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, 2015, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_layout.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 #ifndef EXPORT_SYMTAB
  32 # define EXPORT_SYMTAB
  33 #endif
  34 #define DEBUG_SUBSYSTEM S_LFSCK
  35
  36 #include <linux/bitops.h>
  37 #include <linux/rbtree.h>
  38
  39 #include <lustre/lustre_idl.h>
  40 #include <lu_object.h>
  41 #include <dt_object.h>
  42 #include <lustre_fid.h>
  43 #include <lustre_lib.h>
  44 #include <lustre_net.h>
  45 #include <lustre/lustre_user.h>
  46 #include <md_object.h>
  47 #include <obd_class.h>
  48
  49 #include "lfsck_internal.h"
  50
  51 #define LFSCK_LAYOUT_MAGIC_V1           0xB173AE14
  52 #define LFSCK_LAYOUT_MAGIC_V2           0xB1734D76
  53
  54 #define LFSCK_LAYOUT_MAGIC              LFSCK_LAYOUT_MAGIC_V2
  55
  56 struct lfsck_layout_seq {
  57         struct list_head         lls_list;
  58         __u64                    lls_seq;
  59         __u64                    lls_lastid;
  60         __u64                    lls_lastid_known;
  61         struct dt_object        *lls_lastid_obj;
  62         unsigned int             lls_dirty:1;
  63 };
  64
  65 struct lfsck_layout_slave_target {
  66         /* link into lfsck_layout_slave_data::llsd_master_list. */
  67         struct list_head        llst_list;
  68         /* The position for next record in the rbtree for iteration. */
  69         struct lu_fid           llst_fid;
  70         /* Dummy hash for iteration against the rbtree. */
  71         __u64                   llst_hash;
  72         __u64                   llst_gen;
  73         atomic_t                llst_ref;
  74         __u32                   llst_index;
  75         /* How many times we have failed to get the master status. */
  76         int                     llst_failures;
  77 };
  78
  79 struct lfsck_layout_slave_data {
  80         /* list for lfsck_layout_seq */
  81         struct list_head         llsd_seq_list;
  82
  83         /* list for the masters involve layout verification. */
  84         struct list_head         llsd_master_list;
  85         spinlock_t               llsd_lock;
  86         __u64                    llsd_touch_gen;
  87         struct dt_object        *llsd_rb_obj;
  88         struct rb_root           llsd_rb_root;
  89         rwlock_t                 llsd_rb_lock;
  90         unsigned int             llsd_rbtree_valid:1;
  91 };
  92
  93 struct lfsck_layout_slave_async_args {
  94         struct obd_export                *llsaa_exp;
  95         struct lfsck_component           *llsaa_com;
  96         struct lfsck_layout_slave_target *llsaa_llst;
  97 };
  98
  99 static inline void
 100 lfsck_layout_llst_put(struct lfsck_layout_slave_target *llst)
 101 {
 102         if (atomic_dec_and_test(&llst->llst_ref)) {
 103                 LASSERT(list_empty(&llst->llst_list));
 104
 105                 OBD_FREE_PTR(llst);
 106         }
 107 }
 108
 109 static inline int
 110 lfsck_layout_llst_add(struct lfsck_layout_slave_data *llsd, __u32 index)
 111 {
 112         struct lfsck_layout_slave_target *llst;
 113         struct lfsck_layout_slave_target *tmp;
 114         int                               rc   = 0;
 115
 116         OBD_ALLOC_PTR(llst);
 117         if (llst == NULL)
 118                 return -ENOMEM;
 119
 120         INIT_LIST_HEAD(&llst->llst_list);
 121         llst->llst_gen = 0;
 122         llst->llst_index = index;
 123         atomic_set(&llst->llst_ref, 1);
 124
 125         spin_lock(&llsd->llsd_lock);
 126         list_for_each_entry(tmp, &llsd->llsd_master_list, llst_list) {
 127                 if (tmp->llst_index == index) {
 128                         rc = -EALREADY;
 129                         break;
 130                 }
 131         }
 132         if (rc == 0)
 133                 list_add_tail(&llst->llst_list, &llsd->llsd_master_list);
 134         spin_unlock(&llsd->llsd_lock);
 135
 136         if (rc != 0)
 137                 OBD_FREE_PTR(llst);
 138
 139         return rc;
 140 }
 141
 142 static inline void
 143 lfsck_layout_llst_del(struct lfsck_layout_slave_data *llsd,
 144                       struct lfsck_layout_slave_target *llst)
 145 {
 146         bool del = false;
 147
 148         spin_lock(&llsd->llsd_lock);
 149         if (!list_empty(&llst->llst_list)) {
 150                 list_del_init(&llst->llst_list);
 151                 del = true;
 152         }
 153         spin_unlock(&llsd->llsd_lock);
 154
 155         if (del)
 156                 lfsck_layout_llst_put(llst);
 157 }
 158
 159 static inline struct lfsck_layout_slave_target *
 160 lfsck_layout_llst_find_and_del(struct lfsck_layout_slave_data *llsd,
 161                                __u32 index, bool unlink)
 162 {
 163         struct lfsck_layout_slave_target *llst;
 164
 165         spin_lock(&llsd->llsd_lock);
 166         list_for_each_entry(llst, &llsd->llsd_master_list, llst_list) {
 167                 if (llst->llst_index == index) {
 168                         if (unlink)
 169                                 list_del_init(&llst->llst_list);
 170                         else
 171                                 atomic_inc(&llst->llst_ref);
 172                         spin_unlock(&llsd->llsd_lock);
 173
 174                         return llst;
 175                 }
 176         }
 177         spin_unlock(&llsd->llsd_lock);
 178
 179         return NULL;
 180 }
 181
 182 static struct lfsck_layout_req *
 183 lfsck_layout_assistant_req_init(struct lfsck_assistant_object *lso,
 184                                 struct dt_object *child, __u32 ost_idx,
 185                                 __u32 lov_idx)
 186 {
 187         struct lfsck_layout_req *llr;
 188
 189         OBD_ALLOC_PTR(llr);
 190         if (llr == NULL)
 191                 return ERR_PTR(-ENOMEM);
 192
 193         INIT_LIST_HEAD(&llr->llr_lar.lar_list);
 194         llr->llr_lar.lar_parent = lfsck_assistant_object_get(lso);
 195         llr->llr_child = child;
 196         llr->llr_ost_idx = ost_idx;
 197         llr->llr_lov_idx = lov_idx;
 198
 199         return llr;
 200 }
 201
 202 static void lfsck_layout_assistant_req_fini(const struct lu_env *env,
 203                                             struct lfsck_assistant_req *lar)
 204 {
 205         struct lfsck_layout_req *llr =
 206                         container_of0(lar, struct lfsck_layout_req, llr_lar);
 207
 208         lfsck_object_put(env, llr->llr_child);
 209         lfsck_assistant_object_put(env, lar->lar_parent);
 210         OBD_FREE_PTR(llr);
 211 }
 212
 213 static int
 214 lfsck_layout_assistant_sync_failures_interpret(const struct lu_env *env,
 215                                                struct ptlrpc_request *req,
 216                                                void *args, int rc)
 217 {
 218         if (rc == 0) {
 219                 struct lfsck_async_interpret_args *laia = args;
 220                 struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
 221
 222                 ltd->ltd_synced_failures = 1;
 223                 atomic_dec(laia->laia_count);
 224         }
 225
 226         return 0;
 227 }
 228
 229 /**
 230  * Notify remote LFSCK instances about former failures.
 231  *
 232  * The local LFSCK instance has recorded which OSTs have ever failed to respond
 233  * some LFSCK verification requests (maybe because of network issues or the OST
 234  * itself trouble). During the respond gap, the OST may missed some OST-objects
 235  * verification, then the OST cannot know whether related OST-objects have been
 236  * referenced by related MDT-objects or not, then in the second-stage scanning,
 237  * these OST-objects will be regarded as orphan, if the OST-object contains bad
 238  * parent FID for back reference, then it will misguide the LFSCK to make wrong
 239  * fixing for the fake orphan.
 240  *
 241  * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
 242  * it will scan the bitmap for the ever failed OSTs, and notify them that they
 243  * have ever missed some OST-object verification and should skip the handling
 244  * for orphan OST-objects on all MDTs that are in the layout LFSCK.
 245  *
 246  * \param[in] env       pointer to the thread context
 247  * \param[in] com       pointer to the lfsck component
 248  * \param[in] lr        pointer to the lfsck request
 249  */
 250 static void lfsck_layout_assistant_sync_failures(const struct lu_env *env,
 251                                                  struct lfsck_component *com,
 252                                                  struct lfsck_request *lr)
 253 {
 254         struct lfsck_async_interpret_args *laia  =
 255                                 &lfsck_env_info(env)->lti_laia2;
 256         struct lfsck_assistant_data       *lad   = com->lc_data;
 257         struct lfsck_layout               *lo    = com->lc_file_ram;
 258         struct lfsck_instance             *lfsck = com->lc_lfsck;
 259         struct lfsck_tgt_descs            *ltds  = &lfsck->li_ost_descs;
 260         struct lfsck_tgt_desc             *ltd;
 261         struct ptlrpc_request_set         *set;
 262         atomic_t                           count;
 263         __u32                              idx;
 264         int                                rc    = 0;
 265         ENTRY;
 266
 267         if (!lad->lad_incomplete || lo->ll_flags & LF_INCOMPLETE)
 268                 RETURN_EXIT;
 269
 270         /* If the MDT has ever failed to verfiy some OST-objects,
 271          * then sync failures with them firstly. */
 272         lr->lr_flags2 = lo->ll_flags | LF_INCOMPLETE;
 273
 274         atomic_set(&count, 0);
 275         memset(laia, 0, sizeof(*laia));
 276         laia->laia_count = &count;
 277         set = ptlrpc_prep_set();
 278         if (set == NULL)
 279                 GOTO(out, rc = -ENOMEM);
 280
 281         down_read(&ltds->ltd_rw_sem);
 282         cfs_foreach_bit(lad->lad_bitmap, idx) {
 283                 ltd = lfsck_ltd2tgt(ltds, idx);
 284                 LASSERT(ltd != NULL);
 285
 286                 laia->laia_ltd = ltd;
 287                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
 288                                 lfsck_layout_assistant_sync_failures_interpret,
 289                                 laia, LFSCK_NOTIFY);
 290                 if (rc != 0) {
 291                         CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
 292                                "notify target %x for %s phase1 done: "
 293                                "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
 294                                ltd->ltd_index, lad->lad_name, rc);
 295
 296                         break;
 297                 }
 298
 299                 atomic_inc(&count);
 300         }
 301         up_read(&ltds->ltd_rw_sem);
 302
 303         if (rc == 0 && atomic_read(&count) > 0)
 304                 rc = ptlrpc_set_wait(set);
 305
 306         ptlrpc_set_destroy(set);
 307
 308         if (rc == 0 && atomic_read(&count) > 0)
 309                 rc = -EINVAL;
 310
 311         GOTO(out, rc);
 312
 313 out:
 314         if (rc != 0)
 315                 /* If failed to sync failures with the OSTs, then have to
 316                  * mark the whole LFSCK as LF_INCOMPLETE to skip the whole
 317                  * subsequent orphan OST-object handling. */
 318                 lo->ll_flags |= LF_INCOMPLETE;
 319
 320         lr->lr_flags2 = lo->ll_flags;
 321 }
 322
 323 static int lfsck_layout_get_lovea(const struct lu_env *env,
 324                                   struct dt_object *obj, struct lu_buf *buf)
 325 {
 326         int rc;
 327
 328 again:
 329         rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LOV);
 330         if (rc == -ERANGE) {
 331                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV);
 332                 if (rc <= 0)
 333                         return rc;
 334
 335                 lu_buf_realloc(buf, rc);
 336                 if (buf->lb_buf == NULL)
 337                         return -ENOMEM;
 338
 339                 goto again;
 340         }
 341
 342         if (rc == -ENODATA)
 343                 rc = 0;
 344
 345         if (rc <= 0)
 346                 return rc;
 347
 348         if (unlikely(buf->lb_buf == NULL)) {
 349                 lu_buf_alloc(buf, rc);
 350                 if (buf->lb_buf == NULL)
 351                         return -ENOMEM;
 352
 353                 goto again;
 354         }
 355
 356         return rc;
 357 }
 358
 359 static int lfsck_layout_verify_header(struct lov_mds_md_v1 *lmm)
 360 {
 361         __u32 magic;
 362         __u32 pattern;
 363
 364         magic = le32_to_cpu(lmm->lmm_magic);
 365         /* If magic crashed, keep it there. Sometime later, during OST-object
 366          * orphan handling, if some OST-object(s) back-point to it, it can be
 367          * verified and repaired. */
 368         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) {
 369                 struct ost_id   oi;
 370                 int             rc;
 371
 372                 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
 373                 if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
 374                         rc = -EOPNOTSUPP;
 375                 else
 376                         rc = -EINVAL;
 377
 378                 CDEBUG(D_LFSCK, "%s LOV EA magic %u on "DOSTID"\n",
 379                        rc == -EINVAL ? "Unknown" : "Unsupported",
 380                        magic, POSTID(&oi));
 381
 382                 return rc;
 383         }
 384
 385         pattern = le32_to_cpu(lmm->lmm_pattern);
 386         /* XXX: currently, we only support LOV_PATTERN_RAID0. */
 387         if (lov_pattern(pattern) != LOV_PATTERN_RAID0) {
 388                 struct ost_id oi;
 389
 390                 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
 391                 CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u on "DOSTID"\n",
 392                        pattern, POSTID(&oi));
 393
 394                 return -EOPNOTSUPP;
 395         }
 396
 397         return 0;
 398 }
 399
 400 #define LFSCK_RBTREE_BITMAP_SIZE        PAGE_CACHE_SIZE
 401 #define LFSCK_RBTREE_BITMAP_WIDTH       (LFSCK_RBTREE_BITMAP_SIZE << 3)
 402 #define LFSCK_RBTREE_BITMAP_MASK        (LFSCK_RBTREE_BITMAP_WIDTH - 1)
 403
 404 struct lfsck_rbtree_node {
 405         struct rb_node   lrn_node;
 406         __u64            lrn_seq;
 407         __u32            lrn_first_oid;
 408         atomic_t         lrn_known_count;
 409         atomic_t         lrn_accessed_count;
 410         void            *lrn_known_bitmap;
 411         void            *lrn_accessed_bitmap;
 412 };
 413
 414 static inline int lfsck_rbtree_cmp(struct lfsck_rbtree_node *lrn,
 415                                    __u64 seq, __u32 oid)
 416 {
 417         if (seq < lrn->lrn_seq)
 418                 return -1;
 419
 420         if (seq > lrn->lrn_seq)
 421                 return 1;
 422
 423         if (oid < lrn->lrn_first_oid)
 424                 return -1;
 425
 426         if (oid - lrn->lrn_first_oid >= LFSCK_RBTREE_BITMAP_WIDTH)
 427                 return 1;
 428
 429         return 0;
 430 }
 431
 432 /* The caller should hold llsd->llsd_rb_lock. */
 433 static struct lfsck_rbtree_node *
 434 lfsck_rbtree_search(struct lfsck_layout_slave_data *llsd,
 435                     const struct lu_fid *fid, bool *exact)
 436 {
 437         struct rb_node           *node  = llsd->llsd_rb_root.rb_node;
 438         struct rb_node           *prev  = NULL;
 439         struct lfsck_rbtree_node *lrn   = NULL;
 440         int                       rc    = 0;
 441
 442         if (exact != NULL)
 443                 *exact = true;
 444
 445         while (node != NULL) {
 446                 prev = node;
 447                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 448                 rc = lfsck_rbtree_cmp(lrn, fid_seq(fid), fid_oid(fid));
 449                 if (rc < 0)
 450                         node = node->rb_left;
 451                 else if (rc > 0)
 452                         node = node->rb_right;
 453                 else
 454                         return lrn;
 455         }
 456
 457         if (exact == NULL)
 458                 return NULL;
 459
 460         /* If there is no exactly matched one, then to the next valid one. */
 461         *exact = false;
 462
 463         /* The rbtree is empty. */
 464         if (rc == 0)
 465                 return NULL;
 466
 467         if (rc < 0)
 468                 return lrn;
 469
 470         node = rb_next(prev);
 471
 472         /* The end of the rbtree. */
 473         if (node == NULL)
 474                 return NULL;
 475
 476         lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 477
 478         return lrn;
 479 }
 480
 481 static struct lfsck_rbtree_node *lfsck_rbtree_new(const struct lu_env *env,
 482                                                   const struct lu_fid *fid)
 483 {
 484         struct lfsck_rbtree_node *lrn;
 485
 486         OBD_ALLOC_PTR(lrn);
 487         if (lrn == NULL)
 488                 return ERR_PTR(-ENOMEM);
 489
 490         OBD_ALLOC(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 491         if (lrn->lrn_known_bitmap == NULL) {
 492                 OBD_FREE_PTR(lrn);
 493
 494                 return ERR_PTR(-ENOMEM);
 495         }
 496
 497         OBD_ALLOC(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 498         if (lrn->lrn_accessed_bitmap == NULL) {
 499                 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 500                 OBD_FREE_PTR(lrn);
 501
 502                 return ERR_PTR(-ENOMEM);
 503         }
 504
 505         RB_CLEAR_NODE(&lrn->lrn_node);
 506         lrn->lrn_seq = fid_seq(fid);
 507         lrn->lrn_first_oid = fid_oid(fid) & ~LFSCK_RBTREE_BITMAP_MASK;
 508         atomic_set(&lrn->lrn_known_count, 0);
 509         atomic_set(&lrn->lrn_accessed_count, 0);
 510
 511         return lrn;
 512 }
 513
 514 static void lfsck_rbtree_free(struct lfsck_rbtree_node *lrn)
 515 {
 516         OBD_FREE(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 517         OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 518         OBD_FREE_PTR(lrn);
 519 }
 520
 521 /* The caller should hold lock. */
 522 static struct lfsck_rbtree_node *
 523 lfsck_rbtree_insert(struct lfsck_layout_slave_data *llsd,
 524                     struct lfsck_rbtree_node *lrn)
 525 {
 526         struct rb_node           **pos    = &llsd->llsd_rb_root.rb_node;
 527         struct rb_node            *parent = NULL;
 528         struct lfsck_rbtree_node  *tmp;
 529         int                        rc;
 530
 531         while (*pos != NULL) {
 532                 parent = *pos;
 533                 tmp = rb_entry(parent, struct lfsck_rbtree_node, lrn_node);
 534                 rc = lfsck_rbtree_cmp(tmp, lrn->lrn_seq, lrn->lrn_first_oid);
 535                 if (rc < 0)
 536                         pos = &(*pos)->rb_left;
 537                 else if (rc > 0)
 538                         pos = &(*pos)->rb_right;
 539                 else
 540                         return tmp;
 541         }
 542
 543         rb_link_node(&lrn->lrn_node, parent, pos);
 544         rb_insert_color(&lrn->lrn_node, &llsd->llsd_rb_root);
 545
 546         return lrn;
 547 }
 548
 549 extern const struct dt_index_operations lfsck_orphan_index_ops;
 550
 551 static int lfsck_rbtree_setup(const struct lu_env *env,
 552                               struct lfsck_component *com)
 553 {
 554         struct lu_fid                   *fid    = &lfsck_env_info(env)->lti_fid;
 555         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 556         struct dt_device                *dev    = lfsck->li_bottom;
 557         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
 558         struct dt_object                *obj;
 559
 560         fid->f_seq = FID_SEQ_LAYOUT_RBTREE;
 561         fid->f_oid = lfsck_dev_idx(lfsck);
 562         fid->f_ver = 0;
 563         obj = dt_locate(env, dev, fid);
 564         if (IS_ERR(obj))
 565                 RETURN(PTR_ERR(obj));
 566
 567         /* Generate an in-RAM object to stand for the layout rbtree.
 568          * Scanning the layout rbtree will be via the iteration over
 569          * the object. In the future, the rbtree may be written onto
 570          * disk with the object.
 571          *
 572          * Mark the object to be as exist. */
 573         obj->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 574         obj->do_index_ops = &lfsck_orphan_index_ops;
 575         llsd->llsd_rb_obj = obj;
 576         llsd->llsd_rbtree_valid = 1;
 577         dev->dd_record_fid_accessed = 1;
 578
 579         CDEBUG(D_LFSCK, "%s: layout LFSCK init OST-objects accessing bitmap\n",
 580                lfsck_lfsck2name(lfsck));
 581
 582         return 0;
 583 }
 584
 585 static void lfsck_rbtree_cleanup(const struct lu_env *env,
 586                                  struct lfsck_component *com)
 587 {
 588         struct lfsck_instance           *lfsck = com->lc_lfsck;
 589         struct lfsck_layout_slave_data  *llsd  = com->lc_data;
 590         struct rb_node                  *node  = rb_first(&llsd->llsd_rb_root);
 591         struct rb_node                  *next;
 592         struct lfsck_rbtree_node        *lrn;
 593
 594         lfsck->li_bottom->dd_record_fid_accessed = 0;
 595         /* Invalid the rbtree, then no others will use it. */
 596         write_lock(&llsd->llsd_rb_lock);
 597         llsd->llsd_rbtree_valid = 0;
 598         write_unlock(&llsd->llsd_rb_lock);
 599
 600         while (node != NULL) {
 601                 next = rb_next(node);
 602                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 603                 rb_erase(node, &llsd->llsd_rb_root);
 604                 lfsck_rbtree_free(lrn);
 605                 node = next;
 606         }
 607
 608         if (llsd->llsd_rb_obj != NULL) {
 609                 lfsck_object_put(env, llsd->llsd_rb_obj);
 610                 llsd->llsd_rb_obj = NULL;
 611         }
 612
 613         CDEBUG(D_LFSCK, "%s: layout LFSCK fini OST-objects accessing bitmap\n",
 614                lfsck_lfsck2name(lfsck));
 615 }
 616
 617 static void lfsck_rbtree_update_bitmap(const struct lu_env *env,
 618                                        struct lfsck_component *com,
 619                                        const struct lu_fid *fid,
 620                                        bool accessed)
 621 {
 622         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
 623         struct lfsck_rbtree_node        *lrn;
 624         bool                             insert = false;
 625         int                              idx;
 626         int                              rc     = 0;
 627         ENTRY;
 628
 629         if (unlikely(!fid_is_sane(fid) || fid_is_last_id(fid)))
 630                 RETURN_EXIT;
 631
 632         if (!fid_is_idif(fid) && !fid_is_norm(fid))
 633                 RETURN_EXIT;
 634
 635         read_lock(&llsd->llsd_rb_lock);
 636         if (!llsd->llsd_rbtree_valid)
 637                 GOTO(unlock, rc = 0);
 638
 639         lrn = lfsck_rbtree_search(llsd, fid, NULL);
 640         if (lrn == NULL) {
 641                 struct lfsck_rbtree_node *tmp;
 642
 643                 LASSERT(!insert);
 644
 645                 read_unlock(&llsd->llsd_rb_lock);
 646                 tmp = lfsck_rbtree_new(env, fid);
 647                 if (IS_ERR(tmp))
 648                         GOTO(out, rc = PTR_ERR(tmp));
 649
 650                 insert = true;
 651                 write_lock(&llsd->llsd_rb_lock);
 652                 if (!llsd->llsd_rbtree_valid) {
 653                         lfsck_rbtree_free(tmp);
 654                         GOTO(unlock, rc = 0);
 655                 }
 656
 657                 lrn = lfsck_rbtree_insert(llsd, tmp);
 658                 if (lrn != tmp)
 659                         lfsck_rbtree_free(tmp);
 660         }
 661
 662         idx = fid_oid(fid) & LFSCK_RBTREE_BITMAP_MASK;
 663         /* Any accessed object must be a known object. */
 664         if (!test_and_set_bit(idx, lrn->lrn_known_bitmap))
 665                 atomic_inc(&lrn->lrn_known_count);
 666         if (accessed && !test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
 667                 atomic_inc(&lrn->lrn_accessed_count);
 668
 669         GOTO(unlock, rc = 0);
 670
 671 unlock:
 672         if (insert)
 673                 write_unlock(&llsd->llsd_rb_lock);
 674         else
 675                 read_unlock(&llsd->llsd_rb_lock);
 676 out:
 677         if (rc != 0 && accessed) {
 678                 struct lfsck_layout *lo = com->lc_file_ram;
 679
 680                 CDEBUG(D_LFSCK, "%s: fail to update OST-objects accessing "
 681                        "bitmap, and will cause incorrect LFSCK OST-object "
 682                        "handling, so disable it to cancel orphan handling "
 683                        "for related device. rc = %d\n",
 684                        lfsck_lfsck2name(com->lc_lfsck), rc);
 685
 686                 lo->ll_flags |= LF_INCOMPLETE;
 687                 lfsck_rbtree_cleanup(env, com);
 688         }
 689 }
 690
 691 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
 692                                    const struct lfsck_layout *src)
 693 {
 694         int i;
 695
 696         des->ll_magic = le32_to_cpu(src->ll_magic);
 697         des->ll_status = le32_to_cpu(src->ll_status);
 698         des->ll_flags = le32_to_cpu(src->ll_flags);
 699         des->ll_success_count = le32_to_cpu(src->ll_success_count);
 700         des->ll_run_time_phase1 = le32_to_cpu(src->ll_run_time_phase1);
 701         des->ll_run_time_phase2 = le32_to_cpu(src->ll_run_time_phase2);
 702         des->ll_time_last_complete = le64_to_cpu(src->ll_time_last_complete);
 703         des->ll_time_latest_start = le64_to_cpu(src->ll_time_latest_start);
 704         des->ll_time_last_checkpoint =
 705                                 le64_to_cpu(src->ll_time_last_checkpoint);
 706         des->ll_pos_latest_start = le64_to_cpu(src->ll_pos_latest_start);
 707         des->ll_pos_last_checkpoint = le64_to_cpu(src->ll_pos_last_checkpoint);
 708         des->ll_pos_first_inconsistent =
 709                         le64_to_cpu(src->ll_pos_first_inconsistent);
 710         des->ll_objs_checked_phase1 = le64_to_cpu(src->ll_objs_checked_phase1);
 711         des->ll_objs_failed_phase1 = le64_to_cpu(src->ll_objs_failed_phase1);
 712         des->ll_objs_checked_phase2 = le64_to_cpu(src->ll_objs_checked_phase2);
 713         des->ll_objs_failed_phase2 = le64_to_cpu(src->ll_objs_failed_phase2);
 714         for (i = 0; i < LLIT_MAX; i++)
 715                 des->ll_objs_repaired[i] =
 716                                 le64_to_cpu(src->ll_objs_repaired[i]);
 717         des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
 718         des->ll_bitmap_size = le32_to_cpu(src->ll_bitmap_size);
 719 }
 720
 721 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
 722                                    const struct lfsck_layout *src)
 723 {
 724         int i;
 725
 726         des->ll_magic = cpu_to_le32(src->ll_magic);
 727         des->ll_status = cpu_to_le32(src->ll_status);
 728         des->ll_flags = cpu_to_le32(src->ll_flags);
 729         des->ll_success_count = cpu_to_le32(src->ll_success_count);
 730         des->ll_run_time_phase1 = cpu_to_le32(src->ll_run_time_phase1);
 731         des->ll_run_time_phase2 = cpu_to_le32(src->ll_run_time_phase2);
 732         des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
 733         des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
 734         des->ll_time_last_checkpoint =
 735                                 cpu_to_le64(src->ll_time_last_checkpoint);
 736         des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
 737         des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
 738         des->ll_pos_first_inconsistent =
 739                         cpu_to_le64(src->ll_pos_first_inconsistent);
 740         des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
 741         des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
 742         des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
 743         des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
 744         for (i = 0; i < LLIT_MAX; i++)
 745                 des->ll_objs_repaired[i] =
 746                                 cpu_to_le64(src->ll_objs_repaired[i]);
 747         des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
 748         des->ll_bitmap_size = cpu_to_le32(src->ll_bitmap_size);
 749 }
 750
 751 /**
 752  * Load the OST bitmap from the lfsck_layout trace file.
