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) 2013, 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              0xB173AE14
  52
  53 static const char lfsck_layout_name[] = "lfsck_layout";
  54
  55 struct lfsck_layout_seq {
  56         struct list_head         lls_list;
  57         __u64                    lls_seq;
  58         __u64                    lls_lastid;
  59         __u64                    lls_lastid_known;
  60         struct dt_object        *lls_lastid_obj;
  61         unsigned int             lls_dirty:1;
  62 };
  63
  64 struct lfsck_layout_slave_target {
  65         /* link into lfsck_layout_slave_data::llsd_master_list. */
  66         struct list_head        llst_list;
  67         /* The position for next record in the rbtree for iteration. */
  68         struct lu_fid           llst_fid;
  69         /* Dummy hash for iteration against the rbtree. */
  70         __u64                   llst_hash;
  71         __u64                   llst_gen;
  72         atomic_t                llst_ref;
  73         __u32                   llst_index;
  74 };
  75
  76 struct lfsck_layout_slave_data {
  77         /* list for lfsck_layout_seq */
  78         struct list_head         llsd_seq_list;
  79
  80         /* list for the masters involve layout verification. */
  81         struct list_head         llsd_master_list;
  82         spinlock_t               llsd_lock;
  83         __u64                    llsd_touch_gen;
  84         struct dt_object        *llsd_rb_obj;
  85         struct rb_root           llsd_rb_root;
  86         rwlock_t                 llsd_rb_lock;
  87         unsigned int             llsd_rbtree_valid:1;
  88 };
  89
  90 struct lfsck_layout_object {
  91         struct dt_object        *llo_obj;
  92         struct lu_attr           llo_attr;
  93         atomic_t                 llo_ref;
  94         __u16                    llo_gen;
  95 };
  96
  97 struct lfsck_layout_req {
  98         struct list_head                 llr_list;
  99         struct lfsck_layout_object      *llr_parent;
 100         struct dt_object                *llr_child;
 101         __u32                            llr_ost_idx;
 102         __u32                            llr_lov_idx; /* offset in LOV EA */
 103 };
 104
 105 struct lfsck_layout_master_data {
 106         spinlock_t              llmd_lock;
 107         struct list_head        llmd_req_list;
 108
 109         /* list for the ost targets involve layout verification. */
 110         struct list_head        llmd_ost_list;
 111
 112         /* list for the ost targets in phase1 scanning. */
 113         struct list_head        llmd_ost_phase1_list;
 114
 115         /* list for the ost targets in phase1 scanning. */
 116         struct list_head        llmd_ost_phase2_list;
 117
 118         /* list for the mdt targets involve layout verification. */
 119         struct list_head        llmd_mdt_list;
 120
 121         /* list for the mdt targets in phase1 scanning. */
 122         struct list_head        llmd_mdt_phase1_list;
 123
 124         /* list for the mdt targets in phase1 scanning. */
 125         struct list_head        llmd_mdt_phase2_list;
 126
 127         struct ptlrpc_thread    llmd_thread;
 128         __u32                   llmd_touch_gen;
 129         int                     llmd_prefetched;
 130         int                     llmd_assistant_status;
 131         int                     llmd_post_result;
 132         unsigned int            llmd_to_post:1,
 133                                 llmd_to_double_scan:1,
 134                                 llmd_in_double_scan:1,
 135                                 llmd_exit:1;
 136 };
 137
 138 struct lfsck_layout_slave_async_args {
 139         struct obd_export                *llsaa_exp;
 140         struct lfsck_component           *llsaa_com;
 141         struct lfsck_layout_slave_target *llsaa_llst;
 142 };
 143
 144 static struct lfsck_layout_object *
 145 lfsck_layout_object_init(const struct lu_env *env, struct dt_object *obj,
 146                          __u16 gen)
 147 {
 148         struct lfsck_layout_object *llo;
 149         int                         rc;
 150
 151         OBD_ALLOC_PTR(llo);
 152         if (llo == NULL)
 153                 return ERR_PTR(-ENOMEM);
 154
 155         rc = dt_attr_get(env, obj, &llo->llo_attr, BYPASS_CAPA);
 156         if (rc != 0) {
 157                 OBD_FREE_PTR(llo);
 158
 159                 return ERR_PTR(rc);
 160         }
 161
 162         lu_object_get(&obj->do_lu);
 163         llo->llo_obj = obj;
 164         /* The gen can be used to check whether some others have changed the
 165          * file layout after LFSCK pre-fetching but before real verification. */
 166         llo->llo_gen = gen;
 167         atomic_set(&llo->llo_ref, 1);
 168
 169         return llo;
 170 }
 171
 172 static inline void
 173 lfsck_layout_llst_put(struct lfsck_layout_slave_target *llst)
 174 {
 175         if (atomic_dec_and_test(&llst->llst_ref)) {
 176                 LASSERT(list_empty(&llst->llst_list));
 177
 178                 OBD_FREE_PTR(llst);
 179         }
 180 }
 181
 182 static inline int
 183 lfsck_layout_llst_add(struct lfsck_layout_slave_data *llsd, __u32 index)
 184 {
 185         struct lfsck_layout_slave_target *llst;
 186         struct lfsck_layout_slave_target *tmp;
 187         int                               rc   = 0;
 188
 189         OBD_ALLOC_PTR(llst);
 190         if (llst == NULL)
 191                 return -ENOMEM;
 192
 193         INIT_LIST_HEAD(&llst->llst_list);
 194         llst->llst_gen = 0;
 195         llst->llst_index = index;
 196         atomic_set(&llst->llst_ref, 1);
 197
 198         spin_lock(&llsd->llsd_lock);
 199         list_for_each_entry(tmp, &llsd->llsd_master_list, llst_list) {
 200                 if (tmp->llst_index == index) {
 201                         rc = -EALREADY;
 202                         break;
 203                 }
 204         }
 205         if (rc == 0)
 206                 list_add_tail(&llst->llst_list, &llsd->llsd_master_list);
