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, 2016, 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_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         fid_le_to_cpu(&des->ll_fid_latest_scanned_phase2,
 720                       &src->ll_fid_latest_scanned_phase2);
 721 }
 722
 723 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
 724                                    const struct lfsck_layout *src)
 725 {
 726         int i;
 727
 728         des->ll_magic = cpu_to_le32(src->ll_magic);
 729         des->ll_status = cpu_to_le32(src->ll_status);
 730         des->ll_flags = cpu_to_le32(src->ll_flags);
 731         des->ll_success_count = cpu_to_le32(src->ll_success_count);
 732         des->ll_run_time_phase1 = cpu_to_le32(src->ll_run_time_phase1);
 733         des->ll_run_time_phase2 = cpu_to_le32(src->ll_run_time_phase2);
 734         des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
 735         des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
 736         des->ll_time_last_checkpoint =
 737                                 cpu_to_le64(src->ll_time_last_checkpoint);
 738         des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
 739         des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
 740         des->ll_pos_first_inconsistent =
 741                         cpu_to_le64(src->ll_pos_first_inconsistent);
 742         des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
 743         des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
 744         des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
 745         des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
 746         for (i = 0; i < LLIT_MAX; i++)
 747                 des->ll_objs_repaired[i] =
 748                                 cpu_to_le64(src->ll_objs_repaired[i]);
 749         des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
 750         des->ll_bitmap_size = cpu_to_le32(src->ll_bitmap_size);
 751         fid_cpu_to_le(&des->ll_fid_latest_scanned_phase2,
 752                       &src->ll_fid_latest_scanned_phase2);
 753 }
 754
 755 /**
 756  * Load the OST bitmap from the lfsck_layout trace file.
 757  *
 758  * \param[in] env       pointer to the thread context
 759  * \param[in] com       pointer to the lfsck component
 760  *
 761  * \retval              0 for success
 762  * \retval              negative error number on failure or data corruption
 763  */
 764 static int lfsck_layout_load_bitmap(const struct lu_env *env,
 765                                     struct lfsck_component *com)
 766 {
 767         struct dt_object                *obj    = com->lc_obj;
 768         struct lfsck_assistant_data     *lad    = com->lc_data;
 769         struct lfsck_layout             *lo     = com->lc_file_ram;
 770         struct cfs_bitmap                       *bitmap = lad->lad_bitmap;
 771         loff_t                           pos    = com->lc_file_size;
 772         ssize_t                          size;
 773         __u32                            nbits;
 774         int                              rc;
 775         ENTRY;
 776
 777         if (com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size >
 778             lo->ll_bitmap_size)
 779                 nbits = com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size;
 780         else
 781                 nbits = lo->ll_bitmap_size;
 782
 783         if (unlikely(nbits < BITS_PER_LONG))
 784                 nbits = BITS_PER_LONG;
 785
 786         if (nbits > bitmap->size) {
 787                 __u32 new_bits = bitmap->size;
 788                 struct cfs_bitmap *new_bitmap;
 789
 790                 while (new_bits < nbits)
 791                         new_bits <<= 1;
 792
 793                 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
 794                 if (new_bitmap == NULL)
 795                         RETURN(-ENOMEM);
 796
 797                 lad->lad_bitmap = new_bitmap;
 798                 CFS_FREE_BITMAP(bitmap);
 799                 bitmap = new_bitmap;
 800         }
 801
 802         if (lo->ll_bitmap_size == 0) {
 803                 lad->lad_incomplete = 0;
 804                 CFS_RESET_BITMAP(bitmap);
 805
 806                 RETURN(0);
 807         }
 808
 809         size = (lo->ll_bitmap_size + 7) >> 3;
 810         rc = dt_read(env, obj, lfsck_buf_get(env, bitmap->data, size), &pos);
 811         if (rc != size)
 812                 RETURN(rc >= 0 ? -EINVAL : rc);
 813
 814         if (cfs_bitmap_check_empty(bitmap))
 815                 lad->lad_incomplete = 0;
 816         else
 817                 lad->lad_incomplete = 1;
 818
 819         RETURN(0);
 820 }
 821
 822 /**
 823  * Load the layout LFSCK trace file from disk.
 824  *
 825  * The layout LFSCK trace file records the layout LFSCK status information
 826  * and other statistics, such as how many objects have been scanned, and how
 827  * many objects have been repaired, and etc. It also contains the bitmap for
 828  * failed OSTs during the layout LFSCK. All these information will be loaded
 829  * from disk to RAM when the layout LFSCK component setup.
 830  *
 831  * \param[in] env       pointer to the thread context
 832  * \param[in] com       pointer to the lfsck component
 833  *
 834  * \retval              positive number for file data corruption, the caller
 835  *                      should reset the layout LFSCK trace file
 836  * \retval              0 for success
 837  * \retval              negative error number on failure
 838  */
 839 static int lfsck_layout_load(const struct lu_env *env,
 840                              struct lfsck_component *com)
 841 {
 842         struct lfsck_layout             *lo     = com->lc_file_ram;
 843         ssize_t                          size   = com->lc_file_size;
 844         loff_t                           pos    = 0;
 845         int                              rc;
 846
 847         rc = dt_read(env, com->lc_obj,
 848                      lfsck_buf_get(env, com->lc_file_disk, size), &pos);
 849         if (rc == 0) {
 850                 return -ENOENT;
 851         } else if (rc < 0) {
 852                 CDEBUG(D_LFSCK, "%s: failed to load lfsck_layout: rc = %d\n",
 853                        lfsck_lfsck2name(com->lc_lfsck), rc);
 854                 return rc;
 855         } else if (rc != size) {
 856                 CDEBUG(D_LFSCK, "%s: lfsck_layout size %u != %u; reset it\n",
 857                        lfsck_lfsck2name(com->lc_lfsck), rc, (unsigned int)size);
 858                 return 1;
 859         }
 860
 861         lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
 862         if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
 863                 CDEBUG(D_LFSCK, "%s: invalid lfsck_layout magic %#x != %#x, "
 864                        "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
 865                        lo->ll_magic, LFSCK_LAYOUT_MAGIC);
 866                 return 1;
 867         }
 868
 869         return 0;
 870 }
 871
 872 /**
 873  * Store the layout LFSCK trace file on disk.
 874  *
 875  * The layout LFSCK trace file records the layout LFSCK status information
 876  * and other statistics, such as how many objects have been scanned, and how
 877  * many objects have been repaired, and etc. It also contains the bitmap for
 878  * failed OSTs during the layout LFSCK. All these information will be synced
 879  * from RAM to disk periodically.
 880  *
 881  * \param[in] env       pointer to the thread context
 882  * \param[in] com       pointer to the lfsck component
 883  *
 884  * \retval              0 for success
 885  * \retval              negative error number on failure
 886  */
 887 static int lfsck_layout_store(const struct lu_env *env,
 888                               struct lfsck_component *com)
 889 {
 890         struct dt_object        *obj    = com->lc_obj;
 891         struct lfsck_instance   *lfsck  = com->lc_lfsck;
 892         struct lfsck_layout     *lo_ram = com->lc_file_ram;
 893         struct lfsck_layout     *lo     = com->lc_file_disk;
 894         struct thandle          *th;
 895         struct dt_device        *dev    = lfsck_obj2dev(obj);
 896         struct cfs_bitmap       *bitmap = NULL;
 897         loff_t                   pos;
 898         ssize_t                  size   = com->lc_file_size;
 899         __u32                    nbits  = 0;
 900         int                      rc;
 901         ENTRY;
 902
 903         if (lfsck->li_master) {
 904                 struct lfsck_assistant_data *lad = com->lc_data;
 905
 906                 bitmap = lad->lad_bitmap;
 907                 nbits = bitmap->size;
 908
 909                 LASSERT(nbits > 0);
 910                 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
 911         }
 912
 913         lo_ram->ll_bitmap_size = nbits;
 914         lfsck_layout_cpu_to_le(lo, lo_ram);
 915         th = dt_trans_create(env, dev);
 916         if (IS_ERR(th))
 917                 GOTO(log, rc = PTR_ERR(th));
 918
 919         rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
 920                                      (loff_t)0, th);
 921         if (rc != 0)
 922                 GOTO(out, rc);
 923
 924         if (bitmap != NULL) {
 925                 rc = dt_declare_record_write(env, obj,
 926                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
 927                                 (loff_t)size, th);
 928                 if (rc != 0)
 929                         GOTO(out, rc);
 930         }
 931
 932         rc = dt_trans_start_local(env, dev, th);
 933         if (rc != 0)
 934                 GOTO(out, rc);
 935
 936         pos = 0;
 937         rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos, th);
 938         if (rc != 0)
 939                 GOTO(out, rc);
 940
 941         if (bitmap != NULL) {
 942                 pos = size;
 943                 rc = dt_record_write(env, obj,
 944                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
 945                                 &pos, th);
 946         }
 947
 948         GOTO(out, rc);
 949
 950 out:
 951         dt_trans_stop(env, dev, th);
 952
 953 log:
 954         if (rc != 0)
 955                 CDEBUG(D_LFSCK, "%s: fail to store lfsck_layout: rc = %d\n",
 956                        lfsck_lfsck2name(lfsck), rc);
 957
 958         return rc;
 959 }
 960
 961 static int lfsck_layout_init(const struct lu_env *env,
 962                              struct lfsck_component *com)
 963 {
 964         struct lfsck_layout *lo = com->lc_file_ram;
 965         int rc;
 966
 967         memset(lo, 0, com->lc_file_size);
 968         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
 969         lo->ll_status = LS_INIT;
 970         down_write(&com->lc_sem);
 971         rc = lfsck_layout_store(env, com);
 972         if (rc == 0 && com->lc_lfsck->li_master)
 973                 rc = lfsck_load_sub_trace_files(env, com,
 974                         &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
 975         up_write(&com->lc_sem);
 976
 977         return rc;
 978 }
 979
 980 static int fid_is_for_ostobj(const struct lu_env *env,
 981                              struct lfsck_instance *lfsck,
 982                              struct dt_object *obj, const struct lu_fid *fid)
 983 {
 984         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
 985         struct lu_seq_range     *range  = &lfsck_env_info(env)->lti_range;
 986         struct lustre_mdt_attrs *lma;
 987         int                      rc;
 988
 989         fld_range_set_any(range);
 990         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
 991         if (rc == 0) {
 992                 if (fld_range_is_ost(range))
 993                         return 1;
 994
 995                 return 0;
 996         }
 997
 998         lma = &lfsck_env_info(env)->lti_lma;
 999         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, lma, sizeof(*lma)),
1000                           XATTR_NAME_LMA);
1001         if (rc == sizeof(*lma)) {
1002                 lustre_lma_swab(lma);
1003
1004                 return lma->lma_compat & LMAC_FID_ON_OST ? 1 : 0;
1005         }
1006
1007         rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID);
1008
1009         return rc > 0;
1010 }
1011
1012 static struct lfsck_layout_seq *
1013 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
1014 {
1015         struct lfsck_layout_seq *lls;
1016
1017         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1018                 if (lls->lls_seq == seq)
1019                         return lls;
1020
1021                 if (lls->lls_seq > seq)
1022                         return NULL;
1023         }
1024
1025         return NULL;
1026 }
1027
1028 static void
1029 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
1030                         struct lfsck_layout_seq *lls)
1031 {
1032         struct lfsck_layout_seq *tmp;
1033         struct list_head        *pos = &llsd->llsd_seq_list;
1034
1035         list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
1036                 if (lls->lls_seq < tmp->lls_seq) {
1037                         pos = &tmp->lls_list;
1038                         break;
1039                 }
1040         }
1041         list_add_tail(&lls->lls_list, pos);
1042 }
1043
1044 static int
1045 lfsck_layout_lastid_create(const struct lu_env *env,
1046                            struct lfsck_instance *lfsck,
1047                            struct dt_object *obj)
1048 {
1049         struct lfsck_thread_info *info   = lfsck_env_info(env);
1050         struct lu_attr           *la     = &info->lti_la;
1051         struct dt_object_format  *dof    = &info->lti_dof;
1052         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
1053         struct dt_device         *dt     = lfsck_obj2dev(obj);
1054         struct thandle           *th;
1055         __u64                     lastid = 0;
1056         loff_t                    pos    = 0;
1057         int                       rc;
1058         ENTRY;
1059
1060         if (bk->lb_param & LPF_DRYRUN)
1061                 return 0;
1062
1063         memset(la, 0, sizeof(*la));
1064         la->la_mode = S_IFREG |  S_IRUGO | S_IWUSR;
1065         la->la_valid = LA_MODE | LA_UID | LA_GID;
1066         memset(dof, 0, sizeof(*dof));
1067         dof->dof_type = dt_mode_to_dft(S_IFREG);
1068
1069         th = dt_trans_create(env, dt);
1070         if (IS_ERR(th))
1071                 GOTO(log, rc = PTR_ERR(th));
1072
1073         rc = dt_declare_create(env, obj, la, NULL, dof, th);
1074         if (rc != 0)
1075                 GOTO(stop, rc);
1076
1077         rc = dt_declare_record_write(env, obj,
1078                                      lfsck_buf_get(env, &lastid,
1079                                                    sizeof(lastid)),
1080                                      pos, th);
1081         if (rc != 0)
1082                 GOTO(stop, rc);
1083
1084         rc = dt_trans_start_local(env, dt, th);
1085         if (rc != 0)
1086                 GOTO(stop, rc);
1087
1088         dt_write_lock(env, obj, 0);
1089         if (likely(dt_object_exists(obj) == 0)) {
1090                 rc = dt_create(env, obj, la, NULL, dof, th);
1091                 if (rc == 0)
1092                         rc = dt_record_write(env, obj,
1093                                 lfsck_buf_get(env, &lastid, sizeof(lastid)),
1094                                 &pos, th);
1095         }
1096         dt_write_unlock(env, obj);
1097
1098         GOTO(stop, rc);
1099
1100 stop:
1101         dt_trans_stop(env, dt, th);
1102
1103 log:
1104         CDEBUG(D_LFSCK, "%s: layout LFSCK will create LAST_ID for <seq> "
1105                "%#llx: rc = %d\n",
1106                lfsck_lfsck2name(lfsck), fid_seq(lfsck_dto2fid(obj)), rc);
1107
1108         return rc;
1109 }
1110
1111 static int
1112 lfsck_layout_lastid_reload(const struct lu_env *env,
1113                            struct lfsck_component *com,
1114                            struct lfsck_layout_seq *lls)
1115 {
1116         __u64   lastid;
1117         loff_t  pos     = 0;
1118         int     rc;
1119
1120         dt_read_lock(env, lls->lls_lastid_obj, 0);
1121         rc = dt_record_read(env, lls->lls_lastid_obj,
1122                             lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
1123         dt_read_unlock(env, lls->lls_lastid_obj);
1124         if (unlikely(rc != 0))
1125                 return rc;
1126
1127         lastid = le64_to_cpu(lastid);
1128         if (lastid < lls->lls_lastid_known) {
1129                 struct lfsck_instance   *lfsck  = com->lc_lfsck;
1130                 struct lfsck_layout     *lo     = com->lc_file_ram;
1131
1132                 lls->lls_lastid = lls->lls_lastid_known;
1133                 lls->lls_dirty = 1;
1134                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1135                         LASSERT(lfsck->li_out_notify != NULL);
1136
1137                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1138                                              LE_LASTID_REBUILDING);
1139                         lo->ll_flags |= LF_CRASHED_LASTID;
1140
1141                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
1142                                "LAST_ID file (1) for the sequence %#llx"
1143                                ", old value %llu, known value %llu\n",
1144                                lfsck_lfsck2name(lfsck), lls->lls_seq,
1145                                lastid, lls->lls_lastid);
1146                 }
1147         } else if (lastid >= lls->lls_lastid) {
1148                 lls->lls_lastid = lastid;
1149                 lls->lls_dirty = 0;
1150         }
1151
1152         return 0;
1153 }
1154
1155 static int
1156 lfsck_layout_lastid_store(const struct lu_env *env,
1157                           struct lfsck_component *com)
1158 {
1159         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1160         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
1161         struct dt_device                *dt     = lfsck->li_bottom;
1162         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
1163         struct lfsck_layout_seq         *lls;
1164         struct thandle                  *th;
1165         __u64                            lastid;
1166         int                              rc     = 0;
1167         int                              rc1    = 0;
1168
1169         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1170                 loff_t pos = 0;
1171
1172                 if (!lls->lls_dirty)
1173                         continue;
1174
1175                 CDEBUG(D_LFSCK, "%s: layout LFSCK will sync the LAST_ID for "
1176                        "<seq> %#llx as <oid> %llu\n",
1177                        lfsck_lfsck2name(lfsck), lls->lls_seq, lls->lls_lastid);
1178
1179                 if (bk->lb_param & LPF_DRYRUN) {
1180                         lls->lls_dirty = 0;
1181                         continue;
1182                 }
1183
1184                 th = dt_trans_create(env, dt);
1185                 if (IS_ERR(th)) {
1186                         rc1 = PTR_ERR(th);
1187                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1188                                "the LAST_ID for <seq> %#llx(1): rc = %d\n",
1189                                lfsck_lfsck2name(com->lc_lfsck),
1190                                lls->lls_seq, rc1);
1191                         continue;
1192                 }
1193
1194                 lastid = cpu_to_le64(lls->lls_lastid);
1195                 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1196                                              lfsck_buf_get(env, &lastid,
1197                                                            sizeof(lastid)),
1198                                              pos, th);
1199                 if (rc != 0)
1200                         goto stop;
1201
1202                 rc = dt_trans_start_local(env, dt, th);
1203                 if (rc != 0)
1204                         goto stop;
1205
1206                 dt_write_lock(env, lls->lls_lastid_obj, 0);
1207                 rc = dt_record_write(env, lls->lls_lastid_obj,
1208                                      lfsck_buf_get(env, &lastid,
1209                                      sizeof(lastid)), &pos, th);
1210                 dt_write_unlock(env, lls->lls_lastid_obj);
1211                 if (rc == 0)
1212                         lls->lls_dirty = 0;
1213
1214 stop:
1215                 dt_trans_stop(env, dt, th);
1216                 if (rc != 0) {
1217                         rc1 = rc;
1218                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1219                                "the LAST_ID for <seq> %#llx(2): rc = %d\n",
1220                                lfsck_lfsck2name(com->lc_lfsck),
1221                                lls->lls_seq, rc1);
1222                 }
1223         }
1224
1225         return rc1;
1226 }
1227
1228 static int
1229 lfsck_layout_lastid_load(const struct lu_env *env,
1230                          struct lfsck_component *com,
1231                          struct lfsck_layout_seq *lls)
1232 {
1233         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1234         struct lfsck_layout     *lo     = com->lc_file_ram;
1235         struct lu_fid           *fid    = &lfsck_env_info(env)->lti_fid;
1236         struct dt_object        *obj;
1237         loff_t                   pos    = 0;
1238         int                      rc;
1239         ENTRY;
1240
1241         lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck));
1242         obj = dt_locate(env, lfsck->li_bottom, fid);
1243         if (IS_ERR(obj))
1244                 RETURN(PTR_ERR(obj));
1245
1246         /* LAST_ID crashed, to be rebuilt */
1247         if (dt_object_exists(obj) == 0) {
1248                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1249                         LASSERT(lfsck->li_out_notify != NULL);
1250
1251                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1252                                              LE_LASTID_REBUILDING);
1253                         lo->ll_flags |= LF_CRASHED_LASTID;
1254
1255                         CDEBUG(D_LFSCK, "%s: layout LFSCK cannot find the "
1256                                "LAST_ID file for sequence %#llx\n",
1257                                lfsck_lfsck2name(lfsck), lls->lls_seq);
1258
1259                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1260                             cfs_fail_val > 0) {
1261                                 struct l_wait_info lwi = LWI_TIMEOUT(
1262                                                 cfs_time_seconds(cfs_fail_val),
1263                                                 NULL, NULL);
1264
1265                                 /* Some others may changed the cfs_fail_val
1266                                  * as zero after above check, re-check it for
1267                                  * sure to avoid falling into wait for ever. */
1268                                 if (likely(lwi.lwi_timeout > 0)) {
1269                                         struct ptlrpc_thread *thread =
1270                                                 &lfsck->li_thread;
1271
1272                                         up_write(&com->lc_sem);
1273                                         l_wait_event(thread->t_ctl_waitq,
1274                                                      !thread_is_running(thread),
1275                                                      &lwi);
1276                                         down_write(&com->lc_sem);
1277                                 }
1278                         }
1279                 }
1280
1281                 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1282         } else {
1283                 dt_read_lock(env, obj, 0);
1284                 rc = dt_read(env, obj,
1285                         lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1286                         &pos);
1287                 dt_read_unlock(env, obj);
1288                 if (rc != 0 && rc != sizeof(__u64))
1289                         GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1290
1291                 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1292                         LASSERT(lfsck->li_out_notify != NULL);
1293
1294                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1295                                              LE_LASTID_REBUILDING);
1296                         lo->ll_flags |= LF_CRASHED_LASTID;
1297
1298                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds invalid "
1299                                "LAST_ID file for the sequence %#llx"
1300                                ": rc = %d\n",
1301                                lfsck_lfsck2name(lfsck), lls->lls_seq, rc);
1302                 }
1303
1304                 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1305                 rc = 0;
1306         }
1307
1308         GOTO(out, rc);
1309
1310 out:
1311         if (rc != 0)
1312                 lfsck_object_put(env, obj);
1313         else
1314                 lls->lls_lastid_obj = obj;
1315
1316         return rc;
1317 }
1318
1319 static void lfsck_layout_record_failure(const struct lu_env *env,
1320                                         struct lfsck_instance *lfsck,
1321                                         struct lfsck_layout *lo)
1322 {
1323         __u64 cookie;
1324
1325         lo->ll_objs_failed_phase1++;
1326         cookie = lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
1327                                                         lfsck->li_di_oit);
1328         if (lo->ll_pos_first_inconsistent == 0 ||
1329             lo->ll_pos_first_inconsistent < cookie) {
1330                 lo->ll_pos_first_inconsistent = cookie;
1331
1332                 CDEBUG(D_LFSCK, "%s: layout LFSCK hit first non-repaired "
1333                        "inconsistency at the pos [%llu]\n",
1334                        lfsck_lfsck2name(lfsck),
1335                        lo->ll_pos_first_inconsistent);
1336         }
1337 }
1338
1339 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1340                                            struct lfsck_component *com,
1341                                            int rc)
1342 {
1343         struct lfsck_instance   *lfsck = com->lc_lfsck;
1344         struct lfsck_layout     *lo    = com->lc_file_ram;
1345
1346         down_write(&com->lc_sem);
1347         lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1348                                   HALF_SEC - com->lc_time_last_checkpoint);
1349         lo->ll_time_last_checkpoint = cfs_time_current_sec();
1350         lo->ll_objs_checked_phase2 += com->lc_new_checked;
1351
1352         if (rc > 0) {
1353                 if (lo->ll_flags & LF_INCOMPLETE) {
1354                         lo->ll_status = LS_PARTIAL;
1355                 } else {
1356                         if (lfsck->li_master) {
1357                                 struct lfsck_assistant_data *lad = com->lc_data;
1358
1359                                 if (lad->lad_incomplete)
1360                                         lo->ll_status = LS_PARTIAL;
1361                                 else
1362                                         lo->ll_status = LS_COMPLETED;
1363                         } else {
1364                                 lo->ll_status = LS_COMPLETED;
1365                         }
1366                 }
1367                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
1368                         lo->ll_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
1369                 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1370                 lo->ll_success_count++;
1371         } else if (rc == 0) {
1372                 if (lfsck->li_status != 0)
1373                         lo->ll_status = lfsck->li_status;
1374                 else
1375                         lo->ll_status = LS_STOPPED;
1376         } else {
1377                 lo->ll_status = LS_FAILED;
1378         }
1379
1380         rc = lfsck_layout_store(env, com);
1381         up_write(&com->lc_sem);
1382
1383         return rc;
1384 }
1385
1386 static int lfsck_layout_trans_stop(const struct lu_env *env,
1387                                    struct dt_device *dev,
1388                                    struct thandle *handle, int result)
1389 {
1390         int rc;
1391
1392         /* XXX: If there is something worng or it needs to repair nothing,
1393          *      then notify the lower to stop the modification. Currently,
1394          *      we use th_result for such purpose, that may be replaced by
1395          *      some rollback mechanism in the future. */
1396         handle->th_result = result;
1397         rc = dt_trans_stop(env, dev, handle);
1398         if (result != 0)
1399                 return result > 0 ? 0 : result;
1400
1401         return rc == 0 ? 1 : rc;
1402 }
1403
1404 static int lfsck_layout_ins_dangling_rec(const struct lu_env *env,
1405                                          struct lfsck_component *com,
1406                                          const struct lu_fid *pfid,
1407                                          const struct lu_fid *cfid,
1408                                          __u32 ea_off, __u32 ost_idx)
1409 {
1410         struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
1411         struct lu_fid *rec = &lfsck_env_info(env)->lti_fid4;
1412         struct dt_device *dev;
1413         struct dt_object *obj;
1414         struct thandle *th = NULL;
1415         int idx;
1416         int rc = 0;
1417         ENTRY;
1418
1419         idx = lfsck_sub_trace_file_fid2idx(pfid);
1420         obj = com->lc_sub_trace_objs[idx].lsto_obj;
1421         dev = lfsck_obj2dev(obj);
1422         fid_cpu_to_be(key, pfid);
1423         key->f_ver = cpu_to_be32(ea_off);
1424         fid_cpu_to_be(rec, cfid);
1425         rec->f_ver = cpu_to_be32(ost_idx);
1426
1427         mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1428
1429         th = dt_trans_create(env, dev);
1430         if (IS_ERR(th))
1431                 GOTO(unlock, rc = PTR_ERR(th));
1432
1433         rc = dt_declare_insert(env, obj,
1434                                (const struct dt_rec *)rec,
1435                                (const struct dt_key *)key, th);
1436         if (rc)
1437                 GOTO(unlock, rc);
1438
1439         rc = dt_trans_start_local(env, dev, th);
1440         if (rc)
1441                 GOTO(unlock, rc);
1442
1443         rc = dt_insert(env, obj, (const struct dt_rec *)rec,
1444                        (const struct dt_key *)key, th, 1);
1445
1446         GOTO(unlock, rc);
1447
1448 unlock:
1449         if (th != NULL && !IS_ERR(th))
1450                 dt_trans_stop(env, dev, th);
1451
1452         mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1453
1454         CDEBUG(D_LFSCK, "%s: insert the paris "DFID" => "DFID", ea_off = %u, "
1455                "ost_idx = %u, into the trace file for further dangling check: "
1456                "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
1457                PFID(pfid), PFID(cfid), ea_off, ost_idx, rc);
1458
1459         return rc;
1460 }
1461
1462 static int lfsck_layout_del_dangling_rec(const struct lu_env *env,
1463                                          struct lfsck_component *com,
1464                                          const struct lu_fid *fid,
1465                                          __u32 ea_off)
1466 {
1467         struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
1468         struct dt_device *dev;
1469         struct dt_object *obj;
1470         struct thandle *th = NULL;
1471         int idx;
1472         int rc = 0;
1473         ENTRY;
1474
1475         idx = lfsck_sub_trace_file_fid2idx(fid);
1476         obj = com->lc_sub_trace_objs[idx].lsto_obj;
1477         dev = lfsck_obj2dev(obj);
1478         fid_cpu_to_be(key, fid);
1479         key->f_ver = cpu_to_be32(ea_off);
1480
1481         mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1482
1483         th = dt_trans_create(env, dev);
1484         if (IS_ERR(th))
1485                 GOTO(unlock, rc = PTR_ERR(th));
1486
1487         rc = dt_declare_delete(env, obj, (const struct dt_key *)key, th);
1488         if (rc)
1489                 GOTO(unlock, rc);
1490
1491         rc = dt_trans_start_local(env, dev, th);
1492         if (rc)
1493                 GOTO(unlock, rc);
1494
1495         rc = dt_delete(env, obj, (const struct dt_key *)key, th);
1496
1497         GOTO(unlock, rc);
1498
1499 unlock:
1500         if (th != NULL && !IS_ERR(th))
1501                 dt_trans_stop(env, dev, th);
1502
1503         mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1504
1505         CDEBUG(D_LFSCK, "%s: delete the dangling record for "DFID
1506                ", ea_off = %u from the trace file: rc = %d\n",
1507                lfsck_lfsck2name(com->lc_lfsck), PFID(fid), ea_off, rc);
1508
1509         return rc;
1510 }
1511
1512 /**
1513  * Get the system default stripe size.