 753  *
 754  * \param[in] env       pointer to the thread context
 755  * \param[in] com       pointer to the lfsck component
 756  *
 757  * \retval              0 for success
 758  * \retval              negative error number on failure or data corruption
 759  */
 760 static int lfsck_layout_load_bitmap(const struct lu_env *env,
 761                                     struct lfsck_component *com)
 762 {
 763         struct dt_object                *obj    = com->lc_obj;
 764         struct lfsck_assistant_data     *lad    = com->lc_data;
 765         struct lfsck_layout             *lo     = com->lc_file_ram;
 766         cfs_bitmap_t                    *bitmap = lad->lad_bitmap;
 767         loff_t                           pos    = com->lc_file_size;
 768         ssize_t                          size;
 769         __u32                            nbits;
 770         int                              rc;
 771         ENTRY;
 772
 773         if (com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size >
 774             lo->ll_bitmap_size)
 775                 nbits = com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size;
 776         else
 777                 nbits = lo->ll_bitmap_size;
 778
 779         if (unlikely(nbits < BITS_PER_LONG))
 780                 nbits = BITS_PER_LONG;
 781
 782         if (nbits > bitmap->size) {
 783                 __u32 new_bits = bitmap->size;
 784                 cfs_bitmap_t *new_bitmap;
 785
 786                 while (new_bits < nbits)
 787                         new_bits <<= 1;
 788
 789                 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
 790                 if (new_bitmap == NULL)
 791                         RETURN(-ENOMEM);
 792
 793                 lad->lad_bitmap = new_bitmap;
 794                 CFS_FREE_BITMAP(bitmap);
 795                 bitmap = new_bitmap;
 796         }
 797
 798         if (lo->ll_bitmap_size == 0) {
 799                 lad->lad_incomplete = 0;
 800                 CFS_RESET_BITMAP(bitmap);
 801
 802                 RETURN(0);
 803         }
 804
 805         size = (lo->ll_bitmap_size + 7) >> 3;
 806         rc = dt_read(env, obj, lfsck_buf_get(env, bitmap->data, size), &pos);
 807         if (rc != size)
 808                 RETURN(rc >= 0 ? -EINVAL : rc);
 809
 810         if (cfs_bitmap_check_empty(bitmap))
 811                 lad->lad_incomplete = 0;
 812         else
 813                 lad->lad_incomplete = 1;
 814
 815         RETURN(0);
 816 }
 817
 818 /**
 819  * Load the layout LFSCK trace file from disk.
 820  *
 821  * The layout LFSCK trace file records the layout LFSCK status information
 822  * and other statistics, such as how many objects have been scanned, and how
 823  * many objects have been repaired, and etc. It also contains the bitmap for
 824  * failed OSTs during the layout LFSCK. All these information will be loaded
 825  * from disk to RAM when the layout LFSCK component setup.
 826  *
 827  * \param[in] env       pointer to the thread context
 828  * \param[in] com       pointer to the lfsck component
 829  *
 830  * \retval              positive number for file data corruption, the caller
 831  *                      should reset the layout LFSCK trace file
 832  * \retval              0 for success
 833  * \retval              negative error number on failure
 834  */
 835 static int lfsck_layout_load(const struct lu_env *env,
 836                              struct lfsck_component *com)
 837 {
 838         struct lfsck_layout             *lo     = com->lc_file_ram;
 839         ssize_t                          size   = com->lc_file_size;
 840         loff_t                           pos    = 0;
 841         int                              rc;
 842
 843         rc = dt_read(env, com->lc_obj,
 844                      lfsck_buf_get(env, com->lc_file_disk, size), &pos);
 845         if (rc == 0) {
 846                 return -ENOENT;
 847         } else if (rc < 0) {
 848                 CDEBUG(D_LFSCK, "%s: failed to load lfsck_layout: rc = %d\n",
 849                        lfsck_lfsck2name(com->lc_lfsck), rc);
 850                 return rc;
 851         } else if (rc != size) {
 852                 CDEBUG(D_LFSCK, "%s: lfsck_layout size %u != %u; reset it\n",
 853                        lfsck_lfsck2name(com->lc_lfsck), rc, (unsigned int)size);
 854                 return 1;
 855         }
 856
 857         lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
 858         if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
 859                 CDEBUG(D_LFSCK, "%s: invalid lfsck_layout magic %#x != %#x, "
 860                        "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
 861                        lo->ll_magic, LFSCK_LAYOUT_MAGIC);
 862                 return 1;
 863         }
 864
 865         return 0;
 866 }
 867
 868 /**
 869  * Store the layout LFSCK trace file on disk.
 870  *
 871  * The layout LFSCK trace file records the layout LFSCK status information
 872  * and other statistics, such as how many objects have been scanned, and how
 873  * many objects have been repaired, and etc. It also contains the bitmap for
 874  * failed OSTs during the layout LFSCK. All these information will be synced
 875  * from RAM to disk periodically.
 876  *
 877  * \param[in] env       pointer to the thread context
 878  * \param[in] com       pointer to the lfsck component
 879  *
 880  * \retval              0 for success
 881  * \retval              negative error number on failure
 882  */
 883 static int lfsck_layout_store(const struct lu_env *env,
 884                               struct lfsck_component *com)
 885 {
 886         struct dt_object        *obj    = com->lc_obj;
 887         struct lfsck_instance   *lfsck  = com->lc_lfsck;
 888         struct lfsck_layout     *lo_ram = com->lc_file_ram;
 889         struct lfsck_layout     *lo     = com->lc_file_disk;
 890         struct thandle          *th;
 891         struct dt_device        *dev    = lfsck_obj2dev(obj);
 892         cfs_bitmap_t            *bitmap = NULL;
 893         loff_t                   pos;
 894         ssize_t                  size   = com->lc_file_size;
 895         __u32                    nbits  = 0;
 896         int                      rc;
 897         ENTRY;
 898
 899         if (lfsck->li_master) {
 900                 struct lfsck_assistant_data *lad = com->lc_data;
 901
 902                 bitmap = lad->lad_bitmap;
 903                 nbits = bitmap->size;
 904
 905                 LASSERT(nbits > 0);
 906                 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
 907         }
 908
 909         lo_ram->ll_bitmap_size = nbits;
 910         lfsck_layout_cpu_to_le(lo, lo_ram);
 911         th = dt_trans_create(env, dev);
 912         if (IS_ERR(th))
 913                 GOTO(log, rc = PTR_ERR(th));
 914
 915         rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
 916                                      (loff_t)0, th);
 917         if (rc != 0)
 918                 GOTO(out, rc);
 919
 920         if (bitmap != NULL) {
 921                 rc = dt_declare_record_write(env, obj,
 922                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
 923                                 (loff_t)size, th);
 924                 if (rc != 0)
 925                         GOTO(out, rc);
 926         }
 927
 928         rc = dt_trans_start_local(env, dev, th);
 929         if (rc != 0)
 930                 GOTO(out, rc);
 931
 932         pos = 0;
 933         rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos, th);
 934         if (rc != 0)
 935                 GOTO(out, rc);
 936
 937         if (bitmap != NULL) {
 938                 pos = size;
 939                 rc = dt_record_write(env, obj,
 940                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
 941                                 &pos, th);
 942         }
 943
 944         GOTO(out, rc);
 945
 946 out:
 947         dt_trans_stop(env, dev, th);
 948
 949 log:
 950         if (rc != 0)
 951                 CDEBUG(D_LFSCK, "%s: fail to store lfsck_layout: rc = %d\n",
 952                        lfsck_lfsck2name(lfsck), rc);
 953
 954         return rc;
 955 }
 956
 957 static int lfsck_layout_init(const struct lu_env *env,
 958                              struct lfsck_component *com)
 959 {
 960         struct lfsck_layout *lo = com->lc_file_ram;
 961         int rc;
 962
 963         memset(lo, 0, com->lc_file_size);
 964         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
 965         lo->ll_status = LS_INIT;
 966         down_write(&com->lc_sem);
 967         rc = lfsck_layout_store(env, com);
 968         up_write(&com->lc_sem);
 969
 970         return rc;
 971 }
 972
 973 static int fid_is_for_ostobj(const struct lu_env *env,
 974                              struct lfsck_instance *lfsck,
 975                              struct dt_object *obj, const struct lu_fid *fid)
 976 {
 977         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
 978         struct lu_seq_range     *range  = &lfsck_env_info(env)->lti_range;
 979         struct lustre_mdt_attrs *lma;
 980         int                      rc;
 981
 982         fld_range_set_any(range);
 983         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
 984         if (rc == 0) {
 985                 if (fld_range_is_ost(range))
 986                         return 1;
 987
 988                 return 0;
 989         }
 990
 991         lma = &lfsck_env_info(env)->lti_lma;
 992         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, lma, sizeof(*lma)),
 993                           XATTR_NAME_LMA);
 994         if (rc == sizeof(*lma)) {
 995                 lustre_lma_swab(lma);
 996
 997                 return lma->lma_compat & LMAC_FID_ON_OST ? 1 : 0;
 998         }
 999
1000         rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID);
1001
1002         return rc > 0;
1003 }
1004
1005 static struct lfsck_layout_seq *
1006 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
1007 {
1008         struct lfsck_layout_seq *lls;
1009
1010         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1011                 if (lls->lls_seq == seq)
1012                         return lls;
1013
1014                 if (lls->lls_seq > seq)
1015                         return NULL;
1016         }
1017
1018         return NULL;
1019 }
1020
1021 static void
1022 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
1023                         struct lfsck_layout_seq *lls)
1024 {
1025         struct lfsck_layout_seq *tmp;
1026         struct list_head        *pos = &llsd->llsd_seq_list;
1027
1028         list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
1029                 if (lls->lls_seq < tmp->lls_seq) {
1030                         pos = &tmp->lls_list;
1031                         break;
1032                 }
1033         }
1034         list_add_tail(&lls->lls_list, pos);
1035 }
1036
1037 static int
1038 lfsck_layout_lastid_create(const struct lu_env *env,
1039                            struct lfsck_instance *lfsck,
1040                            struct dt_object *obj)
1041 {
1042         struct lfsck_thread_info *info   = lfsck_env_info(env);
1043         struct lu_attr           *la     = &info->lti_la;
1044         struct dt_object_format  *dof    = &info->lti_dof;
1045         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
1046         struct dt_device         *dt     = lfsck_obj2dev(obj);
1047         struct thandle           *th;
1048         __u64                     lastid = 0;
1049         loff_t                    pos    = 0;
1050         int                       rc;
1051         ENTRY;
1052
1053         if (bk->lb_param & LPF_DRYRUN)
1054                 return 0;
1055
1056         memset(la, 0, sizeof(*la));
1057         la->la_mode = S_IFREG |  S_IRUGO | S_IWUSR;
1058         la->la_valid = LA_MODE | LA_UID | LA_GID;
1059         memset(dof, 0, sizeof(*dof));
1060         dof->dof_type = dt_mode_to_dft(S_IFREG);
1061
1062         th = dt_trans_create(env, dt);
1063         if (IS_ERR(th))
1064                 GOTO(log, rc = PTR_ERR(th));
1065
1066         rc = dt_declare_create(env, obj, la, NULL, dof, th);
1067         if (rc != 0)
1068                 GOTO(stop, rc);
1069
1070         rc = dt_declare_record_write(env, obj,
1071                                      lfsck_buf_get(env, &lastid,
1072                                                    sizeof(lastid)),
1073                                      pos, th);
1074         if (rc != 0)
1075                 GOTO(stop, rc);
1076
1077         rc = dt_trans_start_local(env, dt, th);
1078         if (rc != 0)
1079                 GOTO(stop, rc);
1080
1081         dt_write_lock(env, obj, 0);
1082         if (likely(dt_object_exists(obj) == 0)) {
1083                 rc = dt_create(env, obj, la, NULL, dof, th);
1084                 if (rc == 0)
1085                         rc = dt_record_write(env, obj,
1086                                 lfsck_buf_get(env, &lastid, sizeof(lastid)),
1087                                 &pos, th);
1088         }
1089         dt_write_unlock(env, obj);
1090
1091         GOTO(stop, rc);
1092
1093 stop:
1094         dt_trans_stop(env, dt, th);
1095
1096 log:
1097         CDEBUG(D_LFSCK, "%s: layout LFSCK will create LAST_ID for <seq> "
1098                LPX64": rc = %d\n",
1099                lfsck_lfsck2name(lfsck), fid_seq(lfsck_dto2fid(obj)), rc);
1100
1101         return rc;
1102 }
1103
1104 static int
1105 lfsck_layout_lastid_reload(const struct lu_env *env,
1106                            struct lfsck_component *com,
1107                            struct lfsck_layout_seq *lls)
1108 {
1109         __u64   lastid;
1110         loff_t  pos     = 0;
1111         int     rc;
1112
1113         dt_read_lock(env, lls->lls_lastid_obj, 0);
1114         rc = dt_record_read(env, lls->lls_lastid_obj,
1115                             lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
1116         dt_read_unlock(env, lls->lls_lastid_obj);
1117         if (unlikely(rc != 0))
1118                 return rc;
1119
1120         lastid = le64_to_cpu(lastid);
1121         if (lastid < lls->lls_lastid_known) {
1122                 struct lfsck_instance   *lfsck  = com->lc_lfsck;
1123                 struct lfsck_layout     *lo     = com->lc_file_ram;
1124
1125                 lls->lls_lastid = lls->lls_lastid_known;
1126                 lls->lls_dirty = 1;
1127                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1128                         LASSERT(lfsck->li_out_notify != NULL);
1129
1130                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1131                                              LE_LASTID_REBUILDING);
1132                         lo->ll_flags |= LF_CRASHED_LASTID;
1133
1134                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
1135                                "LAST_ID file (1) for the sequence "LPX64
1136                                ", old value "LPU64", known value "LPU64"\n",
1137                                lfsck_lfsck2name(lfsck), lls->lls_seq,
1138                                lastid, lls->lls_lastid);
1139                 }
1140         } else if (lastid >= lls->lls_lastid) {
1141                 lls->lls_lastid = lastid;
1142                 lls->lls_dirty = 0;
1143         }
1144
1145         return 0;
1146 }
1147
1148 static int
1149 lfsck_layout_lastid_store(const struct lu_env *env,
1150                           struct lfsck_component *com)
1151 {
1152         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1153         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
1154         struct dt_device                *dt     = lfsck->li_bottom;
1155         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
1156         struct lfsck_layout_seq         *lls;
1157         struct thandle                  *th;
1158         __u64                            lastid;
1159         int                              rc     = 0;
1160         int                              rc1    = 0;
1161
1162         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1163                 loff_t pos = 0;
1164
1165                 if (!lls->lls_dirty)
1166                         continue;
1167
1168                 CDEBUG(D_LFSCK, "%s: layout LFSCK will sync the LAST_ID for "
1169                        "<seq> "LPX64" as <oid> "LPU64"\n",
1170                        lfsck_lfsck2name(lfsck), lls->lls_seq, lls->lls_lastid);
1171
1172                 if (bk->lb_param & LPF_DRYRUN) {
1173                         lls->lls_dirty = 0;
1174                         continue;
1175                 }
1176
1177                 th = dt_trans_create(env, dt);
1178                 if (IS_ERR(th)) {
1179                         rc1 = PTR_ERR(th);
1180                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1181                                "the LAST_ID for <seq> "LPX64"(1): rc = %d\n",
1182                                lfsck_lfsck2name(com->lc_lfsck),
1183                                lls->lls_seq, rc1);
1184                         continue;
1185                 }
1186
1187                 lastid = cpu_to_le64(lls->lls_lastid);
1188                 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1189                                              lfsck_buf_get(env, &lastid,
1190                                                            sizeof(lastid)),
1191                                              pos, th);
1192                 if (rc != 0)
1193                         goto stop;
1194
1195                 rc = dt_trans_start_local(env, dt, th);
1196                 if (rc != 0)
1197                         goto stop;
1198
1199                 dt_write_lock(env, lls->lls_lastid_obj, 0);
1200                 rc = dt_record_write(env, lls->lls_lastid_obj,
1201                                      lfsck_buf_get(env, &lastid,
1202                                      sizeof(lastid)), &pos, th);
1203                 dt_write_unlock(env, lls->lls_lastid_obj);
1204                 if (rc == 0)
1205                         lls->lls_dirty = 0;
1206
1207 stop:
1208                 dt_trans_stop(env, dt, th);
1209                 if (rc != 0) {
1210                         rc1 = rc;
1211                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1212                                "the LAST_ID for <seq> "LPX64"(2): rc = %d\n",
1213                                lfsck_lfsck2name(com->lc_lfsck),
1214                                lls->lls_seq, rc1);
1215                 }
1216         }
1217
1218         return rc1;
1219 }
1220
1221 static int
1222 lfsck_layout_lastid_load(const struct lu_env *env,
1223                          struct lfsck_component *com,
1224                          struct lfsck_layout_seq *lls)
1225 {
1226         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1227         struct lfsck_layout     *lo     = com->lc_file_ram;
1228         struct lu_fid           *fid    = &lfsck_env_info(env)->lti_fid;
1229         struct dt_object        *obj;
1230         loff_t                   pos    = 0;
1231         int                      rc;
1232         ENTRY;
1233
1234         lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck));
1235         obj = dt_locate(env, lfsck->li_bottom, fid);
1236         if (IS_ERR(obj))
1237                 RETURN(PTR_ERR(obj));
1238
1239         /* LAST_ID crashed, to be rebuilt */
1240         if (dt_object_exists(obj) == 0) {
1241                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1242                         LASSERT(lfsck->li_out_notify != NULL);
1243
1244                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1245                                              LE_LASTID_REBUILDING);
1246                         lo->ll_flags |= LF_CRASHED_LASTID;
1247
1248                         CDEBUG(D_LFSCK, "%s: layout LFSCK cannot find the "
1249                                "LAST_ID file for sequence "LPX64"\n",
1250                                lfsck_lfsck2name(lfsck), lls->lls_seq);
1251
1252                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1253                             cfs_fail_val > 0) {
1254                                 struct l_wait_info lwi = LWI_TIMEOUT(
1255                                                 cfs_time_seconds(cfs_fail_val),
1256                                                 NULL, NULL);
1257
1258                                 /* Some others may changed the cfs_fail_val
1259                                  * as zero after above check, re-check it for
1260                                  * sure to avoid falling into wait for ever. */
1261                                 if (likely(lwi.lwi_timeout > 0)) {
1262                                         struct ptlrpc_thread *thread =
1263                                                 &lfsck->li_thread;
1264
1265                                         up_write(&com->lc_sem);
1266                                         l_wait_event(thread->t_ctl_waitq,
1267                                                      !thread_is_running(thread),
1268                                                      &lwi);
1269                                         down_write(&com->lc_sem);
1270                                 }
1271                         }
1272                 }
1273
1274                 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1275         } else {
1276                 dt_read_lock(env, obj, 0);
1277                 rc = dt_read(env, obj,
1278                         lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1279                         &pos);
1280                 dt_read_unlock(env, obj);
1281                 if (rc != 0 && rc != sizeof(__u64))
1282                         GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1283
1284                 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1285                         LASSERT(lfsck->li_out_notify != NULL);
1286
1287                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1288                                              LE_LASTID_REBUILDING);
1289                         lo->ll_flags |= LF_CRASHED_LASTID;
1290
1291                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds invalid "
1292                                "LAST_ID file for the sequence "LPX64
1293                                ": rc = %d\n",
1294                                lfsck_lfsck2name(lfsck), lls->lls_seq, rc);
1295                 }
1296
1297                 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1298                 rc = 0;
1299         }
1300
1301         GOTO(out, rc);
1302
1303 out:
1304         if (rc != 0)
1305                 lfsck_object_put(env, obj);
1306         else
1307                 lls->lls_lastid_obj = obj;
1308
1309         return rc;
1310 }
1311
1312 static void lfsck_layout_record_failure(const struct lu_env *env,
1313                                         struct lfsck_instance *lfsck,
1314                                         struct lfsck_layout *lo)
1315 {
1316         __u64 cookie;
1317
1318         lo->ll_objs_failed_phase1++;
1319         cookie = lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
1320                                                         lfsck->li_di_oit);
1321         if (lo->ll_pos_first_inconsistent == 0 ||
1322             lo->ll_pos_first_inconsistent < cookie) {
1323                 lo->ll_pos_first_inconsistent = cookie;
1324
1325                 CDEBUG(D_LFSCK, "%s: layout LFSCK hit first non-repaired "
1326                        "inconsistency at the pos ["LPU64"]\n",
1327                        lfsck_lfsck2name(lfsck),
1328                        lo->ll_pos_first_inconsistent);
1329         }
1330 }
1331
1332 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1333                                            struct lfsck_component *com,
1334                                            int rc)
1335 {
1336         struct lfsck_instance   *lfsck = com->lc_lfsck;
1337         struct lfsck_layout     *lo    = com->lc_file_ram;
1338
1339         down_write(&com->lc_sem);
1340         lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1341                                   HALF_SEC - com->lc_time_last_checkpoint);
1342         lo->ll_time_last_checkpoint = cfs_time_current_sec();
1343         lo->ll_objs_checked_phase2 += com->lc_new_checked;
1344
1345         if (rc > 0) {
1346                 if (lo->ll_flags & LF_INCOMPLETE) {
1347                         lo->ll_status = LS_PARTIAL;
1348                 } else {
1349                         if (lfsck->li_master) {
1350                                 struct lfsck_assistant_data *lad = com->lc_data;
1351
1352                                 if (lad->lad_incomplete)
1353                                         lo->ll_status = LS_PARTIAL;
1354                                 else
1355                                         lo->ll_status = LS_COMPLETED;
1356                         } else {
1357                                 lo->ll_status = LS_COMPLETED;
1358                         }
1359                 }
1360                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
1361                         lo->ll_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
1362                 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1363                 lo->ll_success_count++;
1364         } else if (rc == 0) {
1365                 if (lfsck->li_status != 0)
1366                         lo->ll_status = lfsck->li_status;
1367                 else
1368                         lo->ll_status = LS_STOPPED;
1369         } else {
1370                 lo->ll_status = LS_FAILED;
1371         }
1372
1373         rc = lfsck_layout_store(env, com);
1374         up_write(&com->lc_sem);
1375
1376         return rc;
1377 }
1378
1379 static int lfsck_layout_trans_stop(const struct lu_env *env,
1380                                    struct dt_device *dev,
1381                                    struct thandle *handle, int result)
1382 {
1383         int rc;
1384
1385         /* XXX: If there is something worng or it needs to repair nothing,
1386          *      then notify the lower to stop the modification. Currently,
1387          *      we use th_result for such purpose, that may be replaced by
1388          *      some rollback mechanism in the future. */
1389         handle->th_result = result;
1390         rc = dt_trans_stop(env, dev, handle);
1391         if (result != 0)
1392                 return result > 0 ? 0 : result;
1393
1394         return rc == 0 ? 1 : rc;
1395 }
1396
1397 /**
1398  * Get the system default stripe size.
1399  *
1400  * \param[in] env       pointer to the thread context
1401  * \param[in] lfsck     pointer to the lfsck instance
1402  * \param[out] size     pointer to the default stripe size
1403  *
1404  * \retval              0 for success
1405  * \retval              negative error number on failure
1406  */
1407 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1408                                            struct lfsck_instance *lfsck,
1409                                            __u32 *size)
1410 {
1411         struct lov_user_md      *lum = &lfsck_env_info(env)->lti_lum;
1412         struct dt_object        *root;
1413         int                      rc;
1414
1415         root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1416         if (IS_ERR(root))
1417                 return PTR_ERR(root);
1418
1419         /* Get the default stripe size via xattr_get on the backend root. */
1420         rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1421                           XATTR_NAME_LOV);
1422         if (rc > 0) {
1423                 /* The lum->lmm_stripe_size is LE mode. The *size also
1424                  * should be LE mode. So it is unnecessary to convert. */
1425                 *size = lum->lmm_stripe_size;
1426                 rc = 0;
1427         } else if (unlikely(rc == 0)) {
1428                 rc = -EINVAL;
1429         }
1430
1431         lfsck_object_put(env, root);
1432
1433         return rc;
1434 }
1435
1436 /**
1437  * \retval       +1: repaired
1438  * \retval        0: did nothing
1439  * \retval      -ve: on error
1440  */
1441 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1442                                      struct thandle *handle,
1443                                      struct dt_object *parent,
1444                                      struct lu_fid *cfid,
1445                                      struct lu_buf *buf,
1446                                      struct lov_ost_data_v1 *slot,
1447                                      int fl, __u32 ost_idx)
1448 {
1449         struct ost_id           *oi     = &lfsck_env_info(env)->lti_oi;
1450         struct lov_mds_md_v1    *lmm    = buf->lb_buf;
1451         struct lu_buf            ea_buf;
1452         int                      rc;
1453         __u32                    magic;
1454         __u16                    count;
1455         ENTRY;
1456
1457         magic = le32_to_cpu(lmm->lmm_magic);
1458         count = le16_to_cpu(lmm->lmm_stripe_count);
1459
1460         fid_to_ostid(cfid, oi);
1461         ostid_cpu_to_le(oi, &slot->l_ost_oi);
1462         slot->l_ost_gen = cpu_to_le32(0);
1463         slot->l_ost_idx = cpu_to_le32(ost_idx);
1464
1465         if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE) {
1466                 struct lov_ost_data_v1 *objs;
1467                 int                     i;
1468
1469                 if (magic == LOV_MAGIC_V1)
1470                         objs = &lmm->lmm_objects[0];
1471                 else
1472                         objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1473                 for (i = 0; i < count; i++, objs++) {
1474                         if (objs != slot && lovea_slot_is_dummy(objs))
1475                                 break;
1476                 }
1477
1478                 /* If the @slot is the last dummy slot to be refilled,
1479                  * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1480                 if (i == count)
1481                         lmm->lmm_pattern &= ~cpu_to_le32(LOV_PATTERN_F_HOLE);
1482         }
1483
1484         lfsck_buf_init(&ea_buf, lmm, lov_mds_md_size(count, magic));
1485         rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle);
1486         if (rc == 0)
1487                 rc = 1;
1488
1489         RETURN(rc);
1490 }
1491
1492 /**
1493  * \retval       +1: repaired
1494  * \retval        0: did nothing
1495  * \retval      -ve: on error
1496  */
1497 static int lfsck_layout_extend_lovea(const struct lu_env *env,
1498                                      struct lfsck_instance *lfsck,
1499                                      struct thandle *handle,
1500                                      struct dt_object *parent,
1501                                      struct lu_fid *cfid,
1502                                      struct lu_buf *buf, int fl,
1503                                      __u32 ost_idx, __u32 ea_off, bool reset)
1504 {
1505         struct lov_mds_md_v1    *lmm    = buf->lb_buf;
1506         struct lov_ost_data_v1  *objs;
1507         int                      rc;
1508         __u16                    count;
1509         bool                     hole   = false;
1510         ENTRY;
1511
1512         if (fl == LU_XATTR_CREATE || reset) {
1513                 __u32 pattern = LOV_PATTERN_RAID0;
1514
1515                 count = ea_off + 1;
1516                 LASSERT(buf->lb_len >= lov_mds_md_size(count, LOV_MAGIC_V1));
1517
1518                 if (ea_off != 0 || reset) {
1519                         pattern |= LOV_PATTERN_F_HOLE;
1520                         hole = true;
1521                 }
1522
1523                 memset(lmm, 0, buf->lb_len);
1524                 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1525                 lmm->lmm_pattern = cpu_to_le32(pattern);
1526                 fid_to_lmm_oi(lfsck_dto2fid(parent), &lmm->lmm_oi);
1527                 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1528
1529                 rc = lfsck_layout_get_def_stripesize(env, lfsck,
1530                                                      &lmm->lmm_stripe_size);
1531                 if (rc != 0)
1532                         RETURN(rc);
1533
1534                 objs = &lmm->lmm_objects[ea_off];
1535         } else {
1536                 __u32   magic = le32_to_cpu(lmm->lmm_magic);
1537                 int     gap;
1538
1539                 count = le16_to_cpu(lmm->lmm_stripe_count);
1540                 if (magic == LOV_MAGIC_V1)
1541                         objs = &lmm->lmm_objects[count];
1542                 else
1543                         objs = &((struct lov_mds_md_v3 *)lmm)->
1544                                                         lmm_objects[count];
1545
1546                 gap = ea_off - count;
1547                 if (gap >= 0)
1548                         count = ea_off + 1;
1549                 LASSERT(buf->lb_len >= lov_mds_md_size(count, magic));
1550
1551                 if (gap > 0) {
1552                         memset(objs, 0, gap * sizeof(*objs));
1553                         lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
1554                         hole = true;
1555                 }
1556
1557                 lmm->lmm_layout_gen =
1558                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
1559                 objs += gap;
1560         }
1561
1562         lmm->lmm_stripe_count = cpu_to_le16(count);
1563         rc = lfsck_layout_refill_lovea(env, handle, parent, cfid, buf, objs,
1564                                        fl, ost_idx);
1565
1566         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
1567                DFID": parent "DFID", OST-index %u, stripe-index %u, fl %d, "
1568                "reset %s, %s LOV EA hole: rc = %d\n",
1569                lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1570                ost_idx, ea_off, fl, reset ? "yes" : "no",
1571                hole ? "with" : "without", rc);
1572
1573         RETURN(rc);
1574 }
1575
1576 static int __lfsck_layout_update_pfid(const struct lu_env *env,
1577                                       struct dt_object *child,
1578                                       const struct lu_fid *pfid, __u32 offset)
1579 {
1580         struct dt_device        *dev    = lfsck_obj2dev(child);
1581         struct filter_fid       *ff     = &lfsck_env_info(env)->lti_new_pfid;
1582         struct thandle          *handle;
1583         struct lu_buf            buf    = { NULL };
1584         int                      rc;
1585
1586         ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
1587         ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
1588         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
1589          * MDT-object's FID::f_ver, instead it is the OST-object index in its
1590          * parent MDT-object's layout EA. */
1591         ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
1592         lfsck_buf_init(&buf, ff, sizeof(struct filter_fid));
1593
1594         handle = dt_trans_create(env, dev);
1595         if (IS_ERR(handle))
1596                 RETURN(PTR_ERR(handle));
1597
1598         rc = dt_declare_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
1599         if (rc != 0)
1600                 GOTO(stop, rc);
1601
1602         rc = dt_trans_start_local(env, dev, handle);
1603         if (rc != 0)
1604                 GOTO(stop, rc);
1605
1606         rc = dt_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
1607
1608         GOTO(stop, rc);
1609
1610 stop:
1611         dt_trans_stop(env, dev, handle);
1612
1613         return rc;
1614 }
1615
1616 /**
1617  * \retval       +1: repaired
1618  * \retval        0: did nothing
1619  * \retval      -ve: on error
1620  */
1621 static int lfsck_layout_update_pfid(const struct lu_env *env,
1622                                     struct lfsck_component *com,
1623                                     struct dt_object *parent,
1624                                     struct lu_fid *cfid,
1625                                     struct dt_device *cdev, __u32 ea_off)
1626 {
1627         struct dt_object        *child;
1628         int                      rc     = 0;
1629         ENTRY;
1630
1631         child = lfsck_object_find_by_dev(env, cdev, cfid);
1632         if (IS_ERR(child))
1633                 RETURN(PTR_ERR(child));
1634
1635         rc = __lfsck_layout_update_pfid(env, child,
1636                                         lu_object_fid(&parent->do_lu), ea_off);
1637         lfsck_object_put(env, child);
1638
1639         RETURN(rc == 0 ? 1 : rc);
1640 }
1641
1642 /**
1643  * This function will create the MDT-object with the given (partial) LOV EA.
