lustre/lfsck/lfsck_layout.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2014, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_layout.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 #ifndef EXPORT_SYMTAB
  32 # define EXPORT_SYMTAB
  33 #endif
  34 #define DEBUG_SUBSYSTEM S_LFSCK
  35
  36 #include <linux/bitops.h>
  37 #include <linux/rbtree.h>
  38
  39 #include <lustre/lustre_idl.h>
  40 #include <lu_object.h>
  41 #include <dt_object.h>
  42 #include <lustre_fid.h>
  43 #include <lustre_lib.h>
  44 #include <lustre_net.h>
  45 #include <lustre/lustre_user.h>
  46 #include <md_object.h>
  47 #include <obd_class.h>
  48
  49 #include "lfsck_internal.h"
  50
  51 #define LFSCK_LAYOUT_MAGIC_V1           0xB173AE14
  52 #define LFSCK_LAYOUT_MAGIC_V2           0xB1734D76
  53
  54 #define LFSCK_LAYOUT_MAGIC              LFSCK_LAYOUT_MAGIC_V2
  55
  56 struct lfsck_layout_seq {
  57         struct list_head         lls_list;
  58         __u64                    lls_seq;
  59         __u64                    lls_lastid;
  60         __u64                    lls_lastid_known;
  61         struct dt_object        *lls_lastid_obj;
  62         unsigned int             lls_dirty:1;
  63 };
  64
  65 struct lfsck_layout_slave_target {
  66         /* link into lfsck_layout_slave_data::llsd_master_list. */
  67         struct list_head        llst_list;
  68         /* The position for next record in the rbtree for iteration. */
  69         struct lu_fid           llst_fid;
  70         /* Dummy hash for iteration against the rbtree. */
  71         __u64                   llst_hash;
  72         __u64                   llst_gen;
  73         atomic_t                llst_ref;
  74         __u32                   llst_index;
  75 };
  76
  77 struct lfsck_layout_slave_data {
  78         /* list for lfsck_layout_seq */
  79         struct list_head         llsd_seq_list;
  80
  81         /* list for the masters involve layout verification. */
  82         struct list_head         llsd_master_list;
  83         spinlock_t               llsd_lock;
  84         __u64                    llsd_touch_gen;
  85         struct dt_object        *llsd_rb_obj;
  86         struct rb_root           llsd_rb_root;
  87         rwlock_t                 llsd_rb_lock;
  88         unsigned int             llsd_rbtree_valid:1;
  89 };
  90
  91 struct lfsck_layout_slave_async_args {
  92         struct obd_export                *llsaa_exp;
  93         struct lfsck_component           *llsaa_com;
  94         struct lfsck_layout_slave_target *llsaa_llst;
  95 };
  96
  97 static inline void
  98 lfsck_layout_llst_put(struct lfsck_layout_slave_target *llst)
  99 {
 100         if (atomic_dec_and_test(&llst->llst_ref)) {
 101                 LASSERT(list_empty(&llst->llst_list));
 102
 103                 OBD_FREE_PTR(llst);
 104         }
 105 }
 106
 107 static inline int
 108 lfsck_layout_llst_add(struct lfsck_layout_slave_data *llsd, __u32 index)
 109 {
 110         struct lfsck_layout_slave_target *llst;
 111         struct lfsck_layout_slave_target *tmp;
 112         int                               rc   = 0;
 113
 114         OBD_ALLOC_PTR(llst);
 115         if (llst == NULL)
 116                 return -ENOMEM;
 117
 118         INIT_LIST_HEAD(&llst->llst_list);
 119         llst->llst_gen = 0;
 120         llst->llst_index = index;
 121         atomic_set(&llst->llst_ref, 1);
 122
 123         spin_lock(&llsd->llsd_lock);
 124         list_for_each_entry(tmp, &llsd->llsd_master_list, llst_list) {
 125                 if (tmp->llst_index == index) {
 126                         rc = -EALREADY;
 127                         break;
 128                 }
 129         }
 130         if (rc == 0)
 131                 list_add_tail(&llst->llst_list, &llsd->llsd_master_list);
 132         spin_unlock(&llsd->llsd_lock);
 133
 134         if (rc != 0)
 135                 OBD_FREE_PTR(llst);
 136
 137         return rc;
 138 }
 139
 140 static inline void
 141 lfsck_layout_llst_del(struct lfsck_layout_slave_data *llsd,
 142                       struct lfsck_layout_slave_target *llst)
 143 {
 144         bool del = false;
 145
 146         spin_lock(&llsd->llsd_lock);
 147         if (!list_empty(&llst->llst_list)) {
 148                 list_del_init(&llst->llst_list);
 149                 del = true;
 150         }
 151         spin_unlock(&llsd->llsd_lock);
 152
 153         if (del)
 154                 lfsck_layout_llst_put(llst);
 155 }
 156
 157 static inline struct lfsck_layout_slave_target *
 158 lfsck_layout_llst_find_and_del(struct lfsck_layout_slave_data *llsd,
 159                                __u32 index, bool unlink)
 160 {
 161         struct lfsck_layout_slave_target *llst;
 162
 163         spin_lock(&llsd->llsd_lock);
 164         list_for_each_entry(llst, &llsd->llsd_master_list, llst_list) {
 165                 if (llst->llst_index == index) {
 166                         if (unlink)
 167                                 list_del_init(&llst->llst_list);
 168                         else
 169                                 atomic_inc(&llst->llst_ref);
 170                         spin_unlock(&llsd->llsd_lock);
 171
 172                         return llst;
 173                 }
 174         }
 175         spin_unlock(&llsd->llsd_lock);
 176
 177         return NULL;
 178 }
 179
 180 static struct lfsck_layout_req *
 181 lfsck_layout_assistant_req_init(struct lfsck_assistant_object *lso,
 182                                 struct dt_object *child, __u32 ost_idx,
 183                                 __u32 lov_idx)
 184 {
 185         struct lfsck_layout_req *llr;
 186
 187         OBD_ALLOC_PTR(llr);
 188         if (llr == NULL)
 189                 return ERR_PTR(-ENOMEM);
 190
 191         INIT_LIST_HEAD(&llr->llr_lar.lar_list);
 192         llr->llr_lar.lar_parent = lfsck_assistant_object_get(lso);
 193         llr->llr_child = child;
 194         llr->llr_ost_idx = ost_idx;
 195         llr->llr_lov_idx = lov_idx;
 196
 197         return llr;
 198 }
 199
 200 static void lfsck_layout_assistant_req_fini(const struct lu_env *env,
 201                                             struct lfsck_assistant_req *lar)
 202 {
 203         struct lfsck_layout_req *llr =
 204                         container_of0(lar, struct lfsck_layout_req, llr_lar);
 205
 206         lfsck_object_put(env, llr->llr_child);
 207         lfsck_assistant_object_put(env, lar->lar_parent);
 208         OBD_FREE_PTR(llr);
 209 }
 210
 211 static int
 212 lfsck_layout_assistant_sync_failures_interpret(const struct lu_env *env,
 213                                                struct ptlrpc_request *req,
 214                                                void *args, int rc)
 215 {
 216         if (rc == 0) {
 217                 struct lfsck_async_interpret_args *laia = args;
 218                 struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
 219
 220                 ltd->ltd_synced_failures = 1;
 221                 atomic_dec(laia->laia_count);
 222         }
 223
 224         return 0;
 225 }
 226
 227 /**
 228  * Notify remote LFSCK instances about former failures.
 229  *
 230  * The local LFSCK instance has recorded which OSTs have ever failed to respond
 231  * some LFSCK verification requests (maybe because of network issues or the OST
 232  * itself trouble). During the respond gap, the OST may missed some OST-objects
 233  * verification, then the OST cannot know whether related OST-objects have been
 234  * referenced by related MDT-objects or not, then in the second-stage scanning,
 235  * these OST-objects will be regarded as orphan, if the OST-object contains bad
 236  * parent FID for back reference, then it will misguide the LFSCK to make wrong
 237  * fixing for the fake orphan.
 238  *
 239  * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
 240  * it will scan the bitmap for the ever failed OSTs, and notify them that they
 241  * have ever missed some OST-object verification and should skip the handling
 242  * for orphan OST-objects on all MDTs that are in the layout LFSCK.
 243  *
 244  * \param[in] env       pointer to the thread context
 245  * \param[in] com       pointer to the lfsck component
 246  * \param[in] lr        pointer to the lfsck request
 247  */
 248 static void lfsck_layout_assistant_sync_failures(const struct lu_env *env,
 249                                                  struct lfsck_component *com,
 250                                                  struct lfsck_request *lr)
 251 {
 252         struct lfsck_async_interpret_args *laia  =
 253                                 &lfsck_env_info(env)->lti_laia2;
 254         struct lfsck_assistant_data       *lad   = com->lc_data;
 255         struct lfsck_layout               *lo    = com->lc_file_ram;
 256         struct lfsck_instance             *lfsck = com->lc_lfsck;
 257         struct lfsck_tgt_descs            *ltds  = &lfsck->li_ost_descs;
 258         struct lfsck_tgt_desc             *ltd;
 259         struct ptlrpc_request_set         *set;
 260         atomic_t                           count;
 261         __u32                              idx;
 262         int                                rc    = 0;
 263         ENTRY;
 264
 265         if (!lad->lad_incomplete || lo->ll_flags & LF_INCOMPLETE)
 266                 RETURN_EXIT;
 267
 268         /* If the MDT has ever failed to verfiy some OST-objects,
 269          * then sync failures with them firstly. */
 270         lr->lr_flags2 = lo->ll_flags | LF_INCOMPLETE;
 271
 272         atomic_set(&count, 0);
 273         memset(laia, 0, sizeof(*laia));
 274         laia->laia_count = &count;
 275         set = ptlrpc_prep_set();
 276         if (set == NULL)
 277                 GOTO(out, rc = -ENOMEM);
 278
 279         down_read(&ltds->ltd_rw_sem);
 280         cfs_foreach_bit(lad->lad_bitmap, idx) {
 281                 ltd = lfsck_ltd2tgt(ltds, idx);
 282                 LASSERT(ltd != NULL);
 283
 284                 laia->laia_ltd = ltd;
 285                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
 286                                 lfsck_layout_assistant_sync_failures_interpret,
 287                                 laia, LFSCK_NOTIFY);
 288                 if (rc != 0) {
 289                         CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
 290                                "notify target %x for %s phase1 done: "
 291                                "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
 292                                ltd->ltd_index, lad->lad_name, rc);
 293
 294                         break;
 295                 }
 296
 297                 atomic_inc(&count);
 298         }
 299         up_read(&ltds->ltd_rw_sem);
 300
 301         if (rc == 0 && atomic_read(&count) > 0)
 302                 rc = ptlrpc_set_wait(set);
 303
 304         ptlrpc_set_destroy(set);
 305
 306         if (rc == 0 && atomic_read(&count) > 0)
 307                 rc = -EINVAL;
 308
 309         GOTO(out, rc);
 310
 311 out:
 312         if (rc != 0)
 313                 /* If failed to sync failures with the OSTs, then have to
 314                  * mark the whole LFSCK as LF_INCOMPLETE to skip the whole
 315                  * subsequent orphan OST-object handling. */
 316                 lo->ll_flags |= LF_INCOMPLETE;
 317
 318         lr->lr_flags2 = lo->ll_flags;
 319 }
 320
 321 static int lfsck_layout_get_lovea(const struct lu_env *env,
 322                                   struct dt_object *obj, struct lu_buf *buf)
 323 {
 324         int rc;
 325
 326 again:
 327         rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LOV);
 328         if (rc == -ERANGE) {
 329                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV);
 330                 if (rc <= 0)
 331                         return rc;
 332
 333                 lu_buf_realloc(buf, rc);
 334                 if (buf->lb_buf == NULL)
 335                         return -ENOMEM;
 336
 337                 goto again;
 338         }
 339
 340         if (rc == -ENODATA)
 341                 rc = 0;
 342
 343         if (rc <= 0)
 344                 return rc;
 345
 346         if (unlikely(buf->lb_buf == NULL)) {
 347                 lu_buf_alloc(buf, rc);
 348                 if (buf->lb_buf == NULL)
 349                         return -ENOMEM;
 350
 351                 goto again;
 352         }
 353
 354         return rc;
 355 }
 356
 357 static int lfsck_layout_verify_header(struct lov_mds_md_v1 *lmm)
 358 {
 359         __u32 magic;
 360         __u32 pattern;
 361
 362         magic = le32_to_cpu(lmm->lmm_magic);
 363         /* If magic crashed, keep it there. Sometime later, during OST-object
 364          * orphan handling, if some OST-object(s) back-point to it, it can be
 365          * verified and repaired. */
 366         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) {
 367                 struct ost_id   oi;
 368                 int             rc;
 369
 370                 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
 371                 if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
 372                         rc = -EOPNOTSUPP;
 373                 else
 374                         rc = -EINVAL;
 375
 376                 CDEBUG(D_LFSCK, "%s LOV EA magic %u on "DOSTID"\n",
 377                        rc == -EINVAL ? "Unknown" : "Unsupported",
 378                        magic, POSTID(&oi));
 379
 380                 return rc;
 381         }
 382
 383         pattern = le32_to_cpu(lmm->lmm_pattern);
 384         /* XXX: currently, we only support LOV_PATTERN_RAID0. */
 385         if (lov_pattern(pattern) != LOV_PATTERN_RAID0) {
 386                 struct ost_id oi;
 387
 388                 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
 389                 CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u on "DOSTID"\n",
 390                        pattern, POSTID(&oi));
 391
 392                 return -EOPNOTSUPP;
 393         }
 394
 395         return 0;
 396 }
 397
 398 #define LFSCK_RBTREE_BITMAP_SIZE        PAGE_CACHE_SIZE
 399 #define LFSCK_RBTREE_BITMAP_WIDTH       (LFSCK_RBTREE_BITMAP_SIZE << 3)
 400 #define LFSCK_RBTREE_BITMAP_MASK        (LFSCK_RBTREE_BITMAP_WIDTH - 1)
 401
 402 struct lfsck_rbtree_node {
 403         struct rb_node   lrn_node;
 404         __u64            lrn_seq;
 405         __u32            lrn_first_oid;
 406         atomic_t         lrn_known_count;
 407         atomic_t         lrn_accessed_count;
 408         void            *lrn_known_bitmap;
 409         void            *lrn_accessed_bitmap;
 410 };
 411
 412 static inline int lfsck_rbtree_cmp(struct lfsck_rbtree_node *lrn,
 413                                    __u64 seq, __u32 oid)
 414 {
 415         if (seq < lrn->lrn_seq)
 416                 return -1;
 417
 418         if (seq > lrn->lrn_seq)
 419                 return 1;
 420
 421         if (oid < lrn->lrn_first_oid)
 422                 return -1;
 423
 424         if (oid - lrn->lrn_first_oid >= LFSCK_RBTREE_BITMAP_WIDTH)
 425                 return 1;
 426
 427         return 0;
 428 }
 429
 430 /* The caller should hold llsd->llsd_rb_lock. */
 431 static struct lfsck_rbtree_node *
 432 lfsck_rbtree_search(struct lfsck_layout_slave_data *llsd,
 433                     const struct lu_fid *fid, bool *exact)
 434 {
 435         struct rb_node           *node  = llsd->llsd_rb_root.rb_node;
 436         struct rb_node           *prev  = NULL;
 437         struct lfsck_rbtree_node *lrn   = NULL;
 438         int                       rc    = 0;
 439
 440         if (exact != NULL)
 441                 *exact = true;
 442
 443         while (node != NULL) {
 444                 prev = node;
 445                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 446                 rc = lfsck_rbtree_cmp(lrn, fid_seq(fid), fid_oid(fid));
 447                 if (rc < 0)
 448                         node = node->rb_left;
 449                 else if (rc > 0)
 450                         node = node->rb_right;
 451                 else
 452                         return lrn;
 453         }
 454
 455         if (exact == NULL)
 456                 return NULL;
 457
 458         /* If there is no exactly matched one, then to the next valid one. */
 459         *exact = false;
 460
 461         /* The rbtree is empty. */
 462         if (rc == 0)
 463                 return NULL;
 464
 465         if (rc < 0)
 466                 return lrn;
 467
 468         node = rb_next(prev);
 469
 470         /* The end of the rbtree. */
 471         if (node == NULL)
 472                 return NULL;
 473
 474         lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 475
 476         return lrn;
 477 }
 478
 479 static struct lfsck_rbtree_node *lfsck_rbtree_new(const struct lu_env *env,
 480                                                   const struct lu_fid *fid)
 481 {
 482         struct lfsck_rbtree_node *lrn;
 483
 484         OBD_ALLOC_PTR(lrn);
 485         if (lrn == NULL)
 486                 return ERR_PTR(-ENOMEM);
 487
 488         OBD_ALLOC(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 489         if (lrn->lrn_known_bitmap == NULL) {
 490                 OBD_FREE_PTR(lrn);
 491
 492                 return ERR_PTR(-ENOMEM);
 493         }
 494
 495         OBD_ALLOC(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 496         if (lrn->lrn_accessed_bitmap == NULL) {
 497                 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 498                 OBD_FREE_PTR(lrn);
 499
 500                 return ERR_PTR(-ENOMEM);
 501         }
 502
 503         RB_CLEAR_NODE(&lrn->lrn_node);
 504         lrn->lrn_seq = fid_seq(fid);
 505         lrn->lrn_first_oid = fid_oid(fid) & ~LFSCK_RBTREE_BITMAP_MASK;
 506         atomic_set(&lrn->lrn_known_count, 0);
 507         atomic_set(&lrn->lrn_accessed_count, 0);
 508
 509         return lrn;
 510 }
 511
 512 static void lfsck_rbtree_free(struct lfsck_rbtree_node *lrn)
 513 {
 514         OBD_FREE(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 515         OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 516         OBD_FREE_PTR(lrn);
 517 }
 518
 519 /* The caller should hold lock. */
 520 static struct lfsck_rbtree_node *
 521 lfsck_rbtree_insert(struct lfsck_layout_slave_data *llsd,
 522                     struct lfsck_rbtree_node *lrn)
 523 {
 524         struct rb_node           **pos    = &llsd->llsd_rb_root.rb_node;
 525         struct rb_node            *parent = NULL;
 526         struct lfsck_rbtree_node  *tmp;
 527         int                        rc;
 528
 529         while (*pos != NULL) {
 530                 parent = *pos;
 531                 tmp = rb_entry(parent, struct lfsck_rbtree_node, lrn_node);
 532                 rc = lfsck_rbtree_cmp(tmp, lrn->lrn_seq, lrn->lrn_first_oid);
 533                 if (rc < 0)
 534                         pos = &(*pos)->rb_left;
 535                 else if (rc > 0)
 536                         pos = &(*pos)->rb_right;
 537                 else
 538                         return tmp;
 539         }
 540
 541         rb_link_node(&lrn->lrn_node, parent, pos);
 542         rb_insert_color(&lrn->lrn_node, &llsd->llsd_rb_root);
 543
 544         return lrn;
 545 }
 546
 547 extern const struct dt_index_operations lfsck_orphan_index_ops;
 548
 549 static int lfsck_rbtree_setup(const struct lu_env *env,
 550                               struct lfsck_component *com)
 551 {
 552         struct lu_fid                   *fid    = &lfsck_env_info(env)->lti_fid;
 553         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 554         struct dt_device                *dev    = lfsck->li_bottom;
 555         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
 556         struct dt_object                *obj;
 557
 558         fid->f_seq = FID_SEQ_LAYOUT_RBTREE;
 559         fid->f_oid = lfsck_dev_idx(lfsck);
 560         fid->f_ver = 0;
 561         obj = dt_locate(env, dev, fid);
 562         if (IS_ERR(obj))
 563                 RETURN(PTR_ERR(obj));
 564
 565         /* Generate an in-RAM object to stand for the layout rbtree.
 566          * Scanning the layout rbtree will be via the iteration over
 567          * the object. In the future, the rbtree may be written onto
 568          * disk with the object.
 569          *
 570          * Mark the object to be as exist. */
 571         obj->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 572         obj->do_index_ops = &lfsck_orphan_index_ops;
 573         llsd->llsd_rb_obj = obj;
 574         llsd->llsd_rbtree_valid = 1;
 575         dev->dd_record_fid_accessed = 1;
 576
 577         CDEBUG(D_LFSCK, "%s: layout LFSCK init OST-objects accessing bitmap\n",
 578                lfsck_lfsck2name(lfsck));
 579
 580         return 0;
 581 }
 582
 583 static void lfsck_rbtree_cleanup(const struct lu_env *env,
 584                                  struct lfsck_component *com)
 585 {
 586         struct lfsck_instance           *lfsck = com->lc_lfsck;
 587         struct lfsck_layout_slave_data  *llsd  = com->lc_data;
 588         struct rb_node                  *node  = rb_first(&llsd->llsd_rb_root);
 589         struct rb_node                  *next;
 590         struct lfsck_rbtree_node        *lrn;
 591
 592         lfsck->li_bottom->dd_record_fid_accessed = 0;
 593         /* Invalid the rbtree, then no others will use it. */
 594         write_lock(&llsd->llsd_rb_lock);
 595         llsd->llsd_rbtree_valid = 0;
 596         write_unlock(&llsd->llsd_rb_lock);
 597
 598         while (node != NULL) {
 599                 next = rb_next(node);
 600                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 601                 rb_erase(node, &llsd->llsd_rb_root);
 602                 lfsck_rbtree_free(lrn);
 603                 node = next;
 604         }
 605
 606         if (llsd->llsd_rb_obj != NULL) {
 607                 lfsck_object_put(env, llsd->llsd_rb_obj);
 608                 llsd->llsd_rb_obj = NULL;
 609         }
 610
 611         CDEBUG(D_LFSCK, "%s: layout LFSCK fini OST-objects accessing bitmap\n",
 612                lfsck_lfsck2name(lfsck));
 613 }
 614
 615 static void lfsck_rbtree_update_bitmap(const struct lu_env *env,
 616                                        struct lfsck_component *com,
 617                                        const struct lu_fid *fid,
 618                                        bool accessed)
 619 {
 620         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
 621         struct lfsck_rbtree_node        *lrn;
 622         bool                             insert = false;
 623         int                              idx;
 624         int                              rc     = 0;
 625         ENTRY;
 626
 627         if (unlikely(!fid_is_sane(fid) || fid_is_last_id(fid)))
 628                 RETURN_EXIT;
 629
 630         if (!fid_is_idif(fid) && !fid_is_norm(fid))
 631                 RETURN_EXIT;
 632
 633         read_lock(&llsd->llsd_rb_lock);
 634         if (!llsd->llsd_rbtree_valid)
 635                 GOTO(unlock, rc = 0);
 636
 637         lrn = lfsck_rbtree_search(llsd, fid, NULL);
 638         if (lrn == NULL) {
 639                 struct lfsck_rbtree_node *tmp;
 640
 641                 LASSERT(!insert);
 642
 643                 read_unlock(&llsd->llsd_rb_lock);
 644                 tmp = lfsck_rbtree_new(env, fid);
 645                 if (IS_ERR(tmp))
 646                         GOTO(out, rc = PTR_ERR(tmp));
 647
 648                 insert = true;
 649                 write_lock(&llsd->llsd_rb_lock);
 650                 if (!llsd->llsd_rbtree_valid) {
 651                         lfsck_rbtree_free(tmp);
 652                         GOTO(unlock, rc = 0);
 653                 }
 654
 655                 lrn = lfsck_rbtree_insert(llsd, tmp);
 656                 if (lrn != tmp)
 657                         lfsck_rbtree_free(tmp);
 658         }
 659
 660         idx = fid_oid(fid) & LFSCK_RBTREE_BITMAP_MASK;
 661         /* Any accessed object must be a known object. */
 662         if (!test_and_set_bit(idx, lrn->lrn_known_bitmap))
 663                 atomic_inc(&lrn->lrn_known_count);
 664         if (accessed && !test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
 665                 atomic_inc(&lrn->lrn_accessed_count);
 666
 667         GOTO(unlock, rc = 0);
 668
 669 unlock:
 670         if (insert)
 671                 write_unlock(&llsd->llsd_rb_lock);
 672         else
 673                 read_unlock(&llsd->llsd_rb_lock);
 674 out:
 675         if (rc != 0 && accessed) {
 676                 struct lfsck_layout *lo = com->lc_file_ram;
 677
 678                 CDEBUG(D_LFSCK, "%s: fail to update OST-objects accessing "
 679                        "bitmap, and will cause incorrect LFSCK OST-object "
 680                        "handling, so disable it to cancel orphan handling "
 681                        "for related device. rc = %d\n",
 682                        lfsck_lfsck2name(com->lc_lfsck), rc);
 683
 684                 lo->ll_flags |= LF_INCOMPLETE;
 685                 lfsck_rbtree_cleanup(env, com);
 686         }
 687 }
 688
 689 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
 690                                    const struct lfsck_layout *src)
 691 {
 692         int i;
 693
 694         des->ll_magic = le32_to_cpu(src->ll_magic);
 695         des->ll_status = le32_to_cpu(src->ll_status);
 696         des->ll_flags = le32_to_cpu(src->ll_flags);
 697         des->ll_success_count = le32_to_cpu(src->ll_success_count);
 698         des->ll_run_time_phase1 = le32_to_cpu(src->ll_run_time_phase1);
 699         des->ll_run_time_phase2 = le32_to_cpu(src->ll_run_time_phase2);
 700         des->ll_time_last_complete = le64_to_cpu(src->ll_time_last_complete);
 701         des->ll_time_latest_start = le64_to_cpu(src->ll_time_latest_start);
 702         des->ll_time_last_checkpoint =
 703                                 le64_to_cpu(src->ll_time_last_checkpoint);
 704         des->ll_pos_latest_start = le64_to_cpu(src->ll_pos_latest_start);
 705         des->ll_pos_last_checkpoint = le64_to_cpu(src->ll_pos_last_checkpoint);
 706         des->ll_pos_first_inconsistent =
 707                         le64_to_cpu(src->ll_pos_first_inconsistent);
 708         des->ll_objs_checked_phase1 = le64_to_cpu(src->ll_objs_checked_phase1);
 709         des->ll_objs_failed_phase1 = le64_to_cpu(src->ll_objs_failed_phase1);
 710         des->ll_objs_checked_phase2 = le64_to_cpu(src->ll_objs_checked_phase2);
 711         des->ll_objs_failed_phase2 = le64_to_cpu(src->ll_objs_failed_phase2);
 712         for (i = 0; i < LLIT_MAX; i++)
 713                 des->ll_objs_repaired[i] =
 714                                 le64_to_cpu(src->ll_objs_repaired[i]);
 715         des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
 716         des->ll_bitmap_size = le32_to_cpu(src->ll_bitmap_size);
 717 }
 718
 719 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
 720                                    const struct lfsck_layout *src)
 721 {
 722         int i;
 723
 724         des->ll_magic = cpu_to_le32(src->ll_magic);
 725         des->ll_status = cpu_to_le32(src->ll_status);
 726         des->ll_flags = cpu_to_le32(src->ll_flags);
 727         des->ll_success_count = cpu_to_le32(src->ll_success_count);
 728         des->ll_run_time_phase1 = cpu_to_le32(src->ll_run_time_phase1);
 729         des->ll_run_time_phase2 = cpu_to_le32(src->ll_run_time_phase2);
 730         des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
 731         des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
 732         des->ll_time_last_checkpoint =
 733                                 cpu_to_le64(src->ll_time_last_checkpoint);
 734         des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
 735         des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
 736         des->ll_pos_first_inconsistent =
 737                         cpu_to_le64(src->ll_pos_first_inconsistent);
 738         des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
 739         des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
 740         des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
 741         des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
 742         for (i = 0; i < LLIT_MAX; i++)
 743                 des->ll_objs_repaired[i] =
 744                                 cpu_to_le64(src->ll_objs_repaired[i]);
 745         des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
 746         des->ll_bitmap_size = cpu_to_le32(src->ll_bitmap_size);
 747 }
 748
 749 /**
 750  * Load the OST bitmap from the lfsck_layout trace file.