 207         spin_unlock(&llsd->llsd_lock);
 208
 209         if (rc != 0)
 210                 OBD_FREE_PTR(llst);
 211
 212         return rc;
 213 }
 214
 215 static inline void
 216 lfsck_layout_llst_del(struct lfsck_layout_slave_data *llsd,
 217                       struct lfsck_layout_slave_target *llst)
 218 {
 219         bool del = false;
 220
 221         spin_lock(&llsd->llsd_lock);
 222         if (!list_empty(&llst->llst_list)) {
 223                 list_del_init(&llst->llst_list);
 224                 del = true;
 225         }
 226         spin_unlock(&llsd->llsd_lock);
 227
 228         if (del)
 229                 lfsck_layout_llst_put(llst);
 230 }
 231
 232 static inline struct lfsck_layout_slave_target *
 233 lfsck_layout_llst_find_and_del(struct lfsck_layout_slave_data *llsd,
 234                                __u32 index, bool unlink)
 235 {
 236         struct lfsck_layout_slave_target *llst;
 237
 238         spin_lock(&llsd->llsd_lock);
 239         list_for_each_entry(llst, &llsd->llsd_master_list, llst_list) {
 240                 if (llst->llst_index == index) {
 241                         if (unlink)
 242                                 list_del_init(&llst->llst_list);
 243                         else
 244                                 atomic_inc(&llst->llst_ref);
 245                         spin_unlock(&llsd->llsd_lock);
 246
 247                         return llst;
 248                 }
 249         }
 250         spin_unlock(&llsd->llsd_lock);
 251
 252         return NULL;
 253 }
 254
 255 static inline void lfsck_layout_object_put(const struct lu_env *env,
 256                                            struct lfsck_layout_object *llo)
 257 {
 258         if (atomic_dec_and_test(&llo->llo_ref)) {
 259                 lfsck_object_put(env, llo->llo_obj);
 260                 OBD_FREE_PTR(llo);
 261         }
 262 }
 263
 264 static struct lfsck_layout_req *
 265 lfsck_layout_req_init(struct lfsck_layout_object *parent,
 266                       struct dt_object *child, __u32 ost_idx, __u32 lov_idx)
 267 {
 268         struct lfsck_layout_req *llr;
 269
 270         OBD_ALLOC_PTR(llr);
 271         if (llr == NULL)
 272                 return ERR_PTR(-ENOMEM);
 273
 274         INIT_LIST_HEAD(&llr->llr_list);
 275         atomic_inc(&parent->llo_ref);
 276         llr->llr_parent = parent;
 277         llr->llr_child = child;
 278         llr->llr_ost_idx = ost_idx;
 279         llr->llr_lov_idx = lov_idx;
 280
 281         return llr;
 282 }
 283
 284 static inline void lfsck_layout_req_fini(const struct lu_env *env,
 285                                          struct lfsck_layout_req *llr)
 286 {
 287         lu_object_put(env, &llr->llr_child->do_lu);
 288         lfsck_layout_object_put(env, llr->llr_parent);
 289         OBD_FREE_PTR(llr);
 290 }
 291
 292 static inline bool lfsck_layout_req_empty(struct lfsck_layout_master_data *llmd)
 293 {
 294         bool empty = false;
 295
 296         spin_lock(&llmd->llmd_lock);
 297         if (list_empty(&llmd->llmd_req_list))
 298                 empty = true;
 299         spin_unlock(&llmd->llmd_lock);
 300
 301         return empty;
 302 }
 303
 304 static int lfsck_layout_get_lovea(const struct lu_env *env,
 305                                   struct dt_object *obj,
 306                                   struct lu_buf *buf, ssize_t *buflen)
 307 {
 308         int rc;
 309
 310 again:
 311         rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LOV, BYPASS_CAPA);
 312         if (rc == -ERANGE) {
 313                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV,
 314                                   BYPASS_CAPA);
 315                 if (rc <= 0)
 316                         return rc;
 317
 318                 lu_buf_realloc(buf, rc);
 319                 if (buflen != NULL)
 320                         *buflen = buf->lb_len;
 321
 322                 if (buf->lb_buf == NULL)
 323                         return -ENOMEM;
 324
 325                 goto again;
 326         }
 327
 328         if (rc == -ENODATA)
 329                 rc = 0;
 330
 331         if (rc <= 0)
 332                 return rc;
 333
 334         if (unlikely(buf->lb_buf == NULL)) {
 335                 lu_buf_alloc(buf, rc);
 336                 if (buflen != NULL)
 337                         *buflen = buf->lb_len;
 338
 339                 if (buf->lb_buf == NULL)
 340                         return -ENOMEM;
 341
 342                 goto again;
 343         }
 344
 345         return rc;
 346 }
 347
 348 static int lfsck_layout_verify_header(struct lov_mds_md_v1 *lmm)
 349 {
 350         __u32 magic;
 351         __u32 pattern;
 352
 353         magic = le32_to_cpu(lmm->lmm_magic);
 354         /* If magic crashed, keep it there. Sometime later, during OST-object
 355          * orphan handling, if some OST-object(s) back-point to it, it can be
 356          * verified and repaired. */
 357         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) {
 358                 struct ost_id   oi;
 359                 int             rc;
 360
 361                 lmm_oi_cpu_to_le(&oi, &lmm->lmm_oi);
 362                 if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
 363                         rc = -EOPNOTSUPP;
 364                 else
 365                         rc = -EINVAL;
 366
 367                 CDEBUG(D_LFSCK, "%s LOV EA magic %u on "DOSTID"\n",
 368                        rc == -EINVAL ? "Unknown" : "Unsupported",
 369                        magic, POSTID(&oi));
 370
 371                 return rc;
 372         }
 373
 374         pattern = le32_to_cpu(lmm->lmm_pattern);
 375         /* XXX: currently, we only support LOV_PATTERN_RAID0. */
 376         if (lov_pattern(pattern) != LOV_PATTERN_RAID0) {
 377                 struct ost_id oi;
 378
 379                 lmm_oi_cpu_to_le(&oi, &lmm->lmm_oi);
 380                 CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u on "DOSTID"\n",
 381                        pattern, POSTID(&oi));
 382
 383                 return -EOPNOTSUPP;
 384         }
 385
 386         return 0;
 387 }
 388