1644  *
1645  * Under some data corruption cases, the MDT-object of the file may be lost,
1646  * but its OST-objects, or some of them are there. The layout LFSCK needs to
1647  * re-create the MDT-object with the orphan OST-object(s) information.
1648  *
1649  * On the other hand, the LFSCK may has created some OST-object for repairing
1650  * dangling LOV EA reference, but as the LFSCK processing, it may find that
1651  * the old OST-object is there and should replace the former new created OST
1652  * object. Unfortunately, some others have modified such newly created object.
1653  * To keep the data (both new and old), the LFSCK will create MDT-object with
1654  * new FID to reference the original OST-object.
1655  *
1656  * \param[in] env       pointer to the thread context
1657  * \param[in] com       pointer to the lfsck component
1658  * \param[in] ltd       pointer to target device descriptor
1659  * \param[in] rec       pointer to the record for the orphan OST-object
1660  * \param[in] cfid      pointer to FID for the orphan OST-object
1661  * \param[in] infix     additional information, such as the FID for original
1662  *                      MDT-object and the stripe offset in the LOV EA
1663  * \param[in] type      the type for describing why the orphan MDT-object is
1664  *                      created. The rules are as following:
1665  *
1666  *  type "C":           Multiple OST-objects claim the same MDT-object and the
1667  *                      same slot in the layout EA. Then the LFSCK will create
1668  *                      new MDT-object(s) to hold the conflict OST-object(s).
1669  *
1670  *  type "N":           The orphan OST-object does not know which one was the
1671  *                      real parent MDT-object, so the LFSCK uses new FID for
1672  *                      its parent MDT-object.
1673  *
1674  *  type "R":           The orphan OST-object knows its parent MDT-object FID,
1675  *                      but does not know the position (the file name) in the
1676  *                      layout.
1677  *
1678  *  type "D":           The MDT-object is a directory, it may knows its parent
1679  *                      but because there is no valid linkEA, the LFSCK cannot
1680  *                      know where to put it back to the namespace.
1681  *  type "O":           The MDT-object has no linkEA, and there is no name
1682  *                      entry that references the MDT-object.
1683  *
1684  *  type "P":           The orphan object to be created was a parent directory
1685  *                      of some MDT-object which linkEA shows that the @orphan
1686  *                      object is missing.
1687  *
1688  * The orphan name will be like:
1689  * ${FID}-${infix}-${type}-${conflict_version}
1690  *
1691  * \param[in] ea_off    the stripe offset in the LOV EA
1692  *
1693  * \retval              positive on repaired something
1694  * \retval              0 if needs to repair nothing
1695  * \retval              negative error number on failure
1696  */
1697 static int lfsck_layout_recreate_parent(const struct lu_env *env,
1698                                         struct lfsck_component *com,
1699                                         struct lfsck_tgt_desc *ltd,
1700                                         struct lu_orphan_rec *rec,
1701                                         struct lu_fid *cfid,
1702                                         const char *infix,
1703                                         const char *type,
1704                                         __u32 ea_off)
1705 {
1706         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1707         struct dt_insert_rec            *dtrec  = &info->lti_dt_rec;
1708         char                            *name   = info->lti_key;
1709         struct lu_attr                  *la     = &info->lti_la2;
1710         struct dt_object_format         *dof    = &info->lti_dof;
1711         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1712         struct lu_fid                   *pfid   = &rec->lor_fid;
1713         struct lu_fid                   *tfid   = &info->lti_fid3;
1714         struct dt_device                *dev    = lfsck->li_bottom;
1715         struct dt_object                *lpf    = lfsck->li_lpf_obj;
1716         struct dt_object                *pobj   = NULL;
1717         struct dt_object                *cobj   = NULL;
1718         struct thandle                  *th     = NULL;
1719         struct lu_buf                   *ea_buf = &info->lti_big_buf;
1720         struct lu_buf                    lov_buf;
1721         struct lfsck_lock_handle        *llh    = &info->lti_llh;
1722         struct linkea_data               ldata  = { NULL };
1723         struct lu_buf                    linkea_buf;
1724         const struct lu_name            *pname;
1725         int                              size   = 0;
1726         int                              idx    = 0;
1727         int                              rc     = 0;
1728         ENTRY;
1729
1730         if (unlikely(lpf == NULL))
1731                 GOTO(log, rc = -ENXIO);
1732
1733         /* We use two separated transactions to repair the inconsistency.
1734          *
1735          * 1) create the MDT-object locally.
1736          * 2) update the OST-object's PFID EA if necessary.
1737          *
1738          * If 1) succeed, but 2) failed, then the OST-object's PFID EA will be
1739          * updated when the layout LFSCK run next time.
1740          *
1741          * If 1) failed, but 2) succeed, then such MDT-object will be re-created
1742          * when the layout LFSCK run next time. */
1743
1744         if (fid_is_zero(pfid)) {
1745                 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
1746                 if (rc != 0)
1747                         GOTO(log, rc);
1748
1749                 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
1750                 if (IS_ERR(cobj))
1751                         GOTO(log, rc = PTR_ERR(cobj));
1752         }
1753
1754         pobj = lfsck_object_find_by_dev(env, dev, pfid);
1755         if (IS_ERR(pobj))
1756                 GOTO(log, rc = PTR_ERR(pobj));
1757
1758         LASSERT(infix != NULL);
1759         LASSERT(type != NULL);
1760
1761         memset(la, 0, sizeof(*la));
1762         la->la_uid = rec->lor_uid;
1763         la->la_gid = rec->lor_gid;
1764         la->la_mode = S_IFREG | S_IRUSR;
1765         la->la_valid = LA_MODE | LA_UID | LA_GID;
1766
1767         memset(dof, 0, sizeof(*dof));
1768         dof->dof_type = dt_mode_to_dft(S_IFREG);
1769         /* Because the dof->dof_reg.striped = 0, the LOD will not create
1770          * the stripe(s). The LFSCK will specify the LOV EA via
1771          * lfsck_layout_extend_lovea(). */
1772
1773         size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
1774         if (ea_buf->lb_len < size) {
1775                 lu_buf_realloc(ea_buf, size);
1776                 if (ea_buf->lb_buf == NULL)
1777                         GOTO(log, rc = -ENOMEM);
1778         }
1779
1780 again:
1781         do {
1782                 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
1783                          type, idx++);
1784                 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1785                                (const struct dt_key *)name);
1786                 if (rc != 0 && rc != -ENOENT)
1787                         GOTO(log, rc);
1788         } while (rc == 0);
1789
1790         rc = lfsck_lock(env, lfsck, lfsck->li_lpf_obj, name, llh,
1791                         MDS_INODELOCK_UPDATE, LCK_PW);
1792         if (rc != 0)
1793                 GOTO(log, rc);
1794
1795         /* Re-check whether the name conflict with othrs after taken
1796          * the ldlm lock. */
1797         rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1798                        (const struct dt_key *)name);
1799         if (unlikely(rc == 0)) {
1800                 lfsck_unlock(llh);
1801                 goto again;
1802         }
1803
1804         if (rc != -ENOENT)
1805                 GOTO(unlock, rc);
1806
1807         rc = linkea_data_new(&ldata,
1808                              &lfsck_env_info(env)->lti_linkea_buf);
1809         if (rc != 0)
1810                 GOTO(unlock, rc);
1811
1812         pname = lfsck_name_get_const(env, name, strlen(name));
1813         rc = linkea_add_buf(&ldata, pname, lfsck_dto2fid(lfsck->li_lpf_obj));
1814         if (rc != 0)
1815                 GOTO(unlock, rc);
1816
1817         /* The 1st transaction. */
1818         th = dt_trans_create(env, dev);
1819         if (IS_ERR(th))
1820                 GOTO(unlock, rc = PTR_ERR(th));
1821
1822         rc = dt_declare_create(env, pobj, la, NULL, dof, th);
1823         if (rc != 0)
1824                 GOTO(stop, rc);
1825
1826         lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
1827         rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
1828                                   LU_XATTR_REPLACE, th);
1829         if (rc != 0)
1830                 GOTO(stop, rc);
1831
1832         dtrec->rec_fid = pfid;
1833         dtrec->rec_type = S_IFREG;
1834         rc = dt_declare_insert(env, lpf,
1835                                (const struct dt_rec *)dtrec,
1836                                (const struct dt_key *)name, th);
1837         if (rc != 0)
1838                 GOTO(stop, rc);
1839
1840         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1841                        ldata.ld_leh->leh_len);
1842         rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
1843                                   XATTR_NAME_LINK, 0, th);
1844         if (rc != 0)
1845                 GOTO(stop, rc);
1846
1847         rc = dt_trans_start_local(env, dev, th);
1848         if (rc != 0)
1849                 GOTO(stop, rc);
1850
1851         dt_write_lock(env, pobj, 0);
1852         rc = dt_create(env, pobj, la, NULL, dof, th);
1853         if (rc == 0)
1854                 rc = lfsck_layout_extend_lovea(env, lfsck, th, pobj, cfid,
1855                                 &lov_buf, 0, ltd->ltd_index, ea_off, true);
1856         dt_write_unlock(env, pobj);
1857         if (rc < 0)
1858                 GOTO(stop, rc);
1859
1860         rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
1861                        (const struct dt_key *)name, th, 1);
1862         if (rc != 0)
1863                 GOTO(stop, rc);
1864
1865         rc = dt_xattr_set(env, pobj, &linkea_buf, XATTR_NAME_LINK, 0, th);
1866         if (rc == 0 && cobj != NULL) {
1867                 dt_trans_stop(env, dev, th);
1868                 th = NULL;
1869
1870                 /* The 2nd transaction. */
1871                 rc = __lfsck_layout_update_pfid(env, cobj, pfid, ea_off);
1872         }
1873
1874         GOTO(stop, rc);
1875
1876 stop:
1877         if (th != NULL)
1878                 dt_trans_stop(env, dev, th);
1879
1880 unlock:
1881         lfsck_unlock(llh);
1882
1883 log:
1884         if (cobj != NULL && !IS_ERR(cobj))
1885                 lfsck_object_put(env, cobj);
1886         if (pobj != NULL && !IS_ERR(pobj))
1887                 lfsck_object_put(env, pobj);
1888
1889         if (rc < 0)
1890                 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
1891                        "recreate the lost MDT-object: parent "DFID
1892                        ", child "DFID", OST-index %u, stripe-index %u, "
1893                        "infix %s, type %s: rc = %d\n",
1894                        lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
1895                        ltd->ltd_index, ea_off, infix, type, rc);
1896
1897         return rc >= 0 ? 1 : rc;
1898 }
1899
1900 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
1901                                                    struct lfsck_component *com,
1902                                                    const struct lu_fid *fid,
1903                                                    __u32 index)
1904 {
1905         struct lfsck_thread_info *info  = lfsck_env_info(env);
1906         struct lfsck_request     *lr    = &info->lti_lr;
1907         struct lfsck_instance    *lfsck = com->lc_lfsck;
1908         struct lfsck_tgt_desc    *ltd;
1909         struct ptlrpc_request    *req;
1910         struct lfsck_request     *tmp;
1911         struct obd_export        *exp;
1912         int                       rc    = 0;
1913         ENTRY;
1914
1915         ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
1916         if (unlikely(ltd == NULL))
1917                 RETURN(-ENXIO);
1918
1919         exp = ltd->ltd_exp;
1920         if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
1921                 GOTO(put, rc = -EOPNOTSUPP);
1922
1923         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
1924         if (req == NULL)
1925                 GOTO(put, rc = -ENOMEM);
1926
1927         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
1928         if (rc != 0) {
1929                 ptlrpc_request_free(req);
1930
1931                 GOTO(put, rc);
1932         }
1933
1934         memset(lr, 0, sizeof(*lr));
1935         lr->lr_event = LE_CONDITIONAL_DESTROY;
1936         lr->lr_active = LFSCK_TYPE_LAYOUT;
1937         lr->lr_fid = *fid;
1938
1939         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
1940         *tmp = *lr;
1941         ptlrpc_request_set_replen(req);
1942
1943         rc = ptlrpc_queue_wait(req);
1944         ptlrpc_req_finished(req);
1945
1946         GOTO(put, rc);
1947
1948 put:
1949         lfsck_tgt_put(ltd);
1950
1951         return rc;
1952 }
1953
1954 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
1955                                                   struct lfsck_component *com,
1956                                                   struct lfsck_request *lr)
1957 {
1958         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1959         struct lu_attr                  *la     = &info->lti_la;
1960         union ldlm_policy_data          *policy = &info->lti_policy;
1961         struct ldlm_res_id              *resid  = &info->lti_resid;
1962         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1963         struct dt_device                *dev    = lfsck->li_bottom;
1964         struct lu_fid                   *fid    = &lr->lr_fid;
1965         struct dt_object                *obj;
1966         struct thandle                  *th     = NULL;
1967         struct lustre_handle             lh     = { 0 };
1968         __u64                            flags  = 0;
1969         int                              rc     = 0;
1970         ENTRY;
1971
1972         obj = lfsck_object_find_by_dev(env, dev, fid);
1973         if (IS_ERR(obj))
1974                 RETURN(PTR_ERR(obj));
1975
1976         dt_read_lock(env, obj, 0);
1977         if (dt_object_exists(obj) == 0 ||
1978             lfsck_is_dead_obj(obj)) {
1979                 dt_read_unlock(env, obj);
1980
1981                 GOTO(put, rc = -ENOENT);
1982         }
1983
1984         /* Get obj's attr without lock firstly. */
1985         rc = dt_attr_get(env, obj, la);
1986         dt_read_unlock(env, obj);
1987         if (rc != 0)
1988                 GOTO(put, rc);
1989
1990         if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
1991                 GOTO(put, rc = -ETXTBSY);
1992
1993         /* Acquire extent lock on [0, EOF] to sync with all possible written. */
1994         LASSERT(lfsck->li_namespace != NULL);
1995
1996         memset(policy, 0, sizeof(*policy));
1997         policy->l_extent.end = OBD_OBJECT_EOF;
1998         ost_fid_build_resid(fid, resid);
1999         rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
2000                                     policy, LCK_EX, &flags, ldlm_blocking_ast,
2001                                     ldlm_completion_ast, NULL, NULL, 0,
2002                                     LVB_T_NONE, NULL, &lh);
2003         if (rc != ELDLM_OK)
2004                 GOTO(put, rc = -EIO);
2005
2006         dt_write_lock(env, obj, 0);
2007         /* Get obj's attr within lock again. */
2008         rc = dt_attr_get(env, obj, la);
2009         if (rc != 0)
2010                 GOTO(unlock, rc);
2011
2012         if (la->la_ctime != 0)
2013                 GOTO(unlock, rc = -ETXTBSY);
2014
2015         th = dt_trans_create(env, dev);
2016         if (IS_ERR(th))
2017                 GOTO(unlock, rc = PTR_ERR(th));
2018
2019         rc = dt_declare_ref_del(env, obj, th);
2020         if (rc != 0)
2021                 GOTO(stop, rc);
2022
2023         rc = dt_declare_destroy(env, obj, th);
2024         if (rc != 0)
2025                 GOTO(stop, rc);
2026
2027         rc = dt_trans_start_local(env, dev, th);
2028         if (rc != 0)
2029                 GOTO(stop, rc);
2030
2031         rc = dt_ref_del(env, obj, th);
2032         if (rc != 0)
2033                 GOTO(stop, rc);
2034
2035         rc = dt_destroy(env, obj, th);
2036         if (rc == 0)
2037                 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2038                        "OST-object "DFID" that was created for reparing "
2039                        "dangling referenced case. But the original missing "
2040                        "OST-object is found now.\n",
2041                        lfsck_lfsck2name(lfsck), PFID(fid));
2042
2043         GOTO(stop, rc);
2044
2045 stop:
2046         dt_trans_stop(env, dev, th);
2047
2048 unlock:
2049         dt_write_unlock(env, obj);
2050         ldlm_lock_decref(&lh, LCK_EX);
2051
2052 put:
2053         lfsck_object_put(env, obj);
2054
2055         return rc;
2056 }
2057
2058 /**
2059  * Some OST-object has occupied the specified layout EA slot.
2060  * Such OST-object may be generated by the LFSCK when repair
2061  * dangling referenced MDT-object, which can be indicated by
2062  * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2063  * is true and such OST-object has not been modified yet, we
2064  * will replace it with the orphan OST-object; otherwise the
2065  * LFSCK will create new MDT-object to reference the orphan.
2066  *
2067  * \retval       +1: repaired
2068  * \retval        0: did nothing
2069  * \retval      -ve: on error
2070  */
2071 static int lfsck_layout_conflict_create(const struct lu_env *env,
2072                                         struct lfsck_component *com,
2073                                         struct lfsck_tgt_desc *ltd,
2074                                         struct lu_orphan_rec *rec,
2075                                         struct dt_object *parent,
2076                                         struct lu_fid *cfid,
2077                                         struct lu_buf *ea_buf,
2078                                         struct lov_ost_data_v1 *slot,
2079                                         __u32 ea_off)
2080 {
2081         struct lfsck_thread_info *info          = lfsck_env_info(env);
2082         struct lu_fid            *cfid2         = &info->lti_fid2;
2083         struct ost_id            *oi            = &info->lti_oi;
2084         struct lov_mds_md_v1     *lmm           = ea_buf->lb_buf;
2085         struct dt_device         *dev           = lfsck_obj2dev(parent);
2086         struct thandle           *th            = NULL;
2087         struct lustre_handle      lh            = { 0 };
2088         __u32                     ost_idx2      = le32_to_cpu(slot->l_ost_idx);
2089         int                       rc            = 0;
2090         ENTRY;
2091
2092         ostid_le_to_cpu(&slot->l_ost_oi, oi);
2093         rc = ostid_to_fid(cfid2, oi, ost_idx2);
2094         if (rc != 0)
2095                 GOTO(out, rc);
2096
2097         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2098                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2099                               LCK_EX);
2100         if (rc != 0)
2101                 GOTO(out, rc);
2102
2103         rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2104
2105         /* If the conflict OST-obejct is not created for fixing dangling
2106          * referenced MDT-object in former LFSCK check/repair, or it has
2107          * been modified by others, then we cannot destroy it. Re-create
2108          * a new MDT-object for the orphan OST-object. */
2109         if (rc == -ETXTBSY) {
2110                 /* No need the layout lock on the original parent. */
2111                 lfsck_ibits_unlock(&lh, LCK_EX);
2112
2113                 fid_zero(&rec->lor_fid);
2114                 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2115                          "-"DFID"-%x", PFID(lu_object_fid(&parent->do_lu)),
2116                          ea_off);
2117                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2118                                                 info->lti_tmpbuf, "C", ea_off);
2119
2120                 RETURN(rc);
2121         }
2122
2123         if (rc != 0 && rc != -ENOENT)
2124                 GOTO(unlock, rc);
2125
2126         th = dt_trans_create(env, dev);
2127         if (IS_ERR(th))
2128                 GOTO(unlock, rc = PTR_ERR(th));
2129
2130         rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2131                                   LU_XATTR_REPLACE, th);
2132         if (rc != 0)
2133                 GOTO(stop, rc);
2134
2135         rc = dt_trans_start_local(env, dev, th);
2136         if (rc != 0)
2137                 GOTO(stop, rc);
2138
2139         dt_write_lock(env, parent, 0);
2140         lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2141         rc = lfsck_layout_refill_lovea(env, th, parent, cfid, ea_buf, slot,
2142                                        LU_XATTR_REPLACE, ltd->ltd_index);
2143         dt_write_unlock(env, parent);
2144
2145         GOTO(stop, rc);
2146
2147 stop:
2148         dt_trans_stop(env, dev, th);
2149
2150 unlock:
2151         lfsck_ibits_unlock(&lh, LCK_EX);
2152
2153 out:
2154         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2155                "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2156                "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2157                lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2158                PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2159                ea_off, rc);
2160
2161         return rc >= 0 ? 1 : rc;
2162 }
2163
2164 /**
2165  * \retval       +1: repaired
2166  * \retval        0: did nothing
2167  * \retval      -ve: on error
2168  */
2169 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2170                                        struct lfsck_component *com,
2171                                        struct lfsck_tgt_desc *ltd,
2172                                        struct lu_orphan_rec *rec,
2173                                        struct dt_object *parent,
2174                                        struct lu_fid *cfid,
2175                                        __u32 ost_idx, __u32 ea_off)
2176 {
2177         struct lfsck_thread_info *info          = lfsck_env_info(env);
2178         struct lu_buf            *buf           = &info->lti_big_buf;
2179         struct lu_fid            *fid           = &info->lti_fid2;
2180         struct ost_id            *oi            = &info->lti_oi;
2181         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2182         struct dt_device         *dt            = lfsck_obj2dev(parent);
2183         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2184         struct thandle            *handle       = NULL;
2185         size_t                    lovea_size;
2186         struct lov_mds_md_v1     *lmm;
2187         struct lov_ost_data_v1   *objs;
2188         struct lustre_handle      lh            = { 0 };
2189         __u32                     magic;
2190         int                       fl            = 0;
2191         int                       rc            = 0;
2192         int                       rc1;
2193         int                       i;
2194         __u16                     count;
2195         bool                      locked        = false;
2196         ENTRY;
2197
2198         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2199                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2200                               LCK_EX);
2201         if (rc != 0) {
2202                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2203                        "LOV EA for "DFID": parent "DFID", OST-index %u, "
2204                        "stripe-index %u: rc = %d\n",
2205                        lfsck_lfsck2name(lfsck), PFID(cfid),
2206                        PFID(lfsck_dto2fid(parent)), ost_idx, ea_off, rc);
2207
2208                 RETURN(rc);
2209         }
2210
2211 again:
2212         if (locked) {
2213                 dt_write_unlock(env, parent);
2214                 locked = false;
2215         }
2216
2217         if (handle != NULL) {
2218                 dt_trans_stop(env, dt, handle);
2219                 handle = NULL;
2220         }
2221
2222         if (rc < 0)
2223                 GOTO(unlock_layout, rc);
2224
2225         lovea_size = rc;
2226         if (buf->lb_len < lovea_size) {
2227                 lu_buf_realloc(buf, lovea_size);
2228                 if (buf->lb_buf == NULL)
2229                         GOTO(unlock_layout, rc = -ENOMEM);
2230         }
2231
2232         if (!(bk->lb_param & LPF_DRYRUN)) {
2233                 handle = dt_trans_create(env, dt);
2234                 if (IS_ERR(handle))
2235                         GOTO(unlock_layout, rc = PTR_ERR(handle));
2236
2237                 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2238                                           fl, handle);
2239                 if (rc != 0)
2240                         GOTO(stop, rc);
2241
2242                 rc = dt_trans_start_local(env, dt, handle);
2243                 if (rc != 0)
2244                         GOTO(stop, rc);
2245         }
2246
2247         dt_write_lock(env, parent, 0);
2248         locked = true;
2249         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2250         if (rc == -ERANGE) {
2251                 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV);
2252                 LASSERT(rc != 0);
2253                 goto again;
2254         } else if (rc == -ENODATA || rc == 0) {
2255                 lovea_size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2256                 /* If the declared is not big enough, re-try. */
2257                 if (buf->lb_len < lovea_size) {
2258                         rc = lovea_size;
2259                         goto again;
2260                 }
2261                 fl = LU_XATTR_CREATE;
2262         } else if (rc < 0) {
2263                 GOTO(unlock_parent, rc);
2264         } else if (unlikely(buf->lb_len == 0)) {
2265                 goto again;
2266         } else {
2267                 fl = LU_XATTR_REPLACE;
2268                 lovea_size = rc;
2269         }
2270
2271         if (fl == LU_XATTR_CREATE) {
2272                 if (bk->lb_param & LPF_DRYRUN)
2273                         GOTO(unlock_parent, rc = 1);
2274
2275                 LASSERT(buf->lb_len >= lovea_size);
2276
2277                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2278                                                buf, fl, ost_idx, ea_off, false);
2279
2280                 GOTO(unlock_parent, rc);
2281         }
2282
2283         lmm = buf->lb_buf;
2284         rc1 = lfsck_layout_verify_header(lmm);
2285
2286         /* If the LOV EA crashed, the rebuild it. */
2287         if (rc1 == -EINVAL) {
2288                 if (bk->lb_param & LPF_DRYRUN)
2289                         GOTO(unlock_parent, rc = 1);
2290
2291                 LASSERT(buf->lb_len >= lovea_size);
2292
2293                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2294                                                buf, fl, ost_idx, ea_off, true);
2295
2296                 GOTO(unlock_parent, rc);
2297         }
2298
2299         /* For other unknown magic/pattern, keep the current LOV EA. */
2300         if (rc1 != 0)
2301                 GOTO(unlock_parent, rc = rc1);
2302
2303         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2304          * been verified in lfsck_layout_verify_header() already. If some
2305          * new magic introduced in the future, then layout LFSCK needs to
2306          * be updated also. */
2307         magic = le32_to_cpu(lmm->lmm_magic);
2308         if (magic == LOV_MAGIC_V1) {
2309                 objs = &lmm->lmm_objects[0];
2310         } else {
2311                 LASSERT(magic == LOV_MAGIC_V3);
2312                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2313         }
2314
2315         count = le16_to_cpu(lmm->lmm_stripe_count);
2316         if (count == 0)
2317                 GOTO(unlock_parent, rc = -EINVAL);
2318         LASSERT(count > 0);
2319
2320         /* Exceed the current end of MDT-object layout EA. Then extend it. */
2321         if (count <= ea_off) {
2322                 if (bk->lb_param & LPF_DRYRUN)
2323                         GOTO(unlock_parent, rc = 1);
2324
2325                 lovea_size = lov_mds_md_size(ea_off + 1, magic);
2326                 /* If the declared is not big enough, re-try. */
2327                 if (buf->lb_len < lovea_size) {
2328                         rc = lovea_size;
2329                         goto again;
2330                 }
2331
2332                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2333                                                buf, fl, ost_idx, ea_off, false);
2334
2335                 GOTO(unlock_parent, rc);
2336         }
2337
2338         LASSERTF(rc > 0, "invalid rc = %d\n", rc);
2339
2340         for (i = 0; i < count; i++, objs++) {
2341                 /* The MDT-object was created via lfsck_layout_recover_create()
2342                  * by others before, and we fill the dummy layout EA. */
2343                 if (lovea_slot_is_dummy(objs)) {
2344                         if (i != ea_off)
2345                                 continue;
2346
2347                         if (bk->lb_param & LPF_DRYRUN)
2348                                 GOTO(unlock_parent, rc = 1);
2349
2350                         lmm->lmm_layout_gen =
2351                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2352                         rc = lfsck_layout_refill_lovea(env, handle, parent,
2353                                                        cfid, buf, objs, fl,
2354                                                        ost_idx);
2355
2356                         CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
2357                                "dummy layout slot for "DFID": parent "DFID
2358                                ", OST-index %u, stripe-index %u: rc = %d\n",
2359                                lfsck_lfsck2name(lfsck), PFID(cfid),
2360                                PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
2361
2362                         GOTO(unlock_parent, rc);
2363                 }
2364
2365                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2366                 rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
2367                 if (rc != 0) {
2368                         CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
2369                                "invalid layout EA at the slot %d, index %u\n",
2370                                lfsck_lfsck2name(lfsck),
2371                                PFID(lfsck_dto2fid(parent)), i,
2372                                le32_to_cpu(objs->l_ost_idx));
2373
2374                         GOTO(unlock_parent, rc);
2375                 }
2376
2377                 /* It should be rare case, the slot is there, but the LFSCK
2378                  * does not handle it during the first-phase cycle scanning. */
2379                 if (unlikely(lu_fid_eq(fid, cfid))) {
2380                         if (i == ea_off) {
2381                                 GOTO(unlock_parent, rc = 0);
2382                         } else {
2383                                 /* Rare case that the OST-object index
2384                                  * does not match the parent MDT-object
2385                                  * layout EA. We trust the later one. */
2386                                 if (bk->lb_param & LPF_DRYRUN)
2387                                         GOTO(unlock_parent, rc = 1);
2388
2389                                 dt_write_unlock(env, parent);
2390                                 if (handle != NULL)
2391                                         dt_trans_stop(env, dt, handle);
2392                                 lfsck_ibits_unlock(&lh, LCK_EX);
2393                                 rc = lfsck_layout_update_pfid(env, com, parent,
2394                                                         cfid, ltd->ltd_tgt, i);
2395
2396                                 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
2397                                        "updated OST-object's pfid for "DFID
2398                                        ": parent "DFID", OST-index %u, "
2399                                        "stripe-index %u: rc = %d\n",
2400                                        lfsck_lfsck2name(lfsck), PFID(cfid),
2401                                        PFID(lfsck_dto2fid(parent)),
2402                                        ltd->ltd_index, i, rc);
2403
2404                                 RETURN(rc);
2405                         }
2406                 }
2407         }
2408
2409         /* The MDT-object exists, but related layout EA slot is occupied