 751  *
 752  * \param[in] env       pointer to the thread context
 753  * \param[in] com       pointer to the lfsck component
 754  *
 755  * \retval              0 for success
 756  * \retval              negative error number on failure or data corruption
 757  */
 758 static int lfsck_layout_load_bitmap(const struct lu_env *env,
 759                                     struct lfsck_component *com)
 760 {
 761         struct dt_object                *obj    = com->lc_obj;
 762         struct lfsck_assistant_data     *lad    = com->lc_data;
 763         struct lfsck_layout             *lo     = com->lc_file_ram;
 764         cfs_bitmap_t                    *bitmap = lad->lad_bitmap;
 765         loff_t                           pos    = com->lc_file_size;
 766         ssize_t                          size;
 767         __u32                            nbits;
 768         int                              rc;
 769         ENTRY;
 770
 771         if (com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size >
 772             lo->ll_bitmap_size)
 773                 nbits = com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size;
 774         else
 775                 nbits = lo->ll_bitmap_size;
 776
 777         if (unlikely(nbits < BITS_PER_LONG))
 778                 nbits = BITS_PER_LONG;
 779
 780         if (nbits > bitmap->size) {
 781                 __u32 new_bits = bitmap->size;
 782                 cfs_bitmap_t *new_bitmap;
 783
 784                 while (new_bits < nbits)
 785                         new_bits <<= 1;
 786
 787                 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
 788                 if (new_bitmap == NULL)
 789                         RETURN(-ENOMEM);
 790
 791                 lad->lad_bitmap = new_bitmap;
 792                 CFS_FREE_BITMAP(bitmap);
 793                 bitmap = new_bitmap;
 794         }
 795
 796         if (lo->ll_bitmap_size == 0) {
 797                 lad->lad_incomplete = 0;
 798                 CFS_RESET_BITMAP(bitmap);
 799
 800                 RETURN(0);
 801         }
 802
 803         size = (lo->ll_bitmap_size + 7) >> 3;
 804         rc = dt_read(env, obj, lfsck_buf_get(env, bitmap->data, size), &pos);
 805         if (rc != size)
 806                 RETURN(rc >= 0 ? -EINVAL : rc);
 807
 808         if (cfs_bitmap_check_empty(bitmap))
 809                 lad->lad_incomplete = 0;
 810         else
 811                 lad->lad_incomplete = 1;
 812
 813         RETURN(0);
 814 }
 815
 816 /**
 817  * Load the layout LFSCK trace file from disk.
 818  *
 819  * The layout LFSCK trace file records the layout LFSCK status information
 820  * and other statistics, such as how many objects have been scanned, and how
 821  * many objects have been repaired, and etc. It also contains the bitmap for
 822  * failed OSTs during the layout LFSCK. All these information will be loaded
 823  * from disk to RAM when the layout LFSCK component setup.
 824  *
 825  * \param[in] env       pointer to the thread context
 826  * \param[in] com       pointer to the lfsck component
 827  *
 828  * \retval              positive number for file data corruption, the caller
 829  *                      should reset the layout LFSCK trace file
 830  * \retval              0 for success
 831  * \retval              negative error number on failure
 832  */
 833 static int lfsck_layout_load(const struct lu_env *env,
 834                              struct lfsck_component *com)
 835 {
 836         struct lfsck_layout             *lo     = com->lc_file_ram;
 837         ssize_t                          size   = com->lc_file_size;
 838         loff_t                           pos    = 0;
 839         int                              rc;
 840
 841         rc = dt_read(env, com->lc_obj,
 842                      lfsck_buf_get(env, com->lc_file_disk, size), &pos);
 843         if (rc == 0) {
 844                 return -ENOENT;
 845         } else if (rc < 0) {
 846                 CDEBUG(D_LFSCK, "%s: failed to load lfsck_layout: rc = %d\n",
 847                        lfsck_lfsck2name(com->lc_lfsck), rc);
 848                 return rc;
 849         } else if (rc != size) {
 850                 CDEBUG(D_LFSCK, "%s: lfsck_layout size %u != %u; reset it\n",
 851                        lfsck_lfsck2name(com->lc_lfsck), rc, (unsigned int)size);
 852                 return 1;
 853         }
 854
 855         lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
 856         if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
 857                 CDEBUG(D_LFSCK, "%s: invalid lfsck_layout magic %#x != %#x, "
 858                        "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
 859                        lo->ll_magic, LFSCK_LAYOUT_MAGIC);
 860                 return 1;
 861         }
 862
 863         return 0;
 864 }
 865
 866 /**
 867  * Store the layout LFSCK trace file on disk.
 868  *
 869  * The layout LFSCK trace file records the layout LFSCK status information
 870  * and other statistics, such as how many objects have been scanned, and how
 871  * many objects have been repaired, and etc. It also contains the bitmap for
 872  * failed OSTs during the layout LFSCK. All these information will be synced
 873  * from RAM to disk periodically.
 874  *
 875  * \param[in] env       pointer to the thread context
 876  * \param[in] com       pointer to the lfsck component
 877  *
 878  * \retval              0 for success
 879  * \retval              negative error number on failure
 880  */
 881 static int lfsck_layout_store(const struct lu_env *env,
 882                               struct lfsck_component *com)
 883 {
 884         struct dt_object        *obj    = com->lc_obj;
 885         struct lfsck_instance   *lfsck  = com->lc_lfsck;
 886         struct lfsck_layout     *lo_ram = com->lc_file_ram;
 887         struct lfsck_layout     *lo     = com->lc_file_disk;
 888         struct thandle          *th;
 889         struct dt_device        *dev    = lfsck_obj2dev(obj);
 890         cfs_bitmap_t            *bitmap = NULL;
 891         loff_t                   pos;
 892         ssize_t                  size   = com->lc_file_size;
 893         __u32                    nbits  = 0;
 894         int                      rc;
 895         ENTRY;
 896
 897         if (lfsck->li_master) {
 898                 struct lfsck_assistant_data *lad = com->lc_data;
 899
 900                 bitmap = lad->lad_bitmap;
 901                 nbits = bitmap->size;
 902
 903                 LASSERT(nbits > 0);
 904                 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
 905         }
 906
 907         lo_ram->ll_bitmap_size = nbits;
 908         lfsck_layout_cpu_to_le(lo, lo_ram);
 909         th = dt_trans_create(env, dev);
 910         if (IS_ERR(th))
 911                 GOTO(log, rc = PTR_ERR(th));
 912
 913         rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
 914                                      (loff_t)0, th);
 915         if (rc != 0)
 916                 GOTO(out, rc);
 917
 918         if (bitmap != NULL) {
 919                 rc = dt_declare_record_write(env, obj,
 920                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
 921                                 (loff_t)size, th);
 922                 if (rc != 0)
 923                         GOTO(out, rc);
 924         }
 925
 926         rc = dt_trans_start_local(env, dev, th);
 927         if (rc != 0)
 928                 GOTO(out, rc);
 929
 930         pos = 0;
 931         rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos, th);
 932         if (rc != 0)
 933                 GOTO(out, rc);
 934
 935         if (bitmap != NULL) {
 936                 pos = size;
 937                 rc = dt_record_write(env, obj,
 938                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
 939                                 &pos, th);
 940         }
 941
 942         GOTO(out, rc);
 943
 944 out:
 945         dt_trans_stop(env, dev, th);
 946
 947 log:
 948         if (rc != 0)
 949                 CDEBUG(D_LFSCK, "%s: fail to store lfsck_layout: rc = %d\n",
 950                        lfsck_lfsck2name(lfsck), rc);
 951
 952         return rc;
 953 }
 954
 955 static int lfsck_layout_init(const struct lu_env *env,
 956                              struct lfsck_component *com)
 957 {
 958         struct lfsck_layout *lo = com->lc_file_ram;
 959         int rc;
 960
 961         memset(lo, 0, com->lc_file_size);
 962         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
 963         lo->ll_status = LS_INIT;
 964         down_write(&com->lc_sem);
 965         rc = lfsck_layout_store(env, com);
 966         up_write(&com->lc_sem);
 967
 968         return rc;
 969 }
 970
 971 static int fid_is_for_ostobj(const struct lu_env *env,
 972                              struct lfsck_instance *lfsck,
 973                              struct dt_object *obj, const struct lu_fid *fid)
 974 {
 975         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
 976         struct lu_seq_range     *range  = &lfsck_env_info(env)->lti_range;
 977         struct lustre_mdt_attrs *lma;
 978         int                      rc;
 979
 980         fld_range_set_any(range);
 981         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
 982         if (rc == 0) {
 983                 if (fld_range_is_ost(range))
 984                         return 1;
 985
 986                 return 0;
 987         }
 988
 989         lma = &lfsck_env_info(env)->lti_lma;
 990         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, lma, sizeof(*lma)),
 991                           XATTR_NAME_LMA);
 992         if (rc == sizeof(*lma)) {
 993                 lustre_lma_swab(lma);
 994
 995                 return lma->lma_compat & LMAC_FID_ON_OST ? 1 : 0;
 996         }
 997
 998         rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID);
 999
1000         return rc > 0;
1001 }
1002
1003 static struct lfsck_layout_seq *
1004 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
1005 {
1006         struct lfsck_layout_seq *lls;
1007
1008         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1009                 if (lls->lls_seq == seq)
1010                         return lls;
1011
1012                 if (lls->lls_seq > seq)
1013                         return NULL;
1014         }
1015
1016         return NULL;
1017 }
1018
1019 static void
1020 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
1021                         struct lfsck_layout_seq *lls)
1022 {
1023         struct lfsck_layout_seq *tmp;
1024         struct list_head        *pos = &llsd->llsd_seq_list;
1025
1026         list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
1027                 if (lls->lls_seq < tmp->lls_seq) {
1028                         pos = &tmp->lls_list;
1029                         break;
1030                 }
1031         }
1032         list_add_tail(&lls->lls_list, pos);
1033 }
1034
1035 static int
1036 lfsck_layout_lastid_create(const struct lu_env *env,
1037                            struct lfsck_instance *lfsck,
1038                            struct dt_object *obj)
1039 {
1040         struct lfsck_thread_info *info   = lfsck_env_info(env);
1041         struct lu_attr           *la     = &info->lti_la;
1042         struct dt_object_format  *dof    = &info->lti_dof;
1043         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
1044         struct dt_device         *dt     = lfsck_obj2dev(obj);
1045         struct thandle           *th;
1046         __u64                     lastid = 0;
1047         loff_t                    pos    = 0;
1048         int                       rc;
1049         ENTRY;
1050
1051         if (bk->lb_param & LPF_DRYRUN)
1052                 return 0;
1053
1054         memset(la, 0, sizeof(*la));
1055         la->la_mode = S_IFREG |  S_IRUGO | S_IWUSR;
1056         la->la_valid = LA_MODE | LA_UID | LA_GID;
1057         memset(dof, 0, sizeof(*dof));
1058         dof->dof_type = dt_mode_to_dft(S_IFREG);
1059
1060         th = dt_trans_create(env, dt);
1061         if (IS_ERR(th))
1062                 GOTO(log, rc = PTR_ERR(th));
1063
1064         rc = dt_declare_create(env, obj, la, NULL, dof, th);
1065         if (rc != 0)
1066                 GOTO(stop, rc);
1067
1068         rc = dt_declare_record_write(env, obj,
1069                                      lfsck_buf_get(env, &lastid,
1070                                                    sizeof(lastid)),
1071                                      pos, th);
1072         if (rc != 0)
1073                 GOTO(stop, rc);
1074
1075         rc = dt_trans_start_local(env, dt, th);
1076         if (rc != 0)
1077                 GOTO(stop, rc);
1078
1079         dt_write_lock(env, obj, 0);
1080         if (likely(dt_object_exists(obj) == 0)) {
1081                 rc = dt_create(env, obj, la, NULL, dof, th);
1082                 if (rc == 0)
1083                         rc = dt_record_write(env, obj,
1084                                 lfsck_buf_get(env, &lastid, sizeof(lastid)),
1085                                 &pos, th);
1086         }
1087         dt_write_unlock(env, obj);
1088
1089         GOTO(stop, rc);
1090
1091 stop:
1092         dt_trans_stop(env, dt, th);
1093
1094 log:
1095         CDEBUG(D_LFSCK, "%s: layout LFSCK will create LAST_ID for <seq> "
1096                LPX64": rc = %d\n",
1097                lfsck_lfsck2name(lfsck), fid_seq(lfsck_dto2fid(obj)), rc);
1098
1099         return rc;
1100 }
1101
1102 static int
1103 lfsck_layout_lastid_reload(const struct lu_env *env,
1104                            struct lfsck_component *com,
1105                            struct lfsck_layout_seq *lls)
1106 {
1107         __u64   lastid;
1108         loff_t  pos     = 0;
1109         int     rc;
1110
1111         dt_read_lock(env, lls->lls_lastid_obj, 0);
1112         rc = dt_record_read(env, lls->lls_lastid_obj,
1113                             lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
1114         dt_read_unlock(env, lls->lls_lastid_obj);
1115         if (unlikely(rc != 0))
1116                 return rc;
1117
1118         lastid = le64_to_cpu(lastid);
1119         if (lastid < lls->lls_lastid_known) {
1120                 struct lfsck_instance   *lfsck  = com->lc_lfsck;
1121                 struct lfsck_layout     *lo     = com->lc_file_ram;
1122
1123                 lls->lls_lastid = lls->lls_lastid_known;
1124                 lls->lls_dirty = 1;
1125                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1126                         LASSERT(lfsck->li_out_notify != NULL);
1127
1128                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1129                                              LE_LASTID_REBUILDING);
1130                         lo->ll_flags |= LF_CRASHED_LASTID;
1131
1132                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
1133                                "LAST_ID file (1) for the sequence "LPX64
1134                                ", old value "LPU64", known value "LPU64"\n",
1135                                lfsck_lfsck2name(lfsck), lls->lls_seq,
1136                                lastid, lls->lls_lastid);
1137                 }
1138         } else if (lastid >= lls->lls_lastid) {
1139                 lls->lls_lastid = lastid;
1140                 lls->lls_dirty = 0;
1141         }
1142
1143         return 0;
1144 }
1145
1146 static int
1147 lfsck_layout_lastid_store(const struct lu_env *env,
1148                           struct lfsck_component *com)
1149 {
1150         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1151         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
1152         struct dt_device                *dt     = lfsck->li_bottom;
1153         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
1154         struct lfsck_layout_seq         *lls;
1155         struct thandle                  *th;
1156         __u64                            lastid;
1157         int                              rc     = 0;
1158         int                              rc1    = 0;
1159
1160         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1161                 loff_t pos = 0;
1162
1163                 if (!lls->lls_dirty)
1164                         continue;
1165
1166                 CDEBUG(D_LFSCK, "%s: layout LFSCK will sync the LAST_ID for "
1167                        "<seq> "LPX64" as <oid> "LPU64"\n",
1168                        lfsck_lfsck2name(lfsck), lls->lls_seq, lls->lls_lastid);
1169
1170                 if (bk->lb_param & LPF_DRYRUN) {
1171                         lls->lls_dirty = 0;
1172                         continue;
1173                 }
1174
1175                 th = dt_trans_create(env, dt);
1176                 if (IS_ERR(th)) {
1177                         rc1 = PTR_ERR(th);
1178                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1179                                "the LAST_ID for <seq> "LPX64"(1): rc = %d\n",
1180                                lfsck_lfsck2name(com->lc_lfsck),
1181                                lls->lls_seq, rc1);
1182                         continue;
1183                 }
1184
1185                 lastid = cpu_to_le64(lls->lls_lastid);
1186                 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1187                                              lfsck_buf_get(env, &lastid,
1188                                                            sizeof(lastid)),
1189                                              pos, th);
1190                 if (rc != 0)
1191                         goto stop;
1192
1193                 rc = dt_trans_start_local(env, dt, th);
1194                 if (rc != 0)
1195                         goto stop;
1196
1197                 dt_write_lock(env, lls->lls_lastid_obj, 0);
1198                 rc = dt_record_write(env, lls->lls_lastid_obj,
1199                                      lfsck_buf_get(env, &lastid,
1200                                      sizeof(lastid)), &pos, th);
1201                 dt_write_unlock(env, lls->lls_lastid_obj);
1202                 if (rc == 0)
1203                         lls->lls_dirty = 0;
1204
1205 stop:
1206                 dt_trans_stop(env, dt, th);
1207                 if (rc != 0) {
1208                         rc1 = rc;
1209                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1210                                "the LAST_ID for <seq> "LPX64"(2): rc = %d\n",
1211                                lfsck_lfsck2name(com->lc_lfsck),
1212                                lls->lls_seq, rc1);
1213                 }
1214         }
1215
1216         return rc1;
1217 }
1218
1219 static int
1220 lfsck_layout_lastid_load(const struct lu_env *env,
1221                          struct lfsck_component *com,
1222                          struct lfsck_layout_seq *lls)
1223 {
1224         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1225         struct lfsck_layout     *lo     = com->lc_file_ram;
1226         struct lu_fid           *fid    = &lfsck_env_info(env)->lti_fid;
1227         struct dt_object        *obj;
1228         loff_t                   pos    = 0;
1229         int                      rc;
1230         ENTRY;
1231
1232         lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck));
1233         obj = dt_locate(env, lfsck->li_bottom, fid);
1234         if (IS_ERR(obj))
1235                 RETURN(PTR_ERR(obj));
1236
1237         /* LAST_ID crashed, to be rebuilt */
1238         if (dt_object_exists(obj) == 0) {
1239                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1240                         LASSERT(lfsck->li_out_notify != NULL);
1241
1242                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1243                                              LE_LASTID_REBUILDING);
1244                         lo->ll_flags |= LF_CRASHED_LASTID;
1245
1246                         CDEBUG(D_LFSCK, "%s: layout LFSCK cannot find the "
1247                                "LAST_ID file for sequence "LPX64"\n",
1248                                lfsck_lfsck2name(lfsck), lls->lls_seq);
1249
1250                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1251                             cfs_fail_val > 0) {
1252                                 struct l_wait_info lwi = LWI_TIMEOUT(
1253                                                 cfs_time_seconds(cfs_fail_val),
1254                                                 NULL, NULL);
1255
1256                                 /* Some others may changed the cfs_fail_val
1257                                  * as zero after above check, re-check it for
1258                                  * sure to avoid falling into wait for ever. */
1259                                 if (likely(lwi.lwi_timeout > 0)) {
1260                                         struct ptlrpc_thread *thread =
1261                                                 &lfsck->li_thread;
1262
1263                                         up_write(&com->lc_sem);
1264                                         l_wait_event(thread->t_ctl_waitq,
1265                                                      !thread_is_running(thread),
1266                                                      &lwi);
1267                                         down_write(&com->lc_sem);
1268                                 }
1269                         }
1270                 }
1271
1272                 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1273         } else {
1274                 dt_read_lock(env, obj, 0);
1275                 rc = dt_read(env, obj,
1276                         lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1277                         &pos);
1278                 dt_read_unlock(env, obj);
1279                 if (rc != 0 && rc != sizeof(__u64))
1280                         GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1281
1282                 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1283                         LASSERT(lfsck->li_out_notify != NULL);
1284
1285                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1286                                              LE_LASTID_REBUILDING);
1287                         lo->ll_flags |= LF_CRASHED_LASTID;
1288
1289                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds invalid "
1290                                "LAST_ID file for the sequence "LPX64
1291                                ": rc = %d\n",
1292                                lfsck_lfsck2name(lfsck), lls->lls_seq, rc);
1293                 }
1294
1295                 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1296                 rc = 0;
1297         }
1298
1299         GOTO(out, rc);
1300
1301 out:
1302         if (rc != 0)
1303                 lfsck_object_put(env, obj);
1304         else
1305                 lls->lls_lastid_obj = obj;
1306
1307         return rc;
1308 }
1309
1310 static void lfsck_layout_record_failure(const struct lu_env *env,
1311                                         struct lfsck_instance *lfsck,
1312                                         struct lfsck_layout *lo)
1313 {
1314         __u64 cookie;
1315
1316         lo->ll_objs_failed_phase1++;
1317         cookie = lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
1318                                                         lfsck->li_di_oit);
1319         if (lo->ll_pos_first_inconsistent == 0 ||
1320             lo->ll_pos_first_inconsistent < cookie) {
1321                 lo->ll_pos_first_inconsistent = cookie;
1322
1323                 CDEBUG(D_LFSCK, "%s: layout LFSCK hit first non-repaired "
1324                        "inconsistency at the pos ["LPU64"]\n",
1325                        lfsck_lfsck2name(lfsck),
1326                        lo->ll_pos_first_inconsistent);
1327         }
1328 }
1329
1330 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1331                                            struct lfsck_component *com,
1332                                            int rc)
1333 {
1334         struct lfsck_instance   *lfsck = com->lc_lfsck;
1335         struct lfsck_layout     *lo    = com->lc_file_ram;
1336
1337         down_write(&com->lc_sem);
1338         lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1339                                   HALF_SEC - com->lc_time_last_checkpoint);
1340         lo->ll_time_last_checkpoint = cfs_time_current_sec();
1341         lo->ll_objs_checked_phase2 += com->lc_new_checked;
1342
1343         if (rc > 0) {
1344                 if (lo->ll_flags & LF_INCOMPLETE) {
1345                         lo->ll_status = LS_PARTIAL;
1346                 } else {
1347                         if (lfsck->li_master) {
1348                                 struct lfsck_assistant_data *lad = com->lc_data;
1349
1350                                 if (lad->lad_incomplete)
1351                                         lo->ll_status = LS_PARTIAL;
1352                                 else
1353                                         lo->ll_status = LS_COMPLETED;
1354                         } else {
1355                                 lo->ll_status = LS_COMPLETED;
1356                         }
1357                 }
1358                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
1359                         lo->ll_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
1360                 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1361                 lo->ll_success_count++;
1362         } else if (rc == 0) {
1363                 if (lfsck->li_status != 0)
1364                         lo->ll_status = lfsck->li_status;
1365                 else
1366                         lo->ll_status = LS_STOPPED;
1367         } else {
1368                 lo->ll_status = LS_FAILED;
1369         }
1370
1371         rc = lfsck_layout_store(env, com);
1372         up_write(&com->lc_sem);
1373
1374         return rc;
1375 }
1376
1377 static int lfsck_layout_trans_stop(const struct lu_env *env,
1378                                    struct dt_device *dev,
1379                                    struct thandle *handle, int result)
1380 {
1381         int rc;
1382
1383         /* XXX: If there is something worng or it needs to repair nothing,
1384          *      then notify the lower to stop the modification. Currently,
1385          *      we use th_result for such purpose, that may be replaced by
1386          *      some rollback mechanism in the future. */
1387         handle->th_result = result;
1388         rc = dt_trans_stop(env, dev, handle);
1389         if (result != 0)
1390                 return result > 0 ? 0 : result;
1391
1392         return rc == 0 ? 1 : rc;
1393 }
1394
1395 /**
1396  * Get the system default stripe size.
1397  *
1398  * \param[in] env       pointer to the thread context
1399  * \param[in] lfsck     pointer to the lfsck instance
1400  * \param[out] size     pointer to the default stripe size
1401  *
1402  * \retval              0 for success
1403  * \retval              negative error number on failure
1404  */
1405 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1406                                            struct lfsck_instance *lfsck,
1407                                            __u32 *size)
1408 {
1409         struct lov_user_md      *lum = &lfsck_env_info(env)->lti_lum;
1410         struct dt_object        *root;
1411         int                      rc;
1412
1413         root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1414         if (IS_ERR(root))
1415                 return PTR_ERR(root);
1416
1417         /* Get the default stripe size via xattr_get on the backend root. */
1418         rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1419                           XATTR_NAME_LOV);
1420         if (rc > 0) {
1421                 /* The lum->lmm_stripe_size is LE mode. The *size also
1422                  * should be LE mode. So it is unnecessary to convert. */
1423                 *size = lum->lmm_stripe_size;
1424                 rc = 0;
1425         } else if (unlikely(rc == 0)) {
1426                 rc = -EINVAL;
1427         }
1428
1429         lfsck_object_put(env, root);
1430
1431         return rc;
1432 }
1433
1434 /**
1435  * \retval       +1: repaired
1436  * \retval        0: did nothing
1437  * \retval      -ve: on error
1438  */
1439 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1440                                      struct thandle *handle,
1441                                      struct dt_object *parent,
1442                                      struct lu_fid *cfid,
1443                                      struct lu_buf *buf,
1444                                      struct lov_ost_data_v1 *slot,
1445                                      int fl, __u32 ost_idx)
1446 {
1447         struct ost_id           *oi     = &lfsck_env_info(env)->lti_oi;
1448         struct lov_mds_md_v1    *lmm    = buf->lb_buf;
1449         struct lu_buf            ea_buf;
1450         int                      rc;
1451         __u32                    magic;
1452         __u16                    count;
1453         ENTRY;
1454
1455         magic = le32_to_cpu(lmm->lmm_magic);
1456         count = le16_to_cpu(lmm->lmm_stripe_count);
1457
1458         fid_to_ostid(cfid, oi);
1459         ostid_cpu_to_le(oi, &slot->l_ost_oi);
1460         slot->l_ost_gen = cpu_to_le32(0);
1461         slot->l_ost_idx = cpu_to_le32(ost_idx);
1462
1463         if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE) {
1464                 struct lov_ost_data_v1 *objs;
1465                 int                     i;
1466
1467                 if (magic == LOV_MAGIC_V1)
1468                         objs = &lmm->lmm_objects[0];
1469                 else
1470                         objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1471                 for (i = 0; i < count; i++, objs++) {
1472                         if (objs != slot && lovea_slot_is_dummy(objs))
1473                                 break;
1474                 }
1475
1476                 /* If the @slot is the last dummy slot to be refilled,
1477                  * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1478                 if (i == count)
1479                         lmm->lmm_pattern &= ~cpu_to_le32(LOV_PATTERN_F_HOLE);
1480         }
1481
1482         lfsck_buf_init(&ea_buf, lmm, lov_mds_md_size(count, magic));
1483         rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle);
1484         if (rc == 0)
1485                 rc = 1;
1486
1487         RETURN(rc);
1488 }
1489
1490 /**
1491  * \retval       +1: repaired
1492  * \retval        0: did nothing
1493  * \retval      -ve: on error
1494  */
1495 static int lfsck_layout_extend_lovea(const struct lu_env *env,
1496                                      struct lfsck_instance *lfsck,
1497                                      struct thandle *handle,
1498                                      struct dt_object *parent,
1499                                      struct lu_fid *cfid,
1500                                      struct lu_buf *buf, int fl,
1501                                      __u32 ost_idx, __u32 ea_off, bool reset)
1502 {
1503         struct lov_mds_md_v1    *lmm    = buf->lb_buf;
1504         struct lov_ost_data_v1  *objs;
1505         int                      rc;
1506         __u16                    count;
1507         bool                     hole   = false;
1508         ENTRY;
1509
1510         if (fl == LU_XATTR_CREATE || reset) {
1511                 __u32 pattern = LOV_PATTERN_RAID0;
1512
1513                 count = ea_off + 1;
1514                 LASSERT(buf->lb_len >= lov_mds_md_size(count, LOV_MAGIC_V1));
1515
1516                 if (ea_off != 0 || reset) {
1517                         pattern |= LOV_PATTERN_F_HOLE;
1518                         hole = true;
1519                 }
1520
1521                 memset(lmm, 0, buf->lb_len);
1522                 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1523                 lmm->lmm_pattern = cpu_to_le32(pattern);
1524                 fid_to_lmm_oi(lfsck_dto2fid(parent), &lmm->lmm_oi);
1525                 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1526
1527                 rc = lfsck_layout_get_def_stripesize(env, lfsck,
1528                                                      &lmm->lmm_stripe_size);
1529                 if (rc != 0)
1530                         RETURN(rc);
1531
1532                 objs = &lmm->lmm_objects[ea_off];
1533         } else {
1534                 __u32   magic = le32_to_cpu(lmm->lmm_magic);
1535                 int     gap;
1536
1537                 count = le16_to_cpu(lmm->lmm_stripe_count);
1538                 if (magic == LOV_MAGIC_V1)
1539                         objs = &lmm->lmm_objects[count];
1540                 else
1541                         objs = &((struct lov_mds_md_v3 *)lmm)->
1542                                                         lmm_objects[count];
1543
1544                 gap = ea_off - count;
1545                 if (gap >= 0)
1546                         count = ea_off + 1;
1547                 LASSERT(buf->lb_len >= lov_mds_md_size(count, magic));
1548
1549                 if (gap > 0) {
1550                         memset(objs, 0, gap * sizeof(*objs));
1551                         lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
1552                         hole = true;
1553                 }
1554
1555                 lmm->lmm_layout_gen =
1556                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
1557                 objs += gap;
1558         }
1559
1560         lmm->lmm_stripe_count = cpu_to_le16(count);
1561         rc = lfsck_layout_refill_lovea(env, handle, parent, cfid, buf, objs,
1562                                        fl, ost_idx);
1563
1564         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
1565                DFID": parent "DFID", OST-index %u, stripe-index %u, fl %d, "
1566                "reset %s, %s LOV EA hole: rc = %d\n",
1567                lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1568                ost_idx, ea_off, fl, reset ? "yes" : "no",
1569                hole ? "with" : "without", rc);
1570
1571         RETURN(rc);
1572 }
1573
1574 static int __lfsck_layout_update_pfid(const struct lu_env *env,
1575                                       struct dt_object *child,
1576                                       const struct lu_fid *pfid, __u32 offset)
1577 {
1578         struct dt_device        *dev    = lfsck_obj2dev(child);
1579         struct filter_fid       *ff     = &lfsck_env_info(env)->lti_new_pfid;
1580         struct thandle          *handle;
1581         struct lu_buf            buf    = { NULL };
1582         int                      rc;
1583
1584         ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
1585         ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
1586         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
1587          * MDT-object's FID::f_ver, instead it is the OST-object index in its
1588          * parent MDT-object's layout EA. */
1589         ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
1590         lfsck_buf_init(&buf, ff, sizeof(struct filter_fid));
1591
1592         handle = dt_trans_create(env, dev);
1593         if (IS_ERR(handle))
1594                 RETURN(PTR_ERR(handle));
1595
1596         rc = dt_declare_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
1597         if (rc != 0)
1598                 GOTO(stop, rc);
1599
1600         rc = dt_trans_start_local(env, dev, handle);
1601         if (rc != 0)
1602                 GOTO(stop, rc);
1603
1604         rc = dt_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
1605
1606         GOTO(stop, rc);
1607
1608 stop:
1609         dt_trans_stop(env, dev, handle);
1610
1611         return rc;
1612 }
1613
1614 /**
1615  * \retval       +1: repaired
1616  * \retval        0: did nothing
1617  * \retval      -ve: on error
1618  */
1619 static int lfsck_layout_update_pfid(const struct lu_env *env,
1620                                     struct lfsck_component *com,
1621                                     struct dt_object *parent,
1622                                     struct lu_fid *cfid,
1623                                     struct dt_device *cdev, __u32 ea_off)
1624 {
1625         struct dt_object        *child;
1626         int                      rc     = 0;
1627         ENTRY;
1628
1629         child = lfsck_object_find_by_dev(env, cdev, cfid);
1630         if (IS_ERR(child))
1631                 RETURN(PTR_ERR(child));
1632
1633         rc = __lfsck_layout_update_pfid(env, child,
1634                                         lu_object_fid(&parent->do_lu), ea_off);
1635         lfsck_object_put(env, child);
1636
1637         RETURN(rc == 0 ? 1 : rc);
1638 }
1639
1640 /**
1641  * This function will create the MDT-object with the given (partial) LOV EA.