 389 #define LFSCK_RBTREE_BITMAP_SIZE        PAGE_CACHE_SIZE
 390 #define LFSCK_RBTREE_BITMAP_WIDTH       (LFSCK_RBTREE_BITMAP_SIZE << 3)
 391 #define LFSCK_RBTREE_BITMAP_MASK        (LFSCK_RBTREE_BITMAP_WIDTH - 1)
 392
 393 struct lfsck_rbtree_node {
 394         struct rb_node   lrn_node;
 395         __u64            lrn_seq;
 396         __u32            lrn_first_oid;
 397         atomic_t         lrn_known_count;
 398         atomic_t         lrn_accessed_count;
 399         void            *lrn_known_bitmap;
 400         void            *lrn_accessed_bitmap;
 401 };
 402
 403 static inline int lfsck_rbtree_cmp(struct lfsck_rbtree_node *lrn,
 404                                    __u64 seq, __u32 oid)
 405 {
 406         if (seq < lrn->lrn_seq)
 407                 return -1;
 408
 409         if (seq > lrn->lrn_seq)
 410                 return 1;
 411
 412         if (oid < lrn->lrn_first_oid)
 413                 return -1;
 414
 415         if (oid - lrn->lrn_first_oid >= LFSCK_RBTREE_BITMAP_WIDTH)
 416                 return 1;
 417
 418         return 0;
 419 }
 420
 421 /* The caller should hold llsd->llsd_rb_lock. */
 422 static struct lfsck_rbtree_node *
 423 lfsck_rbtree_search(struct lfsck_layout_slave_data *llsd,
 424                     const struct lu_fid *fid, bool *exact)
 425 {
 426         struct rb_node           *node  = llsd->llsd_rb_root.rb_node;
 427         struct rb_node           *prev  = NULL;
 428         struct lfsck_rbtree_node *lrn   = NULL;
 429         int                       rc    = 0;
 430
 431         if (exact != NULL)
 432                 *exact = true;
 433
 434         while (node != NULL) {
 435                 prev = node;
 436                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 437                 rc = lfsck_rbtree_cmp(lrn, fid_seq(fid), fid_oid(fid));
 438                 if (rc < 0)
 439                         node = node->rb_left;
 440                 else if (rc > 0)
 441                         node = node->rb_right;
 442                 else
 443                         return lrn;
 444         }
 445
 446         if (exact == NULL)
 447                 return NULL;
 448
 449         /* If there is no exactly matched one, then to the next valid one. */
 450         *exact = false;
 451
 452         /* The rbtree is empty. */
 453         if (rc == 0)
 454                 return NULL;
 455
 456         if (rc < 0)
 457                 return lrn;
 458
 459         node = rb_next(prev);
 460
 461         /* The end of the rbtree. */
 462         if (node == NULL)
 463                 return NULL;
 464
 465         lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 466
 467         return lrn;
 468 }
 469
 470 static struct lfsck_rbtree_node *lfsck_rbtree_new(const struct lu_env *env,
 471                                                   const struct lu_fid *fid)
 472 {
 473         struct lfsck_rbtree_node *lrn;
 474
 475         OBD_ALLOC_PTR(lrn);
 476         if (lrn == NULL)
 477                 return ERR_PTR(-ENOMEM);
 478
 479         OBD_ALLOC(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 480         if (lrn->lrn_known_bitmap == NULL) {
 481                 OBD_FREE_PTR(lrn);
 482
 483                 return ERR_PTR(-ENOMEM);
 484         }
 485
 486         OBD_ALLOC(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 487         if (lrn->lrn_accessed_bitmap == NULL) {
 488                 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 489                 OBD_FREE_PTR(lrn);
 490
 491                 return ERR_PTR(-ENOMEM);
 492         }
 493
 494         RB_CLEAR_NODE(&lrn->lrn_node);
 495         lrn->lrn_seq = fid_seq(fid);
 496         lrn->lrn_first_oid = fid_oid(fid) & ~LFSCK_RBTREE_BITMAP_MASK;
 497         atomic_set(&lrn->lrn_known_count, 0);
 498         atomic_set(&lrn->lrn_accessed_count, 0);
 499
 500         return lrn;
 501 }
 502
 503 static void lfsck_rbtree_free(struct lfsck_rbtree_node *lrn)
 504 {
 505         OBD_FREE(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 506         OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 507         OBD_FREE_PTR(lrn);
 508 }
 509
 510 /* The caller should hold lock. */
 511 static struct lfsck_rbtree_node *
 512 lfsck_rbtree_insert(struct lfsck_layout_slave_data *llsd,
 513                     struct lfsck_rbtree_node *lrn)
 514 {
 515         struct rb_node           **pos    = &llsd->llsd_rb_root.rb_node;
 516         struct rb_node            *parent = NULL;
 517         struct lfsck_rbtree_node  *tmp;
 518         int                        rc;
 519
 520         while (*pos != NULL) {
 521                 parent = *pos;
 522                 tmp = rb_entry(parent, struct lfsck_rbtree_node, lrn_node);
 523                 rc = lfsck_rbtree_cmp(tmp, lrn->lrn_seq, lrn->lrn_first_oid);
 524                 if (rc < 0)
 525                         pos = &(*pos)->rb_left;
 526                 else if (rc > 0)
 527                         pos = &(*pos)->rb_right;
 528                 else
 529                         return tmp;
 530         }
 531
 532         rb_link_node(&lrn->lrn_node, parent, pos);
 533         rb_insert_color(&lrn->lrn_node, &llsd->llsd_rb_root);
 534
 535         return lrn;
 536 }
 537
 538 extern const struct dt_index_operations lfsck_orphan_index_ops;
 539
 540 static int lfsck_rbtree_setup(const struct lu_env *env,
 541                               struct lfsck_component *com)
 542 {
 543         struct lu_fid                   *fid    = &lfsck_env_info(env)->lti_fid;
 544         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 545         struct dt_device                *dev    = lfsck->li_bottom;
 546         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
 547         struct dt_object                *obj;
 548
 549         fid->f_seq = FID_SEQ_LAYOUT_RBTREE;
 550         fid->f_oid = lfsck_dev_idx(dev);
 551         fid->f_ver = 0;
 552         obj = dt_locate(env, dev, fid);
 553         if (IS_ERR(obj))
 554                 RETURN(PTR_ERR(obj));
 555
 556         /* Generate an in-RAM object to stand for the layout rbtree.