2410          * by others. */
2411         if (bk->lb_param & LPF_DRYRUN)
2412                 GOTO(unlock_parent, rc = 1);
2413
2414         dt_write_unlock(env, parent);
2415         if (handle != NULL)
2416                 dt_trans_stop(env, dt, handle);
2417         lfsck_ibits_unlock(&lh, LCK_EX);
2418         if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
2419                 objs = &lmm->lmm_objects[ea_off];
2420         else
2421                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
2422         rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
2423                                           buf, objs, ea_off);
2424
2425         RETURN(rc);
2426
2427 unlock_parent:
2428         if (locked)
2429                 dt_write_unlock(env, parent);
2430
2431 stop:
2432         if (handle != NULL)
2433                 dt_trans_stop(env, dt, handle);
2434
2435 unlock_layout:
2436         lfsck_ibits_unlock(&lh, LCK_EX);
2437
2438         return rc;
2439 }
2440
2441 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
2442                                         struct lfsck_component *com,
2443                                         struct lfsck_tgt_desc *ltd,
2444                                         struct lu_orphan_rec *rec,
2445                                         struct lu_fid *cfid)
2446 {
2447         struct lfsck_layout     *lo     = com->lc_file_ram;
2448         struct lu_fid           *pfid   = &rec->lor_fid;
2449         struct dt_object        *parent = NULL;
2450         __u32                    ea_off = pfid->f_stripe_idx;
2451         int                      rc     = 0;
2452         ENTRY;
2453
2454         if (!fid_is_sane(cfid))
2455                 GOTO(out, rc = -EINVAL);
2456
2457         if (fid_is_zero(pfid)) {
2458                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2459                                                   "", "N", ea_off);
2460                 GOTO(out, rc);
2461         }
2462
2463         pfid->f_ver = 0;
2464         if (!fid_is_sane(pfid))
2465                 GOTO(out, rc = -EINVAL);
2466
2467         parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
2468         if (IS_ERR(parent))
2469                 GOTO(out, rc = PTR_ERR(parent));
2470
2471         if (unlikely(dt_object_remote(parent) != 0))
2472                 GOTO(put, rc = -EXDEV);
2473
2474         if (dt_object_exists(parent) == 0) {
2475                 lfsck_object_put(env, parent);
2476                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2477                                                   "", "R", ea_off);
2478                 GOTO(out, rc);
2479         }
2480
2481         if (!S_ISREG(lu_object_attr(&parent->do_lu)))
2482                 GOTO(put, rc = -EISDIR);
2483
2484         rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
2485                                          ltd->ltd_index, ea_off);
2486
2487         GOTO(put, rc);
2488
2489 put:
2490         if (rc <= 0)
2491                 lfsck_object_put(env, parent);
2492         else
2493                 /* The layout EA is changed, need to be reloaded next time. */
2494                 lu_object_put_nocache(env, &parent->do_lu);
2495
2496 out:
2497         down_write(&com->lc_sem);
2498         com->lc_new_scanned++;
2499         com->lc_new_checked++;
2500         if (rc > 0) {
2501                 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
2502                 rc = 0;
2503         } else if (rc < 0) {
2504                 lo->ll_objs_failed_phase2++;
2505         }
2506         up_write(&com->lc_sem);
2507
2508         return rc;
2509 }
2510
2511 static int lfsck_layout_scan_orphan(const struct lu_env *env,
2512                                     struct lfsck_component *com,
2513                                     struct lfsck_tgt_desc *ltd)
2514 {
2515         struct lfsck_assistant_data     *lad    = com->lc_data;
2516         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2517         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
2518         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2519         struct ost_id                   *oi     = &info->lti_oi;
2520         struct lu_fid                   *fid    = &info->lti_fid;
2521         struct dt_object                *obj;
2522         const struct dt_it_ops          *iops;
2523         struct dt_it                    *di;
2524         int                              rc     = 0;
2525         ENTRY;
2526
2527         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
2528                "scanning for OST%04x\n",
2529                lfsck_lfsck2name(lfsck), ltd->ltd_index);
2530
2531         if (cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
2532                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
2533                        "scanning for OST%04x\n",
2534                        lfsck_lfsck2name(lfsck), ltd->ltd_index);
2535
2536                 RETURN(0);
2537         }
2538
2539         ostid_set_seq(oi, FID_SEQ_IDIF);
2540         ostid_set_id(oi, 0);
2541         rc = ostid_to_fid(fid, oi, ltd->ltd_index);
2542         if (rc != 0)
2543                 GOTO(log, rc);
2544
2545         obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
2546         if (unlikely(IS_ERR(obj)))
2547                 GOTO(log, rc = PTR_ERR(obj));
2548
2549         rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_orphan_features);
2550         if (rc != 0)
2551                 GOTO(put, rc);
2552
2553         iops = &obj->do_index_ops->dio_it;
2554         di = iops->init(env, obj, 0);
2555         if (IS_ERR(di))
2556                 GOTO(put, rc = PTR_ERR(di));
2557
2558         rc = iops->load(env, di, 0);
2559         if (rc == -ESRCH) {
2560                 /* -ESRCH means that the orphan OST-objects rbtree has been
2561                  * cleanup because of the OSS server restart or other errors. */
2562                 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
2563                 GOTO(fini, rc);
2564         }
2565
2566         if (rc == 0)
2567                 rc = iops->next(env, di);
2568         else if (rc > 0)
2569                 rc = 0;
2570
2571         if (rc < 0)
2572                 GOTO(fini, rc);
2573
2574         if (rc > 0)
2575                 GOTO(fini, rc = 0);
2576
2577         do {
2578                 struct dt_key           *key;
2579                 struct lu_orphan_rec    *rec = &info->lti_rec;
2580
2581                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
2582                     unlikely(!thread_is_running(&lfsck->li_thread)))
2583                         break;
2584
2585                 key = iops->key(env, di);
2586                 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
2587                 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
2588                 if (rc == 0)
2589                         rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
2590                                         &com->lc_fid_latest_scanned_phase2);
2591                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2592                         GOTO(fini, rc);
2593
2594                 lfsck_control_speed_by_self(com);
2595                 do {
2596                         rc = iops->next(env, di);
2597                 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
2598         } while (rc == 0);
2599
2600         GOTO(fini, rc);
2601
2602 fini:
2603         iops->put(env, di);
2604         iops->fini(env, di);
2605 put:
2606         lfsck_object_put(env, obj);
2607
2608 log:
2609         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
2610                "scanning for OST%04x: rc = %d\n",
2611                lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
2612
2613         return rc > 0 ? 0 : rc;
2614 }
2615
2616 /* For the MDT-object with dangling reference, we need to repare the
2617  * inconsistency according to the LFSCK sponsor's requirement:
2618  *
2619  * 1) Keep the inconsistency there and report the inconsistency case,
2620  *    then give the chance to the application to find related issues,
2621  *    and the users can make the decision about how to handle it with
2622  *    more human knownledge. (by default)
2623  *
2624  * 2) Re-create the missing OST-object with the FID/owner information. */
2625 static int lfsck_layout_repair_dangling(const struct lu_env *env,
2626                                         struct lfsck_component *com,
2627                                         struct dt_object *parent,
2628                                         struct lfsck_layout_req *llr,
2629                                         struct lu_attr *la)
2630 {
2631         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2632         struct filter_fid               *pfid   = &info->lti_new_pfid;
2633         struct dt_object_format         *dof    = &info->lti_dof;
2634         struct dt_object                *child  = llr->llr_child;
2635         struct dt_device                *dev    = lfsck_obj2dev(child);
2636         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
2637         struct thandle                  *handle;
2638         struct lu_buf                   *buf;
2639         struct lustre_handle             lh     = { 0 };
2640         int                              rc;
2641         bool                             create;
2642         ENTRY;
2643
2644         if (com->lc_lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ)
2645                 create = true;
2646         else
2647                 create = false;
2648
2649         if (!create)
2650                 GOTO(log, rc = 1);
2651
2652         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2653                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2654                               LCK_EX);
2655         if (rc != 0)
2656                 GOTO(log, rc);
2657
2658         rc = dt_attr_get(env, parent, la);
2659         if (rc != 0)
2660                 GOTO(unlock1, rc);
2661
2662         la->la_mode = S_IFREG | 0666;
2663         la->la_atime = la->la_mtime = la->la_ctime = 0;
2664         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
2665                        LA_ATIME | LA_MTIME | LA_CTIME;
2666         memset(dof, 0, sizeof(*dof));
2667         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
2668         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
2669         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2670          * MDT-object's FID::f_ver, instead it is the OST-object index in its
2671          * parent MDT-object's layout EA. */
2672         pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
2673         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
2674
2675         handle = dt_trans_create(env, dev);
2676         if (IS_ERR(handle))
2677                 GOTO(unlock1, rc = PTR_ERR(handle));
2678
2679         rc = dt_declare_create(env, child, la, NULL, dof, handle);
2680         if (rc != 0)
2681                 GOTO(stop, rc);
2682
2683         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
2684                                   LU_XATTR_CREATE, handle);
2685         if (rc != 0)
2686                 GOTO(stop, rc);
2687
2688         rc = dt_trans_start_local(env, dev, handle);
2689         if (rc != 0)
2690                 GOTO(stop, rc);
2691
2692         dt_read_lock(env, parent, 0);
2693         if (unlikely(lfsck_is_dead_obj(parent)))
2694                 GOTO(unlock2, rc = 1);
2695
2696         rc = dt_create(env, child, la, NULL, dof, handle);
2697         if (rc != 0)
2698                 GOTO(unlock2, rc);
2699
2700         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
2701                           handle);
2702
2703         GOTO(unlock2, rc);
2704
2705 unlock2:
2706         dt_read_unlock(env, parent);
2707
2708 stop:
2709         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2710
2711 unlock1:
2712         lfsck_ibits_unlock(&lh, LCK_EX);
2713
2714 log:
2715         if (rc != 0)
2716                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
2717                        "dangling reference for: parent "DFID", child "DFID
2718                        ", OST-index %u, stripe-index %u, owner %u/%u. %s: "
2719                        "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
2720                        PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
2721                        llr->llr_ost_idx, llr->llr_lov_idx,
2722                        la->la_uid, la->la_gid,
2723                        create ? "Create the lost OST-object as required" :
2724                                 "Keep the MDT-object there by default", rc);
2725
2726         return rc;
2727 }
2728
2729 /* If the OST-object does not recognize the MDT-object as its parent, and
2730  * there is no other MDT-object claims as its parent, then just trust the
2731  * given MDT-object as its parent. So update the OST-object filter_fid. */
2732 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
2733                                               struct lfsck_component *com,
2734                                               struct dt_object *parent,
2735                                               struct lfsck_layout_req *llr,
2736                                               struct lu_attr *la)
2737 {
2738         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2739         struct filter_fid               *pfid   = &info->lti_new_pfid;
2740         struct dt_object                *child  = llr->llr_child;
2741         struct dt_device                *dev    = lfsck_obj2dev(child);
2742         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
2743         struct thandle                  *handle;
2744         struct lu_buf                   *buf;
2745         struct lustre_handle             lh     = { 0 };
2746         int                              rc;
2747         ENTRY;
2748
2749         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2750                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2751                               LCK_EX);
2752         if (rc != 0)
2753                 GOTO(log, rc);
2754
2755         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
2756         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
2757         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2758          * MDT-object's FID::f_ver, instead it is the OST-object index in its
2759          * parent MDT-object's layout EA. */
2760         pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
2761         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
2762
2763         handle = dt_trans_create(env, dev);
2764         if (IS_ERR(handle))
2765                 GOTO(unlock1, rc = PTR_ERR(handle));
2766
2767         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
2768         if (rc != 0)
2769                 GOTO(stop, rc);
2770
2771         rc = dt_attr_get(env, parent, la);
2772         if (rc != 0)
2773                 GOTO(stop, rc);
2774
2775         la->la_valid = LA_UID | LA_GID;
2776         rc = dt_declare_attr_set(env, child, la, handle);
2777         if (rc != 0)
2778                 GOTO(stop, rc);
2779
2780         rc = dt_trans_start_local(env, dev, handle);
2781         if (rc != 0)
2782                 GOTO(stop, rc);
2783
2784         dt_write_lock(env, parent, 0);
2785         if (unlikely(lfsck_is_dead_obj(parent)))
2786                 GOTO(unlock2, rc = 1);
2787
2788         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
2789         if (rc != 0)
2790                 GOTO(unlock2, rc);
2791
2792         /* Get the latest parent's owner. */
2793         rc = dt_attr_get(env, parent, la);
2794         if (rc != 0)
2795                 GOTO(unlock2, rc);
2796
2797         la->la_valid = LA_UID | LA_GID;
2798         rc = dt_attr_set(env, child, la, handle);
2799
2800         GOTO(unlock2, rc);
2801
2802 unlock2:
2803         dt_write_unlock(env, parent);
2804
2805 stop:
2806         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2807
2808 unlock1:
2809         lfsck_ibits_unlock(&lh, LCK_EX);
2810
2811 log:
2812         if (rc != 0)
2813                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
2814                        "unmatched MDT-OST pair for: parent "DFID
2815                        ", child "DFID", OST-index %u, stripe-index %u, "
2816                        "owner %u/%u: rc = %d\n",
2817                        lfsck_lfsck2name(com->lc_lfsck),
2818                        PFID(lfsck_dto2fid(parent)),
2819                        PFID(lfsck_dto2fid(child)),
2820                        llr->llr_ost_idx, llr->llr_lov_idx,
2821                        la->la_uid, la->la_gid, rc);
2822
2823         return rc;
2824 }
2825
2826 /* If there are more than one MDT-objects claim as the OST-object's parent,
2827  * and the OST-object only recognizes one of them, then we need to generate
2828  * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
2829 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
2830                                                    struct lfsck_component *com,
2831                                                    struct dt_object *parent,
2832                                                    struct lfsck_layout_req *llr,
2833                                                    struct lu_attr *la,
2834                                                    struct lu_buf *buf)
2835 {
2836         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2837         struct dt_allocation_hint       *hint   = &info->lti_hint;
2838         struct dt_object_format         *dof    = &info->lti_dof;
2839         struct ost_id                   *oi     = &info->lti_oi;
2840         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2841         struct dt_device                *dev;
2842         struct lu_device                *d      =
2843                                 &lfsck_obj2dev(llr->llr_child)->dd_lu_dev;
2844         struct lu_object                *o;
2845         struct lu_object                *n;
2846         struct dt_object                *child  = NULL;
2847         struct thandle                  *handle = NULL;
2848         struct lov_mds_md_v1            *lmm;
2849         struct lov_ost_data_v1          *objs;
2850         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
2851         struct lu_fid                    tfid;
2852         struct lustre_handle             lh     = { 0 };
2853         struct lu_buf                    ea_buf;
2854         __u32                            magic;
2855         __u32                            index;
2856         int                              rc;
2857         ENTRY;
2858
2859         /* We use two separated transactions to repair the inconsistency.
2860          *
2861          * 1) create the child (OST-object).
2862          * 2) update the parent LOV EA according to the child's FID.
2863          *
2864          * If 1) succeed, but 2) failed or aborted, then such OST-object will be
2865          * handled as orphan when the layout LFSCK run next time.
2866          *
2867          * If 1) failed, but 2) succeed, then such OST-object will be re-created
2868          * as dangling referened case when the layout LFSCK run next time. */
2869
2870         /* The 1st transaction. */
2871         o = lu_object_anon(env, d, NULL);
2872         if (IS_ERR(o))
2873                 GOTO(log, rc = PTR_ERR(o));
2874
2875         n = lu_object_locate(o->lo_header, d->ld_type);
2876         if (unlikely(n == NULL)) {
2877                 lu_object_put_nocache(env, o);
2878
2879                 GOTO(log, rc = -EINVAL);
2880         }
2881
2882         child = container_of(n, struct dt_object, do_lu);
2883         memset(hint, 0, sizeof(*hint));
2884         rc = dt_attr_get(env, parent, la);
2885         if (rc != 0)
2886                 GOTO(log, rc);
2887
2888         la->la_valid = LA_UID | LA_GID;
2889         memset(dof, 0, sizeof(*dof));
2890
2891         dev = lfsck_obj2dev(child);
2892         handle = dt_trans_create(env, dev);
2893         if (IS_ERR(handle))
2894                 GOTO(log, rc = PTR_ERR(handle));
2895
2896         rc = dt_declare_create(env, child, la, hint, dof, handle);
2897         if (rc != 0)
2898                 GOTO(stop, rc);
2899
2900         rc = dt_trans_start_local(env, dev, handle);
2901         if (rc != 0)
2902                 GOTO(stop, rc);
2903
2904         rc = dt_create(env, child, la, hint, dof, handle);
2905         dt_trans_stop(env, dev, handle);
2906         handle = NULL;
2907         if (rc != 0)
2908                 GOTO(log, rc);
2909
2910         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2911                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2912                               LCK_EX);
2913         if (rc != 0)
2914                 GOTO(log, rc);
2915
2916         /* The 2nd transaction. */
2917
2918         /* XXX: Generally, we should use bottom device (OSD) to update parent
2919          *      LOV EA. But because the LOD-object still references the wrong
2920          *      OSP-object that should be detached after the parent's LOV EA
2921          *      refreshed. Unfortunately, there is no suitable API for that.
2922          *      So we have to make the LOD to re-load the OSP-object(s) via
2923          *      replacing the LOV EA against the LOD-object.
2924          *
2925          *      Once the DNE2 patches have been landed, we can replace the
2926          *      LOD device with the OSD device. LU-6230. */
2927
2928         dev = lfsck->li_next;
2929         parent = lfsck_object_locate(dev, parent);
2930         if (IS_ERR(parent))
2931                 GOTO(log, rc = PTR_ERR(parent));
2932
2933         handle = dt_trans_create(env, dev);
2934         if (IS_ERR(handle))
2935                 GOTO(log, rc = PTR_ERR(handle));
2936
2937         rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2938                                   LU_XATTR_REPLACE, handle);
2939         if (rc != 0)
2940                 GOTO(stop, rc);
2941
2942         rc = dt_trans_start_local(env, dev, handle);
2943         if (rc != 0)
2944                 GOTO(stop, rc);
2945
2946         dt_write_lock(env, parent, 0);
2947         if (unlikely(lfsck_is_dead_obj(parent)))
2948                 GOTO(unlock, rc = 0);
2949
2950         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2951         if (unlikely(rc == 0 || rc == -ENODATA || rc == -ERANGE))
2952                 GOTO(unlock, rc = 0);
2953
2954         lmm = buf->lb_buf;
2955         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2956          * been verified in lfsck_layout_verify_header() already. If some
2957          * new magic introduced in the future, then layout LFSCK needs to
2958          * be updated also. */
2959         magic = le32_to_cpu(lmm->lmm_magic);
2960         if (magic == LOV_MAGIC_V1) {
2961                 objs = &lmm->lmm_objects[llr->llr_lov_idx];
2962         } else {
2963                 LASSERT(magic == LOV_MAGIC_V3);
2964                 objs =
2965                 &((struct lov_mds_md_v3 *)lmm)->lmm_objects[llr->llr_lov_idx];
2966         }
2967
2968         ostid_le_to_cpu(&objs->l_ost_oi, oi);
2969         index = le32_to_cpu(objs->l_ost_idx);
2970         rc = ostid_to_fid(&tfid, oi, index);
2971         /* Someone changed layout during the LFSCK, no need to repair then. */
2972         if (rc == 0 && !lu_fid_eq(&tfid, lu_object_fid(&llr->llr_child->do_lu)))
2973                 GOTO(unlock, rc = 0);
2974
2975         lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2976         fid_to_ostid(lu_object_fid(&child->do_lu), oi);
2977         ostid_cpu_to_le(oi, &objs->l_ost_oi);
2978         objs->l_ost_gen = cpu_to_le32(0);
2979         objs->l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
2980         lfsck_buf_init(&ea_buf, lmm,
2981                        lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
2982                                        magic));
2983         rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV,
2984                           LU_XATTR_REPLACE, handle);
2985
2986         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2987
2988 unlock:
2989         dt_write_unlock(env, parent);
2990
2991 stop:
2992         if (handle != NULL)
2993                 dt_trans_stop(env, dev, handle);
2994
2995 log:
2996         lfsck_ibits_unlock(&lh, LCK_EX);
2997         if (child != NULL)
2998                 lfsck_object_put(env, child);
2999
3000         if (rc != 0)
3001                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3002                        "multiple references for: parent "DFID", OST-index %u, "
3003                        "stripe-index %u, owner %u/%u: rc = %d\n",
3004                        lfsck_lfsck2name(lfsck), PFID(pfid), llr->llr_ost_idx,
3005                        llr->llr_lov_idx, la->la_uid, la->la_gid, rc);
3006
3007         return rc;
3008 }
3009
3010 /* If the MDT-object and the OST-object have different owner information,
3011  * then trust the MDT-object, because the normal chown/chgrp handle order
3012  * is from MDT to OST, and it is possible that some chown/chgrp operation
3013  * is partly done. */
3014 static int lfsck_layout_repair_owner(const struct lu_env *env,
3015                                      struct lfsck_component *com,
3016                                      struct dt_object *parent,
3017                                      struct lfsck_layout_req *llr,
3018                                      struct lu_attr *pla,
3019                                      const struct lu_attr *cla)
3020 {
3021         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3022         struct lu_attr                  *tla    = &info->lti_la2;
3023         struct dt_object                *child  = llr->llr_child;
3024         struct dt_device                *dev    = lfsck_obj2dev(child);
3025         struct thandle                  *handle;
3026         int                              rc;
3027         ENTRY;
3028
3029         tla->la_uid = pla->la_uid;
3030         tla->la_gid = pla->la_gid;
3031         tla->la_valid = LA_UID | LA_GID;
3032         handle = dt_trans_create(env, dev);
3033         if (IS_ERR(handle))
3034                 GOTO(log, rc = PTR_ERR(handle));
3035
3036         rc = dt_declare_attr_set(env, child, tla, handle);
3037         if (rc != 0)
3038                 GOTO(stop, rc);
3039
3040         rc = dt_trans_start_local(env, dev, handle);
3041         if (rc != 0)
3042                 GOTO(stop, rc);
3043
3044         /* Use the dt_object lock to serialize with destroy and attr_set. */
3045         dt_read_lock(env, parent, 0);
3046         if (unlikely(lfsck_is_dead_obj(parent)))
3047                 GOTO(unlock, rc = 1);
3048
3049         /* Get the latest parent's owner. */
3050         rc = dt_attr_get(env, parent, pla);
3051         if (rc != 0)
3052                 GOTO(unlock, rc);
3053
3054         /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
3055         if (unlikely(tla->la_uid != pla->la_uid ||
3056                      tla->la_gid != pla->la_gid))
3057                 rc = 1;
3058         else
3059                 rc = dt_attr_set(env, child, tla, handle);
3060
3061         GOTO(unlock, rc);
3062
3063 unlock:
3064         dt_read_unlock(env, parent);
3065
3066 stop:
3067         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3068
3069 log:
3070         if (rc != 0)
3071                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3072                        "inconsistent file owner for: parent "DFID", child "DFID
3073                        ", OST-index %u, stripe-index %u, old owner %u/%u, "
3074                        "new owner %u/%u: rc = %d\n",
3075                        lfsck_lfsck2name(com->lc_lfsck),
3076                        PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
3077                        llr->llr_ost_idx, llr->llr_lov_idx,
3078                        cla->la_uid, cla->la_gid, tla->la_uid, tla->la_gid, rc);
3079
3080         return rc;
3081 }
3082
3083 /* Check whether the OST-object correctly back points to the
3084  * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
3085 static int lfsck_layout_check_parent(const struct lu_env *env,
3086                                      struct lfsck_component *com,
3087                                      struct lfsck_assistant_object *lso,
3088                                      const struct lu_fid *pfid,
3089                                      const struct lu_fid *cfid,
3090                                      const struct lu_attr *cla,
3091                                      struct lfsck_layout_req *llr,
3092                                      struct lu_buf *lov_ea, __u32 idx)
3093 {
3094         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3095         struct lu_buf                   *buf    = &info->lti_big_buf;
3096         struct dt_object                *tobj;
3097         struct lov_mds_md_v1            *lmm;
3098         struct lov_ost_data_v1          *objs;
3099         struct lustre_handle             lh     = { 0 };
3100         int                              rc;
3101         int                              i;
3102         __u32                            magic;
3103         __u16                            count;
3104         ENTRY;
3105
3106         if (unlikely(!fid_is_sane(pfid)))
3107                 RETURN(LLIT_UNMATCHED_PAIR);
3108
3109         if (lu_fid_eq(pfid, &lso->lso_fid)) {
3110                 if (likely(llr->llr_lov_idx == idx))
3111                         RETURN(0);
3112
3113                 RETURN(LLIT_UNMATCHED_PAIR);
3114         }
3115
3116         tobj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3117         if (IS_ERR(tobj))
3118                 RETURN(PTR_ERR(tobj));
3119
3120         if (dt_object_exists(tobj) == 0 || lfsck_is_dead_obj(tobj) ||
3121             !S_ISREG(lfsck_object_type(tobj)))
3122                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3123
3124         /* Load the tobj's layout EA, in spite of it is a local MDT-object or
3125          * remote one on another MDT. Then check whether the given OST-object
3126          * is in such layout. If yes, it is multiple referenced, otherwise it
3127          * is unmatched referenced case. */
3128         rc = lfsck_layout_get_lovea(env, tobj, buf);
3129         if (rc == 0 || rc == -ENOENT)
3130                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3131
3132         if (rc < 0)
3133                 GOTO(out, rc);
3134
3135         lmm = buf->lb_buf;
3136         magic = le32_to_cpu(lmm->lmm_magic);
3137         if (magic == LOV_MAGIC_V1) {
3138                 objs = &lmm->lmm_objects[0];
3139         } else {
3140                 LASSERT(magic == LOV_MAGIC_V3);
3141                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3142         }
3143
3144         count = le16_to_cpu(lmm->lmm_stripe_count);
3145         for (i = 0; i < count; i++, objs++) {
3146                 struct lu_fid           *tfid   = &info->lti_fid2;
3147                 struct ost_id           *oi     = &info->lti_oi;
3148                 __u32                    idx2;
3149
3150                 if (lovea_slot_is_dummy(objs))
3151                         continue;
3152
3153                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3154                 idx2 = le32_to_cpu(objs->l_ost_idx);
3155                 rc = ostid_to_fid(tfid, oi, idx2);
3156                 if (rc != 0) {
3157                         CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
3158                                "invalid layout EA at the slot %d, index %u\n",
3159                                lfsck_lfsck2name(com->lc_lfsck),
3160                                PFID(pfid), i, idx2);
3161
3162                         GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3163                 }
3164
3165                 if (lu_fid_eq(cfid, tfid)) {
3166                         rc = lfsck_ibits_lock(env, com->lc_lfsck, tobj, &lh,
3167                                               MDS_INODELOCK_UPDATE |
3168                                               MDS_INODELOCK_LAYOUT |
3169                                               MDS_INODELOCK_XATTR,
3170                                               LCK_EX);
3171                         if (rc != 0)
3172                                 GOTO(out, rc);
3173
3174                         dt_read_lock(env, tobj, 0);
3175
3176                         /* For local MDT-object, re-check existence
3177                          * after taken the lock. */
3178                         if (!dt_object_remote(tobj)) {
3179                                 if (dt_object_exists(tobj) == 0 ||
3180                                     lfsck_is_dead_obj(tobj)) {
3181                                         rc = LLIT_UNMATCHED_PAIR;
3182                                 } else {
3183                                         *lov_ea = *buf;
3184                                         rc = LLIT_MULTIPLE_REFERENCED;
3185                                 }
3186
3187                                 GOTO(unlock, rc);
3188                         }
3189
3190                         /* For migration case, the new MDT-object and old
3191                          * MDT-object may reference the same OST-object at
3192                          * some migration internal time.
