid < lrn->lrn_first_oid)
 424                 return -1;
 425
 426         if (oid - lrn->lrn_first_oid >= LFSCK_RBTREE_BITMAP_WIDTH)
 427                 return 1;
 428
 429         return 0;
 430 }
 431
 432 /* The caller should hold llsd->llsd_rb_lock. */
 433 static struct lfsck_rbtree_node *
 434 lfsck_rbtree_search(struct lfsck_layout_slave_data *llsd,
 435                     const struct lu_fid *fid, bool *exact)
 436 {
 437         struct rb_node           *node  = llsd->llsd_rb_root.rb_node;
 438         struct rb_node           *prev  = NULL;
 439         struct lfsck_rbtree_node *lrn   = NULL;
 440         int                       rc    = 0;
 441
 442         if (exact != NULL)
 443                 *exact = true;
 444
 445         while (node != NULL) {
 446                 prev = node;
 447                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 448                 rc = lfsck_rbtree_cmp(lrn, fid_seq(fid), fid_oid(fid));
 449                 if (rc < 0)
 450                         node = node->rb_left;
 451                 else if (rc > 0)
 452                         node = node->rb_right;
 453                 else
 454                         return lrn;
 455         }
 456
 457         if (exact == NULL)
 458                 return NULL;
 459
 460         /* If there is no exactly matched one, then to the next valid one. */
 461         *exact = false;
 462
 463         /* The rbtree is empty. */
 464         if (rc == 0)
 465                 return NULL;
 466
 467         if (rc < 0)
 468                 return lrn;
 469
 470         node = rb_next(prev);
 471
 472         /* The end of the rbtree. */
 473         if (node == NULL)
 474                 return NULL;
 475
 476         lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 477
 478         return lrn;
 479 }
 480
 481 static struct lfsck_rbtree_node *lfsck_rbtree_new(const struct lu_env *env,
 482                                                   const struct lu_fid *fid)
 483 {
 484         struct lfsck_rbtree_node *lrn;
 485
 486         OBD_ALLOC_PTR(lrn);
 487         if (lrn == NULL)
 488                 return ERR_PTR(-ENOMEM);
 489
 490         OBD_ALLOC(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 491         if (lrn->lrn_known_bitmap == NULL) {
 492                 OBD_FREE_PTR(lrn);
 493
 494                 return ERR_PTR(-ENOMEM);
 495         }
 496
 497         OBD_ALLOC(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 498         if (lrn->lrn_accessed_bitmap == NULL) {
 499                 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 500                 OBD_FREE_PTR(lrn);
 501
 502                 return ERR_PTR(-ENOMEM);
 503         }
 504
 505         RB_CLEAR_NODE(&lrn->lrn_node);
 506         lrn->lrn_seq = fid_seq(fid);
 507         lrn->lrn_first_oid = fid_oid(fid) & ~LFSCK_RBTREE_BITMAP_MASK;
 508         atomic_set(&lrn->lrn_known_count, 0);
 509         atomic_set(&lrn->lrn_accessed_count, 0);
 510
 511         return lrn;
 512 }
 513
 514 static void lfsck_rbtree_free(struct lfsck_rbtree_node *lrn)
 515 {
 516         OBD_FREE(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 517         OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 518         OBD_FREE_PTR(lrn);
 519 }
 520
 521 /* The caller should hold lock. */
 522 static struct lfsck_rbtree_node *
 523 lfsck_rbtree_insert(struct lfsck_layout_slave_data *llsd,
 524                     struct lfsck_rbtree_node *lrn)
 525 {
 526         struct rb_node           **pos    = &llsd->llsd_rb_root.rb_node;
 527         struct rb_node            *parent = NULL;
 528         struct lfsck_rbtree_node  *tmp;
 529         int                        rc;
 530
 531         while (*pos != NULL) {
 532                 parent = *pos;
 533                 tmp = rb_entry(parent, struct lfsck_rbtree_node, lrn_node);
 534                 rc = lfsck_rbtree_cmp(tmp, lrn->lrn_seq, lrn->lrn_first_oid);
 535                 if (rc < 0)
 536                         pos = &(*pos)->rb_left;
 537                 else if (rc > 0)
 538                         pos = &(*pos)->rb_right;
 539                 else
 540                         return tmp;
 541         }
 542
 543         rb_link_node(&lrn->lrn_node, parent, pos);
 544         rb_insert_color(&lrn->lrn_node, &llsd->llsd_rb_root);
 545
 546         return lrn;
 547 }
 548
 549 extern const struct dt_index_operations lfsck_orphan_index_ops;
 550
 551 static int lfsck_rbtree_setup(const struct lu_env *env,
 552                               struct lfsck_component *com)
 553 {
 554         struct lu_fid                   *fid    = &lfsck_env_info(env)->lti_fid;
 555         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 556         struct dt_device                *dev    = lfsck->li_bottom;
 557         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
 558         struct dt_object                *obj;
 559
 560         fid->f_seq = FID_SEQ_LAYOUT_RBTREE;
 561         fid->f_oid = lfsck_dev_idx(lfsck);
 562         fid->f_ver = 0;
 563         obj = dt_locate(env, dev, fid);
 564         if (IS_ERR(obj))
 565                 RETURN(PTR_ERR(obj));
 566
 567         /* Generate an in-RAM object to stand for the layout rbtree.
 568          * Scanning the layout rbtree will be via the iteration over
 569          * the object. In the future, the rbtree may be written onto
 570          * disk with the object.
 571          *
 572          * Mark the object to be as exist. */
 573         obj->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 574         obj->do_index_ops = &lfsck_orphan_index_ops;
 575         llsd->llsd_rb_obj = obj;
 576         llsd->llsd_rbtree_valid = 1;
 577         dev->dd_record_fid_accessed = 1;
 578
 579         CDEBUG(D_LFSCK, "%s: layout LFSCK init OST-objects accessing bitmap\n",
 580                lfsck_lfsck2name(lfsck));
 581
 582         return 0;
 583 }
 584
 585 static void lfsck_rbtree_cleanup(const struct lu_env *env,
 586                                  struct lfsck_component *com)
 587 {
 588         struct lfsck_instance           *lfsck = com->lc_lfsck;
 589         struct lfsck_layout_slave_data  *llsd  = com->lc_data;
 590         struct rb_node                  *node  = rb_first(&llsd->llsd_rb_root);
 591         struct rb_node                  *next;
 592         struct lfsck_rbtree_node        *lrn;
 593
 594         lfsck->li_bottom->dd_record_fid_accessed = 0;
 595         /* Invalid the rbtree, then no others will use it. */
 596         write_lock(&llsd->llsd_rb_lock);
 597         llsd->llsd_rbtree_valid = 0;
 598         write_unlock(&llsd->llsd_rb_lock);
 599
 600         while (node != NULL) {
 601                 next = rb_next(node);
 602                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 603                 rb_erase(node, &llsd->llsd_rb_root);
 604                 lfsck_rbtree_free(lrn);
 605                 node = next;
 606         }
 607
 608         if (llsd->llsd_rb_obj != NULL) {
 609                 lfsck_object_put(env, llsd->llsd_rb_obj);
 610                 llsd->llsd_rb_obj = NULL;
 611         }
 612
 613         CDEBUG(D_LFSCK, "%s: layout LFSCK fini OST-objects accessing bitmap\n",
 614                lfsck_lfsck2name(lfsck));
 615 }
 616
 617 static void lfsck_rbtree_update_bitmap(const struct lu_env *env,
 618                                        struct lfsck_component *com,
 619                                        const struct lu_fid *fid,
 620                                        bool accessed)
 621 {
 622         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
 623         struct lfsck_rbtree_node        *lrn;
 624         bool                             insert = false;
 625         int                              idx;
 626         int                              rc     = 0;
 627         ENTRY;
 628
 629         if (unlikely(!fid_is_sane(fid) || fid_is_last_id(fid)))
 630                 RETURN_EXIT;
 631
 632         if (!fid_is_idif(fid) && !fid_is_norm(fid))
 633                 RETURN_EXIT;
 634
 635         read_lock(&llsd->llsd_rb_lock);
 636         if (!llsd->llsd_rbtree_valid)
 637                 GOTO(unlock, rc = 0);
 638
 639         lrn = lfsck_rbtree_search(llsd, fid, NULL);
 640         if (lrn == NULL) {
 641                 struct lfsck_rbtree_node *tmp;
 642
 643                 LASSERT(!insert);
 644
 645                 read_unlock(&llsd->llsd_rb_lock);
 646                 tmp = lfsck_rbtree_new(env, fid);
 647                 if (IS_ERR(tmp))
 648                         GOTO(out, rc = PTR_ERR(tmp));
 649
 650                 insert = true;
 651                 write_lock(&llsd->llsd_rb_lock);
 652                 if (!llsd->llsd_rbtree_valid) {
 653                         lfsck_rbtree_free(tmp);
 654                         GOTO(unlock, rc = 0);
 655                 }
 656
 657                 lrn = lfsck_rbtree_insert(llsd, tmp);
 658                 if (lrn != tmp)
 659                         lfsck_rbtree_free(tmp);
 660         }
 661
 662         idx = fid_oid(fid) & LFSCK_RBTREE_BITMAP_MASK;
 663         /* Any accessed object must be a known object. */
 664         if (!test_and_set_bit(idx, lrn->lrn_known_bitmap))
 665                 atomic_inc(&lrn->lrn_known_count);
 666         if (accessed && !test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
 667                 atomic_inc(&lrn->lrn_accessed_count);
 668
 669         GOTO(unlock, rc = 0);
 670
 671 unlock:
 672         if (insert)
 673                 write_unlock(&llsd->llsd_rb_lock);
 674         else
 675                 read_unlock(&llsd->llsd_rb_lock);
 676 out:
 677         if (rc != 0 && accessed) {
 678                 struct lfsck_layout *lo = com->lc_file_ram;
 679
 680                 CDEBUG(D_LFSCK, "%s: fail to update OST-objects accessing "
 681                        "bitmap, and will cause incorrect LFSCK OST-object "
 682                        "handling, so disable it to cancel orphan handling "
 683                        "for related device. rc = %d\n",
 684                        lfsck_lfsck2name(com->lc_lfsck), rc);
 685
 686                 lo->ll_flags |= LF_INCOMPLETE;
 687                 lfsck_rbtree_cleanup(env, com);
 688         }
 689 }
 690
 691 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
 692                                    const struct lfsck_layout *src)
 693 {
 694         int i;
 695
 696         des->ll_magic = le32_to_cpu(src->ll_magic);
 697         des->ll_status = le32_to_cpu(src->ll_status);
 698         des->ll_flags = le32_to_cpu(src->ll_flags);
 699         des->ll_success_count = le32_to_cpu(src->ll_success_count);
 700         des->ll_run_time_phase1 = le32_to_cpu(src->ll_run_time_phase1);
 701         des->ll_run_time_phase2 = le32_to_cpu(src->ll_run_time_phase2);
 702         des->ll_time_last_complete = le64_to_cpu(src->ll_time_last_complete);
 703         des->ll_time_latest_start = le64_to_cpu(src->ll_time_latest_start);
 704         des->ll_time_last_checkpoint =
 705                                 le64_to_cpu(src->ll_time_last_checkpoint);
 706         des->ll_pos_latest_start = le64_to_cpu(src->ll_pos_latest_start);
 707         des->ll_pos_last_checkpoint = le64_to_cpu(src->ll_pos_last_checkpoint);
 708         des->ll_pos_first_inconsistent =
 709                         le64_to_cpu(src->ll_pos_first_inconsistent);
 710         des->ll_objs_checked_phase1 = le64_to_cpu(src->ll_objs_checked_phase1);
 711         des->ll_objs_failed_phase1 = le64_to_cpu(src->ll_objs_failed_phase1);
 712         des->ll_objs_checked_phase2 = le64_to_cpu(src->ll_objs_checked_phase2);
 713         des->ll_objs_failed_phase2 = le64_to_cpu(src->ll_objs_failed_phase2);
 714         for (i = 0; i < LLIT_MAX; i++)
 715                 des->ll_objs_repaired[i] =
 716                                 le64_to_cpu(src->ll_objs_repaired[i]);
 717         des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
 718         des->ll_bitmap_size = le32_to_cpu(src->ll_bitmap_size);
 719 }
 720
 721 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
 722                                    const struct lfsck_layout *src)
 723 {
 724         int i;
 725
 726         des->ll_magic = cpu_to_le32(src->ll_magic);
 727         des->ll_status = cpu_to_le32(src->ll_status);
 728         des->ll_flags = cpu_to_le32(src->ll_flags);
 729         des->ll_success_count = cpu_to_le32(src->ll_success_count);
 730         des->ll_run_time_phase1 = cpu_to_le32(src->ll_run_time_phase1);
 731         des->ll_run_time_phase2 = cpu_to_le32(src->ll_run_time_phase2);
 732         des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
 733         des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
 734         des->ll_time_last_checkpoint =
 735                                 cpu_to_le64(src->ll_time_last_checkpoint);
 736         des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
 737         des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
 738         des->ll_pos_first_inconsistent =
 739                         cpu_to_le64(src->ll_pos_first_inconsistent);
 740         des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
 741         des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
 742         des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
 743         des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
 744         for (i = 0; i < LLIT_MAX; i++)
 745                 des->ll_objs_repaired[i] =
 746                                 cpu_to_le64(src->ll_objs_repaired[i]);
 747         des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
 748         des->ll_bitmap_size = cpu_to_le32(src->ll_bitmap_size);
 749 }
 750
 751 /**
 752  * Load the OST bitmap from the lfsck_layout trace file.
 753  *
 754  * \param[in] env       pointer to the thread context
 755  * \param[in] com       pointer to the lfsck component
 756  *
 757  * \retval              0 for success
 758  * \retval              negative error number on failure or data corruption
 759  */
 760 static int lfsck_layout_load_bitmap(const struct lu_env *env,
 761                                     struct lfsck_component *com)
 762 {
 763         struct dt_object                *obj    = com->lc_obj;
 764         struct lfsck_assistant_data     *lad    = com->lc_data;
 765         struct lfsck_layout             *lo     = com->lc_file_ram;
 766         cfs_bitmap_t                    *bitmap = lad->lad_bitmap;
 767         loff_t                           pos    = com->lc_file_size;
 768         ssize_t                          size;
 769         __u32                            nbits;
 770         int                              rc;
 771         ENTRY;
 772
 773         if (com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size >
 774             lo->ll_bitmap_size)
 775                 nbits = com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size;
 776         else
 777                 nbits = lo->ll_bitmap_size;
 778
 779         if (unlikely(nbits < BITS_PER_LONG))
 780                 nbits = BITS_PER_LONG;
 781
 782         if (nbits > bitmap->size) {
 783                 __u32 new_bits = bitmap->size;
 784                 cfs_bitmap_t *new_bitmap;
 785
 786                 while (new_bits < nbits)
 787                         new_bits <<= 1;
 788
 789                 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
 790                 if (new_bitmap == NULL)
 791                         RETURN(-ENOMEM);
 792
 793                 lad->lad_bitmap = new_bitmap;
 794                 CFS_FREE_BITMAP(bitmap);
 795                 bitmap = new_bitmap;
 796         }
 797
 798         if (lo->ll_bitmap_size == 0) {
 799                 lad->lad_incomplete = 0;
 800                 CFS_RESET_BITMAP(bitmap);
 801
 802                 RETURN(0);
 803         }
 804
 805         size = (lo->ll_bitmap_size + 7) >> 3;
 806         rc = dt_read(env, obj, lfsck_buf_get(env, bitmap->data, size), &pos);
 807         if (rc != size)
 808                 RETURN(rc >= 0 ? -EINVAL : rc);
 809
 810         if (cfs_bitmap_check_empty(bitmap))
 811                 lad->lad_incomplete = 0;
 812         else
 813                 lad->lad_incomplete = 1;
 814
 815         RETURN(0);
 816 }
 817
 818 /**
 819  * Load the layout LFSCK trace file from disk.
 820  *
 821  * The layout LFSCK trace file records the layout LFSCK status information
 822  * and other statistics, such as how many objects have been scanned, and how
 823  * many objects have been repaired, and etc. It also contains the bitmap for
 824  * failed OSTs during the layout LFSCK. All these information will be loaded
 825  * from disk to RAM when the layout LFSCK component setup.
 826  *
 827  * \param[in] env       pointer to the thread context
 828  * \param[in] com       pointer to the lfsck component
 829  *
 830  * \retval              positive number for file data corruption, the caller
 831  *                      should reset the layout LFSCK trace file
 832  * \retval              0 for success
 833  * \retval              negative error number on failure
 834  */
 835 static int lfsck_layout_load(const struct lu_env *env,
 836                              struct lfsck_component *com)
 837 {
 838         struct lfsck_layout             *lo     = com->lc_file_ram;
 839         ssize_t                          size   = com->lc_file_size;
 840         loff_t                           pos    = 0;
 841         int                              rc;
 842
 843         rc = dt_read(env, com->lc_obj,
 844                      lfsck_buf_get(env, com->lc_file_disk, size), &pos);
 845         if (rc == 0) {
 846                 return -ENOENT;
 847         } else if (rc < 0) {
 848                 CDEBUG(D_LFSCK, "%s: failed to load lfsck_layout: rc = %d\n",
 849                        lfsck_lfsck2name(com->lc_lfsck), rc);
 850                 return rc;
 851         } else if (rc != size) {
 852                 CDEBUG(D_LFSCK, "%s: lfsck_layout size %u != %u; reset it\n",
 853                        lfsck_lfsck2name(com->lc_lfsck), rc, (unsigned int)size);
 854                 return 1;
 855         }
 856
 857         lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
 858         if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
 859                 CDEBUG(D_LFSCK, "%s: invalid lfsck_layout magic %#x != %#x, "
 860                        "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
 861                        lo->ll_magic, LFSCK_LAYOUT_MAGIC);
 862                 return 1;
 863         }
 864
 865         return 0;
 866 }
 867
 868 /**
 869  * Store the layout LFSCK trace file on disk.
 870  *
 871  * The layout LFSCK trace file records the layout LFSCK status information
 872  * and other statistics, such as how many objects have been scanned, and how
 873  * many objects have been repaired, and etc. It also contains the bitmap for
 874  * failed OSTs during the layout LFSCK. All these information will be synced
 875  * from RAM to disk periodically.
 876  *
 877  * \param[in] env       pointer to the thread context
 878  * \param[in] com       pointer to the lfsck component
 879  *
 880  * \retval              0 for success
 881  * \retval              negative error number on failure
 882  */
 883 static int lfsck_layout_store(const struct lu_env *env,
 884                               struct lfsck_component *com)
 885 {
 886         struct dt_object        *obj    = com->lc_obj;
 887         struct lfsck_instance   *lfsck  = com->lc_lfsck;
 888         struct lfsck_layout     *lo_ram = com->lc_file_ram;
 889         struct lfsck_layout     *lo     = com->lc_file_disk;
 890         struct thandle          *th;
 891         struct dt_device        *dev    = lfsck_obj2dev(obj);
 892         cfs_bitmap_t            *bitmap = NULL;
 893         loff_t                   pos;
 894         ssize_t                  size   = com->lc_file_size;
 895         __u32                    nbits  = 0;
 896         int                      rc;
 897         ENTRY;
 898
 899         if (lfsck->li_master) {
 900                 struct lfsck_assistant_data *lad = com->lc_data;
 901
 902                 bitmap = lad->lad_bitmap;
 903                 nbits = bitmap->size;
 904
 905                 LASSERT(nbits > 0);
 906                 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
 907         }
 908
 909         lo_ram->ll_bitmap_size = nbits;
 910         lfsck_layout_cpu_to_le(lo, lo_ram);
 911         th = dt_trans_create(env, dev);
 912         if (IS_ERR(th))
 913                 GOTO(log, rc = PTR_ERR(th));
 914
 915         rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
 916                                      (loff_t)0, th);
 917         if (rc != 0)
 918                 GOTO(out, rc);
 919
 920         if (bitmap != NULL) {
 921                 rc = dt_declare_record_write(env, obj,
 922                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
 923                                 (loff_t)size, th);
 924                 if (rc != 0)
 925                         GOTO(out, rc);
 926         }
 927
 928         rc = dt_trans_start_local(env, dev, th);
 929         if (rc != 0)
 930                 GOTO(out, rc);
 931
 932         pos = 0;
 933         rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos, th);
 934         if (rc != 0)
 935                 GOTO(out, rc);
 936
 937         if (bitmap != NULL) {
 938                 pos = size;
 939                 rc = dt_record_write(env, obj,
 940                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
 941                                 &pos, th);
 942         }
 943
 944         GOTO(out, rc);
 945
 946 out:
 947         dt_trans_stop(env, dev, th);
 948
 949 log:
 950         if (rc != 0)
 951                 CDEBUG(D_LFSCK, "%s: fail to store lfsck_layout: rc = %d\n",
 952                        lfsck_lfsck2name(lfsck), rc);
 953
 954         return rc;
 955 }
 956
 957 static int lfsck_layout_init(const struct lu_env *env,
 958                              struct lfsck_component *com)
 959 {
 960         struct lfsck_layout *lo = com->lc_file_ram;
 961         int rc;
 962
 963         memset(lo, 0, com->lc_file_size);
 964         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
 965         lo->ll_status = LS_INIT;
 966         down_write(&com->lc_sem);
 967         rc = lfsck_layout_store(env, com);
 968         up_write(&com->lc_sem);
 969
 970         return rc;
 971 }
 972
 973 static int fid_is_for_ostobj(const struct lu_env *env,
 974                              struct lfsck_instance *lfsck,
 975                              struct dt_object *obj, const struct lu_fid *fid)
 976 {
 977         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
 978         struct lu_seq_range     *range  = &lfsck_env_info(env)->lti_range;
 979         struct lustre_mdt_attrs *lma;
 980         int                      rc;
 981
 982         fld_range_set_any(range);
 983         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
 984         if (rc == 0) {
 985                 if (fld_range_is_ost(range))
 986                         return 1;
 987
 988                 return 0;
 989         }
 990
 991         lma = &lfsck_env_info(env)->lti_lma;
 992         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, lma, sizeof(*lma)),
 993                           XATTR_NAME_LMA);
 994         if (rc == sizeof(*lma)) {
 995                 lustre_lma_swab(lma);
 996
 997                 return lma->lma_compat & LMAC_FID_ON_OST ? 1 : 0;
 998         }
 999
1000         rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID);
1001
1002         return rc > 0;
1003 }
1004
1005 static struct lfsck_layout_seq *
1006 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
1007 {
1008         struct lfsck_layout_seq *lls;
1009
1010         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1011                 if (lls->lls_seq == seq)
1012                         return lls;
1013
1014                 if (lls->lls_seq > seq)
1015                         return NULL;
1016         }
1017
1018         return NULL;
1019 }
1020
1021 static void
1022 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
1023                         struct lfsck_layout_seq *lls)
1024 {
1025         struct lfsck_layout_seq *tmp;
1026         struct list_head        *pos = &llsd->llsd_seq_list;
1027
1028         list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
1029                 if (lls->lls_seq < tmp->lls_seq) {
1030                         pos = &tmp->lls_list;
1031                         break;
1032                 }
1033         }
1034         list_add_tail(&lls->lls_list, pos);
1035 }
1036
1037 static int
1038 lfsck_layout_lastid_create(const struct lu_env *env,
1039                            struct lfsck_instance *lfsck,
1040                            struct dt_object *obj)
1041 {
1042         struct lfsck_thread_info *info   = lfsck_env_info(env);
1043         struct lu_attr           *la     = &info->lti_la;
1044         struct dt_object_format  *dof    = &info->lti_dof;
1045         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
1046         struct dt_device         *dt     = lfsck_obj2dev(obj);
1047         struct thandle           *th;
1048         __u64                     lastid = 0;
1049         loff_t                    pos    = 0;
1050         int                       rc;
1051         ENTRY;
1052
1053         if (bk->lb_param & LPF_DRYRUN)
1054                 return 0;
1055
1056         memset(la, 0, sizeof(*la));
1057         la->la_mode = S_IFREG |  S_IRUGO | S_IWUSR;
1058         la->la_valid = LA_MODE | LA_UID | LA_GID;
1059         memset(dof, 0, sizeof(*dof));
1060         dof->dof_type = dt_mode_to_dft(S_IFREG);
1061
1062         th = dt_trans_create(env, dt);
1063         if (IS_ERR(th))
1064                 GOTO(log, rc = PTR_ERR(th));
1065
1066         rc = dt_declare_create(env, obj, la, NULL, dof, th);
1067         if (rc != 0)
1068                 GOTO(stop, rc);
1069
1070         rc = dt_declare_record_write(env, obj,
1071                                      lfsck_buf_get(env, &lastid,
1072                                                    sizeof(lastid)),
1073                                      pos, th);
1074         if (rc != 0)
1075                 GOTO(stop, rc);
1076
1077         rc = dt_trans_start_local(env, dt, th);
1078         if (rc != 0)
1079                 GOTO(stop, rc);
1080
1081         dt_write_lock(env, obj, 0);
1082         if (likely(dt_object_exists(obj) == 0)) {
1083                 rc = dt_create(env, obj, la, NULL, dof, th);
1084                 if (rc == 0)
1085                         rc = dt_record_write(env, obj,
1086                                 lfsck_buf_get(env, &lastid, sizeof(lastid)),
1087                                 &pos, th);
1088         }
1089         dt_write_unlock(env, obj);
1090
1091         GOTO(stop, rc);
1092
1093 stop:
1094         dt_trans_stop(env, dt, th);
1095
1096 log:
1097         CDEBUG(D_LFSCK, "%s: layout LFSCK will create LAST_ID for <seq> "
1098                LPX64": rc = %d\n",
1099                lfsck_lfsck2name(lfsck), fid_seq(lfsck_dto2fid(obj)), rc);
1100
1101         return rc;
1102 }
1103
1104 static int
1105 lfsck_layout_lastid_reload(const struct lu_env *env,
1106                            struct lfsck_component *com,
1107                            struct lfsck_layout_seq *lls)
1108 {
1109         __u64   lastid;
1110         loff_t  pos     = 0;
1111         int     rc;
1112
1113         dt_read_lock(env, lls->lls_lastid_obj, 0);
1114         rc = dt_record_read(env, lls->lls_lastid_obj,
1115                             lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
1116         dt_read_unlock(env, lls->lls_lastid_obj);
1117         if (unlikely(rc != 0))
1118                 return rc;
1119
1120         lastid = le64_to_cpu(lastid);
1121         if (lastid < lls->lls_lastid_known) {
1122                 struct lfsck_instance   *lfsck  = com->lc_lfsck;