 557          * Scanning the layout rbtree will be via the iteration over
 558          * the object. In the future, the rbtree may be written onto
 559          * disk with the object.
 560          *
 561          * Mark the object to be as exist. */
 562         obj->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 563         obj->do_index_ops = &lfsck_orphan_index_ops;
 564         llsd->llsd_rb_obj = obj;
 565         llsd->llsd_rbtree_valid = 1;
 566         dev->dd_record_fid_accessed = 1;
 567
 568         CDEBUG(D_LFSCK, "%s: layout LFSCK init OST-objects accessing bitmap\n",
 569                lfsck_lfsck2name(lfsck));
 570
 571         return 0;
 572 }
 573
 574 static void lfsck_rbtree_cleanup(const struct lu_env *env,
 575                                  struct lfsck_component *com)
 576 {
 577         struct lfsck_instance           *lfsck = com->lc_lfsck;
 578         struct lfsck_layout_slave_data  *llsd  = com->lc_data;
 579         struct rb_node                  *node  = rb_first(&llsd->llsd_rb_root);
 580         struct rb_node                  *next;
 581         struct lfsck_rbtree_node        *lrn;
 582
 583         lfsck->li_bottom->dd_record_fid_accessed = 0;
 584         /* Invalid the rbtree, then no others will use it. */
 585         write_lock(&llsd->llsd_rb_lock);
 586         llsd->llsd_rbtree_valid = 0;
 587         write_unlock(&llsd->llsd_rb_lock);
 588
 589         while (node != NULL) {
 590                 next = rb_next(node);
 591                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 592                 rb_erase(node, &llsd->llsd_rb_root);
 593                 lfsck_rbtree_free(lrn);
 594                 node = next;
 595         }
 596
 597         if (llsd->llsd_rb_obj != NULL) {
 598                 lu_object_put(env, &llsd->llsd_rb_obj->do_lu);
 599                 llsd->llsd_rb_obj = NULL;
 600         }
 601
 602         CDEBUG(D_LFSCK, "%s: layout LFSCK fini OST-objects accessing bitmap\n",
 603                lfsck_lfsck2name(lfsck));
 604 }
 605
 606 static void lfsck_rbtree_update_bitmap(const struct lu_env *env,
 607                                        struct lfsck_component *com,
 608                                        const struct lu_fid *fid,
 609                                        bool accessed)
 610 {
 611         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
 612         struct lfsck_rbtree_node        *lrn;
 613         bool                             insert = false;
 614         int                              idx;
 615         int                              rc     = 0;
 616         ENTRY;
 617
 618         if (unlikely(!fid_is_sane(fid) || fid_is_last_id(fid)))
 619                 RETURN_EXIT;
 620
 621         if (!fid_is_idif(fid) && !fid_is_norm(fid))
 622                 RETURN_EXIT;
 623
 624         read_lock(&llsd->llsd_rb_lock);
 625         if (!llsd->llsd_rbtree_valid)
 626                 GOTO(unlock, rc = 0);
 627
 628         lrn = lfsck_rbtree_search(llsd, fid, NULL);
 629         if (lrn == NULL) {
 630                 struct lfsck_rbtree_node *tmp;
 631
 632                 LASSERT(!insert);
 633
 634                 read_unlock(&llsd->llsd_rb_lock);
 635                 tmp = lfsck_rbtree_new(env, fid);
 636                 if (IS_ERR(tmp))
 637                         GOTO(out, rc = PTR_ERR(tmp));
 638
 639                 insert = true;
 640                 write_lock(&llsd->llsd_rb_lock);
 641                 if (!llsd->llsd_rbtree_valid) {
 642                         lfsck_rbtree_free(tmp);
 643                         GOTO(unlock, rc = 0);
 644                 }
 645
 646                 lrn = lfsck_rbtree_insert(llsd, tmp);
 647                 if (lrn != tmp)
 648                         lfsck_rbtree_free(tmp);
 649         }
 650
 651         idx = fid_oid(fid) & LFSCK_RBTREE_BITMAP_MASK;
 652         /* Any accessed object must be a known object. */
 653         if (!test_and_set_bit(idx, lrn->lrn_known_bitmap))
 654                 atomic_inc(&lrn->lrn_known_count);
 655         if (accessed && !test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
 656                 atomic_inc(&lrn->lrn_accessed_count);
 657
 658         GOTO(unlock, rc = 0);
 659
 660 unlock:
 661         if (insert)
 662                 write_unlock(&llsd->llsd_rb_lock);
 663         else
 664                 read_unlock(&llsd->llsd_rb_lock);
 665 out:
 666         if (rc != 0 && accessed) {
 667                 struct lfsck_layout *lo = com->lc_file_ram;
 668
 669                 CDEBUG(D_LFSCK, "%s: fail to update OST-objects accessing "
 670                        "bitmap, and will cause incorrect LFSCK OST-object "
 671                        "handling, so disable it to cancel orphan handling "
 672                        "for related device. rc = %d\n",
 673                        lfsck_lfsck2name(com->lc_lfsck), rc);
 674
 675                 lo->ll_flags |= LF_INCOMPLETE;
 676                 lfsck_rbtree_cleanup(env, com);
 677         }
 678 }
 679
 680 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
 681                                    const struct lfsck_layout *src)
 682 {
 683         int i;
 684
 685         des->ll_magic = le32_to_cpu(src->ll_magic);
 686         des->ll_status = le32_to_cpu(src->ll_status);
 687         des->ll_flags = le32_to_cpu(src->ll_flags);
 688         des->ll_success_count = le32_to_cpu(src->ll_success_count);
 689         des->ll_run_time_phase1 = le32_to_cpu(src->ll_run_time_phase1);
 690         des->ll_run_time_phase2 = le32_to_cpu(src->ll_run_time_phase2);
 691         des->ll_time_last_complete = le64_to_cpu(src->ll_time_last_complete);
 692         des->ll_time_latest_start = le64_to_cpu(src->ll_time_latest_start);
 693         des->ll_time_last_checkpoint =
 694                                 le64_to_cpu(src->ll_time_last_checkpoint);
 695         des->ll_pos_latest_start = le64_to_cpu(src->ll_pos_latest_start);
 696         des->ll_pos_last_checkpoint = le64_to_cpu(src->ll_pos_last_checkpoint);
 697         des->ll_pos_first_inconsistent =
 698                         le64_to_cpu(src->ll_pos_first_inconsistent);
 699         des->ll_objs_checked_phase1 = le64_to_cpu(src->ll_objs_checked_phase1);
 700         des->ll_objs_failed_phase1 = le64_to_cpu(src->ll_objs_failed_phase1);
 701         des->ll_objs_checked_phase2 = le64_to_cpu(src->ll_objs_checked_phase2);
 702         des->ll_objs_failed_phase2 = le64_to_cpu(src->ll_objs_failed_phase2);