3193                          *
3194                          * For remote MDT-object, the local MDT may not know
3195                          * whether it has been removed or not.  Try checking
3196                          * for a non-existent xattr to check if this object
3197                          * has been been removed or not. */
3198                         rc = dt_xattr_get(env, tobj, &LU_BUF_NULL,
3199                                           XATTR_NAME_DUMMY);
3200                         if (unlikely(rc == -ENOENT || rc >= 0)) {
3201                                 rc = LLIT_UNMATCHED_PAIR;
3202                         } else if (rc == -ENODATA) {
3203                                 *lov_ea = *buf;
3204                                 rc = LLIT_MULTIPLE_REFERENCED;
3205                         }
3206
3207                         GOTO(unlock, rc);
3208                 }
3209         }
3210
3211         GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3212
3213 unlock:
3214         if (lustre_handle_is_used(&lh)) {
3215                 dt_read_unlock(env, tobj);
3216                 lfsck_ibits_unlock(&lh, LCK_EX);
3217         }
3218
3219 out:
3220         lfsck_object_put(env, tobj);
3221
3222         return rc;
3223 }
3224
3225 static int lfsck_layout_assistant_handler_p1(const struct lu_env *env,
3226                                              struct lfsck_component *com,
3227                                              struct lfsck_assistant_req *lar)
3228 {
3229         struct lfsck_layout_req              *llr    =
3230                         container_of0(lar, struct lfsck_layout_req, llr_lar);
3231         struct lfsck_assistant_object        *lso    = lar->lar_parent;
3232         struct lfsck_layout                  *lo     = com->lc_file_ram;
3233         struct lfsck_thread_info             *info   = lfsck_env_info(env);
3234         struct filter_fid_old                *pea    = &info->lti_old_pfid;
3235         struct lu_fid                        *pfid   = &info->lti_fid;
3236         struct lu_buf                         buf    = { NULL };
3237         struct dt_object                     *parent = NULL;
3238         struct dt_object                     *child  = llr->llr_child;
3239         struct lu_attr                       *pla    = &lso->lso_attr;
3240         struct lu_attr                       *cla    = &info->lti_la;
3241         struct lfsck_instance                *lfsck  = com->lc_lfsck;
3242         struct lfsck_bookmark                *bk     = &lfsck->li_bookmark_ram;
3243         enum lfsck_layout_inconsistency_type  type   = LLIT_NONE;
3244         __u32                                 idx    = 0;
3245         int                                   rc;
3246         ENTRY;
3247
3248         if (lso->lso_dead)
3249                 RETURN(0);
3250
3251         CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_ASSISTANT_DIRECT, cfs_fail_val);
3252
3253         rc = dt_attr_get(env, child, cla);
3254         if (rc == -ENOENT) {
3255                 parent = lfsck_assistant_object_load(env, lfsck, lso);
3256                 if (IS_ERR(parent)) {
3257                         rc = PTR_ERR(parent);
3258
3259                         RETURN(rc == -ENOENT ? 0 : rc);
3260                 }
3261
3262                 type = LLIT_DANGLING;
3263                 goto repair;
3264         }
3265
3266         if (rc != 0)
3267                 GOTO(out, rc);
3268
3269         lfsck_buf_init(&buf, pea, sizeof(struct filter_fid_old));
3270         rc = dt_xattr_get(env, child, &buf, XATTR_NAME_FID);
3271         if (unlikely(rc > 0 && rc != sizeof(struct filter_fid_old) &&
3272                      rc != sizeof(struct filter_fid))) {
3273                 type = LLIT_UNMATCHED_PAIR;
3274                 goto repair;
3275         }
3276
3277         if (rc < 0 && rc != -ENODATA)
3278                 GOTO(out, rc);
3279
3280         if (rc == 0 || rc == -ENODATA)
3281                 GOTO(check_owner, rc = 0);
3282
3283         fid_le_to_cpu(pfid, &pea->ff_parent);
3284         /* Currently, the filter_fid::ff_parent::f_ver is not the
3285          * real parent MDT-object's FID::f_ver, instead it is the
3286          * OST-object index in its parent MDT-object's layout EA. */
3287         idx = pfid->f_stripe_idx;
3288         pfid->f_ver = 0;
3289         rc = lfsck_layout_check_parent(env, com, lso, pfid,
3290                                        lu_object_fid(&child->do_lu),
3291                                        cla, llr, &buf, idx);
3292         if (rc > 0) {
3293                 type = rc;
3294                 goto repair;
3295         }
3296
3297         if (rc < 0)
3298                 GOTO(out, rc);
3299
3300 check_owner:
3301         /* Someone may has changed the owner after the parent attr pre-loaded.
3302          * It can be handled later inside the lfsck_layout_repair_owner(). */
3303         if (unlikely(cla->la_uid != pla->la_uid ||
3304                      cla->la_gid != pla->la_gid)) {
3305                 type = LLIT_INCONSISTENT_OWNER;
3306                 goto repair;
3307         }
3308
3309 repair:
3310         if (type == LLIT_NONE)
3311                 GOTO(out, rc = 0);
3312
3313         if (bk->lb_param & LPF_DRYRUN)
3314                 GOTO(out, rc = 1);
3315
3316         if (parent == NULL) {
3317                 parent = lfsck_assistant_object_load(env, lfsck, lso);
3318                 if (IS_ERR(parent)) {
3319                         rc = PTR_ERR(parent);
3320
3321                         if (rc == -ENOENT)
3322                                 RETURN(0);
3323
3324                         GOTO(out, rc);
3325                 }
3326         }
3327
3328         switch (type) {
3329         case LLIT_DANGLING:
3330                 rc = lfsck_layout_repair_dangling(env, com, parent, llr, pla);
3331                 break;
3332         case LLIT_UNMATCHED_PAIR:
3333                 rc = lfsck_layout_repair_unmatched_pair(env, com, parent,
3334                                                         llr, pla);
3335                 break;
3336         case LLIT_MULTIPLE_REFERENCED:
3337                 rc = lfsck_layout_repair_multiple_references(env, com, parent,
3338                                                              llr, pla, &buf);
3339                 break;
3340         case LLIT_INCONSISTENT_OWNER:
3341                 rc = lfsck_layout_repair_owner(env, com, parent, llr, pla, cla);
3342                 break;
3343         default:
3344                 rc = 0;
3345                 break;
3346         }
3347
3348         GOTO(out, rc);
3349
3350 out:
3351         down_write(&com->lc_sem);
3352         if (rc < 0) {
3353                 struct lfsck_assistant_data *lad = com->lc_data;
3354
3355                 if (unlikely(lad->lad_exit)) {
3356                         rc = 0;
3357                 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
3358                            rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
3359                            rc == -EHOSTUNREACH) {
3360                         /* If cannot touch the target server,
3361                          * mark the LFSCK as INCOMPLETE. */
3362                         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
3363                                "talk with OST %x: rc = %d\n",
3364                                lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
3365                         lfsck_lad_set_bitmap(env, com, llr->llr_ost_idx);
3366                         lo->ll_objs_skipped++;
3367                         rc = 0;
3368                 } else {
3369                         lfsck_layout_record_failure(env, lfsck, lo);
3370                 }
3371         } else if (rc > 0) {
3372                 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
3373                          "unknown type = %d\n", type);
3374
3375                 lo->ll_objs_repaired[type - 1]++;
3376                 if (bk->lb_param & LPF_DRYRUN &&
3377                     unlikely(lo->ll_pos_first_inconsistent == 0))
3378                         lo->ll_pos_first_inconsistent =
3379                         lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
3380                                                         lfsck->li_di_oit);
3381         }
3382         up_write(&com->lc_sem);
3383
3384         if (parent != NULL && !IS_ERR(parent))
3385                 lfsck_object_put(env, parent);
3386
3387         return rc;
3388 }
3389
3390 static int lfsck_layout_assistant_handler_p2(const struct lu_env *env,
3391                                              struct lfsck_component *com)
3392 {
3393         struct lfsck_assistant_data     *lad    = com->lc_data;
3394         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3395         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
3396         struct lfsck_tgt_descs          *ltds   = &lfsck->li_ost_descs;
3397         struct lfsck_tgt_desc           *ltd;
3398         int                              rc     = 0;
3399         ENTRY;
3400
3401         CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan start\n",
3402                lfsck_lfsck2name(lfsck));
3403
3404         spin_lock(&ltds->ltd_lock);
3405         while (!list_empty(&lad->lad_ost_phase2_list)) {
3406                 ltd = list_entry(lad->lad_ost_phase2_list.next,
3407                                  struct lfsck_tgt_desc,
3408                                  ltd_layout_phase_list);
3409                 list_del_init(&ltd->ltd_layout_phase_list);
3410                 if (bk->lb_param & LPF_ALL_TGT) {
3411                         spin_unlock(&ltds->ltd_lock);
3412                         rc = lfsck_layout_scan_orphan(env, com, ltd);
3413                         if (rc != 0 && bk->lb_param & LPF_FAILOUT)
3414                                 RETURN(rc);
3415
3416                         if (unlikely(lad->lad_exit ||
3417                                      !thread_is_running(&lfsck->li_thread)))
3418                                 RETURN(0);
3419                         spin_lock(&ltds->ltd_lock);
3420                 }
3421         }
3422
3423         if (list_empty(&lad->lad_ost_phase1_list))
3424                 rc = 1;
3425         else
3426                 rc = 0;
3427         spin_unlock(&ltds->ltd_lock);
3428
3429         CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan stop: rc = %d\n",
3430                lfsck_lfsck2name(lfsck), rc);
3431
3432         RETURN(rc);
3433 }
3434
3435 static int
3436 lfsck_layout_slave_async_interpret(const struct lu_env *env,
3437                                    struct ptlrpc_request *req,
3438                                    void *args, int rc)
3439 {
3440         struct lfsck_layout_slave_async_args *llsaa = args;
3441         struct obd_export                    *exp   = llsaa->llsaa_exp;
3442         struct lfsck_component               *com   = llsaa->llsaa_com;
3443         struct lfsck_layout_slave_target     *llst  = llsaa->llsaa_llst;
3444         struct lfsck_layout_slave_data       *llsd  = com->lc_data;
3445         struct lfsck_reply                   *lr    = NULL;
3446         bool                                  done  = false;
3447
3448         if (rc != 0) {
3449                 /* It is probably caused by network trouble, or target crash,
3450                  * it will try several times (depends on the obd_timeout, and
3451                  * will not less than 3 times). But to make the LFSCK can go
3452                  * ahead, we should not try for ever. After some try but still
3453                  * hit failure, it will assume that the target exit the LFSCK
3454                  * prcoessing and stop try. */
3455                 if (rc == -ENOTCONN || rc == -ESHUTDOWN) {
3456                         int max_try = max_t(int, obd_timeout / 30, 3);
3457
3458                         if (++(llst->llst_failures) > max_try)
3459                                 done = true;
3460                 } else {
3461                         done = true;
3462                 }
3463         } else {
3464                 llst->llst_failures = 0;
3465                 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
3466                 if (lr->lr_status != LS_SCANNING_PHASE1 &&
3467                     lr->lr_status != LS_SCANNING_PHASE2)
3468                         done = true;
3469         }
3470
3471         if (done) {
3472                 CDEBUG(D_LFSCK, "%s: layout LFSCK slave gets the MDT %x "
3473                        "status %d, failures_try %d\n", lfsck_lfsck2name(com->lc_lfsck),
3474                        llst->llst_index, lr != NULL ? lr->lr_status : rc,
3475                        llst->llst_failures);
3476
3477                 lfsck_layout_llst_del(llsd, llst);
3478         }
3479
3480         lfsck_layout_llst_put(llst);
3481         lfsck_component_put(env, com);
3482         class_export_put(exp);
3483
3484         return 0;
3485 }
3486
3487 static int lfsck_layout_async_query(const struct lu_env *env,
3488                                     struct lfsck_component *com,
3489                                     struct obd_export *exp,
3490                                     struct lfsck_layout_slave_target *llst,
3491                                     struct lfsck_request *lr,
3492                                     struct ptlrpc_request_set *set)
3493 {
3494         struct lfsck_layout_slave_async_args *llsaa;
3495         struct ptlrpc_request                *req;
3496         struct lfsck_request                 *tmp;
3497         int                                   rc;
3498         ENTRY;
3499
3500         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
3501         if (req == NULL)
3502                 RETURN(-ENOMEM);
3503
3504         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
3505         if (rc != 0) {
3506                 ptlrpc_request_free(req);
3507                 RETURN(rc);
3508         }
3509
3510         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3511         *tmp = *lr;
3512         ptlrpc_request_set_replen(req);
3513
3514         llsaa = ptlrpc_req_async_args(req);
3515         llsaa->llsaa_exp = exp;
3516         llsaa->llsaa_com = lfsck_component_get(com);
3517         llsaa->llsaa_llst = llst;
3518         req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
3519         ptlrpc_set_add_req(set, req);
3520
3521         RETURN(0);
3522 }
3523
3524 static int lfsck_layout_async_notify(const struct lu_env *env,
3525                                      struct obd_export *exp,
3526                                      struct lfsck_request *lr,
3527                                      struct ptlrpc_request_set *set)
3528 {
3529         struct ptlrpc_request   *req;
3530         struct lfsck_request    *tmp;
3531         int                      rc;
3532         ENTRY;
3533
3534         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
3535         if (req == NULL)
3536                 RETURN(-ENOMEM);
3537
3538         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
3539         if (rc != 0) {
3540                 ptlrpc_request_free(req);
3541                 RETURN(rc);
3542         }
3543
3544         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3545         *tmp = *lr;
3546         ptlrpc_request_set_replen(req);
3547         ptlrpc_set_add_req(set, req);
3548
3549         RETURN(0);
3550 }
3551
3552 static int
3553 lfsck_layout_slave_query_master(const struct lu_env *env,
3554                                 struct lfsck_component *com)
3555 {
3556         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
3557         struct lfsck_instance            *lfsck = com->lc_lfsck;
3558         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3559         struct lfsck_layout_slave_target *llst;
3560         struct obd_export                *exp;
3561         struct ptlrpc_request_set        *set;
3562         int                               rc    = 0;
3563         int                               rc1   = 0;
3564         ENTRY;
3565
3566         set = ptlrpc_prep_set();
3567         if (set == NULL)
3568                 GOTO(log, rc = -ENOMEM);
3569
3570         memset(lr, 0, sizeof(*lr));
3571         lr->lr_event = LE_QUERY;
3572         lr->lr_active = LFSCK_TYPE_LAYOUT;
3573
3574         llsd->llsd_touch_gen++;
3575         spin_lock(&llsd->llsd_lock);
3576         while (!list_empty(&llsd->llsd_master_list)) {
3577                 llst = list_entry(llsd->llsd_master_list.next,
3578                                   struct lfsck_layout_slave_target,
3579                                   llst_list);
3580                 if (llst->llst_gen == llsd->llsd_touch_gen)
3581                         break;
3582
3583                 llst->llst_gen = llsd->llsd_touch_gen;
3584                 list_move_tail(&llst->llst_list,
3585                                &llsd->llsd_master_list);
3586                 atomic_inc(&llst->llst_ref);
3587                 spin_unlock(&llsd->llsd_lock);
3588
3589                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
3590                                                llst->llst_index);
3591                 if (exp == NULL) {
3592                         lfsck_layout_llst_del(llsd, llst);
3593                         lfsck_layout_llst_put(llst);
3594                         spin_lock(&llsd->llsd_lock);
3595                         continue;
3596                 }
3597
3598                 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
3599                 if (rc != 0) {
3600                         CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
3601                                "query %s for layout: rc = %d\n",
3602                                lfsck_lfsck2name(lfsck),
3603                                exp->exp_obd->obd_name, rc);
3604
3605                         rc1 = rc;
3606                         lfsck_layout_llst_put(llst);
3607                         class_export_put(exp);
3608                 }
3609                 spin_lock(&llsd->llsd_lock);
3610         }
3611         spin_unlock(&llsd->llsd_lock);
3612
3613         rc = ptlrpc_set_wait(set);
3614         ptlrpc_set_destroy(set);
3615
3616         GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
3617
3618 log:
3619         CDEBUG(D_LFSCK, "%s: layout LFSCK slave queries master: rc = %d\n",
3620                lfsck_lfsck2name(com->lc_lfsck), rc);
3621
3622         return rc;
3623 }
3624
3625 static void
3626 lfsck_layout_slave_notify_master(const struct lu_env *env,
3627                                  struct lfsck_component *com,
3628                                  enum lfsck_events event, int result)
3629 {
3630         struct lfsck_layout              *lo    = com->lc_file_ram;
3631         struct lfsck_instance            *lfsck = com->lc_lfsck;
3632         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3633         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
3634         struct lfsck_layout_slave_target *llst;
3635         struct obd_export                *exp;
3636         struct ptlrpc_request_set        *set;
3637         int                               rc;
3638         ENTRY;
3639
3640         CDEBUG(D_LFSCK, "%s: layout LFSCK slave notifies master\n",
3641                lfsck_lfsck2name(com->lc_lfsck));
3642
3643         set = ptlrpc_prep_set();
3644         if (set == NULL)
3645                 RETURN_EXIT;
3646
3647         memset(lr, 0, sizeof(*lr));
3648         lr->lr_event = event;
3649         lr->lr_flags = LEF_FROM_OST;
3650         lr->lr_status = result;
3651         lr->lr_index = lfsck_dev_idx(lfsck);
3652         lr->lr_active = LFSCK_TYPE_LAYOUT;
3653         lr->lr_flags2 = lo->ll_flags;
3654         llsd->llsd_touch_gen++;
3655         spin_lock(&llsd->llsd_lock);
3656         while (!list_empty(&llsd->llsd_master_list)) {
3657                 llst = list_entry(llsd->llsd_master_list.next,
3658                                   struct lfsck_layout_slave_target,
3659                                   llst_list);
3660                 if (llst->llst_gen == llsd->llsd_touch_gen)
3661                         break;
3662
3663                 llst->llst_gen = llsd->llsd_touch_gen;
3664                 list_move_tail(&llst->llst_list,
3665                                &llsd->llsd_master_list);
3666                 atomic_inc(&llst->llst_ref);
3667                 spin_unlock(&llsd->llsd_lock);
3668
3669                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
3670                                                llst->llst_index);
3671                 if (exp == NULL) {
3672                         lfsck_layout_llst_del(llsd, llst);
3673                         lfsck_layout_llst_put(llst);
3674                         spin_lock(&llsd->llsd_lock);
3675                         continue;
3676                 }
3677
3678                 rc = lfsck_layout_async_notify(env, exp, lr, set);
3679                 if (rc != 0)
3680                         CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
3681                                "notify %s for layout: rc = %d\n",
3682                                lfsck_lfsck2name(lfsck),
3683                                exp->exp_obd->obd_name, rc);
3684
3685                 lfsck_layout_llst_put(llst);
3686                 class_export_put(exp);
3687                 spin_lock(&llsd->llsd_lock);
3688         }
3689         spin_unlock(&llsd->llsd_lock);
3690
3691         ptlrpc_set_wait(set);
3692         ptlrpc_set_destroy(set);
3693
3694         RETURN_EXIT;
3695 }
3696
3697 /*
3698  * \ret -ENODATA: unrecognized stripe
3699  * \ret = 0     : recognized stripe
3700  * \ret < 0     : other failures
3701  */
3702 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
3703                                            struct lfsck_component *com,
3704                                            struct lu_fid *cfid,
3705                                            struct lu_fid *pfid)
3706 {
3707         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3708         struct lu_buf                   *buf    = &info->lti_big_buf;
3709         struct ost_id                   *oi     = &info->lti_oi;
3710         struct dt_object                *obj;
3711         struct lov_mds_md_v1            *lmm;
3712         struct lov_ost_data_v1          *objs;
3713         __u32                            idx    = pfid->f_stripe_idx;
3714         __u32                            magic;
3715         int                              rc     = 0;
3716         int                              i;
3717         __u16                            count;
3718         ENTRY;
3719
3720         pfid->f_ver = 0;
3721         obj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3722         if (IS_ERR(obj))
3723                 RETURN(PTR_ERR(obj));
3724
3725         dt_read_lock(env, obj, 0);
3726         if (unlikely(dt_object_exists(obj) == 0 ||
3727                      lfsck_is_dead_obj(obj)))
3728                 GOTO(unlock, rc = -ENOENT);
3729
3730         if (!S_ISREG(lfsck_object_type(obj)))
3731                 GOTO(unlock, rc = -ENODATA);
3732
3733         rc = lfsck_layout_get_lovea(env, obj, buf);
3734         if (rc < 0)
3735                 GOTO(unlock, rc);
3736
3737         if (rc == 0)
3738                 GOTO(unlock, rc = -ENODATA);
3739
3740         lmm = buf->lb_buf;
3741         rc = lfsck_layout_verify_header(lmm);
3742         if (rc != 0)
3743                 GOTO(unlock, rc);
3744
3745         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
3746          * been verified in lfsck_layout_verify_header() already. If some
3747          * new magic introduced in the future, then layout LFSCK needs to
3748          * be updated also. */
3749         magic = le32_to_cpu(lmm->lmm_magic);
3750         if (magic == LOV_MAGIC_V1) {
3751                 objs = &lmm->lmm_objects[0];
3752         } else {
3753                 LASSERT(magic == LOV_MAGIC_V3);
3754                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3755         }
3756
3757         fid_to_ostid(cfid, oi);
3758         count = le16_to_cpu(lmm->lmm_stripe_count);
3759         for (i = 0; i < count; i++, objs++) {
3760                 struct ost_id oi2;
3761
3762                 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
3763                 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
3764                         GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
3765         }
3766
3767         GOTO(unlock, rc = -ENODATA);
3768
3769 unlock:
3770         dt_read_unlock(env, obj);
3771         lfsck_object_put(env, obj);
3772
3773         return rc;
3774 }
3775
3776 /*
3777  * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
3778  * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
3779  * layout EA from MDT to OST. On one hand, the OST no need to understand
3780  * the layout EA structure; on the other hand, it may cause trouble when
3781  * transfer large layout EA from MDT to OST via normal OUT RPC.