1123                 struct lfsck_layout     *lo     = com->lc_file_ram;
1124
1125                 lls->lls_lastid = lls->lls_lastid_known;
1126                 lls->lls_dirty = 1;
1127                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1128                         LASSERT(lfsck->li_out_notify != NULL);
1129
1130                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1131                                              LE_LASTID_REBUILDING);
1132                         lo->ll_flags |= LF_CRASHED_LASTID;
1133
1134                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
1135                                "LAST_ID file (1) for the sequence "LPX64
1136                                ", old value "LPU64", known value "LPU64"\n",
1137                                lfsck_lfsck2name(lfsck), lls->lls_seq,
1138                                lastid, lls->lls_lastid);
1139                 }
1140         } else if (lastid >= lls->lls_lastid) {
1141                 lls->lls_lastid = lastid;
1142                 lls->lls_dirty = 0;
1143         }
1144
1145         return 0;
1146 }
1147
1148 static int
1149 lfsck_layout_lastid_store(const struct lu_env *env,
1150                           struct lfsck_component *com)
1151 {
1152         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1153         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
1154         struct dt_device                *dt     = lfsck->li_bottom;
1155         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
1156         struct lfsck_layout_seq         *lls;
1157         struct thandle                  *th;
1158         __u64                            lastid;
1159         int                              rc     = 0;
1160         int                              rc1    = 0;
1161
1162         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1163                 loff_t pos = 0;
1164
1165                 if (!lls->lls_dirty)
1166                         continue;
1167
1168                 CDEBUG(D_LFSCK, "%s: layout LFSCK will sync the LAST_ID for "
1169                        "<seq> "LPX64" as <oid> "LPU64"\n",
1170                        lfsck_lfsck2name(lfsck), lls->lls_seq, lls->lls_lastid);
1171
1172                 if (bk->lb_param & LPF_DRYRUN) {
1173                         lls->lls_dirty = 0;
1174                         continue;
1175                 }
1176
1177                 th = dt_trans_create(env, dt);
1178                 if (IS_ERR(th)) {
1179                         rc1 = PTR_ERR(th);
1180                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1181                                "the LAST_ID for <seq> "LPX64"(1): rc = %d\n",
1182                                lfsck_lfsck2name(com->lc_lfsck),
1183                                lls->lls_seq, rc1);
1184                         continue;
1185                 }
1186
1187                 lastid = cpu_to_le64(lls->lls_lastid);
1188                 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1189                                              lfsck_buf_get(env, &lastid,
1190                                                            sizeof(lastid)),
1191                                              pos, th);
1192                 if (rc != 0)
1193                         goto stop;
1194
1195                 rc = dt_trans_start_local(env, dt, th);
1196                 if (rc != 0)
1197                         goto stop;
1198
1199                 dt_write_lock(env, lls->lls_lastid_obj, 0);
1200                 rc = dt_record_write(env, lls->lls_lastid_obj,
1201                                      lfsck_buf_get(env, &lastid,
1202                                      sizeof(lastid)), &pos, th);
1203                 dt_write_unlock(env, lls->lls_lastid_obj);
1204                 if (rc == 0)
1205                         lls->lls_dirty = 0;
1206
1207 stop:
1208                 dt_trans_stop(env, dt, th);
1209                 if (rc != 0) {
1210                         rc1 = rc;
1211                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1212                                "the LAST_ID for <seq> "LPX64"(2): rc = %d\n",
1213                                lfsck_lfsck2name(com->lc_lfsck),
1214                                lls->lls_seq, rc1);
1215                 }
1216         }
1217
1218         return rc1;
1219 }
1220
1221 static int
1222 lfsck_layout_lastid_load(const struct lu_env *env,
1223                          struct lfsck_component *com,
1224                          struct lfsck_layout_seq *lls)
1225 {
1226         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1227         struct lfsck_layout     *lo     = com->lc_file_ram;
1228         struct lu_fid           *fid    = &lfsck_env_info(env)->lti_fid;
1229         struct dt_object        *obj;
1230         loff_t                   pos    = 0;
1231         int                      rc;
1232         ENTRY;
1233
1234         lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck));
1235         obj = dt_locate(env, lfsck->li_bottom, fid);
1236         if (IS_ERR(obj))
1237                 RETURN(PTR_ERR(obj));
1238
1239         /* LAST_ID crashed, to be rebuilt */
1240         if (dt_object_exists(obj) == 0) {
1241                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1242                         LASSERT(lfsck->li_out_notify != NULL);
1243
1244                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1245                                              LE_LASTID_REBUILDING);
1246                         lo->ll_flags |= LF_CRASHED_LASTID;
1247
1248                         CDEBUG(D_LFSCK, "%s: layout LFSCK cannot find the "
1249                                "LAST_ID file for sequence "LPX64"\n",
1250                                lfsck_lfsck2name(lfsck), lls->lls_seq);
1251
1252                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1253                             cfs_fail_val > 0) {
1254                                 struct l_wait_info lwi = LWI_TIMEOUT(
1255                                                 cfs_time_seconds(cfs_fail_val),
1256                                                 NULL, NULL);
1257
1258                                 /* Some others may changed the cfs_fail_val
1259                                  * as zero after above check, re-check it for
1260                                  * sure to avoid falling into wait for ever. */
1261                                 if (likely(lwi.lwi_timeout > 0)) {
1262                                         struct ptlrpc_thread *thread =
1263                                                 &lfsck->li_thread;
1264
1265                                         up_write(&com->lc_sem);
1266                                         l_wait_event(thread->t_ctl_waitq,
1267                                                      !thread_is_running(thread),
1268                                                      &lwi);
1269                                         down_write(&com->lc_sem);
1270                                 }
1271                         }
1272                 }
1273
1274                 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1275         } else {
1276                 dt_read_lock(env, obj, 0);
1277                 rc = dt_read(env, obj,
1278                         lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1279                         &pos);
1280                 dt_read_unlock(env, obj);
1281                 if (rc != 0 && rc != sizeof(__u64))
1282                         GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1283
1284                 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1285                         LASSERT(lfsck->li_out_notify != NULL);
1286
1287                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1288                                              LE_LASTID_REBUILDING);
1289                         lo->ll_flags |= LF_CRASHED_LASTID;
1290
1291                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds invalid "
1292                                "LAST_ID file for the sequence "LPX64
1293                                ": rc = %d\n",
1294                                lfsck_lfsck2name(lfsck), lls->lls_seq, rc);
1295                 }
1296
1297                 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1298                 rc = 0;
1299         }
1300
1301         GOTO(out, rc);
1302
1303 out:
1304         if (rc != 0)
1305                 lfsck_object_put(env, obj);
1306         else
1307                 lls->lls_lastid_obj = obj;
1308
1309         return rc;
1310 }
1311
1312 static void lfsck_layout_record_failure(const struct lu_env *env,
1313                                         struct lfsck_instance *lfsck,
1314                                         struct lfsck_layout *lo)
1315 {
1316         __u64 cookie;
1317
1318         lo->ll_objs_failed_phase1++;
1319         cookie = lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
1320                                                         lfsck->li_di_oit);
1321         if (lo->ll_pos_first_inconsistent == 0 ||
1322             lo->ll_pos_first_inconsistent < cookie) {
1323                 lo->ll_pos_first_inconsistent = cookie;
1324
1325                 CDEBUG(D_LFSCK, "%s: layout LFSCK hit first non-repaired "
1326                        "inconsistency at the pos ["LPU64"]\n",
1327                        lfsck_lfsck2name(lfsck),
1328                        lo->ll_pos_first_inconsistent);
1329         }
1330 }
1331
1332 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1333                                            struct lfsck_component *com,
1334                                            int rc)
1335 {
1336         struct lfsck_instance   *lfsck = com->lc_lfsck;
1337         struct lfsck_layout     *lo    = com->lc_file_ram;
1338
1339         down_write(&com->lc_sem);
1340         lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1341                                   HALF_SEC - com->lc_time_last_checkpoint);
1342         lo->ll_time_last_checkpoint = cfs_time_current_sec();
1343         lo->ll_objs_checked_phase2 += com->lc_new_checked;
1344
1345         if (rc > 0) {
1346                 if (lo->ll_flags & LF_INCOMPLETE) {
1347                         lo->ll_status = LS_PARTIAL;
1348                 } else {
1349                         if (lfsck->li_master) {
1350                                 struct lfsck_assistant_data *lad = com->lc_data;
1351
1352                                 if (lad->lad_incomplete)
1353                                         lo->ll_status = LS_PARTIAL;
1354                                 else
1355                                         lo->ll_status = LS_COMPLETED;
1356                         } else {
1357                                 lo->ll_status = LS_COMPLETED;
1358                         }
1359                 }
1360                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
1361                         lo->ll_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
1362                 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1363                 lo->ll_success_count++;
1364         } else if (rc == 0) {
1365                 if (lfsck->li_status != 0)
1366                         lo->ll_status = lfsck->li_status;
1367                 else
1368                         lo->ll_status = LS_STOPPED;
1369         } else {
1370                 lo->ll_status = LS_FAILED;
1371         }
1372
1373         rc = lfsck_layout_store(env, com);
1374         up_write(&com->lc_sem);
1375
1376         return rc;
1377 }
1378
1379 static int lfsck_layout_trans_stop(const struct lu_env *env,
1380                                    struct dt_device *dev,
1381                                    struct thandle *handle, int result)
1382 {
1383         int rc;
1384
1385         /* XXX: If there is something worng or it needs to repair nothing,
1386          *      then notify the lower to stop the modification. Currently,
1387          *      we use th_result for such purpose, that may be replaced by
1388          *      some rollback mechanism in the future. */
1389         handle->th_result = result;
1390         rc = dt_trans_stop(env, dev, handle);
1391         if (result != 0)
1392                 return result > 0 ? 0 : result;
1393
1394         return rc == 0 ? 1 : rc;
1395 }
1396
1397 /**
1398  * Get the system default stripe size.
1399  *
1400  * \param[in] env       pointer to the thread context
1401  * \param[in] lfsck     pointer to the lfsck instance
1402  * \param[out] size     pointer to the default stripe size
1403  *
1404  * \retval              0 for success
1405  * \retval              negative error number on failure
1406  */
1407 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1408                                            struct lfsck_instance *lfsck,
1409                                            __u32 *size)
1410 {
1411         struct lov_user_md      *lum = &lfsck_env_info(env)->lti_lum;
1412         struct dt_object        *root;
1413         int                      rc;
1414
1415         root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1416         if (IS_ERR(root))
1417                 return PTR_ERR(root);
1418
1419         /* Get the default stripe size via xattr_get on the backend root. */
1420         rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1421                           XATTR_NAME_LOV);
1422         if (rc > 0) {
1423                 /* The lum->lmm_stripe_size is LE mode. The *size also
1424                  * should be LE mode. So it is unnecessary to convert. */
1425                 *size = lum->lmm_stripe_size;
1426                 rc = 0;
1427         } else if (unlikely(rc == 0)) {
1428                 rc = -EINVAL;
1429         }
1430
1431         lfsck_object_put(env, root);
1432
1433         return rc;
1434 }
1435
1436 /**
1437  * \retval       +1: repaired
1438  * \retval        0: did nothing
1439  * \retval      -ve: on error
1440  */
1441 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1442                                      struct thandle *handle,
1443                                      struct dt_object *parent,
1444                                      struct lu_fid *cfid,
1445                                      struct lu_buf *buf,
1446                                      struct lov_ost_data_v1 *slot,
1447                                      int fl, __u32 ost_idx)
1448 {
1449         struct ost_id           *oi     = &lfsck_env_info(env)->lti_oi;
1450         struct lov_mds_md_v1    *lmm    = buf->lb_buf;
1451         struct lu_buf            ea_buf;
1452         int                      rc;
1453         __u32                    magic;
1454         __u16                    count;
1455         ENTRY;
1456
1457         magic = le32_to_cpu(lmm->lmm_magic);
1458         count = le16_to_cpu(lmm->lmm_stripe_count);
1459
1460         fid_to_ostid(cfid, oi);
1461         ostid_cpu_to_le(oi, &slot->l_ost_oi);
1462         slot->l_ost_gen = cpu_to_le32(0);
1463         slot->l_ost_idx = cpu_to_le32(ost_idx);
1464
1465         if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE) {
1466                 struct lov_ost_data_v1 *objs;
1467                 int                     i;
1468
1469                 if (magic == LOV_MAGIC_V1)
1470                         objs = &lmm->lmm_objects[0];
1471                 else
1472                         objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1473                 for (i = 0; i < count; i++, objs++) {
1474                         if (objs != slot && lovea_slot_is_dummy(objs))
1475                                 break;
1476                 }
1477
1478                 /* If the @slot is the last dummy slot to be refilled,
1479                  * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1480                 if (i == count)
1481                         lmm->lmm_pattern &= ~cpu_to_le32(LOV_PATTERN_F_HOLE);
1482         }
1483
1484         lfsck_buf_init(&ea_buf, lmm, lov_mds_md_size(count, magic));
1485         rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle);
1486         if (rc == 0)
1487                 rc = 1;
1488
1489         RETURN(rc);
1490 }
1491
1492 /**
1493  * \retval       +1: repaired
1494  * \retval        0: did nothing
1495  * \retval      -ve: on error
1496  */
1497 static int lfsck_layout_extend_lovea(const struct lu_env *env,
1498                                      struct lfsck_instance *lfsck,
1499                                      struct thandle *handle,
1500                                      struct dt_object *parent,
1501                                      struct lu_fid *cfid,
1502                                      struct lu_buf *buf, int fl,
1503                                      __u32 ost_idx, __u32 ea_off, bool reset)
1504 {
1505         struct lov_mds_md_v1    *lmm    = buf->lb_buf;
1506         struct lov_ost_data_v1  *objs;
1507         int                      rc;
1508         __u16                    count;
1509         bool                     hole   = false;
1510         ENTRY;
1511
1512         if (fl == LU_XATTR_CREATE || reset) {
1513                 __u32 pattern = LOV_PATTERN_RAID0;
1514
1515                 count = ea_off + 1;
1516                 LASSERT(buf->lb_len >= lov_mds_md_size(count, LOV_MAGIC_V1));
1517
1518                 if (ea_off != 0 || reset) {
1519                         pattern |= LOV_PATTERN_F_HOLE;
1520                         hole = true;
1521                 }
1522
1523                 memset(lmm, 0, buf->lb_len);
1524                 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1525                 lmm->lmm_pattern = cpu_to_le32(pattern);
1526                 fid_to_lmm_oi(lfsck_dto2fid(parent), &lmm->lmm_oi);
1527                 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1528
1529                 rc = lfsck_layout_get_def_stripesize(env, lfsck,
1530                                                      &lmm->lmm_stripe_size);
1531                 if (rc != 0)
1532                         RETURN(rc);
1533
1534                 objs = &lmm->lmm_objects[ea_off];
1535         } else {
1536                 __u32   magic = le32_to_cpu(lmm->lmm_magic);
1537                 int     gap;
1538
1539                 count = le16_to_cpu(lmm->lmm_stripe_count);
1540                 if (magic == LOV_MAGIC_V1)
1541                         objs = &lmm->lmm_objects[count];
1542                 else
1543                         objs = &((struct lov_mds_md_v3 *)lmm)->
1544                                                         lmm_objects[count];
1545
1546                 gap = ea_off - count;
1547                 if (gap >= 0)
1548                         count = ea_off + 1;
1549                 LASSERT(buf->lb_len >= lov_mds_md_size(count, magic));
1550
1551                 if (gap > 0) {
1552                         memset(objs, 0, gap * sizeof(*objs));
1553                         lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
1554                         hole = true;
1555                 }
1556
1557                 lmm->lmm_layout_gen =
1558                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
1559                 objs += gap;
1560         }
1561
1562         lmm->lmm_stripe_count = cpu_to_le16(count);
1563         rc = lfsck_layout_refill_lovea(env, handle, parent, cfid, buf, objs,
1564                                        fl, ost_idx);
1565
1566         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
1567                DFID": parent "DFID", OST-index %u, stripe-index %u, fl %d, "
1568                "reset %s, %s LOV EA hole: rc = %d\n",
1569                lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1570                ost_idx, ea_off, fl, reset ? "yes" : "no",
1571                hole ? "with" : "without", rc);
1572
1573         RETURN(rc);
1574 }
1575
1576 static int __lfsck_layout_update_pfid(const struct lu_env *env,
1577                                       struct dt_object *child,
1578                                       const struct lu_fid *pfid, __u32 offset)
1579 {
1580         struct dt_device        *dev    = lfsck_obj2dev(child);
1581         struct filter_fid       *ff     = &lfsck_env_info(env)->lti_new_pfid;
1582         struct thandle          *handle;
1583         struct lu_buf            buf    = { NULL };
1584         int                      rc;
1585
1586         ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
1587         ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
1588         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
1589          * MDT-object's FID::f_ver, instead it is the OST-object index in its
1590          * parent MDT-object's layout EA. */
1591         ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
1592         lfsck_buf_init(&buf, ff, sizeof(struct filter_fid));
1593
1594         handle = dt_trans_create(env, dev);
1595         if (IS_ERR(handle))
1596                 RETURN(PTR_ERR(handle));
1597
1598         rc = dt_declare_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
1599         if (rc != 0)
1600                 GOTO(stop, rc);
1601
1602         rc = dt_trans_start_local(env, dev, handle);
1603         if (rc != 0)
1604                 GOTO(stop, rc);
1605
1606         rc = dt_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
1607
1608         GOTO(stop, rc);
1609
1610 stop:
1611         dt_trans_stop(env, dev, handle);
1612
1613         return rc;
1614 }
1615
1616 /**
1617  * \retval       +1: repaired
1618  * \retval        0: did nothing
1619  * \retval      -ve: on error
1620  */
1621 static int lfsck_layout_update_pfid(const struct lu_env *env,
1622                                     struct lfsck_component *com,
1623                                     struct dt_object *parent,
1624                                     struct lu_fid *cfid,
1625                                     struct dt_device *cdev, __u32 ea_off)
1626 {
1627         struct dt_object        *child;
1628         int                      rc     = 0;
1629         ENTRY;
1630
1631         child = lfsck_object_find_by_dev(env, cdev, cfid);
1632         if (IS_ERR(child))
1633                 RETURN(PTR_ERR(child));
1634
1635         rc = __lfsck_layout_update_pfid(env, child,
1636                                         lu_object_fid(&parent->do_lu), ea_off);
1637         lfsck_object_put(env, child);
1638
1639         RETURN(rc == 0 ? 1 : rc);
1640 }
1641
1642 /**
1643  * This function will create the MDT-object with the given (partial) LOV EA.
1644  *
1645  * Under some data corruption cases, the MDT-object of the file may be lost,
1646  * but its OST-objects, or some of them are there. The layout LFSCK needs to
1647  * re-create the MDT-object with the orphan OST-object(s) information.
1648  *
1649  * On the other hand, the LFSCK may has created some OST-object for repairing
1650  * dangling LOV EA reference, but as the LFSCK processing, it may find that
1651  * the old OST-object is there and should replace the former new created OST
1652  * object. Unfortunately, some others have modified such newly created object.
1653  * To keep the data (both new and old), the LFSCK will create MDT-object with
1654  * new FID to reference the original OST-object.
1655  *
1656  * \param[in] env       pointer to the thread context
1657  * \param[in] com       pointer to the lfsck component
1658  * \param[in] ltd       pointer to target device descriptor
1659  * \param[in] rec       pointer to the record for the orphan OST-object
1660  * \param[in] cfid      pointer to FID for the orphan OST-object
1661  * \param[in] infix     additional information, such as the FID for original
1662  *                      MDT-object and the stripe offset in the LOV EA
1663  * \param[in] type      the type for describing why the orphan MDT-object is
1664  *                      created. The rules are as following:
1665  *
1666  *  type "C":           Multiple OST-objects claim the same MDT-object and the
1667  *                      same slot in the layout EA. Then the LFSCK will create
1668  *                      new MDT-object(s) to hold the conflict OST-object(s).
1669  *
1670  *  type "N":           The orphan OST-object does not know which one was the
1671  *                      real parent MDT-object, so the LFSCK uses new FID for
1672  *                      its parent MDT-object.
1673  *
1674  *  type "R":           The orphan OST-object knows its parent MDT-object FID,
1675  *                      but does not know the position (the file name) in the
1676  *                      layout.
1677  *
1678  *  type "D":           The MDT-object is a directory, it may knows its parent
1679  *                      but because there is no valid linkEA, the LFSCK cannot
1680  *                      know where to put it back to the namespace.
1681  *  type "O":           The MDT-object has no linkEA, and there is no name
1682  *                      entry that references the MDT-object.
1683  *
1684  *  type "P":           The orphan object to be created was a parent directory
1685  *                      of some MDT-object which linkEA shows that the @orphan
1686  *                      object is missing.
1687  *
1688  * The orphan name will be like:
1689  * ${FID}-${infix}-${type}-${conflict_version}
1690  *
1691  * \param[in] ea_off    the stripe offset in the LOV EA
1692  *
1693  * \retval              positive on repaired something
1694  * \retval              0 if needs to repair nothing
1695  * \retval              negative error number on failure
1696  */
1697 static int lfsck_layout_recreate_parent(const struct lu_env *env,
1698                                         struct lfsck_component *com,
1699                                         struct lfsck_tgt_desc *ltd,
1700                                         struct lu_orphan_rec *rec,
1701                                         struct lu_fid *cfid,
1702                                         const char *infix,
1703                                         const char *type,
1704                                         __u32 ea_off)
1705 {
1706         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1707         struct dt_insert_rec            *dtrec  = &info->lti_dt_rec;
1708         char                            *name   = info->lti_key;
1709         struct lu_attr                  *la     = &info->lti_la2;
1710         struct dt_object_format         *dof    = &info->lti_dof;
1711         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1712         struct lu_fid                   *pfid   = &rec->lor_fid;
1713         struct lu_fid                   *tfid   = &info->lti_fid3;
1714         struct dt_device                *dev    = lfsck->li_bottom;
1715         struct dt_object                *lpf    = lfsck->li_lpf_obj;
1716         struct dt_object                *pobj   = NULL;
1717         struct dt_object                *cobj   = NULL;
1718         struct thandle                  *th     = NULL;
1719         struct lu_buf                   *ea_buf = &info->lti_big_buf;
1720         struct lu_buf                    lov_buf;
1721         struct lfsck_lock_handle        *llh    = &info->lti_llh;
1722         struct linkea_data               ldata  = { NULL };
1723         struct lu_buf                    linkea_buf;
1724         const struct lu_name            *pname;
1725         int                              size   = 0;
1726         int                              idx    = 0;
1727         int                              rc     = 0;
1728         ENTRY;
1729
1730         if (unlikely(lpf == NULL))
1731                 GOTO(log, rc = -ENXIO);
1732
1733         /* We use two separated transactions to repair the inconsistency.
1734          *
1735          * 1) create the MDT-object locally.
1736          * 2) update the OST-object's PFID EA if necessary.
1737          *
1738          * If 1) succeed, but 2) failed, then the OST-object's PFID EA will be
1739          * updated when the layout LFSCK run next time.