1642  *
1643  * Under some data corruption cases, the MDT-object of the file may be lost,
1644  * but its OST-objects, or some of them are there. The layout LFSCK needs to
1645  * re-create the MDT-object with the orphan OST-object(s) information.
1646  *
1647  * On the other hand, the LFSCK may has created some OST-object for repairing
1648  * dangling LOV EA reference, but as the LFSCK processing, it may find that
1649  * the old OST-object is there and should replace the former new created OST
1650  * object. Unfortunately, some others have modified such newly created object.
1651  * To keep the data (both new and old), the LFSCK will create MDT-object with
1652  * new FID to reference the original OST-object.
1653  *
1654  * \param[in] env       pointer to the thread context
1655  * \param[in] com       pointer to the lfsck component
1656  * \param[in] ltd       pointer to target device descriptor
1657  * \param[in] rec       pointer to the record for the orphan OST-object
1658  * \param[in] cfid      pointer to FID for the orphan OST-object
1659  * \param[in] infix     additional information, such as the FID for original
1660  *                      MDT-object and the stripe offset in the LOV EA
1661  * \param[in] type      the type for describing why the orphan MDT-object is
1662  *                      created. The rules are as following:
1663  *
1664  *  type "C":           Multiple OST-objects claim the same MDT-object and the
1665  *                      same slot in the layout EA. Then the LFSCK will create
1666  *                      new MDT-object(s) to hold the conflict OST-object(s).
1667  *
1668  *  type "N":           The orphan OST-object does not know which one was the
1669  *                      real parent MDT-object, so the LFSCK uses new FID for
1670  *                      its parent MDT-object.
1671  *
1672  *  type "R":           The orphan OST-object knows its parent MDT-object FID,
1673  *                      but does not know the position (the file name) in the
1674  *                      layout.
1675  *
1676  *  type "D":           The MDT-object is a directory, it may knows its parent
1677  *                      but because there is no valid linkEA, the LFSCK cannot
1678  *                      know where to put it back to the namespace.
1679  *  type "O":           The MDT-object has no linkEA, and there is no name
1680  *                      entry that references the MDT-object.
1681  *
1682  *  type "P":           The orphan object to be created was a parent directory
1683  *                      of some MDT-object which linkEA shows that the @orphan
1684  *                      object is missing.
1685  *
1686  * The orphan name will be like:
1687  * ${FID}-${infix}-${type}-${conflict_version}
1688  *
1689  * \param[in] ea_off    the stripe offset in the LOV EA
1690  *
1691  * \retval              positive on repaired something
1692  * \retval              0 if needs to repair nothing
1693  * \retval              negative error number on failure
1694  */
1695 static int lfsck_layout_recreate_parent(const struct lu_env *env,
1696                                         struct lfsck_component *com,
1697                                         struct lfsck_tgt_desc *ltd,
1698                                         struct lu_orphan_rec *rec,
1699                                         struct lu_fid *cfid,
1700                                         const char *infix,
1701                                         const char *type,
1702                                         __u32 ea_off)
1703 {
1704         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1705         struct dt_insert_rec            *dtrec  = &info->lti_dt_rec;
1706         char                            *name   = info->lti_key;
1707         struct lu_attr                  *la     = &info->lti_la2;
1708         struct dt_object_format         *dof    = &info->lti_dof;
1709         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1710         struct lu_fid                   *pfid   = &rec->lor_fid;
1711         struct lu_fid                   *tfid   = &info->lti_fid3;
1712         struct dt_device                *dev    = lfsck->li_bottom;
1713         struct dt_object                *lpf    = lfsck->li_lpf_obj;
1714         struct dt_object                *pobj   = NULL;
1715         struct dt_object                *cobj   = NULL;
1716         struct thandle                  *th     = NULL;
1717         struct lu_buf                   *ea_buf = &info->lti_big_buf;
1718         struct lu_buf                    lov_buf;
1719         struct lfsck_lock_handle        *llh    = &info->lti_llh;
1720         struct linkea_data               ldata  = { NULL };
1721         struct lu_buf                    linkea_buf;
1722         const struct lu_name            *pname;
1723         int                              size   = 0;
1724         int                              idx    = 0;
1725         int                              rc     = 0;
1726         ENTRY;
1727
1728         if (unlikely(lpf == NULL))
1729                 GOTO(log, rc = -ENXIO);
1730
1731         /* We use two separated transactions to repair the inconsistency.
1732          *
1733          * 1) create the MDT-object locally.
1734          * 2) update the OST-object's PFID EA if necessary.
1735          *
1736          * If 1) succeed, but 2) failed, then the OST-object's PFID EA will be
1737          * updated when the layout LFSCK run next time.
1738          *
1739          * If 1) failed, but 2) succeed, then such MDT-object will be re-created
1740          * when the layout LFSCK run next time. */
1741
1742         if (fid_is_zero(pfid)) {
1743                 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
1744                 if (rc != 0)
1745                         GOTO(log, rc);
1746
1747                 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
1748                 if (IS_ERR(cobj))
1749                         GOTO(log, rc = PTR_ERR(cobj));
1750         }
1751
1752         pobj = lfsck_object_find_by_dev(env, dev, pfid);
1753         if (IS_ERR(pobj))
1754                 GOTO(log, rc = PTR_ERR(pobj));
1755
1756         LASSERT(infix != NULL);
1757         LASSERT(type != NULL);
1758
1759         memset(la, 0, sizeof(*la));
1760         la->la_uid = rec->lor_uid;
1761         la->la_gid = rec->lor_gid;
1762         la->la_mode = S_IFREG | S_IRUSR;
1763         la->la_valid = LA_MODE | LA_UID | LA_GID;
1764
1765         memset(dof, 0, sizeof(*dof));
1766         dof->dof_type = dt_mode_to_dft(S_IFREG);
1767         /* Because the dof->dof_reg.striped = 0, the LOD will not create
1768          * the stripe(s). The LFSCK will specify the LOV EA via
1769          * lfsck_layout_extend_lovea(). */
1770
1771         size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
1772         if (ea_buf->lb_len < size) {
1773                 lu_buf_realloc(ea_buf, size);
1774                 if (ea_buf->lb_buf == NULL)
1775                         GOTO(log, rc = -ENOMEM);
1776         }
1777
1778 again:
1779         do {
1780                 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
1781                          type, idx++);
1782                 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1783                                (const struct dt_key *)name);
1784                 if (rc != 0 && rc != -ENOENT)
1785                         GOTO(log, rc);
1786         } while (rc == 0);
1787
1788         rc = lfsck_lock(env, lfsck, lfsck->li_lpf_obj, name, llh,
1789                         MDS_INODELOCK_UPDATE, LCK_PW);
1790         if (rc != 0)
1791                 GOTO(log, rc);
1792
1793         /* Re-check whether the name conflict with othrs after taken
1794          * the ldlm lock. */
1795         rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1796                        (const struct dt_key *)name);
1797         if (unlikely(rc == 0)) {
1798                 lfsck_unlock(llh);
1799                 goto again;
1800         }
1801
1802         if (rc != -ENOENT)
1803                 GOTO(unlock, rc);
1804
1805         rc = linkea_data_new(&ldata,
1806                              &lfsck_env_info(env)->lti_linkea_buf);
1807         if (rc != 0)
1808                 GOTO(unlock, rc);
1809
1810         pname = lfsck_name_get_const(env, name, strlen(name));
1811         rc = linkea_add_buf(&ldata, pname, lfsck_dto2fid(lfsck->li_lpf_obj));
1812         if (rc != 0)
1813                 GOTO(unlock, rc);
1814
1815         /* The 1st transaction. */
1816         th = dt_trans_create(env, dev);
1817         if (IS_ERR(th))
1818                 GOTO(unlock, rc = PTR_ERR(th));
1819
1820         rc = dt_declare_create(env, pobj, la, NULL, dof, th);
1821         if (rc != 0)
1822                 GOTO(stop, rc);
1823
1824         lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
1825         rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
1826                                   LU_XATTR_REPLACE, th);
1827         if (rc != 0)
1828                 GOTO(stop, rc);
1829
1830         dtrec->rec_fid = pfid;
1831         dtrec->rec_type = S_IFREG;
1832         rc = dt_declare_insert(env, lpf,
1833                                (const struct dt_rec *)dtrec,
1834                                (const struct dt_key *)name, th);
1835         if (rc != 0)
1836                 GOTO(stop, rc);
1837
1838         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1839                        ldata.ld_leh->leh_len);
1840         rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
1841                                   XATTR_NAME_LINK, 0, th);
1842         if (rc != 0)
1843                 GOTO(stop, rc);
1844
1845         rc = dt_trans_start_local(env, dev, th);
1846         if (rc != 0)
1847                 GOTO(stop, rc);
1848
1849         dt_write_lock(env, pobj, 0);
1850         rc = dt_create(env, pobj, la, NULL, dof, th);
1851         if (rc == 0)
1852                 rc = lfsck_layout_extend_lovea(env, lfsck, th, pobj, cfid,
1853                                 &lov_buf, 0, ltd->ltd_index, ea_off, true);
1854         dt_write_unlock(env, pobj);
1855         if (rc < 0)
1856                 GOTO(stop, rc);
1857
1858         rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
1859                        (const struct dt_key *)name, th, 1);
1860         if (rc != 0)
1861                 GOTO(stop, rc);
1862
1863         rc = dt_xattr_set(env, pobj, &linkea_buf, XATTR_NAME_LINK, 0, th);
1864         if (rc == 0 && cobj != NULL) {
1865                 dt_trans_stop(env, dev, th);
1866                 th = NULL;
1867
1868                 /* The 2nd transaction. */
1869                 rc = __lfsck_layout_update_pfid(env, cobj, pfid, ea_off);
1870         }
1871
1872         GOTO(stop, rc);
1873
1874 stop:
1875         if (th != NULL)
1876                 dt_trans_stop(env, dev, th);
1877
1878 unlock:
1879         lfsck_unlock(llh);
1880
1881 log:
1882         if (cobj != NULL && !IS_ERR(cobj))
1883                 lfsck_object_put(env, cobj);
1884         if (pobj != NULL && !IS_ERR(pobj))
1885                 lfsck_object_put(env, pobj);
1886
1887         if (rc < 0)
1888                 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
1889                        "recreate the lost MDT-object: parent "DFID
1890                        ", child "DFID", OST-index %u, stripe-index %u, "
1891                        "infix %s, type %s: rc = %d\n",
1892                        lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
1893                        ltd->ltd_index, ea_off, infix, type, rc);
1894
1895         return rc >= 0 ? 1 : rc;
1896 }
1897
1898 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
1899                                                    struct lfsck_component *com,
1900                                                    const struct lu_fid *fid,
1901                                                    __u32 index)
1902 {
1903         struct lfsck_thread_info *info  = lfsck_env_info(env);
1904         struct lfsck_request     *lr    = &info->lti_lr;
1905         struct lfsck_instance    *lfsck = com->lc_lfsck;
1906         struct lfsck_tgt_desc    *ltd;
1907         struct ptlrpc_request    *req;
1908         struct lfsck_request     *tmp;
1909         struct obd_export        *exp;
1910         int                       rc    = 0;
1911         ENTRY;
1912
1913         ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
1914         if (unlikely(ltd == NULL))
1915                 RETURN(-ENXIO);
1916
1917         exp = ltd->ltd_exp;
1918         if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
1919                 GOTO(put, rc = -EOPNOTSUPP);
1920
1921         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
1922         if (req == NULL)
1923                 GOTO(put, rc = -ENOMEM);
1924
1925         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
1926         if (rc != 0) {
1927                 ptlrpc_request_free(req);
1928
1929                 GOTO(put, rc);
1930         }
1931
1932         memset(lr, 0, sizeof(*lr));
1933         lr->lr_event = LE_CONDITIONAL_DESTROY;
1934         lr->lr_active = LFSCK_TYPE_LAYOUT;
1935         lr->lr_fid = *fid;
1936
1937         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
1938         *tmp = *lr;
1939         ptlrpc_request_set_replen(req);
1940
1941         rc = ptlrpc_queue_wait(req);
1942         ptlrpc_req_finished(req);
1943
1944         GOTO(put, rc);
1945
1946 put:
1947         lfsck_tgt_put(ltd);
1948
1949         return rc;
1950 }
1951
1952 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
1953                                                   struct lfsck_component *com,
1954                                                   struct lfsck_request *lr)
1955 {
1956         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1957         struct lu_attr                  *la     = &info->lti_la;
1958         ldlm_policy_data_t              *policy = &info->lti_policy;
1959         struct ldlm_res_id              *resid  = &info->lti_resid;
1960         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1961         struct dt_device                *dev    = lfsck->li_bottom;
1962         struct lu_fid                   *fid    = &lr->lr_fid;
1963         struct dt_object                *obj;
1964         struct thandle                  *th     = NULL;
1965         struct lustre_handle             lh     = { 0 };
1966         __u64                            flags  = 0;
1967         int                              rc     = 0;
1968         ENTRY;
1969
1970         obj = lfsck_object_find_by_dev(env, dev, fid);
1971         if (IS_ERR(obj))
1972                 RETURN(PTR_ERR(obj));
1973
1974         dt_read_lock(env, obj, 0);
1975         if (dt_object_exists(obj) == 0 ||
1976             lfsck_is_dead_obj(obj)) {
1977                 dt_read_unlock(env, obj);
1978
1979                 GOTO(put, rc = -ENOENT);
1980         }
1981
1982         /* Get obj's attr without lock firstly. */
1983         rc = dt_attr_get(env, obj, la);
1984         dt_read_unlock(env, obj);
1985         if (rc != 0)
1986                 GOTO(put, rc);
1987
1988         if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
1989                 GOTO(put, rc = -ETXTBSY);
1990
1991         /* Acquire extent lock on [0, EOF] to sync with all possible written. */
1992         LASSERT(lfsck->li_namespace != NULL);
1993
1994         memset(policy, 0, sizeof(*policy));
1995         policy->l_extent.end = OBD_OBJECT_EOF;
1996         ost_fid_build_resid(fid, resid);
1997         rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
1998                                     policy, LCK_EX, &flags, ldlm_blocking_ast,
1999                                     ldlm_completion_ast, NULL, NULL, 0,
2000                                     LVB_T_NONE, NULL, &lh);
2001         if (rc != ELDLM_OK)
2002                 GOTO(put, rc = -EIO);
2003
2004         dt_write_lock(env, obj, 0);
2005         /* Get obj's attr within lock again. */
2006         rc = dt_attr_get(env, obj, la);
2007         if (rc != 0)
2008                 GOTO(unlock, rc);
2009
2010         if (la->la_ctime != 0)
2011                 GOTO(unlock, rc = -ETXTBSY);
2012
2013         th = dt_trans_create(env, dev);
2014         if (IS_ERR(th))
2015                 GOTO(unlock, rc = PTR_ERR(th));
2016
2017         rc = dt_declare_ref_del(env, obj, th);
2018         if (rc != 0)
2019                 GOTO(stop, rc);
2020
2021         rc = dt_declare_destroy(env, obj, th);
2022         if (rc != 0)
2023                 GOTO(stop, rc);
2024
2025         rc = dt_trans_start_local(env, dev, th);
2026         if (rc != 0)
2027                 GOTO(stop, rc);
2028
2029         rc = dt_ref_del(env, obj, th);
2030         if (rc != 0)
2031                 GOTO(stop, rc);
2032
2033         rc = dt_destroy(env, obj, th);
2034         if (rc == 0)
2035                 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2036                        "OST-object "DFID" that was created for reparing "
2037                        "dangling referenced case. But the original missing "
2038                        "OST-object is found now.\n",
2039                        lfsck_lfsck2name(lfsck), PFID(fid));
2040
2041         GOTO(stop, rc);
2042
2043 stop:
2044         dt_trans_stop(env, dev, th);
2045
2046 unlock:
2047         dt_write_unlock(env, obj);
2048         ldlm_lock_decref(&lh, LCK_EX);
2049
2050 put:
2051         lfsck_object_put(env, obj);
2052
2053         return rc;
2054 }
2055
2056 /**
2057  * Some OST-object has occupied the specified layout EA slot.
2058  * Such OST-object may be generated by the LFSCK when repair
2059  * dangling referenced MDT-object, which can be indicated by
2060  * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2061  * is true and such OST-object has not been modified yet, we
2062  * will replace it with the orphan OST-object; otherwise the
2063  * LFSCK will create new MDT-object to reference the orphan.
2064  *
2065  * \retval       +1: repaired
2066  * \retval        0: did nothing
2067  * \retval      -ve: on error
2068  */
2069 static int lfsck_layout_conflict_create(const struct lu_env *env,
2070                                         struct lfsck_component *com,
2071                                         struct lfsck_tgt_desc *ltd,
2072                                         struct lu_orphan_rec *rec,
2073                                         struct dt_object *parent,
2074                                         struct lu_fid *cfid,
2075                                         struct lu_buf *ea_buf,
2076                                         struct lov_ost_data_v1 *slot,
2077                                         __u32 ea_off)
2078 {
2079         struct lfsck_thread_info *info          = lfsck_env_info(env);
2080         struct lu_fid            *cfid2         = &info->lti_fid2;
2081         struct ost_id            *oi            = &info->lti_oi;
2082         struct lov_mds_md_v1     *lmm           = ea_buf->lb_buf;
2083         struct dt_device         *dev           = lfsck_obj2dev(parent);
2084         struct thandle           *th            = NULL;
2085         struct lustre_handle      lh            = { 0 };
2086         __u32                     ost_idx2      = le32_to_cpu(slot->l_ost_idx);
2087         int                       rc            = 0;
2088         ENTRY;
2089
2090         ostid_le_to_cpu(&slot->l_ost_oi, oi);
2091         rc = ostid_to_fid(cfid2, oi, ost_idx2);
2092         if (rc != 0)
2093                 GOTO(out, rc);
2094
2095         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2096                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2097                               LCK_EX);
2098         if (rc != 0)
2099                 GOTO(out, rc);
2100
2101         rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2102
2103         /* If the conflict OST-obejct is not created for fixing dangling
2104          * referenced MDT-object in former LFSCK check/repair, or it has
2105          * been modified by others, then we cannot destroy it. Re-create
2106          * a new MDT-object for the orphan OST-object. */
2107         if (rc == -ETXTBSY) {
2108                 /* No need the layout lock on the original parent. */
2109                 lfsck_ibits_unlock(&lh, LCK_EX);
2110
2111                 fid_zero(&rec->lor_fid);
2112                 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2113                          "-"DFID"-%x", PFID(lu_object_fid(&parent->do_lu)),
2114                          ea_off);
2115                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2116                                                 info->lti_tmpbuf, "C", ea_off);
2117
2118                 RETURN(rc);
2119         }
2120
2121         if (rc != 0 && rc != -ENOENT)
2122                 GOTO(unlock, rc);
2123
2124         th = dt_trans_create(env, dev);
2125         if (IS_ERR(th))
2126                 GOTO(unlock, rc = PTR_ERR(th));
2127
2128         rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2129                                   LU_XATTR_REPLACE, th);
2130         if (rc != 0)
2131                 GOTO(stop, rc);
2132
2133         rc = dt_trans_start_local(env, dev, th);
2134         if (rc != 0)
2135                 GOTO(stop, rc);
2136
2137         dt_write_lock(env, parent, 0);
2138         lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2139         rc = lfsck_layout_refill_lovea(env, th, parent, cfid, ea_buf, slot,
2140                                        LU_XATTR_REPLACE, ltd->ltd_index);
2141         dt_write_unlock(env, parent);
2142
2143         GOTO(stop, rc);
2144
2145 stop:
2146         dt_trans_stop(env, dev, th);
2147
2148 unlock:
2149         lfsck_ibits_unlock(&lh, LCK_EX);
2150
2151 out:
2152         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2153                "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2154                "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2155                lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2156                PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2157                ea_off, rc);
2158
2159         return rc >= 0 ? 1 : rc;
2160 }
2161
2162 /**
2163  * \retval       +1: repaired
2164  * \retval        0: did nothing
2165  * \retval      -ve: on error
2166  */
2167 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2168                                        struct lfsck_component *com,
2169                                        struct lfsck_tgt_desc *ltd,
2170                                        struct lu_orphan_rec *rec,
2171                                        struct dt_object *parent,
2172                                        struct lu_fid *cfid,
2173                                        __u32 ost_idx, __u32 ea_off)
2174 {
2175         struct lfsck_thread_info *info          = lfsck_env_info(env);
2176         struct lu_buf            *buf           = &info->lti_big_buf;
2177         struct lu_fid            *fid           = &info->lti_fid2;
2178         struct ost_id            *oi            = &info->lti_oi;
2179         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2180         struct dt_device         *dt            = lfsck_obj2dev(parent);
2181         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2182         struct thandle            *handle       = NULL;
2183         size_t                    lovea_size;
2184         struct lov_mds_md_v1     *lmm;
2185         struct lov_ost_data_v1   *objs;
2186         struct lustre_handle      lh            = { 0 };
2187         __u32                     magic;
2188         int                       fl            = 0;
2189         int                       rc            = 0;
2190         int                       rc1;
2191         int                       i;
2192         __u16                     count;
2193         bool                      locked        = false;
2194         ENTRY;
2195
2196         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2197                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2198                               LCK_EX);
2199         if (rc != 0) {
2200                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2201                        "LOV EA for "DFID": parent "DFID", OST-index %u, "
2202                        "stripe-index %u: rc = %d\n",
2203                        lfsck_lfsck2name(lfsck), PFID(cfid),
2204                        PFID(lfsck_dto2fid(parent)), ost_idx, ea_off, rc);
2205
2206                 RETURN(rc);
2207         }
2208
2209 again:
2210         if (locked) {
2211                 dt_write_unlock(env, parent);
2212                 locked = false;
2213         }
2214
2215         if (handle != NULL) {
2216                 dt_trans_stop(env, dt, handle);
2217                 handle = NULL;
2218         }
2219
2220         if (rc < 0)
2221                 GOTO(unlock_layout, rc);
2222
2223         lovea_size = rc;
2224         if (buf->lb_len < lovea_size) {
2225                 lu_buf_realloc(buf, lovea_size);
2226                 if (buf->lb_buf == NULL)
2227                         GOTO(unlock_layout, rc = -ENOMEM);
2228         }
2229
2230         if (!(bk->lb_param & LPF_DRYRUN)) {
2231                 handle = dt_trans_create(env, dt);
2232                 if (IS_ERR(handle))
2233                         GOTO(unlock_layout, rc = PTR_ERR(handle));
2234
2235                 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2236                                           fl, handle);
2237                 if (rc != 0)
2238                         GOTO(stop, rc);
2239
2240                 rc = dt_trans_start_local(env, dt, handle);
2241                 if (rc != 0)
2242                         GOTO(stop, rc);
2243         }
2244
2245         dt_write_lock(env, parent, 0);
2246         locked = true;
2247         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2248         if (rc == -ERANGE) {
2249                 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV);
2250                 LASSERT(rc != 0);
2251                 goto again;
2252         } else if (rc == -ENODATA || rc == 0) {
2253                 lovea_size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2254                 /* If the declared is not big enough, re-try. */
2255                 if (buf->lb_len < lovea_size) {
2256                         rc = lovea_size;
2257                         goto again;
2258                 }
2259                 fl = LU_XATTR_CREATE;
2260         } else if (rc < 0) {
2261                 GOTO(unlock_parent, rc);
2262         } else if (unlikely(buf->lb_len == 0)) {
2263                 goto again;
2264         } else {
2265                 fl = LU_XATTR_REPLACE;
2266                 lovea_size = rc;
2267         }
2268
2269         if (fl == LU_XATTR_CREATE) {
2270                 if (bk->lb_param & LPF_DRYRUN)
2271                         GOTO(unlock_parent, rc = 1);
2272
2273                 LASSERT(buf->lb_len >= lovea_size);
2274
2275                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2276                                                buf, fl, ost_idx, ea_off, false);
2277
2278                 GOTO(unlock_parent, rc);
2279         }
2280
2281         lmm = buf->lb_buf;
2282         rc1 = lfsck_layout_verify_header(lmm);
2283
2284         /* If the LOV EA crashed, the rebuild it. */
2285         if (rc1 == -EINVAL) {
2286                 if (bk->lb_param & LPF_DRYRUN)
2287                         GOTO(unlock_parent, rc = 1);
2288
2289                 LASSERT(buf->lb_len >= lovea_size);
2290
2291                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2292                                                buf, fl, ost_idx, ea_off, true);
2293
2294                 GOTO(unlock_parent, rc);
2295         }
2296
2297         /* For other unknown magic/pattern, keep the current LOV EA. */
2298         if (rc1 != 0)
2299                 GOTO(unlock_parent, rc = rc1);
2300
2301         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2302          * been verified in lfsck_layout_verify_header() already. If some
2303          * new magic introduced in the future, then layout LFSCK needs to
2304          * be updated also. */
2305         magic = le32_to_cpu(lmm->lmm_magic);
2306         if (magic == LOV_MAGIC_V1) {
2307                 objs = &lmm->lmm_objects[0];
2308         } else {
2309                 LASSERT(magic == LOV_MAGIC_V3);
2310                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2311         }
2312
2313         count = le16_to_cpu(lmm->lmm_stripe_count);
2314         if (count == 0)
2315                 GOTO(unlock_parent, rc = -EINVAL);
2316         LASSERT(count > 0);
2317
2318         /* Exceed the current end of MDT-object layout EA. Then extend it. */
2319         if (count <= ea_off) {
2320                 if (bk->lb_param & LPF_DRYRUN)
2321                         GOTO(unlock_parent, rc = 1);
2322
2323                 lovea_size = lov_mds_md_size(ea_off + 1, magic);
2324                 /* If the declared is not big enough, re-try. */
2325                 if (buf->lb_len < lovea_size) {
2326                         rc = lovea_size;
2327                         goto again;
2328                 }
2329
2330                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2331                                                buf, fl, ost_idx, ea_off, false);
2332
2333                 GOTO(unlock_parent, rc);
2334         }
2335
2336         LASSERTF(rc > 0, "invalid rc = %d\n", rc);
2337
2338         for (i = 0; i < count; i++, objs++) {
2339                 /* The MDT-object was created via lfsck_layout_recover_create()
2340                  * by others before, and we fill the dummy layout EA. */
2341                 if (lovea_slot_is_dummy(objs)) {
2342                         if (i != ea_off)
2343                                 continue;
2344
2345                         if (bk->lb_param & LPF_DRYRUN)
2346                                 GOTO(unlock_parent, rc = 1);
2347
2348                         lmm->lmm_layout_gen =
2349                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2350                         rc = lfsck_layout_refill_lovea(env, handle, parent,
2351                                                        cfid, buf, objs, fl,
2352                                                        ost_idx);
2353
2354                         CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
2355                                "dummy layout slot for "DFID": parent "DFID
2356                                ", OST-index %u, stripe-index %u: rc = %d\n",
2357                                lfsck_lfsck2name(lfsck), PFID(cfid),
2358                                PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
2359
2360                         GOTO(unlock_parent, rc);
2361                 }
2362
2363                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2364                 rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
2365                 if (rc != 0) {
2366                         CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
2367                                "invalid layout EA at the slot %d, index %u\n",
2368                                lfsck_lfsck2name(lfsck),
2369                                PFID(lfsck_dto2fid(parent)), i,
2370                                le32_to_cpu(objs->l_ost_idx));
2371
2372                         GOTO(unlock_parent, rc);
2373                 }
2374
2375                 /* It should be rare case, the slot is there, but the LFSCK
2376                  * does not handle it during the first-phase cycle scanning. */
2377                 if (unlikely(lu_fid_eq(fid, cfid))) {
2378                         if (i == ea_off) {
2379                                 GOTO(unlock_parent, rc = 0);
2380                         } else {
2381                                 /* Rare case that the OST-object index
2382                                  * does not match the parent MDT-object
2383                                  * layout EA. We trust the later one. */
2384                                 if (bk->lb_param & LPF_DRYRUN)
2385                                         GOTO(unlock_parent, rc = 1);
2386
2387                                 dt_write_unlock(env, parent);
2388                                 if (handle != NULL)
2389                                         dt_trans_stop(env, dt, handle);
2390                                 lfsck_ibits_unlock(&lh, LCK_EX);
2391                                 rc = lfsck_layout_update_pfid(env, com, parent,
2392                                                         cfid, ltd->ltd_tgt, i);
2393
2394                                 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
2395                                        "updated OST-object's pfid for "DFID
2396                                        ": parent "DFID", OST-index %u, "
2397                                        "stripe-index %u: rc = %d\n",
2398                                        lfsck_lfsck2name(lfsck), PFID(cfid),
2399                                        PFID(lfsck_dto2fid(parent)),
2400                                        ltd->ltd_index, i, rc);
2401
2402                                 RETURN(rc);
2403                         }
2404                 }
2405         }
2406
2407         /* The MDT-object exists, but related layout EA slot is occupied