1514  *
1515  * \param[in] env       pointer to the thread context
1516  * \param[in] lfsck     pointer to the lfsck instance
1517  * \param[out] size     pointer to the default stripe size
1518  *
1519  * \retval              0 for success
1520  * \retval              negative error number on failure
1521  */
1522 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1523                                            struct lfsck_instance *lfsck,
1524                                            __u32 *size)
1525 {
1526         struct lov_user_md      *lum = &lfsck_env_info(env)->lti_lum;
1527         struct dt_object        *root;
1528         int                      rc;
1529
1530         root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1531         if (IS_ERR(root))
1532                 return PTR_ERR(root);
1533
1534         /* Get the default stripe size via xattr_get on the backend root. */
1535         rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1536                           XATTR_NAME_LOV);
1537         if (rc > 0) {
1538                 /* The lum->lmm_stripe_size is LE mode. The *size also
1539                  * should be LE mode. So it is unnecessary to convert. */
1540                 *size = lum->lmm_stripe_size;
1541                 rc = 0;
1542         } else if (unlikely(rc == 0)) {
1543                 rc = -EINVAL;
1544         }
1545
1546         lfsck_object_put(env, root);
1547
1548         return rc;
1549 }
1550
1551 /**
1552  * \retval       +1: repaired
1553  * \retval        0: did nothing
1554  * \retval      -ve: on error
1555  */
1556 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1557                                      struct thandle *handle,
1558                                      struct dt_object *parent,
1559                                      struct lu_fid *cfid,
1560                                      struct lu_buf *buf,
1561                                      struct lov_ost_data_v1 *slot,
1562                                      int fl, __u32 ost_idx)
1563 {
1564         struct ost_id           *oi     = &lfsck_env_info(env)->lti_oi;
1565         struct lov_mds_md_v1    *lmm    = buf->lb_buf;
1566         struct lu_buf            ea_buf;
1567         int                      rc;
1568         __u32                    magic;
1569         __u16                    count;
1570         ENTRY;
1571
1572         magic = le32_to_cpu(lmm->lmm_magic);
1573         count = le16_to_cpu(lmm->lmm_stripe_count);
1574
1575         fid_to_ostid(cfid, oi);
1576         ostid_cpu_to_le(oi, &slot->l_ost_oi);
1577         slot->l_ost_gen = cpu_to_le32(0);
1578         slot->l_ost_idx = cpu_to_le32(ost_idx);
1579
1580         if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE) {
1581                 struct lov_ost_data_v1 *objs;
1582                 int                     i;
1583
1584                 if (magic == LOV_MAGIC_V1)
1585                         objs = &lmm->lmm_objects[0];
1586                 else
1587                         objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1588                 for (i = 0; i < count; i++, objs++) {
1589                         if (objs != slot && lovea_slot_is_dummy(objs))
1590                                 break;
1591                 }
1592
1593                 /* If the @slot is the last dummy slot to be refilled,
1594                  * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1595                 if (i == count)
1596                         lmm->lmm_pattern &= ~cpu_to_le32(LOV_PATTERN_F_HOLE);
1597         }
1598
1599         lfsck_buf_init(&ea_buf, lmm, lov_mds_md_size(count, magic));
1600         rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle);
1601         if (rc == 0)
1602                 rc = 1;
1603
1604         RETURN(rc);
1605 }
1606
1607 /**
1608  * \retval       +1: repaired
1609  * \retval        0: did nothing
1610  * \retval      -ve: on error
1611  */
1612 static int lfsck_layout_extend_lovea(const struct lu_env *env,
1613                                      struct lfsck_instance *lfsck,
1614                                      struct thandle *handle,
1615                                      struct dt_object *parent,
1616                                      struct lu_fid *cfid,
1617                                      struct lu_buf *buf, int fl,
1618                                      __u32 ost_idx, __u32 ea_off, bool reset)
1619 {
1620         struct lov_mds_md_v1    *lmm    = buf->lb_buf;
1621         struct lov_ost_data_v1  *objs;
1622         int                      rc;
1623         __u16                    count;
1624         bool                     hole   = false;
1625         ENTRY;
1626
1627         if (fl == LU_XATTR_CREATE || reset) {
1628                 __u32 pattern = LOV_PATTERN_RAID0;
1629
1630                 count = ea_off + 1;
1631                 LASSERT(buf->lb_len >= lov_mds_md_size(count, LOV_MAGIC_V1));
1632
1633                 if (ea_off != 0 || reset) {
1634                         pattern |= LOV_PATTERN_F_HOLE;
1635                         hole = true;
1636                 }
1637
1638                 memset(lmm, 0, buf->lb_len);
1639                 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1640                 lmm->lmm_pattern = cpu_to_le32(pattern);
1641                 fid_to_lmm_oi(lfsck_dto2fid(parent), &lmm->lmm_oi);
1642                 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1643
1644                 rc = lfsck_layout_get_def_stripesize(env, lfsck,
1645                                                      &lmm->lmm_stripe_size);
1646                 if (rc != 0)
1647                         RETURN(rc);
1648
1649                 objs = &lmm->lmm_objects[ea_off];
1650         } else {
1651                 __u32   magic = le32_to_cpu(lmm->lmm_magic);
1652                 int     gap;
1653
1654                 count = le16_to_cpu(lmm->lmm_stripe_count);
1655                 if (magic == LOV_MAGIC_V1)
1656                         objs = &lmm->lmm_objects[count];
1657                 else
1658                         objs = &((struct lov_mds_md_v3 *)lmm)->
1659                                                         lmm_objects[count];
1660
1661                 gap = ea_off - count;
1662                 if (gap >= 0)
1663                         count = ea_off + 1;
1664                 LASSERT(buf->lb_len >= lov_mds_md_size(count, magic));
1665
1666                 if (gap > 0) {
1667                         memset(objs, 0, gap * sizeof(*objs));
1668                         lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
1669                         hole = true;
1670                 }
1671
1672                 lmm->lmm_layout_gen =
1673                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
1674                 objs += gap;
1675         }
1676
1677         lmm->lmm_stripe_count = cpu_to_le16(count);
1678         rc = lfsck_layout_refill_lovea(env, handle, parent, cfid, buf, objs,
1679                                        fl, ost_idx);
1680
1681         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
1682                DFID": parent "DFID", OST-index %u, stripe-index %u, fl %d, "
1683                "reset %s, %s LOV EA hole: rc = %d\n",
1684                lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1685                ost_idx, ea_off, fl, reset ? "yes" : "no",
1686                hole ? "with" : "without", rc);
1687
1688         RETURN(rc);
1689 }
1690
1691 static int __lfsck_layout_update_pfid(const struct lu_env *env,
1692                                       struct dt_object *child,
1693                                       const struct lu_fid *pfid, __u32 offset)
1694 {
1695         struct dt_device        *dev    = lfsck_obj2dev(child);
1696         struct filter_fid       *ff     = &lfsck_env_info(env)->lti_new_pfid;
1697         struct thandle          *handle;
1698         struct lu_buf            buf    = { NULL };
1699         int                      rc;
1700
1701         ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
1702         ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
1703         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
1704          * MDT-object's FID::f_ver, instead it is the OST-object index in its
1705          * parent MDT-object's layout EA. */
1706         ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
1707         lfsck_buf_init(&buf, ff, sizeof(struct filter_fid));
1708
1709         handle = dt_trans_create(env, dev);
1710         if (IS_ERR(handle))
1711                 RETURN(PTR_ERR(handle));
1712
1713         rc = dt_declare_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
1714         if (rc != 0)
1715                 GOTO(stop, rc);
1716
1717         rc = dt_trans_start_local(env, dev, handle);
1718         if (rc != 0)
1719                 GOTO(stop, rc);
1720
1721         rc = dt_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
1722
1723         GOTO(stop, rc);
1724
1725 stop:
1726         dt_trans_stop(env, dev, handle);
1727
1728         return rc;
1729 }
1730
1731 /**
1732  * \retval       +1: repaired
1733  * \retval        0: did nothing
1734  * \retval      -ve: on error
1735  */
1736 static int lfsck_layout_update_pfid(const struct lu_env *env,
1737                                     struct lfsck_component *com,
1738                                     struct dt_object *parent,
1739                                     struct lu_fid *cfid,
1740                                     struct dt_device *cdev, __u32 ea_off)
1741 {
1742         struct dt_object        *child;
1743         int                      rc     = 0;
1744         ENTRY;
1745
1746         child = lfsck_object_find_by_dev(env, cdev, cfid);
1747         if (IS_ERR(child))
1748                 RETURN(PTR_ERR(child));
1749
1750         rc = __lfsck_layout_update_pfid(env, child,
1751                                         lu_object_fid(&parent->do_lu), ea_off);
1752         lfsck_object_put(env, child);
1753
1754         RETURN(rc == 0 ? 1 : rc);
1755 }
1756
1757 /**
1758  * This function will create the MDT-object with the given (partial) LOV EA.
1759  *
1760  * Under some data corruption cases, the MDT-object of the file may be lost,
1761  * but its OST-objects, or some of them are there. The layout LFSCK needs to
1762  * re-create the MDT-object with the orphan OST-object(s) information.
1763  *
1764  * On the other hand, the LFSCK may has created some OST-object for repairing
1765  * dangling LOV EA reference, but as the LFSCK processing, it may find that
1766  * the old OST-object is there and should replace the former new created OST
1767  * object. Unfortunately, some others have modified such newly created object.
1768  * To keep the data (both new and old), the LFSCK will create MDT-object with
1769  * new FID to reference the original OST-object.
1770  *
1771  * \param[in] env       pointer to the thread context
1772  * \param[in] com       pointer to the lfsck component
1773  * \param[in] ltd       pointer to target device descriptor
1774  * \param[in] rec       pointer to the record for the orphan OST-object
1775  * \param[in] cfid      pointer to FID for the orphan OST-object
1776  * \param[in] infix     additional information, such as the FID for original
1777  *                      MDT-object and the stripe offset in the LOV EA
1778  * \param[in] type      the type for describing why the orphan MDT-object is
1779  *                      created. The rules are as following:
1780  *
1781  *  type "C":           Multiple OST-objects claim the same MDT-object and the
1782  *                      same slot in the layout EA. Then the LFSCK will create
1783  *                      new MDT-object(s) to hold the conflict OST-object(s).
1784  *
1785  *  type "N":           The orphan OST-object does not know which one was the
1786  *                      real parent MDT-object, so the LFSCK uses new FID for
1787  *                      its parent MDT-object.
1788  *
1789  *  type "R":           The orphan OST-object knows its parent MDT-object FID,
1790  *                      but does not know the position (the file name) in the
1791  *                      layout.
1792  *
1793  *  type "D":           The MDT-object is a directory, it may knows its parent
1794  *                      but because there is no valid linkEA, the LFSCK cannot
1795  *                      know where to put it back to the namespace.
1796  *  type "O":           The MDT-object has no linkEA, and there is no name
1797  *                      entry that references the MDT-object.
1798  *
1799  *  type "P":           The orphan object to be created was a parent directory
1800  *                      of some MDT-object which linkEA shows that the @orphan
1801  *                      object is missing.
1802  *
1803  * The orphan name will be like:
1804  * ${FID}-${infix}-${type}-${conflict_version}
1805  *
1806  * \param[in] ea_off    the stripe offset in the LOV EA
1807  *
1808  * \retval              positive on repaired something
1809  * \retval              0 if needs to repair nothing
1810  * \retval              negative error number on failure
1811  */
1812 static int lfsck_layout_recreate_parent(const struct lu_env *env,
1813                                         struct lfsck_component *com,
1814                                         struct lfsck_tgt_desc *ltd,
1815                                         struct lu_orphan_rec *rec,
1816                                         struct lu_fid *cfid,
1817                                         const char *infix,
1818                                         const char *type,
1819                                         __u32 ea_off)
1820 {
1821         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1822         struct dt_insert_rec            *dtrec  = &info->lti_dt_rec;
1823         char                            *name   = info->lti_key;
1824         struct lu_attr                  *la     = &info->lti_la2;
1825         struct dt_object_format         *dof    = &info->lti_dof;
1826         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1827         struct lu_fid                   *pfid   = &rec->lor_fid;
1828         struct lu_fid                   *tfid   = &info->lti_fid3;
1829         struct dt_device                *dev    = lfsck->li_bottom;
1830         struct dt_object                *lpf    = lfsck->li_lpf_obj;
1831         struct dt_object                *pobj   = NULL;
1832         struct dt_object                *cobj   = NULL;
1833         struct thandle                  *th     = NULL;
1834         struct lu_buf                   *ea_buf = &info->lti_big_buf;
1835         struct lu_buf                    lov_buf;
1836         struct lfsck_lock_handle        *llh    = &info->lti_llh;
1837         struct linkea_data               ldata  = { NULL };
1838         struct lu_buf                    linkea_buf;
1839         const struct lu_name            *pname;
1840         int                              size   = 0;
1841         int                              idx    = 0;
1842         int                              rc     = 0;
1843         ENTRY;
1844
1845         if (unlikely(lpf == NULL))
1846                 GOTO(log, rc = -ENXIO);
1847
1848         /* We use two separated transactions to repair the inconsistency.
1849          *
1850          * 1) create the MDT-object locally.
1851          * 2) update the OST-object's PFID EA if necessary.
1852          *
1853          * If 1) succeed, but 2) failed, then the OST-object's PFID EA will be
1854          * updated when the layout LFSCK run next time.
1855          *
1856          * If 1) failed, but 2) succeed, then such MDT-object will be re-created
1857          * when the layout LFSCK run next time. */
1858
1859         if (fid_is_zero(pfid)) {
1860                 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
1861                 if (rc != 0)
1862                         GOTO(log, rc);
1863
1864                 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
1865                 if (IS_ERR(cobj))
1866                         GOTO(log, rc = PTR_ERR(cobj));
1867         }
1868
1869         pobj = lfsck_object_find_by_dev(env, dev, pfid);
1870         if (IS_ERR(pobj))
1871                 GOTO(log, rc = PTR_ERR(pobj));
1872
1873         LASSERT(infix != NULL);
1874         LASSERT(type != NULL);
1875
1876         memset(la, 0, sizeof(*la));
1877         la->la_uid = rec->lor_uid;
1878         la->la_gid = rec->lor_gid;
1879         la->la_mode = S_IFREG | S_IRUSR;
1880         la->la_valid = LA_MODE | LA_UID | LA_GID;
1881
1882         memset(dof, 0, sizeof(*dof));
1883         dof->dof_type = dt_mode_to_dft(S_IFREG);
1884         /* Because the dof->dof_reg.striped = 0, the LOD will not create
1885          * the stripe(s). The LFSCK will specify the LOV EA via
1886          * lfsck_layout_extend_lovea(). */
1887
1888         size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
1889         if (ea_buf->lb_len < size) {
1890                 lu_buf_realloc(ea_buf, size);
1891                 if (ea_buf->lb_buf == NULL)
1892                         GOTO(log, rc = -ENOMEM);
1893         }
1894
1895 again:
1896         do {
1897                 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
1898                          type, idx++);
1899                 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1900                                (const struct dt_key *)name);
1901                 if (rc != 0 && rc != -ENOENT)
1902                         GOTO(log, rc);
1903         } while (rc == 0);
1904
1905         rc = lfsck_lock(env, lfsck, lfsck->li_lpf_obj, name, llh,
1906                         MDS_INODELOCK_UPDATE, LCK_PW);
1907         if (rc != 0)
1908                 GOTO(log, rc);
1909
1910         /* Re-check whether the name conflict with othrs after taken
1911          * the ldlm lock. */
1912         rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1913                        (const struct dt_key *)name);
1914         if (unlikely(rc == 0)) {
1915                 lfsck_unlock(llh);
1916                 goto again;
1917         }
1918
1919         if (rc != -ENOENT)
1920                 GOTO(unlock, rc);
1921
1922         pname = lfsck_name_get_const(env, name, strlen(name));
1923         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
1924                               pname, lfsck_dto2fid(lfsck->li_lpf_obj));
1925         if (rc != 0)
1926                 GOTO(unlock, rc);
1927
1928         /* The 1st transaction. */
1929         th = dt_trans_create(env, dev);
1930         if (IS_ERR(th))
1931                 GOTO(unlock, rc = PTR_ERR(th));
1932
1933         rc = dt_declare_create(env, pobj, la, NULL, dof, th);
1934         if (rc != 0)
1935                 GOTO(stop, rc);
1936
1937         lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
1938         rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
1939                                   LU_XATTR_REPLACE, th);
1940         if (rc != 0)
1941                 GOTO(stop, rc);
1942
1943         dtrec->rec_fid = pfid;
1944         dtrec->rec_type = S_IFREG;
1945         rc = dt_declare_insert(env, lpf,
1946                                (const struct dt_rec *)dtrec,
1947                                (const struct dt_key *)name, th);
1948         if (rc != 0)
1949                 GOTO(stop, rc);
1950
1951         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1952                        ldata.ld_leh->leh_len);
1953         rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
1954                                   XATTR_NAME_LINK, 0, th);
1955         if (rc != 0)
1956                 GOTO(stop, rc);
1957
1958         rc = dt_trans_start_local(env, dev, th);
1959         if (rc != 0)
1960                 GOTO(stop, rc);
1961
1962         dt_write_lock(env, pobj, 0);
1963         rc = dt_create(env, pobj, la, NULL, dof, th);
1964         if (rc == 0)
1965                 rc = lfsck_layout_extend_lovea(env, lfsck, th, pobj, cfid,
1966                                 &lov_buf, 0, ltd->ltd_index, ea_off, true);
1967         dt_write_unlock(env, pobj);
1968         if (rc < 0)
1969                 GOTO(stop, rc);
1970
1971         rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
1972                        (const struct dt_key *)name, th, 1);
1973         if (rc != 0)
1974                 GOTO(stop, rc);
1975
1976         rc = dt_xattr_set(env, pobj, &linkea_buf, XATTR_NAME_LINK, 0, th);
1977         if (rc == 0 && cobj != NULL) {
1978                 dt_trans_stop(env, dev, th);
1979                 th = NULL;
1980
1981                 /* The 2nd transaction. */
1982                 rc = __lfsck_layout_update_pfid(env, cobj, pfid, ea_off);
1983         }
1984
1985         GOTO(stop, rc);
1986
1987 stop:
1988         if (th != NULL)
1989                 dt_trans_stop(env, dev, th);
1990
1991 unlock:
1992         lfsck_unlock(llh);
1993
1994 log:
1995         if (cobj != NULL && !IS_ERR(cobj))
1996                 lfsck_object_put(env, cobj);
1997         if (pobj != NULL && !IS_ERR(pobj))
1998                 lfsck_object_put(env, pobj);
1999
2000         if (rc < 0)
2001                 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
2002                        "recreate the lost MDT-object: parent "DFID
2003                        ", child "DFID", OST-index %u, stripe-index %u, "
2004                        "infix %s, type %s: rc = %d\n",
2005                        lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2006                        ltd->ltd_index, ea_off, infix, type, rc);
2007
2008         return rc >= 0 ? 1 : rc;
2009 }
2010
2011 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
2012                                                    struct lfsck_component *com,
2013                                                    const struct lu_fid *fid,
2014                                                    __u32 index)
2015 {
2016         struct lfsck_thread_info *info  = lfsck_env_info(env);
2017         struct lfsck_request     *lr    = &info->lti_lr;
2018         struct lfsck_instance    *lfsck = com->lc_lfsck;
2019         struct lfsck_tgt_desc    *ltd;
2020         struct ptlrpc_request    *req;
2021         struct lfsck_request     *tmp;
2022         struct obd_export        *exp;
2023         int                       rc    = 0;
2024         ENTRY;
2025
2026         ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
2027         if (unlikely(ltd == NULL))
2028                 RETURN(-ENXIO);
2029
2030         exp = ltd->ltd_exp;
2031         if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
2032                 GOTO(put, rc = -EOPNOTSUPP);
2033
2034         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
2035         if (req == NULL)
2036                 GOTO(put, rc = -ENOMEM);
2037
2038         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
2039         if (rc != 0) {
2040                 ptlrpc_request_free(req);
2041
2042                 GOTO(put, rc);
2043         }
2044
2045         memset(lr, 0, sizeof(*lr));
2046         lr->lr_event = LE_CONDITIONAL_DESTROY;
2047         lr->lr_active = LFSCK_TYPE_LAYOUT;
2048         lr->lr_fid = *fid;
2049
2050         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2051         *tmp = *lr;
2052         ptlrpc_request_set_replen(req);
2053
2054         rc = ptlrpc_queue_wait(req);
2055         ptlrpc_req_finished(req);
2056
2057         GOTO(put, rc);
2058
2059 put:
2060         lfsck_tgt_put(ltd);
2061
2062         return rc;
2063 }
2064
2065 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
2066                                                   struct lfsck_component *com,
2067                                                   struct lfsck_request *lr)
2068 {
2069         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2070         struct lu_attr                  *la     = &info->lti_la;
2071         union ldlm_policy_data          *policy = &info->lti_policy;
2072         struct ldlm_res_id              *resid  = &info->lti_resid;
2073         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2074         struct dt_device                *dev    = lfsck->li_bottom;
2075         struct lu_fid                   *fid    = &lr->lr_fid;
2076         struct dt_object                *obj;
2077         struct thandle                  *th     = NULL;
2078         struct lustre_handle             lh     = { 0 };
2079         __u64                            flags  = 0;
2080         int                              rc     = 0;
2081         ENTRY;
2082
2083         obj = lfsck_object_find_by_dev(env, dev, fid);
2084         if (IS_ERR(obj))
2085                 RETURN(PTR_ERR(obj));
2086
2087         dt_read_lock(env, obj, 0);
2088         if (dt_object_exists(obj) == 0 ||
2089             lfsck_is_dead_obj(obj)) {
2090                 dt_read_unlock(env, obj);
2091
2092                 GOTO(put, rc = -ENOENT);
2093         }
2094
2095         /* Get obj's attr without lock firstly. */
2096         rc = dt_attr_get(env, obj, la);
2097         dt_read_unlock(env, obj);
2098         if (rc != 0)
2099                 GOTO(put, rc);
2100
2101         if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
2102                 GOTO(put, rc = -ETXTBSY);
2103
2104         /* Acquire extent lock on [0, EOF] to sync with all possible written. */
2105         LASSERT(lfsck->li_namespace != NULL);
2106
2107         memset(policy, 0, sizeof(*policy));
2108         policy->l_extent.end = OBD_OBJECT_EOF;
2109         ost_fid_build_resid(fid, resid);
2110         rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
2111                                     policy, LCK_EX, &flags, ldlm_blocking_ast,
2112                                     ldlm_completion_ast, NULL, NULL, 0,
2113                                     LVB_T_NONE, NULL, &lh);
2114         if (rc != ELDLM_OK)
2115                 GOTO(put, rc = -EIO);
2116
2117         dt_write_lock(env, obj, 0);
2118         /* Get obj's attr within lock again. */
2119         rc = dt_attr_get(env, obj, la);
2120         if (rc != 0)
2121                 GOTO(unlock, rc);
2122
2123         if (la->la_ctime != 0)
2124                 GOTO(unlock, rc = -ETXTBSY);
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_ref_del(env, obj, th);
2131         if (rc != 0)
2132                 GOTO(stop, rc);
2133
2134         rc = dt_declare_destroy(env, obj, th);
2135         if (rc != 0)
2136                 GOTO(stop, rc);
2137
2138         rc = dt_trans_start_local(env, dev, th);
2139         if (rc != 0)
2140                 GOTO(stop, rc);
2141
2142         rc = dt_ref_del(env, obj, th);
2143         if (rc != 0)
2144                 GOTO(stop, rc);
2145
2146         rc = dt_destroy(env, obj, th);
2147         if (rc == 0)
2148                 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2149                        "OST-object "DFID" that was created for reparing "
2150                        "dangling referenced case. But the original missing "
2151                        "OST-object is found now.\n",
2152                        lfsck_lfsck2name(lfsck), PFID(fid));
2153
2154         GOTO(stop, rc);
2155
2156 stop:
2157         dt_trans_stop(env, dev, th);
2158
2159 unlock:
2160         dt_write_unlock(env, obj);
2161         ldlm_lock_decref(&lh, LCK_EX);
2162
2163 put:
2164         lfsck_object_put(env, obj);
2165
2166         return rc;
2167 }
2168
2169 /**
2170  * Some OST-object has occupied the specified layout EA slot.
2171  * Such OST-object may be generated by the LFSCK when repair
2172  * dangling referenced MDT-object, which can be indicated by
2173  * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2174  * is true and such OST-object has not been modified yet, we
2175  * will replace it with the orphan OST-object; otherwise the
2176  * LFSCK will create new MDT-object to reference the orphan.