 703         for (i = 0; i < LLIT_MAX; i++)
 704                 des->ll_objs_repaired[i] =
 705                                 le64_to_cpu(src->ll_objs_repaired[i]);
 706         des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
 707 }
 708
 709 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
 710                                    const struct lfsck_layout *src)
 711 {
 712         int i;
 713
 714         des->ll_magic = cpu_to_le32(src->ll_magic);
 715         des->ll_status = cpu_to_le32(src->ll_status);
 716         des->ll_flags = cpu_to_le32(src->ll_flags);
 717         des->ll_success_count = cpu_to_le32(src->ll_success_count);
 718         des->ll_run_time_phase1 = cpu_to_le32(src->ll_run_time_phase1);
 719         des->ll_run_time_phase2 = cpu_to_le32(src->ll_run_time_phase2);
 720         des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
 721         des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
 722         des->ll_time_last_checkpoint =
 723                                 cpu_to_le64(src->ll_time_last_checkpoint);
 724         des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
 725         des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
 726         des->ll_pos_first_inconsistent =
 727                         cpu_to_le64(src->ll_pos_first_inconsistent);
 728         des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
 729         des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
 730         des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
 731         des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
 732         for (i = 0; i < LLIT_MAX; i++)
 733                 des->ll_objs_repaired[i] =
 734                                 cpu_to_le64(src->ll_objs_repaired[i]);
 735         des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
 736 }
 737
 738 /**
 739  * \retval +ve: the lfsck_layout is broken, the caller should reset it.
 740  * \retval 0: succeed.
 741  * \retval -ve: failed cases.
 742  */
 743 static int lfsck_layout_load(const struct lu_env *env,
 744                              struct lfsck_component *com)
 745 {
 746         struct lfsck_layout             *lo     = com->lc_file_ram;
 747         const struct dt_body_operations *dbo    = com->lc_obj->do_body_ops;
 748         ssize_t                          size   = com->lc_file_size;
 749         loff_t                           pos    = 0;
 750         int                              rc;
 751
 752         rc = dbo->dbo_read(env, com->lc_obj,
 753                            lfsck_buf_get(env, com->lc_file_disk, size), &pos,
 754                            BYPASS_CAPA);
 755         if (rc == 0) {
 756                 return -ENOENT;
 757         } else if (rc < 0) {
 758                 CDEBUG(D_LFSCK, "%s: failed to load lfsck_layout: rc = %d\n",
 759                        lfsck_lfsck2name(com->lc_lfsck), rc);
 760                 return rc;
 761         } else if (rc != size) {
 762                 CDEBUG(D_LFSCK, "%s: lfsck_layout size %u != %u; reset it\n",
 763                        lfsck_lfsck2name(com->lc_lfsck), rc, (unsigned int)size);
 764                 return 1;
 765         }
 766
 767         lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
 768         if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
 769                 CDEBUG(D_LFSCK, "%s: invalid lfsck_layout magic %#x != %#x, "
 770                        "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
 771                        lo->ll_magic, LFSCK_LAYOUT_MAGIC);
 772                 return 1;
 773         }
 774
 775         return 0;
 776 }
 777
 778 static int lfsck_layout_store(const struct lu_env *env,
 779                               struct lfsck_component *com)
 780 {
 781         struct dt_object         *obj           = com->lc_obj;
 782         struct lfsck_instance    *lfsck         = com->lc_lfsck;
 783         struct lfsck_layout      *lo            = com->lc_file_disk;
 784         struct thandle           *handle;
 785         ssize_t                   size          = com->lc_file_size;
 786         loff_t                    pos           = 0;
 787         int                       rc;
 788         ENTRY;
 789
 790         lfsck_layout_cpu_to_le(lo, com->lc_file_ram);
 791         handle = dt_trans_create(env, lfsck->li_bottom);
 792         if (IS_ERR(handle))
 793                 GOTO(log, rc = PTR_ERR(handle));
 794
 795         rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
 796                                      pos, handle);
 797         if (rc != 0)
 798                 GOTO(out, rc);
 799
 800         rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
 801         if (rc != 0)
 802                 GOTO(out, rc);
 803
 804         rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos,
 805                              handle);
 806
 807         GOTO(out, rc);
 808
 809 out:
 810         dt_trans_stop(env, lfsck->li_bottom, handle);
 811
 812 log:
 813         if (rc != 0)
 814                 CDEBUG(D_LFSCK, "%s: fail to store lfsck_layout: rc = %d\n",
 815                        lfsck_lfsck2name(lfsck), rc);
 816         return rc;
 817 }
 818
 819 static int lfsck_layout_init(const struct lu_env *env,
 820                              struct lfsck_component *com)
 821 {
 822         struct lfsck_layout *lo = com->lc_file_ram;
 823         int rc;
 824
 825         memset(lo, 0, com->lc_file_size);
 826         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
 827         lo->ll_status = LS_INIT;
 828         down_write(&com->lc_sem);
 829         rc = lfsck_layout_store(env, com);
 830         up_write(&com->lc_sem);
 831
 832         return rc;
 833 }
 834
 835 static int fid_is_for_ostobj(const struct lu_env *env, struct dt_device *dt,
 836                              struct dt_object *obj, const struct lu_fid *fid)
 837 {
 838         struct seq_server_site  *ss     = lu_site2seq(dt->dd_lu_dev.ld_site);
 839         struct lu_seq_range      range  = { 0 };
 840         struct lustre_mdt_attrs *lma;
 841         int                      rc;
 842
 843         fld_range_set_any(&range);
 844         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), &range);
 845         if (rc == 0) {
 846                 if (fld_range_is_ost(&range))
 847                         return 1;
 848
 849                 return 0;
 850         }