2408          * by others. */
2409         if (bk->lb_param & LPF_DRYRUN)
2410                 GOTO(unlock_parent, rc = 1);
2411
2412         dt_write_unlock(env, parent);
2413         if (handle != NULL)
2414                 dt_trans_stop(env, dt, handle);
2415         lfsck_ibits_unlock(&lh, LCK_EX);
2416         if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
2417                 objs = &lmm->lmm_objects[ea_off];
2418         else
2419                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
2420         rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
2421                                           buf, objs, ea_off);
2422
2423         RETURN(rc);
2424
2425 unlock_parent:
2426         if (locked)
2427                 dt_write_unlock(env, parent);
2428
2429 stop:
2430         if (handle != NULL)
2431                 dt_trans_stop(env, dt, handle);
2432
2433 unlock_layout:
2434         lfsck_ibits_unlock(&lh, LCK_EX);
2435
2436         return rc;
2437 }
2438
2439 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
2440                                         struct lfsck_component *com,
2441                                         struct lfsck_tgt_desc *ltd,
2442                                         struct lu_orphan_rec *rec,
2443                                         struct lu_fid *cfid)
2444 {
2445         struct lfsck_layout     *lo     = com->lc_file_ram;
2446         struct lu_fid           *pfid   = &rec->lor_fid;
2447         struct dt_object        *parent = NULL;
2448         __u32                    ea_off = pfid->f_stripe_idx;
2449         int                      rc     = 0;
2450         ENTRY;
2451
2452         if (!fid_is_sane(cfid))
2453                 GOTO(out, rc = -EINVAL);
2454
2455         if (fid_is_zero(pfid)) {
2456                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2457                                                   "", "N", ea_off);
2458                 GOTO(out, rc);
2459         }
2460
2461         pfid->f_ver = 0;
2462         if (!fid_is_sane(pfid))
2463                 GOTO(out, rc = -EINVAL);
2464
2465         parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
2466         if (IS_ERR(parent))
2467                 GOTO(out, rc = PTR_ERR(parent));
2468
2469         if (unlikely(dt_object_remote(parent) != 0))
2470                 GOTO(put, rc = -EXDEV);
2471
2472         if (dt_object_exists(parent) == 0) {
2473                 lfsck_object_put(env, parent);
2474                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2475                                                   "", "R", ea_off);
2476                 GOTO(out, rc);
2477         }
2478
2479         if (!S_ISREG(lu_object_attr(&parent->do_lu)))
2480                 GOTO(put, rc = -EISDIR);
2481
2482         rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
2483                                          ltd->ltd_index, ea_off);
2484
2485         GOTO(put, rc);
2486
2487 put:
2488         if (rc <= 0)
2489                 lfsck_object_put(env, parent);
2490         else
2491                 /* The layout EA is changed, need to be reloaded next time. */
2492                 lu_object_put_nocache(env, &parent->do_lu);
2493
2494 out:
2495         down_write(&com->lc_sem);
2496         com->lc_new_scanned++;
2497         com->lc_new_checked++;
2498         if (rc > 0) {
2499                 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
2500                 rc = 0;
2501         } else if (rc < 0) {
2502                 lo->ll_objs_failed_phase2++;
2503         }
2504         up_write(&com->lc_sem);
2505
2506         return rc;
2507 }
2508
2509 static int lfsck_layout_scan_orphan(const struct lu_env *env,
2510                                     struct lfsck_component *com,
2511                                     struct lfsck_tgt_desc *ltd)
2512 {
2513         struct lfsck_assistant_data     *lad    = com->lc_data;
2514         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2515         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
2516         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2517         struct ost_id                   *oi     = &info->lti_oi;
2518         struct lu_fid                   *fid    = &info->lti_fid;
2519         struct dt_object                *obj;
2520         const struct dt_it_ops          *iops;
2521         struct dt_it                    *di;
2522         int                              rc     = 0;
2523         ENTRY;
2524
2525         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
2526                "scanning for OST%04x\n",
2527                lfsck_lfsck2name(lfsck), ltd->ltd_index);
2528
2529         if (cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
2530                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
2531                        "scanning for OST%04x\n",
2532                        lfsck_lfsck2name(lfsck), ltd->ltd_index);
2533
2534                 RETURN(0);
2535         }
2536
2537         ostid_set_seq(oi, FID_SEQ_IDIF);
2538         ostid_set_id(oi, 0);
2539         rc = ostid_to_fid(fid, oi, ltd->ltd_index);
2540         if (rc != 0)
2541                 GOTO(log, rc);
2542
2543         obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
2544         if (unlikely(IS_ERR(obj)))
2545                 GOTO(log, rc = PTR_ERR(obj));
2546
2547         rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_orphan_features);
2548         if (rc != 0)
2549                 GOTO(put, rc);
2550
2551         iops = &obj->do_index_ops->dio_it;
2552         di = iops->init(env, obj, 0);
2553         if (IS_ERR(di))
2554                 GOTO(put, rc = PTR_ERR(di));
2555
2556         rc = iops->load(env, di, 0);
2557         if (rc == -ESRCH) {
2558                 /* -ESRCH means that the orphan OST-objects rbtree has been
2559                  * cleanup because of the OSS server restart or other errors. */
2560                 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
2561                 GOTO(fini, rc);
2562         }
2563
2564         if (rc == 0)
2565                 rc = iops->next(env, di);
2566         else if (rc > 0)
2567                 rc = 0;
2568
2569         if (rc < 0)
2570                 GOTO(fini, rc);
2571
2572         if (rc > 0)
2573                 GOTO(fini, rc = 0);
2574
2575         do {
2576                 struct dt_key           *key;
2577                 struct lu_orphan_rec    *rec = &info->lti_rec;
2578
2579                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
2580                     unlikely(!thread_is_running(&lfsck->li_thread)))
2581                         break;
2582
2583                 key = iops->key(env, di);
2584                 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
2585                 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
2586                 if (rc == 0)
2587                         rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
2588                                         &com->lc_fid_latest_scanned_phase2);
2589                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2590                         GOTO(fini, rc);
2591
2592                 lfsck_control_speed_by_self(com);
2593                 do {
2594                         rc = iops->next(env, di);
2595                 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
2596         } while (rc == 0);
2597
2598         GOTO(fini, rc);
2599
2600 fini:
2601         iops->put(env, di);
2602         iops->fini(env, di);
2603 put:
2604         lfsck_object_put(env, obj);
2605
2606 log:
2607         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
2608                "scanning for OST%04x: rc = %d\n",
2609                lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
2610
2611         return rc > 0 ? 0 : rc;
2612 }
2613
2614 /* For the MDT-object with dangling reference, we need to repare the
2615  * inconsistency according to the LFSCK sponsor's requirement:
2616  *
2617  * 1) Keep the inconsistency there and report the inconsistency case,
2618  *    then give the chance to the application to find related issues,
2619  *    and the users can make the decision about how to handle it with
2620  *    more human knownledge. (by default)
2621  *
2622  * 2) Re-create the missing OST-object with the FID/owner information. */
2623 static int lfsck_layout_repair_dangling(const struct lu_env *env,
2624                                         struct lfsck_component *com,
2625                                         struct dt_object *parent,
2626                                         struct lfsck_layout_req *llr,
2627                                         struct lu_attr *la)
2628 {
2629         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2630         struct filter_fid               *pfid   = &info->lti_new_pfid;
2631         struct dt_object_format         *dof    = &info->lti_dof;
2632         struct dt_object                *child  = llr->llr_child;
2633         struct dt_device                *dev    = lfsck_obj2dev(child);
2634         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
2635         struct thandle                  *handle;
2636         struct lu_buf                   *buf;
2637         struct lustre_handle             lh     = { 0 };
2638         int                              rc;
2639         bool                             create;
2640         ENTRY;
2641
2642         if (com->lc_lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ)
2643                 create = true;
2644         else
2645                 create = false;
2646
2647         if (!create)
2648                 GOTO(log, rc = 1);
2649
2650         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2651                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2652                               LCK_EX);
2653         if (rc != 0)
2654                 GOTO(log, rc);
2655
2656         rc = dt_attr_get(env, parent, la);
2657         if (rc != 0)
2658                 GOTO(unlock1, rc);
2659
2660         la->la_mode = S_IFREG | 0666;
2661         la->la_atime = la->la_mtime = la->la_ctime = 0;
2662         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
2663                        LA_ATIME | LA_MTIME | LA_CTIME;
2664         memset(dof, 0, sizeof(*dof));
2665         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
2666         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
2667         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2668          * MDT-object's FID::f_ver, instead it is the OST-object index in its
2669          * parent MDT-object's layout EA. */
2670         pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
2671         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
2672
2673         handle = dt_trans_create(env, dev);
2674         if (IS_ERR(handle))
2675                 GOTO(unlock1, rc = PTR_ERR(handle));
2676
2677         rc = dt_declare_create(env, child, la, NULL, dof, handle);
2678         if (rc != 0)
2679                 GOTO(stop, rc);
2680
2681         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
2682                                   LU_XATTR_CREATE, handle);
2683         if (rc != 0)
2684                 GOTO(stop, rc);
2685
2686         rc = dt_trans_start_local(env, dev, handle);
2687         if (rc != 0)
2688                 GOTO(stop, rc);
2689
2690         dt_read_lock(env, parent, 0);
2691         if (unlikely(lfsck_is_dead_obj(parent)))
2692                 GOTO(unlock2, rc = 1);
2693
2694         rc = dt_create(env, child, la, NULL, dof, handle);
2695         if (rc != 0)
2696                 GOTO(unlock2, rc);
2697
2698         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
2699                           handle);
2700
2701         GOTO(unlock2, rc);
2702
2703 unlock2:
2704         dt_read_unlock(env, parent);
2705
2706 stop:
2707         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2708
2709 unlock1:
2710         lfsck_ibits_unlock(&lh, LCK_EX);
2711
2712 log:
2713         if (rc != 0)
2714                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
2715                        "dangling reference for: parent "DFID", child "DFID
2716                        ", OST-index %u, stripe-index %u, owner %u/%u. %s: "
2717                        "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
2718                        PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
2719                        llr->llr_ost_idx, llr->llr_lov_idx,
2720                        la->la_uid, la->la_gid,
2721                        create ? "Create the lost OST-object as required" :
2722                                 "Keep the MDT-object there by default", rc);
2723
2724         return rc;
2725 }
2726
2727 /* If the OST-object does not recognize the MDT-object as its parent, and
2728  * there is no other MDT-object claims as its parent, then just trust the
2729  * given MDT-object as its parent. So update the OST-object filter_fid. */
2730 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
2731                                               struct lfsck_component *com,
2732                                               struct dt_object *parent,
2733                                               struct lfsck_layout_req *llr,
2734                                               struct lu_attr *la)
2735 {
2736         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2737         struct filter_fid               *pfid   = &info->lti_new_pfid;
2738         struct dt_object                *child  = llr->llr_child;
2739         struct dt_device                *dev    = lfsck_obj2dev(child);
2740         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
2741         struct thandle                  *handle;
2742         struct lu_buf                   *buf;
2743         struct lustre_handle             lh     = { 0 };
2744         int                              rc;
2745         ENTRY;
2746
2747         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2748                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2749                               LCK_EX);
2750         if (rc != 0)
2751                 GOTO(log, rc);
2752
2753         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
2754         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
2755         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2756          * MDT-object's FID::f_ver, instead it is the OST-object index in its
2757          * parent MDT-object's layout EA. */
2758         pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
2759         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
2760
2761         handle = dt_trans_create(env, dev);
2762         if (IS_ERR(handle))
2763                 GOTO(unlock1, rc = PTR_ERR(handle));
2764
2765         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
2766         if (rc != 0)
2767                 GOTO(stop, rc);
2768
2769         rc = dt_attr_get(env, parent, la);
2770         if (rc != 0)
2771                 GOTO(stop, rc);
2772
2773         la->la_valid = LA_UID | LA_GID;
2774         rc = dt_declare_attr_set(env, child, la, handle);
2775         if (rc != 0)
2776                 GOTO(stop, rc);
2777
2778         rc = dt_trans_start_local(env, dev, handle);
2779         if (rc != 0)
2780                 GOTO(stop, rc);
2781
2782         dt_write_lock(env, parent, 0);
2783         if (unlikely(lfsck_is_dead_obj(parent)))
2784                 GOTO(unlock2, rc = 1);
2785
2786         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
2787         if (rc != 0)
2788                 GOTO(unlock2, rc);
2789
2790         /* Get the latest parent's owner. */
2791         rc = dt_attr_get(env, parent, la);
2792         if (rc != 0)
2793                 GOTO(unlock2, rc);
2794
2795         la->la_valid = LA_UID | LA_GID;
2796         rc = dt_attr_set(env, child, la, handle);
2797
2798         GOTO(unlock2, rc);
2799
2800 unlock2:
2801         dt_write_unlock(env, parent);
2802
2803 stop:
2804         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2805
2806 unlock1:
2807         lfsck_ibits_unlock(&lh, LCK_EX);
2808
2809 log:
2810         if (rc != 0)
2811                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
2812                        "unmatched MDT-OST pair for: parent "DFID
2813                        ", child "DFID", OST-index %u, stripe-index %u, "
2814                        "owner %u/%u: rc = %d\n",
2815                        lfsck_lfsck2name(com->lc_lfsck),
2816                        PFID(lfsck_dto2fid(parent)),
2817                        PFID(lfsck_dto2fid(child)),
2818                        llr->llr_ost_idx, llr->llr_lov_idx,
2819                        la->la_uid, la->la_gid, rc);
2820
2821         return rc;
2822 }
2823
2824 /* If there are more than one MDT-objects claim as the OST-object's parent,
2825  * and the OST-object only recognizes one of them, then we need to generate
2826  * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
2827 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
2828                                                    struct lfsck_component *com,
2829                                                    struct dt_object *parent,
2830                                                    struct lfsck_layout_req *llr,
2831                                                    struct lu_attr *la,
2832                                                    struct lu_buf *buf)
2833 {
2834         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2835         struct dt_allocation_hint       *hint   = &info->lti_hint;
2836         struct dt_object_format         *dof    = &info->lti_dof;
2837         struct ost_id                   *oi     = &info->lti_oi;
2838         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2839         struct dt_device                *dev;
2840         struct lu_device                *d      =
2841                                 &lfsck_obj2dev(llr->llr_child)->dd_lu_dev;
2842         struct lu_object                *o;
2843         struct lu_object                *n;
2844         struct dt_object                *child  = NULL;
2845         struct thandle                  *handle = NULL;
2846         struct lov_mds_md_v1            *lmm;
2847         struct lov_ost_data_v1          *objs;
2848         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
2849         struct lu_fid                    tfid;
2850         struct lustre_handle             lh     = { 0 };
2851         struct lu_buf                    ea_buf;
2852         __u32                            magic;
2853         __u32                            index;
2854         int                              rc;
2855         ENTRY;
2856
2857         /* We use two separated transactions to repair the inconsistency.
2858          *
2859          * 1) create the child (OST-object).
2860          * 2) update the parent LOV EA according to the child's FID.
2861          *
2862          * If 1) succeed, but 2) failed or aborted, then such OST-object will be
2863          * handled as orphan when the layout LFSCK run next time.
2864          *
2865          * If 1) failed, but 2) succeed, then such OST-object will be re-created
2866          * as dangling referened case when the layout LFSCK run next time. */
2867
2868         /* The 1st transaction. */
2869         o = lu_object_anon(env, d, NULL);
2870         if (IS_ERR(o))
2871                 GOTO(log, rc = PTR_ERR(o));
2872
2873         n = lu_object_locate(o->lo_header, d->ld_type);
2874         if (unlikely(n == NULL)) {
2875                 lu_object_put_nocache(env, o);
2876
2877                 GOTO(log, rc = -EINVAL);
2878         }
2879
2880         child = container_of(n, struct dt_object, do_lu);
2881         memset(hint, 0, sizeof(*hint));
2882         rc = dt_attr_get(env, parent, la);
2883         if (rc != 0)
2884                 GOTO(log, rc);
2885
2886         la->la_valid = LA_UID | LA_GID;
2887         memset(dof, 0, sizeof(*dof));
2888
2889         dev = lfsck_obj2dev(child);
2890         handle = dt_trans_create(env, dev);
2891         if (IS_ERR(handle))
2892                 GOTO(log, rc = PTR_ERR(handle));
2893
2894         rc = dt_declare_create(env, child, la, hint, dof, handle);
2895         if (rc != 0)
2896                 GOTO(stop, rc);
2897
2898         rc = dt_trans_start_local(env, dev, handle);
2899         if (rc != 0)
2900                 GOTO(stop, rc);
2901
2902         rc = dt_create(env, child, la, hint, dof, handle);
2903         dt_trans_stop(env, dev, handle);
2904         handle = NULL;
2905         if (rc != 0)
2906                 GOTO(log, rc);
2907
2908         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2909                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2910                               LCK_EX);
2911         if (rc != 0)
2912                 GOTO(log, rc);
2913
2914         /* The 2nd transaction. */
2915
2916         /* XXX: Generally, we should use bottom device (OSD) to update parent
2917          *      LOV EA. But because the LOD-object still references the wrong
2918          *      OSP-object that should be detached after the parent's LOV EA
2919          *      refreshed. Unfortunately, there is no suitable API for that.
2920          *      So we have to make the LOD to re-load the OSP-object(s) via
2921          *      replacing the LOV EA against the LOD-object.
2922          *
2923          *      Once the DNE2 patches have been landed, we can replace the
2924          *      LOD device with the OSD device. LU-6230. */
2925
2926         dev = lfsck->li_next;
2927         parent = lfsck_object_locate(dev, parent);
2928         if (IS_ERR(parent))
2929                 GOTO(log, rc = PTR_ERR(parent));
2930
2931         handle = dt_trans_create(env, dev);
2932         if (IS_ERR(handle))
2933                 GOTO(log, rc = PTR_ERR(handle));
2934
2935         rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2936                                   LU_XATTR_REPLACE, handle);
2937         if (rc != 0)
2938                 GOTO(stop, rc);
2939
2940         rc = dt_trans_start_local(env, dev, handle);
2941         if (rc != 0)
2942                 GOTO(stop, rc);
2943
2944         dt_write_lock(env, parent, 0);
2945         if (unlikely(lfsck_is_dead_obj(parent)))
2946                 GOTO(unlock, rc = 0);
2947
2948         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2949         if (unlikely(rc == 0 || rc == -ENODATA || rc == -ERANGE))
2950                 GOTO(unlock, rc = 0);
2951
2952         lmm = buf->lb_buf;
2953         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2954          * been verified in lfsck_layout_verify_header() already. If some
2955          * new magic introduced in the future, then layout LFSCK needs to
2956          * be updated also. */
2957         magic = le32_to_cpu(lmm->lmm_magic);
2958         if (magic == LOV_MAGIC_V1) {
2959                 objs = &lmm->lmm_objects[llr->llr_lov_idx];
2960         } else {
2961                 LASSERT(magic == LOV_MAGIC_V3);
2962                 objs =
2963                 &((struct lov_mds_md_v3 *)lmm)->lmm_objects[llr->llr_lov_idx];
2964         }
2965
2966         ostid_le_to_cpu(&objs->l_ost_oi, oi);
2967         index = le32_to_cpu(objs->l_ost_idx);
2968         rc = ostid_to_fid(&tfid, oi, index);
2969         /* Someone changed layout during the LFSCK, no need to repair then. */
2970         if (rc == 0 && !lu_fid_eq(&tfid, lu_object_fid(&llr->llr_child->do_lu)))
2971                 GOTO(unlock, rc = 0);
2972
2973         lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2974         fid_to_ostid(lu_object_fid(&child->do_lu), oi);
2975         ostid_cpu_to_le(oi, &objs->l_ost_oi);
2976         objs->l_ost_gen = cpu_to_le32(0);
2977         objs->l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
2978         lfsck_buf_init(&ea_buf, lmm,
2979                        lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
2980                                        magic));
2981         rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV,
2982                           LU_XATTR_REPLACE, handle);
2983
2984         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2985
2986 unlock:
2987         dt_write_unlock(env, parent);
2988
2989 stop:
2990         if (handle != NULL)
2991                 dt_trans_stop(env, dev, handle);
2992
2993 log:
2994         lfsck_ibits_unlock(&lh, LCK_EX);
2995         if (child != NULL)
2996                 lfsck_object_put(env, child);
2997
2998         if (rc != 0)
2999                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3000                        "multiple references for: parent "DFID", OST-index %u, "
3001                        "stripe-index %u, owner %u/%u: rc = %d\n",
3002                        lfsck_lfsck2name(lfsck), PFID(pfid), llr->llr_ost_idx,
3003                        llr->llr_lov_idx, la->la_uid, la->la_gid, rc);
3004
3005         return rc;
3006 }
3007
3008 /* If the MDT-object and the OST-object have different owner information,
3009  * then trust the MDT-object, because the normal chown/chgrp handle order
3010  * is from MDT to OST, and it is possible that some chown/chgrp operation
3011  * is partly done. */
3012 static int lfsck_layout_repair_owner(const struct lu_env *env,
3013                                      struct lfsck_component *com,
3014                                      struct dt_object *parent,
3015                                      struct lfsck_layout_req *llr,
3016                                      struct lu_attr *pla,
3017                                      const struct lu_attr *cla)
3018 {
3019         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3020         struct lu_attr                  *tla    = &info->lti_la2;
3021         struct dt_object                *child  = llr->llr_child;
3022         struct dt_device                *dev    = lfsck_obj2dev(child);
3023         struct thandle                  *handle;
3024         int                              rc;
3025         ENTRY;
3026
3027         tla->la_uid = pla->la_uid;
3028         tla->la_gid = pla->la_gid;
3029         tla->la_valid = LA_UID | LA_GID;
3030         handle = dt_trans_create(env, dev);
3031         if (IS_ERR(handle))
3032                 GOTO(log, rc = PTR_ERR(handle));
3033
3034         rc = dt_declare_attr_set(env, child, tla, handle);
3035         if (rc != 0)
3036                 GOTO(stop, rc);
3037
3038         rc = dt_trans_start_local(env, dev, handle);
3039         if (rc != 0)
3040                 GOTO(stop, rc);
3041
3042         /* Use the dt_object lock to serialize with destroy and attr_set. */
3043         dt_read_lock(env, parent, 0);
3044         if (unlikely(lfsck_is_dead_obj(parent)))
3045                 GOTO(unlock, rc = 1);
3046
3047         /* Get the latest parent's owner. */
3048         rc = dt_attr_get(env, parent, pla);
3049         if (rc != 0)
3050                 GOTO(unlock, rc);
3051
3052         /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
3053         if (unlikely(tla->la_uid != pla->la_uid ||
3054                      tla->la_gid != pla->la_gid))
3055                 rc = 1;
3056         else
3057                 rc = dt_attr_set(env, child, tla, handle);
3058
3059         GOTO(unlock, rc);
3060
3061 unlock:
3062         dt_read_unlock(env, parent);
3063
3064 stop:
3065         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3066
3067 log:
3068         if (rc != 0)
3069                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3070                        "inconsistent file owner for: parent "DFID", child "DFID
3071                        ", OST-index %u, stripe-index %u, old owner %u/%u, "
3072                        "new owner %u/%u: rc = %d\n",
3073                        lfsck_lfsck2name(com->lc_lfsck),
3074                        PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
3075                        llr->llr_ost_idx, llr->llr_lov_idx,
3076                        cla->la_uid, cla->la_gid, tla->la_uid, tla->la_gid, rc);
3077
3078         return rc;
3079 }
3080
3081 /* Check whether the OST-object correctly back points to the
3082  * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
3083 static int lfsck_layout_check_parent(const struct lu_env *env,
3084                                      struct lfsck_component *com,
3085                                      struct lfsck_assistant_object *lso,
3086                                      const struct lu_fid *pfid,
3087                                      const struct lu_fid *cfid,
3088                                      const struct lu_attr *cla,
3089                                      struct lfsck_layout_req *llr,
3090                                      struct lu_buf *lov_ea, __u32 idx)
3091 {
3092         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3093         struct lu_buf                   *buf    = &info->lti_big_buf;
3094         struct dt_object                *tobj;
3095         struct lov_mds_md_v1            *lmm;
3096         struct lov_ost_data_v1          *objs;
3097         struct lustre_handle             lh     = { 0 };
3098         int                              rc;
3099         int                              i;
3100         __u32                            magic;
3101         __u16                            count;
3102         ENTRY;
3103
3104         if (unlikely(!fid_is_sane(pfid)))
3105                 RETURN(LLIT_UNMATCHED_PAIR);
3106
3107         if (lu_fid_eq(pfid, &lso->lso_fid)) {
3108                 if (likely(llr->llr_lov_idx == idx))
3109                         RETURN(0);
3110
3111                 RETURN(LLIT_UNMATCHED_PAIR);
3112         }
3113
3114         tobj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3115         if (IS_ERR(tobj))
3116                 RETURN(PTR_ERR(tobj));
3117
3118         if (dt_object_exists(tobj) == 0 || lfsck_is_dead_obj(tobj) ||
3119             !S_ISREG(lfsck_object_type(tobj)))
3120                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3121
3122         /* Load the tobj's layout EA, in spite of it is a local MDT-object or
3123          * remote one on another MDT. Then check whether the given OST-object
3124          * is in such layout. If yes, it is multiple referenced, otherwise it
3125          * is unmatched referenced case. */
3126         rc = lfsck_layout_get_lovea(env, tobj, buf);
3127         if (rc == 0 || rc == -ENOENT)
3128                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3129
3130         if (rc < 0)
3131                 GOTO(out, rc);
3132
3133         lmm = buf->lb_buf;
3134         magic = le32_to_cpu(lmm->lmm_magic);
3135         if (magic == LOV_MAGIC_V1) {
3136                 objs = &lmm->lmm_objects[0];
3137         } else {
3138                 LASSERT(magic == LOV_MAGIC_V3);
3139                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3140         }
3141
3142         count = le16_to_cpu(lmm->lmm_stripe_count);
3143         for (i = 0; i < count; i++, objs++) {
3144                 struct lu_fid           *tfid   = &info->lti_fid2;
3145                 struct ost_id           *oi     = &info->lti_oi;
3146                 __u32                    idx2;
3147
3148                 if (lovea_slot_is_dummy(objs))
3149                         continue;
3150
3151                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3152                 idx2 = le32_to_cpu(objs->l_ost_idx);
3153                 rc = ostid_to_fid(tfid, oi, idx2);
3154                 if (rc != 0) {
3155                         CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
3156                                "invalid layout EA at the slot %d, index %u\n",
3157                                lfsck_lfsck2name(com->lc_lfsck),
3158                                PFID(pfid), i, idx2);
3159
3160                         GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3161                 }
3162
3163                 if (lu_fid_eq(cfid, tfid)) {
3164                         rc = lfsck_ibits_lock(env, com->lc_lfsck, tobj, &lh,
3165                                               MDS_INODELOCK_UPDATE |
3166                                               MDS_INODELOCK_LAYOUT |
3167                                               MDS_INODELOCK_XATTR,
3168                                               LCK_EX);
3169                         if (rc != 0)
3170                                 GOTO(out, rc);
3171
3172                         dt_read_lock(env, tobj, 0);
3173
3174                         /* For local MDT-object, re-check existence
3175                          * after taken the lock. */
3176                         if (!dt_object_remote(tobj)) {
3177                                 if (dt_object_exists(tobj) == 0 ||
3178                                     lfsck_is_dead_obj(tobj)) {
3179                                         rc = LLIT_UNMATCHED_PAIR;
3180                                 } else {
3181                                         *lov_ea = *buf;
3182                                         rc = LLIT_MULTIPLE_REFERENCED;
3183                                 }
3184
3185                                 GOTO(unlock, rc);
3186                         }
3187
3188                         /* For migration case, the new MDT-object and old
3189                          * MDT-object may reference the same OST-object at
3190                          * some migration internal time.