2177  *
2178  * \retval       +1: repaired
2179  * \retval        0: did nothing
2180  * \retval      -ve: on error
2181  */
2182 static int lfsck_layout_conflict_create(const struct lu_env *env,
2183                                         struct lfsck_component *com,
2184                                         struct lfsck_tgt_desc *ltd,
2185                                         struct lu_orphan_rec *rec,
2186                                         struct dt_object *parent,
2187                                         struct lu_fid *cfid,
2188                                         struct lu_buf *ea_buf,
2189                                         struct lov_ost_data_v1 *slot,
2190                                         __u32 ea_off)
2191 {
2192         struct lfsck_thread_info *info          = lfsck_env_info(env);
2193         struct lu_fid            *cfid2         = &info->lti_fid2;
2194         struct ost_id            *oi            = &info->lti_oi;
2195         struct lov_mds_md_v1     *lmm           = ea_buf->lb_buf;
2196         struct dt_device         *dev           = lfsck_obj2dev(parent);
2197         struct thandle           *th            = NULL;
2198         struct lustre_handle      lh            = { 0 };
2199         __u32                     ost_idx2      = le32_to_cpu(slot->l_ost_idx);
2200         int                       rc            = 0;
2201         ENTRY;
2202
2203         while (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val)) {
2204                 if (unlikely(!thread_is_running(&com->lc_lfsck->li_thread)))
2205                         RETURN(0);
2206         }
2207
2208         ostid_le_to_cpu(&slot->l_ost_oi, oi);
2209         rc = ostid_to_fid(cfid2, oi, ost_idx2);
2210         if (rc != 0)
2211                 GOTO(out, rc);
2212
2213         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2214                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2215                               LCK_EX);
2216         if (rc != 0)
2217                 GOTO(out, rc);
2218
2219         rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2220
2221         /* If the conflict OST-obejct is not created for fixing dangling
2222          * referenced MDT-object in former LFSCK check/repair, or it has
2223          * been modified by others, then we cannot destroy it. Re-create
2224          * a new MDT-object for the orphan OST-object. */
2225         if (rc == -ETXTBSY) {
2226                 /* No need the layout lock on the original parent. */
2227                 lfsck_ibits_unlock(&lh, LCK_EX);
2228
2229                 fid_zero(&rec->lor_fid);
2230                 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2231                          "-"DFID"-%x", PFID(lu_object_fid(&parent->do_lu)),
2232                          ea_off);
2233                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2234                                                 info->lti_tmpbuf, "C", ea_off);
2235
2236                 RETURN(rc);
2237         }
2238
2239         if (rc != 0 && rc != -ENOENT)
2240                 GOTO(unlock, rc);
2241
2242         th = dt_trans_create(env, dev);
2243         if (IS_ERR(th))
2244                 GOTO(unlock, rc = PTR_ERR(th));
2245
2246         rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2247                                   LU_XATTR_REPLACE, th);
2248         if (rc != 0)
2249                 GOTO(stop, rc);
2250
2251         rc = dt_trans_start_local(env, dev, th);
2252         if (rc != 0)
2253                 GOTO(stop, rc);
2254
2255         dt_write_lock(env, parent, 0);
2256         lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2257         rc = lfsck_layout_refill_lovea(env, th, parent, cfid, ea_buf, slot,
2258                                        LU_XATTR_REPLACE, ltd->ltd_index);
2259         dt_write_unlock(env, parent);
2260
2261         GOTO(stop, rc);
2262
2263 stop:
2264         dt_trans_stop(env, dev, th);
2265
2266 unlock:
2267         lfsck_ibits_unlock(&lh, LCK_EX);
2268
2269 out:
2270         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2271                "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2272                "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2273                lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2274                PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2275                ea_off, rc);
2276
2277         return rc >= 0 ? 1 : rc;
2278 }
2279
2280 /**
2281  * \retval       +1: repaired
2282  * \retval        0: did nothing
2283  * \retval      -ve: on error
2284  */
2285 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2286                                        struct lfsck_component *com,
2287                                        struct lfsck_tgt_desc *ltd,
2288                                        struct lu_orphan_rec *rec,
2289                                        struct dt_object *parent,
2290                                        struct lu_fid *cfid,
2291                                        __u32 ost_idx, __u32 ea_off)
2292 {
2293         struct lfsck_thread_info *info          = lfsck_env_info(env);
2294         struct lu_buf            *buf           = &info->lti_big_buf;
2295         struct lu_fid            *fid           = &info->lti_fid2;
2296         struct ost_id            *oi            = &info->lti_oi;
2297         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2298         struct dt_device         *dt            = lfsck_obj2dev(parent);
2299         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2300         struct thandle            *handle       = NULL;
2301         size_t                    lovea_size;
2302         struct lov_mds_md_v1     *lmm;
2303         struct lov_ost_data_v1   *objs;
2304         struct lustre_handle      lh            = { 0 };
2305         __u32                     magic;
2306         int                       fl            = 0;
2307         int                       rc            = 0;
2308         int                       rc1;
2309         int                       i;
2310         __u16                     count;
2311         bool                      locked        = false;
2312         ENTRY;
2313
2314         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2315                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2316                               LCK_EX);
2317         if (rc != 0) {
2318                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2319                        "LOV EA for "DFID": parent "DFID", OST-index %u, "
2320                        "stripe-index %u: rc = %d\n",
2321                        lfsck_lfsck2name(lfsck), PFID(cfid),
2322                        PFID(lfsck_dto2fid(parent)), ost_idx, ea_off, rc);
2323
2324                 RETURN(rc);
2325         }
2326
2327 again:
2328         if (locked) {
2329                 dt_write_unlock(env, parent);
2330                 locked = false;
2331         }
2332
2333         if (handle != NULL) {
2334                 dt_trans_stop(env, dt, handle);
2335                 handle = NULL;
2336         }
2337
2338         if (rc < 0)
2339                 GOTO(unlock_layout, rc);
2340
2341         lovea_size = rc;
2342         if (buf->lb_len < lovea_size) {
2343                 lu_buf_realloc(buf, lovea_size);
2344                 if (buf->lb_buf == NULL)
2345                         GOTO(unlock_layout, rc = -ENOMEM);
2346         }
2347
2348         if (!(bk->lb_param & LPF_DRYRUN)) {
2349                 handle = dt_trans_create(env, dt);
2350                 if (IS_ERR(handle))
2351                         GOTO(unlock_layout, rc = PTR_ERR(handle));
2352
2353                 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2354                                           fl, handle);
2355                 if (rc != 0)
2356                         GOTO(stop, rc);
2357
2358                 rc = dt_trans_start_local(env, dt, handle);
2359                 if (rc != 0)
2360                         GOTO(stop, rc);
2361         }
2362
2363         dt_write_lock(env, parent, 0);
2364         locked = true;
2365         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2366         if (rc == -ERANGE) {
2367                 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV);
2368                 LASSERT(rc != 0);
2369                 goto again;
2370         } else if (rc == -ENODATA || rc == 0) {
2371                 lovea_size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2372                 /* If the declared is not big enough, re-try. */
2373                 if (buf->lb_len < lovea_size) {
2374                         rc = lovea_size;
2375                         goto again;
2376                 }
2377                 fl = LU_XATTR_CREATE;
2378         } else if (rc < 0) {
2379                 GOTO(unlock_parent, rc);
2380         } else if (unlikely(buf->lb_len == 0)) {
2381                 goto again;
2382         } else {
2383                 fl = LU_XATTR_REPLACE;
2384                 lovea_size = rc;
2385         }
2386
2387         if (fl == LU_XATTR_CREATE) {
2388                 if (bk->lb_param & LPF_DRYRUN)
2389                         GOTO(unlock_parent, rc = 1);
2390
2391                 LASSERT(buf->lb_len >= lovea_size);
2392
2393                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2394                                                buf, fl, ost_idx, ea_off, false);
2395
2396                 GOTO(unlock_parent, rc);
2397         }
2398
2399         lmm = buf->lb_buf;
2400         rc1 = lfsck_layout_verify_header(lmm);
2401
2402         /* If the LOV EA crashed, the rebuild it. */
2403         if (rc1 == -EINVAL) {
2404                 if (bk->lb_param & LPF_DRYRUN)
2405                         GOTO(unlock_parent, rc = 1);
2406
2407                 LASSERT(buf->lb_len >= lovea_size);
2408
2409                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2410                                                buf, fl, ost_idx, ea_off, true);
2411
2412                 GOTO(unlock_parent, rc);
2413         }
2414
2415         /* For other unknown magic/pattern, keep the current LOV EA. */
2416         if (rc1 != 0)
2417                 GOTO(unlock_parent, rc = rc1);
2418
2419         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2420          * been verified in lfsck_layout_verify_header() already. If some
2421          * new magic introduced in the future, then layout LFSCK needs to
2422          * be updated also. */
2423         magic = le32_to_cpu(lmm->lmm_magic);
2424         if (magic == LOV_MAGIC_V1) {
2425                 objs = &lmm->lmm_objects[0];
2426         } else {
2427                 LASSERT(magic == LOV_MAGIC_V3);
2428                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2429         }
2430
2431         count = le16_to_cpu(lmm->lmm_stripe_count);
2432         if (count == 0)
2433                 GOTO(unlock_parent, rc = -EINVAL);
2434         LASSERT(count > 0);
2435
2436         /* Exceed the current end of MDT-object layout EA. Then extend it. */
2437         if (count <= ea_off) {
2438                 if (bk->lb_param & LPF_DRYRUN)
2439                         GOTO(unlock_parent, rc = 1);
2440
2441                 lovea_size = lov_mds_md_size(ea_off + 1, magic);
2442                 /* If the declared is not big enough, re-try. */
2443                 if (buf->lb_len < lovea_size) {
2444                         rc = lovea_size;
2445                         goto again;
2446                 }
2447
2448                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2449                                                buf, fl, ost_idx, ea_off, false);
2450
2451                 GOTO(unlock_parent, rc);
2452         }
2453
2454         LASSERTF(rc > 0, "invalid rc = %d\n", rc);
2455
2456         for (i = 0; i < count; i++, objs++) {
2457                 /* The MDT-object was created via lfsck_layout_recover_create()
2458                  * by others before, and we fill the dummy layout EA. */
2459                 if (lovea_slot_is_dummy(objs)) {
2460                         if (i != ea_off)
2461                                 continue;
2462
2463                         if (bk->lb_param & LPF_DRYRUN)
2464                                 GOTO(unlock_parent, rc = 1);
2465
2466                         lmm->lmm_layout_gen =
2467                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2468                         rc = lfsck_layout_refill_lovea(env, handle, parent,
2469                                                        cfid, buf, objs, fl,
2470                                                        ost_idx);
2471
2472                         CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
2473                                "dummy layout slot for "DFID": parent "DFID
2474                                ", OST-index %u, stripe-index %u: rc = %d\n",
2475                                lfsck_lfsck2name(lfsck), PFID(cfid),
2476                                PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
2477
2478                         GOTO(unlock_parent, rc);
2479                 }
2480
2481                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2482                 rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
2483                 if (rc != 0) {
2484                         CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
2485                                "invalid layout EA at the slot %d, index %u\n",
2486                                lfsck_lfsck2name(lfsck),
2487                                PFID(lfsck_dto2fid(parent)), i,
2488                                le32_to_cpu(objs->l_ost_idx));
2489
2490                         GOTO(unlock_parent, rc);
2491                 }
2492
2493                 /* It should be rare case, the slot is there, but the LFSCK
2494                  * does not handle it during the first-phase cycle scanning. */
2495                 if (unlikely(lu_fid_eq(fid, cfid))) {
2496                         if (i == ea_off) {
2497                                 GOTO(unlock_parent, rc = 0);
2498                         } else {
2499                                 /* Rare case that the OST-object index
2500                                  * does not match the parent MDT-object
2501                                  * layout EA. We trust the later one. */
2502                                 if (bk->lb_param & LPF_DRYRUN)
2503                                         GOTO(unlock_parent, rc = 1);
2504
2505                                 dt_write_unlock(env, parent);
2506                                 if (handle != NULL)
2507                                         dt_trans_stop(env, dt, handle);
2508                                 lfsck_ibits_unlock(&lh, LCK_EX);
2509                                 rc = lfsck_layout_update_pfid(env, com, parent,
2510                                                         cfid, ltd->ltd_tgt, i);
2511
2512                                 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
2513                                        "updated OST-object's pfid for "DFID
2514                                        ": parent "DFID", OST-index %u, "
2515                                        "stripe-index %u: rc = %d\n",
2516                                        lfsck_lfsck2name(lfsck), PFID(cfid),
2517                                        PFID(lfsck_dto2fid(parent)),
2518                                        ltd->ltd_index, i, rc);
2519
2520                                 RETURN(rc);
2521                         }
2522                 }
2523         }
2524
2525         /* The MDT-object exists, but related layout EA slot is occupied
2526          * by others. */
2527         if (bk->lb_param & LPF_DRYRUN)
2528                 GOTO(unlock_parent, rc = 1);
2529
2530         dt_write_unlock(env, parent);
2531         if (handle != NULL)
2532                 dt_trans_stop(env, dt, handle);
2533         lfsck_ibits_unlock(&lh, LCK_EX);
2534         if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
2535                 objs = &lmm->lmm_objects[ea_off];
2536         else
2537                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
2538         rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
2539                                           buf, objs, ea_off);
2540
2541         RETURN(rc);
2542
2543 unlock_parent:
2544         if (locked)
2545                 dt_write_unlock(env, parent);
2546
2547 stop:
2548         if (handle != NULL)
2549                 dt_trans_stop(env, dt, handle);
2550
2551 unlock_layout:
2552         lfsck_ibits_unlock(&lh, LCK_EX);
2553
2554         return rc;
2555 }
2556
2557 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
2558                                         struct lfsck_component *com,
2559                                         struct lfsck_tgt_desc *ltd,
2560                                         struct lu_orphan_rec *rec,
2561                                         struct lu_fid *cfid)
2562 {
2563         struct lfsck_layout     *lo     = com->lc_file_ram;
2564         struct lu_fid           *pfid   = &rec->lor_fid;
2565         struct dt_object        *parent = NULL;
2566         __u32                    ea_off = pfid->f_stripe_idx;
2567         int                      rc     = 0;
2568         ENTRY;
2569
2570         if (!fid_is_sane(cfid))
2571                 GOTO(out, rc = -EINVAL);
2572
2573         if (fid_is_zero(pfid)) {
2574                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2575                                                   "", "N", ea_off);
2576                 GOTO(out, rc);
2577         }
2578
2579         pfid->f_ver = 0;
2580         if (!fid_is_sane(pfid))
2581                 GOTO(out, rc = -EINVAL);
2582
2583         parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
2584         if (IS_ERR(parent))
2585                 GOTO(out, rc = PTR_ERR(parent));
2586
2587         if (unlikely(dt_object_remote(parent) != 0))
2588                 GOTO(put, rc = -EXDEV);
2589
2590         if (dt_object_exists(parent) == 0) {
2591                 lfsck_object_put(env, parent);
2592                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2593                                                   "", "R", ea_off);
2594                 GOTO(out, rc);
2595         }
2596
2597         if (!S_ISREG(lu_object_attr(&parent->do_lu)))
2598                 GOTO(put, rc = -EISDIR);
2599
2600         /* The orphan OST-object claims to be the parent's stripe, then
2601          * related dangling record in the trace file is meaningless. */
2602         rc = lfsck_layout_del_dangling_rec(env, com, pfid, ea_off);
2603         if (rc != 0 && rc != -ENOENT)
2604                 GOTO(put, rc);
2605
2606         rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
2607                                          ltd->ltd_index, ea_off);
2608
2609         GOTO(put, rc);
2610
2611 put:
2612         if (rc <= 0)
2613                 lfsck_object_put(env, parent);
2614         else
2615                 /* The layout EA is changed, need to be reloaded next time. */
2616                 lu_object_put_nocache(env, &parent->do_lu);
2617
2618 out:
2619         down_write(&com->lc_sem);
2620         com->lc_new_scanned++;
2621         com->lc_new_checked++;
2622         if (rc > 0) {
2623                 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
2624                 rc = 0;
2625         } else if (rc < 0) {
2626                 lo->ll_objs_failed_phase2++;
2627         }
2628         up_write(&com->lc_sem);
2629
2630         return rc;
2631 }
2632
2633 static int lfsck_layout_scan_orphan(const struct lu_env *env,
2634                                     struct lfsck_component *com,
2635                                     struct lfsck_tgt_desc *ltd)
2636 {
2637         struct lfsck_assistant_data     *lad    = com->lc_data;
2638         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2639         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
2640         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2641         struct ost_id                   *oi     = &info->lti_oi;
2642         struct lu_fid                   *fid    = &info->lti_fid;
2643         struct dt_object                *obj;
2644         const struct dt_it_ops          *iops;
2645         struct dt_it                    *di;
2646         int                              rc     = 0;
2647         ENTRY;
2648
2649         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
2650                "scanning for OST%04x\n",
2651                lfsck_lfsck2name(lfsck), ltd->ltd_index);
2652
2653         if (cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
2654                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
2655                        "scanning for OST%04x\n",
2656                        lfsck_lfsck2name(lfsck), ltd->ltd_index);
2657
2658                 RETURN(0);
2659         }
2660
2661         ostid_set_seq(oi, FID_SEQ_IDIF);
2662         ostid_set_id(oi, 0);
2663         rc = ostid_to_fid(fid, oi, ltd->ltd_index);
2664         if (rc != 0)
2665                 GOTO(log, rc);
2666
2667         obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
2668         if (unlikely(IS_ERR(obj)))
2669                 GOTO(log, rc = PTR_ERR(obj));
2670
2671         rc = obj->do_ops->do_index_try(env, obj,
2672                                        &dt_lfsck_layout_orphan_features);
2673         if (rc != 0)
2674                 GOTO(put, rc);
2675
2676         iops = &obj->do_index_ops->dio_it;
2677         di = iops->init(env, obj, 0);
2678         if (IS_ERR(di))
2679                 GOTO(put, rc = PTR_ERR(di));
2680
2681         rc = iops->load(env, di, 0);
2682         if (rc == -ESRCH) {
2683                 /* -ESRCH means that the orphan OST-objects rbtree has been
2684                  * cleanup because of the OSS server restart or other errors. */
2685                 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
2686                 GOTO(fini, rc);
2687         }
2688
2689         if (rc == 0)
2690                 rc = iops->next(env, di);
2691         else if (rc > 0)
2692                 rc = 0;
2693
2694         if (rc < 0)
2695                 GOTO(fini, rc);
2696
2697         if (rc > 0)
2698                 GOTO(fini, rc = 0);
2699
2700         do {
2701                 struct dt_key           *key;
2702                 struct lu_orphan_rec    *rec = &info->lti_rec;
2703
2704                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
2705                     unlikely(!thread_is_running(&lfsck->li_thread)))
2706                         break;
2707
2708                 key = iops->key(env, di);
2709                 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
2710                 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
2711                 if (rc == 0)
2712                         rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
2713                                         &com->lc_fid_latest_scanned_phase2);
2714                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2715                         GOTO(fini, rc);
2716
2717                 lfsck_control_speed_by_self(com);
2718                 do {
2719                         rc = iops->next(env, di);
2720                 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
2721         } while (rc == 0);
2722
2723         GOTO(fini, rc);
2724
2725 fini:
2726         iops->put(env, di);
2727         iops->fini(env, di);
2728 put:
2729         lfsck_object_put(env, obj);
2730
2731 log:
2732         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
2733                "scanning for OST%04x: rc = %d\n",
2734                lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
2735
2736         return rc > 0 ? 0 : rc;
2737 }
2738
2739 /**
2740  * Repair the MDT-object with dangling LOV EA reference.
2741  *
2742  * we need to repair the inconsistency according to the users' requirement:
2743  *
2744  * 1) Keep the inconsistency there and report the inconsistency case,
2745  *    then give the chance to the application to find related issues,
2746  *    and the users can make the decision about how to handle it with
2747  *    more human knownledge. (by default)
2748  *
2749  * 2) Re-create the missing OST-object with the FID/owner information.
2750  *
2751  * \param[in] env       pointer to the thread context
2752  * \param[in] com       the layout LFSCK component
2753  * \param[in] parent    the MDT-object with dangling LOV EA reference
2754  * \param[in] child     the OST-object to be created
2755  * \param[in] ea_off    the offset of the OST-object in the LOV EA
2756  * \param[in] ost_idx   the index of OST on which the OST-object resides
2757  *
2758  * \retval              +1 for repair successfully
2759  * \retval              0 for did nothing
2760  * \retval              negative error number on failure
2761  */
2762 static int __lfsck_layout_repair_dangling(const struct lu_env *env,
2763                                           struct lfsck_component *com,
2764                                           struct dt_object *parent,
2765                                           struct dt_object *child,
2766                                           __u32 ea_off, __u32 ost_idx, bool log)
2767 {
2768         struct lfsck_thread_info *info = lfsck_env_info(env);
2769         struct filter_fid *ff = &info->lti_new_pfid;
2770         struct dt_object_format *dof = &info->lti_dof;
2771         struct lu_attr *la = &info->lti_la;
2772         struct lfsck_instance *lfsck = com->lc_lfsck;
2773         struct dt_device *dev = lfsck_obj2dev(child);
2774         const struct lu_fid *pfid = lfsck_dto2fid(parent);
2775         const struct lu_fid *cfid = lfsck_dto2fid(child);
2776         struct thandle *handle;
2777         struct lu_buf *buf;
2778         struct lustre_handle lh = { 0 };
2779         int rc;
2780         ENTRY;
2781
2782         if (!(lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ))
2783                 GOTO(log, rc = 1);
2784
2785         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2786                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2787                               LCK_EX);
2788         if (rc != 0)
2789                 GOTO(log, rc);
2790
2791         rc = dt_attr_get(env, parent, la);
2792         if (rc != 0)
2793                 GOTO(unlock1, rc);
2794
2795         la->la_mode = S_IFREG | 0666;
2796         la->la_atime = la->la_mtime = la->la_ctime = 0;
2797         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
2798                        LA_ATIME | LA_MTIME | LA_CTIME;
2799         memset(dof, 0, sizeof(*dof));
2800         ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
2801         ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
2802         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2803          * MDT-object's FID::f_ver, instead it is the OST-object index in its
2804          * parent MDT-object's layout EA. */
2805         ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
2806         buf = lfsck_buf_get(env, ff, sizeof(struct filter_fid));
2807
2808         handle = dt_trans_create(env, dev);
2809         if (IS_ERR(handle))
2810                 GOTO(unlock1, rc = PTR_ERR(handle));
2811
2812         rc = dt_declare_create(env, child, la, NULL, dof, handle);
2813         if (rc != 0)
2814                 GOTO(stop, rc);
2815
2816         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
2817                                   LU_XATTR_CREATE, handle);
2818         if (rc != 0)
2819                 GOTO(stop, rc);
2820
2821         rc = dt_trans_start_local(env, dev, handle);
2822         if (rc != 0)
2823                 GOTO(stop, rc);
2824
2825         dt_read_lock(env, parent, 0);
2826         if (unlikely(lfsck_is_dead_obj(parent)))
2827                 GOTO(unlock2, rc = 0);
2828
2829         if (lfsck->li_bookmark_ram.lb_param & LPF_DELAY_CREATE_OSTOBJ) {
2830                 struct ost_id *oi = &info->lti_oi;
2831                 struct lu_fid *tfid = &info->lti_fid2;
2832                 struct lu_buf *lovea = &info->lti_big_buf;
2833                 struct lov_mds_md_v1 *lmm;
2834                 struct lov_ost_data_v1 *objs;
2835                 __u32 magic;
2836                 int count;
2837                 int idx2;
2838
2839                 rc = lfsck_layout_get_lovea(env, parent, lovea);
2840                 if (rc <= 0)
2841                         GOTO(unlock2, rc);
2842
2843                 lmm = lovea->lb_buf;
2844                 rc = lfsck_layout_verify_header(lmm);
2845                 if (unlikely(rc != 0))
2846                         GOTO(unlock2, rc);
2847
2848                 count = le16_to_cpu(lmm->lmm_stripe_count);
2849                 /* Someone changed the LOV EA, do nothing. */
2850                 if (count <= ea_off)
2851                         GOTO(unlock2, rc = 0);
2852
2853                 /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which
2854                  * has been verified in lfsck_layout_verify_header() already.
2855                  * If some new magic introduced in the future, then the layout
2856                  * LFSCK needs to be updated also. */
2857                 magic = le32_to_cpu(lmm->lmm_magic);
2858                 if (magic == LOV_MAGIC_V1) {
2859                         objs = &lmm->lmm_objects[ea_off];
2860                 } else {
2861                         LASSERT(magic == LOV_MAGIC_V3);
2862
2863                         objs = &((struct lov_mds_md_v3 *)lmm)->\
2864                                                         lmm_objects[ea_off];
2865                 }
2866
2867                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2868                 idx2 = le32_to_cpu(objs->l_ost_idx);
2869                 rc = ostid_to_fid(tfid, oi, idx2);
2870                 /* Someone changed the LOV EA, do nothing. */
2871                 if (rc != 0 || !lu_fid_eq(tfid, cfid))
2872                         GOTO(unlock2, rc);
2873         }
2874
2875         rc = dt_create(env, child, la, NULL, dof, handle);
2876         if (rc != 0)
2877                 GOTO(unlock2, rc);
2878
2879         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
2880                           handle);
2881
2882         GOTO(unlock2, rc);
2883
2884 unlock2:
2885         dt_read_unlock(env, parent);
2886
2887 stop:
2888         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2889
2890 unlock1:
2891         lfsck_ibits_unlock(&lh, LCK_EX);
2892
2893 log:
2894         if (rc != 0 && log)
2895                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
2896                        "dangling reference for: parent "DFID", child "
2897                        DFID", ea_off %u, ost_idx %u, %s: rc = %d\n",
2898                        lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2899                        ea_off, ost_idx,
2900                        (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
2901                                 "Create the lost OST-object as required" :
2902                                 "Keep the MDT-object there by default", rc);
2903
2904         return rc;
2905 }
2906
2907 /**
2908  * Repair the MDT-object with dangling LOV EA reference.
2909  *
2910  * Prepare parameters and call __lfsck_layout_repair_dangling()
2911  * to repair the dangling LOV EA reference.
2912  *
2913  * \param[in] env       pointer to the thread context
2914  * \param[in] com       the layout LFSCK component
2915  * \param[in] pfid      the MDT-object's FID
2916  * \param[in] cfid      the FID for the OST-object to be created
2917  * \param[in] ea_off    the offset of the OST-object in the LOV EA
2918  * \param[in] ost_idx   the index of OST on which the OST-object resides
2919  *
2920  * \retval              +1 for repair successfully
2921  * \retval              0 for did nothing
2922  * \retval              negative error number on failure
2923  */
2924 static int lfsck_layout_repair_dangling(const struct lu_env *env,
2925                                         struct lfsck_component *com,
2926                                         const struct lu_fid *pfid,
2927                                         const struct lu_fid *cfid,
2928                                         __u32 ea_off, __u32 ost_idx)
2929 {
2930         struct lfsck_instance *lfsck = com->lc_lfsck;
2931         struct dt_object *parent = NULL;
2932         struct dt_object *child = NULL;
2933         struct lfsck_tgt_desc *ltd;
2934         int rc;
2935         ENTRY;
2936
2937         parent = lfsck_object_find_bottom(env, lfsck, pfid);
2938         if (IS_ERR(parent))
2939                 GOTO(log, rc = PTR_ERR(parent));
2940
2941         /* The MDT-object has been removed. */
2942         if (dt_object_exists(parent) == 0)
2943                 GOTO(log, rc = 0);
2944
2945         ltd = lfsck_ltd2tgt(&lfsck->li_ost_descs, ost_idx);
2946         if (unlikely(ltd == NULL))
2947                 GOTO(log, rc = -ENODEV);
2948
2949         child = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
2950         if (IS_ERR(child))
2951                 GOTO(log, rc = PTR_ERR(child));
2952
2953         /* The OST-object has been created. */
2954         if (unlikely(dt_object_exists(child) != 0))
2955                 GOTO(log, rc = 0);
2956
2957         rc = __lfsck_layout_repair_dangling(env, com, parent, child,
2958                                             ea_off, ost_idx, false);
2959
2960         GOTO(log, rc);
2961
2962 log:
2963         if (child != NULL && !IS_ERR(child))
2964                 lfsck_object_put(env, child);
2965
2966         if (parent != NULL && !IS_ERR(parent))
2967                 lfsck_object_put(env, parent);
2968
2969         if (rc != 0)
2970                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
2971                        "dangling reference for: parent "DFID", child "
2972                        DFID", ea_off %u, ost_idx %u, %s: rc = %d\n",
2973                        lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2974                        ea_off, ost_idx,
2975                        (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
2976                                 "Create the lost OST-object as required" :
2977                                 "Keep the MDT-object there by default", rc);
2978
2979         return rc;
2980 }
2981
2982 /* If the OST-object does not recognize the MDT-object as its parent, and
2983  * there is no other MDT-object claims as its parent, then just trust the
2984  * given MDT-object as its parent. So update the OST-object filter_fid. */
2985 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
2986                                               struct lfsck_component *com,
2987                                               struct dt_object *parent,
2988                                               struct lfsck_layout_req *llr,
2989                                               struct lu_attr *la)
2990 {
2991         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2992         struct filter_fid               *pfid   = &info->lti_new_pfid;
2993         struct dt_object                *child  = llr->llr_child;
2994         struct dt_device                *dev    = lfsck_obj2dev(child);
2995         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
2996         struct thandle                  *handle;
2997         struct lu_buf                   *buf;
2998         struct lustre_handle             lh     = { 0 };
2999         int                              rc;
3000         ENTRY;
3001
3002         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
3003                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3004                               LCK_EX);
3005         if (rc != 0)
3006                 GOTO(log, rc);
3007
3008         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
3009         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
3010         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3011          * MDT-object's FID::f_ver, instead it is the OST-object index in its
3012          * parent MDT-object's layout EA. */
3013         pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
3014         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
3015
3016         handle = dt_trans_create(env, dev);
3017         if (IS_ERR(handle))
3018                 GOTO(unlock1, rc = PTR_ERR(handle));
3019
3020         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3021         if (rc != 0)
3022                 GOTO(stop, rc);
3023
3024         rc = dt_attr_get(env, parent, la);
3025         if (rc != 0)
3026                 GOTO(stop, rc);
3027
3028         la->la_valid = LA_UID | LA_GID;
3029         rc = dt_declare_attr_set(env, child, la, handle);
3030         if (rc != 0)
3031                 GOTO(stop, rc);
3032
3033         rc = dt_trans_start_local(env, dev, handle);
3034         if (rc != 0)
3035                 GOTO(stop, rc);
3036
3037         dt_write_lock(env, parent, 0);
3038         if (unlikely(lfsck_is_dead_obj(parent)))
3039                 GOTO(unlock2, rc = 1);
3040
3041         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3042         if (rc != 0)
3043                 GOTO(unlock2, rc);
3044
3045         /* Get the latest parent's owner. */
3046         rc = dt_attr_get(env, parent, la);
3047         if (rc != 0)
3048                 GOTO(unlock2, rc);
3049
3050         la->la_valid = LA_UID | LA_GID;
3051         rc = dt_attr_set(env, child, la, handle);
3052
3053         GOTO(unlock2, rc);
3054
3055 unlock2:
3056         dt_write_unlock(env, parent);
3057
3058 stop:
3059         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3060
3061 unlock1:
3062         lfsck_ibits_unlock(&lh, LCK_EX);
3063
3064 log:
3065         if (rc != 0)
3066                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3067                        "unmatched MDT-OST pair for: parent "DFID
3068                        ", child "DFID", OST-index %u, stripe-index %u, "
3069                        "owner %u/%u: rc = %d\n",
3070                        lfsck_lfsck2name(com->lc_lfsck),
3071                        PFID(lfsck_dto2fid(parent)),
3072                        PFID(lfsck_dto2fid(child)),
3073                        llr->llr_ost_idx, llr->llr_lov_idx,
3074                        la->la_uid, la->la_gid, rc);
3075
3076         return rc;
3077 }
3078
3079 /* If there are more than one MDT-objects claim as the OST-object's parent,
3080  * and the OST-object only recognizes one of them, then we need to generate
3081  * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
3082 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
3083                                                    struct lfsck_component *com,
3084                                                    struct dt_object *parent,
3085                                                    struct lfsck_layout_req *llr,
3086                                                    struct lu_attr *la,
3087                                                    struct lu_buf *buf)
3088 {
3089         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3090         struct dt_allocation_hint       *hint   = &info->lti_hint;
3091         struct dt_object_format         *dof    = &info->lti_dof;
3092         struct ost_id                   *oi     = &info->lti_oi;
3093         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3094         struct dt_device                *dev;
3095         struct lu_device                *d      =
3096                                 &lfsck_obj2dev(llr->llr_child)->dd_lu_dev;
3097         struct lu_object                *o;
3098         struct lu_object                *n;
3099         struct dt_object                *child  = NULL;
3100         struct thandle                  *handle = NULL;
3101         struct lov_mds_md_v1            *lmm;
3102         struct lov_ost_data_v1          *objs;
3103         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
3104         struct lu_fid                    tfid;
3105         struct lustre_handle             lh     = { 0 };
3106         struct lu_buf                    ea_buf;
3107         __u32                            magic;
3108         __u32                            index;
3109         int                              rc;
3110         ENTRY;
3111
3112         /* We use two separated transactions to repair the inconsistency.