1740          *
1741          * If 1) failed, but 2) succeed, then such MDT-object will be re-created
1742          * when the layout LFSCK run next time. */
1743
1744         if (fid_is_zero(pfid)) {
1745                 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
1746                 if (rc != 0)
1747                         GOTO(log, rc);
1748
1749                 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
1750                 if (IS_ERR(cobj))
1751                         GOTO(log, rc = PTR_ERR(cobj));
1752         }
1753
1754         pobj = lfsck_object_find_by_dev(env, dev, pfid);
1755         if (IS_ERR(pobj))
1756                 GOTO(log, rc = PTR_ERR(pobj));
1757
1758         LASSERT(infix != NULL);
1759         LASSERT(type != NULL);
1760
1761         memset(la, 0, sizeof(*la));
1762         la->la_uid = rec->lor_uid;
1763         la->la_gid = rec->lor_gid;
1764         la->la_mode = S_IFREG | S_IRUSR;
1765         la->la_valid = LA_MODE | LA_UID | LA_GID;
1766
1767         memset(dof, 0, sizeof(*dof));
1768         dof->dof_type = dt_mode_to_dft(S_IFREG);
1769         /* Because the dof->dof_reg.striped = 0, the LOD will not create
1770          * the stripe(s). The LFSCK will specify the LOV EA via
1771          * lfsck_layout_extend_lovea(). */
1772
1773         size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
1774         if (ea_buf->lb_len < size) {
1775                 lu_buf_realloc(ea_buf, size);
1776                 if (ea_buf->lb_buf == NULL)
1777                         GOTO(log, rc = -ENOMEM);
1778         }
1779
1780 again:
1781         do {
1782                 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
1783                          type, idx++);
1784                 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1785                                (const struct dt_key *)name);
1786                 if (rc != 0 && rc != -ENOENT)
1787                         GOTO(log, rc);
1788         } while (rc == 0);
1789
1790         rc = lfsck_lock(env, lfsck, lfsck->li_lpf_obj, name, llh,
1791                         MDS_INODELOCK_UPDATE, LCK_PW);
1792         if (rc != 0)
1793                 GOTO(log, rc);
1794
1795         /* Re-check whether the name conflict with othrs after taken
1796          * the ldlm lock. */
1797         rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1798                        (const struct dt_key *)name);
1799         if (unlikely(rc == 0)) {
1800                 lfsck_unlock(llh);
1801                 goto again;
1802         }
1803
1804         if (rc != -ENOENT)
1805                 GOTO(unlock, rc);
1806
1807         rc = linkea_data_new(&ldata,
1808                              &lfsck_env_info(env)->lti_linkea_buf);
1809         if (rc != 0)
1810                 GOTO(unlock, rc);
1811
1812         pname = lfsck_name_get_const(env, name, strlen(name));
1813         rc = linkea_add_buf(&ldata, pname, lfsck_dto2fid(lfsck->li_lpf_obj));
1814         if (rc != 0)
1815                 GOTO(unlock, rc);
1816
1817         /* The 1st transaction. */
1818         th = dt_trans_create(env, dev);
1819         if (IS_ERR(th))
1820                 GOTO(unlock, rc = PTR_ERR(th));
1821
1822         rc = dt_declare_create(env, pobj, la, NULL, dof, th);
1823         if (rc != 0)
1824                 GOTO(stop, rc);
1825
1826         lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
1827         rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
1828                                   LU_XATTR_REPLACE, th);
1829         if (rc != 0)
1830                 GOTO(stop, rc);
1831
1832         dtrec->rec_fid = pfid;
1833         dtrec->rec_type = S_IFREG;
1834         rc = dt_declare_insert(env, lpf,
1835                                (const struct dt_rec *)dtrec,
1836                                (const struct dt_key *)name, th);
1837         if (rc != 0)
1838                 GOTO(stop, rc);
1839
1840         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1841                        ldata.ld_leh->leh_len);
1842         rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
1843                                   XATTR_NAME_LINK, 0, th);
1844         if (rc != 0)
1845                 GOTO(stop, rc);
1846
1847         rc = dt_trans_start_local(env, dev, th);
1848         if (rc != 0)
1849                 GOTO(stop, rc);
1850
1851         dt_write_lock(env, pobj, 0);
1852         rc = dt_create(env, pobj, la, NULL, dof, th);
1853         if (rc == 0)
1854                 rc = lfsck_layout_extend_lovea(env, lfsck, th, pobj, cfid,
1855                                 &lov_buf, 0, ltd->ltd_index, ea_off, true);
1856         dt_write_unlock(env, pobj);
1857         if (rc < 0)
1858                 GOTO(stop, rc);
1859
1860         rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
1861                        (const struct dt_key *)name, th, 1);
1862         if (rc != 0)
1863                 GOTO(stop, rc);
1864
1865         rc = dt_xattr_set(env, pobj, &linkea_buf, XATTR_NAME_LINK, 0, th);
1866         if (rc == 0 && cobj != NULL) {
1867                 dt_trans_stop(env, dev, th);
1868                 th = NULL;
1869
1870                 /* The 2nd transaction. */
1871                 rc = __lfsck_layout_update_pfid(env, cobj, pfid, ea_off);
1872         }
1873
1874         GOTO(stop, rc);
1875
1876 stop:
1877         if (th != NULL)
1878                 dt_trans_stop(env, dev, th);
1879
1880 unlock:
1881         lfsck_unlock(llh);
1882
1883 log:
1884         if (cobj != NULL && !IS_ERR(cobj))
1885                 lfsck_object_put(env, cobj);
1886         if (pobj != NULL && !IS_ERR(pobj))
1887                 lfsck_object_put(env, pobj);
1888
1889         if (rc < 0)
1890                 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
1891                        "recreate the lost MDT-object: parent "DFID
1892                        ", child "DFID", OST-index %u, stripe-index %u, "
1893                        "infix %s, type %s: rc = %d\n",
1894                        lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
1895                        ltd->ltd_index, ea_off, infix, type, rc);
1896
1897         return rc >= 0 ? 1 : rc;
1898 }
1899
1900 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
1901                                                    struct lfsck_component *com,
1902                                                    const struct lu_fid *fid,
1903                                                    __u32 index)
1904 {
1905         struct lfsck_thread_info *info  = lfsck_env_info(env);
1906         struct lfsck_request     *lr    = &info->lti_lr;
1907         struct lfsck_instance    *lfsck = com->lc_lfsck;
1908         struct lfsck_tgt_desc    *ltd;
1909         struct ptlrpc_request    *req;
1910         struct lfsck_request     *tmp;
1911         struct obd_export        *exp;
1912         int                       rc    = 0;
1913         ENTRY;
1914
1915         ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
1916         if (unlikely(ltd == NULL))
1917                 RETURN(-ENXIO);
1918
1919         exp = ltd->ltd_exp;
1920         if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
1921                 GOTO(put, rc = -EOPNOTSUPP);
1922
1923         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
1924         if (req == NULL)
1925                 GOTO(put, rc = -ENOMEM);
1926
1927         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
1928         if (rc != 0) {
1929                 ptlrpc_request_free(req);
1930
1931                 GOTO(put, rc);
1932         }
1933
1934         memset(lr, 0, sizeof(*lr));
1935         lr->lr_event = LE_CONDITIONAL_DESTROY;
1936         lr->lr_active = LFSCK_TYPE_LAYOUT;
1937         lr->lr_fid = *fid;
1938
1939         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
1940         *tmp = *lr;
1941         ptlrpc_request_set_replen(req);
1942
1943         rc = ptlrpc_queue_wait(req);
1944         ptlrpc_req_finished(req);
1945
1946         GOTO(put, rc);
1947
1948 put:
1949         lfsck_tgt_put(ltd);
1950
1951         return rc;
1952 }
1953
1954 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
1955                                                   struct lfsck_component *com,
1956                                                   struct lfsck_request *lr)
1957 {
1958         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1959         struct lu_attr                  *la     = &info->lti_la;
1960         union ldlm_policy_data          *policy = &info->lti_policy;
1961         struct ldlm_res_id              *resid  = &info->lti_resid;
1962         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1963         struct dt_device                *dev    = lfsck->li_bottom;
1964         struct lu_fid                   *fid    = &lr->lr_fid;
1965         struct dt_object                *obj;
1966         struct thandle                  *th     = NULL;
1967         struct lustre_handle             lh     = { 0 };
1968         __u64                            flags  = 0;
1969         int                              rc     = 0;
1970         ENTRY;
1971
1972         obj = lfsck_object_find_by_dev(env, dev, fid);
1973         if (IS_ERR(obj))
1974                 RETURN(PTR_ERR(obj));
1975
1976         dt_read_lock(env, obj, 0);
1977         if (dt_object_exists(obj) == 0 ||
1978             lfsck_is_dead_obj(obj)) {
1979                 dt_read_unlock(env, obj);
1980
1981                 GOTO(put, rc = -ENOENT);
1982         }
1983
1984         /* Get obj's attr without lock firstly. */
1985         rc = dt_attr_get(env, obj, la);
1986         dt_read_unlock(env, obj);
1987         if (rc != 0)
1988                 GOTO(put, rc);
1989
1990         if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
1991                 GOTO(put, rc = -ETXTBSY);
1992
1993         /* Acquire extent lock on [0, EOF] to sync with all possible written. */
1994         LASSERT(lfsck->li_namespace != NULL);
1995
1996         memset(policy, 0, sizeof(*policy));
1997         policy->l_extent.end = OBD_OBJECT_EOF;
1998         ost_fid_build_resid(fid, resid);
1999         rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
2000                                     policy, LCK_EX, &flags, ldlm_blocking_ast,
2001                                     ldlm_completion_ast, NULL, NULL, 0,
2002                                     LVB_T_NONE, NULL, &lh);
2003         if (rc != ELDLM_OK)
2004                 GOTO(put, rc = -EIO);
2005
2006         dt_write_lock(env, obj, 0);
2007         /* Get obj's attr within lock again. */
2008         rc = dt_attr_get(env, obj, la);
2009         if (rc != 0)
2010                 GOTO(unlock, rc);
2011
2012         if (la->la_ctime != 0)
2013                 GOTO(unlock, rc = -ETXTBSY);
2014
2015         th = dt_trans_create(env, dev);
2016         if (IS_ERR(th))
2017                 GOTO(unlock, rc = PTR_ERR(th));
2018
2019         rc = dt_declare_ref_del(env, obj, th);
2020         if (rc != 0)
2021                 GOTO(stop, rc);
2022
2023         rc = dt_declare_destroy(env, obj, th);
2024         if (rc != 0)
2025                 GOTO(stop, rc);
2026
2027         rc = dt_trans_start_local(env, dev, th);
2028         if (rc != 0)
2029                 GOTO(stop, rc);
2030
2031         rc = dt_ref_del(env, obj, th);
2032         if (rc != 0)
2033                 GOTO(stop, rc);
2034
2035         rc = dt_destroy(env, obj, th);
2036         if (rc == 0)
2037                 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2038                        "OST-object "DFID" that was created for reparing "
2039                        "dangling referenced case. But the original missing "
2040                        "OST-object is found now.\n",
2041                        lfsck_lfsck2name(lfsck), PFID(fid));
2042
2043         GOTO(stop, rc);
2044
2045 stop:
2046         dt_trans_stop(env, dev, th);
2047
2048 unlock:
2049         dt_write_unlock(env, obj);
2050         ldlm_lock_decref(&lh, LCK_EX);
2051
2052 put:
2053         lfsck_object_put(env, obj);
2054
2055         return rc;
2056 }
2057
2058 /**
2059  * Some OST-object has occupied the specified layout EA slot.
2060  * Such OST-object may be generated by the LFSCK when repair
2061  * dangling referenced MDT-object, which can be indicated by
2062  * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2063  * is true and such OST-object has not been modified yet, we
2064  * will replace it with the orphan OST-object; otherwise the
2065  * LFSCK will create new MDT-object to reference the orphan.
2066  *
2067  * \retval       +1: repaired
2068  * \retval        0: did nothing
2069  * \retval      -ve: on error
2070  */
2071 static int lfsck_layout_conflict_create(const struct lu_env *env,
2072                                         struct lfsck_component *com,
2073                                         struct lfsck_tgt_desc *ltd,
2074                                         struct lu_orphan_rec *rec,
2075                                         struct dt_object *parent,
2076                                         struct lu_fid *cfid,
2077                                         struct lu_buf *ea_buf,
2078                                         struct lov_ost_data_v1 *slot,
2079                                         __u32 ea_off)
2080 {
2081         struct lfsck_thread_info *info          = lfsck_env_info(env);
2082         struct lu_fid            *cfid2         = &info->lti_fid2;
2083         struct ost_id            *oi            = &info->lti_oi;
2084         struct lov_mds_md_v1     *lmm           = ea_buf->lb_buf;
2085         struct dt_device         *dev           = lfsck_obj2dev(parent);
2086         struct thandle           *th            = NULL;
2087         struct lustre_handle      lh            = { 0 };
2088         __u32                     ost_idx2      = le32_to_cpu(slot->l_ost_idx);
2089         int                       rc            = 0;
2090         ENTRY;
2091
2092         ostid_le_to_cpu(&slot->l_ost_oi, oi);
2093         rc = ostid_to_fid(cfid2, oi, ost_idx2);
2094         if (rc != 0)
2095                 GOTO(out, rc);
2096
2097         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2098                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2099                               LCK_EX);
2100         if (rc != 0)
2101                 GOTO(out, rc);
2102
2103         rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2104
2105         /* If the conflict OST-obejct is not created for fixing dangling
2106          * referenced MDT-object in former LFSCK check/repair, or it has
2107          * been modified by others, then we cannot destroy it. Re-create
2108          * a new MDT-object for the orphan OST-object. */
2109         if (rc == -ETXTBSY) {
2110                 /* No need the layout lock on the original parent. */
2111                 lfsck_ibits_unlock(&lh, LCK_EX);
2112
2113                 fid_zero(&rec->lor_fid);
2114                 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2115                          "-"DFID"-%x", PFID(lu_object_fid(&parent->do_lu)),
2116                          ea_off);
2117                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2118                                                 info->lti_tmpbuf, "C", ea_off);
2119
2120                 RETURN(rc);
2121         }
2122
2123         if (rc != 0 && rc != -ENOENT)
2124                 GOTO(unlock, rc);
2125
2126         th = dt_trans_create(env, dev);
2127         if (IS_ERR(th))
2128                 GOTO(unlock, rc = PTR_ERR(th));
2129
2130         rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2131                                   LU_XATTR_REPLACE, th);
2132         if (rc != 0)
2133                 GOTO(stop, rc);
2134
2135         rc = dt_trans_start_local(env, dev, th);
2136         if (rc != 0)
2137                 GOTO(stop, rc);
2138
2139         dt_write_lock(env, parent, 0);
2140         lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2141         rc = lfsck_layout_refill_lovea(env, th, parent, cfid, ea_buf, slot,
2142                                        LU_XATTR_REPLACE, ltd->ltd_index);
2143         dt_write_unlock(env, parent);
2144
2145         GOTO(stop, rc);
2146
2147 stop:
2148         dt_trans_stop(env, dev, th);
2149
2150 unlock:
2151         lfsck_ibits_unlock(&lh, LCK_EX);
2152
2153 out:
2154         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2155                "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2156                "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2157                lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2158                PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2159                ea_off, rc);
2160
2161         return rc >= 0 ? 1 : rc;
2162 }
2163
2164 /**
2165  * \retval       +1: repaired
2166  * \retval        0: did nothing
2167  * \retval      -ve: on error
2168  */
2169 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2170                                        struct lfsck_component *com,
2171                                        struct lfsck_tgt_desc *ltd,
2172                                        struct lu_orphan_rec *rec,
2173                                        struct dt_object *parent,
2174                                        struct lu_fid *cfid,
2175                                        __u32 ost_idx, __u32 ea_off)
2176 {
2177         struct lfsck_thread_info *info          = lfsck_env_info(env);
2178         struct lu_buf            *buf           = &info->lti_big_buf;
2179         struct lu_fid            *fid           = &info->lti_fid2;
2180         struct ost_id            *oi            = &info->lti_oi;
2181         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2182         struct dt_device         *dt            = lfsck_obj2dev(parent);
2183         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2184         struct thandle            *handle       = NULL;
2185         size_t                    lovea_size;
2186         struct lov_mds_md_v1     *lmm;
2187         struct lov_ost_data_v1   *objs;
2188         struct lustre_handle      lh            = { 0 };
2189         __u32                     magic;
2190         int                       fl            = 0;
2191         int                       rc            = 0;
2192         int                       rc1;
2193         int                       i;
2194         __u16                     count;
2195         bool                      locked        = false;
2196         ENTRY;
2197
2198         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2199                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2200                               LCK_EX);
2201         if (rc != 0) {
2202                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2203                        "LOV EA for "DFID": parent "DFID", OST-index %u, "
2204                        "stripe-index %u: rc = %d\n",
2205                        lfsck_lfsck2name(lfsck), PFID(cfid),
2206                        PFID(lfsck_dto2fid(parent)), ost_idx, ea_off, rc);
2207
2208                 RETURN(rc);
2209         }
2210
2211 again:
2212         if (locked) {
2213                 dt_write_unlock(env, parent);
2214                 locked = false;
2215         }
2216
2217         if (handle != NULL) {
2218                 dt_trans_stop(env, dt, handle);
2219                 handle = NULL;
2220         }
2221
2222         if (rc < 0)
2223                 GOTO(unlock_layout, rc);
2224
2225         lovea_size = rc;
2226         if (buf->lb_len < lovea_size) {
2227                 lu_buf_realloc(buf, lovea_size);
2228                 if (buf->lb_buf == NULL)
2229                         GOTO(unlock_layout, rc = -ENOMEM);
2230         }
2231
2232         if (!(bk->lb_param & LPF_DRYRUN)) {
2233                 handle = dt_trans_create(env, dt);
2234                 if (IS_ERR(handle))
2235                         GOTO(unlock_layout, rc = PTR_ERR(handle));
2236
2237                 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2238                                           fl, handle);
2239                 if (rc != 0)
2240                         GOTO(stop, rc);
2241
2242                 rc = dt_trans_start_local(env, dt, handle);
2243                 if (rc != 0)
2244                         GOTO(stop, rc);
2245         }
2246
2247         dt_write_lock(env, parent, 0);
2248         locked = true;
2249         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2250         if (rc == -ERANGE) {
2251                 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV);
2252                 LASSERT(rc != 0);
2253                 goto again;
2254         } else if (rc == -ENODATA || rc == 0) {
2255                 lovea_size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2256                 /* If the declared is not big enough, re-try. */
2257                 if (buf->lb_len < lovea_size) {
2258                         rc = lovea_size;
2259                         goto again;
2260                 }
2261                 fl = LU_XATTR_CREATE;
2262         } else if (rc < 0) {
2263                 GOTO(unlock_parent, rc);
2264         } else if (unlikely(buf->lb_len == 0)) {
2265                 goto again;
2266         } else {
2267                 fl = LU_XATTR_REPLACE;
2268                 lovea_size = rc;
2269         }
2270
2271         if (fl == LU_XATTR_CREATE) {
2272                 if (bk->lb_param & LPF_DRYRUN)
2273                         GOTO(unlock_parent, rc = 1);
2274
2275                 LASSERT(buf->lb_len >= lovea_size);
2276
2277                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2278                                                buf, fl, ost_idx, ea_off, false);
2279
2280                 GOTO(unlock_parent, rc);
2281         }
2282
2283         lmm = buf->lb_buf;
2284         rc1 = lfsck_layout_verify_header(lmm);
2285
2286         /* If the LOV EA crashed, the rebuild it. */
2287         if (rc1 == -EINVAL) {
2288                 if (bk->lb_param & LPF_DRYRUN)
2289                         GOTO(unlock_parent, rc = 1);
2290
2291                 LASSERT(buf->lb_len >= lovea_size);
2292
2293                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2294                                                buf, fl, ost_idx, ea_off, true);
2295
2296                 GOTO(unlock_parent, rc);
2297         }
2298
2299         /* For other unknown magic/pattern, keep the current LOV EA. */
2300         if (rc1 != 0)
2301                 GOTO(unlock_parent, rc = rc1);
2302
2303         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2304          * been verified in lfsck_layout_verify_header() already. If some
2305          * new magic introduced in the future, then layout LFSCK needs to
2306          * be updated also. */
2307         magic = le32_to_cpu(lmm->lmm_magic);
2308         if (magic == LOV_MAGIC_V1) {
2309                 objs = &lmm->lmm_objects[0];
2310         } else {
2311                 LASSERT(magic == LOV_MAGIC_V3);
2312                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2313         }
2314
2315         count = le16_to_cpu(lmm->lmm_stripe_count);
2316         if (count == 0)
2317                 GOTO(unlock_parent, rc = -EINVAL);
2318         LASSERT(count > 0);
2319
2320         /* Exceed the current end of MDT-object layout EA. Then extend it. */
2321         if (count <= ea_off) {
2322                 if (bk->lb_param & LPF_DRYRUN)
2323                         GOTO(unlock_parent, rc = 1);
2324
2325                 lovea_size = lov_mds_md_size(ea_off + 1, magic);
2326                 /* If the declared is not big enough, re-try. */
2327                 if (buf->lb_len < lovea_size) {
2328                         rc = lovea_size;
2329                         goto again;
2330                 }
2331
2332                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2333                                                buf, fl, ost_idx, ea_off, false);
2334
2335                 GOTO(unlock_parent, rc);
2336         }
2337
2338         LASSERTF(rc > 0, "invalid rc = %d\n", rc);
2339
2340         for (i = 0; i < count; i++, objs++) {
2341                 /* The MDT-object was created via lfsck_layout_recover_create()
2342                  * by others before, and we fill the dummy layout EA. */
2343                 if (lovea_slot_is_dummy(objs)) {
2344                         if (i != ea_off)
2345                                 continue;
2346
2347                         if (bk->lb_param & LPF_DRYRUN)
2348                                 GOTO(unlock_parent, rc = 1);
2349
2350                         lmm->lmm_layout_gen =
2351                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2352                         rc = lfsck_layout_refill_lovea(env, handle, parent,
2353                                                        cfid, buf, objs, fl,
2354                                                        ost_idx);
2355
2356                         CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
2357                                "dummy layout slot for "DFID": parent "DFID
2358                                ", OST-index %u, stripe-index %u: rc = %d\n",
2359                                lfsck_lfsck2name(lfsck), PFID(cfid),
2360                                PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
2361
2362                         GOTO(unlock_parent, rc);
2363                 }
2364
2365                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2366                 rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
2367                 if (rc != 0) {
2368                         CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
2369                                "invalid layout EA at the slot %d, index %u\n",
2370                                lfsck_lfsck2name(lfsck),
2371                                PFID(lfsck_dto2fid(parent)), i,
2372                                le32_to_cpu(objs->l_ost_idx));
2373
2374                         GOTO(unlock_parent, rc);
2375                 }
2376
2377                 /* It should be rare case, the slot is there, but the LFSCK
2378                  * does not handle it during the first-phase cycle scanning. */
2379                 if (unlikely(lu_fid_eq(fid, cfid))) {
2380                         if (i == ea_off) {
2381                                 GOTO(unlock_parent, rc = 0);
2382                         } else {
2383                                 /* Rare case that the OST-object index
2384                                  * does not match the parent MDT-object
2385                                  * layout EA. We trust the later one. */
2386                                 if (bk->lb_param & LPF_DRYRUN)
2387                                         GOTO(unlock_parent, rc = 1);
2388
2389                                 dt_write_unlock(env, parent);
2390                                 if (handle != NULL)
2391                                         dt_trans_stop(env, dt, handle);
2392                                 lfsck_ibits_unlock(&lh, LCK_EX);
2393                                 rc = lfsck_layout_update_pfid(env, com, parent,
2394                                                         cfid, ltd->ltd_tgt, i);
2395
2396                                 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
2397                                        "updated OST-object's pfid for "DFID
2398                                        ": parent "DFID", OST-index %u, "
2399                                        "stripe-index %u: rc = %d\n",
2400                                        lfsck_lfsck2name(lfsck), PFID(cfid),
2401                                        PFID(lfsck_dto2fid(parent)),
2402                                        ltd->ltd_index, i, rc);
2403
2404                                 RETURN(rc);
2405                         }
2406                 }
2407         }
2408
2409         /* The MDT-object exists, but related layout EA slot is occupied
2410          * by others. */
2411         if (bk->lb_param & LPF_DRYRUN)
2412                 GOTO(unlock_parent, rc = 1);
2413
2414         dt_write_unlock(env, parent);
2415         if (handle != NULL)
2416                 dt_trans_stop(env, dt, handle);
2417         lfsck_ibits_unlock(&lh, LCK_EX);
2418         if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
2419                 objs = &lmm->lmm_objects[ea_off];
2420         else
2421                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
2422         rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
2423                                           buf, objs, ea_off);
2424
2425         RETURN(rc);
2426
2427 unlock_parent:
2428         if (locked)
2429                 dt_write_unlock(env, parent);
2430
2431 stop:
2432         if (handle != NULL)
2433                 dt_trans_stop(env, dt, handle);
2434
2435 unlock_layout:
2436         lfsck_ibits_unlock(&lh, LCK_EX);
2437
2438         return rc;
2439 }
2440
2441 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
2442                                         struct lfsck_component *com,
2443                                         struct lfsck_tgt_desc *ltd,
2444                                         struct lu_orphan_rec *rec,
2445                                         struct lu_fid *cfid)
2446 {
2447         struct lfsck_layout     *lo     = com->lc_file_ram;
2448         struct lu_fid           *pfid   = &rec->lor_fid;
2449         struct dt_object        *parent = NULL;
2450         __u32                    ea_off = pfid->f_stripe_idx;
2451         int                      rc     = 0;
2452         ENTRY;
2453
2454         if (!fid_is_sane(cfid))
2455                 GOTO(out, rc = -EINVAL);
2456
2457         if (fid_is_zero(pfid)) {
2458                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2459                                                   "", "N", ea_off);
2460                 GOTO(out, rc);
2461         }
2462
2463         pfid->f_ver = 0;
2464         if (!fid_is_sane(pfid))
2465                 GOTO(out, rc = -EINVAL);
2466
2467         parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
2468         if (IS_ERR(parent))
2469                 GOTO(out, rc = PTR_ERR(parent));
2470
2471         if (unlikely(dt_object_remote(parent) != 0))
2472                 GOTO(put, rc = -EXDEV);
2473
2474         if (dt_object_exists(parent) == 0) {
2475                 lfsck_object_put(env, parent);
2476                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2477                                                   "", "R", ea_off);
2478                 GOTO(out, rc);
2479         }
2480
2481         if (!S_ISREG(lu_object_attr(&parent->do_lu)))
2482                 GOTO(put, rc = -EISDIR);
2483
2484         rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
2485                                          ltd->ltd_index, ea_off);
2486
2487         GOTO(put, rc);
2488
2489 put:
2490         if (rc <= 0)
2491                 lfsck_object_put(env, parent);
2492         else
2493                 /* The layout EA is changed, need to be reloaded next time. */
2494                 lu_object_put_nocache(env, &parent->do_lu);
2495
2496 out:
2497         down_write(&com->lc_sem);
2498         com->lc_new_scanned++;
2499         com->lc_new_checked++;
2500         if (rc > 0) {
2501                 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
2502                 rc = 0;
2503         } else if (rc < 0) {
2504                 lo->ll_objs_failed_phase2++;
2505         }
2506         up_write(&com->lc_sem);
2507
2508         return rc;
2509 }
2510
2511 static int lfsck_layout_scan_orphan(const struct lu_env *env,
2512                                     struct lfsck_component *com,
2513                                     struct lfsck_tgt_desc *ltd)
2514 {
2515         struct lfsck_assistant_data     *lad    = com->lc_data;
2516         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2517         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
2518         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2519         struct ost_id                   *oi     = &info->lti_oi;
2520         struct lu_fid                   *fid    = &info->lti_fid;
2521         struct dt_object                *obj;
2522         const struct dt_it_ops          *iops;
2523         struct dt_it                    *di;
2524         int                              rc     = 0;
2525         ENTRY;
2526
2527         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
2528                "scanning for OST%04x\n",
2529                lfsck_lfsck2name(lfsck), ltd->ltd_index);
2530
2531         if (cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
2532                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
2533                        "scanning for OST%04x\n",
2534                        lfsck_lfsck2name(lfsck), ltd->ltd_index);
2535
2536                 RETURN(0);
2537         }
2538
2539         ostid_set_seq(oi, FID_SEQ_IDIF);
2540         ostid_set_id(oi, 0);