 851
 852         lma = &lfsck_env_info(env)->lti_lma;
 853         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, lma, sizeof(*lma)),
 854                           XATTR_NAME_LMA, BYPASS_CAPA);
 855         if (rc == sizeof(*lma)) {
 856                 lustre_lma_swab(lma);
 857
 858                 return lma->lma_compat & LMAC_FID_ON_OST ? 1 : 0;
 859         }
 860
 861         rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID, BYPASS_CAPA);
 862
 863         return rc > 0;
 864 }
 865
 866 static struct lfsck_layout_seq *
 867 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
 868 {
 869         struct lfsck_layout_seq *lls;
 870
 871         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
 872                 if (lls->lls_seq == seq)
 873                         return lls;
 874
 875                 if (lls->lls_seq > seq)
 876                         return NULL;
 877         }
 878
 879         return NULL;
 880 }
 881
 882 static void
 883 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
 884                         struct lfsck_layout_seq *lls)
 885 {
 886         struct lfsck_layout_seq *tmp;
 887         struct list_head        *pos = &llsd->llsd_seq_list;
 888
 889         list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
 890                 if (lls->lls_seq < tmp->lls_seq) {
 891                         pos = &tmp->lls_list;
 892                         break;
 893                 }
 894         }
 895         list_add_tail(&lls->lls_list, pos);
 896 }
 897
 898 static int
 899 lfsck_layout_lastid_create(const struct lu_env *env,
 900                            struct lfsck_instance *lfsck,
 901                            struct dt_object *obj)
 902 {
 903         struct lfsck_thread_info *info   = lfsck_env_info(env);
 904         struct lu_attr           *la     = &info->lti_la;
 905         struct dt_object_format  *dof    = &info->lti_dof;
 906         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
 907         struct dt_device         *dt     = lfsck->li_bottom;
 908         struct thandle           *th;
 909         __u64                     lastid = 0;
 910         loff_t                    pos    = 0;
 911         int                       rc;
 912         ENTRY;
 913
 914         if (bk->lb_param & LPF_DRYRUN)
 915                 return 0;
 916
 917         memset(la, 0, sizeof(*la));
 918         la->la_mode = S_IFREG |  S_IRUGO | S_IWUSR;
 919         la->la_valid = LA_MODE | LA_UID | LA_GID;
 920         dof->dof_type = dt_mode_to_dft(S_IFREG);
 921
 922         th = dt_trans_create(env, dt);
 923         if (IS_ERR(th))
 924                 GOTO(log, rc = PTR_ERR(th));
 925
 926         rc = dt_declare_create(env, obj, la, NULL, dof, th);
 927         if (rc != 0)
 928                 GOTO(stop, rc);
 929
 930         rc = dt_declare_record_write(env, obj,
 931                                      lfsck_buf_get(env, &lastid,
 932                                                    sizeof(lastid)),
 933                                      pos, th);
 934         if (rc != 0)
 935                 GOTO(stop, rc);
 936
 937         rc = dt_trans_start_local(env, dt, th);
 938         if (rc != 0)
 939                 GOTO(stop, rc);
 940
 941         dt_write_lock(env, obj, 0);
 942         if (likely(!dt_object_exists(obj))) {
 943                 rc = dt_create(env, obj, la, NULL, dof, th);
 944                 if (rc == 0)
 945                         rc = dt_record_write(env, obj,
 946                                 lfsck_buf_get(env, &lastid, sizeof(lastid)),
 947                                 &pos, th);
 948         }
 949         dt_write_unlock(env, obj);
 950
 951         GOTO(stop, rc);
 952
 953 stop:
 954         dt_trans_stop(env, dt, th);
 955
 956 log:
 957         CDEBUG(D_LFSCK, "%s: layout LFSCK will create LAST_ID for <seq> "
 958                LPX64": rc = %d\n",
 959                lfsck_lfsck2name(lfsck), fid_seq(lfsck_dto2fid(obj)), rc);
 960
 961         return rc;
 962 }
 963
 964 static int
 965 lfsck_layout_lastid_reload(const struct lu_env *env,
 966                            struct lfsck_component *com,
 967                            struct lfsck_layout_seq *lls)
 968 {
 969         __u64   lastid;
 970         loff_t  pos     = 0;
 971         int     rc;
 972
 973         dt_read_lock(env, lls->lls_lastid_obj, 0);
 974         rc = dt_record_read(env, lls->lls_lastid_obj,
 975                             lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
 976         dt_read_unlock(env, lls->lls_lastid_obj);
 977         if (unlikely(rc != 0))
 978                 return rc;
 979
 980         lastid = le64_to_cpu(lastid);
 981         if (lastid < lls->lls_lastid_known) {
 982                 struct lfsck_instance   *lfsck  = com->lc_lfsck;
 983                 struct lfsck_layout     *lo     = com->lc_file_ram;
 984
 985                 lls->lls_lastid = lls->lls_lastid_known;
 986                 lls->lls_dirty = 1;
 987                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
 988                         LASSERT(lfsck->li_out_notify != NULL);
 989
 990                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
 991                                              LE_LASTID_REBUILDING);
 992                         lo->ll_flags |= LF_CRASHED_LASTID;
 993                 }
 994         } else if (lastid >= lls->lls_lastid) {
 995                 lls->lls_lastid = lastid;
 996                 lls->lls_dirty = 0;
 997         }
 998
 999         return 0;
1000 }
1001
1002 static int
1003 lfsck_layout_lastid_store(const struct lu_env *env,
1004                           struct lfsck_component *com)
1005 {
1006         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1007         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
1008         struct dt_device                *dt     = lfsck->li_bottom;
1009         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
1010         struct lfsck_layout_seq         *lls;
1011         struct thandle                  *th;
1012         __u64                            lastid;
1013         int                              rc     = 0;
1014         int                              rc1    = 0;
1015
1016         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1017                 loff_t pos = 0;
1018
1019                 /* XXX: Add the code back if we really found related