3113          *
3114          * 1) create the child (OST-object).
3115          * 2) update the parent LOV EA according to the child's FID.
3116          *
3117          * If 1) succeed, but 2) failed or aborted, then such OST-object will be
3118          * handled as orphan when the layout LFSCK run next time.
3119          *
3120          * If 1) failed, but 2) succeed, then such OST-object will be re-created
3121          * as dangling referened case when the layout LFSCK run next time. */
3122
3123         /* The 1st transaction. */
3124         o = lu_object_anon(env, d, NULL);
3125         if (IS_ERR(o))
3126                 GOTO(log, rc = PTR_ERR(o));
3127
3128         n = lu_object_locate(o->lo_header, d->ld_type);
3129         if (unlikely(n == NULL)) {
3130                 lu_object_put_nocache(env, o);
3131
3132                 GOTO(log, rc = -EINVAL);
3133         }
3134
3135         child = container_of(n, struct dt_object, do_lu);
3136         memset(hint, 0, sizeof(*hint));
3137         rc = dt_attr_get(env, parent, la);
3138         if (rc != 0)
3139                 GOTO(log, rc);
3140
3141         la->la_valid = LA_UID | LA_GID;
3142         memset(dof, 0, sizeof(*dof));
3143
3144         dev = lfsck_obj2dev(child);
3145         handle = dt_trans_create(env, dev);
3146         if (IS_ERR(handle))
3147                 GOTO(log, rc = PTR_ERR(handle));
3148
3149         rc = dt_declare_create(env, child, la, hint, dof, handle);
3150         if (rc != 0)
3151                 GOTO(stop, rc);
3152
3153         rc = dt_trans_start_local(env, dev, handle);
3154         if (rc != 0)
3155                 GOTO(stop, rc);
3156
3157         rc = dt_create(env, child, la, hint, dof, handle);
3158         dt_trans_stop(env, dev, handle);
3159         handle = NULL;
3160         if (rc != 0)
3161                 GOTO(log, rc);
3162
3163         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3164                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3165                               LCK_EX);
3166         if (rc != 0)
3167                 GOTO(log, rc);
3168
3169         /* The 2nd transaction. */
3170
3171         /* XXX: Generally, we should use bottom device (OSD) to update parent
3172          *      LOV EA. But because the LOD-object still references the wrong
3173          *      OSP-object that should be detached after the parent's LOV EA
3174          *      refreshed. Unfortunately, there is no suitable API for that.
3175          *      So we have to make the LOD to re-load the OSP-object(s) via
3176          *      replacing the LOV EA against the LOD-object.
3177          *
3178          *      Once the DNE2 patches have been landed, we can replace the
3179          *      LOD device with the OSD device. LU-6230. */
3180
3181         dev = lfsck->li_next;
3182         parent = lfsck_object_locate(dev, parent);
3183         if (IS_ERR(parent))
3184                 GOTO(log, rc = PTR_ERR(parent));
3185
3186         handle = dt_trans_create(env, dev);
3187         if (IS_ERR(handle))
3188                 GOTO(log, rc = PTR_ERR(handle));
3189
3190         rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3191                                   LU_XATTR_REPLACE, handle);
3192         if (rc != 0)
3193                 GOTO(stop, rc);
3194
3195         rc = dt_trans_start_local(env, dev, handle);
3196         if (rc != 0)
3197                 GOTO(stop, rc);
3198
3199         dt_write_lock(env, parent, 0);
3200         if (unlikely(lfsck_is_dead_obj(parent)))
3201                 GOTO(unlock, rc = 0);
3202
3203         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
3204         if (unlikely(rc == 0 || rc == -ENODATA || rc == -ERANGE))
3205                 GOTO(unlock, rc = 0);
3206
3207         lmm = buf->lb_buf;
3208         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
3209          * been verified in lfsck_layout_verify_header() already. If some
3210          * new magic introduced in the future, then layout LFSCK needs to
3211          * be updated also. */
3212         magic = le32_to_cpu(lmm->lmm_magic);
3213         if (magic == LOV_MAGIC_V1) {
3214                 objs = &lmm->lmm_objects[llr->llr_lov_idx];
3215         } else {
3216                 LASSERT(magic == LOV_MAGIC_V3);
3217                 objs =
3218                 &((struct lov_mds_md_v3 *)lmm)->lmm_objects[llr->llr_lov_idx];
3219         }
3220
3221         ostid_le_to_cpu(&objs->l_ost_oi, oi);
3222         index = le32_to_cpu(objs->l_ost_idx);
3223         rc = ostid_to_fid(&tfid, oi, index);
3224         /* Someone changed layout during the LFSCK, no need to repair then. */
3225         if (rc == 0 && !lu_fid_eq(&tfid, lu_object_fid(&llr->llr_child->do_lu)))
3226                 GOTO(unlock, rc = 0);
3227
3228         lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
3229         fid_to_ostid(lu_object_fid(&child->do_lu), oi);
3230         ostid_cpu_to_le(oi, &objs->l_ost_oi);
3231         objs->l_ost_gen = cpu_to_le32(0);
3232         objs->l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
3233         lfsck_buf_init(&ea_buf, lmm,
3234                        lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3235                                        magic));
3236         rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV,
3237                           LU_XATTR_REPLACE, handle);
3238
3239         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
3240
3241 unlock:
3242         dt_write_unlock(env, parent);
3243
3244 stop:
3245         if (handle != NULL)
3246                 dt_trans_stop(env, dev, handle);
3247
3248 log:
3249         lfsck_ibits_unlock(&lh, LCK_EX);
3250         if (child != NULL)
3251                 lfsck_object_put(env, child);
3252
3253         if (rc != 0)
3254                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3255                        "multiple references for: parent "DFID", OST-index %u, "
3256                        "stripe-index %u, owner %u/%u: rc = %d\n",
3257                        lfsck_lfsck2name(lfsck), PFID(pfid), llr->llr_ost_idx,
3258                        llr->llr_lov_idx, la->la_uid, la->la_gid, rc);
3259
3260         return rc;
3261 }
3262
3263 /* If the MDT-object and the OST-object have different owner information,
3264  * then trust the MDT-object, because the normal chown/chgrp handle order
3265  * is from MDT to OST, and it is possible that some chown/chgrp operation
3266  * is partly done. */
3267 static int lfsck_layout_repair_owner(const struct lu_env *env,
3268                                      struct lfsck_component *com,
3269                                      struct dt_object *parent,
3270                                      struct lfsck_layout_req *llr,
3271                                      struct lu_attr *pla,
3272                                      const struct lu_attr *cla)
3273 {
3274         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3275         struct lu_attr                  *tla    = &info->lti_la2;
3276         struct dt_object                *child  = llr->llr_child;
3277         struct dt_device                *dev    = lfsck_obj2dev(child);
3278         struct thandle                  *handle;
3279         int                              rc;
3280         ENTRY;
3281
3282         tla->la_uid = pla->la_uid;
3283         tla->la_gid = pla->la_gid;
3284         tla->la_valid = LA_UID | LA_GID;
3285         handle = dt_trans_create(env, dev);
3286         if (IS_ERR(handle))
3287                 GOTO(log, rc = PTR_ERR(handle));
3288
3289         rc = dt_declare_attr_set(env, child, tla, handle);
3290         if (rc != 0)
3291                 GOTO(stop, rc);
3292
3293         rc = dt_trans_start_local(env, dev, handle);
3294         if (rc != 0)
3295                 GOTO(stop, rc);
3296
3297         /* Use the dt_object lock to serialize with destroy and attr_set. */
3298         dt_read_lock(env, parent, 0);
3299         if (unlikely(lfsck_is_dead_obj(parent)))
3300                 GOTO(unlock, rc = 1);
3301
3302         /* Get the latest parent's owner. */
3303         rc = dt_attr_get(env, parent, pla);
3304         if (rc != 0)
3305                 GOTO(unlock, rc);
3306
3307         /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
3308         if (unlikely(tla->la_uid != pla->la_uid ||
3309                      tla->la_gid != pla->la_gid))
3310                 rc = 1;
3311         else
3312                 rc = dt_attr_set(env, child, tla, handle);
3313
3314         GOTO(unlock, rc);
3315
3316 unlock:
3317         dt_read_unlock(env, parent);
3318
3319 stop:
3320         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3321
3322 log:
3323         if (rc != 0)
3324                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3325                        "inconsistent file owner for: parent "DFID", child "DFID
3326                        ", OST-index %u, stripe-index %u, old owner %u/%u, "
3327                        "new owner %u/%u: rc = %d\n",
3328                        lfsck_lfsck2name(com->lc_lfsck),
3329                        PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
3330                        llr->llr_ost_idx, llr->llr_lov_idx,
3331                        cla->la_uid, cla->la_gid, tla->la_uid, tla->la_gid, rc);
3332
3333         return rc;
3334 }
3335
3336 /* Check whether the OST-object correctly back points to the
3337  * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
3338 static int lfsck_layout_check_parent(const struct lu_env *env,
3339                                      struct lfsck_component *com,
3340                                      struct lfsck_assistant_object *lso,
3341                                      const struct lu_fid *pfid,
3342                                      const struct lu_fid *cfid,
3343                                      const struct lu_attr *cla,
3344                                      struct lfsck_layout_req *llr,
3345                                      struct lu_buf *lov_ea, __u32 idx)
3346 {
3347         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3348         struct lu_buf                   *buf    = &info->lti_big_buf;
3349         struct dt_object                *tobj;
3350         struct lov_mds_md_v1            *lmm;
3351         struct lov_ost_data_v1          *objs;
3352         struct lustre_handle             lh     = { 0 };
3353         int                              rc;
3354         int                              i;
3355         __u32                            magic;
3356         __u16                            count;
3357         ENTRY;
3358
3359         if (unlikely(!fid_is_sane(pfid)))
3360                 RETURN(LLIT_UNMATCHED_PAIR);
3361
3362         if (lu_fid_eq(pfid, &lso->lso_fid)) {
3363                 if (likely(llr->llr_lov_idx == idx))
3364                         RETURN(0);
3365
3366                 RETURN(LLIT_UNMATCHED_PAIR);
3367         }
3368
3369         tobj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3370         if (IS_ERR(tobj))
3371                 RETURN(PTR_ERR(tobj));
3372
3373         if (dt_object_exists(tobj) == 0 || lfsck_is_dead_obj(tobj) ||
3374             !S_ISREG(lfsck_object_type(tobj)))
3375                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3376
3377         /* Load the tobj's layout EA, in spite of it is a local MDT-object or
3378          * remote one on another MDT. Then check whether the given OST-object
3379          * is in such layout. If yes, it is multiple referenced, otherwise it
3380          * is unmatched referenced case. */
3381         rc = lfsck_layout_get_lovea(env, tobj, buf);
3382         if (rc == 0 || rc == -ENOENT)
3383                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3384
3385         if (rc < 0)
3386                 GOTO(out, rc);
3387
3388         lmm = buf->lb_buf;
3389         magic = le32_to_cpu(lmm->lmm_magic);
3390         if (magic == LOV_MAGIC_V1) {
3391                 objs = &lmm->lmm_objects[0];
3392         } else {
3393                 LASSERT(magic == LOV_MAGIC_V3);
3394                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3395         }
3396
3397         count = le16_to_cpu(lmm->lmm_stripe_count);
3398         for (i = 0; i < count; i++, objs++) {
3399                 struct lu_fid           *tfid   = &info->lti_fid2;
3400                 struct ost_id           *oi     = &info->lti_oi;
3401                 __u32                    idx2;
3402
3403                 if (lovea_slot_is_dummy(objs))
3404                         continue;
3405
3406                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3407                 idx2 = le32_to_cpu(objs->l_ost_idx);
3408                 rc = ostid_to_fid(tfid, oi, idx2);
3409                 if (rc != 0) {
3410                         CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
3411                                "invalid layout EA at the slot %d, index %u\n",
3412                                lfsck_lfsck2name(com->lc_lfsck),
3413                                PFID(pfid), i, idx2);
3414
3415                         GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3416                 }
3417
3418                 if (lu_fid_eq(cfid, tfid)) {
3419                         rc = lfsck_ibits_lock(env, com->lc_lfsck, tobj, &lh,
3420                                               MDS_INODELOCK_UPDATE |
3421                                               MDS_INODELOCK_LAYOUT |
3422                                               MDS_INODELOCK_XATTR,
3423                                               LCK_EX);
3424                         if (rc != 0)
3425                                 GOTO(out, rc);
3426
3427                         dt_read_lock(env, tobj, 0);
3428
3429                         /* For local MDT-object, re-check existence
3430                          * after taken the lock. */
3431                         if (!dt_object_remote(tobj)) {
3432                                 if (dt_object_exists(tobj) == 0 ||
3433                                     lfsck_is_dead_obj(tobj)) {
3434                                         rc = LLIT_UNMATCHED_PAIR;
3435                                 } else {
3436                                         *lov_ea = *buf;
3437                                         rc = LLIT_MULTIPLE_REFERENCED;
3438                                 }
3439
3440                                 GOTO(unlock, rc);
3441                         }
3442
3443                         /* For migration case, the new MDT-object and old
3444                          * MDT-object may reference the same OST-object at
3445                          * some migration internal time.
3446                          *
3447                          * For remote MDT-object, the local MDT may not know
3448                          * whether it has been removed or not.  Try checking
3449                          * for a non-existent xattr to check if this object
3450                          * has been been removed or not. */
3451                         rc = dt_xattr_get(env, tobj, &LU_BUF_NULL,
3452                                           XATTR_NAME_DUMMY);
3453                         if (unlikely(rc == -ENOENT || rc >= 0)) {
3454                                 rc = LLIT_UNMATCHED_PAIR;
3455                         } else if (rc == -ENODATA) {
3456                                 *lov_ea = *buf;
3457                                 rc = LLIT_MULTIPLE_REFERENCED;
3458                         }
3459
3460                         GOTO(unlock, rc);
3461                 }
3462         }
3463
3464         GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3465
3466 unlock:
3467         if (lustre_handle_is_used(&lh)) {
3468                 dt_read_unlock(env, tobj);
3469                 lfsck_ibits_unlock(&lh, LCK_EX);
3470         }
3471
3472 out:
3473         lfsck_object_put(env, tobj);
3474
3475         return rc;
3476 }
3477
3478 static int lfsck_layout_assistant_handler_p1(const struct lu_env *env,
3479                                              struct lfsck_component *com,
3480                                              struct lfsck_assistant_req *lar)
3481 {
3482         struct lfsck_layout_req              *llr    =
3483                         container_of0(lar, struct lfsck_layout_req, llr_lar);
3484         struct lfsck_assistant_object        *lso    = lar->lar_parent;
3485         struct lfsck_layout                  *lo     = com->lc_file_ram;
3486         struct lfsck_thread_info             *info   = lfsck_env_info(env);
3487         struct filter_fid_old                *pea    = &info->lti_old_pfid;
3488         struct lu_fid                        *pfid   = &info->lti_fid;
3489         struct lu_buf                         buf    = { NULL };
3490         struct dt_object                     *parent = NULL;
3491         struct dt_object                     *child  = llr->llr_child;
3492         struct lu_attr                       *pla    = &lso->lso_attr;
3493         struct lu_attr                       *cla    = &info->lti_la;
3494         struct lfsck_instance                *lfsck  = com->lc_lfsck;
3495         struct lfsck_bookmark                *bk     = &lfsck->li_bookmark_ram;
3496         enum lfsck_layout_inconsistency_type  type   = LLIT_NONE;
3497         __u32                                 idx    = 0;
3498         int                                   rc;
3499         ENTRY;
3500
3501         if (lso->lso_dead)
3502                 RETURN(0);
3503
3504         CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_ASSISTANT_DIRECT, cfs_fail_val);
3505
3506         rc = dt_attr_get(env, child, cla);
3507         if (rc == -ENOENT) {
3508                 parent = lfsck_assistant_object_load(env, lfsck, lso);
3509                 if (IS_ERR(parent)) {
3510                         rc = PTR_ERR(parent);
3511
3512                         RETURN(rc == -ENOENT ? 0 : rc);
3513                 }
3514
3515                 type = LLIT_DANGLING;
3516                 goto repair;
3517         }
3518
3519         if (rc != 0)
3520                 GOTO(out, rc);
3521
3522         lfsck_buf_init(&buf, pea, sizeof(struct filter_fid_old));
3523         rc = dt_xattr_get(env, child, &buf, XATTR_NAME_FID);
3524         if (unlikely(rc > 0 && rc != sizeof(struct filter_fid_old) &&
3525                      rc != sizeof(struct filter_fid))) {
3526                 type = LLIT_UNMATCHED_PAIR;
3527                 goto repair;
3528         }
3529
3530         if (rc < 0 && rc != -ENODATA)
3531                 GOTO(out, rc);
3532
3533         if (rc == 0 || rc == -ENODATA)
3534                 GOTO(check_owner, rc = 0);
3535
3536         fid_le_to_cpu(pfid, &pea->ff_parent);
3537         /* Currently, the filter_fid::ff_parent::f_ver is not the
3538          * real parent MDT-object's FID::f_ver, instead it is the
3539          * OST-object index in its parent MDT-object's layout EA. */
3540         idx = pfid->f_stripe_idx;
3541         pfid->f_ver = 0;
3542         rc = lfsck_layout_check_parent(env, com, lso, pfid,
3543                                        lu_object_fid(&child->do_lu),
3544                                        cla, llr, &buf, idx);
3545         if (rc > 0) {
3546                 type = rc;
3547                 goto repair;
3548         }
3549
3550         if (rc < 0)
3551                 GOTO(out, rc);
3552
3553 check_owner:
3554         /* Someone may has changed the owner after the parent attr pre-loaded.