3782  *
3783  * \ret > 0: unrecognized stripe
3784  * \ret = 0: recognized stripe
3785  * \ret < 0: other failures
3786  */
3787 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
3788                                           struct lfsck_component *com,
3789                                           struct lu_fid *cfid,
3790                                           struct lu_fid *pfid)
3791 {
3792         struct lfsck_instance    *lfsck  = com->lc_lfsck;
3793         struct obd_device        *obd    = lfsck->li_obd;
3794         struct seq_server_site   *ss     = lfsck_dev_site(lfsck);
3795         struct obd_export        *exp    = NULL;
3796         struct ptlrpc_request    *req    = NULL;
3797         struct lfsck_request     *lr;
3798         struct lu_seq_range      *range  = &lfsck_env_info(env)->lti_range;
3799         int                       rc     = 0;
3800         ENTRY;
3801
3802         if (unlikely(fid_is_idif(pfid)))
3803                 RETURN(1);
3804
3805         fld_range_set_any(range);
3806         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), range);
3807         if (rc != 0)
3808                 RETURN(rc == -ENOENT ? 1 : rc);
3809
3810         if (unlikely(!fld_range_is_mdt(range)))
3811                 RETURN(1);
3812
3813         exp = lustre_find_lwp_by_index(obd->obd_name, range->lsr_index);
3814         if (unlikely(exp == NULL))
3815                 RETURN(1);
3816
3817         if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
3818                 GOTO(out, rc = -EOPNOTSUPP);
3819
3820         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
3821         if (req == NULL)
3822                 GOTO(out, rc = -ENOMEM);
3823
3824         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
3825         if (rc != 0) {
3826                 ptlrpc_request_free(req);
3827
3828                 GOTO(out, rc);
3829         }
3830
3831         lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3832         memset(lr, 0, sizeof(*lr));
3833         lr->lr_event = LE_PAIRS_VERIFY;
3834         lr->lr_active = LFSCK_TYPE_LAYOUT;
3835         lr->lr_fid = *cfid; /* OST-object itself FID. */
3836         lr->lr_fid2 = *pfid; /* The claimed parent FID. */
3837
3838         ptlrpc_request_set_replen(req);
3839         rc = ptlrpc_queue_wait(req);
3840         ptlrpc_req_finished(req);
3841
3842         if (rc == -ENOENT || rc == -ENODATA)
3843                 rc = 1;
3844
3845         GOTO(out, rc);
3846
3847 out:
3848         if (exp != NULL)
3849                 class_export_put(exp);
3850
3851         return rc;
3852 }
3853
3854 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
3855                                           struct lfsck_component *com,
3856                                           struct lfsck_request *lr)
3857 {
3858         struct dt_object        *obj;
3859         int                      rc     = 0;
3860         ENTRY;
3861
3862         obj = lfsck_object_find_bottom(env, com->lc_lfsck, &lr->lr_fid);
3863         if (IS_ERR(obj))
3864                 GOTO(log, rc = PTR_ERR(obj));
3865
3866         dt_write_lock(env, obj, 0);
3867         if (unlikely(dt_object_exists(obj) == 0 ||
3868                      lfsck_is_dead_obj(obj)))
3869                 GOTO(unlock, rc = 0);
3870
3871         rc = __lfsck_layout_update_pfid(env, obj, &lr->lr_fid2,
3872                                         lr->lr_fid2.f_ver);
3873
3874         GOTO(unlock, rc);
3875
3876 unlock:
3877         dt_write_unlock(env, obj);
3878         lfsck_object_put(env, obj);
3879
3880 log:
3881         CDEBUG(D_LFSCK, "%s: layout LFSCK slave repaired pfid for "DFID
3882                ", parent "DFID": rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
3883                PFID(&lr->lr_fid), PFID(&lr->lr_fid2), rc);
3884
3885         return rc;
3886 }
3887
3888 /* layout APIs */
3889
3890 static void lfsck_layout_slave_quit(const struct lu_env *env,
3891                                     struct lfsck_component *com);
3892
3893 static int lfsck_layout_reset(const struct lu_env *env,
3894                               struct lfsck_component *com, bool init)
3895 {
3896         struct lfsck_layout     *lo    = com->lc_file_ram;
3897         int                      rc;
3898
3899         down_write(&com->lc_sem);
3900         if (init) {
3901                 memset(lo, 0, com->lc_file_size);
3902         } else {
3903                 __u32 count = lo->ll_success_count;
3904                 __u64 last_time = lo->ll_time_last_complete;
3905
3906                 memset(lo, 0, com->lc_file_size);
3907                 lo->ll_success_count = count;
3908                 lo->ll_time_last_complete = last_time;
3909         }
3910
3911         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
3912         lo->ll_status = LS_INIT;
3913
3914         if (com->lc_lfsck->li_master) {
3915                 struct lfsck_assistant_data *lad = com->lc_data;
3916
3917                 lad->lad_incomplete = 0;
3918                 CFS_RESET_BITMAP(lad->lad_bitmap);
3919         }
3920
3921         rc = lfsck_layout_store(env, com);
3922         up_write(&com->lc_sem);
3923
3924         CDEBUG(D_LFSCK, "%s: layout LFSCK reset: rc = %d\n",
3925                lfsck_lfsck2name(com->lc_lfsck), rc);
3926
3927         return rc;
3928 }
3929
3930 static void lfsck_layout_fail(const struct lu_env *env,
3931                               struct lfsck_component *com, bool new_checked)
3932 {
3933         struct lfsck_layout *lo = com->lc_file_ram;
3934
3935         down_write(&com->lc_sem);
3936         if (new_checked)
3937                 com->lc_new_checked++;
3938         lfsck_layout_record_failure(env, com->lc_lfsck, lo);
3939         up_write(&com->lc_sem);
3940 }
3941
3942 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
3943                                           struct lfsck_component *com, bool init)
3944 {
3945         struct lfsck_instance   *lfsck   = com->lc_lfsck;
3946         struct lfsck_layout     *lo      = com->lc_file_ram;
3947         int                      rc;
3948
3949         if (!init) {
3950                 rc = lfsck_checkpoint_generic(env, com);
3951                 if (rc != 0)
3952                         return rc > 0 ? 0 : rc;
3953         }
3954
3955         down_write(&com->lc_sem);
3956         if (init) {
3957                 lo->ll_pos_latest_start =
3958                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
3959         } else {
3960                 lo->ll_pos_last_checkpoint =
3961                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
3962                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3963                                 HALF_SEC - lfsck->li_time_last_checkpoint);
3964                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
3965                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
3966                 com->lc_new_checked = 0;
3967         }
3968
3969         rc = lfsck_layout_store(env, com);
3970         up_write(&com->lc_sem);
3971
3972         CDEBUG(D_LFSCK, "%s: layout LFSCK master checkpoint at the pos ["
3973                LPU64"], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
3974                lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
3975
3976         return rc;
3977 }
3978
3979 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
3980                                          struct lfsck_component *com, bool init)
3981 {
3982         struct lfsck_instance   *lfsck = com->lc_lfsck;
3983         struct lfsck_layout     *lo    = com->lc_file_ram;
3984         int                      rc;
3985
3986         if (com->lc_new_checked == 0 && !init)
3987                 return 0;
3988
3989         down_write(&com->lc_sem);
3990         if (init) {
3991                 lo->ll_pos_latest_start =
3992                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
3993         } else {
3994                 lo->ll_pos_last_checkpoint =
3995                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
3996                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3997                                 HALF_SEC - lfsck->li_time_last_checkpoint);
3998                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
3999                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4000                 com->lc_new_checked = 0;
4001         }
4002
4003         rc = lfsck_layout_store(env, com);
4004         up_write(&com->lc_sem);
4005
4006         CDEBUG(D_LFSCK, "%s: layout LFSCK slave checkpoint at the pos ["
4007                LPU64"], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
4008                lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
4009
4010         return rc;
4011 }
4012
4013 static int lfsck_layout_prep(const struct lu_env *env,
4014                              struct lfsck_component *com,
4015                              struct lfsck_start *start)
4016 {
4017         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4018         struct lfsck_layout     *lo     = com->lc_file_ram;
4019         struct lfsck_position   *pos    = &com->lc_pos_start;
4020
4021         fid_zero(&pos->lp_dir_parent);
4022         pos->lp_dir_cookie = 0;
4023         if (lo->ll_status == LS_COMPLETED ||
4024             lo->ll_status == LS_PARTIAL ||
4025             /* To handle orphan, must scan from the beginning. */
4026             (start != NULL && start->ls_flags & LPF_OST_ORPHAN)) {
4027                 int rc;
4028
4029                 rc = lfsck_layout_reset(env, com, false);
4030                 if (rc == 0)
4031                         rc = lfsck_set_param(env, lfsck, start, true);
4032
4033                 if (rc != 0) {
4034                         CDEBUG(D_LFSCK, "%s: layout LFSCK prep failed: "
4035                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4036
4037                         return rc;
4038                 }
4039         }
4040
4041         down_write(&com->lc_sem);
4042         lo->ll_time_latest_start = cfs_time_current_sec();
4043         spin_lock(&lfsck->li_lock);
4044         if (lo->ll_flags & LF_SCANNED_ONCE) {
4045                 if (!lfsck->li_drop_dryrun ||
4046                     lo->ll_pos_first_inconsistent == 0) {
4047                         lo->ll_status = LS_SCANNING_PHASE2;
4048                         list_move_tail(&com->lc_link,
4049                                        &lfsck->li_list_double_scan);
4050                         pos->lp_oit_cookie = 0;
4051                 } else {
4052                         int i;
4053
4054                         lo->ll_status = LS_SCANNING_PHASE1;
4055                         lo->ll_run_time_phase1 = 0;
4056                         lo->ll_run_time_phase2 = 0;
4057                         lo->ll_objs_checked_phase1 = 0;
4058                         lo->ll_objs_checked_phase2 = 0;
4059                         lo->ll_objs_failed_phase1 = 0;
4060                         lo->ll_objs_failed_phase2 = 0;
4061                         for (i = 0; i < LLIT_MAX; i++)
4062                                 lo->ll_objs_repaired[i] = 0;
4063
4064                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4065                         fid_zero(&com->lc_fid_latest_scanned_phase2);
4066                 }
4067         } else {
4068                 lo->ll_status = LS_SCANNING_PHASE1;
4069                 if (!lfsck->li_drop_dryrun ||
4070                     lo->ll_pos_first_inconsistent == 0)
4071                         pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
4072                 else
4073                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4074         }
4075         spin_unlock(&lfsck->li_lock);
4076         up_write(&com->lc_sem);
4077
4078         return 0;
4079 }
4080
4081 static int lfsck_layout_slave_prep(const struct lu_env *env,
4082                                    struct lfsck_component *com,
4083                                    struct lfsck_start_param *lsp)
4084 {
4085         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4086         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4087         struct lfsck_layout             *lo     = com->lc_file_ram;
4088         struct lfsck_start              *start  = lsp->lsp_start;
4089         int                              rc;
4090
4091         rc = lfsck_layout_prep(env, com, start);
4092         if (rc != 0)
4093                 return rc;
4094
4095         if (lo->ll_flags & LF_CRASHED_LASTID &&
4096             list_empty(&llsd->llsd_master_list)) {
4097                 LASSERT(lfsck->li_out_notify != NULL);
4098
4099                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4100                                      LE_LASTID_REBUILDING);
4101         }
4102
4103         if (!lsp->lsp_index_valid)
4104                 return 0;
4105
4106         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
4107         if (rc == 0 && start != NULL && start->ls_flags & LPF_OST_ORPHAN) {
4108                 LASSERT(!llsd->llsd_rbtree_valid);
4109
4110                 write_lock(&llsd->llsd_rb_lock);
4111                 rc = lfsck_rbtree_setup(env, com);
4112                 write_unlock(&llsd->llsd_rb_lock);
4113         }
4114
4115         CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
4116                LPU64"]\n", lfsck_lfsck2name(lfsck),
4117                com->lc_pos_start.lp_oit_cookie);
4118
4119         return rc;
4120 }
4121
4122 static int lfsck_layout_master_prep(const struct lu_env *env,
4123                                     struct lfsck_component *com,
4124                                     struct lfsck_start_param *lsp)
4125 {
4126         int rc;
4127         ENTRY;
4128
4129         rc = lfsck_layout_load_bitmap(env, com);
4130         if (rc != 0) {
4131                 rc = lfsck_layout_reset(env, com, false);
4132                 if (rc == 0)
4133                         rc = lfsck_set_param(env, com->lc_lfsck,
4134                                              lsp->lsp_start, true);
4135
4136                 if (rc != 0)
4137                         GOTO(log, rc);
4138         }
4139
4140         rc = lfsck_layout_prep(env, com, lsp->lsp_start);
4141         if (rc != 0)
4142                 RETURN(rc);
4143
4144         rc = lfsck_start_assistant(env, com, lsp);
4145
4146         GOTO(log, rc);
4147
4148 log:
4149         CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
4150                LPU64"]\n", lfsck_lfsck2name(com->lc_lfsck),
4151                com->lc_pos_start.lp_oit_cookie);
4152
4153         return 0;
4154 }
4155
4156 /* Pre-fetch the attribute for each stripe in the given layout EA. */
4157 static int lfsck_layout_scan_stripes(const struct lu_env *env,
4158                                      struct lfsck_component *com,
4159                                      struct dt_object *parent,
4160                                      struct lov_mds_md_v1 *lmm)
4161 {
4162         struct lfsck_thread_info        *info    = lfsck_env_info(env);
4163         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4164         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
4165         struct lfsck_layout             *lo      = com->lc_file_ram;
4166         struct lfsck_assistant_data     *lad     = com->lc_data;
4167         struct lfsck_assistant_object   *lso     = NULL;
4168         struct lov_ost_data_v1          *objs;
4169         struct lfsck_tgt_descs          *ltds    = &lfsck->li_ost_descs;
4170         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4171         struct ptlrpc_thread            *athread = &lad->lad_thread;
4172         struct l_wait_info               lwi     = { 0 };
4173         struct lu_buf                    buf;
4174         int                              rc      = 0;
4175         int                              i;
4176         __u32                            magic;
4177         __u16                            count;
4178         ENTRY;
4179
4180         lfsck_buf_init(&buf, &info->lti_old_pfid,
4181                        sizeof(struct filter_fid_old));
4182         count = le16_to_cpu(lmm->lmm_stripe_count);
4183         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
4184          * been verified in lfsck_layout_verify_header() already. If some
4185          * new magic introduced in the future, then layout LFSCK needs to
4186          * be updated also. */
4187         magic = le32_to_cpu(lmm->lmm_magic);
4188         if (magic == LOV_MAGIC_V1) {
4189                 objs = &lmm->lmm_objects[0];
4190         } else {
4191                 LASSERT(magic == LOV_MAGIC_V3);
4192                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4193         }
4194
4195         for (i = 0; i < count; i++, objs++) {
4196                 struct lu_fid           *fid    = &info->lti_fid;
4197                 struct ost_id           *oi     = &info->lti_oi;
4198                 struct lfsck_layout_req *llr;
4199                 struct lfsck_tgt_desc   *tgt    = NULL;
4200                 struct dt_object        *cobj   = NULL;
4201                 __u32                    index;
4202                 bool                     wakeup = false;
4203
4204                 if (unlikely(lovea_slot_is_dummy(objs)))
4205                         continue;
4206
4207                 l_wait_event(mthread->t_ctl_waitq,
4208                              lad->lad_prefetched < bk->lb_async_windows ||
4209                              !thread_is_running(mthread) ||
4210                              thread_is_stopped(athread),
4211                              &lwi);
4212
4213                 if (unlikely(!thread_is_running(mthread)) ||
4214                              thread_is_stopped(athread))
4215                         GOTO(out, rc = 0);
4216
4217                 if (unlikely(lfsck_is_dead_obj(parent)))
4218                         GOTO(out, rc = 0);
4219
4220                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4221                 index = le32_to_cpu(objs->l_ost_idx);
4222                 rc = ostid_to_fid(fid, oi, index);
4223                 if (rc != 0) {
4224                         CDEBUG(D_LFSCK, "%s: get invalid layout EA for "DFID
4225                                ": "DOSTID", idx:%u\n", lfsck_lfsck2name(lfsck),
4226                                PFID(lfsck_dto2fid(parent)), POSTID(oi), index);
4227                         goto next;
4228                 }
4229
4230                 tgt = lfsck_tgt_get(ltds, index);
4231                 if (unlikely(tgt == NULL)) {
4232                         CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
4233                                "did not join the layout LFSCK\n",
4234                                lfsck_lfsck2name(lfsck), index);
4235                         lfsck_lad_set_bitmap(env, com, index);
4236                         goto next;
4237                 }
4238
4239                 /* There is potential deadlock race condition between object
4240                  * destroy and layout LFSCK. Consider the following scenario:
4241                  *
4242                  * 1) The LFSCK thread obtained the parent object firstly, at
4243                  *    that time, the parent object has not been destroyed yet.
4244                  *
4245                  * 2) One RPC service thread destroyed the parent and all its
4246                  *    children objects. Because the LFSCK is referencing the
4247                  *    parent object, then the parent object will be marked as
4248                  *    dying in RAM. On the other hand, the parent object is
4249                  *    referencing all its children objects, then all children
4250                  *    objects will be marked as dying in RAM also.
4251                  *
4252                  * 3) The LFSCK thread tries to find some child object with
4253                  *    the parent object referenced. Then it will find that the
4254                  *    child object is dying. According to the object visibility
4255                  *    rules: the object with dying flag cannot be returned to
4256                  *    others. So the LFSCK thread has to wait until the dying
4257                  *    object has been purged from RAM, then it can allocate a
4258                  *    new object (with the same FID) in RAM. Unfortunately, the
4259                  *    LFSCK thread itself is referencing the parent object, and
4260                  *    cause the parent object cannot be purged, then cause the
4261                  *    child object cannot be purged also. So the LFSCK thread
4262                  *    will fall into deadlock.
4263                  *
4264                  * We introduce non-blocked version lu_object_find() to allow
4265                  * the LFSCK thread to return failure immediately (instead of
4266                  * wait) when it finds dying (child) object, then the LFSCK
4267                  * thread can check whether the parent object is dying or not.
4268                  * So avoid above deadlock. LU-5395 */
4269                 cobj = lfsck_object_find_by_dev_nowait(env, tgt->ltd_tgt, fid);
4270                 if (IS_ERR(cobj)) {
4271                         if (lfsck_is_dead_obj(parent)) {
4272                                 lfsck_tgt_put(tgt);
4273
4274                                 GOTO(out, rc = 0);
4275                         }
4276
4277                         rc = PTR_ERR(cobj);
4278                         goto next;
4279                 }
4280
4281                 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_ASSISTANT_DIRECT)) {
4282                         rc = dt_declare_attr_get(env, cobj);
4283                         if (rc != 0)
4284                                 goto next;
4285
4286                         rc = dt_declare_xattr_get(env, cobj, &buf,
4287                                                   XATTR_NAME_FID);
4288                         if (rc != 0)
4289                                 goto next;
4290                 }
4291
4292                 if (lso == NULL) {
4293                         struct lu_attr *attr = &info->lti_la;
4294
4295                         rc = dt_attr_get(env, parent, attr);
4296                         if (rc != 0)
4297                                 goto next;
4298
4299                         lso = lfsck_assistant_object_init(env,
4300                                 lfsck_dto2fid(parent), attr,
4301                                 lfsck->li_pos_current.lp_oit_cookie, false);
4302                         if (IS_ERR(lso)) {
4303                                 rc = PTR_ERR(lso);
4304                                 lso = NULL;
4305
4306                                 goto next;
4307                         }
4308                 }
4309
4310                 llr = lfsck_layout_assistant_req_init(lso, cobj, index, i);
4311                 if (IS_ERR(llr)) {
4312                         rc = PTR_ERR(llr);
4313                         goto next;
4314                 }
4315
4316                 cobj = NULL;
4317                 spin_lock(&lad->lad_lock);
4318                 if (lad->lad_assistant_status < 0) {
4319                         spin_unlock(&lad->lad_lock);
4320                         lfsck_layout_assistant_req_fini(env, &llr->llr_lar);
4321                         lfsck_tgt_put(tgt);
4322                         RETURN(lad->lad_assistant_status);
4323                 }
4324
4325                 list_add_tail(&llr->llr_lar.lar_list, &lad->lad_req_list);
4326                 if (lad->lad_prefetched == 0)
4327                         wakeup = true;
4328
4329                 lad->lad_prefetched++;
4330                 spin_unlock(&lad->lad_lock);
4331                 if (wakeup)
4332                         wake_up_all(&athread->t_ctl_waitq);
4333
4334 next:
4335                 down_write(&com->lc_sem);
4336                 com->lc_new_checked++;
4337                 if (rc < 0)
4338                         lfsck_layout_record_failure(env, lfsck, lo);
4339                 up_write(&com->lc_sem);
4340
4341                 if (cobj != NULL && !IS_ERR(cobj))
4342                         lfsck_object_put(env, cobj);
4343
4344                 if (likely(tgt != NULL))
4345                         lfsck_tgt_put(tgt);
4346
4347                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4348                         GOTO(out, rc);
4349         }
4350
4351         GOTO(out, rc = 0);
4352
4353 out:
4354         if (lso != NULL)
4355                 lfsck_assistant_object_put(env, lso);
4356
4357         return rc;
4358 }
4359
4360 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
4361  * the OST-object's attribute and generate an structure lfsck_layout_req on the
4362  * list ::lad_req_list.
4363  *
4364  * For each request on above list, the lfsck_layout_assistant thread compares
4365  * the OST side attribute with local attribute, if inconsistent, then repair it.
4366  *
4367  * All above processing is async mode with pipeline. */
4368 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
4369                                         struct lfsck_component *com,
4370                                         struct dt_object *obj)
4371 {
4372         struct lfsck_thread_info        *info   = lfsck_env_info(env);
4373         struct ost_id                   *oi     = &info->lti_oi;
4374         struct lfsck_layout             *lo     = com->lc_file_ram;
4375         struct lfsck_assistant_data     *lad    = com->lc_data;
4376         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4377         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
4378         struct thandle                  *handle = NULL;
4379         struct lu_buf                   *buf    = &info->lti_big_buf;
4380         struct lov_mds_md_v1            *lmm    = NULL;
4381         struct dt_device                *dev    = lfsck_obj2dev(obj);
4382         struct lustre_handle             lh     = { 0 };
4383         struct lu_buf                    ea_buf = { NULL };
4384         int                              rc     = 0;
4385         int                              size   = 0;
4386         bool                             locked = false;
4387         bool                             stripe = false;
4388         bool                             bad_oi = false;
4389         ENTRY;
4390
4391         if (!S_ISREG(lfsck_object_type(obj)))
4392                 GOTO(out, rc = 0);
4393
4394         if (lad->lad_assistant_status < 0)
4395                 GOTO(out, rc = -ESRCH);
4396
4397         fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
4398         lmm_oi_cpu_to_le(oi, oi);
4399         dt_read_lock(env, obj, 0);
4400         locked = true;
4401
4402 again:
4403         if (dt_object_exists(obj) == 0 ||
4404             lfsck_is_dead_obj(obj))
4405                 GOTO(out, rc = 0);
4406
4407         rc = lfsck_layout_get_lovea(env, obj, buf);
4408         if (rc <= 0)
4409                 GOTO(out, rc);
4410
4411         size = rc;
4412         lmm = buf->lb_buf;
4413         rc = lfsck_layout_verify_header(lmm);
4414         /* If the LOV EA crashed, then it is possible to be rebuilt later
4415          * when handle orphan OST-objects. */
4416         if (rc != 0)
4417                 GOTO(out, rc);
4418
4419         if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0)
4420                 GOTO(out, stripe = true);
4421
4422         /* Inconsistent lmm_oi, should be repaired. */
4423         bad_oi = true;
4424         lmm->lmm_oi = *oi;
4425
4426         if (bk->lb_param & LPF_DRYRUN) {
4427                 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4428
4429                 GOTO(out, stripe = true);
4430         }
4431
4432         if (!lustre_handle_is_used(&lh)) {
4433                 dt_read_unlock(env, obj);
4434                 locked = false;
4435                 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4436                                       MDS_INODELOCK_LAYOUT |
4437                                       MDS_INODELOCK_XATTR, LCK_EX);
4438                 if (rc != 0)
4439                         GOTO(out, rc);
4440
4441                 handle = dt_trans_create(env, dev);
4442                 if (IS_ERR(handle))
4443                         GOTO(out, rc = PTR_ERR(handle));
4444
4445                 lfsck_buf_init(&ea_buf, lmm, size);
4446                 rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
4447                                           LU_XATTR_REPLACE, handle);
4448                 if (rc != 0)
4449                         GOTO(out, rc);
4450
4451                 rc = dt_trans_start_local(env, dev, handle);
4452                 if (rc != 0)
4453                         GOTO(out, rc);
4454
4455                 dt_write_lock(env, obj, 0);
4456                 locked = true;
4457
4458                 goto again;
4459         }
4460
4461         rc = dt_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
4462                           LU_XATTR_REPLACE, handle);
4463         if (rc != 0)
4464                 GOTO(out, rc);
4465
4466         lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4467
4468         GOTO(out, stripe = true);
4469
4470 out:
4471         if (locked) {
4472                 if (lustre_handle_is_used(&lh))
4473                         dt_write_unlock(env, obj);
4474                 else
4475                         dt_read_unlock(env, obj);
4476         }
4477
4478         if (handle != NULL && !IS_ERR(handle))
4479                 dt_trans_stop(env, dev, handle);
4480
4481         lfsck_ibits_unlock(&lh, LCK_EX);
4482
4483         if (bad_oi)
4484                 CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
4485                        DFID": rc = %d\n", lfsck_lfsck2name(lfsck),
4486                        bk->lb_param & LPF_DRYRUN ? "found" : "repaired",
4487                        PFID(lfsck_dto2fid(obj)), rc);
4488
4489         if (stripe) {
4490                 rc = lfsck_layout_scan_stripes(env, com, obj, lmm);
4491         } else {
4492                 down_write(&com->lc_sem);
4493                 com->lc_new_checked++;
4494                 if (rc < 0)
4495                         lfsck_layout_record_failure(env, lfsck, lo);
4496                 up_write(&com->lc_sem);
4497         }
4498
4499         return rc;
4500 }
4501
4502 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
4503                                        struct lfsck_component *com,
4504                                        struct dt_object *obj)
4505 {
4506         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4507         struct lfsck_layout             *lo     = com->lc_file_ram;
4508         const struct lu_fid             *fid    = lfsck_dto2fid(obj);
4509         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4510         struct lfsck_layout_seq         *lls;
4511         __u64                            seq;
4512         __u64                            oid;
4513         int                              rc;
4514         ENTRY;
4515
4516         LASSERT(llsd != NULL);
4517
4518         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
4519             cfs_fail_val == lfsck_dev_idx(lfsck)) {
4520                 struct l_wait_info       lwi = LWI_TIMEOUT(cfs_time_seconds(1),
4521                                                            NULL, NULL);
4522                 struct ptlrpc_thread    *thread = &lfsck->li_thread;
4523
4524                 l_wait_event(thread->t_ctl_waitq,
4525                              !thread_is_running(thread),
4526                              &lwi);
4527         }
4528
4529         lfsck_rbtree_update_bitmap(env, com, fid, false);
4530
4531         down_write(&com->lc_sem);
4532         if (fid_is_idif(fid))
4533                 seq = 0;
4534         else if (!fid_is_norm(fid) ||
4535                  !fid_is_for_ostobj(env, lfsck, obj, fid))
4536                 GOTO(unlock, rc = 0);
4537         else
4538                 seq = fid_seq(fid);
4539         com->lc_new_checked++;
4540
4541         lls = lfsck_layout_seq_lookup(llsd, seq);
4542         if (lls == NULL) {
4543                 OBD_ALLOC_PTR(lls);
4544                 if (unlikely(lls == NULL))
4545                         GOTO(unlock, rc = -ENOMEM);
4546
4547                 INIT_LIST_HEAD(&lls->lls_list);
4548                 lls->lls_seq = seq;
4549                 rc = lfsck_layout_lastid_load(env, com, lls);
4550                 if (rc != 0) {
4551                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
4552                               "load LAST_ID for "LPX64": rc = %d\n",
4553                               lfsck_lfsck2name(com->lc_lfsck), seq, rc);
4554                         lo->ll_objs_failed_phase1++;
4555                         OBD_FREE_PTR(lls);
4556                         GOTO(unlock, rc);
4557                 }
4558
4559                 lfsck_layout_seq_insert(llsd, lls);
4560         }
4561
4562         if (unlikely(fid_is_last_id(fid)))
4563                 GOTO(unlock, rc = 0);
4564
4565         if (fid_is_idif(fid))
4566                 oid = fid_idif_id(fid_seq(fid), fid_oid(fid), fid_ver(fid));
4567         else
4568                 oid = fid_oid(fid);
4569
4570         if (oid > lls->lls_lastid_known)
4571                 lls->lls_lastid_known = oid;
4572
4573         if (oid > lls->lls_lastid) {
4574                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
4575                         /* OFD may create new objects during LFSCK scanning. */
4576                         rc = lfsck_layout_lastid_reload(env, com, lls);
4577                         if (unlikely(rc != 0)) {
4578                                 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
4579                                       "reload LAST_ID for "LPX64": rc = %d\n",
4580                                       lfsck_lfsck2name(com->lc_lfsck),
4581                                       lls->lls_seq, rc);
4582
4583                                 GOTO(unlock, rc);