3191                          *
3192                          * For remote MDT-object, the local MDT may not know
3193                          * whether it has been removed or not.  Try checking
3194                          * for a non-existent xattr to check if this object
3195                          * has been been removed or not. */
3196                         rc = dt_xattr_get(env, tobj, &LU_BUF_NULL,
3197                                           XATTR_NAME_DUMMY);
3198                         if (unlikely(rc == -ENOENT || rc >= 0)) {
3199                                 rc = LLIT_UNMATCHED_PAIR;
3200                         } else if (rc == -ENODATA) {
3201                                 *lov_ea = *buf;
3202                                 rc = LLIT_MULTIPLE_REFERENCED;
3203                         }
3204
3205                         GOTO(unlock, rc);
3206                 }
3207         }
3208
3209         GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3210
3211 unlock:
3212         if (lustre_handle_is_used(&lh)) {
3213                 dt_read_unlock(env, tobj);
3214                 lfsck_ibits_unlock(&lh, LCK_EX);
3215         }
3216
3217 out:
3218         lfsck_object_put(env, tobj);
3219
3220         return rc;
3221 }
3222
3223 static int lfsck_layout_assistant_handler_p1(const struct lu_env *env,
3224                                              struct lfsck_component *com,
3225                                              struct lfsck_assistant_req *lar)
3226 {
3227         struct lfsck_layout_req              *llr    =
3228                         container_of0(lar, struct lfsck_layout_req, llr_lar);
3229         struct lfsck_assistant_object        *lso    = lar->lar_parent;
3230         struct lfsck_layout                  *lo     = com->lc_file_ram;
3231         struct lfsck_thread_info             *info   = lfsck_env_info(env);
3232         struct filter_fid_old                *pea    = &info->lti_old_pfid;
3233         struct lu_fid                        *pfid   = &info->lti_fid;
3234         struct lu_buf                         buf    = { NULL };
3235         struct dt_object                     *parent = NULL;
3236         struct dt_object                     *child  = llr->llr_child;
3237         struct lu_attr                       *pla    = &lso->lso_attr;
3238         struct lu_attr                       *cla    = &info->lti_la;
3239         struct lfsck_instance                *lfsck  = com->lc_lfsck;
3240         struct lfsck_bookmark                *bk     = &lfsck->li_bookmark_ram;
3241         enum lfsck_layout_inconsistency_type  type   = LLIT_NONE;
3242         __u32                                 idx    = 0;
3243         int                                   rc;
3244         ENTRY;
3245
3246         if (lso->lso_dead)
3247                 RETURN(0);
3248
3249         rc = dt_attr_get(env, child, cla);
3250         if (rc == -ENOENT) {
3251                 parent = lfsck_assistant_object_load(env, lfsck, lso);
3252                 if (IS_ERR(parent)) {
3253                         rc = PTR_ERR(parent);
3254
3255                         RETURN(rc == -ENOENT ? 0 : rc);
3256                 }
3257
3258                 type = LLIT_DANGLING;
3259                 goto repair;
3260         }
3261
3262         if (rc != 0)
3263                 GOTO(out, rc);
3264
3265         lfsck_buf_init(&buf, pea, sizeof(struct filter_fid_old));
3266         rc = dt_xattr_get(env, child, &buf, XATTR_NAME_FID);
3267         if (unlikely(rc > 0 && rc != sizeof(struct filter_fid_old) &&
3268                      rc != sizeof(struct filter_fid))) {
3269                 type = LLIT_UNMATCHED_PAIR;
3270                 goto repair;
3271         }
3272
3273         if (rc < 0 && rc != -ENODATA)
3274                 GOTO(out, rc);
3275
3276         if (rc == 0 || rc == -ENODATA)
3277                 GOTO(check_owner, rc = 0);
3278
3279         fid_le_to_cpu(pfid, &pea->ff_parent);
3280         /* Currently, the filter_fid::ff_parent::f_ver is not the
3281          * real parent MDT-object's FID::f_ver, instead it is the
3282          * OST-object index in its parent MDT-object's layout EA. */
3283         idx = pfid->f_stripe_idx;
3284         pfid->f_ver = 0;
3285         rc = lfsck_layout_check_parent(env, com, lso, pfid,
3286                                        lu_object_fid(&child->do_lu),
3287                                        cla, llr, &buf, idx);
3288         if (rc > 0) {
3289                 type = rc;
3290                 goto repair;
3291         }
3292
3293         if (rc < 0)
3294                 GOTO(out, rc);
3295
3296 check_owner:
3297         /* Someone may has changed the owner after the parent attr pre-loaded.
3298          * It can be handled later inside the lfsck_layout_repair_owner(). */
3299         if (unlikely(cla->la_uid != pla->la_uid ||
3300                      cla->la_gid != pla->la_gid)) {
3301                 type = LLIT_INCONSISTENT_OWNER;
3302                 goto repair;
3303         }
3304
3305 repair:
3306         if (type == LLIT_NONE)
3307                 GOTO(out, rc = 0);
3308
3309         if (bk->lb_param & LPF_DRYRUN)
3310                 GOTO(out, rc = 1);
3311
3312         if (parent == NULL) {
3313                 parent = lfsck_assistant_object_load(env, lfsck, lso);
3314                 if (IS_ERR(parent)) {
3315                         rc = PTR_ERR(parent);
3316
3317                         if (rc == -ENOENT)
3318                                 RETURN(0);
3319
3320                         GOTO(out, rc);
3321                 }
3322         }
3323
3324         switch (type) {
3325         case LLIT_DANGLING:
3326                 rc = lfsck_layout_repair_dangling(env, com, parent, llr, pla);
3327                 break;
3328         case LLIT_UNMATCHED_PAIR:
3329                 rc = lfsck_layout_repair_unmatched_pair(env, com, parent,
3330                                                         llr, pla);
3331                 break;
3332         case LLIT_MULTIPLE_REFERENCED:
3333                 rc = lfsck_layout_repair_multiple_references(env, com, parent,
3334                                                              llr, pla, &buf);
3335                 break;
3336         case LLIT_INCONSISTENT_OWNER:
3337                 rc = lfsck_layout_repair_owner(env, com, parent, llr, pla, cla);
3338                 break;
3339         default:
3340                 rc = 0;
3341                 break;
3342         }
3343
3344         GOTO(out, rc);
3345
3346 out:
3347         down_write(&com->lc_sem);
3348         if (rc < 0) {
3349                 struct lfsck_assistant_data *lad = com->lc_data;
3350
3351                 if (unlikely(lad->lad_exit)) {
3352                         rc = 0;
3353                 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
3354                            rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
3355                            rc == -EHOSTUNREACH) {
3356                         /* If cannot touch the target server,
3357                          * mark the LFSCK as INCOMPLETE. */
3358                         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
3359                                "talk with OST %x: rc = %d\n",
3360                                lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
3361                         lfsck_lad_set_bitmap(env, com, llr->llr_ost_idx);
3362                         lo->ll_objs_skipped++;
3363                         rc = 0;
3364                 } else {
3365                         lfsck_layout_record_failure(env, lfsck, lo);
3366                 }
3367         } else if (rc > 0) {
3368                 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
3369                          "unknown type = %d\n", type);
3370
3371                 lo->ll_objs_repaired[type - 1]++;
3372                 if (bk->lb_param & LPF_DRYRUN &&
3373                     unlikely(lo->ll_pos_first_inconsistent == 0))
3374                         lo->ll_pos_first_inconsistent =
3375                         lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
3376                                                         lfsck->li_di_oit);
3377         }
3378         up_write(&com->lc_sem);
3379
3380         if (parent != NULL && !IS_ERR(parent))
3381                 lfsck_object_put(env, parent);
3382
3383         return rc;
3384 }
3385
3386 static int lfsck_layout_assistant_handler_p2(const struct lu_env *env,
3387                                              struct lfsck_component *com)
3388 {
3389         struct lfsck_assistant_data     *lad    = com->lc_data;
3390         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3391         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
3392         struct lfsck_tgt_descs          *ltds   = &lfsck->li_ost_descs;
3393         struct lfsck_tgt_desc           *ltd;
3394         int                              rc     = 0;
3395         ENTRY;
3396
3397         CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan start\n",
3398                lfsck_lfsck2name(lfsck));
3399
3400         spin_lock(&ltds->ltd_lock);
3401         while (!list_empty(&lad->lad_ost_phase2_list)) {
3402                 ltd = list_entry(lad->lad_ost_phase2_list.next,
3403                                  struct lfsck_tgt_desc,
3404                                  ltd_layout_phase_list);
3405                 list_del_init(&ltd->ltd_layout_phase_list);
3406                 if (bk->lb_param & LPF_ALL_TGT) {
3407                         spin_unlock(&ltds->ltd_lock);
3408                         rc = lfsck_layout_scan_orphan(env, com, ltd);
3409                         if (rc != 0 && bk->lb_param & LPF_FAILOUT)
3410                                 RETURN(rc);
3411
3412                         if (unlikely(lad->lad_exit ||
3413                                      !thread_is_running(&lfsck->li_thread)))
3414                                 RETURN(0);
3415                         spin_lock(&ltds->ltd_lock);
3416                 }
3417         }
3418
3419         if (list_empty(&lad->lad_ost_phase1_list))
3420                 rc = 1;
3421         else
3422                 rc = 0;
3423         spin_unlock(&ltds->ltd_lock);
3424
3425         CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan stop: rc = %d\n",
3426                lfsck_lfsck2name(lfsck), rc);
3427
3428         RETURN(rc);
3429 }
3430
3431 static int
3432 lfsck_layout_slave_async_interpret(const struct lu_env *env,
3433                                    struct ptlrpc_request *req,
3434                                    void *args, int rc)
3435 {
3436         struct lfsck_layout_slave_async_args *llsaa = args;
3437         struct obd_export                    *exp   = llsaa->llsaa_exp;
3438         struct lfsck_component               *com   = llsaa->llsaa_com;
3439         struct lfsck_layout_slave_target     *llst  = llsaa->llsaa_llst;
3440         struct lfsck_layout_slave_data       *llsd  = com->lc_data;
3441         struct lfsck_reply                   *lr    = NULL;
3442         bool                                  done  = false;
3443
3444         if (rc != 0) {
3445                 /* It is quite probably caused by target crash,
3446                  * to make the LFSCK can go ahead, assume that
3447                  * the target finished the LFSCK prcoessing. */
3448                 done = true;
3449         } else {
3450                 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
3451                 if (lr->lr_status != LS_SCANNING_PHASE1 &&
3452                     lr->lr_status != LS_SCANNING_PHASE2)
3453                         done = true;
3454         }
3455
3456         if (done) {
3457                 CDEBUG(D_LFSCK, "%s: layout LFSCK slave gets the MDT %x "
3458                        "status %d\n", lfsck_lfsck2name(com->lc_lfsck),
3459                        llst->llst_index, lr != NULL ? lr->lr_status : rc);
3460
3461                 lfsck_layout_llst_del(llsd, llst);
3462         }
3463
3464         lfsck_layout_llst_put(llst);
3465         lfsck_component_put(env, com);
3466         class_export_put(exp);
3467
3468         return 0;
3469 }
3470
3471 static int lfsck_layout_async_query(const struct lu_env *env,
3472                                     struct lfsck_component *com,
3473                                     struct obd_export *exp,
3474                                     struct lfsck_layout_slave_target *llst,
3475                                     struct lfsck_request *lr,
3476                                     struct ptlrpc_request_set *set)
3477 {
3478         struct lfsck_layout_slave_async_args *llsaa;
3479         struct ptlrpc_request                *req;
3480         struct lfsck_request                 *tmp;
3481         int                                   rc;
3482         ENTRY;
3483
3484         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
3485         if (req == NULL)
3486                 RETURN(-ENOMEM);
3487
3488         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
3489         if (rc != 0) {
3490                 ptlrpc_request_free(req);
3491                 RETURN(rc);
3492         }
3493
3494         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3495         *tmp = *lr;
3496         ptlrpc_request_set_replen(req);
3497
3498         llsaa = ptlrpc_req_async_args(req);
3499         llsaa->llsaa_exp = exp;
3500         llsaa->llsaa_com = lfsck_component_get(com);
3501         llsaa->llsaa_llst = llst;
3502         req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
3503         ptlrpc_set_add_req(set, req);
3504
3505         RETURN(0);
3506 }
3507
3508 static int lfsck_layout_async_notify(const struct lu_env *env,
3509                                      struct obd_export *exp,
3510                                      struct lfsck_request *lr,
3511                                      struct ptlrpc_request_set *set)
3512 {
3513         struct ptlrpc_request   *req;
3514         struct lfsck_request    *tmp;
3515         int                      rc;
3516         ENTRY;
3517
3518         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
3519         if (req == NULL)
3520                 RETURN(-ENOMEM);
3521
3522         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
3523         if (rc != 0) {
3524                 ptlrpc_request_free(req);
3525                 RETURN(rc);
3526         }
3527
3528         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3529         *tmp = *lr;
3530         ptlrpc_request_set_replen(req);
3531         ptlrpc_set_add_req(set, req);
3532
3533         RETURN(0);
3534 }
3535
3536 static int
3537 lfsck_layout_slave_query_master(const struct lu_env *env,
3538                                 struct lfsck_component *com)
3539 {
3540         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
3541         struct lfsck_instance            *lfsck = com->lc_lfsck;
3542         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3543         struct lfsck_layout_slave_target *llst;
3544         struct obd_export                *exp;
3545         struct ptlrpc_request_set        *set;
3546         int                               rc    = 0;
3547         int                               rc1   = 0;
3548         ENTRY;
3549
3550         set = ptlrpc_prep_set();
3551         if (set == NULL)
3552                 GOTO(log, rc = -ENOMEM);
3553
3554         memset(lr, 0, sizeof(*lr));
3555         lr->lr_event = LE_QUERY;
3556         lr->lr_active = LFSCK_TYPE_LAYOUT;
3557
3558         llsd->llsd_touch_gen++;
3559         spin_lock(&llsd->llsd_lock);
3560         while (!list_empty(&llsd->llsd_master_list)) {
3561                 llst = list_entry(llsd->llsd_master_list.next,
3562                                   struct lfsck_layout_slave_target,
3563                                   llst_list);
3564                 if (llst->llst_gen == llsd->llsd_touch_gen)
3565                         break;
3566
3567                 llst->llst_gen = llsd->llsd_touch_gen;
3568                 list_move_tail(&llst->llst_list,
3569                                &llsd->llsd_master_list);
3570                 atomic_inc(&llst->llst_ref);
3571                 spin_unlock(&llsd->llsd_lock);
3572
3573                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
3574                                                llst->llst_index);
3575                 if (exp == NULL) {
3576                         lfsck_layout_llst_del(llsd, llst);
3577                         lfsck_layout_llst_put(llst);
3578                         spin_lock(&llsd->llsd_lock);
3579                         continue;
3580                 }
3581
3582                 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
3583                 if (rc != 0) {
3584                         CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
3585                                "query %s for layout: rc = %d\n",
3586                                lfsck_lfsck2name(lfsck),
3587                                exp->exp_obd->obd_name, rc);
3588
3589                         rc1 = rc;
3590                         lfsck_layout_llst_put(llst);
3591                         class_export_put(exp);
3592                 }
3593                 spin_lock(&llsd->llsd_lock);
3594         }
3595         spin_unlock(&llsd->llsd_lock);
3596
3597         rc = ptlrpc_set_wait(set);
3598         ptlrpc_set_destroy(set);
3599
3600         GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
3601
3602 log:
3603         CDEBUG(D_LFSCK, "%s: layout LFSCK slave queries master: rc = %d\n",
3604                lfsck_lfsck2name(com->lc_lfsck), rc);
3605
3606         return rc;
3607 }
3608
3609 static void
3610 lfsck_layout_slave_notify_master(const struct lu_env *env,
3611                                  struct lfsck_component *com,
3612                                  enum lfsck_events event, int result)
3613 {
3614         struct lfsck_layout              *lo    = com->lc_file_ram;
3615         struct lfsck_instance            *lfsck = com->lc_lfsck;
3616         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3617         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
3618         struct lfsck_layout_slave_target *llst;
3619         struct obd_export                *exp;
3620         struct ptlrpc_request_set        *set;
3621         int                               rc;
3622         ENTRY;
3623
3624         CDEBUG(D_LFSCK, "%s: layout LFSCK slave notifies master\n",
3625                lfsck_lfsck2name(com->lc_lfsck));
3626
3627         set = ptlrpc_prep_set();
3628         if (set == NULL)
3629                 RETURN_EXIT;
3630
3631         memset(lr, 0, sizeof(*lr));
3632         lr->lr_event = event;
3633         lr->lr_flags = LEF_FROM_OST;
3634         lr->lr_status = result;
3635         lr->lr_index = lfsck_dev_idx(lfsck);
3636         lr->lr_active = LFSCK_TYPE_LAYOUT;
3637         lr->lr_flags2 = lo->ll_flags;
3638         llsd->llsd_touch_gen++;
3639         spin_lock(&llsd->llsd_lock);
3640         while (!list_empty(&llsd->llsd_master_list)) {
3641                 llst = list_entry(llsd->llsd_master_list.next,
3642                                   struct lfsck_layout_slave_target,
3643                                   llst_list);
3644                 if (llst->llst_gen == llsd->llsd_touch_gen)
3645                         break;
3646
3647                 llst->llst_gen = llsd->llsd_touch_gen;
3648                 list_move_tail(&llst->llst_list,
3649                                &llsd->llsd_master_list);
3650                 atomic_inc(&llst->llst_ref);
3651                 spin_unlock(&llsd->llsd_lock);
3652
3653                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
3654                                                llst->llst_index);
3655                 if (exp == NULL) {
3656                         lfsck_layout_llst_del(llsd, llst);
3657                         lfsck_layout_llst_put(llst);
3658                         spin_lock(&llsd->llsd_lock);
3659                         continue;
3660                 }
3661
3662                 rc = lfsck_layout_async_notify(env, exp, lr, set);
3663                 if (rc != 0)
3664                         CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
3665                                "notify %s for layout: rc = %d\n",
3666                                lfsck_lfsck2name(lfsck),
3667                                exp->exp_obd->obd_name, rc);
3668
3669                 lfsck_layout_llst_put(llst);
3670                 class_export_put(exp);
3671                 spin_lock(&llsd->llsd_lock);
3672         }
3673         spin_unlock(&llsd->llsd_lock);
3674
3675         ptlrpc_set_wait(set);
3676         ptlrpc_set_destroy(set);
3677
3678         RETURN_EXIT;
3679 }
3680
3681 /*
3682  * \ret -ENODATA: unrecognized stripe
3683  * \ret = 0     : recognized stripe
3684  * \ret < 0     : other failures
3685  */
3686 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
3687                                            struct lfsck_component *com,
3688                                            struct lu_fid *cfid,
3689                                            struct lu_fid *pfid)
3690 {
3691         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3692         struct lu_buf                   *buf    = &info->lti_big_buf;
3693         struct ost_id                   *oi     = &info->lti_oi;
3694         struct dt_object                *obj;
3695         struct lov_mds_md_v1            *lmm;
3696         struct lov_ost_data_v1          *objs;
3697         __u32                            idx    = pfid->f_stripe_idx;
3698         __u32                            magic;
3699         int                              rc     = 0;
3700         int                              i;
3701         __u16                            count;
3702         ENTRY;
3703
3704         pfid->f_ver = 0;
3705         obj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3706         if (IS_ERR(obj))
3707                 RETURN(PTR_ERR(obj));
3708
3709         dt_read_lock(env, obj, 0);
3710         if (unlikely(dt_object_exists(obj) == 0 ||
3711                      lfsck_is_dead_obj(obj)))
3712                 GOTO(unlock, rc = -ENOENT);
3713
3714         if (!S_ISREG(lfsck_object_type(obj)))
3715                 GOTO(unlock, rc = -ENODATA);
3716
3717         rc = lfsck_layout_get_lovea(env, obj, buf);
3718         if (rc < 0)
3719                 GOTO(unlock, rc);
3720
3721         if (rc == 0)
3722                 GOTO(unlock, rc = -ENODATA);
3723
3724         lmm = buf->lb_buf;
3725         rc = lfsck_layout_verify_header(lmm);
3726         if (rc != 0)
3727                 GOTO(unlock, rc);
3728
3729         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
3730          * been verified in lfsck_layout_verify_header() already. If some
3731          * new magic introduced in the future, then layout LFSCK needs to
3732          * be updated also. */
3733         magic = le32_to_cpu(lmm->lmm_magic);
3734         if (magic == LOV_MAGIC_V1) {
3735                 objs = &lmm->lmm_objects[0];
3736         } else {
3737                 LASSERT(magic == LOV_MAGIC_V3);
3738                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3739         }
3740
3741         fid_to_ostid(cfid, oi);
3742         count = le16_to_cpu(lmm->lmm_stripe_count);
3743         for (i = 0; i < count; i++, objs++) {
3744                 struct ost_id oi2;
3745
3746                 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
3747                 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
3748                         GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
3749         }
3750
3751         GOTO(unlock, rc = -ENODATA);
3752
3753 unlock:
3754         dt_read_unlock(env, obj);
3755         lfsck_object_put(env, obj);
3756
3757         return rc;
3758 }
3759
3760 /*
3761  * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
3762  * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
3763  * layout EA from MDT to OST. On one hand, the OST no need to understand
3764  * the layout EA structure; on the other hand, it may cause trouble when
3765  * transfer large layout EA from MDT to OST via normal OUT RPC.