3555          * It can be handled later inside the lfsck_layout_repair_owner(). */
3556         if (unlikely(cla->la_uid != pla->la_uid ||
3557                      cla->la_gid != pla->la_gid)) {
3558                 type = LLIT_INCONSISTENT_OWNER;
3559                 goto repair;
3560         }
3561
3562 repair:
3563         if (type == LLIT_NONE)
3564                 GOTO(out, rc = 0);
3565
3566         if (bk->lb_param & LPF_DRYRUN)
3567                 GOTO(out, rc = 1);
3568
3569         if (parent == NULL) {
3570                 parent = lfsck_assistant_object_load(env, lfsck, lso);
3571                 if (IS_ERR(parent)) {
3572                         rc = PTR_ERR(parent);
3573
3574                         if (rc == -ENOENT)
3575                                 RETURN(0);
3576
3577                         GOTO(out, rc);
3578                 }
3579         }
3580
3581         switch (type) {
3582         case LLIT_DANGLING:
3583                 if (bk->lb_param & LPF_DELAY_CREATE_OSTOBJ)
3584                         rc = lfsck_layout_ins_dangling_rec(env, com,
3585                                 lfsck_dto2fid(parent), lfsck_dto2fid(child),
3586                                 llr->llr_lov_idx, llr->llr_ost_idx);
3587                 else
3588                         rc = __lfsck_layout_repair_dangling(env, com, parent,
3589                                         llr->llr_child, llr->llr_lov_idx,
3590                                         llr->llr_ost_idx, true);
3591                 break;
3592         case LLIT_UNMATCHED_PAIR:
3593                 rc = lfsck_layout_repair_unmatched_pair(env, com, parent,
3594                                                         llr, pla);
3595                 break;
3596         case LLIT_MULTIPLE_REFERENCED:
3597                 rc = lfsck_layout_repair_multiple_references(env, com, parent,
3598                                                              llr, pla, &buf);
3599                 break;
3600         case LLIT_INCONSISTENT_OWNER:
3601                 rc = lfsck_layout_repair_owner(env, com, parent, llr, pla, cla);
3602                 break;
3603         default:
3604                 rc = 0;
3605                 break;
3606         }
3607
3608         GOTO(out, rc);
3609
3610 out:
3611         down_write(&com->lc_sem);
3612         if (rc < 0) {
3613                 struct lfsck_assistant_data *lad = com->lc_data;
3614
3615                 if (unlikely(lad->lad_exit)) {
3616                         rc = 0;
3617                 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
3618                            rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
3619                            rc == -EHOSTUNREACH) {
3620                         /* If cannot touch the target server,
3621                          * mark the LFSCK as INCOMPLETE. */
3622                         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
3623                                "talk with OST %x: rc = %d\n",
3624                                lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
3625                         lfsck_lad_set_bitmap(env, com, llr->llr_ost_idx);
3626                         lo->ll_objs_skipped++;
3627                         rc = 0;
3628                 } else {
3629                         lfsck_layout_record_failure(env, lfsck, lo);
3630                 }
3631         } else if (rc > 0 && (type != LLIT_DANGLING ||
3632                               !(bk->lb_param & LPF_DELAY_CREATE_OSTOBJ))) {
3633                 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
3634                          "unknown type = %d\n", type);
3635
3636                 lo->ll_objs_repaired[type - 1]++;
3637                 if (bk->lb_param & LPF_DRYRUN &&
3638                     unlikely(lo->ll_pos_first_inconsistent == 0))
3639                         lo->ll_pos_first_inconsistent =
3640                         lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
3641                                                         lfsck->li_di_oit);
3642         }
3643         up_write(&com->lc_sem);
3644
3645         if (parent != NULL && !IS_ERR(parent))
3646                 lfsck_object_put(env, parent);
3647
3648         return rc;
3649 }
3650
3651 static int
3652 lfsck_layout_double_scan_one_trace_file(const struct lu_env *env,
3653                                         struct lfsck_component *com,
3654                                         struct dt_object *obj, bool first)
3655 {
3656         struct lfsck_instance *lfsck = com->lc_lfsck;
3657         struct ptlrpc_thread *thread = &lfsck->li_thread;
3658         struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
3659         struct lfsck_layout *lo = com->lc_file_ram;
3660         const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
3661         struct dt_it *di;
3662         struct dt_key *key;
3663         struct lu_fid *pfid = &lfsck_env_info(env)->lti_fid3;
3664         struct lu_fid *cfid = &lfsck_env_info(env)->lti_fid4;
3665         __u32 ea_off;
3666         __u32 ost_idx;
3667         int rc;
3668         ENTRY;
3669
3670         di = iops->init(env, obj, 0);
3671         if (IS_ERR(di))
3672                 RETURN(PTR_ERR(di));
3673
3674         if (first)
3675                 fid_cpu_to_be(pfid, &lo->ll_fid_latest_scanned_phase2);
3676         else
3677                 fid_zero(pfid);
3678         rc = iops->get(env, di, (const struct dt_key *)pfid);
3679         if (rc < 0)
3680                 GOTO(fini, rc);
3681
3682         if (first) {
3683                 /* The start one either has been processed or does not exist,
3684                  * skip it. */
3685                 rc = iops->next(env, di);
3686                 if (rc != 0)
3687                         GOTO(put, rc);
3688         }
3689
3690         do {
3691                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
3692                     unlikely(!thread_is_running(thread)))
3693                         GOTO(put, rc = 0);
3694
3695                 key = iops->key(env, di);
3696                 if (IS_ERR(key)) {
3697                         rc = PTR_ERR(key);
3698                         if (rc == -ENOENT)
3699                                 GOTO(put, rc = 1);
3700
3701                         goto checkpoint;
3702                 }
3703
3704                 fid_be_to_cpu(pfid, (const struct lu_fid *)key);
3705                 ea_off = pfid->f_ver;
3706                 pfid->f_ver = 0;
3707                 if (!fid_is_sane(pfid)) {
3708                         rc = 0;
3709                         goto checkpoint;
3710                 }
3711
3712                 rc = iops->rec(env, di, (struct dt_rec *)cfid, 0);
3713                 if (rc == 0) {
3714                         fid_be_to_cpu(cfid, cfid);
3715                         ost_idx = cfid->f_ver;
3716                         cfid->f_ver = 0;
3717                         if (!fid_is_sane(cfid)) {
3718                                 rc = 0;
3719                                 goto checkpoint;
3720                         }
3721
3722                         rc = lfsck_layout_repair_dangling(env, com, pfid, cfid,
3723                                                           ea_off, ost_idx);
3724                 }
3725
3726 checkpoint:
3727                 down_write(&com->lc_sem);
3728                 com->lc_new_checked++;
3729                 com->lc_new_scanned++;
3730                 if (rc >= 0)
3731                         lo->ll_fid_latest_scanned_phase2 = *pfid;
3732
3733                 if (rc > 0)
3734                         lo->ll_objs_repaired[LLIT_DANGLING - 1]++;
3735                 else if (rc < 0)
3736                         lo->ll_objs_failed_phase2++;
3737                 up_write(&com->lc_sem);
3738
3739                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
3740                         GOTO(put, rc);
3741
3742                 if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
3743                                               cfs_time_current())) &&
3744                     com->lc_new_checked != 0) {
3745                         down_write(&com->lc_sem);
3746                         lo->ll_run_time_phase2 +=
3747                                 cfs_duration_sec(cfs_time_current() +
3748                                 HALF_SEC - com->lc_time_last_checkpoint);
3749                         lo->ll_time_last_checkpoint = cfs_time_current_sec();
3750                         lo->ll_objs_checked_phase2 += com->lc_new_checked;
3751                         com->lc_new_checked = 0;
3752                         lfsck_layout_store(env, com);
3753                         up_write(&com->lc_sem);
3754
3755                         com->lc_time_last_checkpoint = cfs_time_current();
3756                         com->lc_time_next_checkpoint =
3757                                 com->lc_time_last_checkpoint +
3758                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
3759                 }
3760
3761                 lfsck_control_speed_by_self(com);
3762                 if (unlikely(!thread_is_running(thread)))
3763                         GOTO(put, rc = 0);
3764
3765                 rc = iops->next(env, di);
3766         } while (rc == 0);
3767
3768         GOTO(put, rc);
3769
3770 put:
3771         iops->put(env, di);
3772
3773 fini:
3774         iops->fini(env, di);
3775
3776         return rc;
3777 }
3778
3779 static int lfsck_layout_assistant_handler_p2(const struct lu_env *env,
3780                                              struct lfsck_component *com)
3781 {
3782         struct lfsck_assistant_data     *lad    = com->lc_data;
3783         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3784         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
3785         struct lfsck_tgt_descs          *ltds   = &lfsck->li_ost_descs;
3786         struct lfsck_tgt_desc           *ltd;
3787         int                              rc     = 0;
3788         ENTRY;
3789
3790         CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan start\n",
3791                lfsck_lfsck2name(lfsck));
3792
3793         spin_lock(&ltds->ltd_lock);
3794         while (!list_empty(&lad->lad_ost_phase2_list)) {
3795                 ltd = list_entry(lad->lad_ost_phase2_list.next,
3796                                  struct lfsck_tgt_desc,
3797                                  ltd_layout_phase_list);
3798                 list_del_init(&ltd->ltd_layout_phase_list);
3799                 if (bk->lb_param & LPF_OST_ORPHAN) {
3800                         spin_unlock(&ltds->ltd_lock);
3801                         rc = lfsck_layout_scan_orphan(env, com, ltd);
3802                         if (rc != 0 && bk->lb_param & LPF_FAILOUT)
3803                                 RETURN(rc);
3804
3805                         if (unlikely(lad->lad_exit ||
3806                                      !thread_is_running(&lfsck->li_thread)))
3807                                 RETURN(0);
3808                         spin_lock(&ltds->ltd_lock);
3809                 }
3810         }
3811
3812         if (list_empty(&lad->lad_ost_phase1_list))
3813                 rc = 1;
3814         else
3815                 rc = 0;
3816         spin_unlock(&ltds->ltd_lock);
3817
3818         if (rc == 1 && bk->lb_param & LPF_OST_ORPHAN) {
3819                 struct lfsck_layout *lo = com->lc_file_ram;
3820                 int i;
3821
3822                 com->lc_new_checked = 0;
3823                 com->lc_new_scanned = 0;
3824                 com->lc_time_last_checkpoint = cfs_time_current();
3825                 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
3826                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
3827
3828                 i = lfsck_sub_trace_file_fid2idx(
3829                                 &lo->ll_fid_latest_scanned_phase2);
3830                 rc = lfsck_layout_double_scan_one_trace_file(env, com,
3831                                 com->lc_sub_trace_objs[i].lsto_obj, true);
3832                 while (rc > 0 && ++i < LFSCK_STF_COUNT)
3833                         rc = lfsck_layout_double_scan_one_trace_file(env, com,
3834                                 com->lc_sub_trace_objs[i].lsto_obj, false);
3835
3836                 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan dangling stop "
3837                        "at the No. %d trace file: rc = %d\n",
3838                        lfsck_lfsck2name(lfsck), i, rc);
3839         }
3840
3841         CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan stop: rc = %d\n",
3842                lfsck_lfsck2name(lfsck), rc);
3843
3844         RETURN(rc);
3845 }
3846
3847 static int
3848 lfsck_layout_slave_async_interpret(const struct lu_env *env,
3849                                    struct ptlrpc_request *req,
3850                                    void *args, int rc)
3851 {
3852         struct lfsck_layout_slave_async_args *llsaa = args;
3853         struct obd_export                    *exp   = llsaa->llsaa_exp;
3854         struct lfsck_component               *com   = llsaa->llsaa_com;
3855         struct lfsck_layout_slave_target     *llst  = llsaa->llsaa_llst;
3856         struct lfsck_layout_slave_data       *llsd  = com->lc_data;
3857         struct lfsck_reply                   *lr    = NULL;
3858         bool                                  done  = false;
3859
3860         if (rc != 0) {
3861                 /* It is probably caused by network trouble, or target crash,
3862                  * it will try several times (depends on the obd_timeout, and
3863                  * will not less than 3 times). But to make the LFSCK can go
3864                  * ahead, we should not try for ever. After some try but still
3865                  * hit failure, it will assume that the target exit the LFSCK
3866                  * prcoessing and stop try. */
3867                 if (rc == -ENOTCONN || rc == -ESHUTDOWN) {
3868                         int max_try = max_t(int, obd_timeout / 30, 3);
3869
3870                         if (++(llst->llst_failures) > max_try)
3871                                 done = true;
3872                 } else {
3873                         done = true;
3874                 }
3875         } else {
3876                 llst->llst_failures = 0;
3877                 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
3878                 if (lr->lr_status != LS_SCANNING_PHASE1 &&
3879                     lr->lr_status != LS_SCANNING_PHASE2)
3880                         done = true;
3881         }
3882
3883         if (done) {
3884                 CDEBUG(D_LFSCK, "%s: layout LFSCK slave gets the MDT %x "
3885                        "status %d, failures_try %d\n", lfsck_lfsck2name(com->lc_lfsck),
3886                        llst->llst_index, lr != NULL ? lr->lr_status : rc,
3887                        llst->llst_failures);
3888
3889                 lfsck_layout_llst_del(llsd, llst);
3890         }
3891
3892         lfsck_layout_llst_put(llst);
3893         lfsck_component_put(env, com);
3894         class_export_put(exp);
3895
3896         return 0;
3897 }
3898
3899 static int lfsck_layout_async_query(const struct lu_env *env,
3900                                     struct lfsck_component *com,
3901                                     struct obd_export *exp,
3902                                     struct lfsck_layout_slave_target *llst,
3903                                     struct lfsck_request *lr,
3904                                     struct ptlrpc_request_set *set)
3905 {
3906         struct lfsck_layout_slave_async_args *llsaa;
3907         struct ptlrpc_request                *req;
3908         struct lfsck_request                 *tmp;
3909         int                                   rc;
3910         ENTRY;
3911
3912         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
3913         if (req == NULL)
3914                 RETURN(-ENOMEM);
3915
3916         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
3917         if (rc != 0) {
3918                 ptlrpc_request_free(req);
3919                 RETURN(rc);
3920         }
3921
3922         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3923         *tmp = *lr;
3924         ptlrpc_request_set_replen(req);
3925
3926         llsaa = ptlrpc_req_async_args(req);
3927         llsaa->llsaa_exp = exp;
3928         llsaa->llsaa_com = lfsck_component_get(com);
3929         llsaa->llsaa_llst = llst;
3930         req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
3931         req->rq_allow_intr = 1;
3932         ptlrpc_set_add_req(set, req);
3933
3934         RETURN(0);
3935 }
3936
3937 static int lfsck_layout_async_notify(const struct lu_env *env,
3938                                      struct obd_export *exp,
3939                                      struct lfsck_request *lr,
3940                                      struct ptlrpc_request_set *set)
3941 {
3942         struct ptlrpc_request   *req;
3943         struct lfsck_request    *tmp;
3944         int                      rc;
3945         ENTRY;
3946
3947         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
3948         if (req == NULL)
3949                 RETURN(-ENOMEM);
3950
3951         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
3952         if (rc != 0) {
3953                 ptlrpc_request_free(req);
3954                 RETURN(rc);
3955         }
3956
3957         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3958         *tmp = *lr;
3959         ptlrpc_request_set_replen(req);
3960         req->rq_allow_intr = 1;
3961         ptlrpc_set_add_req(set, req);
3962
3963         RETURN(0);
3964 }
3965
3966 static int
3967 lfsck_layout_slave_query_master(const struct lu_env *env,
3968                                 struct lfsck_component *com)
3969 {
3970         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
3971         struct lfsck_instance            *lfsck = com->lc_lfsck;
3972         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3973         struct lfsck_layout_slave_target *llst;
3974         struct obd_export                *exp;
3975         struct ptlrpc_request_set        *set;
3976         int                               rc    = 0;
3977         int                               rc1   = 0;
3978         ENTRY;
3979
3980         set = ptlrpc_prep_set();
3981         if (set == NULL)
3982                 GOTO(log, rc = -ENOMEM);
3983
3984         memset(lr, 0, sizeof(*lr));
3985         lr->lr_event = LE_QUERY;
3986         lr->lr_active = LFSCK_TYPE_LAYOUT;
3987
3988         llsd->llsd_touch_gen++;
3989         spin_lock(&llsd->llsd_lock);
3990         while (!list_empty(&llsd->llsd_master_list)) {
3991                 llst = list_entry(llsd->llsd_master_list.next,
3992                                   struct lfsck_layout_slave_target,
3993                                   llst_list);
3994                 if (llst->llst_gen == llsd->llsd_touch_gen)
3995                         break;
3996
3997                 llst->llst_gen = llsd->llsd_touch_gen;
3998                 list_move_tail(&llst->llst_list,
3999                                &llsd->llsd_master_list);
4000                 atomic_inc(&llst->llst_ref);
4001                 spin_unlock(&llsd->llsd_lock);
4002
4003                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4004                                                llst->llst_index);
4005                 if (exp == NULL) {
4006                         lfsck_layout_llst_del(llsd, llst);
4007                         lfsck_layout_llst_put(llst);
4008                         spin_lock(&llsd->llsd_lock);
4009                         continue;
4010                 }
4011
4012                 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
4013                 if (rc != 0) {
4014                         CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4015                                "query %s for layout: rc = %d\n",
4016                                lfsck_lfsck2name(lfsck),
4017                                exp->exp_obd->obd_name, rc);
4018
4019                         rc1 = rc;
4020                         lfsck_layout_llst_put(llst);
4021                         class_export_put(exp);
4022                 }
4023                 spin_lock(&llsd->llsd_lock);
4024         }
4025         spin_unlock(&llsd->llsd_lock);
4026
4027         rc = ptlrpc_set_wait(set);
4028         ptlrpc_set_destroy(set);
4029
4030         GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
4031
4032 log:
4033         CDEBUG(D_LFSCK, "%s: layout LFSCK slave queries master: rc = %d\n",
4034                lfsck_lfsck2name(com->lc_lfsck), rc);
4035
4036         return rc;
4037 }
4038
4039 static void
4040 lfsck_layout_slave_notify_master(const struct lu_env *env,
4041                                  struct lfsck_component *com,
4042                                  enum lfsck_events event, int result)
4043 {
4044         struct lfsck_layout              *lo    = com->lc_file_ram;
4045         struct lfsck_instance            *lfsck = com->lc_lfsck;
4046         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
4047         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
4048         struct lfsck_layout_slave_target *llst;
4049         struct obd_export                *exp;
4050         struct ptlrpc_request_set        *set;
4051         int                               rc;
4052         ENTRY;
4053
4054         CDEBUG(D_LFSCK, "%s: layout LFSCK slave notifies master\n",
4055                lfsck_lfsck2name(com->lc_lfsck));
4056
4057         set = ptlrpc_prep_set();
4058         if (set == NULL)
4059                 RETURN_EXIT;
4060
4061         memset(lr, 0, sizeof(*lr));
4062         lr->lr_event = event;
4063         lr->lr_flags = LEF_FROM_OST;
4064         lr->lr_status = result;
4065         lr->lr_index = lfsck_dev_idx(lfsck);
4066         lr->lr_active = LFSCK_TYPE_LAYOUT;
4067         lr->lr_flags2 = lo->ll_flags;
4068         llsd->llsd_touch_gen++;
4069         spin_lock(&llsd->llsd_lock);
4070         while (!list_empty(&llsd->llsd_master_list)) {
4071                 llst = list_entry(llsd->llsd_master_list.next,
4072                                   struct lfsck_layout_slave_target,
4073                                   llst_list);
4074                 if (llst->llst_gen == llsd->llsd_touch_gen)
4075                         break;
4076
4077                 llst->llst_gen = llsd->llsd_touch_gen;
4078                 list_move_tail(&llst->llst_list,
4079                                &llsd->llsd_master_list);
4080                 atomic_inc(&llst->llst_ref);
4081                 spin_unlock(&llsd->llsd_lock);
4082
4083                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4084                                                llst->llst_index);
4085                 if (exp == NULL) {
4086                         lfsck_layout_llst_del(llsd, llst);
4087                         lfsck_layout_llst_put(llst);
4088                         spin_lock(&llsd->llsd_lock);
4089                         continue;
4090                 }
4091
4092                 rc = lfsck_layout_async_notify(env, exp, lr, set);
4093                 if (rc != 0)
4094                         CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4095                                "notify %s for layout: rc = %d\n",
4096                                lfsck_lfsck2name(lfsck),
4097                                exp->exp_obd->obd_name, rc);
4098
4099                 lfsck_layout_llst_put(llst);
4100                 class_export_put(exp);
4101                 spin_lock(&llsd->llsd_lock);
4102         }
4103         spin_unlock(&llsd->llsd_lock);
4104
4105         ptlrpc_set_wait(set);
4106         ptlrpc_set_destroy(set);
4107
4108         RETURN_EXIT;
4109 }
4110
4111 /*
4112  * \ret -ENODATA: unrecognized stripe
4113  * \ret = 0     : recognized stripe
4114  * \ret < 0     : other failures
4115  */
4116 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
4117                                            struct lfsck_component *com,
4118                                            struct lu_fid *cfid,
4119                                            struct lu_fid *pfid)
4120 {
4121         struct lfsck_thread_info        *info   = lfsck_env_info(env);
4122         struct lu_buf                   *buf    = &info->lti_big_buf;
4123         struct ost_id                   *oi     = &info->lti_oi;
4124         struct dt_object                *obj;
4125         struct lov_mds_md_v1            *lmm;
4126         struct lov_ost_data_v1          *objs;
4127         __u32                            idx    = pfid->f_stripe_idx;
4128         __u32                            magic;
4129         int                              rc     = 0;
4130         int                              i;
4131         __u16                            count;
4132         ENTRY;
4133
4134         pfid->f_ver = 0;
4135         obj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
4136         if (IS_ERR(obj))
4137                 RETURN(PTR_ERR(obj));
4138
4139         dt_read_lock(env, obj, 0);
4140         if (unlikely(dt_object_exists(obj) == 0 ||
4141                      lfsck_is_dead_obj(obj)))
4142                 GOTO(unlock, rc = -ENOENT);
4143
4144         if (!S_ISREG(lfsck_object_type(obj)))
4145                 GOTO(unlock, rc = -ENODATA);
4146
4147         rc = lfsck_layout_get_lovea(env, obj, buf);
4148         if (rc < 0)
4149                 GOTO(unlock, rc);
4150
4151         if (rc == 0)
4152                 GOTO(unlock, rc = -ENODATA);
4153
4154         lmm = buf->lb_buf;
4155         rc = lfsck_layout_verify_header(lmm);
4156         if (rc != 0)
4157                 GOTO(unlock, rc);
4158
4159         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
4160          * been verified in lfsck_layout_verify_header() already. If some
4161          * new magic introduced in the future, then layout LFSCK needs to
4162          * be updated also. */
4163         magic = le32_to_cpu(lmm->lmm_magic);
4164         if (magic == LOV_MAGIC_V1) {
4165                 objs = &lmm->lmm_objects[0];
4166         } else {
4167                 LASSERT(magic == LOV_MAGIC_V3);
4168                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4169         }
4170
4171         fid_to_ostid(cfid, oi);
4172         count = le16_to_cpu(lmm->lmm_stripe_count);
4173         for (i = 0; i < count; i++, objs++) {
4174                 struct ost_id oi2;
4175
4176                 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
4177                 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
4178                         GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
4179         }
4180
4181         GOTO(unlock, rc = -ENODATA);
4182
4183 unlock:
4184         dt_read_unlock(env, obj);
4185         lfsck_object_put(env, obj);
4186
4187         return rc;
4188 }
4189
4190 /*
4191  * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
4192  * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
4193  * layout EA from MDT to OST. On one hand, the OST no need to understand
4194  * the layout EA structure; on the other hand, it may cause trouble when
4195  * transfer large layout EA from MDT to OST via normal OUT RPC.
4196  *
4197  * \ret > 0: unrecognized stripe
4198  * \ret = 0: recognized stripe
4199  * \ret < 0: other failures
4200  */
4201 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
4202                                           struct lfsck_component *com,
4203                                           struct lu_fid *cfid,
4204                                           struct lu_fid *pfid)
4205 {
4206         struct lfsck_instance    *lfsck  = com->lc_lfsck;
4207         struct obd_device        *obd    = lfsck->li_obd;
4208         struct seq_server_site   *ss     = lfsck_dev_site(lfsck);
4209         struct obd_export        *exp    = NULL;
4210         struct ptlrpc_request    *req    = NULL;
4211         struct lfsck_request     *lr;
4212         struct lu_seq_range      *range  = &lfsck_env_info(env)->lti_range;
4213         int                       rc     = 0;
4214         ENTRY;
4215
4216         if (unlikely(fid_is_idif(pfid)))
4217                 RETURN(1);
4218
4219         fld_range_set_any(range);
4220         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), range);
4221         if (rc != 0)
4222                 RETURN(rc == -ENOENT ? 1 : rc);
4223
4224         if (unlikely(!fld_range_is_mdt(range)))
4225                 RETURN(1);
4226
4227         exp = lustre_find_lwp_by_index(obd->obd_name, range->lsr_index);
4228         if (unlikely(exp == NULL))
4229                 RETURN(1);
4230
4231         if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
4232                 GOTO(out, rc = -EOPNOTSUPP);
4233
4234         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4235         if (req == NULL)
4236                 GOTO(out, rc = -ENOMEM);
4237
4238         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4239         if (rc != 0) {
4240                 ptlrpc_request_free(req);
4241
4242                 GOTO(out, rc);
4243         }
4244
4245         lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4246         memset(lr, 0, sizeof(*lr));
4247         lr->lr_event = LE_PAIRS_VERIFY;
4248         lr->lr_active = LFSCK_TYPE_LAYOUT;
4249         lr->lr_fid = *cfid; /* OST-object itself FID. */
4250         lr->lr_fid2 = *pfid; /* The claimed parent FID. */
4251
4252         ptlrpc_request_set_replen(req);
4253         rc = ptlrpc_queue_wait(req);
4254         ptlrpc_req_finished(req);
4255
4256         if (rc == -ENOENT || rc == -ENODATA)
4257                 rc = 1;
4258
4259         GOTO(out, rc);
4260
4261 out:
4262         if (exp != NULL)
4263                 class_export_put(exp);
4264
4265         return rc;
4266 }
4267
4268 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
4269                                           struct lfsck_component *com,
4270                                           struct lfsck_request *lr)
4271 {
4272         struct dt_object        *obj;
4273         int                      rc     = 0;
4274         ENTRY;
4275
4276         obj = lfsck_object_find_bottom(env, com->lc_lfsck, &lr->lr_fid);
4277         if (IS_ERR(obj))
4278                 GOTO(log, rc = PTR_ERR(obj));
4279
4280         dt_write_lock(env, obj, 0);
4281         if (unlikely(dt_object_exists(obj) == 0 ||
4282                      lfsck_is_dead_obj(obj)))
4283                 GOTO(unlock, rc = 0);
4284
4285         rc = __lfsck_layout_update_pfid(env, obj, &lr->lr_fid2,
4286                                         lr->lr_fid2.f_ver);
4287
4288         GOTO(unlock, rc);
4289
4290 unlock:
4291         dt_write_unlock(env, obj);
4292         lfsck_object_put(env, obj);
4293
4294 log:
4295         CDEBUG(D_LFSCK, "%s: layout LFSCK slave repaired pfid for "DFID
4296                ", parent "DFID": rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
4297                PFID(&lr->lr_fid), PFID(&lr->lr_fid2), rc);
4298
4299         return rc;
4300 }
4301
4302 /* layout APIs */
4303
4304 static void lfsck_layout_slave_quit(const struct lu_env *env,
4305                                     struct lfsck_component *com);
4306
4307 static int lfsck_layout_reset(const struct lu_env *env,
4308                               struct lfsck_component *com, bool init)
4309 {
4310         struct lfsck_layout     *lo    = com->lc_file_ram;
4311         int                      rc;
4312
4313         down_write(&com->lc_sem);
4314         if (init) {
4315                 memset(lo, 0, com->lc_file_size);
4316         } else {
4317                 __u32 count = lo->ll_success_count;
4318                 __u64 last_time = lo->ll_time_last_complete;
4319
4320                 memset(lo, 0, com->lc_file_size);
4321                 lo->ll_success_count = count;
4322                 lo->ll_time_last_complete = last_time;
4323         }
4324
4325         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
4326         lo->ll_status = LS_INIT;
4327
4328         if (com->lc_lfsck->li_master) {
4329                 struct lfsck_assistant_data *lad = com->lc_data;
4330
4331                 lad->lad_incomplete = 0;
4332                 CFS_RESET_BITMAP(lad->lad_bitmap);
4333         }
4334
4335         rc = lfsck_layout_store(env, com);
4336         if (rc == 0 && com->lc_lfsck->li_master)
4337                 rc = lfsck_load_sub_trace_files(env, com,
4338                         &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
4339         up_write(&com->lc_sem);
4340
4341         CDEBUG(D_LFSCK, "%s: layout LFSCK reset: rc = %d\n",
4342                lfsck_lfsck2name(com->lc_lfsck), rc);
4343
4344         return rc;
4345 }
4346
4347 static void lfsck_layout_fail(const struct lu_env *env,
4348                               struct lfsck_component *com, bool new_checked)
4349 {
4350         struct lfsck_layout *lo = com->lc_file_ram;
4351
4352         down_write(&com->lc_sem);
4353         if (new_checked)
4354                 com->lc_new_checked++;
4355         lfsck_layout_record_failure(env, com->lc_lfsck, lo);
4356         up_write(&com->lc_sem);
4357 }
4358
4359 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
4360                                           struct lfsck_component *com, bool init)
4361 {
4362         struct lfsck_instance   *lfsck   = com->lc_lfsck;
4363         struct lfsck_layout     *lo      = com->lc_file_ram;
4364         int                      rc;
4365
4366         if (!init) {
4367                 rc = lfsck_checkpoint_generic(env, com);
4368                 if (rc != 0)
4369                         return rc > 0 ? 0 : rc;
4370         }
4371
4372         down_write(&com->lc_sem);
4373         if (init) {
4374                 lo->ll_pos_latest_start =
4375                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
4376         } else {
4377                 lo->ll_pos_last_checkpoint =
4378                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
4379                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4380                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4381                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4382                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4383                 com->lc_new_checked = 0;
4384         }
4385
4386         rc = lfsck_layout_store(env, com);
4387         up_write(&com->lc_sem);
4388
4389         CDEBUG(D_LFSCK, "%s: layout LFSCK master checkpoint at the pos ["
4390                "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
4391                lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
4392
4393         return rc;
4394 }
4395
4396 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
4397                                          struct lfsck_component *com, bool init)
4398 {
4399         struct lfsck_instance   *lfsck = com->lc_lfsck;
4400         struct lfsck_layout     *lo    = com->lc_file_ram;
4401         int                      rc;
4402
4403         if (com->lc_new_checked == 0 && !init)
4404                 return 0;
4405
4406         down_write(&com->lc_sem);
4407         if (init) {
4408                 lo->ll_pos_latest_start =
4409                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
4410         } else {
4411                 lo->ll_pos_last_checkpoint =
4412                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
4413                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4414                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4415                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4416                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4417                 com->lc_new_checked = 0;
4418         }
4419
4420         rc = lfsck_layout_store(env, com);
4421         up_write(&com->lc_sem);
4422
4423         CDEBUG(D_LFSCK, "%s: layout LFSCK slave checkpoint at the pos ["
4424                "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
4425                lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
4426
4427         return rc;
4428 }
4429
4430 static int lfsck_layout_prep(const struct lu_env *env,
4431                              struct lfsck_component *com,
4432                              struct lfsck_start *start)
4433 {
4434         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4435         struct lfsck_layout     *lo     = com->lc_file_ram;
4436         struct lfsck_position   *pos    = &com->lc_pos_start;
4437
4438         fid_zero(&pos->lp_dir_parent);
4439         pos->lp_dir_cookie = 0;
4440         if (lo->ll_status == LS_COMPLETED ||
4441             lo->ll_status == LS_PARTIAL ||
4442             /* To handle orphan, must scan from the beginning. */
4443             (start != NULL && start->ls_flags & LPF_OST_ORPHAN)) {
4444                 int rc;
4445
4446                 rc = lfsck_layout_reset(env, com, false);
4447                 if (rc == 0)
4448                         rc = lfsck_set_param(env, lfsck, start, true);
4449
4450                 if (rc != 0) {
4451                         CDEBUG(D_LFSCK, "%s: layout LFSCK prep failed: "
4452                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4453
4454                         return rc;
4455                 }
4456         }
4457
4458         down_write(&com->lc_sem);
4459         lo->ll_time_latest_start = cfs_time_current_sec();
4460         spin_lock(&lfsck->li_lock);
4461         if (lo->ll_flags & LF_SCANNED_ONCE) {
4462                 if (!lfsck->li_drop_dryrun ||
4463                     lo->ll_pos_first_inconsistent == 0) {
4464                         lo->ll_status = LS_SCANNING_PHASE2;
4465                         list_move_tail(&com->lc_link,
4466                                        &lfsck->li_list_double_scan);
4467                         pos->lp_oit_cookie = 0;
4468                 } else {
4469                         int i;
4470
4471                         lo->ll_status = LS_SCANNING_PHASE1;
4472                         lo->ll_run_time_phase1 = 0;
4473                         lo->ll_run_time_phase2 = 0;
4474                         lo->ll_objs_checked_phase1 = 0;
4475                         lo->ll_objs_checked_phase2 = 0;
4476                         lo->ll_objs_failed_phase1 = 0;
4477                         lo->ll_objs_failed_phase2 = 0;
4478                         for (i = 0; i < LLIT_MAX; i++)
4479                                 lo->ll_objs_repaired[i] = 0;
4480
4481                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4482                         fid_zero(&com->lc_fid_latest_scanned_phase2);
4483                 }
4484         } else {
4485                 lo->ll_status = LS_SCANNING_PHASE1;
4486                 if (!lfsck->li_drop_dryrun ||
4487                     lo->ll_pos_first_inconsistent == 0)
4488                         pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
4489                 else
4490                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4491         }
4492         spin_unlock(&lfsck->li_lock);
4493         up_write(&com->lc_sem);
4494
4495         return 0;
4496 }
4497
4498 static int lfsck_layout_slave_prep(const struct lu_env *env,
4499                                    struct lfsck_component *com,
4500                                    struct lfsck_start_param *lsp)
4501 {
4502         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4503         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4504         struct lfsck_layout             *lo     = com->lc_file_ram;
4505         struct lfsck_start              *start  = lsp->lsp_start;
4506         int                              rc;
4507
4508         rc = lfsck_layout_prep(env, com, start);
4509         if (rc != 0)
4510                 return rc;
4511
4512         if (lo->ll_flags & LF_CRASHED_LASTID &&
4513             list_empty(&llsd->llsd_master_list)) {
4514                 LASSERT(lfsck->li_out_notify != NULL);
4515
4516                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4517                                      LE_LASTID_REBUILDING);
4518         }
4519
4520         if (!lsp->lsp_index_valid)
4521                 return 0;
4522
4523         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
4524         if (rc == 0 && start != NULL && start->ls_flags & LPF_OST_ORPHAN) {
4525                 LASSERT(!llsd->llsd_rbtree_valid);
4526
4527                 write_lock(&llsd->llsd_rb_lock);
4528                 rc = lfsck_rbtree_setup(env, com);
4529                 write_unlock(&llsd->llsd_rb_lock);
4530         }
4531
4532         CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
4533                "%llu]\n", lfsck_lfsck2name(lfsck),
4534                com->lc_pos_start.lp_oit_cookie);
4535
4536         return rc;
4537 }
4538
4539 static int lfsck_layout_master_prep(const struct lu_env *env,
4540                                     struct lfsck_component *com,
4541                                     struct lfsck_start_param *lsp)
4542 {
4543         int rc;
4544         ENTRY;
4545
4546         rc = lfsck_layout_load_bitmap(env, com);
4547         if (rc != 0) {
4548                 rc = lfsck_layout_reset(env, com, false);
4549                 if (rc == 0)
4550                         rc = lfsck_set_param(env, com->lc_lfsck,
4551                                              lsp->lsp_start, true);
4552
4553                 if (rc != 0)
4554                         GOTO(log, rc);
4555         }
4556
4557         rc = lfsck_layout_prep(env, com, lsp->lsp_start);
4558         if (rc != 0)
4559                 RETURN(rc);
4560
4561         rc = lfsck_start_assistant(env, com, lsp);
4562
4563         GOTO(log, rc);
4564
4565 log:
4566         CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
4567                "%llu]\n", lfsck_lfsck2name(com->lc_lfsck),
4568                com->lc_pos_start.lp_oit_cookie);
4569
4570         return 0;
4571 }
4572
4573 /* Pre-fetch the attribute for each stripe in the given layout EA. */
4574 static int lfsck_layout_scan_stripes(const struct lu_env *env,
4575                                      struct lfsck_component *com,
4576                                      struct dt_object *parent,
4577                                      struct lov_mds_md_v1 *lmm)
4578 {
4579         struct lfsck_thread_info        *info    = lfsck_env_info(env);
4580         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4581         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
4582         struct lfsck_layout             *lo      = com->lc_file_ram;
4583         struct lfsck_assistant_data     *lad     = com->lc_data;
4584         struct lfsck_assistant_object   *lso     = NULL;
4585         struct lov_ost_data_v1          *objs;
4586         struct lfsck_tgt_descs          *ltds    = &lfsck->li_ost_descs;
4587         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4588         struct ptlrpc_thread            *athread = &lad->lad_thread;
4589         struct l_wait_info               lwi     = { 0 };
4590         struct lu_buf                    buf;
4591         int                              rc      = 0;
4592         int                              i;
4593         __u32                            magic;
4594         __u16                            count;
4595         ENTRY;
4596
4597         lfsck_buf_init(&buf, &info->lti_old_pfid,
4598                        sizeof(struct filter_fid_old));
4599         count = le16_to_cpu(lmm->lmm_stripe_count);
4600         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
4601          * been verified in lfsck_layout_verify_header() already. If some
4602          * new magic introduced in the future, then layout LFSCK needs to
4603          * be updated also. */
4604         magic = le32_to_cpu(lmm->lmm_magic);
4605         if (magic == LOV_MAGIC_V1) {
4606                 objs = &lmm->lmm_objects[0];
4607         } else {
4608                 LASSERT(magic == LOV_MAGIC_V3);
4609                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4610         }
4611
4612         for (i = 0; i < count; i++, objs++) {
4613                 struct lu_fid           *fid    = &info->lti_fid;
4614                 struct ost_id           *oi     = &info->lti_oi;
4615                 struct lfsck_layout_req *llr;
4616                 struct lfsck_tgt_desc   *tgt    = NULL;
4617                 struct dt_object        *cobj   = NULL;
4618                 __u32                    index;
4619                 bool                     wakeup = false;
4620
4621                 if (unlikely(lovea_slot_is_dummy(objs)))
4622                         continue;
4623
4624                 l_wait_event(mthread->t_ctl_waitq,
4625                              lad->lad_prefetched < bk->lb_async_windows ||
4626                              !thread_is_running(mthread) ||
4627                              thread_is_stopped(athread),
4628                              &lwi);
4629
4630                 if (unlikely(!thread_is_running(mthread)) ||
4631                              thread_is_stopped(athread))
4632                         GOTO(out, rc = 0);
4633
4634                 if (unlikely(lfsck_is_dead_obj(parent)))
4635                         GOTO(out, rc = 0);
4636
4637                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4638                 index = le32_to_cpu(objs->l_ost_idx);
4639                 rc = ostid_to_fid(fid, oi, index);
4640                 if (rc != 0) {
4641                         CDEBUG(D_LFSCK, "%s: get invalid layout EA for "DFID
4642                                ": "DOSTID", idx:%u\n", lfsck_lfsck2name(lfsck),
4643                                PFID(lfsck_dto2fid(parent)), POSTID(oi), index);
4644                         goto next;
4645                 }
4646
4647                 tgt = lfsck_tgt_get(ltds, index);
4648                 if (unlikely(tgt == NULL)) {
4649                         CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
4650                                "did not join the layout LFSCK\n",
4651                                lfsck_lfsck2name(lfsck), index);
4652                         lfsck_lad_set_bitmap(env, com, index);
4653                         goto next;
4654                 }
4655
4656                 /* There is potential deadlock race condition between object
4657                  * destroy and layout LFSCK. Consider the following scenario:
4658                  *
4659                  * 1) The LFSCK thread obtained the parent object firstly, at
4660                  *    that time, the parent object has not been destroyed yet.