2541         rc = ostid_to_fid(fid, oi, ltd->ltd_index);
2542         if (rc != 0)
2543                 GOTO(log, rc);
2544
2545         obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
2546         if (unlikely(IS_ERR(obj)))
2547                 GOTO(log, rc = PTR_ERR(obj));
2548
2549         rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_orphan_features);
2550         if (rc != 0)
2551                 GOTO(put, rc);
2552
2553         iops = &obj->do_index_ops->dio_it;
2554         di = iops->init(env, obj, 0);
2555         if (IS_ERR(di))
2556                 GOTO(put, rc = PTR_ERR(di));
2557
2558         rc = iops->load(env, di, 0);
2559         if (rc == -ESRCH) {
2560                 /* -ESRCH means that the orphan OST-objects rbtree has been
2561                  * cleanup because of the OSS server restart or other errors. */
2562                 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
2563                 GOTO(fini, rc);
2564         }
2565
2566         if (rc == 0)
2567                 rc = iops->next(env, di);
2568         else if (rc > 0)
2569                 rc = 0;
2570
2571         if (rc < 0)
2572                 GOTO(fini, rc);
2573
2574         if (rc > 0)
2575                 GOTO(fini, rc = 0);
2576
2577         do {
2578                 struct dt_key           *key;
2579                 struct lu_orphan_rec    *rec = &info->lti_rec;
2580
2581                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
2582                     unlikely(!thread_is_running(&lfsck->li_thread)))
2583                         break;
2584
2585                 key = iops->key(env, di);
2586                 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
2587                 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
2588                 if (rc == 0)
2589                         rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
2590                                         &com->lc_fid_latest_scanned_phase2);
2591                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2592                         GOTO(fini, rc);
2593
2594                 lfsck_control_speed_by_self(com);
2595                 do {
2596                         rc = iops->next(env, di);
2597                 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
2598         } while (rc == 0);
2599
2600         GOTO(fini, rc);
2601
2602 fini:
2603         iops->put(env, di);
2604         iops->fini(env, di);
2605 put:
2606         lfsck_object_put(env, obj);
2607
2608 log:
2609         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
2610                "scanning for OST%04x: rc = %d\n",
2611                lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
2612
2613         return rc > 0 ? 0 : rc;
2614 }
2615
2616 /* For the MDT-object with dangling reference, we need to repare the
2617  * inconsistency according to the LFSCK sponsor's requirement:
2618  *
2619  * 1) Keep the inconsistency there and report the inconsistency case,
2620  *    then give the chance to the application to find related issues,
2621  *    and the users can make the decision about how to handle it with
2622  *    more human knownledge. (by default)
2623  *
2624  * 2) Re-create the missing OST-object with the FID/owner information. */
2625 static int lfsck_layout_repair_dangling(const struct lu_env *env,
2626                                         struct lfsck_component *com,
2627                                         struct dt_object *parent,
2628                                         struct lfsck_layout_req *llr,
2629                                         struct lu_attr *la)
2630 {
2631         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2632         struct filter_fid               *pfid   = &info->lti_new_pfid;
2633         struct dt_object_format         *dof    = &info->lti_dof;
2634         struct dt_object                *child  = llr->llr_child;
2635         struct dt_device                *dev    = lfsck_obj2dev(child);
2636         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
2637         struct thandle                  *handle;
2638         struct lu_buf                   *buf;
2639         struct lustre_handle             lh     = { 0 };
2640         int                              rc;
2641         bool                             create;
2642         ENTRY;
2643
2644         if (com->lc_lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ)
2645                 create = true;
2646         else
2647                 create = false;
2648
2649         if (!create)
2650                 GOTO(log, rc = 1);
2651
2652         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2653                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2654                               LCK_EX);
2655         if (rc != 0)
2656                 GOTO(log, rc);
2657
2658         rc = dt_attr_get(env, parent, la);
2659         if (rc != 0)
2660                 GOTO(unlock1, rc);
2661
2662         la->la_mode = S_IFREG | 0666;
2663         la->la_atime = la->la_mtime = la->la_ctime = 0;
2664         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
2665                        LA_ATIME | LA_MTIME | LA_CTIME;
2666         memset(dof, 0, sizeof(*dof));
2667         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
2668         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
2669         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2670          * MDT-object's FID::f_ver, instead it is the OST-object index in its
2671          * parent MDT-object's layout EA. */
2672         pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
2673         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
2674
2675         handle = dt_trans_create(env, dev);
2676         if (IS_ERR(handle))
2677                 GOTO(unlock1, rc = PTR_ERR(handle));
2678
2679         rc = dt_declare_create(env, child, la, NULL, dof, handle);
2680         if (rc != 0)
2681                 GOTO(stop, rc);
2682
2683         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
2684                                   LU_XATTR_CREATE, handle);
2685         if (rc != 0)
2686                 GOTO(stop, rc);
2687
2688         rc = dt_trans_start_local(env, dev, handle);
2689         if (rc != 0)
2690                 GOTO(stop, rc);
2691
2692         dt_read_lock(env, parent, 0);
2693         if (unlikely(lfsck_is_dead_obj(parent)))
2694                 GOTO(unlock2, rc = 1);
2695
2696         rc = dt_create(env, child, la, NULL, dof, handle);
2697         if (rc != 0)
2698                 GOTO(unlock2, rc);
2699
2700         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
2701                           handle);
2702
2703         GOTO(unlock2, rc);
2704
2705 unlock2:
2706         dt_read_unlock(env, parent);
2707
2708 stop:
2709         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2710
2711 unlock1:
2712         lfsck_ibits_unlock(&lh, LCK_EX);
2713
2714 log:
2715         if (rc != 0)
2716                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
2717                        "dangling reference for: parent "DFID", child "DFID
2718                        ", OST-index %u, stripe-index %u, owner %u/%u. %s: "
2719                        "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
2720                        PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
2721                        llr->llr_ost_idx, llr->llr_lov_idx,
2722                        la->la_uid, la->la_gid,
2723                        create ? "Create the lost OST-object as required" :
2724                                 "Keep the MDT-object there by default", rc);
2725
2726         return rc;
2727 }
2728
2729 /* If the OST-object does not recognize the MDT-object as its parent, and
2730  * there is no other MDT-object claims as its parent, then just trust the
2731  * given MDT-object as its parent. So update the OST-object filter_fid. */
2732 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
2733                                               struct lfsck_component *com,
2734                                               struct dt_object *parent,
2735                                               struct lfsck_layout_req *llr,
2736                                               struct lu_attr *la)
2737 {
2738         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2739         struct filter_fid               *pfid   = &info->lti_new_pfid;
2740         struct dt_object                *child  = llr->llr_child;
2741         struct dt_device                *dev    = lfsck_obj2dev(child);
2742         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
2743         struct thandle                  *handle;
2744         struct lu_buf                   *buf;
2745         struct lustre_handle             lh     = { 0 };
2746         int                              rc;
2747         ENTRY;
2748
2749         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2750                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2751                               LCK_EX);
2752         if (rc != 0)
2753                 GOTO(log, rc);
2754
2755         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
2756         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
2757         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2758          * MDT-object's FID::f_ver, instead it is the OST-object index in its
2759          * parent MDT-object's layout EA. */
2760         pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
2761         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
2762
2763         handle = dt_trans_create(env, dev);
2764         if (IS_ERR(handle))
2765                 GOTO(unlock1, rc = PTR_ERR(handle));
2766
2767         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
2768         if (rc != 0)
2769                 GOTO(stop, rc);
2770
2771         rc = dt_attr_get(env, parent, la);
2772         if (rc != 0)
2773                 GOTO(stop, rc);
2774
2775         la->la_valid = LA_UID | LA_GID;
2776         rc = dt_declare_attr_set(env, child, la, handle);
2777         if (rc != 0)
2778                 GOTO(stop, rc);
2779
2780         rc = dt_trans_start_local(env, dev, handle);
2781         if (rc != 0)
2782                 GOTO(stop, rc);
2783
2784         dt_write_lock(env, parent, 0);
2785         if (unlikely(lfsck_is_dead_obj(parent)))
2786                 GOTO(unlock2, rc = 1);
2787
2788         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
2789         if (rc != 0)
2790                 GOTO(unlock2, rc);
2791
2792         /* Get the latest parent's owner. */
2793         rc = dt_attr_get(env, parent, la);
2794         if (rc != 0)
2795                 GOTO(unlock2, rc);
2796
2797         la->la_valid = LA_UID | LA_GID;
2798         rc = dt_attr_set(env, child, la, handle);
2799
2800         GOTO(unlock2, rc);
2801
2802 unlock2:
2803         dt_write_unlock(env, parent);
2804
2805 stop:
2806         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2807
2808 unlock1:
2809         lfsck_ibits_unlock(&lh, LCK_EX);
2810
2811 log:
2812         if (rc != 0)
2813                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
2814                        "unmatched MDT-OST pair for: parent "DFID
2815                        ", child "DFID", OST-index %u, stripe-index %u, "
2816                        "owner %u/%u: rc = %d\n",
2817                        lfsck_lfsck2name(com->lc_lfsck),
2818                        PFID(lfsck_dto2fid(parent)),
2819                        PFID(lfsck_dto2fid(child)),
2820                        llr->llr_ost_idx, llr->llr_lov_idx,
2821                        la->la_uid, la->la_gid, rc);
2822
2823         return rc;
2824 }
2825
2826 /* If there are more than one MDT-objects claim as the OST-object's parent,
2827  * and the OST-object only recognizes one of them, then we need to generate
2828  * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
2829 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
2830                                                    struct lfsck_component *com,
2831                                                    struct dt_object *parent,
2832                                                    struct lfsck_layout_req *llr,
2833                                                    struct lu_attr *la,
2834                                                    struct lu_buf *buf)
2835 {
2836         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2837         struct dt_allocation_hint       *hint   = &info->lti_hint;
2838         struct dt_object_format         *dof    = &info->lti_dof;
2839         struct ost_id                   *oi     = &info->lti_oi;
2840         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2841         struct dt_device                *dev;
2842         struct lu_device                *d      =
2843                                 &lfsck_obj2dev(llr->llr_child)->dd_lu_dev;
2844         struct lu_object                *o;
2845         struct lu_object                *n;
2846         struct dt_object                *child  = NULL;
2847         struct thandle                  *handle = NULL;
2848         struct lov_mds_md_v1            *lmm;
2849         struct lov_ost_data_v1          *objs;
2850         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
2851         struct lu_fid                    tfid;
2852         struct lustre_handle             lh     = { 0 };
2853         struct lu_buf                    ea_buf;
2854         __u32                            magic;
2855         __u32                            index;
2856         int                              rc;
2857         ENTRY;
2858
2859         /* We use two separated transactions to repair the inconsistency.
2860          *
2861          * 1) create the child (OST-object).
2862          * 2) update the parent LOV EA according to the child's FID.
2863          *
2864          * If 1) succeed, but 2) failed or aborted, then such OST-object will be
2865          * handled as orphan when the layout LFSCK run next time.
2866          *
2867          * If 1) failed, but 2) succeed, then such OST-object will be re-created
2868          * as dangling referened case when the layout LFSCK run next time. */
2869
2870         /* The 1st transaction. */
2871         o = lu_object_anon(env, d, NULL);
2872         if (IS_ERR(o))
2873                 GOTO(log, rc = PTR_ERR(o));
2874
2875         n = lu_object_locate(o->lo_header, d->ld_type);
2876         if (unlikely(n == NULL)) {
2877                 lu_object_put_nocache(env, o);
2878
2879                 GOTO(log, rc = -EINVAL);
2880         }
2881
2882         child = container_of(n, struct dt_object, do_lu);
2883         memset(hint, 0, sizeof(*hint));
2884         rc = dt_attr_get(env, parent, la);
2885         if (rc != 0)
2886                 GOTO(log, rc);
2887
2888         la->la_valid = LA_UID | LA_GID;
2889         memset(dof, 0, sizeof(*dof));
2890
2891         dev = lfsck_obj2dev(child);
2892         handle = dt_trans_create(env, dev);
2893         if (IS_ERR(handle))
2894                 GOTO(log, rc = PTR_ERR(handle));
2895
2896         rc = dt_declare_create(env, child, la, hint, dof, handle);
2897         if (rc != 0)
2898                 GOTO(stop, rc);
2899
2900         rc = dt_trans_start_local(env, dev, handle);
2901         if (rc != 0)
2902                 GOTO(stop, rc);
2903
2904         rc = dt_create(env, child, la, hint, dof, handle);
2905         dt_trans_stop(env, dev, handle);
2906         handle = NULL;
2907         if (rc != 0)
2908                 GOTO(log, rc);
2909
2910         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2911                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2912                               LCK_EX);
2913         if (rc != 0)
2914                 GOTO(log, rc);
2915
2916         /* The 2nd transaction. */
2917
2918         /* XXX: Generally, we should use bottom device (OSD) to update parent
2919          *      LOV EA. But because the LOD-object still references the wrong
2920          *      OSP-object that should be detached after the parent's LOV EA
2921          *      refreshed. Unfortunately, there is no suitable API for that.
2922          *      So we have to make the LOD to re-load the OSP-object(s) via
2923          *      replacing the LOV EA against the LOD-object.
2924          *
2925          *      Once the DNE2 patches have been landed, we can replace the
2926          *      LOD device with the OSD device. LU-6230. */
2927
2928         dev = lfsck->li_next;
2929         parent = lfsck_object_locate(dev, parent);
2930         if (IS_ERR(parent))
2931                 GOTO(log, rc = PTR_ERR(parent));
2932
2933         handle = dt_trans_create(env, dev);
2934         if (IS_ERR(handle))
2935                 GOTO(log, rc = PTR_ERR(handle));
2936
2937         rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2938                                   LU_XATTR_REPLACE, handle);
2939         if (rc != 0)
2940                 GOTO(stop, rc);
2941
2942         rc = dt_trans_start_local(env, dev, handle);
2943         if (rc != 0)
2944                 GOTO(stop, rc);
2945
2946         dt_write_lock(env, parent, 0);
2947         if (unlikely(lfsck_is_dead_obj(parent)))
2948                 GOTO(unlock, rc = 0);
2949
2950         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2951         if (unlikely(rc == 0 || rc == -ENODATA || rc == -ERANGE))
2952                 GOTO(unlock, rc = 0);
2953
2954         lmm = buf->lb_buf;
2955         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2956          * been verified in lfsck_layout_verify_header() already. If some
2957          * new magic introduced in the future, then layout LFSCK needs to
2958          * be updated also. */
2959         magic = le32_to_cpu(lmm->lmm_magic);
2960         if (magic == LOV_MAGIC_V1) {
2961                 objs = &lmm->lmm_objects[llr->llr_lov_idx];
2962         } else {
2963                 LASSERT(magic == LOV_MAGIC_V3);
2964                 objs =
2965                 &((struct lov_mds_md_v3 *)lmm)->lmm_objects[llr->llr_lov_idx];
2966         }
2967
2968         ostid_le_to_cpu(&objs->l_ost_oi, oi);
2969         index = le32_to_cpu(objs->l_ost_idx);
2970         rc = ostid_to_fid(&tfid, oi, index);
2971         /* Someone changed layout during the LFSCK, no need to repair then. */
2972         if (rc == 0 && !lu_fid_eq(&tfid, lu_object_fid(&llr->llr_child->do_lu)))
2973                 GOTO(unlock, rc = 0);
2974
2975         lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2976         fid_to_ostid(lu_object_fid(&child->do_lu), oi);
2977         ostid_cpu_to_le(oi, &objs->l_ost_oi);
2978         objs->l_ost_gen = cpu_to_le32(0);
2979         objs->l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
2980         lfsck_buf_init(&ea_buf, lmm,
2981                        lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
2982                                        magic));
2983         rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV,
2984                           LU_XATTR_REPLACE, handle);
2985
2986         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2987
2988 unlock:
2989         dt_write_unlock(env, parent);
2990
2991 stop:
2992         if (handle != NULL)
2993                 dt_trans_stop(env, dev, handle);
2994
2995 log:
2996         lfsck_ibits_unlock(&lh, LCK_EX);
2997         if (child != NULL)
2998                 lfsck_object_put(env, child);
2999
3000         if (rc != 0)
3001                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3002                        "multiple references for: parent "DFID", OST-index %u, "
3003                        "stripe-index %u, owner %u/%u: rc = %d\n",
3004                        lfsck_lfsck2name(lfsck), PFID(pfid), llr->llr_ost_idx,
3005                        llr->llr_lov_idx, la->la_uid, la->la_gid, rc);
3006
3007         return rc;
3008 }
3009
3010 /* If the MDT-object and the OST-object have different owner information,
3011  * then trust the MDT-object, because the normal chown/chgrp handle order
3012  * is from MDT to OST, and it is possible that some chown/chgrp operation
3013  * is partly done. */
3014 static int lfsck_layout_repair_owner(const struct lu_env *env,
3015                                      struct lfsck_component *com,
3016                                      struct dt_object *parent,
3017                                      struct lfsck_layout_req *llr,
3018                                      struct lu_attr *pla,
3019                                      const struct lu_attr *cla)
3020 {
3021         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3022         struct lu_attr                  *tla    = &info->lti_la2;
3023         struct dt_object                *child  = llr->llr_child;
3024         struct dt_device                *dev    = lfsck_obj2dev(child);
3025         struct thandle                  *handle;
3026         int                              rc;
3027         ENTRY;
3028
3029         tla->la_uid = pla->la_uid;
3030         tla->la_gid = pla->la_gid;
3031         tla->la_valid = LA_UID | LA_GID;
3032         handle = dt_trans_create(env, dev);
3033         if (IS_ERR(handle))
3034                 GOTO(log, rc = PTR_ERR(handle));
3035
3036         rc = dt_declare_attr_set(env, child, tla, handle);
3037         if (rc != 0)
3038                 GOTO(stop, rc);
3039
3040         rc = dt_trans_start_local(env, dev, handle);
3041         if (rc != 0)
3042                 GOTO(stop, rc);
3043
3044         /* Use the dt_object lock to serialize with destroy and attr_set. */
3045         dt_read_lock(env, parent, 0);
3046         if (unlikely(lfsck_is_dead_obj(parent)))
3047                 GOTO(unlock, rc = 1);
3048
3049         /* Get the latest parent's owner. */
3050         rc = dt_attr_get(env, parent, pla);
3051         if (rc != 0)
3052                 GOTO(unlock, rc);
3053
3054         /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
3055         if (unlikely(tla->la_uid != pla->la_uid ||
3056                      tla->la_gid != pla->la_gid))
3057                 rc = 1;
3058         else
3059                 rc = dt_attr_set(env, child, tla, handle);
3060
3061         GOTO(unlock, rc);
3062
3063 unlock:
3064         dt_read_unlock(env, parent);
3065
3066 stop:
3067         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3068
3069 log:
3070         if (rc != 0)
3071                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3072                        "inconsistent file owner for: parent "DFID", child "DFID
3073                        ", OST-index %u, stripe-index %u, old owner %u/%u, "
3074                        "new owner %u/%u: rc = %d\n",
3075                        lfsck_lfsck2name(com->lc_lfsck),
3076                        PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
3077                        llr->llr_ost_idx, llr->llr_lov_idx,
3078                        cla->la_uid, cla->la_gid, tla->la_uid, tla->la_gid, rc);
3079
3080         return rc;
3081 }
3082
3083 /* Check whether the OST-object correctly back points to the
3084  * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
3085 static int lfsck_layout_check_parent(const struct lu_env *env,
3086                                      struct lfsck_component *com,
3087                                      struct lfsck_assistant_object *lso,
3088                                      const struct lu_fid *pfid,
3089                                      const struct lu_fid *cfid,
3090                                      const struct lu_attr *cla,
3091                                      struct lfsck_layout_req *llr,
3092                                      struct lu_buf *lov_ea, __u32 idx)
3093 {
3094         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3095         struct lu_buf                   *buf    = &info->lti_big_buf;
3096         struct dt_object                *tobj;
3097         struct lov_mds_md_v1            *lmm;
3098         struct lov_ost_data_v1          *objs;
3099         struct lustre_handle             lh     = { 0 };
3100         int                              rc;
3101         int                              i;
3102         __u32                            magic;
3103         __u16                            count;
3104         ENTRY;
3105
3106         if (unlikely(!fid_is_sane(pfid)))
3107                 RETURN(LLIT_UNMATCHED_PAIR);
3108
3109         if (lu_fid_eq(pfid, &lso->lso_fid)) {
3110                 if (likely(llr->llr_lov_idx == idx))
3111                         RETURN(0);
3112
3113                 RETURN(LLIT_UNMATCHED_PAIR);
3114         }
3115
3116         tobj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3117         if (IS_ERR(tobj))
3118                 RETURN(PTR_ERR(tobj));
3119
3120         if (dt_object_exists(tobj) == 0 || lfsck_is_dead_obj(tobj) ||
3121             !S_ISREG(lfsck_object_type(tobj)))
3122                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3123
3124         /* Load the tobj's layout EA, in spite of it is a local MDT-object or
3125          * remote one on another MDT. Then check whether the given OST-object
3126          * is in such layout. If yes, it is multiple referenced, otherwise it
3127          * is unmatched referenced case. */
3128         rc = lfsck_layout_get_lovea(env, tobj, buf);
3129         if (rc == 0 || rc == -ENOENT)
3130                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3131
3132         if (rc < 0)
3133                 GOTO(out, rc);
3134
3135         lmm = buf->lb_buf;
3136         magic = le32_to_cpu(lmm->lmm_magic);
3137         if (magic == LOV_MAGIC_V1) {
3138                 objs = &lmm->lmm_objects[0];
3139         } else {
3140                 LASSERT(magic == LOV_MAGIC_V3);
3141                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3142         }
3143
3144         count = le16_to_cpu(lmm->lmm_stripe_count);
3145         for (i = 0; i < count; i++, objs++) {
3146                 struct lu_fid           *tfid   = &info->lti_fid2;
3147                 struct ost_id           *oi     = &info->lti_oi;
3148                 __u32                    idx2;
3149
3150                 if (lovea_slot_is_dummy(objs))
3151                         continue;
3152
3153                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3154                 idx2 = le32_to_cpu(objs->l_ost_idx);
3155                 rc = ostid_to_fid(tfid, oi, idx2);
3156                 if (rc != 0) {
3157                         CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
3158                                "invalid layout EA at the slot %d, index %u\n",
3159                                lfsck_lfsck2name(com->lc_lfsck),
3160                                PFID(pfid), i, idx2);
3161
3162                         GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3163                 }
3164
3165                 if (lu_fid_eq(cfid, tfid)) {
3166                         rc = lfsck_ibits_lock(env, com->lc_lfsck, tobj, &lh,
3167                                               MDS_INODELOCK_UPDATE |
3168                                               MDS_INODELOCK_LAYOUT |
3169                                               MDS_INODELOCK_XATTR,
3170                                               LCK_EX);
3171                         if (rc != 0)
3172                                 GOTO(out, rc);
3173
3174                         dt_read_lock(env, tobj, 0);
3175
3176                         /* For local MDT-object, re-check existence
3177                          * after taken the lock. */
3178                         if (!dt_object_remote(tobj)) {
3179                                 if (dt_object_exists(tobj) == 0 ||
3180                                     lfsck_is_dead_obj(tobj)) {
3181                                         rc = LLIT_UNMATCHED_PAIR;
3182                                 } else {
3183                                         *lov_ea = *buf;
3184                                         rc = LLIT_MULTIPLE_REFERENCED;
3185                                 }
3186
3187                                 GOTO(unlock, rc);
3188                         }
3189
3190                         /* For migration case, the new MDT-object and old
3191                          * MDT-object may reference the same OST-object at
3192                          * some migration internal time.
3193                          *
3194                          * For remote MDT-object, the local MDT may not know
3195                          * whether it has been removed or not.  Try checking
3196                          * for a non-existent xattr to check if this object
3197                          * has been been removed or not. */
3198                         rc = dt_xattr_get(env, tobj, &LU_BUF_NULL,
3199                                           XATTR_NAME_DUMMY);
3200                         if (unlikely(rc == -ENOENT || rc >= 0)) {
3201                                 rc = LLIT_UNMATCHED_PAIR;
3202                         } else if (rc == -ENODATA) {
3203                                 *lov_ea = *buf;
3204                                 rc = LLIT_MULTIPLE_REFERENCED;
3205                         }
3206
3207                         GOTO(unlock, rc);
3208                 }
3209         }
3210
3211         GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3212
3213 unlock:
3214         if (lustre_handle_is_used(&lh)) {
3215                 dt_read_unlock(env, tobj);
3216                 lfsck_ibits_unlock(&lh, LCK_EX);
3217         }
3218
3219 out:
3220         lfsck_object_put(env, tobj);
3221
3222         return rc;
3223 }
3224
3225 static int lfsck_layout_assistant_handler_p1(const struct lu_env *env,
3226                                              struct lfsck_component *com,
3227                                              struct lfsck_assistant_req *lar)
3228 {
3229         struct lfsck_layout_req              *llr    =
3230                         container_of0(lar, struct lfsck_layout_req, llr_lar);
3231         struct lfsck_assistant_object        *lso    = lar->lar_parent;
3232         struct lfsck_layout                  *lo     = com->lc_file_ram;
3233         struct lfsck_thread_info             *info   = lfsck_env_info(env);
3234         struct filter_fid_old                *pea    = &info->lti_old_pfid;
3235         struct lu_fid                        *pfid   = &info->lti_fid;
3236         struct lu_buf                         buf    = { NULL };
3237         struct dt_object                     *parent = NULL;
3238         struct dt_object                     *child  = llr->llr_child;
3239         struct lu_attr                       *pla    = &lso->lso_attr;
3240         struct lu_attr                       *cla    = &info->lti_la;
3241         struct lfsck_instance                *lfsck  = com->lc_lfsck;
3242         struct lfsck_bookmark                *bk     = &lfsck->li_bookmark_ram;
3243         enum lfsck_layout_inconsistency_type  type   = LLIT_NONE;
3244         __u32                                 idx    = 0;
3245         int                                   rc;
3246         ENTRY;
3247
3248         if (lso->lso_dead)
3249                 RETURN(0);
3250
3251         rc = dt_attr_get(env, child, cla);
3252         if (rc == -ENOENT) {
3253                 parent = lfsck_assistant_object_load(env, lfsck, lso);
3254                 if (IS_ERR(parent)) {
3255                         rc = PTR_ERR(parent);
3256
3257                         RETURN(rc == -ENOENT ? 0 : rc);
3258                 }
3259
3260                 type = LLIT_DANGLING;
3261                 goto repair;
3262         }
3263
3264         if (rc != 0)
3265                 GOTO(out, rc);
3266
3267         lfsck_buf_init(&buf, pea, sizeof(struct filter_fid_old));
3268         rc = dt_xattr_get(env, child, &buf, XATTR_NAME_FID);
3269         if (unlikely(rc > 0 && rc != sizeof(struct filter_fid_old) &&
3270                      rc != sizeof(struct filter_fid))) {
3271                 type = LLIT_UNMATCHED_PAIR;
3272                 goto repair;
3273         }
3274
3275         if (rc < 0 && rc != -ENODATA)
3276                 GOTO(out, rc);
3277
3278         if (rc == 0 || rc == -ENODATA)
3279                 GOTO(check_owner, rc = 0);
3280
3281         fid_le_to_cpu(pfid, &pea->ff_parent);
3282         /* Currently, the filter_fid::ff_parent::f_ver is not the
3283          * real parent MDT-object's FID::f_ver, instead it is the
3284          * OST-object index in its parent MDT-object's layout EA. */
3285         idx = pfid->f_stripe_idx;
3286         pfid->f_ver = 0;
3287         rc = lfsck_layout_check_parent(env, com, lso, pfid,
3288                                        lu_object_fid(&child->do_lu),
3289                                        cla, llr, &buf, idx);
3290         if (rc > 0) {
3291                 type = rc;
3292                 goto repair;
3293         }
3294
3295         if (rc < 0)
3296                 GOTO(out, rc);
3297
3298 check_owner:
3299         /* Someone may has changed the owner after the parent attr pre-loaded.