3766  *
3767  * \ret > 0: unrecognized stripe
3768  * \ret = 0: recognized stripe
3769  * \ret < 0: other failures
3770  */
3771 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
3772                                           struct lfsck_component *com,
3773                                           struct lu_fid *cfid,
3774                                           struct lu_fid *pfid)
3775 {
3776         struct lfsck_instance    *lfsck  = com->lc_lfsck;
3777         struct obd_device        *obd    = lfsck->li_obd;
3778         struct seq_server_site   *ss     = lfsck_dev_site(lfsck);
3779         struct obd_export        *exp    = NULL;
3780         struct ptlrpc_request    *req    = NULL;
3781         struct lfsck_request     *lr;
3782         struct lu_seq_range      *range  = &lfsck_env_info(env)->lti_range;
3783         int                       rc     = 0;
3784         ENTRY;
3785
3786         if (unlikely(fid_is_idif(pfid)))
3787                 RETURN(1);
3788
3789         fld_range_set_any(range);
3790         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), range);
3791         if (rc != 0)
3792                 RETURN(rc == -ENOENT ? 1 : rc);
3793
3794         if (unlikely(!fld_range_is_mdt(range)))
3795                 RETURN(1);
3796
3797         exp = lustre_find_lwp_by_index(obd->obd_name, range->lsr_index);
3798         if (unlikely(exp == NULL))
3799                 RETURN(1);
3800
3801         if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
3802                 GOTO(out, rc = -EOPNOTSUPP);
3803
3804         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
3805         if (req == NULL)
3806                 GOTO(out, rc = -ENOMEM);
3807
3808         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
3809         if (rc != 0) {
3810                 ptlrpc_request_free(req);
3811
3812                 GOTO(out, rc);
3813         }
3814
3815         lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3816         memset(lr, 0, sizeof(*lr));
3817         lr->lr_event = LE_PAIRS_VERIFY;
3818         lr->lr_active = LFSCK_TYPE_LAYOUT;
3819         lr->lr_fid = *cfid; /* OST-object itself FID. */
3820         lr->lr_fid2 = *pfid; /* The claimed parent FID. */
3821
3822         ptlrpc_request_set_replen(req);
3823         rc = ptlrpc_queue_wait(req);
3824         ptlrpc_req_finished(req);
3825
3826         if (rc == -ENOENT || rc == -ENODATA)
3827                 rc = 1;
3828
3829         GOTO(out, rc);
3830
3831 out:
3832         if (exp != NULL)
3833                 class_export_put(exp);
3834
3835         return rc;
3836 }
3837
3838 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
3839                                           struct lfsck_component *com,
3840                                           struct lfsck_request *lr)
3841 {
3842         struct dt_object        *obj;
3843         int                      rc     = 0;
3844         ENTRY;
3845
3846         obj = lfsck_object_find_bottom(env, com->lc_lfsck, &lr->lr_fid);
3847         if (IS_ERR(obj))
3848                 GOTO(log, rc = PTR_ERR(obj));
3849
3850         dt_write_lock(env, obj, 0);
3851         if (unlikely(dt_object_exists(obj) == 0 ||
3852                      lfsck_is_dead_obj(obj)))
3853                 GOTO(unlock, rc = 0);
3854
3855         rc = __lfsck_layout_update_pfid(env, obj, &lr->lr_fid2,
3856                                         lr->lr_fid2.f_ver);
3857
3858         GOTO(unlock, rc);
3859
3860 unlock:
3861         dt_write_unlock(env, obj);
3862         lfsck_object_put(env, obj);
3863
3864 log:
3865         CDEBUG(D_LFSCK, "%s: layout LFSCK slave repaired pfid for "DFID
3866                ", parent "DFID": rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
3867                PFID(&lr->lr_fid), PFID(&lr->lr_fid2), rc);
3868
3869         return rc;
3870 }
3871
3872 /* layout APIs */
3873
3874 static void lfsck_layout_slave_quit(const struct lu_env *env,
3875                                     struct lfsck_component *com);
3876
3877 static int lfsck_layout_reset(const struct lu_env *env,
3878                               struct lfsck_component *com, bool init)
3879 {
3880         struct lfsck_layout     *lo    = com->lc_file_ram;
3881         int                      rc;
3882
3883         down_write(&com->lc_sem);
3884         if (init) {
3885                 memset(lo, 0, com->lc_file_size);
3886         } else {
3887                 __u32 count = lo->ll_success_count;
3888                 __u64 last_time = lo->ll_time_last_complete;
3889
3890                 memset(lo, 0, com->lc_file_size);
3891                 lo->ll_success_count = count;
3892                 lo->ll_time_last_complete = last_time;
3893         }
3894
3895         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
3896         lo->ll_status = LS_INIT;
3897
3898         if (com->lc_lfsck->li_master) {
3899                 struct lfsck_assistant_data *lad = com->lc_data;
3900
3901                 lad->lad_incomplete = 0;
3902                 CFS_RESET_BITMAP(lad->lad_bitmap);
3903         }
3904
3905         rc = lfsck_layout_store(env, com);
3906         up_write(&com->lc_sem);
3907
3908         CDEBUG(D_LFSCK, "%s: layout LFSCK reset: rc = %d\n",
3909                lfsck_lfsck2name(com->lc_lfsck), rc);
3910
3911         return rc;
3912 }
3913
3914 static void lfsck_layout_fail(const struct lu_env *env,
3915                               struct lfsck_component *com, bool new_checked)
3916 {
3917         struct lfsck_layout *lo = com->lc_file_ram;
3918
3919         down_write(&com->lc_sem);
3920         if (new_checked)
3921                 com->lc_new_checked++;
3922         lfsck_layout_record_failure(env, com->lc_lfsck, lo);
3923         up_write(&com->lc_sem);
3924 }
3925
3926 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
3927                                           struct lfsck_component *com, bool init)
3928 {
3929         struct lfsck_instance   *lfsck   = com->lc_lfsck;
3930         struct lfsck_layout     *lo      = com->lc_file_ram;
3931         int                      rc;
3932
3933         if (!init) {
3934                 rc = lfsck_checkpoint_generic(env, com);
3935                 if (rc != 0)
3936                         return rc > 0 ? 0 : rc;
3937         }
3938
3939         down_write(&com->lc_sem);
3940         if (init) {
3941                 lo->ll_pos_latest_start =
3942                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
3943         } else {
3944                 lo->ll_pos_last_checkpoint =
3945                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
3946                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3947                                 HALF_SEC - lfsck->li_time_last_checkpoint);
3948                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
3949                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
3950                 com->lc_new_checked = 0;
3951         }
3952
3953         rc = lfsck_layout_store(env, com);
3954         up_write(&com->lc_sem);
3955
3956         CDEBUG(D_LFSCK, "%s: layout LFSCK master checkpoint at the pos ["
3957                LPU64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
3958                lfsck->li_pos_current.lp_oit_cookie, rc);
3959
3960         return rc;
3961 }
3962
3963 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
3964                                          struct lfsck_component *com, bool init)
3965 {
3966         struct lfsck_instance   *lfsck = com->lc_lfsck;
3967         struct lfsck_layout     *lo    = com->lc_file_ram;
3968         int                      rc;
3969
3970         if (com->lc_new_checked == 0 && !init)
3971                 return 0;
3972
3973         down_write(&com->lc_sem);
3974         if (init) {
3975                 lo->ll_pos_latest_start =
3976                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
3977         } else {
3978                 lo->ll_pos_last_checkpoint =
3979                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
3980                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3981                                 HALF_SEC - lfsck->li_time_last_checkpoint);
3982                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
3983                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
3984                 com->lc_new_checked = 0;
3985         }
3986
3987         rc = lfsck_layout_store(env, com);
3988         up_write(&com->lc_sem);
3989
3990         CDEBUG(D_LFSCK, "%s: layout LFSCK slave checkpoint at the pos ["
3991                LPU64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
3992                lfsck->li_pos_current.lp_oit_cookie, rc);
3993
3994         return rc;
3995 }
3996
3997 static int lfsck_layout_prep(const struct lu_env *env,
3998                              struct lfsck_component *com,
3999                              struct lfsck_start *start)
4000 {
4001         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4002         struct lfsck_layout     *lo     = com->lc_file_ram;
4003         struct lfsck_position   *pos    = &com->lc_pos_start;
4004
4005         fid_zero(&pos->lp_dir_parent);
4006         pos->lp_dir_cookie = 0;
4007         if (lo->ll_status == LS_COMPLETED ||
4008             lo->ll_status == LS_PARTIAL ||
4009             /* To handle orphan, must scan from the beginning. */
4010             (start != NULL && start->ls_flags & LPF_OST_ORPHAN)) {
4011                 int rc;
4012
4013                 rc = lfsck_layout_reset(env, com, false);
4014                 if (rc == 0)
4015                         rc = lfsck_set_param(env, lfsck, start, true);
4016
4017                 if (rc != 0) {
4018                         CDEBUG(D_LFSCK, "%s: layout LFSCK prep failed: "
4019                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4020
4021                         return rc;
4022                 }
4023         }
4024
4025         down_write(&com->lc_sem);
4026         lo->ll_time_latest_start = cfs_time_current_sec();
4027         spin_lock(&lfsck->li_lock);
4028         if (lo->ll_flags & LF_SCANNED_ONCE) {
4029                 if (!lfsck->li_drop_dryrun ||
4030                     lo->ll_pos_first_inconsistent == 0) {
4031                         lo->ll_status = LS_SCANNING_PHASE2;
4032                         list_move_tail(&com->lc_link,
4033                                        &lfsck->li_list_double_scan);
4034                         pos->lp_oit_cookie = 0;
4035                 } else {
4036                         int i;
4037
4038                         lo->ll_status = LS_SCANNING_PHASE1;
4039                         lo->ll_run_time_phase1 = 0;
4040                         lo->ll_run_time_phase2 = 0;
4041                         lo->ll_objs_checked_phase1 = 0;
4042                         lo->ll_objs_checked_phase2 = 0;
4043                         lo->ll_objs_failed_phase1 = 0;
4044                         lo->ll_objs_failed_phase2 = 0;
4045                         for (i = 0; i < LLIT_MAX; i++)
4046                                 lo->ll_objs_repaired[i] = 0;
4047
4048                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4049                         fid_zero(&com->lc_fid_latest_scanned_phase2);
4050                 }
4051         } else {
4052                 lo->ll_status = LS_SCANNING_PHASE1;
4053                 if (!lfsck->li_drop_dryrun ||
4054                     lo->ll_pos_first_inconsistent == 0)
4055                         pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
4056                 else
4057                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4058         }
4059         spin_unlock(&lfsck->li_lock);
4060         up_write(&com->lc_sem);
4061
4062         return 0;
4063 }
4064
4065 static int lfsck_layout_slave_prep(const struct lu_env *env,
4066                                    struct lfsck_component *com,
4067                                    struct lfsck_start_param *lsp)
4068 {
4069         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4070         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4071         struct lfsck_layout             *lo     = com->lc_file_ram;
4072         struct lfsck_start              *start  = lsp->lsp_start;
4073         int                              rc;
4074
4075         rc = lfsck_layout_prep(env, com, start);
4076         if (rc != 0)
4077                 return rc;
4078
4079         if (lo->ll_flags & LF_CRASHED_LASTID &&
4080             list_empty(&llsd->llsd_master_list)) {
4081                 LASSERT(lfsck->li_out_notify != NULL);
4082
4083                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4084                                      LE_LASTID_REBUILDING);
4085         }
4086
4087         if (!lsp->lsp_index_valid)
4088                 return 0;
4089
4090         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
4091         if (rc == 0 && start != NULL && start->ls_flags & LPF_OST_ORPHAN) {
4092                 LASSERT(!llsd->llsd_rbtree_valid);
4093
4094                 write_lock(&llsd->llsd_rb_lock);
4095                 rc = lfsck_rbtree_setup(env, com);
4096                 write_unlock(&llsd->llsd_rb_lock);
4097         }
4098
4099         CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
4100                LPU64"]\n", lfsck_lfsck2name(lfsck),
4101                com->lc_pos_start.lp_oit_cookie);
4102
4103         return rc;
4104 }
4105
4106 static int lfsck_layout_master_prep(const struct lu_env *env,
4107                                     struct lfsck_component *com,
4108                                     struct lfsck_start_param *lsp)
4109 {
4110         int rc;
4111         ENTRY;
4112
4113         rc = lfsck_layout_load_bitmap(env, com);
4114         if (rc != 0) {
4115                 rc = lfsck_layout_reset(env, com, false);
4116                 if (rc == 0)
4117                         rc = lfsck_set_param(env, com->lc_lfsck,
4118                                              lsp->lsp_start, true);
4119
4120                 if (rc != 0)
4121                         GOTO(log, rc);
4122         }
4123
4124         rc = lfsck_layout_prep(env, com, lsp->lsp_start);
4125         if (rc != 0)
4126                 RETURN(rc);
4127
4128         rc = lfsck_start_assistant(env, com, lsp);
4129
4130         GOTO(log, rc);
4131
4132 log:
4133         CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
4134                LPU64"]\n", lfsck_lfsck2name(com->lc_lfsck),
4135                com->lc_pos_start.lp_oit_cookie);
4136
4137         return 0;
4138 }
4139
4140 /* Pre-fetch the attribute for each stripe in the given layout EA. */
4141 static int lfsck_layout_scan_stripes(const struct lu_env *env,
4142                                      struct lfsck_component *com,
4143                                      struct dt_object *parent,
4144                                      struct lov_mds_md_v1 *lmm)
4145 {
4146         struct lfsck_thread_info        *info    = lfsck_env_info(env);
4147         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4148         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
4149         struct lfsck_layout             *lo      = com->lc_file_ram;
4150         struct lfsck_assistant_data     *lad     = com->lc_data;
4151         struct lfsck_assistant_object   *lso     = NULL;
4152         struct lov_ost_data_v1          *objs;
4153         struct lfsck_tgt_descs          *ltds    = &lfsck->li_ost_descs;
4154         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4155         struct ptlrpc_thread            *athread = &lad->lad_thread;
4156         struct l_wait_info               lwi     = { 0 };
4157         struct lu_buf                    buf;
4158         int                              rc      = 0;
4159         int                              i;
4160         __u32                            magic;
4161         __u16                            count;
4162         ENTRY;
4163
4164         lfsck_buf_init(&buf, &info->lti_old_pfid,
4165                        sizeof(struct filter_fid_old));
4166         count = le16_to_cpu(lmm->lmm_stripe_count);
4167         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
4168          * been verified in lfsck_layout_verify_header() already. If some
4169          * new magic introduced in the future, then layout LFSCK needs to
4170          * be updated also. */
4171         magic = le32_to_cpu(lmm->lmm_magic);
4172         if (magic == LOV_MAGIC_V1) {
4173                 objs = &lmm->lmm_objects[0];
4174         } else {
4175                 LASSERT(magic == LOV_MAGIC_V3);
4176                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4177         }
4178
4179         for (i = 0; i < count; i++, objs++) {
4180                 struct lu_fid           *fid    = &info->lti_fid;
4181                 struct ost_id           *oi     = &info->lti_oi;
4182                 struct lfsck_layout_req *llr;
4183                 struct lfsck_tgt_desc   *tgt    = NULL;
4184                 struct dt_object        *cobj   = NULL;
4185                 __u32                    index;
4186                 bool                     wakeup = false;
4187
4188                 if (unlikely(lovea_slot_is_dummy(objs)))
4189                         continue;
4190
4191                 l_wait_event(mthread->t_ctl_waitq,
4192                              lad->lad_prefetched < bk->lb_async_windows ||
4193                              !thread_is_running(mthread) ||
4194                              thread_is_stopped(athread),
4195                              &lwi);
4196
4197                 if (unlikely(!thread_is_running(mthread)) ||
4198                              thread_is_stopped(athread))
4199                         GOTO(out, rc = 0);
4200
4201                 if (unlikely(lfsck_is_dead_obj(parent)))
4202                         GOTO(out, rc = 0);
4203
4204                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4205                 index = le32_to_cpu(objs->l_ost_idx);
4206                 rc = ostid_to_fid(fid, oi, index);
4207                 if (rc != 0) {
4208                         CDEBUG(D_LFSCK, "%s: get invalid layout EA for "DFID
4209                                ": "DOSTID", idx:%u\n", lfsck_lfsck2name(lfsck),
4210                                PFID(lfsck_dto2fid(parent)), POSTID(oi), index);
4211                         goto next;
4212                 }
4213
4214                 tgt = lfsck_tgt_get(ltds, index);
4215                 if (unlikely(tgt == NULL)) {
4216                         CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
4217                                "did not join the layout LFSCK\n",
4218                                lfsck_lfsck2name(lfsck), index);
4219                         lfsck_lad_set_bitmap(env, com, index);
4220                         goto next;
4221                 }
4222
4223                 /* There is potential deadlock race condition between object
4224                  * destroy and layout LFSCK. Consider the following scenario:
4225                  *
4226                  * 1) The LFSCK thread obtained the parent object firstly, at
4227                  *    that time, the parent object has not been destroyed yet.
4228                  *
4229                  * 2) One RPC service thread destroyed the parent and all its
4230                  *    children objects. Because the LFSCK is referencing the
4231                  *    parent object, then the parent object will be marked as
4232                  *    dying in RAM. On the other hand, the parent object is
4233                  *    referencing all its children objects, then all children
4234                  *    objects will be marked as dying in RAM also.
4235                  *
4236                  * 3) The LFSCK thread tries to find some child object with
4237                  *    the parent object referenced. Then it will find that the
4238                  *    child object is dying. According to the object visibility
4239                  *    rules: the object with dying flag cannot be returned to
4240                  *    others. So the LFSCK thread has to wait until the dying
4241                  *    object has been purged from RAM, then it can allocate a
4242                  *    new object (with the same FID) in RAM. Unfortunately, the
4243                  *    LFSCK thread itself is referencing the parent object, and
4244                  *    cause the parent object cannot be purged, then cause the
4245                  *    child object cannot be purged also. So the LFSCK thread
4246                  *    will fall into deadlock.
4247                  *
4248                  * We introduce non-blocked version lu_object_find() to allow
4249                  * the LFSCK thread to return failure immediately (instead of
4250                  * wait) when it finds dying (child) object, then the LFSCK
4251                  * thread can check whether the parent object is dying or not.
4252                  * So avoid above deadlock. LU-5395 */
4253                 cobj = lfsck_object_find_by_dev_nowait(env, tgt->ltd_tgt, fid);
4254                 if (IS_ERR(cobj)) {
4255                         if (lfsck_is_dead_obj(parent)) {
4256                                 lfsck_tgt_put(tgt);
4257
4258                                 GOTO(out, rc = 0);
4259                         }
4260
4261                         rc = PTR_ERR(cobj);
4262                         goto next;
4263                 }
4264
4265                 rc = dt_declare_attr_get(env, cobj);
4266                 if (rc != 0)
4267                         goto next;
4268
4269                 rc = dt_declare_xattr_get(env, cobj, &buf, XATTR_NAME_FID);
4270                 if (rc != 0)
4271                         goto next;
4272
4273                 if (lso == NULL) {
4274                         struct lu_attr *attr = &info->lti_la;
4275
4276                         rc = dt_attr_get(env, parent, attr);
4277                         if (rc != 0)
4278                                 goto next;
4279
4280                         lso = lfsck_assistant_object_init(env,
4281                                 lfsck_dto2fid(parent), attr,
4282                                 lfsck->li_pos_current.lp_oit_cookie, false);
4283                         if (IS_ERR(lso)) {
4284                                 rc = PTR_ERR(lso);
4285                                 lso = NULL;
4286
4287                                 goto next;
4288                         }
4289                 }
4290
4291                 llr = lfsck_layout_assistant_req_init(lso, cobj, index, i);
4292                 if (IS_ERR(llr)) {
4293                         rc = PTR_ERR(llr);
4294                         goto next;
4295                 }
4296
4297                 cobj = NULL;
4298                 spin_lock(&lad->lad_lock);
4299                 if (lad->lad_assistant_status < 0) {
4300                         spin_unlock(&lad->lad_lock);
4301                         lfsck_layout_assistant_req_fini(env, &llr->llr_lar);
4302                         lfsck_tgt_put(tgt);
4303                         RETURN(lad->lad_assistant_status);
4304                 }
4305
4306                 list_add_tail(&llr->llr_lar.lar_list, &lad->lad_req_list);
4307                 if (lad->lad_prefetched == 0)
4308                         wakeup = true;
4309
4310                 lad->lad_prefetched++;
4311                 spin_unlock(&lad->lad_lock);
4312                 if (wakeup)
4313                         wake_up_all(&athread->t_ctl_waitq);
4314
4315 next:
4316                 down_write(&com->lc_sem);
4317                 com->lc_new_checked++;
4318                 if (rc < 0)
4319                         lfsck_layout_record_failure(env, lfsck, lo);
4320                 up_write(&com->lc_sem);
4321
4322                 if (cobj != NULL && !IS_ERR(cobj))
4323                         lfsck_object_put(env, cobj);
4324
4325                 if (likely(tgt != NULL))
4326                         lfsck_tgt_put(tgt);
4327
4328                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4329                         GOTO(out, rc);
4330         }
4331
4332         GOTO(out, rc = 0);
4333
4334 out:
4335         if (lso != NULL)
4336                 lfsck_assistant_object_put(env, lso);
4337
4338         return rc;
4339 }
4340
4341 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
4342  * the OST-object's attribute and generate an structure lfsck_layout_req on the
4343  * list ::lad_req_list.
4344  *
4345  * For each request on above list, the lfsck_layout_assistant thread compares
4346  * the OST side attribute with local attribute, if inconsistent, then repair it.
4347  *
4348  * All above processing is async mode with pipeline. */
4349 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
4350                                         struct lfsck_component *com,
4351                                         struct dt_object *obj)
4352 {
4353         struct lfsck_thread_info        *info   = lfsck_env_info(env);
4354         struct ost_id                   *oi     = &info->lti_oi;
4355         struct lfsck_layout             *lo     = com->lc_file_ram;
4356         struct lfsck_assistant_data     *lad    = com->lc_data;
4357         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4358         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
4359         struct thandle                  *handle = NULL;
4360         struct lu_buf                   *buf    = &info->lti_big_buf;
4361         struct lov_mds_md_v1            *lmm    = NULL;
4362         struct dt_device                *dev    = lfsck_obj2dev(obj);
4363         struct lustre_handle             lh     = { 0 };
4364         struct lu_buf                    ea_buf = { NULL };
4365         int                              rc     = 0;
4366         int                              size   = 0;
4367         bool                             locked = false;
4368         bool                             stripe = false;
4369         bool                             bad_oi = false;
4370         ENTRY;
4371
4372         if (!S_ISREG(lfsck_object_type(obj)))
4373                 GOTO(out, rc = 0);
4374
4375         if (lad->lad_assistant_status < 0)
4376                 GOTO(out, rc = -ESRCH);
4377
4378         fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
4379         lmm_oi_cpu_to_le(oi, oi);
4380         dt_read_lock(env, obj, 0);
4381         locked = true;
4382
4383 again:
4384         if (dt_object_exists(obj) == 0 ||
4385             lfsck_is_dead_obj(obj))
4386                 GOTO(out, rc = 0);
4387
4388         rc = lfsck_layout_get_lovea(env, obj, buf);
4389         if (rc <= 0)
4390                 GOTO(out, rc);
4391
4392         size = rc;
4393         lmm = buf->lb_buf;
4394         rc = lfsck_layout_verify_header(lmm);
4395         /* If the LOV EA crashed, then it is possible to be rebuilt later
4396          * when handle orphan OST-objects. */
4397         if (rc != 0)
4398                 GOTO(out, rc);
4399
4400         if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0)
4401                 GOTO(out, stripe = true);
4402
4403         /* Inconsistent lmm_oi, should be repaired. */
4404         bad_oi = true;
4405         lmm->lmm_oi = *oi;
4406
4407         if (bk->lb_param & LPF_DRYRUN) {
4408                 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4409
4410                 GOTO(out, stripe = true);
4411         }
4412
4413         if (!lustre_handle_is_used(&lh)) {
4414                 dt_read_unlock(env, obj);
4415                 locked = false;
4416                 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4417                                       MDS_INODELOCK_LAYOUT |
4418                                       MDS_INODELOCK_XATTR, LCK_EX);
4419                 if (rc != 0)
4420                         GOTO(out, rc);
4421
4422                 handle = dt_trans_create(env, dev);
4423                 if (IS_ERR(handle))
4424                         GOTO(out, rc = PTR_ERR(handle));
4425
4426                 lfsck_buf_init(&ea_buf, lmm, size);
4427                 rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
4428                                           LU_XATTR_REPLACE, handle);
4429                 if (rc != 0)
4430                         GOTO(out, rc);
4431
4432                 rc = dt_trans_start_local(env, dev, handle);
4433                 if (rc != 0)
4434                         GOTO(out, rc);
4435
4436                 dt_write_lock(env, obj, 0);
4437                 locked = true;
4438
4439                 goto again;
4440         }
4441
4442         rc = dt_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
4443                           LU_XATTR_REPLACE, handle);
4444         if (rc != 0)
4445                 GOTO(out, rc);
4446
4447         lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4448
4449         GOTO(out, stripe = true);
4450
4451 out:
4452         if (locked) {
4453                 if (lustre_handle_is_used(&lh))
4454                         dt_write_unlock(env, obj);
4455                 else
4456                         dt_read_unlock(env, obj);
4457         }
4458
4459         if (handle != NULL && !IS_ERR(handle))
4460                 dt_trans_stop(env, dev, handle);
4461
4462         lfsck_ibits_unlock(&lh, LCK_EX);
4463
4464         if (bad_oi)
4465                 CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
4466                        DFID": rc = %d\n", lfsck_lfsck2name(lfsck),
4467                        bk->lb_param & LPF_DRYRUN ? "found" : "repaired",
4468                        PFID(lfsck_dto2fid(obj)), rc);
4469
4470         if (stripe) {
4471                 rc = lfsck_layout_scan_stripes(env, com, obj, lmm);
4472         } else {
4473                 down_write(&com->lc_sem);
4474                 com->lc_new_checked++;
4475                 if (rc < 0)
4476                         lfsck_layout_record_failure(env, lfsck, lo);
4477                 up_write(&com->lc_sem);
4478         }
4479
4480         return rc;
4481 }
4482
4483 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
4484                                        struct lfsck_component *com,
4485                                        struct dt_object *obj)
4486 {
4487         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4488         struct lfsck_layout             *lo     = com->lc_file_ram;
4489         const struct lu_fid             *fid    = lfsck_dto2fid(obj);
4490         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4491         struct lfsck_layout_seq         *lls;
4492         __u64                            seq;
4493         __u64                            oid;
4494         int                              rc;
4495         ENTRY;
4496
4497         LASSERT(llsd != NULL);
4498
4499         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
4500             cfs_fail_val == lfsck_dev_idx(lfsck)) {
4501                 struct l_wait_info       lwi = LWI_TIMEOUT(cfs_time_seconds(1),
4502                                                            NULL, NULL);
4503                 struct ptlrpc_thread    *thread = &lfsck->li_thread;
4504
4505                 l_wait_event(thread->t_ctl_waitq,
4506                              !thread_is_running(thread),
4507                              &lwi);
4508         }
4509
4510         lfsck_rbtree_update_bitmap(env, com, fid, false);
4511
4512         down_write(&com->lc_sem);
4513         if (fid_is_idif(fid))
4514                 seq = 0;
4515         else if (!fid_is_norm(fid) ||
4516                  !fid_is_for_ostobj(env, lfsck, obj, fid))
4517                 GOTO(unlock, rc = 0);
4518         else
4519                 seq = fid_seq(fid);
4520         com->lc_new_checked++;
4521
4522         lls = lfsck_layout_seq_lookup(llsd, seq);
4523         if (lls == NULL) {
4524                 OBD_ALLOC_PTR(lls);
4525                 if (unlikely(lls == NULL))
4526                         GOTO(unlock, rc = -ENOMEM);
4527
4528                 INIT_LIST_HEAD(&lls->lls_list);
4529                 lls->lls_seq = seq;
4530                 rc = lfsck_layout_lastid_load(env, com, lls);
4531                 if (rc != 0) {
4532                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
4533                               "load LAST_ID for "LPX64": rc = %d\n",
4534                               lfsck_lfsck2name(com->lc_lfsck), seq, rc);
4535                         lo->ll_objs_failed_phase1++;
4536                         OBD_FREE_PTR(lls);
4537                         GOTO(unlock, rc);
4538                 }
4539
4540                 lfsck_layout_seq_insert(llsd, lls);
4541         }
4542
4543         if (unlikely(fid_is_last_id(fid)))
4544                 GOTO(unlock, rc = 0);
4545
4546         if (fid_is_idif(fid))
4547                 oid = fid_idif_id(fid_seq(fid), fid_oid(fid), fid_ver(fid));
4548         else
4549                 oid = fid_oid(fid);
4550
4551         if (oid > lls->lls_lastid_known)
4552                 lls->lls_lastid_known = oid;
4553
4554         if (oid > lls->lls_lastid) {
4555                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
4556                         /* OFD may create new objects during LFSCK scanning. */
4557                         rc = lfsck_layout_lastid_reload(env, com, lls);
4558                         if (unlikely(rc != 0)) {
4559                                 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
4560                                       "reload LAST_ID for "LPX64": rc = %d\n",
4561                                       lfsck_lfsck2name(com->lc_lfsck),
4562                                       lls->lls_seq, rc);
4563
4564                                 GOTO(unlock, rc);
4565                         }
4566
4567                         if (oid <= lls->lls_lastid ||