1020                  *      inconsistent cases in the future. */
1021 #if 0
1022                 if (!lls->lls_dirty) {
1023                         /* In OFD, before the pre-creation, the LAST_ID
1024                          * file will be updated firstly, which may hide
1025                          * some potential crashed cases. For example:
1026                          *
1027                          * The old obj1's ID is higher than old LAST_ID
1028                          * but lower than the new LAST_ID, but the LFSCK
1029                          * have not touch the obj1 until the OFD updated
1030                          * the LAST_ID. So the LFSCK does not regard it
1031                          * as crashed case. But when OFD does not create
1032                          * successfully, it will set the LAST_ID as the
1033                          * real created objects' ID, then LFSCK needs to
1034                          * found related inconsistency. */
1035                         rc = lfsck_layout_lastid_reload(env, com, lls);
1036                         if (likely(!lls->lls_dirty))
1037                                 continue;
1038                 }
1039 #endif
1040
1041                 CDEBUG(D_LFSCK, "%s: layout LFSCK will sync the LAST_ID for "
1042                        "<seq> "LPX64" as <oid> "LPU64"\n",
1043                        lfsck_lfsck2name(lfsck), lls->lls_seq, lls->lls_lastid);
1044
1045                 if (bk->lb_param & LPF_DRYRUN) {
1046                         lls->lls_dirty = 0;
1047                         continue;
1048                 }
1049
1050                 th = dt_trans_create(env, dt);
1051                 if (IS_ERR(th)) {
1052                         rc1 = PTR_ERR(th);
1053                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1054                                "the LAST_ID for <seq> "LPX64"(1): rc = %d\n",
1055                                lfsck_lfsck2name(com->lc_lfsck),
1056                                lls->lls_seq, rc1);
1057                         continue;
1058                 }
1059
1060                 lastid = cpu_to_le64(lls->lls_lastid);
1061                 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1062                                              lfsck_buf_get(env, &lastid,
1063                                                            sizeof(lastid)),
1064                                              pos, th);
1065                 if (rc != 0)
1066                         goto stop;
1067
1068                 rc = dt_trans_start_local(env, dt, th);
1069                 if (rc != 0)
1070                         goto stop;
1071
1072                 dt_write_lock(env, lls->lls_lastid_obj, 0);
1073                 rc = dt_record_write(env, lls->lls_lastid_obj,
1074                                      lfsck_buf_get(env, &lastid,
1075                                      sizeof(lastid)), &pos, th);
1076                 dt_write_unlock(env, lls->lls_lastid_obj);
1077                 if (rc == 0)
1078                         lls->lls_dirty = 0;
1079
1080 stop:
1081                 dt_trans_stop(env, dt, th);
1082                 if (rc != 0) {
1083                         rc1 = rc;
1084                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1085                                "the LAST_ID for <seq> "LPX64"(2): rc = %d\n",
1086                                lfsck_lfsck2name(com->lc_lfsck),
1087                                lls->lls_seq, rc1);
1088                 }
1089         }
1090
1091         return rc1;
1092 }
1093
1094 static int
1095 lfsck_layout_lastid_load(const struct lu_env *env,
1096                          struct lfsck_component *com,
1097                          struct lfsck_layout_seq *lls)
1098 {
1099         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1100         struct lfsck_layout     *lo     = com->lc_file_ram;
1101         struct lu_fid           *fid    = &lfsck_env_info(env)->lti_fid;
1102         struct dt_object        *obj;
1103         loff_t                   pos    = 0;
1104         int                      rc;
1105         ENTRY;
1106
1107         lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck->li_bottom));
1108         obj = dt_locate(env, lfsck->li_bottom, fid);
1109         if (IS_ERR(obj))
1110                 RETURN(PTR_ERR(obj));
1111
1112         /* LAST_ID crashed, to be rebuilt */
1113         if (!dt_object_exists(obj)) {
1114                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1115                         LASSERT(lfsck->li_out_notify != NULL);
1116
1117                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1118                                              LE_LASTID_REBUILDING);
1119                         lo->ll_flags |= LF_CRASHED_LASTID;
1120
1121                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1122                             cfs_fail_val > 0) {
1123                                 struct l_wait_info lwi = LWI_TIMEOUT(
1124                                                 cfs_time_seconds(cfs_fail_val),
1125                                                 NULL, NULL);
1126
1127                                 up_write(&com->lc_sem);
1128                                 l_wait_event(lfsck->li_thread.t_ctl_waitq,
1129                                              !thread_is_running(&lfsck->li_thread),
1130                                              &lwi);
1131                                 down_write(&com->lc_sem);
1132                         }
1133                 }
1134
1135                 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1136         } else {
1137                 dt_read_lock(env, obj, 0);
1138                 rc = dt_read(env, obj,
1139                         lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1140                         &pos);
1141                 dt_read_unlock(env, obj);
1142                 if (rc != 0 && rc != sizeof(__u64))
1143                         GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1144
1145                 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1146                         LASSERT(lfsck->li_out_notify != NULL);
1147
1148                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1149                                              LE_LASTID_REBUILDING);
1150                         lo->ll_flags |= LF_CRASHED_LASTID;
1151                 }
1152