4661                  *
4662                  * 2) One RPC service thread destroyed the parent and all its
4663                  *    children objects. Because the LFSCK is referencing the
4664                  *    parent object, then the parent object will be marked as
4665                  *    dying in RAM. On the other hand, the parent object is
4666                  *    referencing all its children objects, then all children
4667                  *    objects will be marked as dying in RAM also.
4668                  *
4669                  * 3) The LFSCK thread tries to find some child object with
4670                  *    the parent object referenced. Then it will find that the
4671                  *    child object is dying. According to the object visibility
4672                  *    rules: the object with dying flag cannot be returned to
4673                  *    others. So the LFSCK thread has to wait until the dying
4674                  *    object has been purged from RAM, then it can allocate a
4675                  *    new object (with the same FID) in RAM. Unfortunately, the
4676                  *    LFSCK thread itself is referencing the parent object, and
4677                  *    cause the parent object cannot be purged, then cause the
4678                  *    child object cannot be purged also. So the LFSCK thread
4679                  *    will fall into deadlock.
4680                  *
4681                  * We introduce non-blocked version lu_object_find() to allow
4682                  * the LFSCK thread to return failure immediately (instead of
4683                  * wait) when it finds dying (child) object, then the LFSCK
4684                  * thread can check whether the parent object is dying or not.
4685                  * So avoid above deadlock. LU-5395 */
4686                 cobj = lfsck_object_find_by_dev_nowait(env, tgt->ltd_tgt, fid);
4687                 if (IS_ERR(cobj)) {
4688                         if (lfsck_is_dead_obj(parent)) {
4689                                 lfsck_tgt_put(tgt);
4690
4691                                 GOTO(out, rc = 0);
4692                         }
4693
4694                         rc = PTR_ERR(cobj);
4695                         goto next;
4696                 }
4697
4698                 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_ASSISTANT_DIRECT)) {
4699                         rc = dt_declare_attr_get(env, cobj);
4700                         if (rc != 0)
4701                                 goto next;
4702
4703                         rc = dt_declare_xattr_get(env, cobj, &buf,
4704                                                   XATTR_NAME_FID);
4705                         if (rc != 0)
4706                                 goto next;
4707                 }
4708
4709                 if (lso == NULL) {
4710                         struct lu_attr *attr = &info->lti_la;
4711
4712                         rc = dt_attr_get(env, parent, attr);
4713                         if (rc != 0)
4714                                 goto next;
4715
4716                         lso = lfsck_assistant_object_init(env,
4717                                 lfsck_dto2fid(parent), attr,
4718                                 lfsck->li_pos_current.lp_oit_cookie, false);
4719                         if (IS_ERR(lso)) {
4720                                 rc = PTR_ERR(lso);
4721                                 lso = NULL;
4722
4723                                 goto next;
4724                         }
4725                 }
4726
4727                 llr = lfsck_layout_assistant_req_init(lso, cobj, index, i);
4728                 if (IS_ERR(llr)) {
4729                         rc = PTR_ERR(llr);
4730                         goto next;
4731                 }
4732
4733                 cobj = NULL;
4734                 spin_lock(&lad->lad_lock);
4735                 if (lad->lad_assistant_status < 0) {
4736                         spin_unlock(&lad->lad_lock);
4737                         lfsck_layout_assistant_req_fini(env, &llr->llr_lar);
4738                         lfsck_tgt_put(tgt);
4739                         RETURN(lad->lad_assistant_status);
4740                 }
4741
4742                 list_add_tail(&llr->llr_lar.lar_list, &lad->lad_req_list);
4743                 if (lad->lad_prefetched == 0)
4744                         wakeup = true;
4745
4746                 lad->lad_prefetched++;
4747                 spin_unlock(&lad->lad_lock);
4748                 if (wakeup)
4749                         wake_up_all(&athread->t_ctl_waitq);
4750
4751 next:
4752                 down_write(&com->lc_sem);
4753                 com->lc_new_checked++;
4754                 if (rc < 0)
4755                         lfsck_layout_record_failure(env, lfsck, lo);
4756                 up_write(&com->lc_sem);
4757
4758                 if (cobj != NULL && !IS_ERR(cobj))
4759                         lfsck_object_put(env, cobj);
4760
4761                 if (likely(tgt != NULL))
4762                         lfsck_tgt_put(tgt);
4763
4764                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4765                         GOTO(out, rc);
4766         }
4767
4768         GOTO(out, rc = 0);
4769
4770 out:
4771         if (lso != NULL)
4772                 lfsck_assistant_object_put(env, lso);
4773
4774         return rc;
4775 }
4776
4777 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
4778  * the OST-object's attribute and generate an structure lfsck_layout_req on the
4779  * list ::lad_req_list.
4780  *
4781  * For each request on above list, the lfsck_layout_assistant thread compares
4782  * the OST side attribute with local attribute, if inconsistent, then repair it.
4783  *
4784  * All above processing is async mode with pipeline. */
4785 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
4786                                         struct lfsck_component *com,
4787                                         struct dt_object *obj)
4788 {
4789         struct lfsck_thread_info        *info   = lfsck_env_info(env);
4790         struct ost_id                   *oi     = &info->lti_oi;
4791         struct lfsck_layout             *lo     = com->lc_file_ram;
4792         struct lfsck_assistant_data     *lad    = com->lc_data;
4793         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4794         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
4795         struct thandle                  *handle = NULL;
4796         struct lu_buf                   *buf    = &info->lti_big_buf;
4797         struct lov_mds_md_v1            *lmm    = NULL;
4798         struct dt_device                *dev    = lfsck_obj2dev(obj);
4799         struct lustre_handle             lh     = { 0 };
4800         struct lu_buf                    ea_buf = { NULL };
4801         int                              rc     = 0;
4802         int                              size   = 0;
4803         bool                             locked = false;
4804         bool                             stripe = false;
4805         bool                             bad_oi = false;
4806         ENTRY;
4807
4808         if (!S_ISREG(lfsck_object_type(obj)))
4809                 GOTO(out, rc = 0);
4810
4811         if (lad->lad_assistant_status < 0)
4812                 GOTO(out, rc = -ESRCH);
4813
4814         fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
4815         lmm_oi_cpu_to_le(oi, oi);
4816         dt_read_lock(env, obj, 0);
4817         locked = true;
4818
4819 again:
4820         if (dt_object_exists(obj) == 0 ||
4821             lfsck_is_dead_obj(obj))
4822                 GOTO(out, rc = 0);
4823
4824         rc = lfsck_layout_get_lovea(env, obj, buf);
4825         if (rc <= 0)
4826                 GOTO(out, rc);
4827
4828         size = rc;
4829         lmm = buf->lb_buf;
4830         rc = lfsck_layout_verify_header(lmm);
4831         /* If the LOV EA crashed, then it is possible to be rebuilt later
4832          * when handle orphan OST-objects. */
4833         if (rc != 0)
4834                 GOTO(out, rc);
4835
4836         if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0)
4837                 GOTO(out, stripe = true);
4838
4839         /* Inconsistent lmm_oi, should be repaired. */
4840         bad_oi = true;
4841         lmm->lmm_oi = *oi;
4842
4843         if (bk->lb_param & LPF_DRYRUN) {
4844                 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4845
4846                 GOTO(out, stripe = true);
4847         }
4848
4849         if (!lustre_handle_is_used(&lh)) {
4850                 dt_read_unlock(env, obj);
4851                 locked = false;
4852                 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4853                                       MDS_INODELOCK_LAYOUT |
4854                                       MDS_INODELOCK_XATTR, LCK_EX);
4855                 if (rc != 0)
4856                         GOTO(out, rc);
4857
4858                 handle = dt_trans_create(env, dev);
4859                 if (IS_ERR(handle))
4860                         GOTO(out, rc = PTR_ERR(handle));
4861
4862                 lfsck_buf_init(&ea_buf, lmm, size);
4863                 rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
4864                                           LU_XATTR_REPLACE, handle);
4865                 if (rc != 0)
4866                         GOTO(out, rc);
4867
4868                 rc = dt_trans_start_local(env, dev, handle);
4869                 if (rc != 0)
4870                         GOTO(out, rc);
4871
4872                 dt_write_lock(env, obj, 0);
4873                 locked = true;
4874
4875                 goto again;
4876         }
4877
4878         rc = dt_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
4879                           LU_XATTR_REPLACE, handle);
4880         if (rc != 0)
4881                 GOTO(out, rc);
4882
4883         lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4884
4885         GOTO(out, stripe = true);
4886
4887 out:
4888         if (locked) {
4889                 if (lustre_handle_is_used(&lh))
4890                         dt_write_unlock(env, obj);
4891                 else
4892                         dt_read_unlock(env, obj);
4893         }
4894
4895         if (handle != NULL && !IS_ERR(handle))
4896                 dt_trans_stop(env, dev, handle);
4897
4898         lfsck_ibits_unlock(&lh, LCK_EX);
4899
4900         if (bad_oi)
4901                 CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
4902                        DFID": rc = %d\n", lfsck_lfsck2name(lfsck),
4903                        bk->lb_param & LPF_DRYRUN ? "found" : "repaired",
4904                        PFID(lfsck_dto2fid(obj)), rc);
4905
4906         if (stripe) {
4907                 rc = lfsck_layout_scan_stripes(env, com, obj, lmm);
4908         } else {
4909                 down_write(&com->lc_sem);
4910                 com->lc_new_checked++;
4911                 if (rc < 0)
4912                         lfsck_layout_record_failure(env, lfsck, lo);
4913                 up_write(&com->lc_sem);
4914         }
4915
4916         return rc;
4917 }
4918
4919 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
4920                                        struct lfsck_component *com,
4921                                        struct dt_object *obj)
4922 {
4923         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4924         struct lfsck_layout             *lo     = com->lc_file_ram;
4925         const struct lu_fid             *fid    = lfsck_dto2fid(obj);
4926         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4927         struct lfsck_layout_seq         *lls;
4928         __u64                            seq;
4929         __u64                            oid;
4930         int                              rc;
4931         ENTRY;
4932
4933         LASSERT(llsd != NULL);
4934
4935         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
4936             cfs_fail_val == lfsck_dev_idx(lfsck)) {
4937                 struct l_wait_info       lwi = LWI_TIMEOUT(cfs_time_seconds(1),
4938                                                            NULL, NULL);
4939                 struct ptlrpc_thread    *thread = &lfsck->li_thread;
4940
4941                 l_wait_event(thread->t_ctl_waitq,
4942                              !thread_is_running(thread),
4943                              &lwi);
4944         }
4945
4946         lfsck_rbtree_update_bitmap(env, com, fid, false);
4947
4948         down_write(&com->lc_sem);
4949         if (fid_is_idif(fid))
4950                 seq = 0;
4951         else if (!fid_is_norm(fid) ||
4952                  !fid_is_for_ostobj(env, lfsck, obj, fid))
4953                 GOTO(unlock, rc = 0);
4954         else
4955                 seq = fid_seq(fid);
4956         com->lc_new_checked++;
4957
4958         lls = lfsck_layout_seq_lookup(llsd, seq);
4959         if (lls == NULL) {
4960                 OBD_ALLOC_PTR(lls);
4961                 if (unlikely(lls == NULL))
4962                         GOTO(unlock, rc = -ENOMEM);
4963
4964                 INIT_LIST_HEAD(&lls->lls_list);
4965                 lls->lls_seq = seq;
4966                 rc = lfsck_layout_lastid_load(env, com, lls);
4967                 if (rc != 0) {
4968                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
4969                               "load LAST_ID for %#llx: rc = %d\n",
4970                               lfsck_lfsck2name(com->lc_lfsck), seq, rc);
4971                         lo->ll_objs_failed_phase1++;
4972                         OBD_FREE_PTR(lls);
4973                         GOTO(unlock, rc);
4974                 }
4975
4976                 lfsck_layout_seq_insert(llsd, lls);
4977         }
4978
4979         if (unlikely(fid_is_last_id(fid)))
4980                 GOTO(unlock, rc = 0);
4981
4982         if (fid_is_idif(fid))
4983                 oid = fid_idif_id(fid_seq(fid), fid_oid(fid), fid_ver(fid));
4984         else
4985                 oid = fid_oid(fid);
4986
4987         if (oid > lls->lls_lastid_known)
4988                 lls->lls_lastid_known = oid;
4989
4990         if (oid > lls->lls_lastid) {
4991                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
4992                         /* OFD may create new objects during LFSCK scanning. */
4993                         rc = lfsck_layout_lastid_reload(env, com, lls);
4994                         if (unlikely(rc != 0)) {
4995                                 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
4996                                       "reload LAST_ID for %#llx: rc = %d\n",
4997                                       lfsck_lfsck2name(com->lc_lfsck),
4998                                       lls->lls_seq, rc);
4999
5000                                 GOTO(unlock, rc);
5001                         }
5002
5003                         if (oid <= lls->lls_lastid ||
5004                             lo->ll_flags & LF_CRASHED_LASTID)
5005                                 GOTO(unlock, rc = 0);
5006
5007                         LASSERT(lfsck->li_out_notify != NULL);
5008
5009                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5010                                              LE_LASTID_REBUILDING);
5011                         lo->ll_flags |= LF_CRASHED_LASTID;
5012
5013                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
5014                                "LAST_ID file (2) for the sequence %#llx"
5015                                ", old value %llu, known value %llu\n",
5016                                lfsck_lfsck2name(lfsck), lls->lls_seq,
5017                                lls->lls_lastid, oid);
5018                 }
5019
5020                 lls->lls_lastid = oid;
5021                 lls->lls_dirty = 1;
5022         }
5023
5024         GOTO(unlock, rc = 0);
5025
5026 unlock:
5027         up_write(&com->lc_sem);
5028
5029         return rc;
5030 }
5031
5032 static int lfsck_layout_exec_dir(const struct lu_env *env,
5033                                  struct lfsck_component *com,
5034                                  struct lfsck_assistant_object *lso,
5035                                  struct lu_dirent *ent, __u16 type)
5036 {
5037         return 0;
5038 }
5039
5040 static int lfsck_layout_master_post(const struct lu_env *env,
5041                                     struct lfsck_component *com,
5042                                     int result, bool init)
5043 {
5044         struct lfsck_instance   *lfsck  = com->lc_lfsck;
5045         struct lfsck_layout     *lo     = com->lc_file_ram;
5046         int                      rc;
5047         ENTRY;
5048
5049         lfsck_post_generic(env, com, &result);
5050
5051         down_write(&com->lc_sem);
5052         spin_lock(&lfsck->li_lock);
5053         if (!init)
5054                 lo->ll_pos_last_checkpoint =
5055                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
5056
5057         if (result > 0) {
5058                 if (lo->ll_flags & LF_INCOMPLETE)
5059                         lo->ll_status = LS_PARTIAL;
5060                 else
5061                         lo->ll_status = LS_SCANNING_PHASE2;
5062                 lo->ll_flags |= LF_SCANNED_ONCE;
5063                 lo->ll_flags &= ~LF_UPGRADE;
5064                 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
5065         } else if (result == 0) {
5066                 if (lfsck->li_status != 0)
5067                         lo->ll_status = lfsck->li_status;
5068                 else
5069                         lo->ll_status = LS_STOPPED;
5070                 if (lo->ll_status != LS_PAUSED)
5071                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5072         } else {
5073                 lo->ll_status = LS_FAILED;
5074                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5075         }
5076         spin_unlock(&lfsck->li_lock);
5077
5078         if (!init) {
5079                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
5080                                 HALF_SEC - lfsck->li_time_last_checkpoint);
5081                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
5082                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5083                 com->lc_new_checked = 0;
5084         }
5085
5086         rc = lfsck_layout_store(env, com);
5087         up_write(&com->lc_sem);
5088
5089         CDEBUG(D_LFSCK, "%s: layout LFSCK master post done: rc = %d\n",
5090                lfsck_lfsck2name(lfsck), rc);
5091
5092         RETURN(rc);
5093 }
5094
5095 static int lfsck_layout_slave_post(const struct lu_env *env,
5096                                    struct lfsck_component *com,
5097                                    int result, bool init)
5098 {
5099         struct lfsck_instance   *lfsck = com->lc_lfsck;
5100         struct lfsck_layout     *lo    = com->lc_file_ram;
5101         int                      rc;
5102         bool                     done  = false;
5103
5104         down_write(&com->lc_sem);
5105         rc = lfsck_layout_lastid_store(env, com);
5106         if (rc != 0)
5107                 result = rc;
5108
5109         LASSERT(lfsck->li_out_notify != NULL);
5110
5111         spin_lock(&lfsck->li_lock);
5112         if (!init)
5113                 lo->ll_pos_last_checkpoint =
5114                                 lfsck->li_pos_checkpoint.lp_oit_cookie;
5115
5116         if (result > 0) {
5117                 lo->ll_status = LS_SCANNING_PHASE2;
5118                 lo->ll_flags |= LF_SCANNED_ONCE;
5119                 if (lo->ll_flags & LF_CRASHED_LASTID) {
5120                         done = true;
5121                         lo->ll_flags &= ~LF_CRASHED_LASTID;
5122
5123                         CDEBUG(D_LFSCK, "%s: layout LFSCK has rebuilt "
5124                                "crashed LAST_ID files successfully\n",
5125                                lfsck_lfsck2name(lfsck));
5126                 }
5127                 lo->ll_flags &= ~LF_UPGRADE;
5128                 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
5129         } else if (result == 0) {
5130                 if (lfsck->li_status != 0)
5131                         lo->ll_status = lfsck->li_status;
5132                 else
5133                         lo->ll_status = LS_STOPPED;
5134                 if (lo->ll_status != LS_PAUSED)
5135                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5136         } else {
5137                 lo->ll_status = LS_FAILED;
5138                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5139         }
5140         spin_unlock(&lfsck->li_lock);
5141
5142         if (done)
5143                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5144                                      LE_LASTID_REBUILT);
5145
5146         if (!init) {
5147                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
5148                                 HALF_SEC - lfsck->li_time_last_checkpoint);
5149                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
5150                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5151                 com->lc_new_checked = 0;
5152         }
5153
5154         rc = lfsck_layout_store(env, com);
5155         up_write(&com->lc_sem);
5156
5157         lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
5158
5159         CDEBUG(D_LFSCK, "%s: layout LFSCK slave post done: rc = %d\n",
5160                lfsck_lfsck2name(lfsck), rc);
5161
5162         return rc;
5163 }
5164
5165 static void lfsck_layout_dump(const struct lu_env *env,
5166                               struct lfsck_component *com, struct seq_file *m)
5167 {
5168         struct lfsck_instance   *lfsck = com->lc_lfsck;
5169         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
5170         struct lfsck_layout     *lo    = com->lc_file_ram;
5171
5172         down_read(&com->lc_sem);
5173         seq_printf(m, "name: lfsck_layout\n"
5174                    "magic: %#x\n"
5175                    "version: %d\n"
5176                    "status: %s\n",
5177                    lo->ll_magic,
5178                    bk->lb_version,
5179                    lfsck_status2name(lo->ll_status));
5180
5181         lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
5182
5183         lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
5184
5185         lfsck_time_dump(m, lo->ll_time_last_complete, "last_completed");
5186
5187         lfsck_time_dump(m, lo->ll_time_latest_start, "latest_start");
5188
5189         lfsck_time_dump(m, lo->ll_time_last_checkpoint, "last_checkpoint");
5190
5191         seq_printf(m, "latest_start_position: %llu\n"
5192                    "last_checkpoint_position: %llu\n"
5193                    "first_failure_position: %llu\n",
5194                    lo->ll_pos_latest_start,
5195                    lo->ll_pos_last_checkpoint,
5196                    lo->ll_pos_first_inconsistent);
5197
5198         seq_printf(m, "success_count: %u\n"
5199                    "repaired_dangling: %llu\n"
5200                    "repaired_unmatched_pair: %llu\n"
5201                    "repaired_multiple_referenced: %llu\n"
5202                    "repaired_orphan: %llu\n"
5203                    "repaired_inconsistent_owner: %llu\n"
5204                    "repaired_others: %llu\n"
5205                    "skipped: %llu\n"
5206                    "failed_phase1: %llu\n"
5207                    "failed_phase2: %llu\n",
5208                    lo->ll_success_count,
5209                    lo->ll_objs_repaired[LLIT_DANGLING - 1],
5210                    lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
5211                    lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
5212                    lo->ll_objs_repaired[LLIT_ORPHAN - 1],
5213                    lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
5214                    lo->ll_objs_repaired[LLIT_OTHERS - 1],
5215                    lo->ll_objs_skipped,
5216                    lo->ll_objs_failed_phase1,
5217                    lo->ll_objs_failed_phase2);
5218
5219         if (lo->ll_status == LS_SCANNING_PHASE1) {
5220                 __u64 pos;
5221                 cfs_duration_t duration = cfs_time_current() -
5222                                           lfsck->li_time_last_checkpoint;
5223                 __u64 checked = lo->ll_objs_checked_phase1 +
5224                                 com->lc_new_checked;
5225                 __u64 speed = checked;
5226                 __u64 new_checked = com->lc_new_checked *
5227                                     msecs_to_jiffies(MSEC_PER_SEC);
5228                 __u32 rtime = lo->ll_run_time_phase1 +
5229                               cfs_duration_sec(duration + HALF_SEC);
5230
5231                 if (duration != 0)
5232                         do_div(new_checked, duration);
5233                 if (rtime != 0)
5234                         do_div(speed, rtime);
5235                 seq_printf(m, "checked_phase1: %llu\n"
5236                            "checked_phase2: %llu\n"
5237                            "run_time_phase1: %u seconds\n"
5238                            "run_time_phase2: %u seconds\n"
5239                            "average_speed_phase1: %llu items/sec\n"
5240                            "average_speed_phase2: N/A\n"
5241                            "real-time_speed_phase1: %llu items/sec\n"
5242                            "real-time_speed_phase2: N/A\n",
5243                            checked,
5244                            lo->ll_objs_checked_phase2,
5245                            rtime,
5246                            lo->ll_run_time_phase2,
5247                            speed,
5248                            new_checked);
5249
5250                 if (likely(lfsck->li_di_oit)) {
5251                         const struct dt_it_ops *iops =
5252                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
5253
5254                         /* The low layer otable-based iteration position may NOT
5255                          * exactly match the layout-based directory traversal
5256                          * cookie. Generally, it is not a serious issue. But the
5257                          * caller should NOT make assumption on that. */
5258                         pos = iops->store(env, lfsck->li_di_oit);
5259                         if (!lfsck->li_current_oit_processed)
5260                                 pos--;
5261                 } else {
5262                         pos = lo->ll_pos_last_checkpoint;
5263                 }
5264
5265                 seq_printf(m, "current_position: %llu\n", pos);
5266         } else if (lo->ll_status == LS_SCANNING_PHASE2) {
5267                 cfs_duration_t duration = cfs_time_current() -
5268                                           com->lc_time_last_checkpoint;
5269                 __u64 checked = lo->ll_objs_checked_phase2 +
5270                                 com->lc_new_checked;
5271                 __u64 speed1 = lo->ll_objs_checked_phase1;
5272                 __u64 speed2 = checked;
5273                 __u64 new_checked = com->lc_new_checked *
5274                                     msecs_to_jiffies(MSEC_PER_SEC);
5275                 __u32 rtime = lo->ll_run_time_phase2 +
5276                               cfs_duration_sec(duration + HALF_SEC);
5277
5278                 if (duration != 0)
5279                         do_div(new_checked, duration);
5280                 if (lo->ll_run_time_phase1 != 0)
5281                         do_div(speed1, lo->ll_run_time_phase1);
5282                 if (rtime != 0)
5283                         do_div(speed2, rtime);
5284                 seq_printf(m, "checked_phase1: %llu\n"
5285                            "checked_phase2: %llu\n"
5286                            "run_time_phase1: %u seconds\n"
5287                            "run_time_phase2: %u seconds\n"
5288                            "average_speed_phase1: %llu items/sec\n"
5289                            "average_speed_phase2: %llu items/sec\n"
5290                            "real-time_speed_phase1: N/A\n"
5291                            "real-time_speed_phase2: %llu items/sec\n"
5292                            "current_position: "DFID"\n",
5293                            lo->ll_objs_checked_phase1,
5294                            checked,
5295                            lo->ll_run_time_phase1,
5296                            rtime,
5297                            speed1,
5298                            speed2,
5299                            new_checked,
5300                            PFID(&com->lc_fid_latest_scanned_phase2));
5301         } else {
5302                 __u64 speed1 = lo->ll_objs_checked_phase1;
5303                 __u64 speed2 = lo->ll_objs_checked_phase2;
5304
5305                 if (lo->ll_run_time_phase1 != 0)
5306                         do_div(speed1, lo->ll_run_time_phase1);
5307                 if (lo->ll_run_time_phase2 != 0)
5308                         do_div(speed2, lo->ll_run_time_phase2);
5309                 seq_printf(m, "checked_phase1: %llu\n"
5310                            "checked_phase2: %llu\n"
5311                            "run_time_phase1: %u seconds\n"
5312                            "run_time_phase2: %u seconds\n"
5313                            "average_speed_phase1: %llu items/sec\n"
5314                            "average_speed_phase2: %llu objs/sec\n"
5315                            "real-time_speed_phase1: N/A\n"
5316                            "real-time_speed_phase2: N/A\n"
5317                            "current_position: N/A\n",
5318                            lo->ll_objs_checked_phase1,
5319                            lo->ll_objs_checked_phase2,
5320                            lo->ll_run_time_phase1,
5321                            lo->ll_run_time_phase2,
5322                            speed1,
5323                            speed2);
5324         }
5325
5326         up_read(&com->lc_sem);
5327 }
5328
5329 static int lfsck_layout_master_double_scan(const struct lu_env *env,
5330                                            struct lfsck_component *com)
5331 {
5332         struct lfsck_layout             *lo     = com->lc_file_ram;
5333         struct lfsck_assistant_data     *lad    = com->lc_data;
5334         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5335         struct lfsck_tgt_descs          *ltds;
5336         struct lfsck_tgt_desc           *ltd;
5337         struct lfsck_tgt_desc           *next;
5338         int                              rc;
5339
5340         rc = lfsck_double_scan_generic(env, com, lo->ll_status);
5341
5342         if (thread_is_stopped(&lad->lad_thread)) {
5343                 LASSERT(list_empty(&lad->lad_req_list));
5344                 LASSERT(list_empty(&lad->lad_ost_phase1_list));
5345                 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
5346
5347                 ltds = &lfsck->li_ost_descs;
5348                 spin_lock(&ltds->ltd_lock);
5349                 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5350                                          ltd_layout_phase_list) {
5351                         list_del_init(&ltd->ltd_layout_phase_list);
5352                 }
5353                 spin_unlock(&ltds->ltd_lock);
5354
5355                 ltds = &lfsck->li_mdt_descs;
5356                 spin_lock(&ltds->ltd_lock);