3300          * It can be handled later inside the lfsck_layout_repair_owner(). */
3301         if (unlikely(cla->la_uid != pla->la_uid ||
3302                      cla->la_gid != pla->la_gid)) {
3303                 type = LLIT_INCONSISTENT_OWNER;
3304                 goto repair;
3305         }
3306
3307 repair:
3308         if (type == LLIT_NONE)
3309                 GOTO(out, rc = 0);
3310
3311         if (bk->lb_param & LPF_DRYRUN)
3312                 GOTO(out, rc = 1);
3313
3314         if (parent == NULL) {
3315                 parent = lfsck_assistant_object_load(env, lfsck, lso);
3316                 if (IS_ERR(parent)) {
3317                         rc = PTR_ERR(parent);
3318
3319                         if (rc == -ENOENT)
3320                                 RETURN(0);
3321
3322                         GOTO(out, rc);
3323                 }
3324         }
3325
3326         switch (type) {
3327         case LLIT_DANGLING:
3328                 rc = lfsck_layout_repair_dangling(env, com, parent, llr, pla);
3329                 break;
3330         case LLIT_UNMATCHED_PAIR:
3331                 rc = lfsck_layout_repair_unmatched_pair(env, com, parent,
3332                                                         llr, pla);
3333                 break;
3334         case LLIT_MULTIPLE_REFERENCED:
3335                 rc = lfsck_layout_repair_multiple_references(env, com, parent,
3336                                                              llr, pla, &buf);
3337                 break;
3338         case LLIT_INCONSISTENT_OWNER:
3339                 rc = lfsck_layout_repair_owner(env, com, parent, llr, pla, cla);
3340                 break;
3341         default:
3342                 rc = 0;
3343                 break;
3344         }
3345
3346         GOTO(out, rc);
3347
3348 out:
3349         down_write(&com->lc_sem);
3350         if (rc < 0) {
3351                 struct lfsck_assistant_data *lad = com->lc_data;
3352
3353                 if (unlikely(lad->lad_exit)) {
3354                         rc = 0;
3355                 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
3356                            rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
3357                            rc == -EHOSTUNREACH) {
3358                         /* If cannot touch the target server,
3359                          * mark the LFSCK as INCOMPLETE. */
3360                         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
3361                                "talk with OST %x: rc = %d\n",
3362                                lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
3363                         lfsck_lad_set_bitmap(env, com, llr->llr_ost_idx);
3364                         lo->ll_objs_skipped++;
3365                         rc = 0;
3366                 } else {
3367                         lfsck_layout_record_failure(env, lfsck, lo);
3368                 }
3369         } else if (rc > 0) {
3370                 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
3371                          "unknown type = %d\n", type);
3372
3373                 lo->ll_objs_repaired[type - 1]++;
3374                 if (bk->lb_param & LPF_DRYRUN &&
3375                     unlikely(lo->ll_pos_first_inconsistent == 0))
3376                         lo->ll_pos_first_inconsistent =
3377                         lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
3378                                                         lfsck->li_di_oit);
3379         }
3380         up_write(&com->lc_sem);
3381
3382         if (parent != NULL && !IS_ERR(parent))
3383                 lfsck_object_put(env, parent);
3384
3385         return rc;
3386 }
3387
3388 static int lfsck_layout_assistant_handler_p2(const struct lu_env *env,
3389                                              struct lfsck_component *com)
3390 {
3391         struct lfsck_assistant_data     *lad    = com->lc_data;
3392         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3393         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
3394         struct lfsck_tgt_descs          *ltds   = &lfsck->li_ost_descs;
3395         struct lfsck_tgt_desc           *ltd;
3396         int                              rc     = 0;
3397         ENTRY;
3398
3399         CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan start\n",
3400                lfsck_lfsck2name(lfsck));
3401
3402         spin_lock(&ltds->ltd_lock);
3403         while (!list_empty(&lad->lad_ost_phase2_list)) {
3404                 ltd = list_entry(lad->lad_ost_phase2_list.next,
3405                                  struct lfsck_tgt_desc,
3406                                  ltd_layout_phase_list);
3407                 list_del_init(&ltd->ltd_layout_phase_list);
3408                 if (bk->lb_param & LPF_ALL_TGT) {
3409                         spin_unlock(&ltds->ltd_lock);
3410                         rc = lfsck_layout_scan_orphan(env, com, ltd);
3411                         if (rc != 0 && bk->lb_param & LPF_FAILOUT)
3412                                 RETURN(rc);
3413
3414                         if (unlikely(lad->lad_exit ||
3415                                      !thread_is_running(&lfsck->li_thread)))
3416                                 RETURN(0);
3417                         spin_lock(&ltds->ltd_lock);
3418                 }
3419         }
3420
3421         if (list_empty(&lad->lad_ost_phase1_list))
3422                 rc = 1;
3423         else
3424                 rc = 0;
3425         spin_unlock(&ltds->ltd_lock);
3426
3427         CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan stop: rc = %d\n",
3428                lfsck_lfsck2name(lfsck), rc);
3429
3430         RETURN(rc);
3431 }
3432
3433 static int
3434 lfsck_layout_slave_async_interpret(const struct lu_env *env,
3435                                    struct ptlrpc_request *req,
3436                                    void *args, int rc)
3437 {
3438         struct lfsck_layout_slave_async_args *llsaa = args;
3439         struct obd_export                    *exp   = llsaa->llsaa_exp;
3440         struct lfsck_component               *com   = llsaa->llsaa_com;
3441         struct lfsck_layout_slave_target     *llst  = llsaa->llsaa_llst;
3442         struct lfsck_layout_slave_data       *llsd  = com->lc_data;
3443         struct lfsck_reply                   *lr    = NULL;
3444         bool                                  done  = false;
3445
3446         if (rc != 0) {
3447                 /* It is probably caused by network trouble, or target crash,
3448                  * it will try several times (depends on the obd_timeout, and
3449                  * will not less than 3 times). But to make the LFSCK can go
3450                  * ahead, we should not try for ever. After some try but still
3451                  * hit failure, it will assume that the target exit the LFSCK
3452                  * prcoessing and stop try. */
3453                 if (rc == -ENOTCONN || rc == -ESHUTDOWN) {
3454                         int max_try = max_t(int, obd_timeout / 30, 3);
3455
3456                         if (++(llst->llst_failures) > max_try)
3457                                 done = true;
3458                 } else {
3459                         done = true;
3460                 }
3461         } else {
3462                 llst->llst_failures = 0;
3463                 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
3464                 if (lr->lr_status != LS_SCANNING_PHASE1 &&
3465                     lr->lr_status != LS_SCANNING_PHASE2)
3466                         done = true;
3467         }
3468
3469         if (done) {
3470                 CDEBUG(D_LFSCK, "%s: layout LFSCK slave gets the MDT %x "
3471                        "status %d, failures_try %d\n", lfsck_lfsck2name(com->lc_lfsck),
3472                        llst->llst_index, lr != NULL ? lr->lr_status : rc,
3473                        llst->llst_failures);
3474
3475                 lfsck_layout_llst_del(llsd, llst);
3476         }
3477
3478         lfsck_layout_llst_put(llst);
3479         lfsck_component_put(env, com);
3480         class_export_put(exp);
3481
3482         return 0;
3483 }
3484
3485 static int lfsck_layout_async_query(const struct lu_env *env,
3486                                     struct lfsck_component *com,
3487                                     struct obd_export *exp,
3488                                     struct lfsck_layout_slave_target *llst,
3489                                     struct lfsck_request *lr,
3490                                     struct ptlrpc_request_set *set)
3491 {
3492         struct lfsck_layout_slave_async_args *llsaa;
3493         struct ptlrpc_request                *req;
3494         struct lfsck_request                 *tmp;
3495         int                                   rc;
3496         ENTRY;
3497
3498         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
3499         if (req == NULL)
3500                 RETURN(-ENOMEM);
3501
3502         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
3503         if (rc != 0) {
3504                 ptlrpc_request_free(req);
3505                 RETURN(rc);
3506         }
3507
3508         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3509         *tmp = *lr;
3510         ptlrpc_request_set_replen(req);
3511
3512         llsaa = ptlrpc_req_async_args(req);
3513         llsaa->llsaa_exp = exp;
3514         llsaa->llsaa_com = lfsck_component_get(com);
3515         llsaa->llsaa_llst = llst;
3516         req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
3517         ptlrpc_set_add_req(set, req);
3518
3519         RETURN(0);
3520 }
3521
3522 static int lfsck_layout_async_notify(const struct lu_env *env,
3523                                      struct obd_export *exp,
3524                                      struct lfsck_request *lr,
3525                                      struct ptlrpc_request_set *set)
3526 {
3527         struct ptlrpc_request   *req;
3528         struct lfsck_request    *tmp;
3529         int                      rc;
3530         ENTRY;
3531
3532         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
3533         if (req == NULL)
3534                 RETURN(-ENOMEM);
3535
3536         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
3537         if (rc != 0) {
3538                 ptlrpc_request_free(req);
3539                 RETURN(rc);
3540         }
3541
3542         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3543         *tmp = *lr;
3544         ptlrpc_request_set_replen(req);
3545         ptlrpc_set_add_req(set, req);
3546
3547         RETURN(0);
3548 }
3549
3550 static int
3551 lfsck_layout_slave_query_master(const struct lu_env *env,
3552                                 struct lfsck_component *com)
3553 {
3554         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
3555         struct lfsck_instance            *lfsck = com->lc_lfsck;
3556         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3557         struct lfsck_layout_slave_target *llst;
3558         struct obd_export                *exp;
3559         struct ptlrpc_request_set        *set;
3560         int                               rc    = 0;
3561         int                               rc1   = 0;
3562         ENTRY;
3563
3564         set = ptlrpc_prep_set();
3565         if (set == NULL)
3566                 GOTO(log, rc = -ENOMEM);
3567
3568         memset(lr, 0, sizeof(*lr));
3569         lr->lr_event = LE_QUERY;
3570         lr->lr_active = LFSCK_TYPE_LAYOUT;
3571
3572         llsd->llsd_touch_gen++;
3573         spin_lock(&llsd->llsd_lock);
3574         while (!list_empty(&llsd->llsd_master_list)) {
3575                 llst = list_entry(llsd->llsd_master_list.next,
3576                                   struct lfsck_layout_slave_target,
3577                                   llst_list);
3578                 if (llst->llst_gen == llsd->llsd_touch_gen)
3579                         break;
3580
3581                 llst->llst_gen = llsd->llsd_touch_gen;
3582                 list_move_tail(&llst->llst_list,
3583                                &llsd->llsd_master_list);
3584                 atomic_inc(&llst->llst_ref);
3585                 spin_unlock(&llsd->llsd_lock);
3586
3587                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
3588                                                llst->llst_index);
3589                 if (exp == NULL) {
3590                         lfsck_layout_llst_del(llsd, llst);
3591                         lfsck_layout_llst_put(llst);
3592                         spin_lock(&llsd->llsd_lock);
3593                         continue;
3594                 }
3595
3596                 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
3597                 if (rc != 0) {
3598                         CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
3599                                "query %s for layout: rc = %d\n",
3600                                lfsck_lfsck2name(lfsck),
3601                                exp->exp_obd->obd_name, rc);
3602
3603                         rc1 = rc;
3604                         lfsck_layout_llst_put(llst);
3605                         class_export_put(exp);
3606                 }
3607                 spin_lock(&llsd->llsd_lock);
3608         }
3609         spin_unlock(&llsd->llsd_lock);
3610
3611         rc = ptlrpc_set_wait(set);
3612         ptlrpc_set_destroy(set);
3613
3614         GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
3615
3616 log:
3617         CDEBUG(D_LFSCK, "%s: layout LFSCK slave queries master: rc = %d\n",
3618                lfsck_lfsck2name(com->lc_lfsck), rc);
3619
3620         return rc;
3621 }
3622
3623 static void
3624 lfsck_layout_slave_notify_master(const struct lu_env *env,
3625                                  struct lfsck_component *com,
3626                                  enum lfsck_events event, int result)
3627 {
3628         struct lfsck_layout              *lo    = com->lc_file_ram;
3629         struct lfsck_instance            *lfsck = com->lc_lfsck;
3630         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3631         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
3632         struct lfsck_layout_slave_target *llst;
3633         struct obd_export                *exp;
3634         struct ptlrpc_request_set        *set;
3635         int                               rc;
3636         ENTRY;
3637
3638         CDEBUG(D_LFSCK, "%s: layout LFSCK slave notifies master\n",
3639                lfsck_lfsck2name(com->lc_lfsck));
3640
3641         set = ptlrpc_prep_set();
3642         if (set == NULL)
3643                 RETURN_EXIT;
3644
3645         memset(lr, 0, sizeof(*lr));
3646         lr->lr_event = event;
3647         lr->lr_flags = LEF_FROM_OST;
3648         lr->lr_status = result;
3649         lr->lr_index = lfsck_dev_idx(lfsck);
3650         lr->lr_active = LFSCK_TYPE_LAYOUT;
3651         lr->lr_flags2 = lo->ll_flags;
3652         llsd->llsd_touch_gen++;
3653         spin_lock(&llsd->llsd_lock);
3654         while (!list_empty(&llsd->llsd_master_list)) {
3655                 llst = list_entry(llsd->llsd_master_list.next,
3656                                   struct lfsck_layout_slave_target,
3657                                   llst_list);
3658                 if (llst->llst_gen == llsd->llsd_touch_gen)
3659                         break;
3660
3661                 llst->llst_gen = llsd->llsd_touch_gen;
3662                 list_move_tail(&llst->llst_list,
3663                                &llsd->llsd_master_list);
3664                 atomic_inc(&llst->llst_ref);
3665                 spin_unlock(&llsd->llsd_lock);
3666
3667                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
3668                                                llst->llst_index);
3669                 if (exp == NULL) {
3670                         lfsck_layout_llst_del(llsd, llst);
3671                         lfsck_layout_llst_put(llst);
3672                         spin_lock(&llsd->llsd_lock);
3673                         continue;
3674                 }
3675
3676                 rc = lfsck_layout_async_notify(env, exp, lr, set);
3677                 if (rc != 0)
3678                         CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
3679                                "notify %s for layout: rc = %d\n",
3680                                lfsck_lfsck2name(lfsck),
3681                                exp->exp_obd->obd_name, rc);
3682
3683                 lfsck_layout_llst_put(llst);
3684                 class_export_put(exp);
3685                 spin_lock(&llsd->llsd_lock);
3686         }
3687         spin_unlock(&llsd->llsd_lock);
3688
3689         ptlrpc_set_wait(set);
3690         ptlrpc_set_destroy(set);
3691
3692         RETURN_EXIT;
3693 }
3694
3695 /*
3696  * \ret -ENODATA: unrecognized stripe
3697  * \ret = 0     : recognized stripe
3698  * \ret < 0     : other failures
3699  */
3700 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
3701                                            struct lfsck_component *com,
3702                                            struct lu_fid *cfid,
3703                                            struct lu_fid *pfid)
3704 {
3705         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3706         struct lu_buf                   *buf    = &info->lti_big_buf;
3707         struct ost_id                   *oi     = &info->lti_oi;
3708         struct dt_object                *obj;
3709         struct lov_mds_md_v1            *lmm;
3710         struct lov_ost_data_v1          *objs;
3711         __u32                            idx    = pfid->f_stripe_idx;
3712         __u32                            magic;
3713         int                              rc     = 0;
3714         int                              i;
3715         __u16                            count;
3716         ENTRY;
3717
3718         pfid->f_ver = 0;
3719         obj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3720         if (IS_ERR(obj))
3721                 RETURN(PTR_ERR(obj));
3722
3723         dt_read_lock(env, obj, 0);
3724         if (unlikely(dt_object_exists(obj) == 0 ||
3725                      lfsck_is_dead_obj(obj)))
3726                 GOTO(unlock, rc = -ENOENT);
3727
3728         if (!S_ISREG(lfsck_object_type(obj)))
3729                 GOTO(unlock, rc = -ENODATA);
3730
3731         rc = lfsck_layout_get_lovea(env, obj, buf);
3732         if (rc < 0)
3733                 GOTO(unlock, rc);
3734
3735         if (rc == 0)
3736                 GOTO(unlock, rc = -ENODATA);
3737
3738         lmm = buf->lb_buf;
3739         rc = lfsck_layout_verify_header(lmm);
3740         if (rc != 0)
3741                 GOTO(unlock, rc);
3742
3743         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
3744          * been verified in lfsck_layout_verify_header() already. If some
3745          * new magic introduced in the future, then layout LFSCK needs to
3746          * be updated also. */
3747         magic = le32_to_cpu(lmm->lmm_magic);
3748         if (magic == LOV_MAGIC_V1) {
3749                 objs = &lmm->lmm_objects[0];
3750         } else {
3751                 LASSERT(magic == LOV_MAGIC_V3);
3752                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3753         }
3754
3755         fid_to_ostid(cfid, oi);
3756         count = le16_to_cpu(lmm->lmm_stripe_count);
3757         for (i = 0; i < count; i++, objs++) {
3758                 struct ost_id oi2;
3759
3760                 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
3761                 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
3762                         GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
3763         }
3764
3765         GOTO(unlock, rc = -ENODATA);
3766
3767 unlock:
3768         dt_read_unlock(env, obj);
3769         lfsck_object_put(env, obj);
3770
3771         return rc;
3772 }
3773
3774 /*
3775  * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
3776  * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
3777  * layout EA from MDT to OST. On one hand, the OST no need to understand
3778  * the layout EA structure; on the other hand, it may cause trouble when
3779  * transfer large layout EA from MDT to OST via normal OUT RPC.
3780  *
3781  * \ret > 0: unrecognized stripe
3782  * \ret = 0: recognized stripe
3783  * \ret < 0: other failures
3784  */
3785 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
3786                                           struct lfsck_component *com,
3787                                           struct lu_fid *cfid,
3788                                           struct lu_fid *pfid)
3789 {
3790         struct lfsck_instance    *lfsck  = com->lc_lfsck;
3791         struct obd_device        *obd    = lfsck->li_obd;
3792         struct seq_server_site   *ss     = lfsck_dev_site(lfsck);
3793         struct obd_export        *exp    = NULL;
3794         struct ptlrpc_request    *req    = NULL;
3795         struct lfsck_request     *lr;
3796         struct lu_seq_range      *range  = &lfsck_env_info(env)->lti_range;
3797         int                       rc     = 0;
3798         ENTRY;
3799
3800         if (unlikely(fid_is_idif(pfid)))
3801                 RETURN(1);
3802
3803         fld_range_set_any(range);
3804         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), range);
3805         if (rc != 0)
3806                 RETURN(rc == -ENOENT ? 1 : rc);
3807
3808         if (unlikely(!fld_range_is_mdt(range)))
3809                 RETURN(1);
3810
3811         exp = lustre_find_lwp_by_index(obd->obd_name, range->lsr_index);
3812         if (unlikely(exp == NULL))
3813                 RETURN(1);
3814
3815         if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
3816                 GOTO(out, rc = -EOPNOTSUPP);
3817
3818         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
3819         if (req == NULL)
3820                 GOTO(out, rc = -ENOMEM);
3821
3822         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
3823         if (rc != 0) {
3824                 ptlrpc_request_free(req);
3825
3826                 GOTO(out, rc);
3827         }
3828
3829         lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3830         memset(lr, 0, sizeof(*lr));
3831         lr->lr_event = LE_PAIRS_VERIFY;
3832         lr->lr_active = LFSCK_TYPE_LAYOUT;
3833         lr->lr_fid = *cfid; /* OST-object itself FID. */
3834         lr->lr_fid2 = *pfid; /* The claimed parent FID. */
3835
3836         ptlrpc_request_set_replen(req);
3837         rc = ptlrpc_queue_wait(req);
3838         ptlrpc_req_finished(req);
3839
3840         if (rc == -ENOENT || rc == -ENODATA)
3841                 rc = 1;
3842
3843         GOTO(out, rc);
3844
3845 out:
3846         if (exp != NULL)
3847                 class_export_put(exp);
3848
3849         return rc;
3850 }
3851
3852 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
3853                                           struct lfsck_component *com,
3854                                           struct lfsck_request *lr)
3855 {
3856         struct dt_object        *obj;
3857         int                      rc     = 0;
3858         ENTRY;
3859
3860         obj = lfsck_object_find_bottom(env, com->lc_lfsck, &lr->lr_fid);
3861         if (IS_ERR(obj))
3862                 GOTO(log, rc = PTR_ERR(obj));
3863
3864         dt_write_lock(env, obj, 0);
3865         if (unlikely(dt_object_exists(obj) == 0 ||
3866                      lfsck_is_dead_obj(obj)))
3867                 GOTO(unlock, rc = 0);
3868
3869         rc = __lfsck_layout_update_pfid(env, obj, &lr->lr_fid2,
3870                                         lr->lr_fid2.f_ver);
3871
3872         GOTO(unlock, rc);
3873
3874 unlock:
3875         dt_write_unlock(env, obj);
3876         lfsck_object_put(env, obj);
3877
3878 log:
3879         CDEBUG(D_LFSCK, "%s: layout LFSCK slave repaired pfid for "DFID
3880                ", parent "DFID": rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
3881                PFID(&lr->lr_fid), PFID(&lr->lr_fid2), rc);
3882
3883         return rc;
3884 }
3885
3886 /* layout APIs */
3887
3888 static void lfsck_layout_slave_quit(const struct lu_env *env,
3889                                     struct lfsck_component *com);
3890
3891 static int lfsck_layout_reset(const struct lu_env *env,
3892                               struct lfsck_component *com, bool init)
3893 {
3894         struct lfsck_layout     *lo    = com->lc_file_ram;
3895         int                      rc;
3896
3897         down_write(&com->lc_sem);
3898         if (init) {
3899                 memset(lo, 0, com->lc_file_size);
3900         } else {
3901                 __u32 count = lo->ll_success_count;
3902                 __u64 last_time = lo->ll_time_last_complete;
3903
3904                 memset(lo, 0, com->lc_file_size);
3905                 lo->ll_success_count = count;
3906                 lo->ll_time_last_complete = last_time;
3907         }
3908
3909         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
3910         lo->ll_status = LS_INIT;
3911
3912         if (com->lc_lfsck->li_master) {
3913                 struct lfsck_assistant_data *lad = com->lc_data;
3914
3915                 lad->lad_incomplete = 0;
3916                 CFS_RESET_BITMAP(lad->lad_bitmap);
3917         }
3918
3919         rc = lfsck_layout_store(env, com);
3920         up_write(&com->lc_sem);
3921
3922         CDEBUG(D_LFSCK, "%s: layout LFSCK reset: rc = %d\n",
3923                lfsck_lfsck2name(com->lc_lfsck), rc);
3924
3925         return rc;
3926 }
3927
3928 static void lfsck_layout_fail(const struct lu_env *env,
3929                               struct lfsck_component *com, bool new_checked)
3930 {
3931         struct lfsck_layout *lo = com->lc_file_ram;
3932
3933         down_write(&com->lc_sem);
3934         if (new_checked)
3935                 com->lc_new_checked++;
3936         lfsck_layout_record_failure(env, com->lc_lfsck, lo);
3937         up_write(&com->lc_sem);
3938 }
3939
3940 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
3941                                           struct lfsck_component *com, bool init)
3942 {
3943         struct lfsck_instance   *lfsck   = com->lc_lfsck;
3944         struct lfsck_layout     *lo      = com->lc_file_ram;
3945         int                      rc;
3946
3947         if (!init) {
3948                 rc = lfsck_checkpoint_generic(env, com);
3949                 if (rc != 0)
3950                         return rc > 0 ? 0 : rc;
3951         }
3952
3953         down_write(&com->lc_sem);
3954         if (init) {
3955                 lo->ll_pos_latest_start =
3956                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
3957         } else {
3958                 lo->ll_pos_last_checkpoint =
3959                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
3960                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3961                                 HALF_SEC - lfsck->li_time_last_checkpoint);
3962                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
3963                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
3964                 com->lc_new_checked = 0;
3965         }
3966
3967         rc = lfsck_layout_store(env, com);
3968         up_write(&com->lc_sem);
3969
3970         CDEBUG(D_LFSCK, "%s: layout LFSCK master checkpoint at the pos ["
3971                LPU64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
3972                lfsck->li_pos_current.lp_oit_cookie, rc);
3973
3974         return rc;
3975 }
3976
3977 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
3978                                          struct lfsck_component *com, bool init)
3979 {
3980         struct lfsck_instance   *lfsck = com->lc_lfsck;
3981         struct lfsck_layout     *lo    = com->lc_file_ram;
3982         int                      rc;
3983
3984         if (com->lc_new_checked == 0 && !init)
3985                 return 0;
3986
3987         down_write(&com->lc_sem);
3988         if (init) {
3989                 lo->ll_pos_latest_start =
3990                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
3991         } else {
3992                 lo->ll_pos_last_checkpoint =
3993                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
3994                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3995                                 HALF_SEC - lfsck->li_time_last_checkpoint);
3996                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
3997                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
3998                 com->lc_new_checked = 0;
3999         }
4000
4001         rc = lfsck_layout_store(env, com);
4002         up_write(&com->lc_sem);
4003
4004         CDEBUG(D_LFSCK, "%s: layout LFSCK slave checkpoint at the pos ["
4005                LPU64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
4006                lfsck->li_pos_current.lp_oit_cookie, rc);
4007
4008         return rc;
4009 }
4010
4011 static int lfsck_layout_prep(const struct lu_env *env,
4012                              struct lfsck_component *com,
4013                              struct lfsck_start *start)
4014 {
4015         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4016         struct lfsck_layout     *lo     = com->lc_file_ram;
4017         struct lfsck_position   *pos    = &com->lc_pos_start;
4018
4019         fid_zero(&pos->lp_dir_parent);
4020         pos->lp_dir_cookie = 0;
4021         if (lo->ll_status == LS_COMPLETED ||
4022             lo->ll_status == LS_PARTIAL ||
4023             /* To handle orphan, must scan from the beginning. */
4024             (start != NULL && start->ls_flags & LPF_OST_ORPHAN)) {
4025                 int rc;
4026
4027                 rc = lfsck_layout_reset(env, com, false);
4028                 if (rc == 0)
4029                         rc = lfsck_set_param(env, lfsck, start, true);
4030
4031                 if (rc != 0) {
4032                         CDEBUG(D_LFSCK, "%s: layout LFSCK prep failed: "
4033                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4034
4035                         return rc;
4036                 }
4037         }
4038
4039         down_write(&com->lc_sem);
4040         lo->ll_time_latest_start = cfs_time_current_sec();
4041         spin_lock(&lfsck->li_lock);
4042         if (lo->ll_flags & LF_SCANNED_ONCE) {
4043                 if (!lfsck->li_drop_dryrun ||
4044                     lo->ll_pos_first_inconsistent == 0) {
4045                         lo->ll_status = LS_SCANNING_PHASE2;
4046                         list_move_tail(&com->lc_link,
4047                                        &lfsck->li_list_double_scan);
4048                         pos->lp_oit_cookie = 0;
4049                 } else {
4050                         int i;
4051
4052                         lo->ll_status = LS_SCANNING_PHASE1;
4053                         lo->ll_run_time_phase1 = 0;
4054                         lo->ll_run_time_phase2 = 0;
4055                         lo->ll_objs_checked_phase1 = 0;
4056                         lo->ll_objs_checked_phase2 = 0;
4057                         lo->ll_objs_failed_phase1 = 0;
4058                         lo->ll_objs_failed_phase2 = 0;
4059                         for (i = 0; i < LLIT_MAX; i++)
4060                                 lo->ll_objs_repaired[i] = 0;
4061
4062                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4063                         fid_zero(&com->lc_fid_latest_scanned_phase2);
4064                 }
4065         } else {
4066                 lo->ll_status = LS_SCANNING_PHASE1;
4067                 if (!lfsck->li_drop_dryrun ||
4068                     lo->ll_pos_first_inconsistent == 0)
4069                         pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
4070                 else
4071                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4072         }
4073         spin_unlock(&lfsck->li_lock);
4074         up_write(&com->lc_sem);
4075
4076         return 0;
4077 }
4078
4079 static int lfsck_layout_slave_prep(const struct lu_env *env,
4080                                    struct lfsck_component *com,
4081                                    struct lfsck_start_param *lsp)
4082 {
4083         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4084         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4085         struct lfsck_layout             *lo     = com->lc_file_ram;
4086         struct lfsck_start              *start  = lsp->lsp_start;
4087         int                              rc;
4088
4089         rc = lfsck_layout_prep(env, com, start);
4090         if (rc != 0)
4091                 return rc;
4092
4093         if (lo->ll_flags & LF_CRASHED_LASTID &&
4094             list_empty(&llsd->llsd_master_list)) {
4095                 LASSERT(lfsck->li_out_notify != NULL);
4096
4097                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4098                                      LE_LASTID_REBUILDING);
4099         }
4100
4101         if (!lsp->lsp_index_valid)
4102                 return 0;
4103
4104         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
4105         if (rc == 0 && start != NULL && start->ls_flags & LPF_OST_ORPHAN) {
4106                 LASSERT(!llsd->llsd_rbtree_valid);
4107
4108                 write_lock(&llsd->llsd_rb_lock);
4109                 rc = lfsck_rbtree_setup(env, com);
4110                 write_unlock(&llsd->llsd_rb_lock);
4111         }
4112
4113         CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
4114                LPU64"]\n", lfsck_lfsck2name(lfsck),
4115                com->lc_pos_start.lp_oit_cookie);
4116
4117         return rc;
4118 }
4119
4120 static int lfsck_layout_master_prep(const struct lu_env *env,
4121                                     struct lfsck_component *com,
4122                                     struct lfsck_start_param *lsp)
4123 {
4124         int rc;
4125         ENTRY;
4126
4127         rc = lfsck_layout_load_bitmap(env, com);
4128         if (rc != 0) {
4129                 rc = lfsck_layout_reset(env, com, false);
4130                 if (rc == 0)
4131                         rc = lfsck_set_param(env, com->lc_lfsck,
4132                                              lsp->lsp_start, true);
4133
4134                 if (rc != 0)
4135                         GOTO(log, rc);
4136         }
4137
4138         rc = lfsck_layout_prep(env, com, lsp->lsp_start);
4139         if (rc != 0)
4140                 RETURN(rc);
4141
4142         rc = lfsck_start_assistant(env, com, lsp);
4143
4144         GOTO(log, rc);
4145
4146 log:
4147         CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
4148                LPU64"]\n", lfsck_lfsck2name(com->lc_lfsck),
4149                com->lc_pos_start.lp_oit_cookie);
4150
4151         return 0;
4152 }
4153
4154 /* Pre-fetch the attribute for each stripe in the given layout EA. */
4155 static int lfsck_layout_scan_stripes(const struct lu_env *env,
4156                                      struct lfsck_component *com,
4157                                      struct dt_object *parent,
4158                                      struct lov_mds_md_v1 *lmm)
4159 {
4160         struct lfsck_thread_info        *info    = lfsck_env_info(env);
4161         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4162         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
4163         struct lfsck_layout             *lo      = com->lc_file_ram;
4164         struct lfsck_assistant_data     *lad     = com->lc_data;
4165         struct lfsck_assistant_object   *lso     = NULL;
4166         struct lov_ost_data_v1          *objs;
4167         struct lfsck_tgt_descs          *ltds    = &lfsck->li_ost_descs;
4168         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4169         struct ptlrpc_thread            *athread = &lad->lad_thread;
4170         struct l_wait_info               lwi     = { 0 };
4171         struct lu_buf                    buf;
4172         int                              rc      = 0;
4173         int                              i;
4174         __u32                            magic;
4175         __u16                            count;
4176         ENTRY;
4177
4178         lfsck_buf_init(&buf, &info->lti_old_pfid,
4179                        sizeof(struct filter_fid_old));
4180         count = le16_to_cpu(lmm->lmm_stripe_count);
4181         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
4182          * been verified in lfsck_layout_verify_header() already. If some
4183          * new magic introduced in the future, then layout LFSCK needs to
4184          * be updated also. */
4185         magic = le32_to_cpu(lmm->lmm_magic);
4186         if (magic == LOV_MAGIC_V1) {
4187                 objs = &lmm->lmm_objects[0];
4188         } else {
4189                 LASSERT(magic == LOV_MAGIC_V3);
4190                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4191         }
4192
4193         for (i = 0; i < count; i++, objs++) {
4194                 struct lu_fid           *fid    = &info->lti_fid;
4195                 struct ost_id           *oi     = &info->lti_oi;
4196                 struct lfsck_layout_req *llr;
4197                 struct lfsck_tgt_desc   *tgt    = NULL;
4198                 struct dt_object        *cobj   = NULL;
4199                 __u32                    index;
4200                 bool                     wakeup = false;
4201
4202                 if (unlikely(lovea_slot_is_dummy(objs)))
4203                         continue;
4204
4205                 l_wait_event(mthread->t_ctl_waitq,
4206                              lad->lad_prefetched < bk->lb_async_windows ||
4207                              !thread_is_running(mthread) ||
4208                              thread_is_stopped(athread),
4209                              &lwi);
4210
4211                 if (unlikely(!thread_is_running(mthread)) ||
4212                              thread_is_stopped(athread))
4213                         GOTO(out, rc = 0);
4214
4215                 if (unlikely(lfsck_is_dead_obj(parent)))
4216                         GOTO(out, rc = 0);
4217
4218                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4219                 index = le32_to_cpu(objs->l_ost_idx);
4220                 rc = ostid_to_fid(fid, oi, index);
4221                 if (rc != 0) {
4222                         CDEBUG(D_LFSCK, "%s: get invalid layout EA for "DFID
4223                                ": "DOSTID", idx:%u\n", lfsck_lfsck2name(lfsck),
4224                                PFID(lfsck_dto2fid(parent)), POSTID(oi), index);
4225                         goto next;
4226                 }
4227
4228                 tgt = lfsck_tgt_get(ltds, index);
4229                 if (unlikely(tgt == NULL)) {
4230                         CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
4231                                "did not join the layout LFSCK\n",
4232                                lfsck_lfsck2name(lfsck), index);
4233                         lfsck_lad_set_bitmap(env, com, index);
4234                         goto next;
4235                 }
4236
4237                 /* There is potential deadlock race condition between object
4238                  * destroy and layout LFSCK. Consider the following scenario:
4239                  *
4240                  * 1) The LFSCK thread obtained the parent object firstly, at
4241                  *    that time, the parent object has not been destroyed yet.