4584                         }
4585
4586                         if (oid <= lls->lls_lastid ||
4587                             lo->ll_flags & LF_CRASHED_LASTID)
4588                                 GOTO(unlock, rc = 0);
4589
4590                         LASSERT(lfsck->li_out_notify != NULL);
4591
4592                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4593                                              LE_LASTID_REBUILDING);
4594                         lo->ll_flags |= LF_CRASHED_LASTID;
4595
4596                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
4597                                "LAST_ID file (2) for the sequence "LPX64
4598                                ", old value "LPU64", known value "LPU64"\n",
4599                                lfsck_lfsck2name(lfsck), lls->lls_seq,
4600                                lls->lls_lastid, oid);
4601                 }
4602
4603                 lls->lls_lastid = oid;
4604                 lls->lls_dirty = 1;
4605         }
4606
4607         GOTO(unlock, rc = 0);
4608
4609 unlock:
4610         up_write(&com->lc_sem);
4611
4612         return rc;
4613 }
4614
4615 static int lfsck_layout_exec_dir(const struct lu_env *env,
4616                                  struct lfsck_component *com,
4617                                  struct lfsck_assistant_object *lso,
4618                                  struct lu_dirent *ent, __u16 type)
4619 {
4620         return 0;
4621 }
4622
4623 static int lfsck_layout_master_post(const struct lu_env *env,
4624                                     struct lfsck_component *com,
4625                                     int result, bool init)
4626 {
4627         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4628         struct lfsck_layout     *lo     = com->lc_file_ram;
4629         int                      rc;
4630         ENTRY;
4631
4632         lfsck_post_generic(env, com, &result);
4633
4634         down_write(&com->lc_sem);
4635         spin_lock(&lfsck->li_lock);
4636         if (!init)
4637                 lo->ll_pos_last_checkpoint =
4638                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
4639
4640         if (result > 0) {
4641                 if (lo->ll_flags & LF_INCOMPLETE)
4642                         lo->ll_status = LS_PARTIAL;
4643                 else
4644                         lo->ll_status = LS_SCANNING_PHASE2;
4645                 lo->ll_flags |= LF_SCANNED_ONCE;
4646                 lo->ll_flags &= ~LF_UPGRADE;
4647                 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4648         } else if (result == 0) {
4649                 if (lfsck->li_status != 0)
4650                         lo->ll_status = lfsck->li_status;
4651                 else
4652                         lo->ll_status = LS_STOPPED;
4653                 if (lo->ll_status != LS_PAUSED)
4654                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4655         } else {
4656                 lo->ll_status = LS_FAILED;
4657                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4658         }
4659         spin_unlock(&lfsck->li_lock);
4660
4661         if (!init) {
4662                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4663                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4664                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4665                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4666                 com->lc_new_checked = 0;
4667         }
4668
4669         rc = lfsck_layout_store(env, com);
4670         up_write(&com->lc_sem);
4671
4672         CDEBUG(D_LFSCK, "%s: layout LFSCK master post done: rc = %d\n",
4673                lfsck_lfsck2name(lfsck), rc);
4674
4675         RETURN(rc);
4676 }
4677
4678 static int lfsck_layout_slave_post(const struct lu_env *env,
4679                                    struct lfsck_component *com,
4680                                    int result, bool init)
4681 {
4682         struct lfsck_instance   *lfsck = com->lc_lfsck;
4683         struct lfsck_layout     *lo    = com->lc_file_ram;
4684         int                      rc;
4685         bool                     done  = false;
4686
4687         down_write(&com->lc_sem);
4688         rc = lfsck_layout_lastid_store(env, com);
4689         if (rc != 0)
4690                 result = rc;
4691
4692         LASSERT(lfsck->li_out_notify != NULL);
4693
4694         spin_lock(&lfsck->li_lock);
4695         if (!init)
4696                 lo->ll_pos_last_checkpoint =
4697                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
4698
4699         if (result > 0) {
4700                 lo->ll_status = LS_SCANNING_PHASE2;
4701                 lo->ll_flags |= LF_SCANNED_ONCE;
4702                 if (lo->ll_flags & LF_CRASHED_LASTID) {
4703                         done = true;
4704                         lo->ll_flags &= ~LF_CRASHED_LASTID;
4705
4706                         CDEBUG(D_LFSCK, "%s: layout LFSCK has rebuilt "
4707                                "crashed LAST_ID files successfully\n",
4708                                lfsck_lfsck2name(lfsck));
4709                 }
4710                 lo->ll_flags &= ~LF_UPGRADE;
4711                 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4712         } else if (result == 0) {
4713                 if (lfsck->li_status != 0)
4714                         lo->ll_status = lfsck->li_status;
4715                 else
4716                         lo->ll_status = LS_STOPPED;
4717                 if (lo->ll_status != LS_PAUSED)
4718                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4719         } else {
4720                 lo->ll_status = LS_FAILED;
4721                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4722         }
4723         spin_unlock(&lfsck->li_lock);
4724
4725         if (done)
4726                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4727                                      LE_LASTID_REBUILT);
4728
4729         if (!init) {
4730                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4731                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4732                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4733                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4734                 com->lc_new_checked = 0;
4735         }
4736
4737         rc = lfsck_layout_store(env, com);
4738         up_write(&com->lc_sem);
4739
4740         lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
4741
4742         CDEBUG(D_LFSCK, "%s: layout LFSCK slave post done: rc = %d\n",
4743                lfsck_lfsck2name(lfsck), rc);
4744
4745         return rc;
4746 }
4747
4748 static int lfsck_layout_dump(const struct lu_env *env,
4749                              struct lfsck_component *com, struct seq_file *m)
4750 {
4751         struct lfsck_instance   *lfsck = com->lc_lfsck;
4752         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
4753         struct lfsck_layout     *lo    = com->lc_file_ram;
4754         int                      rc;
4755
4756         down_read(&com->lc_sem);
4757         seq_printf(m, "name: lfsck_layout\n"
4758                       "magic: %#x\n"
4759                       "version: %d\n"
4760                       "status: %s\n",
4761                       lo->ll_magic,
4762                       bk->lb_version,
4763                       lfsck_status2names(lo->ll_status));
4764
4765         rc = lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
4766         if (rc < 0)
4767                 goto out;
4768
4769         rc = lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4770         if (rc < 0)
4771                 goto out;
4772
4773         rc = lfsck_time_dump(m, lo->ll_time_last_complete,
4774                              "last_completed");
4775         if (rc < 0)
4776                 goto out;
4777
4778         rc = lfsck_time_dump(m, lo->ll_time_latest_start,
4779                              "latest_start");
4780         if (rc < 0)
4781                 goto out;
4782
4783         rc = lfsck_time_dump(m, lo->ll_time_last_checkpoint,
4784                              "last_checkpoint");
4785         if (rc < 0)
4786                 goto out;
4787
4788         seq_printf(m, "latest_start_position: "LPU64"\n"
4789                       "last_checkpoint_position: "LPU64"\n"
4790                       "first_failure_position: "LPU64"\n",
4791                       lo->ll_pos_latest_start,
4792                       lo->ll_pos_last_checkpoint,
4793                       lo->ll_pos_first_inconsistent);
4794
4795         seq_printf(m, "success_count: %u\n"
4796                       "repaired_dangling: "LPU64"\n"
4797                       "repaired_unmatched_pair: "LPU64"\n"
4798                       "repaired_multiple_referenced: "LPU64"\n"
4799                       "repaired_orphan: "LPU64"\n"
4800                       "repaired_inconsistent_owner: "LPU64"\n"
4801                       "repaired_others: "LPU64"\n"
4802                       "skipped: "LPU64"\n"
4803                       "failed_phase1: "LPU64"\n"
4804                       "failed_phase2: "LPU64"\n",
4805                       lo->ll_success_count,
4806                       lo->ll_objs_repaired[LLIT_DANGLING - 1],
4807                       lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
4808                       lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
4809                       lo->ll_objs_repaired[LLIT_ORPHAN - 1],
4810                       lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
4811                       lo->ll_objs_repaired[LLIT_OTHERS - 1],
4812                       lo->ll_objs_skipped,
4813                       lo->ll_objs_failed_phase1,
4814                       lo->ll_objs_failed_phase2);
4815
4816         if (lo->ll_status == LS_SCANNING_PHASE1) {
4817                 __u64 pos;
4818                 const struct dt_it_ops *iops;
4819                 cfs_duration_t duration = cfs_time_current() -
4820                                           lfsck->li_time_last_checkpoint;
4821                 __u64 checked = lo->ll_objs_checked_phase1 +
4822                                 com->lc_new_checked;
4823                 __u64 speed = checked;
4824                 __u64 new_checked = com->lc_new_checked *
4825                                     msecs_to_jiffies(MSEC_PER_SEC);
4826                 __u32 rtime = lo->ll_run_time_phase1 +
4827                               cfs_duration_sec(duration + HALF_SEC);
4828
4829                 if (duration != 0)
4830                         do_div(new_checked, duration);
4831                 if (rtime != 0)
4832                         do_div(speed, rtime);
4833                 seq_printf(m, "checked_phase1: "LPU64"\n"
4834                               "checked_phase2: "LPU64"\n"
4835                               "run_time_phase1: %u seconds\n"
4836                               "run_time_phase2: %u seconds\n"
4837                               "average_speed_phase1: "LPU64" items/sec\n"
4838                               "average_speed_phase2: N/A\n"
4839                               "real-time_speed_phase1: "LPU64" items/sec\n"
4840                               "real-time_speed_phase2: N/A\n",
4841                               checked,
4842                               lo->ll_objs_checked_phase2,
4843                               rtime,
4844                               lo->ll_run_time_phase2,
4845                               speed,
4846                               new_checked);
4847
4848                 LASSERT(lfsck->li_di_oit != NULL);
4849
4850                 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
4851
4852                 /* The low layer otable-based iteration position may NOT
4853                  * exactly match the layout-based directory traversal
4854                  * cookie. Generally, it is not a serious issue. But the
4855                  * caller should NOT make assumption on that. */
4856                 pos = iops->store(env, lfsck->li_di_oit);
4857                 if (!lfsck->li_current_oit_processed)
4858                         pos--;
4859                 seq_printf(m, "current_position: "LPU64"\n", pos);
4860
4861         } else if (lo->ll_status == LS_SCANNING_PHASE2) {
4862                 cfs_duration_t duration = cfs_time_current() -
4863                                           com->lc_time_last_checkpoint;
4864                 __u64 checked = lo->ll_objs_checked_phase2 +
4865                                 com->lc_new_checked;
4866                 __u64 speed1 = lo->ll_objs_checked_phase1;
4867                 __u64 speed2 = checked;
4868                 __u64 new_checked = com->lc_new_checked *
4869                                     msecs_to_jiffies(MSEC_PER_SEC);
4870                 __u32 rtime = lo->ll_run_time_phase2 +
4871                               cfs_duration_sec(duration + HALF_SEC);
4872
4873                 if (duration != 0)
4874                         do_div(new_checked, duration);
4875                 if (lo->ll_run_time_phase1 != 0)
4876                         do_div(speed1, lo->ll_run_time_phase1);
4877                 if (rtime != 0)
4878                         do_div(speed2, rtime);
4879                 rc = seq_printf(m, "checked_phase1: "LPU64"\n"
4880                                 "checked_phase2: "LPU64"\n"
4881                                 "run_time_phase1: %u seconds\n"
4882                                 "run_time_phase2: %u seconds\n"
4883                                 "average_speed_phase1: "LPU64" items/sec\n"
4884                                 "average_speed_phase2: "LPU64" items/sec\n"
4885                                 "real-time_speed_phase1: N/A\n"
4886                                 "real-time_speed_phase2: "LPU64" items/sec\n"
4887                                 "current_position: "DFID"\n",
4888                                 lo->ll_objs_checked_phase1,
4889                                 checked,
4890                                 lo->ll_run_time_phase1,
4891                                 rtime,
4892                                 speed1,
4893                                 speed2,
4894                                 new_checked,
4895                                 PFID(&com->lc_fid_latest_scanned_phase2));
4896                 if (rc <= 0)
4897                         goto out;
4898
4899         } else {
4900                 __u64 speed1 = lo->ll_objs_checked_phase1;
4901                 __u64 speed2 = lo->ll_objs_checked_phase2;
4902
4903                 if (lo->ll_run_time_phase1 != 0)
4904                         do_div(speed1, lo->ll_run_time_phase1);
4905                 if (lo->ll_run_time_phase2 != 0)
4906                         do_div(speed2, lo->ll_run_time_phase2);
4907                 seq_printf(m, "checked_phase1: "LPU64"\n"
4908                            "checked_phase2: "LPU64"\n"
4909                            "run_time_phase1: %u seconds\n"
4910                            "run_time_phase2: %u seconds\n"
4911                            "average_speed_phase1: "LPU64" items/sec\n"
4912                            "average_speed_phase2: "LPU64" objs/sec\n"
4913                            "real-time_speed_phase1: N/A\n"
4914                            "real-time_speed_phase2: N/A\n"
4915                            "current_position: N/A\n",
4916                            lo->ll_objs_checked_phase1,
4917                            lo->ll_objs_checked_phase2,
4918                            lo->ll_run_time_phase1,
4919                            lo->ll_run_time_phase2,
4920                            speed1,
4921                            speed2);
4922         }
4923 out:
4924         up_read(&com->lc_sem);
4925
4926         return rc;
4927 }
4928
4929 static int lfsck_layout_master_double_scan(const struct lu_env *env,
4930                                            struct lfsck_component *com)
4931 {
4932         struct lfsck_layout             *lo     = com->lc_file_ram;
4933         struct lfsck_assistant_data     *lad    = com->lc_data;
4934         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4935         struct lfsck_tgt_descs          *ltds;
4936         struct lfsck_tgt_desc           *ltd;
4937         struct lfsck_tgt_desc           *next;
4938         int                              rc;
4939
4940         rc = lfsck_double_scan_generic(env, com, lo->ll_status);
4941
4942         if (thread_is_stopped(&lad->lad_thread)) {
4943                 LASSERT(list_empty(&lad->lad_req_list));
4944                 LASSERT(list_empty(&lad->lad_ost_phase1_list));
4945                 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4946
4947                 ltds = &lfsck->li_ost_descs;
4948                 spin_lock(&ltds->ltd_lock);
4949                 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
4950                                          ltd_layout_phase_list) {
4951                         list_del_init(&ltd->ltd_layout_phase_list);
4952                 }
4953                 spin_unlock(&ltds->ltd_lock);
4954
4955                 ltds = &lfsck->li_mdt_descs;
4956                 spin_lock(&ltds->ltd_lock);
4957                 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4958                                          ltd_layout_phase_list) {
4959                         list_del_init(&ltd->ltd_layout_phase_list);
4960                 }
4961                 spin_unlock(&ltds->ltd_lock);
4962         }
4963
4964         return rc;
4965 }
4966
4967 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
4968                                           struct lfsck_component *com)
4969 {
4970         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4971         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4972         struct lfsck_layout             *lo     = com->lc_file_ram;
4973         struct ptlrpc_thread            *thread = &lfsck->li_thread;
4974         int                              rc;
4975         ENTRY;
4976
4977         CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan start\n",
4978                lfsck_lfsck2name(lfsck));
4979
4980         atomic_inc(&lfsck->li_double_scan_count);
4981
4982         if (lo->ll_flags & LF_INCOMPLETE)
4983                 GOTO(done, rc = 1);
4984
4985         com->lc_new_checked = 0;
4986         com->lc_new_scanned = 0;
4987         com->lc_time_last_checkpoint = cfs_time_current();
4988         com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
4989                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
4990
4991         while (1) {
4992                 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(30),
4993                                                      NULL, NULL);
4994
4995                 rc = lfsck_layout_slave_query_master(env, com);
4996                 if (list_empty(&llsd->llsd_master_list)) {
4997                         if (unlikely(!thread_is_running(thread)))
4998                                 rc = 0;
4999                         else
5000                                 rc = 1;
5001
5002                         GOTO(done, rc);
5003                 }
5004
5005                 if (rc < 0)
5006                         GOTO(done, rc);
5007
5008                 rc = l_wait_event(thread->t_ctl_waitq,
5009                                   !thread_is_running(thread) ||
5010                                   lo->ll_flags & LF_INCOMPLETE ||
5011                                   list_empty(&llsd->llsd_master_list),
5012                                   &lwi);
5013                 if (unlikely(!thread_is_running(thread)))
5014                         GOTO(done, rc = 0);
5015
5016                 if (lo->ll_flags & LF_INCOMPLETE)
5017                         GOTO(done, rc = 1);
5018
5019                 if (rc == -ETIMEDOUT)
5020                         continue;
5021
5022                 GOTO(done, rc = (rc < 0 ? rc : 1));
5023         }
5024
5025 done:
5026         rc = lfsck_layout_double_scan_result(env, com, rc);
5027         lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE,
5028                         (rc > 0 && lo->ll_flags & LF_INCOMPLETE) ? 0 : rc);
5029         lfsck_layout_slave_quit(env, com);
5030         if (atomic_dec_and_test(&lfsck->li_double_scan_count))
5031                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5032
5033         CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan finished, "
5034                "status %d: rc = %d\n",
5035                lfsck_lfsck2name(lfsck), lo->ll_status, rc);
5036
5037         return rc;
5038 }
5039
5040 static void lfsck_layout_master_data_release(const struct lu_env *env,
5041                                              struct lfsck_component *com)
5042 {
5043         struct lfsck_assistant_data     *lad    = com->lc_data;
5044         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5045         struct lfsck_tgt_descs          *ltds;
5046         struct lfsck_tgt_desc           *ltd;
5047         struct lfsck_tgt_desc           *next;
5048
5049         LASSERT(lad != NULL);
5050         LASSERT(thread_is_init(&lad->lad_thread) ||
5051                 thread_is_stopped(&lad->lad_thread));
5052         LASSERT(list_empty(&lad->lad_req_list));
5053
5054         com->lc_data = NULL;
5055
5056         ltds = &lfsck->li_ost_descs;
5057         spin_lock(&ltds->ltd_lock);
5058         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
5059                                  ltd_layout_phase_list) {
5060                 list_del_init(&ltd->ltd_layout_phase_list);
5061         }
5062         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5063                                  ltd_layout_phase_list) {
5064                 list_del_init(&ltd->ltd_layout_phase_list);
5065         }
5066         list_for_each_entry_safe(ltd, next, &lad->lad_ost_list,
5067                                  ltd_layout_list) {
5068                 list_del_init(&ltd->ltd_layout_list);
5069         }
5070         spin_unlock(&ltds->ltd_lock);
5071
5072         ltds = &lfsck->li_mdt_descs;
5073         spin_lock(&ltds->ltd_lock);
5074         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5075                                  ltd_layout_phase_list) {
5076                 list_del_init(&ltd->ltd_layout_phase_list);
5077         }
5078         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5079                                  ltd_layout_phase_list) {
5080                 list_del_init(&ltd->ltd_layout_phase_list);
5081         }
5082         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
5083                                  ltd_layout_list) {
5084                 list_del_init(&ltd->ltd_layout_list);
5085         }
5086         spin_unlock(&ltds->ltd_lock);
5087
5088         if (likely(lad->lad_bitmap != NULL))
5089                 CFS_FREE_BITMAP(lad->lad_bitmap);
5090
5091         OBD_FREE_PTR(lad);
5092 }
5093
5094 static void lfsck_layout_slave_data_release(const struct lu_env *env,
5095                                             struct lfsck_component *com)
5096 {
5097         struct lfsck_layout_slave_data *llsd = com->lc_data;
5098
5099         lfsck_layout_slave_quit(env, com);
5100         com->lc_data = NULL;
5101         OBD_FREE_PTR(llsd);
5102 }
5103
5104 static void lfsck_layout_master_quit(const struct lu_env *env,
5105                                      struct lfsck_component *com)
5106 {
5107         struct lfsck_assistant_data     *lad    = com->lc_data;
5108         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5109         struct lfsck_tgt_descs          *ltds;
5110         struct lfsck_tgt_desc           *ltd;
5111         struct lfsck_tgt_desc           *next;
5112
5113         LASSERT(lad != NULL);
5114
5115         lfsck_quit_generic(env, com);
5116
5117         LASSERT(thread_is_init(&lad->lad_thread) ||
5118                 thread_is_stopped(&lad->lad_thread));
5119         LASSERT(list_empty(&lad->lad_req_list));
5120
5121         ltds = &lfsck->li_ost_descs;
5122         spin_lock(&ltds->ltd_lock);
5123         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
5124                                  ltd_layout_phase_list) {
5125                 list_del_init(&ltd->ltd_layout_phase_list);
5126         }
5127         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5128                                  ltd_layout_phase_list) {
5129                 list_del_init(&ltd->ltd_layout_phase_list);
5130         }
5131         spin_unlock(&ltds->ltd_lock);
5132
5133         ltds = &lfsck->li_mdt_descs;
5134         spin_lock(&ltds->ltd_lock);
5135         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5136                                  ltd_layout_phase_list) {
5137                 list_del_init(&ltd->ltd_layout_phase_list);
5138         }
5139         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5140                                  ltd_layout_phase_list) {
5141                 list_del_init(&ltd->ltd_layout_phase_list);
5142         }
5143         spin_unlock(&ltds->ltd_lock);
5144 }
5145
5146 static void lfsck_layout_slave_quit(const struct lu_env *env,
5147                                     struct lfsck_component *com)
5148 {
5149         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5150         struct lfsck_layout_seq          *lls;
5151         struct lfsck_layout_seq          *next;
5152         struct lfsck_layout_slave_target *llst;
5153
5154         LASSERT(llsd != NULL);
5155
5156         down_write(&com->lc_sem);
5157         list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
5158                                  lls_list) {
5159                 list_del_init(&lls->lls_list);
5160                 lfsck_object_put(env, lls->lls_lastid_obj);
5161                 OBD_FREE_PTR(lls);
5162         }
5163         up_write(&com->lc_sem);
5164
5165         spin_lock(&llsd->llsd_lock);
5166         while (!list_empty(&llsd->llsd_master_list)) {
5167                 llst = list_entry(llsd->llsd_master_list.next,
5168                                   struct lfsck_layout_slave_target, llst_list);
5169                 list_del_init(&llst->llst_list);
5170                 spin_unlock(&llsd->llsd_lock);
5171                 lfsck_layout_llst_put(llst);
5172                 spin_lock(&llsd->llsd_lock);
5173         }
5174         spin_unlock(&llsd->llsd_lock);
5175
5176         lfsck_rbtree_cleanup(env, com);
5177 }
5178
5179 static int lfsck_layout_master_in_notify(const struct lu_env *env,
5180                                          struct lfsck_component *com,
5181                                          struct lfsck_request *lr,
5182                                          struct thandle *th)
5183 {
5184         struct lfsck_instance           *lfsck = com->lc_lfsck;
5185         struct lfsck_layout             *lo    = com->lc_file_ram;
5186         struct lfsck_assistant_data     *lad   = com->lc_data;
5187         struct lfsck_tgt_descs          *ltds;
5188         struct lfsck_tgt_desc           *ltd;
5189         bool                             fail  = false;
5190         ENTRY;
5191
5192         if (lr->lr_event == LE_PAIRS_VERIFY) {
5193                 int rc;
5194
5195                 rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
5196                                                      &lr->lr_fid2);
5197
5198                 RETURN(rc);
5199         }
5200
5201         CDEBUG(D_LFSCK, "%s: layout LFSCK master handles notify %u "
5202                "from %s %x, status %d, flags %x, flags2 %x\n",
5203                lfsck_lfsck2name(lfsck), lr->lr_event,
5204                (lr->lr_flags & LEF_FROM_OST) ? "OST" : "MDT",
5205                lr->lr_index, lr->lr_status, lr->lr_flags, lr->lr_flags2);
5206
5207         if (lr->lr_event != LE_PHASE1_DONE &&
5208             lr->lr_event != LE_PHASE2_DONE &&
5209             lr->lr_event != LE_PEER_EXIT)
5210                 RETURN(-EINVAL);
5211
5212         if (lr->lr_flags & LEF_FROM_OST)
5213                 ltds = &lfsck->li_ost_descs;
5214         else
5215                 ltds = &lfsck->li_mdt_descs;
5216         spin_lock(&ltds->ltd_lock);
5217         ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
5218         if (ltd == NULL) {
5219                 spin_unlock(&ltds->ltd_lock);
5220
5221                 RETURN(-ENXIO);
5222         }
5223
5224         list_del_init(&ltd->ltd_layout_phase_list);
5225         switch (lr->lr_event) {
5226         case LE_PHASE1_DONE:
5227                 if (lr->lr_status <= 0 || lr->lr_flags2 & LF_INCOMPLETE) {
5228                         if (lr->lr_flags2 & LF_INCOMPLETE) {
5229                                 if (lr->lr_flags & LEF_FROM_OST)
5230                                         lfsck_lad_set_bitmap(env, com,
5231                                                              ltd->ltd_index);
5232                                 else
5233                                         lo->ll_flags |= LF_INCOMPLETE;
5234                         }
5235                         ltd->ltd_layout_done = 1;
5236                         list_del_init(&ltd->ltd_layout_list);
5237                         fail = true;
5238                         break;
5239                 }
5240
5241                 if (lr->lr_flags & LEF_FROM_OST) {
5242                         if (list_empty(&ltd->ltd_layout_list))
5243                                 list_add_tail(&ltd->ltd_layout_list,
5244                                               &lad->lad_ost_list);
5245                         list_add_tail(&ltd->ltd_layout_phase_list,
5246                                       &lad->lad_ost_phase2_list);
5247                 } else {
5248                         if (list_empty(&ltd->ltd_layout_list))
5249                                 list_add_tail(&ltd->ltd_layout_list,
5250                                               &lad->lad_mdt_list);
5251                         list_add_tail(&ltd->ltd_layout_phase_list,
5252                                       &lad->lad_mdt_phase2_list);
5253                 }
5254                 break;
5255         case LE_PHASE2_DONE:
5256                 ltd->ltd_layout_done = 1;
5257                 if (!list_empty(&ltd->ltd_layout_list)) {
5258                         list_del_init(&ltd->ltd_layout_list);
5259                         if (lr->lr_flags2 & LF_INCOMPLETE) {
5260                                 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
5261                                 fail = true;
5262                         }
5263                 }
5264
5265                 break;
5266         case LE_PEER_EXIT:
5267                 fail = true;
5268                 ltd->ltd_layout_done = 1;
5269                 list_del_init(&ltd->ltd_layout_list);
5270                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) &&
5271                     !(lr->lr_flags & LEF_FROM_OST))
5272                                 lo->ll_flags |= LF_INCOMPLETE;
5273                 break;
5274         default:
5275                 break;
5276         }
5277         spin_unlock(&ltds->ltd_lock);
5278
5279         if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
5280                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5281
5282                 memset(stop, 0, sizeof(*stop));
5283                 stop->ls_status = lr->lr_status;
5284                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5285                 lfsck_stop(env, lfsck->li_bottom, stop);
5286         } else if (lfsck_phase2_next_ready(lad)) {
5287                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
5288         }
5289
5290         RETURN(0);
5291 }
5292
5293 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
5294                                         struct lfsck_component *com,
5295                                         struct lfsck_request *lr,
5296                                         struct thandle *th)
5297 {
5298         struct lfsck_instance            *lfsck = com->lc_lfsck;
5299         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5300         struct lfsck_layout_slave_target *llst;
5301         int                               rc;
5302         ENTRY;
5303
5304         switch (lr->lr_event) {
5305         case LE_FID_ACCESSED:
5306                 lfsck_rbtree_update_bitmap(env, com, &lr->lr_fid, true);
5307                 RETURN(0);
5308         case LE_CONDITIONAL_DESTROY:
5309                 rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
5310                 RETURN(rc);
5311         case LE_PAIRS_VERIFY: {
5312                 lr->lr_status = LPVS_INIT;
5313                 /* Firstly, if the MDT-object which is claimed via OST-object
5314                  * local stored PFID xattr recognizes the OST-object, then it
5315                  * must be that the client given PFID is wrong. */
5316                 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5317                                                     &lr->lr_fid3);
5318                 if (rc <= 0)
5319                         RETURN(0);
5320
5321                 lr->lr_status = LPVS_INCONSISTENT;
5322                 /* The OST-object local stored PFID xattr is stale. We need to
5323                  * check whether the MDT-object that is claimed via the client
5324                  * given PFID information recognizes the OST-object or not. If
5325                  * matches, then need to update the OST-object's PFID xattr. */
5326                 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5327                                                     &lr->lr_fid2);
5328                 /* For rc < 0 case:
5329                  * We are not sure whether the client given PFID information
5330                  * is correct or not, do nothing to avoid improper fixing.
5331                  *
5332                  * For rc > 0 case:
5333                  * The client given PFID information is also invalid, we can
5334                  * NOT fix the OST-object inconsistency.