5357                 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5358                                          ltd_layout_phase_list) {
5359                         list_del_init(&ltd->ltd_layout_phase_list);
5360                 }
5361                 spin_unlock(&ltds->ltd_lock);
5362         }
5363
5364         return rc;
5365 }
5366
5367 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
5368                                           struct lfsck_component *com)
5369 {
5370         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5371         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
5372         struct lfsck_layout             *lo     = com->lc_file_ram;
5373         struct ptlrpc_thread            *thread = &lfsck->li_thread;
5374         int                              rc;
5375         ENTRY;
5376
5377         CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan start\n",
5378                lfsck_lfsck2name(lfsck));
5379
5380         atomic_inc(&lfsck->li_double_scan_count);
5381
5382         if (lo->ll_flags & LF_INCOMPLETE)
5383                 GOTO(done, rc = 1);
5384
5385         com->lc_new_checked = 0;
5386         com->lc_new_scanned = 0;
5387         com->lc_time_last_checkpoint = cfs_time_current();
5388         com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
5389                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
5390
5391         while (1) {
5392                 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(30),
5393                                                      NULL, NULL);
5394
5395                 rc = lfsck_layout_slave_query_master(env, com);
5396                 if (list_empty(&llsd->llsd_master_list)) {
5397                         if (unlikely(!thread_is_running(thread)))
5398                                 rc = 0;
5399                         else
5400                                 rc = 1;
5401
5402                         GOTO(done, rc);
5403                 }
5404
5405                 if (rc < 0)
5406                         GOTO(done, rc);
5407
5408                 rc = l_wait_event(thread->t_ctl_waitq,
5409                                   !thread_is_running(thread) ||
5410                                   lo->ll_flags & LF_INCOMPLETE ||
5411                                   list_empty(&llsd->llsd_master_list),
5412                                   &lwi);
5413                 if (unlikely(!thread_is_running(thread)))
5414                         GOTO(done, rc = 0);
5415
5416                 if (lo->ll_flags & LF_INCOMPLETE)
5417                         GOTO(done, rc = 1);
5418
5419                 if (rc == -ETIMEDOUT)
5420                         continue;
5421
5422                 GOTO(done, rc = (rc < 0 ? rc : 1));
5423         }
5424
5425 done:
5426         rc = lfsck_layout_double_scan_result(env, com, rc);
5427         lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE,
5428                         (rc > 0 && lo->ll_flags & LF_INCOMPLETE) ? 0 : rc);
5429         lfsck_layout_slave_quit(env, com);
5430         if (atomic_dec_and_test(&lfsck->li_double_scan_count))
5431                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5432
5433         CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan finished, "
5434                "status %d: rc = %d\n",
5435                lfsck_lfsck2name(lfsck), lo->ll_status, rc);
5436
5437         return rc;
5438 }
5439
5440 static void lfsck_layout_master_data_release(const struct lu_env *env,
5441                                              struct lfsck_component *com)
5442 {
5443         struct lfsck_assistant_data     *lad    = com->lc_data;
5444         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5445         struct lfsck_tgt_descs          *ltds;
5446         struct lfsck_tgt_desc           *ltd;
5447         struct lfsck_tgt_desc           *next;
5448
5449         LASSERT(lad != NULL);
5450         LASSERT(thread_is_init(&lad->lad_thread) ||
5451                 thread_is_stopped(&lad->lad_thread));
5452         LASSERT(list_empty(&lad->lad_req_list));
5453
5454         com->lc_data = NULL;
5455
5456         ltds = &lfsck->li_ost_descs;
5457         spin_lock(&ltds->ltd_lock);
5458         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
5459                                  ltd_layout_phase_list) {
5460                 list_del_init(&ltd->ltd_layout_phase_list);
5461         }
5462         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5463                                  ltd_layout_phase_list) {
5464                 list_del_init(&ltd->ltd_layout_phase_list);
5465         }
5466         list_for_each_entry_safe(ltd, next, &lad->lad_ost_list,
5467                                  ltd_layout_list) {
5468                 list_del_init(&ltd->ltd_layout_list);
5469         }
5470         spin_unlock(&ltds->ltd_lock);
5471
5472         ltds = &lfsck->li_mdt_descs;
5473         spin_lock(&ltds->ltd_lock);
5474         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5475                                  ltd_layout_phase_list) {
5476                 list_del_init(&ltd->ltd_layout_phase_list);
5477         }
5478         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5479                                  ltd_layout_phase_list) {
5480                 list_del_init(&ltd->ltd_layout_phase_list);
5481         }
5482         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
5483                                  ltd_layout_list) {
5484                 list_del_init(&ltd->ltd_layout_list);
5485         }
5486         spin_unlock(&ltds->ltd_lock);
5487
5488         if (likely(lad->lad_bitmap != NULL))
5489                 CFS_FREE_BITMAP(lad->lad_bitmap);
5490
5491         OBD_FREE_PTR(lad);
5492 }
5493
5494 static void lfsck_layout_slave_data_release(const struct lu_env *env,
5495                                             struct lfsck_component *com)
5496 {
5497         struct lfsck_layout_slave_data *llsd = com->lc_data;
5498
5499         lfsck_layout_slave_quit(env, com);
5500         com->lc_data = NULL;
5501         OBD_FREE_PTR(llsd);
5502 }
5503
5504 static void lfsck_layout_master_quit(const struct lu_env *env,
5505                                      struct lfsck_component *com)
5506 {
5507         struct lfsck_assistant_data     *lad    = com->lc_data;
5508         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5509         struct lfsck_tgt_descs          *ltds;
5510         struct lfsck_tgt_desc           *ltd;
5511         struct lfsck_tgt_desc           *next;
5512
5513         LASSERT(lad != NULL);
5514
5515         lfsck_quit_generic(env, com);
5516
5517         LASSERT(thread_is_init(&lad->lad_thread) ||
5518                 thread_is_stopped(&lad->lad_thread));
5519         LASSERT(list_empty(&lad->lad_req_list));
5520
5521         ltds = &lfsck->li_ost_descs;
5522         spin_lock(&ltds->ltd_lock);
5523         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
5524                                  ltd_layout_phase_list) {
5525                 list_del_init(&ltd->ltd_layout_phase_list);
5526         }
5527         list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5528                                  ltd_layout_phase_list) {
5529                 list_del_init(&ltd->ltd_layout_phase_list);
5530         }
5531         spin_unlock(&ltds->ltd_lock);
5532
5533         ltds = &lfsck->li_mdt_descs;
5534         spin_lock(&ltds->ltd_lock);
5535         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5536                                  ltd_layout_phase_list) {
5537                 list_del_init(&ltd->ltd_layout_phase_list);
5538         }
5539         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5540                                  ltd_layout_phase_list) {
5541                 list_del_init(&ltd->ltd_layout_phase_list);
5542         }
5543         spin_unlock(&ltds->ltd_lock);
5544 }
5545
5546 static void lfsck_layout_slave_quit(const struct lu_env *env,
5547                                     struct lfsck_component *com)
5548 {
5549         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5550         struct lfsck_layout_seq          *lls;
5551         struct lfsck_layout_seq          *next;
5552         struct lfsck_layout_slave_target *llst;
5553
5554         LASSERT(llsd != NULL);
5555
5556         down_write(&com->lc_sem);
5557         list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
5558                                  lls_list) {
5559                 list_del_init(&lls->lls_list);
5560                 lfsck_object_put(env, lls->lls_lastid_obj);
5561                 OBD_FREE_PTR(lls);
5562         }
5563         up_write(&com->lc_sem);
5564
5565         spin_lock(&llsd->llsd_lock);
5566         while (!list_empty(&llsd->llsd_master_list)) {
5567                 llst = list_entry(llsd->llsd_master_list.next,
5568                                   struct lfsck_layout_slave_target, llst_list);
5569                 list_del_init(&llst->llst_list);
5570                 spin_unlock(&llsd->llsd_lock);
5571                 lfsck_layout_llst_put(llst);
5572                 spin_lock(&llsd->llsd_lock);
5573         }
5574         spin_unlock(&llsd->llsd_lock);
5575
5576         lfsck_rbtree_cleanup(env, com);
5577 }
5578
5579 static int lfsck_layout_master_in_notify(const struct lu_env *env,
5580                                          struct lfsck_component *com,
5581                                          struct lfsck_request *lr,
5582                                          struct thandle *th)
5583 {
5584         struct lfsck_instance           *lfsck = com->lc_lfsck;
5585         struct lfsck_layout             *lo    = com->lc_file_ram;
5586         struct lfsck_assistant_data     *lad   = com->lc_data;
5587         struct lfsck_tgt_descs          *ltds;
5588         struct lfsck_tgt_desc           *ltd;
5589         bool                             fail  = false;
5590         ENTRY;
5591
5592         if (lr->lr_event == LE_PAIRS_VERIFY) {
5593                 int rc;
5594
5595                 rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
5596                                                      &lr->lr_fid2);
5597
5598                 RETURN(rc);
5599         }
5600
5601         CDEBUG(D_LFSCK, "%s: layout LFSCK master handles notify %u "
5602                "from %s %x, status %d, flags %x, flags2 %x\n",
5603                lfsck_lfsck2name(lfsck), lr->lr_event,
5604                (lr->lr_flags & LEF_FROM_OST) ? "OST" : "MDT",
5605                lr->lr_index, lr->lr_status, lr->lr_flags, lr->lr_flags2);
5606
5607         if (lr->lr_event != LE_PHASE1_DONE &&
5608             lr->lr_event != LE_PHASE2_DONE &&
5609             lr->lr_event != LE_PEER_EXIT)
5610                 RETURN(-EINVAL);
5611
5612         if (lr->lr_flags & LEF_FROM_OST)
5613                 ltds = &lfsck->li_ost_descs;
5614         else
5615                 ltds = &lfsck->li_mdt_descs;
5616         spin_lock(&ltds->ltd_lock);
5617         ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
5618         if (ltd == NULL) {
5619                 spin_unlock(&ltds->ltd_lock);
5620
5621                 RETURN(-ENXIO);
5622         }
5623
5624         list_del_init(&ltd->ltd_layout_phase_list);
5625         switch (lr->lr_event) {
5626         case LE_PHASE1_DONE:
5627                 if (lr->lr_status <= 0 || lr->lr_flags2 & LF_INCOMPLETE) {
5628                         if (lr->lr_flags2 & LF_INCOMPLETE) {
5629                                 if (lr->lr_flags & LEF_FROM_OST)
5630                                         lfsck_lad_set_bitmap(env, com,
5631                                                              ltd->ltd_index);
5632                                 else
5633                                         lo->ll_flags |= LF_INCOMPLETE;
5634                         }
5635                         ltd->ltd_layout_done = 1;
5636                         list_del_init(&ltd->ltd_layout_list);
5637                         fail = true;
5638                         break;
5639                 }
5640
5641                 if (lr->lr_flags & LEF_FROM_OST) {
5642                         if (list_empty(&ltd->ltd_layout_list))
5643                                 list_add_tail(&ltd->ltd_layout_list,
5644                                               &lad->lad_ost_list);
5645                         list_add_tail(&ltd->ltd_layout_phase_list,
5646                                       &lad->lad_ost_phase2_list);
5647                 } else {
5648                         if (list_empty(&ltd->ltd_layout_list))
5649                                 list_add_tail(&ltd->ltd_layout_list,
5650                                               &lad->lad_mdt_list);
5651                         list_add_tail(&ltd->ltd_layout_phase_list,
5652                                       &lad->lad_mdt_phase2_list);
5653                 }
5654                 break;
5655         case LE_PHASE2_DONE:
5656                 ltd->ltd_layout_done = 1;
5657                 if (!list_empty(&ltd->ltd_layout_list)) {
5658                         list_del_init(&ltd->ltd_layout_list);
5659                         if (lr->lr_flags2 & LF_INCOMPLETE) {
5660                                 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
5661                                 fail = true;
5662                         }
5663                 }
5664
5665                 break;
5666         case LE_PEER_EXIT:
5667                 fail = true;
5668                 ltd->ltd_layout_done = 1;
5669                 list_del_init(&ltd->ltd_layout_list);
5670                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) &&
5671                     !(lr->lr_flags & LEF_FROM_OST))
5672                                 lo->ll_flags |= LF_INCOMPLETE;
5673                 break;
5674         default:
5675                 break;
5676         }
5677         spin_unlock(&ltds->ltd_lock);
5678
5679         if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
5680                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5681
5682                 memset(stop, 0, sizeof(*stop));
5683                 stop->ls_status = lr->lr_status;
5684                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5685                 lfsck_stop(env, lfsck->li_bottom, stop);
5686         } else if (lfsck_phase2_next_ready(lad)) {
5687                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
5688         }
5689
5690         RETURN(0);
5691 }
5692
5693 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
5694                                         struct lfsck_component *com,
5695                                         struct lfsck_request *lr,
5696                                         struct thandle *th)
5697 {
5698         struct lfsck_instance            *lfsck = com->lc_lfsck;
5699         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5700         struct lfsck_layout_slave_target *llst;
5701         int                               rc;
5702         ENTRY;
5703
5704         switch (lr->lr_event) {
5705         case LE_FID_ACCESSED:
5706                 lfsck_rbtree_update_bitmap(env, com, &lr->lr_fid, true);
5707                 RETURN(0);
5708         case LE_CONDITIONAL_DESTROY:
5709                 rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
5710                 RETURN(rc);
5711         case LE_PAIRS_VERIFY: {
5712                 lr->lr_status = LPVS_INIT;
5713                 /* Firstly, if the MDT-object which is claimed via OST-object
5714                  * local stored PFID xattr recognizes the OST-object, then it
5715                  * must be that the client given PFID is wrong. */
5716                 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5717                                                     &lr->lr_fid3);
5718                 if (rc <= 0)
5719                         RETURN(0);
5720
5721                 lr->lr_status = LPVS_INCONSISTENT;
5722                 /* The OST-object local stored PFID xattr is stale. We need to
5723                  * check whether the MDT-object that is claimed via the client
5724                  * given PFID information recognizes the OST-object or not. If
5725                  * matches, then need to update the OST-object's PFID xattr. */
5726                 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5727                                                     &lr->lr_fid2);
5728                 /* For rc < 0 case:
5729                  * We are not sure whether the client given PFID information
5730                  * is correct or not, do nothing to avoid improper fixing.
5731                  *
5732                  * For rc > 0 case:
5733                  * The client given PFID information is also invalid, we can
5734                  * NOT fix the OST-object inconsistency.
5735                  */
5736                 if (rc != 0)
5737                         RETURN(rc);
5738
5739                 lr->lr_status = LPVS_INCONSISTENT_TOFIX;
5740                 rc = lfsck_layout_slave_repair_pfid(env, com, lr);
5741
5742                 RETURN(rc);
5743         }
5744         case LE_PHASE1_DONE: {
5745                 if (lr->lr_flags2 & LF_INCOMPLETE) {
5746                         struct lfsck_layout *lo = com->lc_file_ram;
5747
5748                         lo->ll_flags |= LF_INCOMPLETE;
5749                         llst = lfsck_layout_llst_find_and_del(llsd,
5750                                                               lr->lr_index,
5751                                                               true);
5752                         if (llst != NULL) {
5753                                 lfsck_layout_llst_put(llst);
5754                                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5755                         }
5756                 }
5757
5758                 RETURN(0);
5759         }
5760         case LE_PHASE2_DONE:
5761         case LE_PEER_EXIT:
5762                 CDEBUG(D_LFSCK, "%s: layout LFSCK slave handle notify %u "
5763                        "from MDT %x, status %d\n", lfsck_lfsck2name(lfsck),
5764                        lr->lr_event, lr->lr_index, lr->lr_status);
5765                 break;
5766         default:
5767                 RETURN(-EINVAL);
5768         }
5769
5770         llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
5771         if (llst == NULL)
5772                 RETURN(0);
5773
5774         lfsck_layout_llst_put(llst);
5775         if (list_empty(&llsd->llsd_master_list))
5776                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5777
5778         if (lr->lr_event == LE_PEER_EXIT &&
5779             (lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT ||
5780              (list_empty(&llsd->llsd_master_list) &&
5781               (lr->lr_status == LS_STOPPED ||
5782                lr->lr_status == LS_CO_STOPPED)))) {
5783                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5784
5785                 memset(stop, 0, sizeof(*stop));
5786                 stop->ls_status = lr->lr_status;
5787                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5788                 lfsck_stop(env, lfsck->li_bottom, stop);
5789         }
5790
5791         RETURN(0);
5792 }
5793
5794 static void lfsck_layout_repaired(struct lfsck_layout *lo, __u64 *count)
5795 {
5796         int i;
5797
5798         for (i = 0; i < LLIT_MAX; i++)
5799                 *count += lo->ll_objs_repaired[i];
5800 }
5801
5802 static int lfsck_layout_query_all(const struct lu_env *env,
5803                                   struct lfsck_component *com,
5804                                   __u32 *mdts_count, __u32 *osts_count,
5805                                   __u64 *repaired)
5806 {
5807         struct lfsck_layout *lo = com->lc_file_ram;
5808         struct lfsck_tgt_descs *ltds;
5809         struct lfsck_tgt_desc *ltd;
5810         int idx;
5811         int rc;
5812         ENTRY;
5813
5814         rc = lfsck_query_all(env, com);
5815         if (rc != 0)
5816                 RETURN(rc);
5817
5818         ltds = &com->lc_lfsck->li_mdt_descs;
5819         down_read(&ltds->ltd_rw_sem);
5820         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
5821                 ltd = lfsck_ltd2tgt(ltds, idx);
5822                 LASSERT(ltd != NULL);
5823
5824                 mdts_count[ltd->ltd_layout_status]++;
5825                 *repaired += ltd->ltd_layout_repaired;
5826         }
5827         up_read(&ltds->ltd_rw_sem);
5828
5829         ltds = &com->lc_lfsck->li_ost_descs;
5830         down_read(&ltds->ltd_rw_sem);
5831         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
5832                 ltd = lfsck_ltd2tgt(ltds, idx);
5833                 LASSERT(ltd != NULL);
5834
5835                 osts_count[ltd->ltd_layout_status]++;
5836                 *repaired += ltd->ltd_layout_repaired;
5837         }
5838         up_read(&ltds->ltd_rw_sem);
5839
5840         down_read(&com->lc_sem);
5841         mdts_count[lo->ll_status]++;
5842         lfsck_layout_repaired(lo, repaired);
5843         up_read(&com->lc_sem);
5844
5845         RETURN(0);
5846 }
5847
5848 static int lfsck_layout_query(const struct lu_env *env,
5849                               struct lfsck_component *com,
5850                               struct lfsck_request *req,
5851                               struct lfsck_reply *rep,
5852                               struct lfsck_query *que, int idx)
5853 {
5854         struct lfsck_layout *lo = com->lc_file_ram;
5855         int rc = 0;
5856
5857         if (que != NULL) {
5858                 LASSERT(com->lc_lfsck->li_master);
5859
5860                 rc = lfsck_layout_query_all(env, com,
5861                                             que->lu_mdts_count[idx],
5862                                             que->lu_osts_count[idx],
5863                                             &que->lu_repaired[idx]);
5864         } else {
5865                 down_read(&com->lc_sem);
5866                 rep->lr_status = lo->ll_status;
5867                 if (req->lr_flags & LEF_QUERY_ALL)
5868                         lfsck_layout_repaired(lo, &rep->lr_repaired);
5869                 up_read(&com->lc_sem);
5870         }
5871
5872         return rc;
5873 }
5874
5875 /* with lfsck::li_lock held */
5876 static int lfsck_layout_slave_join(const struct lu_env *env,
5877                                    struct lfsck_component *com,
5878                                    struct lfsck_start_param *lsp)
5879 {
5880         struct lfsck_instance            *lfsck = com->lc_lfsck;
5881         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5882         struct lfsck_layout_slave_target *llst;
5883         struct lfsck_start               *start = lsp->lsp_start;
5884         int                               rc    = 0;
5885         ENTRY;
5886
5887         if (start == NULL || !(start->ls_flags & LPF_OST_ORPHAN))
5888                 RETURN(0);
5889
5890         if (!lsp->lsp_index_valid)
5891                 RETURN(-EINVAL);
5892
5893         /* If someone is running the LFSCK without orphan handling,
5894          * it will not maintain the object accessing rbtree. So we
5895          * cannot join it for orphan handling. */
5896         if (!llsd->llsd_rbtree_valid)
5897                 RETURN(-EBUSY);
5898
5899         spin_unlock(&lfsck->li_lock);
5900         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5901         spin_lock(&lfsck->li_lock);
5902         if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
5903                 spin_unlock(&lfsck->li_lock);
5904                 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
5905                                                       true);
5906                 if (llst != NULL)
5907                         lfsck_layout_llst_put(llst);
5908                 spin_lock(&lfsck->li_lock);
5909                 rc = -EAGAIN;
5910         }
5911
5912         RETURN(rc);
5913 }
5914
5915 static struct lfsck_operations lfsck_layout_master_ops = {
5916         .lfsck_reset            = lfsck_layout_reset,
5917         .lfsck_fail             = lfsck_layout_fail,
5918         .lfsck_checkpoint       = lfsck_layout_master_checkpoint,
5919         .lfsck_prep             = lfsck_layout_master_prep,
5920         .lfsck_exec_oit         = lfsck_layout_master_exec_oit,
5921         .lfsck_exec_dir         = lfsck_layout_exec_dir,
5922         .lfsck_post             = lfsck_layout_master_post,
5923         .lfsck_dump             = lfsck_layout_dump,
5924         .lfsck_double_scan      = lfsck_layout_master_double_scan,
5925         .lfsck_data_release     = lfsck_layout_master_data_release,
5926         .lfsck_quit             = lfsck_layout_master_quit,
5927         .lfsck_in_notify        = lfsck_layout_master_in_notify,
5928         .lfsck_query            = lfsck_layout_query,
5929 };
5930
5931 static struct lfsck_operations lfsck_layout_slave_ops = {
5932         .lfsck_reset            = lfsck_layout_reset,
5933         .lfsck_fail             = lfsck_layout_fail,
5934         .lfsck_checkpoint       = lfsck_layout_slave_checkpoint,
5935         .lfsck_prep             = lfsck_layout_slave_prep,
5936         .lfsck_exec_oit         = lfsck_layout_slave_exec_oit,
5937         .lfsck_exec_dir         = lfsck_layout_exec_dir,
5938         .lfsck_post             = lfsck_layout_slave_post,
5939         .lfsck_dump             = lfsck_layout_dump,
5940         .lfsck_double_scan      = lfsck_layout_slave_double_scan,
5941         .lfsck_data_release     = lfsck_layout_slave_data_release,
5942         .lfsck_quit             = lfsck_layout_slave_quit,
5943         .lfsck_in_notify        = lfsck_layout_slave_in_notify,
5944         .lfsck_query            = lfsck_layout_query,
5945         .lfsck_join             = lfsck_layout_slave_join,
5946 };
5947
5948 static void lfsck_layout_assistant_fill_pos(const struct lu_env *env,
5949                                             struct lfsck_component *com,
5950                                             struct lfsck_position *pos)
5951 {
5952         struct lfsck_assistant_data     *lad = com->lc_data;
5953         struct lfsck_layout_req         *llr;
5954
5955         if (((struct lfsck_layout *)(com->lc_file_ram))->ll_status !=
5956             LS_SCANNING_PHASE1)
5957                 return;
5958
5959         if (list_empty(&lad->lad_req_list))
5960                 return;
5961
5962         llr = list_entry(lad->lad_req_list.next,
5963                          struct lfsck_layout_req,
5964                          llr_lar.lar_list);
5965         pos->lp_oit_cookie = llr->llr_lar.lar_parent->lso_oit_cookie - 1;
5966 }
5967
5968 struct lfsck_assistant_operations lfsck_layout_assistant_ops = {
5969         .la_handler_p1          = lfsck_layout_assistant_handler_p1,
5970         .la_handler_p2          = lfsck_layout_assistant_handler_p2,
5971         .la_fill_pos            = lfsck_layout_assistant_fill_pos,
5972         .la_double_scan_result  = lfsck_layout_double_scan_result,
5973         .la_req_fini            = lfsck_layout_assistant_req_fini,
5974         .la_sync_failures       = lfsck_layout_assistant_sync_failures,
5975 };
5976
5977 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
5978 {
5979         struct lfsck_component  *com;
5980         struct lfsck_layout     *lo;
5981         struct dt_object        *root = NULL;
5982         struct dt_object        *obj;
5983         int                      i;
5984         int                      rc;
5985         ENTRY;
5986
5987         OBD_ALLOC_PTR(com);
5988         if (com == NULL)
5989                 RETURN(-ENOMEM);
5990
5991         INIT_LIST_HEAD(&com->lc_link);
5992         INIT_LIST_HEAD(&com->lc_link_dir);
5993         init_rwsem(&com->lc_sem);
5994         atomic_set(&com->lc_ref, 1);
5995         com->lc_lfsck = lfsck;
5996         com->lc_type = LFSCK_TYPE_LAYOUT;
5997         if (lfsck->li_master) {
5998                 com->lc_ops = &lfsck_layout_master_ops;
5999                 com->lc_data = lfsck_assistant_data_init(
6000                                 &lfsck_layout_assistant_ops,
6001                                 LFSCK_LAYOUT);
6002                 if (com->lc_data == NULL)
6003                         GOTO(out, rc = -ENOMEM);
6004
6005                 for (i = 0; i < LFSCK_STF_COUNT; i++)
6006                         mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6007         } else {
6008                 struct lfsck_layout_slave_data *llsd;
6009
6010                 com->lc_ops = &lfsck_layout_slave_ops;
6011                 OBD_ALLOC_PTR(llsd);
6012                 if (llsd == NULL)
6013                         GOTO(out, rc = -ENOMEM);
6014
6015                 INIT_LIST_HEAD(&llsd->llsd_seq_list);
6016                 INIT_LIST_HEAD(&llsd->llsd_master_list);
6017                 spin_lock_init(&llsd->llsd_lock);
6018                 llsd->llsd_rb_root = RB_ROOT;
6019                 rwlock_init(&llsd->llsd_rb_lock);
6020                 com->lc_data = llsd;
6021         }
6022         com->lc_file_size = sizeof(*lo);
6023         OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6024         if (com->lc_file_ram == NULL)
6025                 GOTO(out, rc = -ENOMEM);
6026
6027         OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6028         if (com->lc_file_disk == NULL)
6029                 GOTO(out, rc = -ENOMEM);
6030
6031         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6032         if (IS_ERR(root))
6033                 GOTO(out, rc = PTR_ERR(root));
6034
6035         if (unlikely(!dt_try_as_dir(env, root)))
6036                 GOTO(out, rc = -ENOTDIR);
6037
6038         obj = local_file_find_or_create(env, lfsck->li_los, root,
6039                                         LFSCK_LAYOUT,
6040                                         S_IFREG | S_IRUGO | S_IWUSR);
6041         if (IS_ERR(obj))
6042                 GOTO(out, rc = PTR_ERR(obj));
6043
6044         com->lc_obj = obj;
6045         rc = lfsck_layout_load(env, com);
6046         if (rc > 0)
6047                 rc = lfsck_layout_reset(env, com, true);
6048         else if (rc == -ENOENT)
6049                 rc = lfsck_layout_init(env, com);
6050         else if (lfsck->li_master)
6051                 rc = lfsck_load_sub_trace_files(env, com,
6052                                 &dt_lfsck_layout_dangling_features,
6053                                 LFSCK_LAYOUT, false);
6054
6055         if (rc != 0)
6056                 GOTO(out, rc);
6057
6058         lo = com->lc_file_ram;
6059         switch (lo->ll_status) {
6060         case LS_INIT:
6061         case LS_COMPLETED:
6062         case LS_FAILED:
6063         case LS_STOPPED:
6064         case LS_PARTIAL:
6065                 spin_lock(&lfsck->li_lock);
6066                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6067                 spin_unlock(&lfsck->li_lock);
6068                 break;
6069         default:
6070                 CERROR("%s: unknown lfsck_layout status %d\n",
6071                        lfsck_lfsck2name(lfsck), lo->ll_status);
6072                 /* fall through */
6073         case LS_SCANNING_PHASE1:
6074         case LS_SCANNING_PHASE2:
6075                 /* No need to store the status to disk right now.