4242                  *
4243                  * 2) One RPC service thread destroyed the parent and all its
4244                  *    children objects. Because the LFSCK is referencing the
4245                  *    parent object, then the parent object will be marked as
4246                  *    dying in RAM. On the other hand, the parent object is
4247                  *    referencing all its children objects, then all children
4248                  *    objects will be marked as dying in RAM also.
4249                  *
4250                  * 3) The LFSCK thread tries to find some child object with
4251                  *    the parent object referenced. Then it will find that the
4252                  *    child object is dying. According to the object visibility
4253                  *    rules: the object with dying flag cannot be returned to
4254                  *    others. So the LFSCK thread has to wait until the dying
4255                  *    object has been purged from RAM, then it can allocate a
4256                  *    new object (with the same FID) in RAM. Unfortunately, the
4257                  *    LFSCK thread itself is referencing the parent object, and
4258                  *    cause the parent object cannot be purged, then cause the
4259                  *    child object cannot be purged also. So the LFSCK thread
4260                  *    will fall into deadlock.
4261                  *
4262                  * We introduce non-blocked version lu_object_find() to allow
4263                  * the LFSCK thread to return failure immediately (instead of
4264                  * wait) when it finds dying (child) object, then the LFSCK
4265                  * thread can check whether the parent object is dying or not.
4266                  * So avoid above deadlock. LU-5395 */
4267                 cobj = lfsck_object_find_by_dev_nowait(env, tgt->ltd_tgt, fid);
4268                 if (IS_ERR(cobj)) {
4269                         if (lfsck_is_dead_obj(parent)) {
4270                                 lfsck_tgt_put(tgt);
4271
4272                                 GOTO(out, rc = 0);
4273                         }
4274
4275                         rc = PTR_ERR(cobj);
4276                         goto next;
4277                 }
4278
4279                 rc = dt_declare_attr_get(env, cobj);
4280                 if (rc != 0)
4281                         goto next;
4282
4283                 rc = dt_declare_xattr_get(env, cobj, &buf, XATTR_NAME_FID);
4284                 if (rc != 0)
4285                         goto next;
4286
4287                 if (lso == NULL) {
4288                         struct lu_attr *attr = &info->lti_la;
4289
4290                         rc = dt_attr_get(env, parent, attr);
4291                         if (rc != 0)
4292                                 goto next;
4293
4294                         lso = lfsck_assistant_object_init(env,
4295                                 lfsck_dto2fid(parent), attr,
4296                                 lfsck->li_pos_current.lp_oit_cookie, false);
4297                         if (IS_ERR(lso)) {
4298                                 rc = PTR_ERR(lso);
4299                                 lso = NULL;
4300
4301                                 goto next;
4302                         }
4303                 }
4304
4305                 llr = lfsck_layout_assistant_req_init(lso, cobj, index, i);
4306                 if (IS_ERR(llr)) {
4307                         rc = PTR_ERR(llr);
4308                         goto next;
4309                 }
4310
4311                 cobj = NULL;
4312                 spin_lock(&lad->lad_lock);
4313                 if (lad->lad_assistant_status < 0) {
4314                         spin_unlock(&lad->lad_lock);
4315                         lfsck_layout_assistant_req_fini(env, &llr->llr_lar);
4316                         lfsck_tgt_put(tgt);
4317                         RETURN(lad->lad_assistant_status);
4318                 }
4319
4320                 list_add_tail(&llr->llr_lar.lar_list, &lad->lad_req_list);
4321                 if (lad->lad_prefetched == 0)
4322                         wakeup = true;
4323
4324                 lad->lad_prefetched++;
4325                 spin_unlock(&lad->lad_lock);
4326                 if (wakeup)
4327                         wake_up_all(&athread->t_ctl_waitq);
4328
4329 next:
4330                 down_write(&com->lc_sem);
4331                 com->lc_new_checked++;
4332                 if (rc < 0)
4333                         lfsck_layout_record_failure(env, lfsck, lo);
4334                 up_write(&com->lc_sem);
4335
4336                 if (cobj != NULL && !IS_ERR(cobj))
4337                         lfsck_object_put(env, cobj);
4338
4339                 if (likely(tgt != NULL))
4340                         lfsck_tgt_put(tgt);
4341
4342                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4343                         GOTO(out, rc);
4344         }
4345
4346         GOTO(out, rc = 0);
4347
4348 out:
4349         if (lso != NULL)
4350                 lfsck_assistant_object_put(env, lso);
4351
4352         return rc;
4353 }
4354
4355 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
4356  * the OST-object's attribute and generate an structure lfsck_layout_req on the
4357  * list ::lad_req_list.
4358  *
4359  * For each request on above list, the lfsck_layout_assistant thread compares
4360  * the OST side attribute with local attribute, if inconsistent, then repair it.
4361  *
4362  * All above processing is async mode with pipeline. */
4363 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
4364                                         struct lfsck_component *com,
4365                                         struct dt_object *obj)
4366 {
4367         struct lfsck_thread_info        *info   = lfsck_env_info(env);
4368         struct ost_id                   *oi     = &info->lti_oi;
4369         struct lfsck_layout             *lo     = com->lc_file_ram;
4370         struct lfsck_assistant_data     *lad    = com->lc_data;
4371         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4372         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
4373         struct thandle                  *handle = NULL;
4374         struct lu_buf                   *buf    = &info->lti_big_buf;
4375         struct lov_mds_md_v1            *lmm    = NULL;
4376         struct dt_device                *dev    = lfsck_obj2dev(obj);
4377         struct lustre_handle             lh     = { 0 };
4378         struct lu_buf                    ea_buf = { NULL };
4379         int                              rc     = 0;
4380         int                              size   = 0;
4381         bool                             locked = false;
4382         bool                             stripe = false;
4383         bool                             bad_oi = false;
4384         ENTRY;
4385
4386         if (!S_ISREG(lfsck_object_type(obj)))
4387                 GOTO(out, rc = 0);
4388
4389         if (lad->lad_assistant_status < 0)
4390                 GOTO(out, rc = -ESRCH);
4391
4392         fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
4393         lmm_oi_cpu_to_le(oi, oi);
4394         dt_read_lock(env, obj, 0);
4395         locked = true;
4396
4397 again:
4398         if (dt_object_exists(obj) == 0 ||
4399             lfsck_is_dead_obj(obj))
4400                 GOTO(out, rc = 0);
4401
4402         rc = lfsck_layout_get_lovea(env, obj, buf);
4403         if (rc <= 0)
4404                 GOTO(out, rc);
4405
4406         size = rc;
4407         lmm = buf->lb_buf;
4408         rc = lfsck_layout_verify_header(lmm);
4409         /* If the LOV EA crashed, then it is possible to be rebuilt later
4410          * when handle orphan OST-objects. */
4411         if (rc != 0)
4412                 GOTO(out, rc);
4413
4414         if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0)
4415                 GOTO(out, stripe = true);
4416
4417         /* Inconsistent lmm_oi, should be repaired. */
4418         bad_oi = true;
4419         lmm->lmm_oi = *oi;
4420
4421         if (bk->lb_param & LPF_DRYRUN) {
4422                 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4423
4424                 GOTO(out, stripe = true);
4425         }
4426
4427         if (!lustre_handle_is_used(&lh)) {
4428                 dt_read_unlock(env, obj);
4429                 locked = false;
4430                 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4431                                       MDS_INODELOCK_LAYOUT |
4432                                       MDS_INODELOCK_XATTR, LCK_EX);
4433                 if (rc != 0)
4434                         GOTO(out, rc);
4435
4436                 handle = dt_trans_create(env, dev);
4437                 if (IS_ERR(handle))
4438                         GOTO(out, rc = PTR_ERR(handle));
4439
4440                 lfsck_buf_init(&ea_buf, lmm, size);
4441                 rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
4442                                           LU_XATTR_REPLACE, handle);
4443                 if (rc != 0)
4444                         GOTO(out, rc);
4445
4446                 rc = dt_trans_start_local(env, dev, handle);
4447                 if (rc != 0)
4448                         GOTO(out, rc);
4449
4450                 dt_write_lock(env, obj, 0);
4451                 locked = true;
4452
4453                 goto again;
4454         }
4455
4456         rc = dt_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
4457                           LU_XATTR_REPLACE, handle);
4458         if (rc != 0)
4459                 GOTO(out, rc);
4460
4461         lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4462
4463         GOTO(out, stripe = true);
4464
4465 out:
4466         if (locked) {
4467                 if (lustre_handle_is_used(&lh))
4468                         dt_write_unlock(env, obj);
4469                 else
4470                         dt_read_unlock(env, obj);
4471         }
4472
4473         if (handle != NULL && !IS_ERR(handle))
4474                 dt_trans_stop(env, dev, handle);
4475
4476         lfsck_ibits_unlock(&lh, LCK_EX);
4477
4478         if (bad_oi)
4479                 CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
4480                        DFID": rc = %d\n", lfsck_lfsck2name(lfsck),
4481                        bk->lb_param & LPF_DRYRUN ? "found" : "repaired",
4482                        PFID(lfsck_dto2fid(obj)), rc);
4483
4484         if (stripe) {
4485                 rc = lfsck_layout_scan_stripes(env, com, obj, lmm);
4486         } else {
4487                 down_write(&com->lc_sem);
4488                 com->lc_new_checked++;
4489                 if (rc < 0)
4490                         lfsck_layout_record_failure(env, lfsck, lo);
4491                 up_write(&com->lc_sem);
4492         }
4493
4494         return rc;
4495 }
4496
4497 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
4498                                        struct lfsck_component *com,
4499                                        struct dt_object *obj)
4500 {
4501         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4502         struct lfsck_layout             *lo     = com->lc_file_ram;
4503         const struct lu_fid             *fid    = lfsck_dto2fid(obj);
4504         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4505         struct lfsck_layout_seq         *lls;
4506         __u64                            seq;
4507         __u64                            oid;
4508         int                              rc;
4509         ENTRY;
4510
4511         LASSERT(llsd != NULL);
4512
4513         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
4514             cfs_fail_val == lfsck_dev_idx(lfsck)) {
4515                 struct l_wait_info       lwi = LWI_TIMEOUT(cfs_time_seconds(1),
4516                                                            NULL, NULL);
4517                 struct ptlrpc_thread    *thread = &lfsck->li_thread;
4518
4519                 l_wait_event(thread->t_ctl_waitq,
4520                              !thread_is_running(thread),
4521                              &lwi);
4522         }
4523
4524         lfsck_rbtree_update_bitmap(env, com, fid, false);
4525
4526         down_write(&com->lc_sem);
4527         if (fid_is_idif(fid))
4528                 seq = 0;
4529         else if (!fid_is_norm(fid) ||
4530                  !fid_is_for_ostobj(env, lfsck, obj, fid))
4531                 GOTO(unlock, rc = 0);
4532         else
4533                 seq = fid_seq(fid);
4534         com->lc_new_checked++;
4535
4536         lls = lfsck_layout_seq_lookup(llsd, seq);
4537         if (lls == NULL) {
4538                 OBD_ALLOC_PTR(lls);
4539                 if (unlikely(lls == NULL))
4540                         GOTO(unlock, rc = -ENOMEM);
4541
4542                 INIT_LIST_HEAD(&lls->lls_list);
4543                 lls->lls_seq = seq;
4544                 rc = lfsck_layout_lastid_load(env, com, lls);
4545                 if (rc != 0) {
4546                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
4547                               "load LAST_ID for "LPX64": rc = %d\n",
4548                               lfsck_lfsck2name(com->lc_lfsck), seq, rc);
4549                         lo->ll_objs_failed_phase1++;
4550                         OBD_FREE_PTR(lls);
4551                         GOTO(unlock, rc);
4552                 }
4553
4554                 lfsck_layout_seq_insert(llsd, lls);
4555         }
4556
4557         if (unlikely(fid_is_last_id(fid)))
4558                 GOTO(unlock, rc = 0);
4559
4560         if (fid_is_idif(fid))
4561                 oid = fid_idif_id(fid_seq(fid), fid_oid(fid), fid_ver(fid));
4562         else
4563                 oid = fid_oid(fid);
4564
4565         if (oid > lls->lls_lastid_known)
4566                 lls->lls_lastid_known = oid;
4567
4568         if (oid > lls->lls_lastid) {
4569                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
4570                         /* OFD may create new objects during LFSCK scanning. */
4571                         rc = lfsck_layout_lastid_reload(env, com, lls);
4572                         if (unlikely(rc != 0)) {
4573                                 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
4574                                       "reload LAST_ID for "LPX64": rc = %d\n",
4575                                       lfsck_lfsck2name(com->lc_lfsck),
4576                                       lls->lls_seq, rc);
4577
4578                                 GOTO(unlock, rc);
4579                         }
4580
4581                         if (oid <= lls->lls_lastid ||
4582                             lo->ll_flags & LF_CRASHED_LASTID)
4583                                 GOTO(unlock, rc = 0);
4584
4585                         LASSERT(lfsck->li_out_notify != NULL);
4586
4587                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4588                                              LE_LASTID_REBUILDING);
4589                         lo->ll_flags |= LF_CRASHED_LASTID;
4590
4591                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
4592                                "LAST_ID file (2) for the sequence "LPX64
4593                                ", old value "LPU64", known value "LPU64"\n",
4594                                lfsck_lfsck2name(lfsck), lls->lls_seq,
4595                                lls->lls_lastid, oid);
4596                 }
4597
4598                 lls->lls_lastid = oid;
4599                 lls->lls_dirty = 1;
4600         }
4601
4602         GOTO(unlock, rc = 0);
4603
4604 unlock:
4605         up_write(&com->lc_sem);
4606
4607         return rc;
4608 }
4609
4610 static int lfsck_layout_exec_dir(const struct lu_env *env,
4611                                  struct lfsck_component *com,
4612                                  struct lfsck_assistant_object *lso,
4613                                  struct lu_dirent *ent, __u16 type)
4614 {
4615         return 0;
4616 }
4617
4618 static int lfsck_layout_master_post(const struct lu_env *env,
4619                                     struct lfsck_component *com,
4620                                     int result, bool init)
4621 {
4622         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4623         struct lfsck_layout     *lo     = com->lc_file_ram;
4624         int                      rc;
4625         ENTRY;
4626
4627         lfsck_post_generic(env, com, &result);
4628
4629         down_write(&com->lc_sem);
4630         spin_lock(&lfsck->li_lock);
4631         if (!init)
4632                 lo->ll_pos_last_checkpoint =
4633                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
4634
4635         if (result > 0) {
4636                 if (lo->ll_flags & LF_INCOMPLETE)
4637                         lo->ll_status = LS_PARTIAL;
4638                 else
4639                         lo->ll_status = LS_SCANNING_PHASE2;
4640                 lo->ll_flags |= LF_SCANNED_ONCE;
4641                 lo->ll_flags &= ~LF_UPGRADE;
4642                 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4643         } else if (result == 0) {
4644                 if (lfsck->li_status != 0)
4645                         lo->ll_status = lfsck->li_status;
4646                 else
4647                         lo->ll_status = LS_STOPPED;
4648                 if (lo->ll_status != LS_PAUSED)
4649                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4650         } else {
4651                 lo->ll_status = LS_FAILED;
4652                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4653         }
4654         spin_unlock(&lfsck->li_lock);
4655
4656         if (!init) {
4657                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4658                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4659                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4660                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4661                 com->lc_new_checked = 0;
4662         }
4663
4664         rc = lfsck_layout_store(env, com);
4665         up_write(&com->lc_sem);
4666
4667         CDEBUG(D_LFSCK, "%s: layout LFSCK master post done: rc = %d\n",
4668                lfsck_lfsck2name(lfsck), rc);
4669
4670         RETURN(rc);
4671 }
4672
4673 static int lfsck_layout_slave_post(const struct lu_env *env,
4674                                    struct lfsck_component *com,
4675                                    int result, bool init)
4676 {
4677         struct lfsck_instance   *lfsck = com->lc_lfsck;
4678         struct lfsck_layout     *lo    = com->lc_file_ram;
4679         int                      rc;
4680         bool                     done  = false;
4681
4682         rc = lfsck_layout_lastid_store(env, com);
4683         if (rc != 0)
4684                 result = rc;
4685
4686         LASSERT(lfsck->li_out_notify != NULL);
4687
4688         down_write(&com->lc_sem);
4689         spin_lock(&lfsck->li_lock);
4690         if (!init)
4691                 lo->ll_pos_last_checkpoint =
4692                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
4693
4694         if (result > 0) {
4695                 lo->ll_status = LS_SCANNING_PHASE2;
4696                 lo->ll_flags |= LF_SCANNED_ONCE;
4697                 if (lo->ll_flags & LF_CRASHED_LASTID) {
4698                         done = true;
4699                         lo->ll_flags &= ~LF_CRASHED_LASTID;
4700
4701                         CDEBUG(D_LFSCK, "%s: layout LFSCK has rebuilt "
4702                                "crashed LAST_ID files successfully\n",
4703                                lfsck_lfsck2name(lfsck));
4704                 }
4705                 lo->ll_flags &= ~LF_UPGRADE;
4706                 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4707         } else if (result == 0) {
4708                 if (lfsck->li_status != 0)
4709                         lo->ll_status = lfsck->li_status;
4710                 else
4711                         lo->ll_status = LS_STOPPED;
4712                 if (lo->ll_status != LS_PAUSED)
4713                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4714         } else {
4715                 lo->ll_status = LS_FAILED;
4716                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4717         }
4718         spin_unlock(&lfsck->li_lock);
4719
4720         if (done)
4721                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4722                                      LE_LASTID_REBUILT);
4723
4724         if (!init) {
4725                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4726                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4727                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4728                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4729                 com->lc_new_checked = 0;
4730         }
4731
4732         rc = lfsck_layout_store(env, com);
4733         up_write(&com->lc_sem);
4734
4735         lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
4736
4737         CDEBUG(D_LFSCK, "%s: layout LFSCK slave post done: rc = %d\n",
4738                lfsck_lfsck2name(lfsck), rc);
4739
4740         return rc;
4741 }
4742
4743 static int lfsck_layout_dump(const struct lu_env *env,
4744                              struct lfsck_component *com, struct seq_file *m)
4745 {
4746         struct lfsck_instance   *lfsck = com->lc_lfsck;
4747         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
4748         struct lfsck_layout     *lo    = com->lc_file_ram;
4749         int                      rc;
4750
4751         down_read(&com->lc_sem);
4752         seq_printf(m, "name: lfsck_layout\n"
4753                       "magic: %#x\n"
4754                       "version: %d\n"
4755                       "status: %s\n",
4756                       lo->ll_magic,
4757                       bk->lb_version,
4758                       lfsck_status2names(lo->ll_status));
4759
4760         rc = lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
4761         if (rc < 0)
4762                 goto out;
4763
4764         rc = lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4765         if (rc < 0)
4766                 goto out;
4767
4768         rc = lfsck_time_dump(m, lo->ll_time_last_complete,
4769                              "last_completed");
4770         if (rc < 0)
4771                 goto out;
4772
4773         rc = lfsck_time_dump(m, lo->ll_time_latest_start,
4774                              "latest_start");
4775         if (rc < 0)
4776                 goto out;
4777
4778         rc = lfsck_time_dump(m, lo->ll_time_last_checkpoint,
4779                              "last_checkpoint");
4780         if (rc < 0)
4781                 goto out;
4782
4783         seq_printf(m, "latest_start_position: "LPU64"\n"
4784                       "last_checkpoint_position: "LPU64"\n"
4785                       "first_failure_position: "LPU64"\n",
4786                       lo->ll_pos_latest_start,
4787                       lo->ll_pos_last_checkpoint,
4788                       lo->ll_pos_first_inconsistent);
4789
4790         seq_printf(m, "success_count: %u\n"
4791                       "repaired_dangling: "LPU64"\n"
4792                       "repaired_unmatched_pair: "LPU64"\n"
4793                       "repaired_multiple_referenced: "LPU64"\n"
4794                       "repaired_orphan: "LPU64"\n"
4795                       "repaired_inconsistent_owner: "LPU64"\n"
4796                       "repaired_others: "LPU64"\n"
4797                       "skipped: "LPU64"\n"
4798                       "failed_phase1: "LPU64"\n"
4799                       "failed_phase2: "LPU64"\n",
4800                       lo->ll_success_count,
4801                       lo->ll_objs_repaired[LLIT_DANGLING - 1],
4802                       lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
4803                       lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
4804                       lo->ll_objs_repaired[LLIT_ORPHAN - 1],
4805                       lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
4806                       lo->ll_objs_repaired[LLIT_OTHERS - 1],
4807                       lo->ll_objs_skipped,
4808                       lo->ll_objs_failed_phase1,
4809                       lo->ll_objs_failed_phase2);
4810
4811         if (lo->ll_status == LS_SCANNING_PHASE1) {
4812                 __u64 pos;
4813                 const struct dt_it_ops *iops;
4814                 cfs_duration_t duration = cfs_time_current() -
4815                                           lfsck->li_time_last_checkpoint;
4816                 __u64 checked = lo->ll_objs_checked_phase1 +
4817                                 com->lc_new_checked;
4818                 __u64 speed = checked;
4819                 __u64 new_checked = com->lc_new_checked *
4820                                     msecs_to_jiffies(MSEC_PER_SEC);
4821                 __u32 rtime = lo->ll_run_time_phase1 +
4822                               cfs_duration_sec(duration + HALF_SEC);
4823
4824                 if (duration != 0)
4825                         do_div(new_checked, duration);
4826                 if (rtime != 0)
4827                         do_div(speed, rtime);
4828                 seq_printf(m, "checked_phase1: "LPU64"\n"
4829                               "checked_phase2: "LPU64"\n"
4830                               "run_time_phase1: %u seconds\n"
4831                               "run_time_phase2: %u seconds\n"
4832                               "average_speed_phase1: "LPU64" items/sec\n"
4833                               "average_speed_phase2: N/A\n"
4834                               "real-time_speed_phase1: "LPU64" items/sec\n"
4835                               "real-time_speed_phase2: N/A\n",
4836                               checked,
4837                               lo->ll_objs_checked_phase2,
4838                               rtime,
4839                               lo->ll_run_time_phase2,
4840                               speed,
4841                               new_checked);
4842
4843                 LASSERT(lfsck->li_di_oit != NULL);
4844
4845                 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
4846
4847                 /* The low layer otable-based iteration position may NOT
4848                  * exactly match the layout-based directory traversal
4849                  * cookie. Generally, it is not a serious issue. But the
4850                  * caller should NOT make assumption on that. */
4851                 pos = iops->store(env, lfsck->li_di_oit);
4852                 if (!lfsck->li_current_oit_processed)
4853                         pos--;
4854                 seq_printf(m, "current_position: "LPU64"\n", pos);
4855
4856         } else if (lo->ll_status == LS_SCANNING_PHASE2) {
4857                 cfs_duration_t duration = cfs_time_current() -
4858                                           com->lc_time_last_checkpoint;
4859                 __u64 checked = lo->ll_objs_checked_phase2 +
4860                                 com->lc_new_checked;
4861                 __u64 speed1 = lo->ll_objs_checked_phase1;
4862                 __u64 speed2 = checked;
4863                 __u64 new_checked = com->lc_new_checked *
4864                                     msecs_to_jiffies(MSEC_PER_SEC);
4865                 __u32 rtime = lo->ll_run_time_phase2 +
4866                               cfs_duration_sec(duration + HALF_SEC);
4867
4868                 if (duration != 0)
4869                         do_div(new_checked, duration);
4870                 if (lo->ll_run_time_phase1 != 0)
4871                         do_div(speed1, lo->ll_run_time_phase1);
4872                 if (rtime != 0)
4873                         do_div(speed2, rtime);
4874                 rc = seq_printf(m, "checked_phase1: "LPU64"\n"
4875                                 "checked_phase2: "LPU64"\n"
4876                                 "run_time_phase1: %u seconds\n"
4877                                 "run_time_phase2: %u seconds\n"
4878                                 "average_speed_phase1: "LPU64" items/sec\n"
4879                                 "average_speed_phase2: "LPU64" items/sec\n"
4880                                 "real-time_speed_phase1: N/A\n"
4881                                 "real-time_speed_phase2: "LPU64" items/sec\n"
4882                                 "current_position: "DFID"\n",
4883                                 lo->ll_objs_checked_phase1,
4884                                 checked,
4885                                 lo->ll_run_time_phase1,
4886                                 rtime,
4887                                 speed1,
4888                                 speed2,
4889                                 new_checked,
4890                                 PFID(&com->lc_fid_latest_scanned_phase2));
4891                 if (rc <= 0)
4892                         goto out;
4893
4894         } else {
4895                 __u64 speed1 = lo->ll_objs_checked_phase1;
4896                 __u64 speed2 = lo->ll_objs_checked_phase2;
4897
4898                 if (lo->ll_run_time_phase1 != 0)
4899                         do_div(speed1, lo->ll_run_time_phase1);
4900                 if (lo->ll_run_time_phase2 != 0)
4901                         do_div(speed2, lo->ll_run_time_phase2);
4902                 seq_printf(m, "checked_phase1: "LPU64"\n"
4903                            "checked_phase2: "LPU64"\n"
4904                            "run_time_phase1: %u seconds\n"
4905                            "run_time_phase2: %u seconds\n"
4906                            "average_speed_phase1: "LPU64" items/sec\n"
4907                            "average_speed_phase2: "LPU64" objs/sec\n"
4908                            "real-time_speed_phase1: N/A\n"