4568                             lo->ll_flags & LF_CRASHED_LASTID)
4569                                 GOTO(unlock, rc = 0);
4570
4571                         LASSERT(lfsck->li_out_notify != NULL);
4572
4573                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4574                                              LE_LASTID_REBUILDING);
4575                         lo->ll_flags |= LF_CRASHED_LASTID;
4576
4577                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
4578                                "LAST_ID file (2) for the sequence "LPX64
4579                                ", old value "LPU64", known value "LPU64"\n",
4580                                lfsck_lfsck2name(lfsck), lls->lls_seq,
4581                                lls->lls_lastid, oid);
4582                 }
4583
4584                 lls->lls_lastid = oid;
4585                 lls->lls_dirty = 1;
4586         }
4587
4588         GOTO(unlock, rc = 0);
4589
4590 unlock:
4591         up_write(&com->lc_sem);
4592
4593         return rc;
4594 }
4595
4596 static int lfsck_layout_exec_dir(const struct lu_env *env,
4597                                  struct lfsck_component *com,
4598                                  struct lfsck_assistant_object *lso,
4599                                  struct lu_dirent *ent, __u16 type)
4600 {
4601         return 0;
4602 }
4603
4604 static int lfsck_layout_master_post(const struct lu_env *env,
4605                                     struct lfsck_component *com,
4606                                     int result, bool init)
4607 {
4608         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4609         struct lfsck_layout     *lo     = com->lc_file_ram;
4610         int                      rc;
4611         ENTRY;
4612
4613         lfsck_post_generic(env, com, &result);
4614
4615         down_write(&com->lc_sem);
4616         spin_lock(&lfsck->li_lock);
4617         if (!init)
4618                 lo->ll_pos_last_checkpoint =
4619                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
4620
4621         if (result > 0) {
4622                 if (lo->ll_flags & LF_INCOMPLETE)
4623                         lo->ll_status = LS_PARTIAL;
4624                 else
4625                         lo->ll_status = LS_SCANNING_PHASE2;
4626                 lo->ll_flags |= LF_SCANNED_ONCE;
4627                 lo->ll_flags &= ~LF_UPGRADE;
4628                 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4629         } else if (result == 0) {
4630                 if (lfsck->li_status != 0)
4631                         lo->ll_status = lfsck->li_status;
4632                 else
4633                         lo->ll_status = LS_STOPPED;
4634                 if (lo->ll_status != LS_PAUSED)
4635                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4636         } else {
4637                 lo->ll_status = LS_FAILED;
4638                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4639         }
4640         spin_unlock(&lfsck->li_lock);
4641
4642         if (!init) {
4643                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4644                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4645                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4646                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4647                 com->lc_new_checked = 0;
4648         }
4649
4650         rc = lfsck_layout_store(env, com);
4651         up_write(&com->lc_sem);
4652
4653         CDEBUG(D_LFSCK, "%s: layout LFSCK master post done: rc = %d\n",
4654                lfsck_lfsck2name(lfsck), rc);
4655
4656         RETURN(rc);
4657 }
4658
4659 static int lfsck_layout_slave_post(const struct lu_env *env,
4660                                    struct lfsck_component *com,
4661                                    int result, bool init)
4662 {
4663         struct lfsck_instance   *lfsck = com->lc_lfsck;
4664         struct lfsck_layout     *lo    = com->lc_file_ram;
4665         int                      rc;
4666         bool                     done  = false;
4667
4668         rc = lfsck_layout_lastid_store(env, com);
4669         if (rc != 0)
4670                 result = rc;
4671
4672         LASSERT(lfsck->li_out_notify != NULL);
4673
4674         down_write(&com->lc_sem);
4675         spin_lock(&lfsck->li_lock);
4676         if (!init)
4677                 lo->ll_pos_last_checkpoint =
4678                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
4679
4680         if (result > 0) {
4681                 lo->ll_status = LS_SCANNING_PHASE2;
4682                 lo->ll_flags |= LF_SCANNED_ONCE;
4683                 if (lo->ll_flags & LF_CRASHED_LASTID) {
4684                         done = true;
4685                         lo->ll_flags &= ~LF_CRASHED_LASTID;
4686
4687                         CDEBUG(D_LFSCK, "%s: layout LFSCK has rebuilt "
4688                                "crashed LAST_ID files successfully\n",
4689                                lfsck_lfsck2name(lfsck));
4690                 }
4691                 lo->ll_flags &= ~LF_UPGRADE;
4692                 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4693         } else if (result == 0) {
4694                 if (lfsck->li_status != 0)
4695                         lo->ll_status = lfsck->li_status;
4696                 else
4697                         lo->ll_status = LS_STOPPED;
4698                 if (lo->ll_status != LS_PAUSED)
4699                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4700         } else {
4701                 lo->ll_status = LS_FAILED;
4702                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4703         }
4704         spin_unlock(&lfsck->li_lock);
4705
4706         if (done)
4707                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4708                                      LE_LASTID_REBUILT);
4709
4710         if (!init) {
4711                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4712                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4713                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4714                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4715                 com->lc_new_checked = 0;
4716         }
4717
4718         rc = lfsck_layout_store(env, com);
4719         up_write(&com->lc_sem);
4720
4721         lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
4722
4723         CDEBUG(D_LFSCK, "%s: layout LFSCK slave post done: rc = %d\n",
4724                lfsck_lfsck2name(lfsck), rc);
4725
4726         return rc;
4727 }
4728
4729 static int lfsck_layout_dump(const struct lu_env *env,
4730                              struct lfsck_component *com, struct seq_file *m)
4731 {
4732         struct lfsck_instance   *lfsck = com->lc_lfsck;
4733         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
4734         struct lfsck_layout     *lo    = com->lc_file_ram;
4735         int                      rc;
4736
4737         down_read(&com->lc_sem);
4738         seq_printf(m, "name: lfsck_layout\n"
4739                       "magic: %#x\n"
4740                       "version: %d\n"
4741                       "status: %s\n",
4742                       lo->ll_magic,
4743                       bk->lb_version,
4744                       lfsck_status2names(lo->ll_status));
4745
4746         rc = lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
4747         if (rc < 0)
4748                 goto out;
4749
4750         rc = lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4751         if (rc < 0)
4752                 goto out;
4753
4754         rc = lfsck_time_dump(m, lo->ll_time_last_complete,
4755                              "last_completed");
4756         if (rc < 0)
4757                 goto out;
4758
4759         rc = lfsck_time_dump(m, lo->ll_time_latest_start,
4760                              "latest_start");
4761         if (rc < 0)
4762                 goto out;
4763
4764         rc = lfsck_time_dump(m, lo->ll_time_last_checkpoint,
4765                              "last_checkpoint");
4766         if (rc < 0)
4767                 goto out;
4768
4769         seq_printf(m, "latest_start_position: "LPU64"\n"
4770                       "last_checkpoint_position: "LPU64"\n"
4771                       "first_failure_position: "LPU64"\n",
4772                       lo->ll_pos_latest_start,
4773                       lo->ll_pos_last_checkpoint,
4774                       lo->ll_pos_first_inconsistent);
4775
4776         seq_printf(m, "success_count: %u\n"
4777                       "repaired_dangling: "LPU64"\n"
4778                       "repaired_unmatched_pair: "LPU64"\n"
4779                       "repaired_multiple_referenced: "LPU64"\n"
4780                       "repaired_orphan: "LPU64"\n"
4781                       "repaired_inconsistent_owner: "LPU64"\n"
4782                       "repaired_others: "LPU64"\n"
4783                       "skipped: "LPU64"\n"
4784                       "failed_phase1: "LPU64"\n"
4785                       "failed_phase2: "LPU64"\n",
4786                       lo->ll_success_count,
4787                       lo->ll_objs_repaired[LLIT_DANGLING - 1],
4788                       lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
4789                       lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
4790                       lo->ll_objs_repaired[LLIT_ORPHAN - 1],
4791                       lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
4792                       lo->ll_objs_repaired[LLIT_OTHERS - 1],
4793                       lo->ll_objs_skipped,
4794                       lo->ll_objs_failed_phase1,
4795                       lo->ll_objs_failed_phase2);
4796
4797         if (lo->ll_status == LS_SCANNING_PHASE1) {
4798                 __u64 pos;
4799                 const struct dt_it_ops *iops;
4800                 cfs_duration_t duration = cfs_time_current() -
4801                                           lfsck->li_time_last_checkpoint;
4802                 __u64 checked = lo->ll_objs_checked_phase1 +
4803                                 com->lc_new_checked;
4804                 __u64 speed = checked;
4805                 __u64 new_checked = com->lc_new_checked *
4806                                     msecs_to_jiffies(MSEC_PER_SEC);
4807                 __u32 rtime = lo->ll_run_time_phase1 +
4808                               cfs_duration_sec(duration + HALF_SEC);
4809
4810                 if (duration != 0)
4811                         do_div(new_checked, duration);
4812                 if (rtime != 0)
4813                         do_div(speed, rtime);
4814                 seq_printf(m, "checked_phase1: "LPU64"\n"
4815                               "checked_phase2: "LPU64"\n"
4816                               "run_time_phase1: %u seconds\n"
4817                               "run_time_phase2: %u seconds\n"
4818                               "average_speed_phase1: "LPU64" items/sec\n"
4819                               "average_speed_phase2: N/A\n"
4820                               "real-time_speed_phase1: "LPU64" items/sec\n"
4821                               "real-time_speed_phase2: N/A\n",
4822                               checked,
4823                               lo->ll_objs_checked_phase2,
4824                               rtime,
4825                               lo->ll_run_time_phase2,
4826                               speed,
4827                               new_checked);
4828
4829                 LASSERT(lfsck->li_di_oit != NULL);
4830
4831                 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
4832
4833                 /* The low layer otable-based iteration position may NOT
4834                  * exactly match the layout-based directory traversal
4835                  * cookie. Generally, it is not a serious issue. But the
4836                  * caller should NOT make assumption on that. */
4837                 pos = iops->store(env, lfsck->li_di_oit);
4838                 if (!lfsck->li_current_oit_processed)
4839                         pos--;
4840                 seq_printf(m, "current_position: "LPU64"\n", pos);
4841
4842         } else if (lo->ll_status == LS_SCANNING_PHASE2) {
4843                 cfs_duration_t duration = cfs_time_current() -
4844                                           com->lc_time_last_checkpoint;
4845                 __u64 checked = lo->ll_objs_checked_phase2 +
4846                                 com->lc_new_checked;
4847                 __u64 speed1 = lo->ll_objs_checked_phase1;
4848                 __u64 speed2 = checked;
4849                 __u64 new_checked = com->lc_new_checked *
4850                                     msecs_to_jiffies(MSEC_PER_SEC);
4851                 __u32 rtime = lo->ll_run_time_phase2 +
4852                               cfs_duration_sec(duration + HALF_SEC);
4853
4854                 if (duration != 0)
4855                         do_div(new_checked, duration);
4856                 if (lo->ll_run_time_phase1 != 0)
4857                         do_div(speed1, lo->ll_run_time_phase1);
4858                 if (rtime != 0)
4859                         do_div(speed2, rtime);
4860                 rc = seq_printf(m, "checked_phase1: "LPU64"\n"
4861                                 "checked_phase2: "LPU64"\n"
4862                                 "run_time_phase1: %u seconds\n"
4863                                 "run_time_phase2: %u seconds\n"
4864                                 "average_speed_phase1: "LPU64" items/sec\n"
4865                                 "average_speed_phase2: "LPU64" items/sec\n"
4866                                 "real-time_speed_phase1: N/A\n"
4867                                 "real-time_speed_phase2: "LPU64" items/sec\n"
4868                                 "current_position: "DFID"\n",
4869                                 lo->ll_objs_checked_phase1,
4870                                 checked,
4871                                 lo->ll_run_time_phase1,
4872                                 rtime,
4873                                 speed1,
4874                                 speed2,
4875                                 new_checked,
4876                                 PFID(&com->lc_fid_latest_scanned_phase2));
4877                 if (rc <= 0)
4878                         goto out;
4879
4880         } else {
4881                 __u64 speed1 = lo->ll_objs_checked_phase1;
4882                 __u64 speed2 = lo->ll_objs_checked_phase2;
4883
4884                 if (lo->ll_run_time_phase1 != 0)
4885                         do_div(speed1, lo->ll_run_time_phase1);
4886                 if (lo->ll_run_time_phase2 != 0)
4887                         do_div(speed2, lo->ll_run_time_phase2);
4888                 seq_printf(m, "checked_phase1: "LPU64"\n"
4889                            "checked_phase2: "LPU64"\n"
4890                            "run_time_phase1: %u seconds\n"
4891                            "run_time_phase2: %u seconds\n"
4892                            "average_speed_phase1: "LPU64" items/sec\n"
4893                            "average_speed_phase2: "LPU64" objs/sec\n"
4894                            "real-time_speed_phase1: N/A\n"
4895                            "real-time_speed_phase2: N/A\n"
4896                            "current_position: N/A\n",
4897                            lo->ll_objs_checked_phase1,
4898                            lo->ll_objs_checked_phase2,
4899                            lo->ll_run_time_phase1,
4900                            lo->ll_run_time_phase2,
4901                            speed1,
4902                            speed2);
4903         }
4904 out:
4905         up_read(&com->lc_sem);
4906
4907         return rc;
4908 }
4909
4910 static int lfsck_layout_master_double_scan(const struct lu_env *env,
4911                                            struct lfsck_component *com)
4912 {
4913         struct lfsck_layout             *lo     = com->lc_file_ram;
4914         struct lfsck_assistant_data     *lad    = com->lc_data;
4915         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4916         struct lfsck_tgt_descs          *ltds;
4917         struct lfsck_tgt_desc           *ltd;
4918         struct lfsck_tgt_desc           *next;
4919         int                              rc;
4920
4921         rc = lfsck_double_scan_generic(env, com, lo->ll_status);
4922
4923         if (thread_is_stopped(&lad->lad_thread)) {
4924                 LASSERT(list_empty(&lad->lad_req_list));
4925                 LASSERT(list_empty(&lad->lad_ost_phase1_list));
4926                 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4927
4928                 ltds = &lfsck->li_ost_descs;
4929                 spin_lock(&ltds->ltd_lock);
4930                 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
4931                                          ltd_layout_phase_list) {
4932                         list_del_init(&ltd->ltd_layout_phase_list);
4933                 }
4934                 spin_unlock(&ltds->ltd_lock);
4935
4936                 ltds = &lfsck->li_mdt_descs;
4937                 spin_lock(&ltds->ltd_lock);
4938                 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4939                                          ltd_layout_phase_list) {
4940                         list_del_init(&ltd->ltd_layout_phase_list);
4941                 }
4942                 spin_unlock(&ltds->ltd_lock);
4943         }
4944
4945         return rc;
4946 }
4947
4948 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
4949                                           struct lfsck_component *com)
4950 {
4951         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4952         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4953         struct lfsck_layout             *lo     = com->lc_file_ram;
4954         struct ptlrpc_thread            *thread = &lfsck->li_thread;
4955         int                              rc;
4956         ENTRY;
4957
4958         CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan start\n",
4959                lfsck_lfsck2name(lfsck));
4960
4961         atomic_inc(&lfsck->li_double_scan_count);
4962
4963         if (lo->ll_flags & LF_INCOMPLETE)
4964                 GOTO(done, rc = 1);
4965
4966         com->lc_new_checked = 0;
4967         com->lc_new_scanned = 0;
4968         com->lc_time_last_checkpoint = cfs_time_current();
4969         com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
4970                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
4971
4972         while (1) {
4973                 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(30),
4974                                                      NULL, NULL);
4975
4976                 rc = lfsck_layout_slave_query_master(env, com);
4977                 if (list_empty(&llsd->llsd_master_list)) {
4978                         if (unlikely(!thread_is_running(thread)))
4979                                 rc = 0;
4980                         else
4981                                 rc = 1;
4982
4983                         GOTO(done, rc);
4984                 }
4985
4986                 if (rc < 0)
4987                         GOTO(done, rc);
4988
4989                 rc = l_wait_event(thread->t_ctl_waitq,
4990                                   !thread_is_running(thread) ||
4991                                   lo->ll_flags & LF_INCOMPLETE ||
4992                                   list_empty(&llsd->llsd_master_list),
4993                                   &lwi);
4994                 if (unlikely(!thread_is_running(thread)))
4995                         GOTO(done, rc = 0);
4996
4997                 if (lo->ll_flags & LF_INCOMPLETE)
4998                         GOTO(done, rc = 1);
4999
5000                 if (rc == -ETIMEDOUT)
5001                         continue;
5002
5003                 GOTO(done, rc = (rc < 0 ? rc : 1));
5004         }
5005
5006 done:
5007         rc = lfsck_layout_double_scan_result(env, com, rc);
5008         lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE,
5009                         (rc > 0 && lo->ll_flags & LF_INCOMPLETE) ? 0 : rc);
5010         lfsck_layout_slave_quit(env, com);
5011         if (atomic_dec_and_test(&lfsck->li_double_scan_count))
5012                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5013
5014         CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan finished, "
5015                "status %d: rc = %d\n",
5016                lfsck_lfsck2name(lfsck), lo->ll_status, rc);
5017
5018         return rc;
5019 }
5020
5021 static void lfsck_layout_master_data_release(const struct lu_env *env,
5022                                              struct lfsck_component *com)
5023 {
5024         struct lfsck_assistant_data     *lad    = com->lc_data;
5025         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5026         struct lfsck_tgt_descs          *ltds;
5027         struct lfsck_tgt_desc           *ltd;
5028         struct lfsck_tgt_desc           *next;
5029
5030         LASSERT(lad != NULL);
5031         LASSERT(thread_is_init(&lad->lad_thread) ||
5032                 thread_is_stopped(&lad->lad_thread));
5033         LASSERT(list_empty(&lad->lad_req_list));
5034
5035         com->lc_data = NULL;
5036
5037         ltds = &lfsck->li_ost_descs;
5038         spin_lock(&ltds->ltd_lock);
5039         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
5040                                  ltd_layout_phase_list) {
5041                 list_del_init(&ltd->ltd_layout_phase_list);
5042         }
5043         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5044                                  ltd_layout_phase_list) {
5045                 list_del_init(&ltd->ltd_layout_phase_list);
5046         }
5047         list_for_each_entry_safe(ltd, next, &lad->lad_ost_list,
5048                                  ltd_layout_list) {
5049                 list_del_init(&ltd->ltd_layout_list);
5050         }
5051         spin_unlock(&ltds->ltd_lock);
5052
5053         ltds = &lfsck->li_mdt_descs;
5054         spin_lock(&ltds->ltd_lock);
5055         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5056                                  ltd_layout_phase_list) {
5057                 list_del_init(&ltd->ltd_layout_phase_list);
5058         }
5059         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5060                                  ltd_layout_phase_list) {
5061                 list_del_init(&ltd->ltd_layout_phase_list);
5062         }
5063         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
5064                                  ltd_layout_list) {
5065                 list_del_init(&ltd->ltd_layout_list);
5066         }
5067         spin_unlock(&ltds->ltd_lock);
5068
5069         if (likely(lad->lad_bitmap != NULL))
5070                 CFS_FREE_BITMAP(lad->lad_bitmap);
5071
5072         OBD_FREE_PTR(lad);
5073 }
5074
5075 static void lfsck_layout_slave_data_release(const struct lu_env *env,
5076                                             struct lfsck_component *com)
5077 {
5078         struct lfsck_layout_slave_data *llsd = com->lc_data;
5079
5080         lfsck_layout_slave_quit(env, com);
5081         com->lc_data = NULL;
5082         OBD_FREE_PTR(llsd);
5083 }
5084
5085 static void lfsck_layout_master_quit(const struct lu_env *env,
5086                                      struct lfsck_component *com)
5087 {
5088         struct lfsck_assistant_data     *lad    = com->lc_data;
5089         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5090         struct lfsck_tgt_descs          *ltds;
5091         struct lfsck_tgt_desc           *ltd;
5092         struct lfsck_tgt_desc           *next;
5093
5094         LASSERT(lad != NULL);
5095
5096         lfsck_quit_generic(env, com);
5097
5098         LASSERT(thread_is_init(&lad->lad_thread) ||
5099                 thread_is_stopped(&lad->lad_thread));
5100         LASSERT(list_empty(&lad->lad_req_list));
5101
5102         ltds = &lfsck->li_ost_descs;
5103         spin_lock(&ltds->ltd_lock);
5104         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
5105                                  ltd_layout_phase_list) {
5106                 list_del_init(&ltd->ltd_layout_phase_list);
5107         }
5108         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5109                                  ltd_layout_phase_list) {
5110                 list_del_init(&ltd->ltd_layout_phase_list);
5111         }
5112         spin_unlock(&ltds->ltd_lock);
5113
5114         ltds = &lfsck->li_mdt_descs;
5115         spin_lock(&ltds->ltd_lock);
5116         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5117                                  ltd_layout_phase_list) {
5118                 list_del_init(&ltd->ltd_layout_phase_list);
5119         }
5120         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5121                                  ltd_layout_phase_list) {
5122                 list_del_init(&ltd->ltd_layout_phase_list);
5123         }
5124         spin_unlock(&ltds->ltd_lock);
5125 }
5126
5127 static void lfsck_layout_slave_quit(const struct lu_env *env,
5128                                     struct lfsck_component *com)
5129 {
5130         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5131         struct lfsck_layout_seq          *lls;
5132         struct lfsck_layout_seq          *next;
5133         struct lfsck_layout_slave_target *llst;
5134
5135         LASSERT(llsd != NULL);
5136
5137         list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
5138                                  lls_list) {
5139                 list_del_init(&lls->lls_list);
5140                 lfsck_object_put(env, lls->lls_lastid_obj);
5141                 OBD_FREE_PTR(lls);
5142         }
5143
5144         spin_lock(&llsd->llsd_lock);
5145         while (!list_empty(&llsd->llsd_master_list)) {
5146                 llst = list_entry(llsd->llsd_master_list.next,
5147                                   struct lfsck_layout_slave_target, llst_list);
5148                 list_del_init(&llst->llst_list);
5149                 spin_unlock(&llsd->llsd_lock);
5150                 lfsck_layout_llst_put(llst);
5151                 spin_lock(&llsd->llsd_lock);
5152         }
5153         spin_unlock(&llsd->llsd_lock);
5154
5155         lfsck_rbtree_cleanup(env, com);
5156 }
5157
5158 static int lfsck_layout_master_in_notify(const struct lu_env *env,
5159                                          struct lfsck_component *com,
5160                                          struct lfsck_request *lr,
5161                                          struct thandle *th)
5162 {
5163         struct lfsck_instance           *lfsck = com->lc_lfsck;
5164         struct lfsck_layout             *lo    = com->lc_file_ram;
5165         struct lfsck_assistant_data     *lad   = com->lc_data;
5166         struct lfsck_tgt_descs          *ltds;
5167         struct lfsck_tgt_desc           *ltd;
5168         bool                             fail  = false;
5169         ENTRY;
5170
5171         if (lr->lr_event == LE_PAIRS_VERIFY) {
5172                 int rc;
5173
5174                 rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
5175                                                      &lr->lr_fid2);
5176
5177                 RETURN(rc);
5178         }
5179
5180         CDEBUG(D_LFSCK, "%s: layout LFSCK master handles notify %u "
5181                "from %s %x, status %d, flags %x, flags2 %x\n",
5182                lfsck_lfsck2name(lfsck), lr->lr_event,
5183                (lr->lr_flags & LEF_FROM_OST) ? "OST" : "MDT",
5184                lr->lr_index, lr->lr_status, lr->lr_flags, lr->lr_flags2);
5185
5186         if (lr->lr_event != LE_PHASE1_DONE &&
5187             lr->lr_event != LE_PHASE2_DONE &&
5188             lr->lr_event != LE_PEER_EXIT)
5189                 RETURN(-EINVAL);
5190
5191         if (lr->lr_flags & LEF_FROM_OST)
5192                 ltds = &lfsck->li_ost_descs;
5193         else
5194                 ltds = &lfsck->li_mdt_descs;
5195         spin_lock(&ltds->ltd_lock);
5196         ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
5197         if (ltd == NULL) {
5198                 spin_unlock(&ltds->ltd_lock);
5199
5200                 RETURN(-ENXIO);
5201         }
5202
5203         list_del_init(&ltd->ltd_layout_phase_list);
5204         switch (lr->lr_event) {
5205         case LE_PHASE1_DONE:
5206                 if (lr->lr_status <= 0 || lr->lr_flags2 & LF_INCOMPLETE) {
5207                         if (lr->lr_flags2 & LF_INCOMPLETE) {
5208                                 if (lr->lr_flags & LEF_FROM_OST)
5209                                         lfsck_lad_set_bitmap(env, com,
5210                                                              ltd->ltd_index);
5211                                 else
5212                                         lo->ll_flags |= LF_INCOMPLETE;
5213                         }
5214                         ltd->ltd_layout_done = 1;
5215                         list_del_init(&ltd->ltd_layout_list);
5216                         fail = true;
5217                         break;
5218                 }
5219
5220                 if (lr->lr_flags & LEF_FROM_OST) {
5221                         if (list_empty(&ltd->ltd_layout_list))
5222                                 list_add_tail(&ltd->ltd_layout_list,
5223                                               &lad->lad_ost_list);
5224                         list_add_tail(&ltd->ltd_layout_phase_list,
5225                                       &lad->lad_ost_phase2_list);
5226                 } else {
5227                         if (list_empty(&ltd->ltd_layout_list))
5228                                 list_add_tail(&ltd->ltd_layout_list,
5229                                               &lad->lad_mdt_list);
5230                         list_add_tail(&ltd->ltd_layout_phase_list,
5231                                       &lad->lad_mdt_phase2_list);
5232                 }
5233                 break;
5234         case LE_PHASE2_DONE:
5235                 ltd->ltd_layout_done = 1;
5236                 if (!list_empty(&ltd->ltd_layout_list)) {
5237                         list_del_init(&ltd->ltd_layout_list);
5238                         if (lr->lr_flags2 & LF_INCOMPLETE) {
5239                                 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
5240                                 fail = true;
5241                         }
5242                 }
5243
5244                 break;
5245         case LE_PEER_EXIT:
5246                 fail = true;
5247                 ltd->ltd_layout_done = 1;
5248                 list_del_init(&ltd->ltd_layout_list);
5249                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) &&
5250                     !(lr->lr_flags & LEF_FROM_OST))
5251                                 lo->ll_flags |= LF_INCOMPLETE;
5252                 break;
5253         default:
5254                 break;
5255         }
5256         spin_unlock(&ltds->ltd_lock);
5257
5258         if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
5259                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5260
5261                 memset(stop, 0, sizeof(*stop));
5262                 stop->ls_status = lr->lr_status;
5263                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5264                 lfsck_stop(env, lfsck->li_bottom, stop);
5265         } else if (lfsck_phase2_next_ready(lad)) {
5266                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
5267         }
5268
5269         RETURN(0);
5270 }
5271
5272 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
5273                                         struct lfsck_component *com,
5274                                         struct lfsck_request *lr,
5275                                         struct thandle *th)
5276 {
5277         struct lfsck_instance            *lfsck = com->lc_lfsck;
5278         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5279         struct lfsck_layout_slave_target *llst;
5280         int                               rc;
5281         ENTRY;
5282
5283         switch (lr->lr_event) {
5284         case LE_FID_ACCESSED:
5285                 lfsck_rbtree_update_bitmap(env, com, &lr->lr_fid, true);
5286                 RETURN(0);
5287         case LE_CONDITIONAL_DESTROY:
5288                 rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
5289                 RETURN(rc);
5290         case LE_PAIRS_VERIFY: {
5291                 lr->lr_status = LPVS_INIT;
5292                 /* Firstly, if the MDT-object which is claimed via OST-object
5293                  * local stored PFID xattr recognizes the OST-object, then it
5294                  * must be that the client given PFID is wrong. */
5295                 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5296                                                     &lr->lr_fid3);
5297                 if (rc <= 0)
5298                         RETURN(0);
5299
5300                 lr->lr_status = LPVS_INCONSISTENT;
5301                 /* The OST-object local stored PFID xattr is stale. We need to
5302                  * check whether the MDT-object that is claimed via the client
5303                  * given PFID information recognizes the OST-object or not. If
5304                  * matches, then need to update the OST-object's PFID xattr. */
5305                 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5306                                                     &lr->lr_fid2);
5307                 /* For rc < 0 case:
5308                  * We are not sure whether the client given PFID information
5309                  * is correct or not, do nothing to avoid improper fixing.
5310                  *
5311                  * For rc > 0 case:
5312                  * The client given PFID information is also invalid, we can
5313                  * NOT fix the OST-object inconsistency.