6076                  * If the system crashed before the status stored,
6077                  * it will be loaded back when next time. */
6078                 lo->ll_status = LS_CRASHED;
6079                 if (!lfsck->li_master)
6080                         lo->ll_flags |= LF_INCOMPLETE;
6081                 /* fall through */
6082         case LS_PAUSED:
6083         case LS_CRASHED:
6084         case LS_CO_FAILED:
6085         case LS_CO_STOPPED:
6086         case LS_CO_PAUSED:
6087                 spin_lock(&lfsck->li_lock);
6088                 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6089                 spin_unlock(&lfsck->li_lock);
6090                 break;
6091         }
6092
6093         if (lo->ll_flags & LF_CRASHED_LASTID) {
6094                 LASSERT(lfsck->li_out_notify != NULL);
6095
6096                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
6097                                      LE_LASTID_REBUILDING);
6098         }
6099
6100         GOTO(out, rc = 0);
6101
6102 out:
6103         if (root != NULL && !IS_ERR(root))
6104                 lfsck_object_put(env, root);
6105
6106         if (rc != 0) {
6107                 lfsck_component_cleanup(env, com);
6108                 CERROR("%s: fail to init layout LFSCK component: rc = %d\n",
6109                        lfsck_lfsck2name(lfsck), rc);
6110         }
6111
6112         return rc;
6113 }
6114
6115 struct lfsck_orphan_it {
6116         struct lfsck_component           *loi_com;
6117         struct lfsck_rbtree_node         *loi_lrn;
6118         struct lfsck_layout_slave_target *loi_llst;
6119         struct lu_fid                     loi_key;
6120         struct lu_orphan_rec              loi_rec;
6121         __u64                             loi_hash;
6122         unsigned int                      loi_over:1;
6123 };
6124
6125 static int lfsck_fid_match_idx(const struct lu_env *env,
6126                                struct lfsck_instance *lfsck,
6127                                const struct lu_fid *fid, int idx)
6128 {
6129         struct seq_server_site  *ss;
6130         struct lu_server_fld    *sf;
6131         struct lu_seq_range     *range = &lfsck_env_info(env)->lti_range;
6132         int                      rc;
6133
6134         /* All abnormal cases will be returned to MDT0. */
6135         if (!fid_is_norm(fid)) {
6136                 if (idx == 0)
6137                         return 1;
6138
6139                 return 0;
6140         }
6141
6142         ss = lfsck_dev_site(lfsck);
6143         if (unlikely(ss == NULL))
6144                 return -ENOTCONN;
6145
6146         sf = ss->ss_server_fld;
6147         LASSERT(sf != NULL);
6148
6149         fld_range_set_any(range);
6150         rc = fld_server_lookup(env, sf, fid_seq(fid), range);
6151         if (rc != 0)
6152                 return rc;
6153
6154         if (!fld_range_is_mdt(range))
6155                 return -EINVAL;
6156
6157         if (range->lsr_index == idx)
6158                 return 1;
6159
6160         return 0;
6161 }
6162
6163 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
6164                                         struct dt_object *obj)
6165 {
6166         struct dt_device        *dev    = lfsck_obj2dev(obj);
6167         struct thandle          *handle;
6168         int                      rc;
6169         ENTRY;
6170
6171         handle = dt_trans_create(env, dev);
6172         if (IS_ERR(handle))
6173                 RETURN_EXIT;
6174
6175         rc = dt_declare_ref_del(env, obj, handle);
6176         if (rc != 0)
6177                 GOTO(stop, rc);
6178
6179         rc = dt_declare_destroy(env, obj, handle);
6180         if (rc != 0)
6181                 GOTO(stop, rc);
6182
6183         rc = dt_trans_start_local(env, dev, handle);
6184         if (rc != 0)
6185                 GOTO(stop, rc);
6186
6187         dt_write_lock(env, obj, 0);
6188         rc = dt_ref_del(env, obj, handle);
6189         if (rc == 0)
6190                 rc = dt_destroy(env, obj, handle);
6191         dt_write_unlock(env, obj);
6192
6193         GOTO(stop, rc);
6194
6195 stop:
6196         dt_trans_stop(env, dev, handle);
6197
6198         CDEBUG(D_LFSCK, "destroy orphan OST-object "DFID": rc = %d\n",
6199                PFID(lfsck_dto2fid(obj)), rc);
6200
6201         RETURN_EXIT;
6202 }
6203
6204 static int lfsck_orphan_index_lookup(const struct lu_env *env,
6205                                      struct dt_object *dt,
6206                                      struct dt_rec *rec,
6207                                      const struct dt_key *key)
6208 {
6209         return -EOPNOTSUPP;
6210 }
6211
6212 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
6213                                              struct dt_object *dt,
6214                                              const struct dt_rec *rec,
6215                                              const struct dt_key *key,
6216                                              struct thandle *handle)
6217 {
6218         return -EOPNOTSUPP;
6219 }
6220
6221 static int lfsck_orphan_index_insert(const struct lu_env *env,
6222                                      struct dt_object *dt,
6223                                      const struct dt_rec *rec,
6224                                      const struct dt_key *key,
6225                                      struct thandle *handle,
6226                                      int ignore_quota)
6227 {
6228         return -EOPNOTSUPP;
6229 }
6230
6231 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
6232                                              struct dt_object *dt,
6233                                              const struct dt_key *key,
6234                                              struct thandle *handle)
6235 {
6236         return -EOPNOTSUPP;
6237 }
6238
6239 static int lfsck_orphan_index_delete(const struct lu_env *env,
6240                                      struct dt_object *dt,
6241                                      const struct dt_key *key,
6242                                      struct thandle *handle)
6243 {
6244         return -EOPNOTSUPP;
6245 }
6246
6247 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
6248                                           struct dt_object *dt,
6249                                           __u32 attr)
6250 {
6251         struct dt_device                *dev    = lu2dt_dev(dt->do_lu.lo_dev);
6252         struct lfsck_instance           *lfsck;
6253         struct lfsck_component          *com    = NULL;
6254         struct lfsck_layout_slave_data  *llsd;
6255         struct lfsck_orphan_it          *it     = NULL;
6256         struct lfsck_layout             *lo;
6257         int                              rc     = 0;
6258         ENTRY;
6259
6260         lfsck = lfsck_instance_find(dev, true, false);
6261         if (unlikely(lfsck == NULL))
6262                 RETURN(ERR_PTR(-ENXIO));
6263
6264         com = lfsck_component_find(lfsck, LFSCK_TYPE_LAYOUT);
6265         if (unlikely(com == NULL))
6266                 GOTO(out, rc = -ENOENT);
6267
6268         lo = com->lc_file_ram;
6269         if (lo->ll_flags & LF_INCOMPLETE)
6270                 GOTO(out, rc = -ESRCH);
6271
6272         llsd = com->lc_data;
6273         if (!llsd->llsd_rbtree_valid)
6274                 GOTO(out, rc = -ESRCH);
6275
6276         OBD_ALLOC_PTR(it);
6277         if (it == NULL)
6278                 GOTO(out, rc = -ENOMEM);
6279
6280         it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
6281         if (it->loi_llst == NULL)
6282                 GOTO(out, rc = -ENXIO);
6283
6284         if (dev->dd_record_fid_accessed) {
6285                 /* The first iteration against the rbtree, scan the whole rbtree
6286                  * to remove the nodes which do NOT need to be handled. */
6287                 write_lock(&llsd->llsd_rb_lock);
6288                 if (dev->dd_record_fid_accessed) {
6289                         struct rb_node                  *node;
6290                         struct rb_node                  *next;
6291                         struct lfsck_rbtree_node        *lrn;
6292
6293                         /* No need to record the fid accessing anymore. */
6294                         dev->dd_record_fid_accessed = 0;
6295
6296                         node = rb_first(&llsd->llsd_rb_root);
6297                         while (node != NULL) {
6298                                 next = rb_next(node);
6299                                 lrn = rb_entry(node, struct lfsck_rbtree_node,
6300                                                lrn_node);
6301                                 if (atomic_read(&lrn->lrn_known_count) <=
6302                                     atomic_read(&lrn->lrn_accessed_count)) {
6303                                         rb_erase(node, &llsd->llsd_rb_root);
6304                                         lfsck_rbtree_free(lrn);
6305                                 }
6306                                 node = next;
6307                         }
6308                 }
6309                 write_unlock(&llsd->llsd_rb_lock);
6310         }
6311
6312         /* read lock the rbtree when init, and unlock when fini */
6313         read_lock(&llsd->llsd_rb_lock);
6314         it->loi_com = com;
6315         com = NULL;
6316
6317         GOTO(out, rc = 0);
6318
6319 out:
6320         if (com != NULL)
6321                 lfsck_component_put(env, com);
6322
6323         CDEBUG(D_LFSCK, "%s: init the orphan iteration: rc = %d\n",
6324                lfsck_lfsck2name(lfsck), rc);
6325
6326         lfsck_instance_put(env, lfsck);
6327         if (rc != 0) {
6328                 if (it != NULL)
6329                         OBD_FREE_PTR(it);
6330
6331                 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
6332         }
6333
6334         return (struct dt_it *)it;
6335 }
6336
6337 static void lfsck_orphan_it_fini(const struct lu_env *env,
6338                                  struct dt_it *di)
6339 {
6340         struct lfsck_orphan_it           *it    = (struct lfsck_orphan_it *)di;
6341         struct lfsck_component           *com   = it->loi_com;
6342         struct lfsck_layout_slave_data   *llsd;
6343         struct lfsck_layout_slave_target *llst;
6344
6345         if (com != NULL) {
6346                 CDEBUG(D_LFSCK, "%s: fini the orphan iteration\n",
6347                        lfsck_lfsck2name(com->lc_lfsck));
6348
6349                 llsd = com->lc_data;
6350                 read_unlock(&llsd->llsd_rb_lock);
6351                 llst = it->loi_llst;
6352                 LASSERT(llst != NULL);
6353
6354                 /* Save the key and hash for iterate next. */
6355                 llst->llst_fid = it->loi_key;
6356                 llst->llst_hash = it->loi_hash;
6357                 lfsck_layout_llst_put(llst);
6358                 lfsck_component_put(env, com);
6359         }
6360         OBD_FREE_PTR(it);
6361 }
6362
6363 /**
6364  * \retval       +1: the iteration finished
6365  * \retval        0: on success, not finished
6366  * \retval      -ve: on error
6367  */
6368 static int lfsck_orphan_it_next(const struct lu_env *env,
6369                                 struct dt_it *di)
6370 {
6371         struct lfsck_thread_info        *info   = lfsck_env_info(env);
6372         struct filter_fid_old           *pfid   = &info->lti_old_pfid;
6373         struct lu_attr                  *la     = &info->lti_la;
6374         struct lfsck_orphan_it          *it     = (struct lfsck_orphan_it *)di;
6375         struct lu_fid                   *key    = &it->loi_key;
6376         struct lu_orphan_rec            *rec    = &it->loi_rec;
6377         struct lfsck_component          *com    = it->loi_com;
6378         struct lfsck_instance           *lfsck  = com->lc_lfsck;
6379         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
6380         struct dt_object                *obj;
6381         struct lfsck_rbtree_node        *lrn;
6382         int                              pos;
6383         int                              rc;
6384         __u32                            save;
6385         __u32                            idx    = it->loi_llst->llst_index;
6386         bool                             exact  = false;
6387         ENTRY;
6388
6389         if (it->loi_over)
6390                 RETURN(1);
6391
6392 again0:
6393         lrn = it->loi_lrn;
6394         if (lrn == NULL) {
6395                 lrn = lfsck_rbtree_search(llsd, key, &exact);
6396                 if (lrn == NULL) {
6397                         it->loi_over = 1;
6398                         RETURN(1);
6399                 }
6400
6401                 it->loi_lrn = lrn;
6402                 if (!exact) {
6403                         key->f_seq = lrn->lrn_seq;
6404                         key->f_oid = lrn->lrn_first_oid;
6405                         key->f_ver = 0;
6406                 }
6407         } else {
6408                 key->f_oid++;
6409                 if (unlikely(key->f_oid == 0)) {
6410                         key->f_seq++;
6411                         it->loi_lrn = NULL;
6412                         goto again0;
6413                 }
6414
6415                 if (key->f_oid >=
6416                     lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
6417                         it->loi_lrn = NULL;
6418                         goto again0;
6419                 }
6420         }
6421
6422         if (unlikely(atomic_read(&lrn->lrn_known_count) <=
6423                      atomic_read(&lrn->lrn_accessed_count))) {
6424                 struct rb_node *next = rb_next(&lrn->lrn_node);
6425
6426                 while (next != NULL) {
6427                         lrn = rb_entry(next, struct lfsck_rbtree_node,
6428                                        lrn_node);
6429                         if (atomic_read(&lrn->lrn_known_count) >
6430                             atomic_read(&lrn->lrn_accessed_count))
6431                                 break;
6432                         next = rb_next(next);
6433                 }
6434
6435                 if (next == NULL) {
6436                         it->loi_over = 1;
6437                         RETURN(1);
6438                 }
6439
6440                 it->loi_lrn = lrn;
6441                 key->f_seq = lrn->lrn_seq;
6442                 key->f_oid = lrn->lrn_first_oid;
6443                 key->f_ver = 0;
6444         }
6445
6446         pos = key->f_oid - lrn->lrn_first_oid;
6447
6448 again1:
6449         pos = find_next_bit(lrn->lrn_known_bitmap,
6450                             LFSCK_RBTREE_BITMAP_WIDTH, pos);
6451         if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
6452                 key->f_oid = lrn->lrn_first_oid + pos;
6453                 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
6454                         key->f_seq++;
6455                         key->f_oid = 0;
6456                 }
6457                 it->loi_lrn = NULL;
6458                 goto again0;
6459         }
6460
6461         if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
6462                 pos++;
6463                 goto again1;
6464         }
6465
6466         key->f_oid = lrn->lrn_first_oid + pos;
6467         obj = lfsck_object_find_bottom(env, lfsck, key);
6468         if (IS_ERR(obj)) {
6469                 rc = PTR_ERR(obj);
6470                 if (rc == -ENOENT) {
6471                         pos++;
6472                         goto again1;
6473                 }
6474                 RETURN(rc);
6475         }
6476
6477         dt_read_lock(env, obj, 0);
6478         if (dt_object_exists(obj) == 0 ||
6479             lfsck_is_dead_obj(obj)) {
6480                 dt_read_unlock(env, obj);
6481                 lfsck_object_put(env, obj);
6482                 pos++;
6483                 goto again1;
6484         }
6485
6486         rc = dt_attr_get(env, obj, la);
6487         if (rc != 0)
6488                 GOTO(out, rc);
6489
6490         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, pfid, sizeof(*pfid)),
6491                           XATTR_NAME_FID);
6492         if (rc == -ENODATA) {
6493                 /* For the pre-created OST-object, update the bitmap to avoid
6494                  * others LFSCK (second phase) iteration to touch it again. */
6495                 if (la->la_ctime == 0) {
6496                         if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
6497                                 atomic_inc(&lrn->lrn_accessed_count);
6498
6499                         /* For the race between repairing dangling referenced
6500                          * MDT-object and unlink the file, it may left orphan
6501                          * OST-object there. Destroy it now! */
6502                         if (unlikely(!(la->la_mode & S_ISUID))) {
6503                                 dt_read_unlock(env, obj);
6504                                 lfsck_layout_destroy_orphan(env, obj);
6505                                 lfsck_object_put(env, obj);
6506                                 pos++;
6507                                 goto again1;
6508                         }
6509                 } else if (idx == 0) {
6510                         /* If the orphan OST-object has no parent information,
6511                          * regard it as referenced by the MDT-object on MDT0. */
6512                         fid_zero(&rec->lor_fid);
6513                         rec->lor_uid = la->la_uid;
6514                         rec->lor_gid = la->la_gid;
6515                         GOTO(out, rc = 0);
6516                 }
6517
6518                 dt_read_unlock(env, obj);
6519                 lfsck_object_put(env, obj);
6520                 pos++;
6521                 goto again1;
6522         }
6523
6524         if (rc < 0)
6525                 GOTO(out, rc);
6526
6527         if (rc != sizeof(struct filter_fid) &&
6528             rc != sizeof(struct filter_fid_old))
6529                 GOTO(out, rc = -EINVAL);
6530
6531         fid_le_to_cpu(&rec->lor_fid, &pfid->ff_parent);
6532         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
6533          * MDT-object's FID::f_ver, instead it is the OST-object index in its
6534          * parent MDT-object's layout EA. */
6535         save = rec->lor_fid.f_stripe_idx;
6536         rec->lor_fid.f_ver = 0;
6537         rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_fid, idx);
6538         /* If the orphan OST-object does not claim the MDT, then next.
6539          *
6540          * If we do not know whether it matches or not, then return it
6541          * to the MDT for further check. */
6542         if (rc == 0) {
6543                 dt_read_unlock(env, obj);
6544                 lfsck_object_put(env, obj);
6545                 pos++;
6546                 goto again1;
6547         }
6548
6549         rec->lor_fid.f_stripe_idx = save;
6550         rec->lor_uid = la->la_uid;
6551         rec->lor_gid = la->la_gid;
6552
6553         CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u\n",
6554                lfsck_lfsck2name(com->lc_lfsck), PFID(key), PFID(&rec->lor_fid),
6555                rec->lor_uid, rec->lor_gid);
6556
6557         GOTO(out, rc = 0);
6558
6559 out:
6560         dt_read_unlock(env, obj);
6561         lfsck_object_put(env, obj);
6562         if (rc == 0)
6563                 it->loi_hash++;
6564
6565         return rc;
6566 }
6567
6568 /**
6569  * \retval       +1: locate to the exactly position
6570  * \retval        0: cannot locate to the exactly position,
6571  *                   call next() to move to a valid position.
6572  * \retval      -ve: on error
6573  */
6574 static int lfsck_orphan_it_get(const struct lu_env *env,
6575                                struct dt_it *di,
6576                                const struct dt_key *key)
6577 {
6578         struct lfsck_orphan_it  *it   = (struct lfsck_orphan_it *)di;
6579         int                      rc;
6580
6581         it->loi_key = *(struct lu_fid *)key;
6582         rc = lfsck_orphan_it_next(env, di);
6583         if (rc == 1)
6584                 return 0;
6585
6586         if (rc == 0)
6587                 return 1;
6588
6589         return rc;
6590 }
6591
6592 static void lfsck_orphan_it_put(const struct lu_env *env,
6593                                 struct dt_it *di)
6594 {
6595 }
6596
6597 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
6598                                           const struct dt_it *di)
6599 {
6600         struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6601
6602         return (struct dt_key *)&it->loi_key;
6603 }
6604
6605 static int lfsck_orphan_it_key_size(const struct lu_env *env,
6606                                     const struct dt_it *di)
6607 {
6608         return sizeof(struct lu_fid);
6609 }
6610
6611 static int lfsck_orphan_it_rec(const struct lu_env *env,
6612                                const struct dt_it *di,
6613                                struct dt_rec *rec,
6614                                __u32 attr)
6615 {
6616         struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6617
6618         *(struct lu_orphan_rec *)rec = it->loi_rec;
6619
6620         return 0;
6621 }
6622
6623 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
6624                                    const struct dt_it *di)
6625 {
6626         struct lfsck_orphan_it  *it   = (struct lfsck_orphan_it *)di;
6627
6628         return it->loi_hash;
6629 }
6630
6631 /**
6632  * \retval       +1: locate to the exactly position
6633  * \retval        0: cannot locate to the exactly position,
6634  *                   call next() to move to a valid position.
6635  * \retval      -ve: on error
6636  */
6637 static int lfsck_orphan_it_load(const struct lu_env *env,
6638                                 const struct dt_it *di,
6639                                 __u64 hash)
6640 {
6641         struct lfsck_orphan_it           *it   = (struct lfsck_orphan_it *)di;
6642         struct lfsck_layout_slave_target *llst = it->loi_llst;
6643         int                               rc;
6644
6645         LASSERT(llst != NULL);
6646
6647         if (hash != llst->llst_hash) {
6648                 CDEBUG(D_LFSCK, "%s: the given hash %llu for orphan "
6649                        "iteration does not match the one when fini "
6650                        "%llu, to be reset.\n",
6651                        lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
6652                        llst->llst_hash);
6653                 fid_zero(&llst->llst_fid);
6654                 llst->llst_hash = 0;
6655         }
6656
6657         it->loi_key = llst->llst_fid;
6658         it->loi_hash = llst->llst_hash;
6659         rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
6660         if (rc == 1)
6661                 return 0;
6662
6663         if (rc == 0)
6664                 return 1;
6665
6666         return rc;
6667 }
6668
6669 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
6670                                    const struct dt_it *di,
6671                                    void *key_rec)
6672 {
6673         return 0;
6674 }
6675
6676 const struct dt_index_operations lfsck_orphan_index_ops = {
6677         .dio_lookup             = lfsck_orphan_index_lookup,
6678         .dio_declare_insert     = lfsck_orphan_index_declare_insert,
6679         .dio_insert             = lfsck_orphan_index_insert,
6680         .dio_declare_delete     = lfsck_orphan_index_declare_delete,
6681         .dio_delete             = lfsck_orphan_index_delete,
6682         .dio_it = {
6683                 .init           = lfsck_orphan_it_init,
6684                 .fini           = lfsck_orphan_it_fini,
6685                 .get            = lfsck_orphan_it_get,
6686                 .put            = lfsck_orphan_it_put,
6687                 .next           = lfsck_orphan_it_next,
6688                 .key            = lfsck_orphan_it_key,
6689                 .key_size       = lfsck_orphan_it_key_size,
6690                 .rec            = lfsck_orphan_it_rec,
6691                 .store          = lfsck_orphan_it_store,
6692                 .load           = lfsck_orphan_it_load,
6693                 .key_rec        = lfsck_orphan_it_key_rec,
6694         }
6695 };