4909                            "real-time_speed_phase2: N/A\n"
4910                            "current_position: N/A\n",
4911                            lo->ll_objs_checked_phase1,
4912                            lo->ll_objs_checked_phase2,
4913                            lo->ll_run_time_phase1,
4914                            lo->ll_run_time_phase2,
4915                            speed1,
4916                            speed2);
4917         }
4918 out:
4919         up_read(&com->lc_sem);
4920
4921         return rc;
4922 }
4923
4924 static int lfsck_layout_master_double_scan(const struct lu_env *env,
4925                                            struct lfsck_component *com)
4926 {
4927         struct lfsck_layout             *lo     = com->lc_file_ram;
4928         struct lfsck_assistant_data     *lad    = com->lc_data;
4929         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4930         struct lfsck_tgt_descs          *ltds;
4931         struct lfsck_tgt_desc           *ltd;
4932         struct lfsck_tgt_desc           *next;
4933         int                              rc;
4934
4935         rc = lfsck_double_scan_generic(env, com, lo->ll_status);
4936
4937         if (thread_is_stopped(&lad->lad_thread)) {
4938                 LASSERT(list_empty(&lad->lad_req_list));
4939                 LASSERT(list_empty(&lad->lad_ost_phase1_list));
4940                 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4941
4942                 ltds = &lfsck->li_ost_descs;
4943                 spin_lock(&ltds->ltd_lock);
4944                 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
4945                                          ltd_layout_phase_list) {
4946                         list_del_init(&ltd->ltd_layout_phase_list);
4947                 }
4948                 spin_unlock(&ltds->ltd_lock);
4949
4950                 ltds = &lfsck->li_mdt_descs;
4951                 spin_lock(&ltds->ltd_lock);
4952                 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4953                                          ltd_layout_phase_list) {
4954                         list_del_init(&ltd->ltd_layout_phase_list);
4955                 }
4956                 spin_unlock(&ltds->ltd_lock);
4957         }
4958
4959         return rc;
4960 }
4961
4962 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
4963                                           struct lfsck_component *com)
4964 {
4965         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4966         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4967         struct lfsck_layout             *lo     = com->lc_file_ram;
4968         struct ptlrpc_thread            *thread = &lfsck->li_thread;
4969         int                              rc;
4970         ENTRY;
4971
4972         CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan start\n",
4973                lfsck_lfsck2name(lfsck));
4974
4975         atomic_inc(&lfsck->li_double_scan_count);
4976
4977         if (lo->ll_flags & LF_INCOMPLETE)
4978                 GOTO(done, rc = 1);
4979
4980         com->lc_new_checked = 0;
4981         com->lc_new_scanned = 0;
4982         com->lc_time_last_checkpoint = cfs_time_current();
4983         com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
4984                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
4985
4986         while (1) {
4987                 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(30),
4988                                                      NULL, NULL);
4989
4990                 rc = lfsck_layout_slave_query_master(env, com);
4991                 if (list_empty(&llsd->llsd_master_list)) {
4992                         if (unlikely(!thread_is_running(thread)))
4993                                 rc = 0;
4994                         else
4995                                 rc = 1;
4996
4997                         GOTO(done, rc);
4998                 }
4999
5000                 if (rc < 0)
5001                         GOTO(done, rc);
5002
5003                 rc = l_wait_event(thread->t_ctl_waitq,
5004                                   !thread_is_running(thread) ||
5005                                   lo->ll_flags & LF_INCOMPLETE ||
5006                                   list_empty(&llsd->llsd_master_list),
5007                                   &lwi);
5008                 if (unlikely(!thread_is_running(thread)))
5009                         GOTO(done, rc = 0);
5010
5011                 if (lo->ll_flags & LF_INCOMPLETE)
5012                         GOTO(done, rc = 1);
5013
5014                 if (rc == -ETIMEDOUT)
5015                         continue;
5016
5017                 GOTO(done, rc = (rc < 0 ? rc : 1));
5018         }
5019
5020 done:
5021         rc = lfsck_layout_double_scan_result(env, com, rc);
5022         lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE,
5023                         (rc > 0 && lo->ll_flags & LF_INCOMPLETE) ? 0 : rc);
5024         lfsck_layout_slave_quit(env, com);
5025         if (atomic_dec_and_test(&lfsck->li_double_scan_count))
5026                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5027
5028         CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan finished, "
5029                "status %d: rc = %d\n",
5030                lfsck_lfsck2name(lfsck), lo->ll_status, rc);
5031
5032         return rc;
5033 }
5034
5035 static void lfsck_layout_master_data_release(const struct lu_env *env,
5036                                              struct lfsck_component *com)
5037 {
5038         struct lfsck_assistant_data     *lad    = com->lc_data;
5039         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5040         struct lfsck_tgt_descs          *ltds;
5041         struct lfsck_tgt_desc           *ltd;
5042         struct lfsck_tgt_desc           *next;
5043
5044         LASSERT(lad != NULL);
5045         LASSERT(thread_is_init(&lad->lad_thread) ||
5046                 thread_is_stopped(&lad->lad_thread));
5047         LASSERT(list_empty(&lad->lad_req_list));
5048
5049         com->lc_data = NULL;
5050
5051         ltds = &lfsck->li_ost_descs;
5052         spin_lock(&ltds->ltd_lock);
5053         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
5054                                  ltd_layout_phase_list) {
5055                 list_del_init(&ltd->ltd_layout_phase_list);
5056         }
5057         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5058                                  ltd_layout_phase_list) {
5059                 list_del_init(&ltd->ltd_layout_phase_list);
5060         }
5061         list_for_each_entry_safe(ltd, next, &lad->lad_ost_list,
5062                                  ltd_layout_list) {
5063                 list_del_init(&ltd->ltd_layout_list);
5064         }
5065         spin_unlock(&ltds->ltd_lock);
5066
5067         ltds = &lfsck->li_mdt_descs;
5068         spin_lock(&ltds->ltd_lock);
5069         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5070                                  ltd_layout_phase_list) {
5071                 list_del_init(&ltd->ltd_layout_phase_list);
5072         }
5073         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5074                                  ltd_layout_phase_list) {
5075                 list_del_init(&ltd->ltd_layout_phase_list);
5076         }
5077         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
5078                                  ltd_layout_list) {
5079                 list_del_init(&ltd->ltd_layout_list);
5080         }
5081         spin_unlock(&ltds->ltd_lock);
5082
5083         if (likely(lad->lad_bitmap != NULL))
5084                 CFS_FREE_BITMAP(lad->lad_bitmap);
5085
5086         OBD_FREE_PTR(lad);
5087 }
5088
5089 static void lfsck_layout_slave_data_release(const struct lu_env *env,
5090                                             struct lfsck_component *com)
5091 {
5092         struct lfsck_layout_slave_data *llsd = com->lc_data;
5093
5094         lfsck_layout_slave_quit(env, com);
5095         com->lc_data = NULL;
5096         OBD_FREE_PTR(llsd);
5097 }
5098
5099 static void lfsck_layout_master_quit(const struct lu_env *env,
5100                                      struct lfsck_component *com)
5101 {
5102         struct lfsck_assistant_data     *lad    = com->lc_data;
5103         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5104         struct lfsck_tgt_descs          *ltds;
5105         struct lfsck_tgt_desc           *ltd;
5106         struct lfsck_tgt_desc           *next;
5107
5108         LASSERT(lad != NULL);
5109
5110         lfsck_quit_generic(env, com);
5111
5112         LASSERT(thread_is_init(&lad->lad_thread) ||
5113                 thread_is_stopped(&lad->lad_thread));
5114         LASSERT(list_empty(&lad->lad_req_list));
5115
5116         ltds = &lfsck->li_ost_descs;
5117         spin_lock(&ltds->ltd_lock);
5118         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
5119                                  ltd_layout_phase_list) {
5120                 list_del_init(&ltd->ltd_layout_phase_list);
5121         }
5122         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5123                                  ltd_layout_phase_list) {
5124                 list_del_init(&ltd->ltd_layout_phase_list);
5125         }
5126         spin_unlock(&ltds->ltd_lock);
5127
5128         ltds = &lfsck->li_mdt_descs;
5129         spin_lock(&ltds->ltd_lock);
5130         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5131                                  ltd_layout_phase_list) {
5132                 list_del_init(&ltd->ltd_layout_phase_list);
5133         }
5134         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5135                                  ltd_layout_phase_list) {
5136                 list_del_init(&ltd->ltd_layout_phase_list);
5137         }
5138         spin_unlock(&ltds->ltd_lock);
5139 }
5140
5141 static void lfsck_layout_slave_quit(const struct lu_env *env,
5142                                     struct lfsck_component *com)
5143 {
5144         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5145         struct lfsck_layout_seq          *lls;
5146         struct lfsck_layout_seq          *next;
5147         struct lfsck_layout_slave_target *llst;
5148
5149         LASSERT(llsd != NULL);
5150
5151         list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
5152                                  lls_list) {
5153                 list_del_init(&lls->lls_list);
5154                 lfsck_object_put(env, lls->lls_lastid_obj);
5155                 OBD_FREE_PTR(lls);
5156         }
5157
5158         spin_lock(&llsd->llsd_lock);
5159         while (!list_empty(&llsd->llsd_master_list)) {
5160                 llst = list_entry(llsd->llsd_master_list.next,
5161                                   struct lfsck_layout_slave_target, llst_list);
5162                 list_del_init(&llst->llst_list);
5163                 spin_unlock(&llsd->llsd_lock);
5164                 lfsck_layout_llst_put(llst);
5165                 spin_lock(&llsd->llsd_lock);
5166         }
5167         spin_unlock(&llsd->llsd_lock);
5168
5169         lfsck_rbtree_cleanup(env, com);
5170 }
5171
5172 static int lfsck_layout_master_in_notify(const struct lu_env *env,
5173                                          struct lfsck_component *com,
5174                                          struct lfsck_request *lr,
5175                                          struct thandle *th)
5176 {
5177         struct lfsck_instance           *lfsck = com->lc_lfsck;
5178         struct lfsck_layout             *lo    = com->lc_file_ram;
5179         struct lfsck_assistant_data     *lad   = com->lc_data;
5180         struct lfsck_tgt_descs          *ltds;
5181         struct lfsck_tgt_desc           *ltd;
5182         bool                             fail  = false;
5183         ENTRY;
5184
5185         if (lr->lr_event == LE_PAIRS_VERIFY) {
5186                 int rc;
5187
5188                 rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
5189                                                      &lr->lr_fid2);
5190
5191                 RETURN(rc);
5192         }
5193
5194         CDEBUG(D_LFSCK, "%s: layout LFSCK master handles notify %u "
5195                "from %s %x, status %d, flags %x, flags2 %x\n",
5196                lfsck_lfsck2name(lfsck), lr->lr_event,
5197                (lr->lr_flags & LEF_FROM_OST) ? "OST" : "MDT",
5198                lr->lr_index, lr->lr_status, lr->lr_flags, lr->lr_flags2);
5199
5200         if (lr->lr_event != LE_PHASE1_DONE &&
5201             lr->lr_event != LE_PHASE2_DONE &&
5202             lr->lr_event != LE_PEER_EXIT)
5203                 RETURN(-EINVAL);
5204
5205         if (lr->lr_flags & LEF_FROM_OST)
5206                 ltds = &lfsck->li_ost_descs;
5207         else
5208                 ltds = &lfsck->li_mdt_descs;
5209         spin_lock(&ltds->ltd_lock);
5210         ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
5211         if (ltd == NULL) {
5212                 spin_unlock(&ltds->ltd_lock);
5213
5214                 RETURN(-ENXIO);
5215         }
5216
5217         list_del_init(&ltd->ltd_layout_phase_list);
5218         switch (lr->lr_event) {
5219         case LE_PHASE1_DONE:
5220                 if (lr->lr_status <= 0 || lr->lr_flags2 & LF_INCOMPLETE) {
5221                         if (lr->lr_flags2 & LF_INCOMPLETE) {
5222                                 if (lr->lr_flags & LEF_FROM_OST)
5223                                         lfsck_lad_set_bitmap(env, com,
5224                                                              ltd->ltd_index);
5225                                 else
5226                                         lo->ll_flags |= LF_INCOMPLETE;
5227                         }
5228                         ltd->ltd_layout_done = 1;
5229                         list_del_init(&ltd->ltd_layout_list);
5230                         fail = true;
5231                         break;
5232                 }
5233
5234                 if (lr->lr_flags & LEF_FROM_OST) {
5235                         if (list_empty(&ltd->ltd_layout_list))
5236                                 list_add_tail(&ltd->ltd_layout_list,
5237                                               &lad->lad_ost_list);
5238                         list_add_tail(&ltd->ltd_layout_phase_list,
5239                                       &lad->lad_ost_phase2_list);
5240                 } else {
5241                         if (list_empty(&ltd->ltd_layout_list))
5242                                 list_add_tail(&ltd->ltd_layout_list,
5243                                               &lad->lad_mdt_list);
5244                         list_add_tail(&ltd->ltd_layout_phase_list,
5245                                       &lad->lad_mdt_phase2_list);
5246                 }
5247                 break;
5248         case LE_PHASE2_DONE:
5249                 ltd->ltd_layout_done = 1;
5250                 if (!list_empty(&ltd->ltd_layout_list)) {
5251                         list_del_init(&ltd->ltd_layout_list);
5252                         if (lr->lr_flags2 & LF_INCOMPLETE) {
5253                                 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
5254                                 fail = true;
5255                         }
5256                 }
5257
5258                 break;
5259         case LE_PEER_EXIT:
5260                 fail = true;
5261                 ltd->ltd_layout_done = 1;
5262                 list_del_init(&ltd->ltd_layout_list);
5263                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) &&
5264                     !(lr->lr_flags & LEF_FROM_OST))
5265                                 lo->ll_flags |= LF_INCOMPLETE;
5266                 break;
5267         default:
5268                 break;
5269         }
5270         spin_unlock(&ltds->ltd_lock);
5271
5272         if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
5273                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5274
5275                 memset(stop, 0, sizeof(*stop));
5276                 stop->ls_status = lr->lr_status;
5277                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5278                 lfsck_stop(env, lfsck->li_bottom, stop);
5279         } else if (lfsck_phase2_next_ready(lad)) {
5280                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
5281         }
5282
5283         RETURN(0);
5284 }
5285
5286 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
5287                                         struct lfsck_component *com,
5288                                         struct lfsck_request *lr,
5289                                         struct thandle *th)
5290 {
5291         struct lfsck_instance            *lfsck = com->lc_lfsck;
5292         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5293         struct lfsck_layout_slave_target *llst;
5294         int                               rc;
5295         ENTRY;
5296
5297         switch (lr->lr_event) {
5298         case LE_FID_ACCESSED:
5299                 lfsck_rbtree_update_bitmap(env, com, &lr->lr_fid, true);
5300                 RETURN(0);
5301         case LE_CONDITIONAL_DESTROY:
5302                 rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
5303                 RETURN(rc);
5304         case LE_PAIRS_VERIFY: {
5305                 lr->lr_status = LPVS_INIT;
5306                 /* Firstly, if the MDT-object which is claimed via OST-object
5307                  * local stored PFID xattr recognizes the OST-object, then it
5308                  * must be that the client given PFID is wrong. */
5309                 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5310                                                     &lr->lr_fid3);
5311                 if (rc <= 0)
5312                         RETURN(0);
5313
5314                 lr->lr_status = LPVS_INCONSISTENT;
5315                 /* The OST-object local stored PFID xattr is stale. We need to
5316                  * check whether the MDT-object that is claimed via the client
5317                  * given PFID information recognizes the OST-object or not. If
5318                  * matches, then need to update the OST-object's PFID xattr. */
5319                 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5320                                                     &lr->lr_fid2);
5321                 /* For rc < 0 case:
5322                  * We are not sure whether the client given PFID information
5323                  * is correct or not, do nothing to avoid improper fixing.
5324                  *
5325                  * For rc > 0 case:
5326                  * The client given PFID information is also invalid, we can
5327                  * NOT fix the OST-object inconsistency.
5328                  */
5329                 if (rc != 0)
5330                         RETURN(rc);
5331
5332                 lr->lr_status = LPVS_INCONSISTENT_TOFIX;
5333                 rc = lfsck_layout_slave_repair_pfid(env, com, lr);
5334
5335                 RETURN(rc);
5336         }
5337         case LE_PHASE1_DONE: {
5338                 if (lr->lr_flags2 & LF_INCOMPLETE) {
5339                         struct lfsck_layout *lo = com->lc_file_ram;
5340
5341                         lo->ll_flags |= LF_INCOMPLETE;
5342                         llst = lfsck_layout_llst_find_and_del(llsd,
5343                                                               lr->lr_index,
5344                                                               true);
5345                         if (llst != NULL) {
5346                                 lfsck_layout_llst_put(llst);
5347                                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5348                         }
5349                 }
5350
5351                 RETURN(0);
5352         }
5353         case LE_PHASE2_DONE:
5354         case LE_PEER_EXIT:
5355                 CDEBUG(D_LFSCK, "%s: layout LFSCK slave handle notify %u "
5356                        "from MDT %x, status %d\n", lfsck_lfsck2name(lfsck),
5357                        lr->lr_event, lr->lr_index, lr->lr_status);
5358                 break;
5359         default:
5360                 RETURN(-EINVAL);
5361         }
5362
5363         llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
5364         if (llst == NULL)
5365                 RETURN(0);
5366
5367         lfsck_layout_llst_put(llst);
5368         if (list_empty(&llsd->llsd_master_list))
5369                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5370
5371         if (lr->lr_event == LE_PEER_EXIT &&
5372             (lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT ||
5373              (list_empty(&llsd->llsd_master_list) &&
5374               (lr->lr_status == LS_STOPPED ||
5375                lr->lr_status == LS_CO_STOPPED)))) {
5376                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5377
5378                 memset(stop, 0, sizeof(*stop));
5379                 stop->ls_status = lr->lr_status;
5380                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5381                 lfsck_stop(env, lfsck->li_bottom, stop);
5382         }
5383
5384         RETURN(0);
5385 }
5386
5387 static int lfsck_layout_query(const struct lu_env *env,
5388                               struct lfsck_component *com)
5389 {
5390         struct lfsck_layout *lo = com->lc_file_ram;
5391
5392         return lo->ll_status;
5393 }
5394
5395 /* with lfsck::li_lock held */
5396 static int lfsck_layout_slave_join(const struct lu_env *env,
5397                                    struct lfsck_component *com,
5398                                    struct lfsck_start_param *lsp)
5399 {
5400         struct lfsck_instance            *lfsck = com->lc_lfsck;
5401         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5402         struct lfsck_layout_slave_target *llst;
5403         struct lfsck_start               *start = lsp->lsp_start;
5404         int                               rc    = 0;
5405         ENTRY;
5406
5407         if (start == NULL || !(start->ls_flags & LPF_OST_ORPHAN))
5408                 RETURN(0);
5409
5410         if (!lsp->lsp_index_valid)
5411                 RETURN(-EINVAL);
5412
5413         /* If someone is running the LFSCK without orphan handling,
5414          * it will not maintain the object accessing rbtree. So we
5415          * cannot join it for orphan handling. */
5416         if (!llsd->llsd_rbtree_valid)
5417                 RETURN(-EBUSY);
5418
5419         spin_unlock(&lfsck->li_lock);
5420         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5421         spin_lock(&lfsck->li_lock);
5422         if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
5423                 spin_unlock(&lfsck->li_lock);
5424                 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
5425                                                       true);
5426                 if (llst != NULL)
5427                         lfsck_layout_llst_put(llst);
5428                 spin_lock(&lfsck->li_lock);
5429                 rc = -EAGAIN;
5430         }
5431
5432         RETURN(rc);
5433 }
5434
5435 static struct lfsck_operations lfsck_layout_master_ops = {
5436         .lfsck_reset            = lfsck_layout_reset,
5437         .lfsck_fail             = lfsck_layout_fail,
5438         .lfsck_checkpoint       = lfsck_layout_master_checkpoint,
5439         .lfsck_prep             = lfsck_layout_master_prep,
5440         .lfsck_exec_oit         = lfsck_layout_master_exec_oit,
5441         .lfsck_exec_dir         = lfsck_layout_exec_dir,
5442         .lfsck_post             = lfsck_layout_master_post,
5443         .lfsck_dump             = lfsck_layout_dump,
5444         .lfsck_double_scan      = lfsck_layout_master_double_scan,
5445         .lfsck_data_release     = lfsck_layout_master_data_release,
5446         .lfsck_quit             = lfsck_layout_master_quit,
5447         .lfsck_in_notify        = lfsck_layout_master_in_notify,
5448         .lfsck_query            = lfsck_layout_query,
5449 };
5450
5451 static struct lfsck_operations lfsck_layout_slave_ops = {
5452         .lfsck_reset            = lfsck_layout_reset,
5453         .lfsck_fail             = lfsck_layout_fail,
5454         .lfsck_checkpoint       = lfsck_layout_slave_checkpoint,
5455         .lfsck_prep             = lfsck_layout_slave_prep,
5456         .lfsck_exec_oit         = lfsck_layout_slave_exec_oit,
5457         .lfsck_exec_dir         = lfsck_layout_exec_dir,
5458         .lfsck_post             = lfsck_layout_slave_post,
5459         .lfsck_dump             = lfsck_layout_dump,
5460         .lfsck_double_scan      = lfsck_layout_slave_double_scan,
5461         .lfsck_data_release     = lfsck_layout_slave_data_release,
5462         .lfsck_quit             = lfsck_layout_slave_quit,
5463         .lfsck_in_notify        = lfsck_layout_slave_in_notify,
5464         .lfsck_query            = lfsck_layout_query,
5465         .lfsck_join             = lfsck_layout_slave_join,
5466 };
5467
5468 static void lfsck_layout_assistant_fill_pos(const struct lu_env *env,
5469                                             struct lfsck_component *com,
5470                                             struct lfsck_position *pos)
5471 {
5472         struct lfsck_assistant_data     *lad = com->lc_data;
5473         struct lfsck_layout_req         *llr;
5474
5475         if (list_empty(&lad->lad_req_list))
5476                 return;
5477
5478         llr = list_entry(lad->lad_req_list.next,
5479                          struct lfsck_layout_req,
5480                          llr_lar.lar_list);
5481         pos->lp_oit_cookie = llr->llr_lar.lar_parent->lso_oit_cookie - 1;
5482 }
5483
5484 struct lfsck_assistant_operations lfsck_layout_assistant_ops = {
5485         .la_handler_p1          = lfsck_layout_assistant_handler_p1,
5486         .la_handler_p2          = lfsck_layout_assistant_handler_p2,
5487         .la_fill_pos            = lfsck_layout_assistant_fill_pos,
5488         .la_double_scan_result  = lfsck_layout_double_scan_result,
5489         .la_req_fini            = lfsck_layout_assistant_req_fini,
5490         .la_sync_failures       = lfsck_layout_assistant_sync_failures,
5491 };
5492
5493 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
5494 {
5495         struct lfsck_component  *com;
5496         struct lfsck_layout     *lo;
5497         struct dt_object        *root = NULL;
5498         struct dt_object        *obj;
5499         int                      rc;
5500         ENTRY;
5501
5502         OBD_ALLOC_PTR(com);
5503         if (com == NULL)
5504                 RETURN(-ENOMEM);
5505
5506         INIT_LIST_HEAD(&com->lc_link);
5507         INIT_LIST_HEAD(&com->lc_link_dir);
5508         init_rwsem(&com->lc_sem);
5509         atomic_set(&com->lc_ref, 1);
5510         com->lc_lfsck = lfsck;
5511         com->lc_type = LFSCK_TYPE_LAYOUT;
5512         if (lfsck->li_master) {
5513                 com->lc_ops = &lfsck_layout_master_ops;
5514                 com->lc_data = lfsck_assistant_data_init(
5515                                 &lfsck_layout_assistant_ops,
5516                                 LFSCK_LAYOUT);
5517                 if (com->lc_data == NULL)
5518                         GOTO(out, rc = -ENOMEM);
5519         } else {
5520                 struct lfsck_layout_slave_data *llsd;
5521
5522                 com->lc_ops = &lfsck_layout_slave_ops;
5523                 OBD_ALLOC_PTR(llsd);
5524                 if (llsd == NULL)
5525                         GOTO(out, rc = -ENOMEM);
5526
5527                 INIT_LIST_HEAD(&llsd->llsd_seq_list);
5528                 INIT_LIST_HEAD(&llsd->llsd_master_list);
5529                 spin_lock_init(&llsd->llsd_lock);
5530                 llsd->llsd_rb_root = RB_ROOT;
5531                 rwlock_init(&llsd->llsd_rb_lock);
5532                 com->lc_data = llsd;
5533         }
5534         com->lc_file_size = sizeof(*lo);
5535         OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
5536         if (com->lc_file_ram == NULL)
5537                 GOTO(out, rc = -ENOMEM);
5538
5539         OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
5540         if (com->lc_file_disk == NULL)
5541                 GOTO(out, rc = -ENOMEM);
5542
5543         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
5544         if (IS_ERR(root))
5545                 GOTO(out, rc = PTR_ERR(root));
5546
5547         if (unlikely(!dt_try_as_dir(env, root)))
5548                 GOTO(out, rc = -ENOTDIR);
5549
5550         obj = local_file_find_or_create(env, lfsck->li_los, root,
5551                                         LFSCK_LAYOUT,
5552                                         S_IFREG | S_IRUGO | S_IWUSR);
5553         if (IS_ERR(obj))
5554                 GOTO(out, rc = PTR_ERR(obj));
5555
5556         com->lc_obj = obj;
5557         rc = lfsck_layout_load(env, com);
5558         if (rc > 0)
5559                 rc = lfsck_layout_reset(env, com, true);
5560         else if (rc == -ENOENT)
5561                 rc = lfsck_layout_init(env, com);
5562
5563         if (rc != 0)
5564                 GOTO(out, rc);
5565
5566         lo = com->lc_file_ram;
5567         switch (lo->ll_status) {
5568         case LS_INIT:
5569         case LS_COMPLETED:
5570         case LS_FAILED:
5571         case LS_STOPPED:
5572         case LS_PARTIAL:
5573                 spin_lock(&lfsck->li_lock);
5574                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
5575                 spin_unlock(&lfsck->li_lock);
5576                 break;
5577         default:
5578                 CERROR("%s: unknown lfsck_layout status %d\n",
5579                        lfsck_lfsck2name(lfsck), lo->ll_status);
5580                 /* fall through */
5581         case LS_SCANNING_PHASE1:
5582         case LS_SCANNING_PHASE2:
5583                 /* No need to store the status to disk right now.