5314                  */
5315                 if (rc != 0)
5316                         RETURN(rc);
5317
5318                 lr->lr_status = LPVS_INCONSISTENT_TOFIX;
5319                 rc = lfsck_layout_slave_repair_pfid(env, com, lr);
5320
5321                 RETURN(rc);
5322         }
5323         case LE_PHASE1_DONE: {
5324                 if (lr->lr_flags2 & LF_INCOMPLETE) {
5325                         struct lfsck_layout *lo = com->lc_file_ram;
5326
5327                         lo->ll_flags |= LF_INCOMPLETE;
5328                         llst = lfsck_layout_llst_find_and_del(llsd,
5329                                                               lr->lr_index,
5330                                                               true);
5331                         if (llst != NULL) {
5332                                 lfsck_layout_llst_put(llst);
5333                                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5334                         }
5335                 }
5336
5337                 RETURN(0);
5338         }
5339         case LE_PHASE2_DONE:
5340         case LE_PEER_EXIT:
5341                 CDEBUG(D_LFSCK, "%s: layout LFSCK slave handle notify %u "
5342                        "from MDT %x, status %d\n", lfsck_lfsck2name(lfsck),
5343                        lr->lr_event, lr->lr_index, lr->lr_status);
5344                 break;
5345         default:
5346                 RETURN(-EINVAL);
5347         }
5348
5349         llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
5350         if (llst == NULL)
5351                 RETURN(0);
5352
5353         lfsck_layout_llst_put(llst);
5354         if (list_empty(&llsd->llsd_master_list))
5355                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5356
5357         if (lr->lr_event == LE_PEER_EXIT &&
5358             (lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT ||
5359              (list_empty(&llsd->llsd_master_list) &&
5360               (lr->lr_status == LS_STOPPED ||
5361                lr->lr_status == LS_CO_STOPPED)))) {
5362                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5363
5364                 memset(stop, 0, sizeof(*stop));
5365                 stop->ls_status = lr->lr_status;
5366                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5367                 lfsck_stop(env, lfsck->li_bottom, stop);
5368         }
5369
5370         RETURN(0);
5371 }
5372
5373 static int lfsck_layout_query(const struct lu_env *env,
5374                               struct lfsck_component *com)
5375 {
5376         struct lfsck_layout *lo = com->lc_file_ram;
5377
5378         return lo->ll_status;
5379 }
5380
5381 /* with lfsck::li_lock held */
5382 static int lfsck_layout_slave_join(const struct lu_env *env,
5383                                    struct lfsck_component *com,
5384                                    struct lfsck_start_param *lsp)
5385 {
5386         struct lfsck_instance            *lfsck = com->lc_lfsck;
5387         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5388         struct lfsck_layout_slave_target *llst;
5389         struct lfsck_start               *start = lsp->lsp_start;
5390         int                               rc    = 0;
5391         ENTRY;
5392
5393         if (start == NULL || !(start->ls_flags & LPF_OST_ORPHAN))
5394                 RETURN(0);
5395
5396         if (!lsp->lsp_index_valid)
5397                 RETURN(-EINVAL);
5398
5399         /* If someone is running the LFSCK without orphan handling,
5400          * it will not maintain the object accessing rbtree. So we
5401          * cannot join it for orphan handling. */
5402         if (!llsd->llsd_rbtree_valid)
5403                 RETURN(-EBUSY);
5404
5405         spin_unlock(&lfsck->li_lock);
5406         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5407         spin_lock(&lfsck->li_lock);
5408         if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
5409                 spin_unlock(&lfsck->li_lock);
5410                 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
5411                                                       true);
5412                 if (llst != NULL)
5413                         lfsck_layout_llst_put(llst);
5414                 spin_lock(&lfsck->li_lock);
5415                 rc = -EAGAIN;
5416         }
5417
5418         RETURN(rc);
5419 }
5420
5421 static struct lfsck_operations lfsck_layout_master_ops = {
5422         .lfsck_reset            = lfsck_layout_reset,
5423         .lfsck_fail             = lfsck_layout_fail,
5424         .lfsck_checkpoint       = lfsck_layout_master_checkpoint,
5425         .lfsck_prep             = lfsck_layout_master_prep,
5426         .lfsck_exec_oit         = lfsck_layout_master_exec_oit,
5427         .lfsck_exec_dir         = lfsck_layout_exec_dir,
5428         .lfsck_post             = lfsck_layout_master_post,
5429         .lfsck_dump             = lfsck_layout_dump,
5430         .lfsck_double_scan      = lfsck_layout_master_double_scan,
5431         .lfsck_data_release     = lfsck_layout_master_data_release,
5432         .lfsck_quit             = lfsck_layout_master_quit,
5433         .lfsck_in_notify        = lfsck_layout_master_in_notify,
5434         .lfsck_query            = lfsck_layout_query,
5435 };
5436
5437 static struct lfsck_operations lfsck_layout_slave_ops = {
5438         .lfsck_reset            = lfsck_layout_reset,
5439         .lfsck_fail             = lfsck_layout_fail,
5440         .lfsck_checkpoint       = lfsck_layout_slave_checkpoint,
5441         .lfsck_prep             = lfsck_layout_slave_prep,
5442         .lfsck_exec_oit         = lfsck_layout_slave_exec_oit,
5443         .lfsck_exec_dir         = lfsck_layout_exec_dir,
5444         .lfsck_post             = lfsck_layout_slave_post,
5445         .lfsck_dump             = lfsck_layout_dump,
5446         .lfsck_double_scan      = lfsck_layout_slave_double_scan,
5447         .lfsck_data_release     = lfsck_layout_slave_data_release,
5448         .lfsck_quit             = lfsck_layout_slave_quit,
5449         .lfsck_in_notify        = lfsck_layout_slave_in_notify,
5450         .lfsck_query            = lfsck_layout_query,
5451         .lfsck_join             = lfsck_layout_slave_join,
5452 };
5453
5454 static void lfsck_layout_assistant_fill_pos(const struct lu_env *env,
5455                                             struct lfsck_component *com,
5456                                             struct lfsck_position *pos)
5457 {
5458         struct lfsck_assistant_data     *lad = com->lc_data;
5459         struct lfsck_layout_req         *llr;
5460
5461         if (list_empty(&lad->lad_req_list))
5462                 return;
5463
5464         llr = list_entry(lad->lad_req_list.next,
5465                          struct lfsck_layout_req,
5466                          llr_lar.lar_list);
5467         pos->lp_oit_cookie = llr->llr_lar.lar_parent->lso_oit_cookie - 1;
5468 }
5469
5470 struct lfsck_assistant_operations lfsck_layout_assistant_ops = {
5471         .la_handler_p1          = lfsck_layout_assistant_handler_p1,
5472         .la_handler_p2          = lfsck_layout_assistant_handler_p2,
5473         .la_fill_pos            = lfsck_layout_assistant_fill_pos,
5474         .la_double_scan_result  = lfsck_layout_double_scan_result,
5475         .la_req_fini            = lfsck_layout_assistant_req_fini,
5476         .la_sync_failures       = lfsck_layout_assistant_sync_failures,
5477 };
5478
5479 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
5480 {
5481         struct lfsck_component  *com;
5482         struct lfsck_layout     *lo;
5483         struct dt_object        *root = NULL;
5484         struct dt_object        *obj;
5485         int                      rc;
5486         ENTRY;
5487
5488         OBD_ALLOC_PTR(com);
5489         if (com == NULL)
5490                 RETURN(-ENOMEM);
5491
5492         INIT_LIST_HEAD(&com->lc_link);
5493         INIT_LIST_HEAD(&com->lc_link_dir);
5494         init_rwsem(&com->lc_sem);
5495         atomic_set(&com->lc_ref, 1);
5496         com->lc_lfsck = lfsck;
5497         com->lc_type = LFSCK_TYPE_LAYOUT;
5498         if (lfsck->li_master) {
5499                 com->lc_ops = &lfsck_layout_master_ops;
5500                 com->lc_data = lfsck_assistant_data_init(
5501                                 &lfsck_layout_assistant_ops,
5502                                 LFSCK_LAYOUT);
5503                 if (com->lc_data == NULL)
5504                         GOTO(out, rc = -ENOMEM);
5505         } else {
5506                 struct lfsck_layout_slave_data *llsd;
5507
5508                 com->lc_ops = &lfsck_layout_slave_ops;
5509                 OBD_ALLOC_PTR(llsd);
5510                 if (llsd == NULL)
5511                         GOTO(out, rc = -ENOMEM);
5512
5513                 INIT_LIST_HEAD(&llsd->llsd_seq_list);
5514                 INIT_LIST_HEAD(&llsd->llsd_master_list);
5515                 spin_lock_init(&llsd->llsd_lock);
5516                 llsd->llsd_rb_root = RB_ROOT;
5517                 rwlock_init(&llsd->llsd_rb_lock);
5518                 com->lc_data = llsd;
5519         }
5520         com->lc_file_size = sizeof(*lo);
5521         OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
5522         if (com->lc_file_ram == NULL)
5523                 GOTO(out, rc = -ENOMEM);
5524
5525         OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
5526         if (com->lc_file_disk == NULL)
5527                 GOTO(out, rc = -ENOMEM);
5528
5529         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
5530         if (IS_ERR(root))
5531                 GOTO(out, rc = PTR_ERR(root));
5532
5533         if (unlikely(!dt_try_as_dir(env, root)))
5534                 GOTO(out, rc = -ENOTDIR);
5535
5536         obj = local_file_find_or_create(env, lfsck->li_los, root,
5537                                         LFSCK_LAYOUT,
5538                                         S_IFREG | S_IRUGO | S_IWUSR);
5539         if (IS_ERR(obj))
5540                 GOTO(out, rc = PTR_ERR(obj));
5541
5542         com->lc_obj = obj;
5543         rc = lfsck_layout_load(env, com);
5544         if (rc > 0)
5545                 rc = lfsck_layout_reset(env, com, true);
5546         else if (rc == -ENOENT)
5547                 rc = lfsck_layout_init(env, com);
5548
5549         if (rc != 0)
5550                 GOTO(out, rc);
5551
5552         lo = com->lc_file_ram;
5553         switch (lo->ll_status) {
5554         case LS_INIT:
5555         case LS_COMPLETED:
5556         case LS_FAILED:
5557         case LS_STOPPED:
5558         case LS_PARTIAL:
5559                 spin_lock(&lfsck->li_lock);
5560                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
5561                 spin_unlock(&lfsck->li_lock);
5562                 break;
5563         default:
5564                 CERROR("%s: unknown lfsck_layout status %d\n",
5565                        lfsck_lfsck2name(lfsck), lo->ll_status);
5566                 /* fall through */
5567         case LS_SCANNING_PHASE1:
5568         case LS_SCANNING_PHASE2:
5569                 /* No need to store the status to disk right now.
5570                  * If the system crashed before the status stored,
5571                  * it will be loaded back when next time. */
5572                 lo->ll_status = LS_CRASHED;
5573                 if (!lfsck->li_master)
5574                         lo->ll_flags |= LF_INCOMPLETE;
5575                 /* fall through */
5576         case LS_PAUSED:
5577         case LS_CRASHED:
5578         case LS_CO_FAILED:
5579         case LS_CO_STOPPED:
5580         case LS_CO_PAUSED:
5581                 spin_lock(&lfsck->li_lock);
5582                 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
5583                 spin_unlock(&lfsck->li_lock);
5584                 break;
5585         }
5586
5587         if (lo->ll_flags & LF_CRASHED_LASTID) {
5588                 LASSERT(lfsck->li_out_notify != NULL);
5589
5590                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5591                                      LE_LASTID_REBUILDING);
5592         }
5593
5594         GOTO(out, rc = 0);
5595
5596 out:
5597         if (root != NULL && !IS_ERR(root))
5598                 lfsck_object_put(env, root);
5599
5600         if (rc != 0) {
5601                 lfsck_component_cleanup(env, com);
5602                 CERROR("%s: fail to init layout LFSCK component: rc = %d\n",
5603                        lfsck_lfsck2name(lfsck), rc);
5604         }
5605
5606         return rc;
5607 }
5608
5609 struct lfsck_orphan_it {
5610         struct lfsck_component           *loi_com;
5611         struct lfsck_rbtree_node         *loi_lrn;
5612         struct lfsck_layout_slave_target *loi_llst;
5613         struct lu_fid                     loi_key;
5614         struct lu_orphan_rec              loi_rec;
5615         __u64                             loi_hash;
5616         unsigned int                      loi_over:1;
5617 };
5618
5619 static int lfsck_fid_match_idx(const struct lu_env *env,
5620                                struct lfsck_instance *lfsck,
5621                                const struct lu_fid *fid, int idx)
5622 {
5623         struct seq_server_site  *ss;
5624         struct lu_server_fld    *sf;
5625         struct lu_seq_range     *range = &lfsck_env_info(env)->lti_range;
5626         int                      rc;
5627
5628         /* All abnormal cases will be returned to MDT0. */
5629         if (!fid_is_norm(fid)) {
5630                 if (idx == 0)
5631                         return 1;
5632
5633                 return 0;
5634         }
5635
5636         ss = lfsck_dev_site(lfsck);
5637         if (unlikely(ss == NULL))
5638                 return -ENOTCONN;
5639
5640         sf = ss->ss_server_fld;
5641         LASSERT(sf != NULL);
5642
5643         fld_range_set_any(range);
5644         rc = fld_server_lookup(env, sf, fid_seq(fid), range);
5645         if (rc != 0)
5646                 return rc;
5647
5648         if (!fld_range_is_mdt(range))
5649                 return -EINVAL;
5650
5651         if (range->lsr_index == idx)
5652                 return 1;
5653
5654         return 0;
5655 }
5656
5657 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
5658                                         struct dt_object *obj)
5659 {
5660         struct dt_device        *dev    = lfsck_obj2dev(obj);
5661         struct thandle          *handle;
5662         int                      rc;
5663         ENTRY;
5664
5665         handle = dt_trans_create(env, dev);
5666         if (IS_ERR(handle))
5667                 RETURN_EXIT;
5668
5669         rc = dt_declare_ref_del(env, obj, handle);
5670         if (rc != 0)
5671                 GOTO(stop, rc);
5672
5673         rc = dt_declare_destroy(env, obj, handle);
5674         if (rc != 0)
5675                 GOTO(stop, rc);
5676
5677         rc = dt_trans_start_local(env, dev, handle);
5678         if (rc != 0)
5679                 GOTO(stop, rc);
5680
5681         dt_write_lock(env, obj, 0);
5682         rc = dt_ref_del(env, obj, handle);
5683         if (rc == 0)
5684                 rc = dt_destroy(env, obj, handle);
5685         dt_write_unlock(env, obj);
5686
5687         GOTO(stop, rc);
5688
5689 stop:
5690         dt_trans_stop(env, dev, handle);
5691
5692         CDEBUG(D_LFSCK, "destroy orphan OST-object "DFID": rc = %d\n",
5693                PFID(lfsck_dto2fid(obj)), rc);
5694
5695         RETURN_EXIT;
5696 }
5697
5698 static int lfsck_orphan_index_lookup(const struct lu_env *env,
5699                                      struct dt_object *dt,
5700                                      struct dt_rec *rec,
5701                                      const struct dt_key *key)
5702 {
5703         return -EOPNOTSUPP;
5704 }
5705
5706 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
5707                                              struct dt_object *dt,
5708                                              const struct dt_rec *rec,
5709                                              const struct dt_key *key,
5710                                              struct thandle *handle)
5711 {
5712         return -EOPNOTSUPP;
5713 }
5714
5715 static int lfsck_orphan_index_insert(const struct lu_env *env,
5716                                      struct dt_object *dt,
5717                                      const struct dt_rec *rec,
5718                                      const struct dt_key *key,
5719                                      struct thandle *handle,
5720                                      int ignore_quota)
5721 {
5722         return -EOPNOTSUPP;
5723 }
5724
5725 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
5726                                              struct dt_object *dt,
5727                                              const struct dt_key *key,
5728                                              struct thandle *handle)
5729 {
5730         return -EOPNOTSUPP;
5731 }
5732
5733 static int lfsck_orphan_index_delete(const struct lu_env *env,
5734                                      struct dt_object *dt,
5735                                      const struct dt_key *key,
5736                                      struct thandle *handle)
5737 {
5738         return -EOPNOTSUPP;
5739 }
5740
5741 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
5742                                           struct dt_object *dt,
5743                                           __u32 attr)
5744 {
5745         struct dt_device                *dev    = lu2dt_dev(dt->do_lu.lo_dev);
5746         struct lfsck_instance           *lfsck;
5747         struct lfsck_component          *com    = NULL;
5748         struct lfsck_layout_slave_data  *llsd;
5749         struct lfsck_orphan_it          *it     = NULL;
5750         struct lfsck_layout             *lo;
5751         int                              rc     = 0;
5752         ENTRY;
5753
5754         lfsck = lfsck_instance_find(dev, true, false);
5755         if (unlikely(lfsck == NULL))
5756                 RETURN(ERR_PTR(-ENXIO));
5757
5758         com = lfsck_component_find(lfsck, LFSCK_TYPE_LAYOUT);
5759         if (unlikely(com == NULL))
5760                 GOTO(out, rc = -ENOENT);
5761
5762         lo = com->lc_file_ram;
5763         if (lo->ll_flags & LF_INCOMPLETE)
5764                 GOTO(out, rc = -ESRCH);
5765
5766         llsd = com->lc_data;
5767         if (!llsd->llsd_rbtree_valid)
5768                 GOTO(out, rc = -ESRCH);
5769
5770         OBD_ALLOC_PTR(it);
5771         if (it == NULL)
5772                 GOTO(out, rc = -ENOMEM);
5773
5774         it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
5775         if (it->loi_llst == NULL)
5776                 GOTO(out, rc = -ENXIO);
5777
5778         if (dev->dd_record_fid_accessed) {
5779                 /* The first iteration against the rbtree, scan the whole rbtree
5780                  * to remove the nodes which do NOT need to be handled. */
5781                 write_lock(&llsd->llsd_rb_lock);
5782                 if (dev->dd_record_fid_accessed) {
5783                         struct rb_node                  *node;
5784                         struct rb_node                  *next;
5785                         struct lfsck_rbtree_node        *lrn;
5786
5787                         /* No need to record the fid accessing anymore. */
5788                         dev->dd_record_fid_accessed = 0;
5789
5790                         node = rb_first(&llsd->llsd_rb_root);
5791                         while (node != NULL) {
5792                                 next = rb_next(node);
5793                                 lrn = rb_entry(node, struct lfsck_rbtree_node,
5794                                                lrn_node);
5795                                 if (atomic_read(&lrn->lrn_known_count) <=
5796                                     atomic_read(&lrn->lrn_accessed_count)) {
5797                                         rb_erase(node, &llsd->llsd_rb_root);
5798                                         lfsck_rbtree_free(lrn);
5799                                 }
5800                                 node = next;
5801                         }
5802                 }
5803                 write_unlock(&llsd->llsd_rb_lock);
5804         }
5805
5806         /* read lock the rbtree when init, and unlock when fini */
5807         read_lock(&llsd->llsd_rb_lock);
5808         it->loi_com = com;
5809         com = NULL;
5810
5811         GOTO(out, rc = 0);
5812
5813 out:
5814         if (com != NULL)
5815                 lfsck_component_put(env, com);
5816
5817         CDEBUG(D_LFSCK, "%s: init the orphan iteration: rc = %d\n",
5818                lfsck_lfsck2name(lfsck), rc);
5819
5820         lfsck_instance_put(env, lfsck);
5821         if (rc != 0) {
5822                 if (it != NULL)
5823                         OBD_FREE_PTR(it);
5824
5825                 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
5826         }
5827
5828         return (struct dt_it *)it;
5829 }
5830
5831 static void lfsck_orphan_it_fini(const struct lu_env *env,
5832                                  struct dt_it *di)
5833 {
5834         struct lfsck_orphan_it           *it    = (struct lfsck_orphan_it *)di;
5835         struct lfsck_component           *com   = it->loi_com;
5836         struct lfsck_layout_slave_data   *llsd;
5837         struct lfsck_layout_slave_target *llst;
5838
5839         if (com != NULL) {
5840                 CDEBUG(D_LFSCK, "%s: fini the orphan iteration\n",
5841                        lfsck_lfsck2name(com->lc_lfsck));
5842
5843                 llsd = com->lc_data;
5844                 read_unlock(&llsd->llsd_rb_lock);
5845                 llst = it->loi_llst;
5846                 LASSERT(llst != NULL);
5847
5848                 /* Save the key and hash for iterate next. */
5849                 llst->llst_fid = it->loi_key;
5850                 llst->llst_hash = it->loi_hash;
5851                 lfsck_layout_llst_put(llst);
5852                 lfsck_component_put(env, com);
5853         }
5854         OBD_FREE_PTR(it);
5855 }
5856
5857 /**
5858  * \retval       +1: the iteration finished
5859  * \retval        0: on success, not finished
5860  * \retval      -ve: on error
5861  */
5862 static int lfsck_orphan_it_next(const struct lu_env *env,
5863                                 struct dt_it *di)
5864 {
5865         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5866         struct filter_fid_old           *pfid   = &info->lti_old_pfid;
5867         struct lu_attr                  *la     = &info->lti_la;
5868         struct lfsck_orphan_it          *it     = (struct lfsck_orphan_it *)di;
5869         struct lu_fid                   *key    = &it->loi_key;
5870         struct lu_orphan_rec            *rec    = &it->loi_rec;
5871         struct lfsck_component          *com    = it->loi_com;
5872         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5873         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
5874         struct dt_object                *obj;
5875         struct lfsck_rbtree_node        *lrn;
5876         int                              pos;
5877         int                              rc;
5878         __u32                            save;
5879         __u32                            idx    = it->loi_llst->llst_index;
5880         bool                             exact  = false;
5881         ENTRY;
5882
5883         if (it->loi_over)
5884                 RETURN(1);
5885
5886 again0:
5887         lrn = it->loi_lrn;
5888         if (lrn == NULL) {
5889                 lrn = lfsck_rbtree_search(llsd, key, &exact);
5890                 if (lrn == NULL) {
5891                         it->loi_over = 1;
5892                         RETURN(1);
5893                 }
5894
5895                 it->loi_lrn = lrn;
5896                 if (!exact) {
5897                         key->f_seq = lrn->lrn_seq;
5898                         key->f_oid = lrn->lrn_first_oid;
5899                         key->f_ver = 0;
5900                 }
5901         } else {
5902                 key->f_oid++;
5903                 if (unlikely(key->f_oid == 0)) {
5904                         key->f_seq++;
5905                         it->loi_lrn = NULL;
5906                         goto again0;
5907                 }
5908
5909                 if (key->f_oid >=
5910                     lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
5911                         it->loi_lrn = NULL;
5912                         goto again0;
5913                 }
5914         }
5915
5916         if (unlikely(atomic_read(&lrn->lrn_known_count) <=
5917                      atomic_read(&lrn->lrn_accessed_count))) {
5918                 struct rb_node *next = rb_next(&lrn->lrn_node);
5919
5920                 while (next != NULL) {
5921                         lrn = rb_entry(next, struct lfsck_rbtree_node,
5922                                        lrn_node);
5923                         if (atomic_read(&lrn->lrn_known_count) >
5924                             atomic_read(&lrn->lrn_accessed_count))
5925                                 break;
5926                         next = rb_next(next);
5927                 }
5928
5929                 if (next == NULL) {
5930                         it->loi_over = 1;
5931                         RETURN(1);
5932                 }
5933
5934                 it->loi_lrn = lrn;
5935                 key->f_seq = lrn->lrn_seq;
5936                 key->f_oid = lrn->lrn_first_oid;
5937                 key->f_ver = 0;
5938         }
5939
5940         pos = key->f_oid - lrn->lrn_first_oid;
5941
5942 again1:
5943         pos = find_next_bit(lrn->lrn_known_bitmap,
5944                             LFSCK_RBTREE_BITMAP_WIDTH, pos);
5945         if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
5946                 key->f_oid = lrn->lrn_first_oid + pos;
5947                 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
5948                         key->f_seq++;
5949                         key->f_oid = 0;
5950                 }
5951                 it->loi_lrn = NULL;
5952                 goto again0;
5953         }
5954
5955         if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
5956                 pos++;
5957                 goto again1;
5958         }
5959
5960         key->f_oid = lrn->lrn_first_oid + pos;
5961         obj = lfsck_object_find_bottom(env, lfsck, key);
5962         if (IS_ERR(obj)) {
5963                 rc = PTR_ERR(obj);
5964                 if (rc == -ENOENT) {
5965                         pos++;
5966                         goto again1;
5967                 }
5968                 RETURN(rc);
5969         }
5970
5971         dt_read_lock(env, obj, 0);
5972         if (dt_object_exists(obj) == 0 ||
5973             lfsck_is_dead_obj(obj)) {
5974                 dt_read_unlock(env, obj);
5975                 lfsck_object_put(env, obj);
5976                 pos++;
5977                 goto again1;
5978         }
5979
5980         rc = dt_attr_get(env, obj, la);
5981         if (rc != 0)
5982                 GOTO(out, rc);
5983
5984         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, pfid, sizeof(*pfid)),
5985                           XATTR_NAME_FID);
5986         if (rc == -ENODATA) {
5987                 /* For the pre-created OST-object, update the bitmap to avoid
5988                  * others LFSCK (second phase) iteration to touch it again. */
5989                 if (la->la_ctime == 0) {
5990                         if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
5991                                 atomic_inc(&lrn->lrn_accessed_count);
5992
5993                         /* For the race between repairing dangling referenced
5994                          * MDT-object and unlink the file, it may left orphan
5995                          * OST-object there. Destroy it now! */
5996                         if (unlikely(!(la->la_mode & S_ISUID))) {
5997                                 dt_read_unlock(env, obj);
5998                                 lfsck_layout_destroy_orphan(env, obj);
5999                                 lfsck_object_put(env, obj);
6000                                 pos++;
6001                                 goto again1;
6002                         }
6003                 } else if (idx == 0) {
6004                         /* If the orphan OST-object has no parent information,
6005                          * regard it as referenced by the MDT-object on MDT0. */
6006                         fid_zero(&rec->lor_fid);
6007                         rec->lor_uid = la->la_uid;
6008                         rec->lor_gid = la->la_gid;
6009                         GOTO(out, rc = 0);
6010                 }
6011
6012                 dt_read_unlock(env, obj);
6013                 lfsck_object_put(env, obj);
6014                 pos++;
6015                 goto again1;
6016         }
6017
6018         if (rc < 0)
6019                 GOTO(out, rc);
6020
6021         if (rc != sizeof(struct filter_fid) &&
6022             rc != sizeof(struct filter_fid_old))
6023                 GOTO(out, rc = -EINVAL);
6024
6025         fid_le_to_cpu(&rec->lor_fid, &pfid->ff_parent);
6026         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
6027          * MDT-object's FID::f_ver, instead it is the OST-object index in its
6028          * parent MDT-object's layout EA. */
6029         save = rec->lor_fid.f_stripe_idx;
6030         rec->lor_fid.f_ver = 0;
6031         rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_fid, idx);
6032         /* If the orphan OST-object does not claim the MDT, then next.
6033          *
6034          * If we do not know whether it matches or not, then return it
6035          * to the MDT for further check. */
6036         if (rc == 0) {
6037                 dt_read_unlock(env, obj);
6038                 lfsck_object_put(env, obj);
6039                 pos++;
6040                 goto again1;
6041         }
6042
6043         rec->lor_fid.f_stripe_idx = save;
6044         rec->lor_uid = la->la_uid;
6045         rec->lor_gid = la->la_gid;
6046
6047         CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u\n",
6048                lfsck_lfsck2name(com->lc_lfsck), PFID(key), PFID(&rec->lor_fid),
6049                rec->lor_uid, rec->lor_gid);
6050
6051         GOTO(out, rc = 0);
6052
6053 out:
6054         dt_read_unlock(env, obj);
6055         lfsck_object_put(env, obj);
6056         if (rc == 0)
6057                 it->loi_hash++;
6058
6059         return rc;
6060 }
6061
6062 /**
6063  * \retval       +1: locate to the exactly position
6064  * \retval        0: cannot locate to the exactly position,
6065  *                   call next() to move to a valid position.
6066  * \retval      -ve: on error
6067  */
6068 static int lfsck_orphan_it_get(const struct lu_env *env,
6069                                struct dt_it *di,
6070                                const struct dt_key *key)
6071 {
6072         struct lfsck_orphan_it  *it   = (struct lfsck_orphan_it *)di;
6073         int                      rc;
6074
6075         it->loi_key = *(struct lu_fid *)key;
6076         rc = lfsck_orphan_it_next(env, di);
6077         if (rc == 1)
6078                 return 0;
6079
6080         if (rc == 0)
6081                 return 1;
6082
6083         return rc;
6084 }
6085
6086 static void lfsck_orphan_it_put(const struct lu_env *env,
6087                                 struct dt_it *di)
6088 {
6089 }
6090
6091 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
6092                                           const struct dt_it *di)
6093 {
6094         struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6095
6096         return (struct dt_key *)&it->loi_key;
6097 }
6098
6099 static int lfsck_orphan_it_key_size(const struct lu_env *env,
6100                                     const struct dt_it *di)
6101 {
6102         return sizeof(struct lu_fid);
6103 }
6104
6105 static int lfsck_orphan_it_rec(const struct lu_env *env,
6106                                const struct dt_it *di,
6107                                struct dt_rec *rec,
6108                                __u32 attr)
6109 {
6110         struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6111
6112         *(struct lu_orphan_rec *)rec = it->loi_rec;
6113
6114         return 0;
6115 }
6116
6117 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
6118                                    const struct dt_it *di)
6119 {
6120         struct lfsck_orphan_it  *it   = (struct lfsck_orphan_it *)di;
6121
6122         return it->loi_hash;
6123 }
6124
6125 /**
6126  * \retval       +1: locate to the exactly position
6127  * \retval        0: cannot locate to the exactly position,
6128  *                   call next() to move to a valid position.
6129  * \retval      -ve: on error
6130  */
6131 static int lfsck_orphan_it_load(const struct lu_env *env,
6132                                 const struct dt_it *di,
6133                                 __u64 hash)
6134 {
6135         struct lfsck_orphan_it           *it   = (struct lfsck_orphan_it *)di;
6136         struct lfsck_layout_slave_target *llst = it->loi_llst;
6137         int                               rc;
6138
6139         LASSERT(llst != NULL);
6140
6141         if (hash != llst->llst_hash) {
6142                 CDEBUG(D_LFSCK, "%s: the given hash "LPU64" for orphan "
6143                        "iteration does not match the one when fini "
6144                        LPU64", to be reset.\n",
6145                        lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
6146                        llst->llst_hash);
6147                 fid_zero(&llst->llst_fid);
6148                 llst->llst_hash = 0;
6149         }
6150
6151         it->loi_key = llst->llst_fid;
6152         it->loi_hash = llst->llst_hash;
6153         rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
6154         if (rc == 1)
6155                 return 0;
6156
6157         if (rc == 0)
6158                 return 1;
6159
6160         return rc;
6161 }
6162
6163 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
6164                                    const struct dt_it *di,
6165                                    void *key_rec)
6166 {
6167         return 0;
6168 }
6169
6170 const struct dt_index_operations lfsck_orphan_index_ops = {
6171         .dio_lookup             = lfsck_orphan_index_lookup,
6172         .dio_declare_insert     = lfsck_orphan_index_declare_insert,
6173         .dio_insert             = lfsck_orphan_index_insert,
6174         .dio_declare_delete     = lfsck_orphan_index_declare_delete,
6175         .dio_delete             = lfsck_orphan_index_delete,
6176         .dio_it = {
6177                 .init           = lfsck_orphan_it_init,
6178                 .fini           = lfsck_orphan_it_fini,
6179                 .get            = lfsck_orphan_it_get,
6180                 .put            = lfsck_orphan_it_put,
6181                 .next           = lfsck_orphan_it_next,
6182                 .key            = lfsck_orphan_it_key,
6183                 .key_size       = lfsck_orphan_it_key_size,
6184                 .rec            = lfsck_orphan_it_rec,
6185                 .store          = lfsck_orphan_it_store,
6186                 .load           = lfsck_orphan_it_load,
6187                 .key_rec        = lfsck_orphan_it_key_rec,
6188         }
6189 };