5335                  */
5336                 if (rc != 0)
5337                         RETURN(rc);
5338
5339                 lr->lr_status = LPVS_INCONSISTENT_TOFIX;
5340                 rc = lfsck_layout_slave_repair_pfid(env, com, lr);
5341
5342                 RETURN(rc);
5343         }
5344         case LE_PHASE1_DONE: {
5345                 if (lr->lr_flags2 & LF_INCOMPLETE) {
5346                         struct lfsck_layout *lo = com->lc_file_ram;
5347
5348                         lo->ll_flags |= LF_INCOMPLETE;
5349                         llst = lfsck_layout_llst_find_and_del(llsd,
5350                                                               lr->lr_index,
5351                                                               true);
5352                         if (llst != NULL) {
5353                                 lfsck_layout_llst_put(llst);
5354                                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5355                         }
5356                 }
5357
5358                 RETURN(0);
5359         }
5360         case LE_PHASE2_DONE:
5361         case LE_PEER_EXIT:
5362                 CDEBUG(D_LFSCK, "%s: layout LFSCK slave handle notify %u "
5363                        "from MDT %x, status %d\n", lfsck_lfsck2name(lfsck),
5364                        lr->lr_event, lr->lr_index, lr->lr_status);
5365                 break;
5366         default:
5367                 RETURN(-EINVAL);
5368         }
5369
5370         llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
5371         if (llst == NULL)
5372                 RETURN(0);
5373
5374         lfsck_layout_llst_put(llst);
5375         if (list_empty(&llsd->llsd_master_list))
5376                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5377
5378         if (lr->lr_event == LE_PEER_EXIT &&
5379             (lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT ||
5380              (list_empty(&llsd->llsd_master_list) &&
5381               (lr->lr_status == LS_STOPPED ||
5382                lr->lr_status == LS_CO_STOPPED)))) {
5383                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5384
5385                 memset(stop, 0, sizeof(*stop));
5386                 stop->ls_status = lr->lr_status;
5387                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5388                 lfsck_stop(env, lfsck->li_bottom, stop);
5389         }
5390
5391         RETURN(0);
5392 }
5393
5394 static int lfsck_layout_query(const struct lu_env *env,
5395                               struct lfsck_component *com)
5396 {
5397         struct lfsck_layout *lo = com->lc_file_ram;
5398
5399         return lo->ll_status;
5400 }
5401
5402 /* with lfsck::li_lock held */
5403 static int lfsck_layout_slave_join(const struct lu_env *env,
5404                                    struct lfsck_component *com,
5405                                    struct lfsck_start_param *lsp)
5406 {
5407         struct lfsck_instance            *lfsck = com->lc_lfsck;
5408         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5409         struct lfsck_layout_slave_target *llst;
5410         struct lfsck_start               *start = lsp->lsp_start;
5411         int                               rc    = 0;
5412         ENTRY;
5413
5414         if (start == NULL || !(start->ls_flags & LPF_OST_ORPHAN))
5415                 RETURN(0);
5416
5417         if (!lsp->lsp_index_valid)
5418                 RETURN(-EINVAL);
5419
5420         /* If someone is running the LFSCK without orphan handling,
5421          * it will not maintain the object accessing rbtree. So we
5422          * cannot join it for orphan handling. */
5423         if (!llsd->llsd_rbtree_valid)
5424                 RETURN(-EBUSY);
5425
5426         spin_unlock(&lfsck->li_lock);
5427         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5428         spin_lock(&lfsck->li_lock);
5429         if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
5430                 spin_unlock(&lfsck->li_lock);
5431                 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
5432                                                       true);
5433                 if (llst != NULL)
5434                         lfsck_layout_llst_put(llst);
5435                 spin_lock(&lfsck->li_lock);
5436                 rc = -EAGAIN;
5437         }
5438
5439         RETURN(rc);
5440 }
5441
5442 static struct lfsck_operations lfsck_layout_master_ops = {
5443         .lfsck_reset            = lfsck_layout_reset,
5444         .lfsck_fail             = lfsck_layout_fail,
5445         .lfsck_checkpoint       = lfsck_layout_master_checkpoint,
5446         .lfsck_prep             = lfsck_layout_master_prep,
5447         .lfsck_exec_oit         = lfsck_layout_master_exec_oit,
5448         .lfsck_exec_dir         = lfsck_layout_exec_dir,
5449         .lfsck_post             = lfsck_layout_master_post,
5450         .lfsck_dump             = lfsck_layout_dump,
5451         .lfsck_double_scan      = lfsck_layout_master_double_scan,
5452         .lfsck_data_release     = lfsck_layout_master_data_release,
5453         .lfsck_quit             = lfsck_layout_master_quit,
5454         .lfsck_in_notify        = lfsck_layout_master_in_notify,
5455         .lfsck_query            = lfsck_layout_query,
5456 };
5457
5458 static struct lfsck_operations lfsck_layout_slave_ops = {
5459         .lfsck_reset            = lfsck_layout_reset,
5460         .lfsck_fail             = lfsck_layout_fail,
5461         .lfsck_checkpoint       = lfsck_layout_slave_checkpoint,
5462         .lfsck_prep             = lfsck_layout_slave_prep,
5463         .lfsck_exec_oit         = lfsck_layout_slave_exec_oit,
5464         .lfsck_exec_dir         = lfsck_layout_exec_dir,
5465         .lfsck_post             = lfsck_layout_slave_post,
5466         .lfsck_dump             = lfsck_layout_dump,
5467         .lfsck_double_scan      = lfsck_layout_slave_double_scan,
5468         .lfsck_data_release     = lfsck_layout_slave_data_release,
5469         .lfsck_quit             = lfsck_layout_slave_quit,
5470         .lfsck_in_notify        = lfsck_layout_slave_in_notify,
5471         .lfsck_query            = lfsck_layout_query,
5472         .lfsck_join             = lfsck_layout_slave_join,
5473 };
5474
5475 static void lfsck_layout_assistant_fill_pos(const struct lu_env *env,
5476                                             struct lfsck_component *com,
5477                                             struct lfsck_position *pos)
5478 {
5479         struct lfsck_assistant_data     *lad = com->lc_data;
5480         struct lfsck_layout_req         *llr;
5481
5482         if (list_empty(&lad->lad_req_list))
5483                 return;
5484
5485         llr = list_entry(lad->lad_req_list.next,
5486                          struct lfsck_layout_req,
5487                          llr_lar.lar_list);
5488         pos->lp_oit_cookie = llr->llr_lar.lar_parent->lso_oit_cookie - 1;
5489 }
5490
5491 struct lfsck_assistant_operations lfsck_layout_assistant_ops = {
5492         .la_handler_p1          = lfsck_layout_assistant_handler_p1,
5493         .la_handler_p2          = lfsck_layout_assistant_handler_p2,
5494         .la_fill_pos            = lfsck_layout_assistant_fill_pos,
5495         .la_double_scan_result  = lfsck_layout_double_scan_result,
5496         .la_req_fini            = lfsck_layout_assistant_req_fini,
5497         .la_sync_failures       = lfsck_layout_assistant_sync_failures,
5498 };
5499
5500 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
5501 {
5502         struct lfsck_component  *com;
5503         struct lfsck_layout     *lo;
5504         struct dt_object        *root = NULL;
5505         struct dt_object        *obj;
5506         int                      rc;
5507         ENTRY;
5508
5509         OBD_ALLOC_PTR(com);
5510         if (com == NULL)
5511                 RETURN(-ENOMEM);
5512
5513         INIT_LIST_HEAD(&com->lc_link);
5514         INIT_LIST_HEAD(&com->lc_link_dir);
5515         init_rwsem(&com->lc_sem);
5516         atomic_set(&com->lc_ref, 1);
5517         com->lc_lfsck = lfsck;
5518         com->lc_type = LFSCK_TYPE_LAYOUT;
5519         if (lfsck->li_master) {
5520                 com->lc_ops = &lfsck_layout_master_ops;
5521                 com->lc_data = lfsck_assistant_data_init(
5522                                 &lfsck_layout_assistant_ops,
5523                                 LFSCK_LAYOUT);
5524                 if (com->lc_data == NULL)
5525                         GOTO(out, rc = -ENOMEM);
5526         } else {
5527                 struct lfsck_layout_slave_data *llsd;
5528
5529                 com->lc_ops = &lfsck_layout_slave_ops;
5530                 OBD_ALLOC_PTR(llsd);
5531                 if (llsd == NULL)
5532                         GOTO(out, rc = -ENOMEM);
5533
5534                 INIT_LIST_HEAD(&llsd->llsd_seq_list);
5535                 INIT_LIST_HEAD(&llsd->llsd_master_list);
5536                 spin_lock_init(&llsd->llsd_lock);
5537                 llsd->llsd_rb_root = RB_ROOT;
5538                 rwlock_init(&llsd->llsd_rb_lock);
5539                 com->lc_data = llsd;
5540         }
5541         com->lc_file_size = sizeof(*lo);
5542         OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
5543         if (com->lc_file_ram == NULL)
5544                 GOTO(out, rc = -ENOMEM);
5545
5546         OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
5547         if (com->lc_file_disk == NULL)
5548                 GOTO(out, rc = -ENOMEM);
5549
5550         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
5551         if (IS_ERR(root))
5552                 GOTO(out, rc = PTR_ERR(root));
5553
5554         if (unlikely(!dt_try_as_dir(env, root)))
5555                 GOTO(out, rc = -ENOTDIR);
5556
5557         obj = local_file_find_or_create(env, lfsck->li_los, root,
5558                                         LFSCK_LAYOUT,
5559                                         S_IFREG | S_IRUGO | S_IWUSR);
5560         if (IS_ERR(obj))
5561                 GOTO(out, rc = PTR_ERR(obj));
5562
5563         com->lc_obj = obj;
5564         rc = lfsck_layout_load(env, com);
5565         if (rc > 0)
5566                 rc = lfsck_layout_reset(env, com, true);
5567         else if (rc == -ENOENT)
5568                 rc = lfsck_layout_init(env, com);
5569
5570         if (rc != 0)
5571                 GOTO(out, rc);
5572
5573         lo = com->lc_file_ram;
5574         switch (lo->ll_status) {
5575         case LS_INIT:
5576         case LS_COMPLETED:
5577         case LS_FAILED:
5578         case LS_STOPPED:
5579         case LS_PARTIAL:
5580                 spin_lock(&lfsck->li_lock);
5581                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
5582                 spin_unlock(&lfsck->li_lock);
5583                 break;
5584         default:
5585                 CERROR("%s: unknown lfsck_layout status %d\n",
5586                        lfsck_lfsck2name(lfsck), lo->ll_status);
5587                 /* fall through */
5588         case LS_SCANNING_PHASE1:
5589         case LS_SCANNING_PHASE2:
5590                 /* No need to store the status to disk right now.
5591                  * If the system crashed before the status stored,
5592                  * it will be loaded back when next time. */
5593                 lo->ll_status = LS_CRASHED;
5594                 if (!lfsck->li_master)
5595                         lo->ll_flags |= LF_INCOMPLETE;
5596                 /* fall through */
5597         case LS_PAUSED:
5598         case LS_CRASHED:
5599         case LS_CO_FAILED:
5600         case LS_CO_STOPPED:
5601         case LS_CO_PAUSED:
5602                 spin_lock(&lfsck->li_lock);
5603                 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
5604                 spin_unlock(&lfsck->li_lock);
5605                 break;
5606         }
5607
5608         if (lo->ll_flags & LF_CRASHED_LASTID) {
5609                 LASSERT(lfsck->li_out_notify != NULL);
5610
5611                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5612                                      LE_LASTID_REBUILDING);
5613         }
5614
5615         GOTO(out, rc = 0);
5616
5617 out:
5618         if (root != NULL && !IS_ERR(root))
5619                 lfsck_object_put(env, root);
5620
5621         if (rc != 0) {
5622                 lfsck_component_cleanup(env, com);
5623                 CERROR("%s: fail to init layout LFSCK component: rc = %d\n",
5624                        lfsck_lfsck2name(lfsck), rc);
5625         }
5626
5627         return rc;
5628 }
5629
5630 struct lfsck_orphan_it {
5631         struct lfsck_component           *loi_com;
5632         struct lfsck_rbtree_node         *loi_lrn;
5633         struct lfsck_layout_slave_target *loi_llst;
5634         struct lu_fid                     loi_key;
5635         struct lu_orphan_rec              loi_rec;
5636         __u64                             loi_hash;
5637         unsigned int                      loi_over:1;
5638 };
5639
5640 static int lfsck_fid_match_idx(const struct lu_env *env,
5641                                struct lfsck_instance *lfsck,
5642                                const struct lu_fid *fid, int idx)
5643 {
5644         struct seq_server_site  *ss;
5645         struct lu_server_fld    *sf;
5646         struct lu_seq_range     *range = &lfsck_env_info(env)->lti_range;
5647         int                      rc;
5648
5649         /* All abnormal cases will be returned to MDT0. */
5650         if (!fid_is_norm(fid)) {
5651                 if (idx == 0)
5652                         return 1;
5653
5654                 return 0;
5655         }
5656
5657         ss = lfsck_dev_site(lfsck);
5658         if (unlikely(ss == NULL))
5659                 return -ENOTCONN;
5660
5661         sf = ss->ss_server_fld;
5662         LASSERT(sf != NULL);
5663
5664         fld_range_set_any(range);
5665         rc = fld_server_lookup(env, sf, fid_seq(fid), range);
5666         if (rc != 0)
5667                 return rc;
5668
5669         if (!fld_range_is_mdt(range))
5670                 return -EINVAL;
5671
5672         if (range->lsr_index == idx)
5673                 return 1;
5674
5675         return 0;
5676 }
5677
5678 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
5679                                         struct dt_object *obj)
5680 {
5681         struct dt_device        *dev    = lfsck_obj2dev(obj);
5682         struct thandle          *handle;
5683         int                      rc;
5684         ENTRY;
5685
5686         handle = dt_trans_create(env, dev);
5687         if (IS_ERR(handle))
5688                 RETURN_EXIT;
5689
5690         rc = dt_declare_ref_del(env, obj, handle);
5691         if (rc != 0)
5692                 GOTO(stop, rc);
5693
5694         rc = dt_declare_destroy(env, obj, handle);
5695         if (rc != 0)
5696                 GOTO(stop, rc);
5697
5698         rc = dt_trans_start_local(env, dev, handle);
5699         if (rc != 0)
5700                 GOTO(stop, rc);
5701
5702         dt_write_lock(env, obj, 0);
5703         rc = dt_ref_del(env, obj, handle);
5704         if (rc == 0)
5705                 rc = dt_destroy(env, obj, handle);
5706         dt_write_unlock(env, obj);
5707
5708         GOTO(stop, rc);
5709
5710 stop:
5711         dt_trans_stop(env, dev, handle);
5712
5713         CDEBUG(D_LFSCK, "destroy orphan OST-object "DFID": rc = %d\n",
5714                PFID(lfsck_dto2fid(obj)), rc);
5715
5716         RETURN_EXIT;
5717 }
5718
5719 static int lfsck_orphan_index_lookup(const struct lu_env *env,
5720                                      struct dt_object *dt,
5721                                      struct dt_rec *rec,
5722                                      const struct dt_key *key)
5723 {
5724         return -EOPNOTSUPP;
5725 }
5726
5727 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
5728                                              struct dt_object *dt,
5729                                              const struct dt_rec *rec,
5730                                              const struct dt_key *key,
5731                                              struct thandle *handle)
5732 {
5733         return -EOPNOTSUPP;
5734 }
5735
5736 static int lfsck_orphan_index_insert(const struct lu_env *env,
5737                                      struct dt_object *dt,
5738                                      const struct dt_rec *rec,
5739                                      const struct dt_key *key,
5740                                      struct thandle *handle,
5741                                      int ignore_quota)
5742 {
5743         return -EOPNOTSUPP;
5744 }
5745
5746 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
5747                                              struct dt_object *dt,
5748                                              const struct dt_key *key,
5749                                              struct thandle *handle)
5750 {
5751         return -EOPNOTSUPP;
5752 }
5753
5754 static int lfsck_orphan_index_delete(const struct lu_env *env,
5755                                      struct dt_object *dt,
5756                                      const struct dt_key *key,
5757                                      struct thandle *handle)
5758 {
5759         return -EOPNOTSUPP;
5760 }
5761
5762 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
5763                                           struct dt_object *dt,
5764                                           __u32 attr)
5765 {
5766         struct dt_device                *dev    = lu2dt_dev(dt->do_lu.lo_dev);
5767         struct lfsck_instance           *lfsck;
5768         struct lfsck_component          *com    = NULL;
5769         struct lfsck_layout_slave_data  *llsd;
5770         struct lfsck_orphan_it          *it     = NULL;
5771         struct lfsck_layout             *lo;
5772         int                              rc     = 0;
5773         ENTRY;
5774
5775         lfsck = lfsck_instance_find(dev, true, false);
5776         if (unlikely(lfsck == NULL))
5777                 RETURN(ERR_PTR(-ENXIO));
5778
5779         com = lfsck_component_find(lfsck, LFSCK_TYPE_LAYOUT);
5780         if (unlikely(com == NULL))
5781                 GOTO(out, rc = -ENOENT);
5782
5783         lo = com->lc_file_ram;
5784         if (lo->ll_flags & LF_INCOMPLETE)
5785                 GOTO(out, rc = -ESRCH);
5786
5787         llsd = com->lc_data;
5788         if (!llsd->llsd_rbtree_valid)
5789                 GOTO(out, rc = -ESRCH);
5790
5791         OBD_ALLOC_PTR(it);
5792         if (it == NULL)
5793                 GOTO(out, rc = -ENOMEM);
5794
5795         it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
5796         if (it->loi_llst == NULL)
5797                 GOTO(out, rc = -ENXIO);
5798
5799         if (dev->dd_record_fid_accessed) {
5800                 /* The first iteration against the rbtree, scan the whole rbtree
5801                  * to remove the nodes which do NOT need to be handled. */
5802                 write_lock(&llsd->llsd_rb_lock);
5803                 if (dev->dd_record_fid_accessed) {
5804                         struct rb_node                  *node;
5805                         struct rb_node                  *next;
5806                         struct lfsck_rbtree_node        *lrn;
5807
5808                         /* No need to record the fid accessing anymore. */
5809                         dev->dd_record_fid_accessed = 0;
5810
5811                         node = rb_first(&llsd->llsd_rb_root);
5812                         while (node != NULL) {
5813                                 next = rb_next(node);
5814                                 lrn = rb_entry(node, struct lfsck_rbtree_node,
5815                                                lrn_node);
5816                                 if (atomic_read(&lrn->lrn_known_count) <=
5817                                     atomic_read(&lrn->lrn_accessed_count)) {
5818                                         rb_erase(node, &llsd->llsd_rb_root);
5819                                         lfsck_rbtree_free(lrn);
5820                                 }
5821                                 node = next;
5822                         }
5823                 }
5824                 write_unlock(&llsd->llsd_rb_lock);
5825         }
5826
5827         /* read lock the rbtree when init, and unlock when fini */
5828         read_lock(&llsd->llsd_rb_lock);
5829         it->loi_com = com;
5830         com = NULL;
5831
5832         GOTO(out, rc = 0);
5833
5834 out:
5835         if (com != NULL)
5836                 lfsck_component_put(env, com);
5837
5838         CDEBUG(D_LFSCK, "%s: init the orphan iteration: rc = %d\n",
5839                lfsck_lfsck2name(lfsck), rc);
5840
5841         lfsck_instance_put(env, lfsck);
5842         if (rc != 0) {
5843                 if (it != NULL)
5844                         OBD_FREE_PTR(it);
5845
5846                 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
5847         }
5848
5849         return (struct dt_it *)it;
5850 }
5851
5852 static void lfsck_orphan_it_fini(const struct lu_env *env,
5853                                  struct dt_it *di)
5854 {
5855         struct lfsck_orphan_it           *it    = (struct lfsck_orphan_it *)di;
5856         struct lfsck_component           *com   = it->loi_com;
5857         struct lfsck_layout_slave_data   *llsd;
5858         struct lfsck_layout_slave_target *llst;
5859
5860         if (com != NULL) {
5861                 CDEBUG(D_LFSCK, "%s: fini the orphan iteration\n",
5862                        lfsck_lfsck2name(com->lc_lfsck));
5863
5864                 llsd = com->lc_data;
5865                 read_unlock(&llsd->llsd_rb_lock);
5866                 llst = it->loi_llst;
5867                 LASSERT(llst != NULL);
5868
5869                 /* Save the key and hash for iterate next. */
5870                 llst->llst_fid = it->loi_key;
5871                 llst->llst_hash = it->loi_hash;
5872                 lfsck_layout_llst_put(llst);
5873                 lfsck_component_put(env, com);
5874         }
5875         OBD_FREE_PTR(it);
5876 }
5877
5878 /**
5879  * \retval       +1: the iteration finished
5880  * \retval        0: on success, not finished
5881  * \retval      -ve: on error
5882  */
5883 static int lfsck_orphan_it_next(const struct lu_env *env,
5884                                 struct dt_it *di)
5885 {
5886         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5887         struct filter_fid_old           *pfid   = &info->lti_old_pfid;
5888         struct lu_attr                  *la     = &info->lti_la;
5889         struct lfsck_orphan_it          *it     = (struct lfsck_orphan_it *)di;
5890         struct lu_fid                   *key    = &it->loi_key;
5891         struct lu_orphan_rec            *rec    = &it->loi_rec;
5892         struct lfsck_component          *com    = it->loi_com;
5893         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5894         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
5895         struct dt_object                *obj;
5896         struct lfsck_rbtree_node        *lrn;
5897         int                              pos;
5898         int                              rc;
5899         __u32                            save;
5900         __u32                            idx    = it->loi_llst->llst_index;
5901         bool                             exact  = false;
5902         ENTRY;
5903
5904         if (it->loi_over)
5905                 RETURN(1);
5906
5907 again0:
5908         lrn = it->loi_lrn;
5909         if (lrn == NULL) {
5910                 lrn = lfsck_rbtree_search(llsd, key, &exact);
5911                 if (lrn == NULL) {
5912                         it->loi_over = 1;
5913                         RETURN(1);
5914                 }
5915
5916                 it->loi_lrn = lrn;
5917                 if (!exact) {
5918                         key->f_seq = lrn->lrn_seq;
5919                         key->f_oid = lrn->lrn_first_oid;
5920                         key->f_ver = 0;
5921                 }
5922         } else {
5923                 key->f_oid++;
5924                 if (unlikely(key->f_oid == 0)) {
5925                         key->f_seq++;
5926                         it->loi_lrn = NULL;
5927                         goto again0;
5928                 }
5929
5930                 if (key->f_oid >=
5931                     lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
5932                         it->loi_lrn = NULL;
5933                         goto again0;
5934                 }
5935         }
5936
5937         if (unlikely(atomic_read(&lrn->lrn_known_count) <=
5938                      atomic_read(&lrn->lrn_accessed_count))) {
5939                 struct rb_node *next = rb_next(&lrn->lrn_node);
5940
5941                 while (next != NULL) {
5942                         lrn = rb_entry(next, struct lfsck_rbtree_node,
5943                                        lrn_node);
5944                         if (atomic_read(&lrn->lrn_known_count) >
5945                             atomic_read(&lrn->lrn_accessed_count))
5946                                 break;
5947                         next = rb_next(next);
5948                 }
5949
5950                 if (next == NULL) {
5951                         it->loi_over = 1;
5952                         RETURN(1);
5953                 }
5954
5955                 it->loi_lrn = lrn;
5956                 key->f_seq = lrn->lrn_seq;
5957                 key->f_oid = lrn->lrn_first_oid;
5958                 key->f_ver = 0;
5959         }
5960
5961         pos = key->f_oid - lrn->lrn_first_oid;
5962
5963 again1:
5964         pos = find_next_bit(lrn->lrn_known_bitmap,
5965                             LFSCK_RBTREE_BITMAP_WIDTH, pos);
5966         if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
5967                 key->f_oid = lrn->lrn_first_oid + pos;
5968                 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
5969                         key->f_seq++;
5970                         key->f_oid = 0;
5971                 }
5972                 it->loi_lrn = NULL;
5973                 goto again0;
5974         }
5975
5976         if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
5977                 pos++;
5978                 goto again1;
5979         }
5980
5981         key->f_oid = lrn->lrn_first_oid + pos;
5982         obj = lfsck_object_find_bottom(env, lfsck, key);
5983         if (IS_ERR(obj)) {
5984                 rc = PTR_ERR(obj);
5985                 if (rc == -ENOENT) {
5986                         pos++;
5987                         goto again1;
5988                 }
5989                 RETURN(rc);
5990         }
5991
5992         dt_read_lock(env, obj, 0);
5993         if (dt_object_exists(obj) == 0 ||
5994             lfsck_is_dead_obj(obj)) {
5995                 dt_read_unlock(env, obj);
5996                 lfsck_object_put(env, obj);
5997                 pos++;
5998                 goto again1;
5999         }
6000
6001         rc = dt_attr_get(env, obj, la);
6002         if (rc != 0)
6003                 GOTO(out, rc);
6004
6005         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, pfid, sizeof(*pfid)),
6006                           XATTR_NAME_FID);
6007         if (rc == -ENODATA) {
6008                 /* For the pre-created OST-object, update the bitmap to avoid
6009                  * others LFSCK (second phase) iteration to touch it again. */
6010                 if (la->la_ctime == 0) {
6011                         if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
6012                                 atomic_inc(&lrn->lrn_accessed_count);
6013
6014                         /* For the race between repairing dangling referenced
6015                          * MDT-object and unlink the file, it may left orphan
6016                          * OST-object there. Destroy it now! */
6017                         if (unlikely(!(la->la_mode & S_ISUID))) {
6018                                 dt_read_unlock(env, obj);
6019                                 lfsck_layout_destroy_orphan(env, obj);
6020                                 lfsck_object_put(env, obj);
6021                                 pos++;
6022                                 goto again1;
6023                         }
6024                 } else if (idx == 0) {
6025                         /* If the orphan OST-object has no parent information,
6026                          * regard it as referenced by the MDT-object on MDT0. */
6027                         fid_zero(&rec->lor_fid);
6028                         rec->lor_uid = la->la_uid;
6029                         rec->lor_gid = la->la_gid;
6030                         GOTO(out, rc = 0);
6031                 }
6032
6033                 dt_read_unlock(env, obj);
6034                 lfsck_object_put(env, obj);
6035                 pos++;
6036                 goto again1;
6037         }
6038
6039         if (rc < 0)
6040                 GOTO(out, rc);
6041
6042         if (rc != sizeof(struct filter_fid) &&
6043             rc != sizeof(struct filter_fid_old))
6044                 GOTO(out, rc = -EINVAL);
6045
6046         fid_le_to_cpu(&rec->lor_fid, &pfid->ff_parent);
6047         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
6048          * MDT-object's FID::f_ver, instead it is the OST-object index in its
6049          * parent MDT-object's layout EA. */
6050         save = rec->lor_fid.f_stripe_idx;
6051         rec->lor_fid.f_ver = 0;
6052         rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_fid, idx);
6053         /* If the orphan OST-object does not claim the MDT, then next.
6054          *
6055          * If we do not know whether it matches or not, then return it
6056          * to the MDT for further check. */
6057         if (rc == 0) {
6058                 dt_read_unlock(env, obj);
6059                 lfsck_object_put(env, obj);
6060                 pos++;
6061                 goto again1;
6062         }
6063
6064         rec->lor_fid.f_stripe_idx = save;
6065         rec->lor_uid = la->la_uid;
6066         rec->lor_gid = la->la_gid;
6067
6068         CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u\n",
6069                lfsck_lfsck2name(com->lc_lfsck), PFID(key), PFID(&rec->lor_fid),
6070                rec->lor_uid, rec->lor_gid);
6071
6072         GOTO(out, rc = 0);
6073
6074 out:
6075         dt_read_unlock(env, obj);
6076         lfsck_object_put(env, obj);
6077         if (rc == 0)
6078                 it->loi_hash++;
6079
6080         return rc;
6081 }
6082
6083 /**
6084  * \retval       +1: locate to the exactly position
6085  * \retval        0: cannot locate to the exactly position,
6086  *                   call next() to move to a valid position.
6087  * \retval      -ve: on error
6088  */
6089 static int lfsck_orphan_it_get(const struct lu_env *env,
6090                                struct dt_it *di,
6091                                const struct dt_key *key)
6092 {
6093         struct lfsck_orphan_it  *it   = (struct lfsck_orphan_it *)di;
6094         int                      rc;
6095
6096         it->loi_key = *(struct lu_fid *)key;
6097         rc = lfsck_orphan_it_next(env, di);
6098         if (rc == 1)
6099                 return 0;
6100
6101         if (rc == 0)
6102                 return 1;
6103
6104         return rc;
6105 }
6106
6107 static void lfsck_orphan_it_put(const struct lu_env *env,
6108                                 struct dt_it *di)
6109 {
6110 }
6111
6112 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
6113                                           const struct dt_it *di)
6114 {
6115         struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6116
6117         return (struct dt_key *)&it->loi_key;
6118 }
6119
6120 static int lfsck_orphan_it_key_size(const struct lu_env *env,
6121                                     const struct dt_it *di)
6122 {
6123         return sizeof(struct lu_fid);
6124 }
6125
6126 static int lfsck_orphan_it_rec(const struct lu_env *env,
6127                                const struct dt_it *di,
6128                                struct dt_rec *rec,
6129                                __u32 attr)
6130 {
6131         struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6132
6133         *(struct lu_orphan_rec *)rec = it->loi_rec;
6134
6135         return 0;
6136 }
6137
6138 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
6139                                    const struct dt_it *di)
6140 {
6141         struct lfsck_orphan_it  *it   = (struct lfsck_orphan_it *)di;
6142
6143         return it->loi_hash;
6144 }
6145
6146 /**
6147  * \retval       +1: locate to the exactly position
6148  * \retval        0: cannot locate to the exactly position,
6149  *                   call next() to move to a valid position.
6150  * \retval      -ve: on error
6151  */
6152 static int lfsck_orphan_it_load(const struct lu_env *env,
6153                                 const struct dt_it *di,
6154                                 __u64 hash)
6155 {
6156         struct lfsck_orphan_it           *it   = (struct lfsck_orphan_it *)di;
6157         struct lfsck_layout_slave_target *llst = it->loi_llst;
6158         int                               rc;
6159
6160         LASSERT(llst != NULL);
6161
6162         if (hash != llst->llst_hash) {
6163                 CDEBUG(D_LFSCK, "%s: the given hash "LPU64" for orphan "
6164                        "iteration does not match the one when fini "
6165                        LPU64", to be reset.\n",
6166                        lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
6167                        llst->llst_hash);
6168                 fid_zero(&llst->llst_fid);
6169                 llst->llst_hash = 0;
6170         }
6171
6172         it->loi_key = llst->llst_fid;
6173         it->loi_hash = llst->llst_hash;
6174         rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
6175         if (rc == 1)
6176                 return 0;
6177
6178         if (rc == 0)
6179                 return 1;
6180
6181         return rc;
6182 }
6183
6184 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
6185                                    const struct dt_it *di,
6186                                    void *key_rec)
6187 {
6188         return 0;
6189 }
6190
6191 const struct dt_index_operations lfsck_orphan_index_ops = {
6192         .dio_lookup             = lfsck_orphan_index_lookup,
6193         .dio_declare_insert     = lfsck_orphan_index_declare_insert,
6194         .dio_insert             = lfsck_orphan_index_insert,
6195         .dio_declare_delete     = lfsck_orphan_index_declare_delete,
6196         .dio_delete             = lfsck_orphan_index_delete,
6197         .dio_it = {
6198                 .init           = lfsck_orphan_it_init,
6199                 .fini           = lfsck_orphan_it_fini,
6200                 .get            = lfsck_orphan_it_get,
6201                 .put            = lfsck_orphan_it_put,
6202                 .next           = lfsck_orphan_it_next,
6203                 .key            = lfsck_orphan_it_key,
6204                 .key_size       = lfsck_orphan_it_key_size,
6205                 .rec            = lfsck_orphan_it_rec,
6206                 .store          = lfsck_orphan_it_store,
6207                 .load           = lfsck_orphan_it_load,
6208                 .key_rec        = lfsck_orphan_it_key_rec,
6209         }
6210 };