5584                  * If the system crashed before the status stored,
5585                  * it will be loaded back when next time. */
5586                 lo->ll_status = LS_CRASHED;
5587                 if (!lfsck->li_master)
5588                         lo->ll_flags |= LF_INCOMPLETE;
5589                 /* fall through */
5590         case LS_PAUSED:
5591         case LS_CRASHED:
5592         case LS_CO_FAILED:
5593         case LS_CO_STOPPED:
5594         case LS_CO_PAUSED:
5595                 spin_lock(&lfsck->li_lock);
5596                 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
5597                 spin_unlock(&lfsck->li_lock);
5598                 break;
5599         }
5600
5601         if (lo->ll_flags & LF_CRASHED_LASTID) {
5602                 LASSERT(lfsck->li_out_notify != NULL);
5603
5604                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5605                                      LE_LASTID_REBUILDING);
5606         }
5607
5608         GOTO(out, rc = 0);
5609
5610 out:
5611         if (root != NULL && !IS_ERR(root))
5612                 lfsck_object_put(env, root);
5613
5614         if (rc != 0) {
5615                 lfsck_component_cleanup(env, com);
5616                 CERROR("%s: fail to init layout LFSCK component: rc = %d\n",
5617                        lfsck_lfsck2name(lfsck), rc);
5618         }
5619
5620         return rc;
5621 }
5622
5623 struct lfsck_orphan_it {
5624         struct lfsck_component           *loi_com;
5625         struct lfsck_rbtree_node         *loi_lrn;
5626         struct lfsck_layout_slave_target *loi_llst;
5627         struct lu_fid                     loi_key;
5628         struct lu_orphan_rec              loi_rec;
5629         __u64                             loi_hash;
5630         unsigned int                      loi_over:1;
5631 };
5632
5633 static int lfsck_fid_match_idx(const struct lu_env *env,
5634                                struct lfsck_instance *lfsck,
5635                                const struct lu_fid *fid, int idx)
5636 {
5637         struct seq_server_site  *ss;
5638         struct lu_server_fld    *sf;
5639         struct lu_seq_range     *range = &lfsck_env_info(env)->lti_range;
5640         int                      rc;
5641
5642         /* All abnormal cases will be returned to MDT0. */
5643         if (!fid_is_norm(fid)) {
5644                 if (idx == 0)
5645                         return 1;
5646
5647                 return 0;
5648         }
5649
5650         ss = lfsck_dev_site(lfsck);
5651         if (unlikely(ss == NULL))
5652                 return -ENOTCONN;
5653
5654         sf = ss->ss_server_fld;
5655         LASSERT(sf != NULL);
5656
5657         fld_range_set_any(range);
5658         rc = fld_server_lookup(env, sf, fid_seq(fid), range);
5659         if (rc != 0)
5660                 return rc;
5661
5662         if (!fld_range_is_mdt(range))
5663                 return -EINVAL;
5664
5665         if (range->lsr_index == idx)
5666                 return 1;
5667
5668         return 0;
5669 }
5670
5671 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
5672                                         struct dt_object *obj)
5673 {
5674         struct dt_device        *dev    = lfsck_obj2dev(obj);
5675         struct thandle          *handle;
5676         int                      rc;
5677         ENTRY;
5678
5679         handle = dt_trans_create(env, dev);
5680         if (IS_ERR(handle))
5681                 RETURN_EXIT;
5682
5683         rc = dt_declare_ref_del(env, obj, handle);
5684         if (rc != 0)
5685                 GOTO(stop, rc);
5686
5687         rc = dt_declare_destroy(env, obj, handle);
5688         if (rc != 0)
5689                 GOTO(stop, rc);
5690
5691         rc = dt_trans_start_local(env, dev, handle);
5692         if (rc != 0)
5693                 GOTO(stop, rc);
5694
5695         dt_write_lock(env, obj, 0);
5696         rc = dt_ref_del(env, obj, handle);
5697         if (rc == 0)
5698                 rc = dt_destroy(env, obj, handle);
5699         dt_write_unlock(env, obj);
5700
5701         GOTO(stop, rc);
5702
5703 stop:
5704         dt_trans_stop(env, dev, handle);
5705
5706         CDEBUG(D_LFSCK, "destroy orphan OST-object "DFID": rc = %d\n",
5707                PFID(lfsck_dto2fid(obj)), rc);
5708
5709         RETURN_EXIT;
5710 }
5711
5712 static int lfsck_orphan_index_lookup(const struct lu_env *env,
5713                                      struct dt_object *dt,
5714                                      struct dt_rec *rec,
5715                                      const struct dt_key *key)
5716 {
5717         return -EOPNOTSUPP;
5718 }
5719
5720 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
5721                                              struct dt_object *dt,
5722                                              const struct dt_rec *rec,
5723                                              const struct dt_key *key,
5724                                              struct thandle *handle)
5725 {
5726         return -EOPNOTSUPP;
5727 }
5728
5729 static int lfsck_orphan_index_insert(const struct lu_env *env,
5730                                      struct dt_object *dt,
5731                                      const struct dt_rec *rec,
5732                                      const struct dt_key *key,
5733                                      struct thandle *handle,
5734                                      int ignore_quota)
5735 {
5736         return -EOPNOTSUPP;
5737 }
5738
5739 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
5740                                              struct dt_object *dt,
5741                                              const struct dt_key *key,
5742                                              struct thandle *handle)
5743 {
5744         return -EOPNOTSUPP;
5745 }
5746
5747 static int lfsck_orphan_index_delete(const struct lu_env *env,
5748                                      struct dt_object *dt,
5749                                      const struct dt_key *key,
5750                                      struct thandle *handle)
5751 {
5752         return -EOPNOTSUPP;
5753 }
5754
5755 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
5756                                           struct dt_object *dt,
5757                                           __u32 attr)
5758 {
5759         struct dt_device                *dev    = lu2dt_dev(dt->do_lu.lo_dev);
5760         struct lfsck_instance           *lfsck;
5761         struct lfsck_component          *com    = NULL;
5762         struct lfsck_layout_slave_data  *llsd;
5763         struct lfsck_orphan_it          *it     = NULL;
5764         struct lfsck_layout             *lo;
5765         int                              rc     = 0;
5766         ENTRY;
5767
5768         lfsck = lfsck_instance_find(dev, true, false);
5769         if (unlikely(lfsck == NULL))
5770                 RETURN(ERR_PTR(-ENXIO));
5771
5772         com = lfsck_component_find(lfsck, LFSCK_TYPE_LAYOUT);
5773         if (unlikely(com == NULL))
5774                 GOTO(out, rc = -ENOENT);
5775
5776         lo = com->lc_file_ram;
5777         if (lo->ll_flags & LF_INCOMPLETE)
5778                 GOTO(out, rc = -ESRCH);
5779
5780         llsd = com->lc_data;
5781         if (!llsd->llsd_rbtree_valid)
5782                 GOTO(out, rc = -ESRCH);
5783
5784         OBD_ALLOC_PTR(it);
5785         if (it == NULL)
5786                 GOTO(out, rc = -ENOMEM);
5787
5788         it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
5789         if (it->loi_llst == NULL)
5790                 GOTO(out, rc = -ENXIO);
5791
5792         if (dev->dd_record_fid_accessed) {
5793                 /* The first iteration against the rbtree, scan the whole rbtree
5794                  * to remove the nodes which do NOT need to be handled. */
5795                 write_lock(&llsd->llsd_rb_lock);
5796                 if (dev->dd_record_fid_accessed) {
5797                         struct rb_node                  *node;
5798                         struct rb_node                  *next;
5799                         struct lfsck_rbtree_node        *lrn;
5800
5801                         /* No need to record the fid accessing anymore. */
5802                         dev->dd_record_fid_accessed = 0;
5803
5804                         node = rb_first(&llsd->llsd_rb_root);
5805                         while (node != NULL) {
5806                                 next = rb_next(node);
5807                                 lrn = rb_entry(node, struct lfsck_rbtree_node,
5808                                                lrn_node);
5809                                 if (atomic_read(&lrn->lrn_known_count) <=
5810                                     atomic_read(&lrn->lrn_accessed_count)) {
5811                                         rb_erase(node, &llsd->llsd_rb_root);
5812                                         lfsck_rbtree_free(lrn);
5813                                 }
5814                                 node = next;
5815                         }
5816                 }
5817                 write_unlock(&llsd->llsd_rb_lock);
5818         }
5819
5820         /* read lock the rbtree when init, and unlock when fini */
5821         read_lock(&llsd->llsd_rb_lock);
5822         it->loi_com = com;
5823         com = NULL;
5824
5825         GOTO(out, rc = 0);
5826
5827 out:
5828         if (com != NULL)
5829                 lfsck_component_put(env, com);
5830
5831         CDEBUG(D_LFSCK, "%s: init the orphan iteration: rc = %d\n",
5832                lfsck_lfsck2name(lfsck), rc);
5833
5834         lfsck_instance_put(env, lfsck);
5835         if (rc != 0) {
5836                 if (it != NULL)
5837                         OBD_FREE_PTR(it);
5838
5839                 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
5840         }
5841
5842         return (struct dt_it *)it;
5843 }
5844
5845 static void lfsck_orphan_it_fini(const struct lu_env *env,
5846                                  struct dt_it *di)
5847 {
5848         struct lfsck_orphan_it           *it    = (struct lfsck_orphan_it *)di;
5849         struct lfsck_component           *com   = it->loi_com;
5850         struct lfsck_layout_slave_data   *llsd;
5851         struct lfsck_layout_slave_target *llst;
5852
5853         if (com != NULL) {
5854                 CDEBUG(D_LFSCK, "%s: fini the orphan iteration\n",
5855                        lfsck_lfsck2name(com->lc_lfsck));
5856
5857                 llsd = com->lc_data;
5858                 read_unlock(&llsd->llsd_rb_lock);
5859                 llst = it->loi_llst;
5860                 LASSERT(llst != NULL);
5861
5862                 /* Save the key and hash for iterate next. */
5863                 llst->llst_fid = it->loi_key;
5864                 llst->llst_hash = it->loi_hash;
5865                 lfsck_layout_llst_put(llst);
5866                 lfsck_component_put(env, com);
5867         }
5868         OBD_FREE_PTR(it);
5869 }
5870
5871 /**
5872  * \retval       +1: the iteration finished
5873  * \retval        0: on success, not finished
5874  * \retval      -ve: on error
5875  */
5876 static int lfsck_orphan_it_next(const struct lu_env *env,
5877                                 struct dt_it *di)
5878 {
5879         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5880         struct filter_fid_old           *pfid   = &info->lti_old_pfid;
5881         struct lu_attr                  *la     = &info->lti_la;
5882         struct lfsck_orphan_it          *it     = (struct lfsck_orphan_it *)di;
5883         struct lu_fid                   *key    = &it->loi_key;
5884         struct lu_orphan_rec            *rec    = &it->loi_rec;
5885         struct lfsck_component          *com    = it->loi_com;
5886         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5887         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
5888         struct dt_object                *obj;
5889         struct lfsck_rbtree_node        *lrn;
5890         int                              pos;
5891         int                              rc;
5892         __u32                            save;
5893         __u32                            idx    = it->loi_llst->llst_index;
5894         bool                             exact  = false;
5895         ENTRY;
5896
5897         if (it->loi_over)
5898                 RETURN(1);
5899
5900 again0:
5901         lrn = it->loi_lrn;
5902         if (lrn == NULL) {
5903                 lrn = lfsck_rbtree_search(llsd, key, &exact);
5904                 if (lrn == NULL) {
5905                         it->loi_over = 1;
5906                         RETURN(1);
5907                 }
5908
5909                 it->loi_lrn = lrn;
5910                 if (!exact) {
5911                         key->f_seq = lrn->lrn_seq;
5912                         key->f_oid = lrn->lrn_first_oid;
5913                         key->f_ver = 0;
5914                 }
5915         } else {
5916                 key->f_oid++;
5917                 if (unlikely(key->f_oid == 0)) {
5918                         key->f_seq++;
5919                         it->loi_lrn = NULL;
5920                         goto again0;
5921                 }
5922
5923                 if (key->f_oid >=
5924                     lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
5925                         it->loi_lrn = NULL;
5926                         goto again0;
5927                 }
5928         }
5929
5930         if (unlikely(atomic_read(&lrn->lrn_known_count) <=
5931                      atomic_read(&lrn->lrn_accessed_count))) {
5932                 struct rb_node *next = rb_next(&lrn->lrn_node);
5933
5934                 while (next != NULL) {
5935                         lrn = rb_entry(next, struct lfsck_rbtree_node,
5936                                        lrn_node);
5937                         if (atomic_read(&lrn->lrn_known_count) >
5938                             atomic_read(&lrn->lrn_accessed_count))
5939                                 break;
5940                         next = rb_next(next);
5941                 }
5942
5943                 if (next == NULL) {
5944                         it->loi_over = 1;
5945                         RETURN(1);
5946                 }
5947
5948                 it->loi_lrn = lrn;
5949                 key->f_seq = lrn->lrn_seq;
5950                 key->f_oid = lrn->lrn_first_oid;
5951                 key->f_ver = 0;
5952         }
5953
5954         pos = key->f_oid - lrn->lrn_first_oid;
5955
5956 again1:
5957         pos = find_next_bit(lrn->lrn_known_bitmap,
5958                             LFSCK_RBTREE_BITMAP_WIDTH, pos);
5959         if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
5960                 key->f_oid = lrn->lrn_first_oid + pos;
5961                 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
5962                         key->f_seq++;
5963                         key->f_oid = 0;
5964                 }
5965                 it->loi_lrn = NULL;
5966                 goto again0;
5967         }
5968
5969         if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
5970                 pos++;
5971                 goto again1;
5972         }
5973
5974         key->f_oid = lrn->lrn_first_oid + pos;
5975         obj = lfsck_object_find_bottom(env, lfsck, key);
5976         if (IS_ERR(obj)) {
5977                 rc = PTR_ERR(obj);
5978                 if (rc == -ENOENT) {
5979                         pos++;
5980                         goto again1;
5981                 }
5982                 RETURN(rc);
5983         }
5984
5985         dt_read_lock(env, obj, 0);
5986         if (dt_object_exists(obj) == 0 ||
5987             lfsck_is_dead_obj(obj)) {
5988                 dt_read_unlock(env, obj);
5989                 lfsck_object_put(env, obj);
5990                 pos++;
5991                 goto again1;
5992         }
5993
5994         rc = dt_attr_get(env, obj, la);
5995         if (rc != 0)
5996                 GOTO(out, rc);
5997
5998         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, pfid, sizeof(*pfid)),
5999                           XATTR_NAME_FID);
6000         if (rc == -ENODATA) {
6001                 /* For the pre-created OST-object, update the bitmap to avoid
6002                  * others LFSCK (second phase) iteration to touch it again. */
6003                 if (la->la_ctime == 0) {
6004                         if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
6005                                 atomic_inc(&lrn->lrn_accessed_count);
6006
6007                         /* For the race between repairing dangling referenced
6008                          * MDT-object and unlink the file, it may left orphan
6009                          * OST-object there. Destroy it now! */
6010                         if (unlikely(!(la->la_mode & S_ISUID))) {
6011                                 dt_read_unlock(env, obj);
6012                                 lfsck_layout_destroy_orphan(env, obj);
6013                                 lfsck_object_put(env, obj);
6014                                 pos++;
6015                                 goto again1;
6016                         }
6017                 } else if (idx == 0) {
6018                         /* If the orphan OST-object has no parent information,
6019                          * regard it as referenced by the MDT-object on MDT0. */
6020                         fid_zero(&rec->lor_fid);
6021                         rec->lor_uid = la->la_uid;
6022                         rec->lor_gid = la->la_gid;
6023                         GOTO(out, rc = 0);
6024                 }
6025
6026                 dt_read_unlock(env, obj);
6027                 lfsck_object_put(env, obj);
6028                 pos++;
6029                 goto again1;
6030         }
6031
6032         if (rc < 0)
6033                 GOTO(out, rc);
6034
6035         if (rc != sizeof(struct filter_fid) &&
6036             rc != sizeof(struct filter_fid_old))
6037                 GOTO(out, rc = -EINVAL);
6038
6039         fid_le_to_cpu(&rec->lor_fid, &pfid->ff_parent);
6040         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
6041          * MDT-object's FID::f_ver, instead it is the OST-object index in its
6042          * parent MDT-object's layout EA. */
6043         save = rec->lor_fid.f_stripe_idx;
6044         rec->lor_fid.f_ver = 0;
6045         rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_fid, idx);
6046         /* If the orphan OST-object does not claim the MDT, then next.
6047          *
6048          * If we do not know whether it matches or not, then return it
6049          * to the MDT for further check. */
6050         if (rc == 0) {
6051                 dt_read_unlock(env, obj);
6052                 lfsck_object_put(env, obj);
6053                 pos++;
6054                 goto again1;
6055         }
6056
6057         rec->lor_fid.f_stripe_idx = save;
6058         rec->lor_uid = la->la_uid;
6059         rec->lor_gid = la->la_gid;
6060
6061         CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u\n",
6062                lfsck_lfsck2name(com->lc_lfsck), PFID(key), PFID(&rec->lor_fid),
6063                rec->lor_uid, rec->lor_gid);
6064
6065         GOTO(out, rc = 0);
6066
6067 out:
6068         dt_read_unlock(env, obj);
6069         lfsck_object_put(env, obj);
6070         if (rc == 0)
6071                 it->loi_hash++;
6072
6073         return rc;
6074 }
6075
6076 /**
6077  * \retval       +1: locate to the exactly position
6078  * \retval        0: cannot locate to the exactly position,
6079  *                   call next() to move to a valid position.
6080  * \retval      -ve: on error
6081  */
6082 static int lfsck_orphan_it_get(const struct lu_env *env,
6083                                struct dt_it *di,
6084                                const struct dt_key *key)
6085 {
6086         struct lfsck_orphan_it  *it   = (struct lfsck_orphan_it *)di;
6087         int                      rc;
6088
6089         it->loi_key = *(struct lu_fid *)key;
6090         rc = lfsck_orphan_it_next(env, di);
6091         if (rc == 1)
6092                 return 0;
6093
6094         if (rc == 0)
6095                 return 1;
6096
6097         return rc;
6098 }
6099
6100 static void lfsck_orphan_it_put(const struct lu_env *env,
6101                                 struct dt_it *di)
6102 {
6103 }
6104
6105 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
6106                                           const struct dt_it *di)
6107 {
6108         struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6109
6110         return (struct dt_key *)&it->loi_key;
6111 }
6112
6113 static int lfsck_orphan_it_key_size(const struct lu_env *env,
6114                                     const struct dt_it *di)
6115 {
6116         return sizeof(struct lu_fid);
6117 }
6118
6119 static int lfsck_orphan_it_rec(const struct lu_env *env,
6120                                const struct dt_it *di,
6121                                struct dt_rec *rec,
6122                                __u32 attr)
6123 {
6124         struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6125
6126         *(struct lu_orphan_rec *)rec = it->loi_rec;
6127
6128         return 0;
6129 }
6130
6131 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
6132                                    const struct dt_it *di)
6133 {
6134         struct lfsck_orphan_it  *it   = (struct lfsck_orphan_it *)di;
6135
6136         return it->loi_hash;
6137 }
6138
6139 /**
6140  * \retval       +1: locate to the exactly position
6141  * \retval        0: cannot locate to the exactly position,
6142  *                   call next() to move to a valid position.
6143  * \retval      -ve: on error
6144  */
6145 static int lfsck_orphan_it_load(const struct lu_env *env,
6146                                 const struct dt_it *di,
6147                                 __u64 hash)
6148 {
6149         struct lfsck_orphan_it           *it   = (struct lfsck_orphan_it *)di;
6150         struct lfsck_layout_slave_target *llst = it->loi_llst;
6151         int                               rc;
6152
6153         LASSERT(llst != NULL);
6154
6155         if (hash != llst->llst_hash) {
6156                 CDEBUG(D_LFSCK, "%s: the given hash "LPU64" for orphan "
6157                        "iteration does not match the one when fini "
6158                        LPU64", to be reset.\n",
6159                        lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
6160                        llst->llst_hash);
6161                 fid_zero(&llst->llst_fid);
6162                 llst->llst_hash = 0;
6163         }
6164
6165         it->loi_key = llst->llst_fid;
6166         it->loi_hash = llst->llst_hash;
6167         rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
6168         if (rc == 1)
6169                 return 0;
6170
6171         if (rc == 0)
6172                 return 1;
6173
6174         return rc;
6175 }
6176
6177 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
6178                                    const struct dt_it *di,
6179                                    void *key_rec)
6180 {
6181         return 0;
6182 }
6183
6184 const struct dt_index_operations lfsck_orphan_index_ops = {
6185         .dio_lookup             = lfsck_orphan_index_lookup,
6186         .dio_declare_insert     = lfsck_orphan_index_declare_insert,
6187         .dio_insert             = lfsck_orphan_index_insert,
6188         .dio_declare_delete     = lfsck_orphan_index_declare_delete,
6189         .dio_delete             = lfsck_orphan_index_delete,
6190         .dio_it = {
6191                 .init           = lfsck_orphan_it_init,
6192                 .fini           = lfsck_orphan_it_fini,
6193                 .get            = lfsck_orphan_it_get,
6194                 .put            = lfsck_orphan_it_put,
6195                 .next           = lfsck_orphan_it_next,
6196                 .key            = lfsck_orphan_it_key,
6197                 .key_size       = lfsck_orphan_it_key_size,
6198                 .rec            = lfsck_orphan_it_rec,
6199                 .store          = lfsck_orphan_it_store,
6200                 .load           = lfsck_orphan_it_load,
6201                 .key_rec        = lfsck_orphan_it_key_rec,
6202         }
6203 };