1153                 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1154                 rc = 0;
1155         }
1156
1157         GOTO(out, rc);
1158
1159 out:
1160         if (rc != 0)
1161                 lfsck_object_put(env, obj);
1162         else
1163                 lls->lls_lastid_obj = obj;
1164
1165         return rc;
1166 }
1167
1168 static void lfsck_layout_record_failure(const struct lu_env *env,
1169                                                  struct lfsck_instance *lfsck,
1170                                                  struct lfsck_layout *lo)
1171 {
1172         lo->ll_objs_failed_phase1++;
1173         if (unlikely(lo->ll_pos_first_inconsistent == 0)) {
1174                 lo->ll_pos_first_inconsistent =
1175                         lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
1176                                                         lfsck->li_di_oit);
1177
1178                 CDEBUG(D_LFSCK, "%s: layout LFSCK hit first non-repaired "
1179                        "inconsistency at the pos ["LPU64"]\n",
1180                        lfsck_lfsck2name(lfsck),
1181                        lo->ll_pos_first_inconsistent);
1182         }
1183 }
1184
1185 static int lfsck_layout_master_async_interpret(const struct lu_env *env,
1186                                                struct ptlrpc_request *req,
1187                                                void *args, int rc)
1188 {
1189         struct lfsck_async_interpret_args *laia = args;
1190         struct lfsck_component            *com  = laia->laia_com;
1191         struct lfsck_layout_master_data   *llmd = com->lc_data;
1192         struct lfsck_tgt_descs            *ltds = laia->laia_ltds;
1193         struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
1194         struct lfsck_request              *lr   = laia->laia_lr;
1195
1196         switch (lr->lr_event) {
1197         case LE_START:
1198                 if (rc != 0) {
1199                         struct lfsck_layout *lo = com->lc_file_ram;
1200
1201                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for layout "
1202                                "start: rc = %d\n",
1203                                lfsck_lfsck2name(com->lc_lfsck),
1204                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1205                                ltd->ltd_index, rc);
1206                         lo->ll_flags |= LF_INCOMPLETE;
1207                         break;
1208                 }
1209
1210                 spin_lock(&ltds->ltd_lock);
1211                 if (ltd->ltd_dead || ltd->ltd_layout_done) {
1212                         spin_unlock(&ltds->ltd_lock);
1213                         break;
1214                 }
1215
1216                 if (lr->lr_flags & LEF_TO_OST) {
1217                         if (list_empty(&ltd->ltd_layout_list))
1218                                 list_add_tail(&ltd->ltd_layout_list,
1219                                               &llmd->llmd_ost_list);
1220                         if (list_empty(&ltd->ltd_layout_phase_list))
1221                                 list_add_tail(&ltd->ltd_layout_phase_list,
1222                                               &llmd->llmd_ost_phase1_list);
1223                 } else {
1224                         if (list_empty(&ltd->ltd_layout_list))
1225                                 list_add_tail(&ltd->ltd_layout_list,
1226                                               &llmd->llmd_mdt_list);
1227                         if (list_empty(&ltd->ltd_layout_phase_list))
1228                                 list_add_tail(&ltd->ltd_layout_phase_list,
1229                                               &llmd->llmd_mdt_phase1_list);
1230                 }
1231                 spin_unlock(&ltds->ltd_lock);
1232                 break;
1233         case LE_STOP:
1234         case LE_PHASE1_DONE:
1235         case LE_PHASE2_DONE:
1236         case LE_PEER_EXIT:
1237                 if (rc != 0 && rc != -EALREADY)
1238                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for layout: "
1239                                "event = %d, rc = %d\n",
1240                                lfsck_lfsck2name(com->lc_lfsck),
1241                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1242                                ltd->ltd_index, lr->lr_event, rc);
1243                 break;
1244         case LE_QUERY: {
1245                 struct lfsck_reply *reply;
1246
1247                 if (rc != 0) {
1248                         spin_lock(&ltds->ltd_lock);
1249                         list_del_init(&ltd->ltd_layout_phase_list);
1250                         list_del_init(&ltd->ltd_layout_list);
1251                         spin_unlock(&ltds->ltd_lock);
1252                         break;
1253                 }
1254
1255                 reply = req_capsule_server_get(&req->rq_pill,
1256                                                &RMF_LFSCK_REPLY);
1257                 if (reply == NULL) {
1258                         rc = -EPROTO;
1259                         CDEBUG(D_LFSCK, "%s:  invalid query reply: rc = %d\n",
1260                                lfsck_lfsck2name(com->lc_lfsck), rc);
1261                         spin_lock(&ltds->ltd_lock);
1262                         list_del_init(&ltd->ltd_layout_phase_list);
1263                         list_del_init(&ltd->ltd_layout_list);
1264                         spin_unlock(&ltds->ltd_lock);
1265                         break;
1266                 }
1267
1268                 switch (reply->lr_status) {
1269                 case LS_SCANNING_PHASE1:
1270                         break;
1271                 case LS_SCANNING_PHASE2:
1272                         spin_lock(&ltds->ltd_lock);
1273                         list_del_init(&ltd->ltd_layout_phase_list);
1274                         if (ltd->ltd_dead || ltd->ltd_layout_done) {
1275                                 spin_unlock(&ltds->ltd_lock);
1276                                 break;
1277                         }
1278
1279                         if (lr->lr_flags & LEF_TO_OST)
1280                                 list_add_tail(&ltd->ltd_layout_phase_list,
1281                                               &llmd->llmd_ost_phase2_list);
1282                         else
1283                                 list_add_tail(&ltd->ltd_layout_phase_list,
1284                                               &llmd->llmd_mdt_phase2_list);
1285                         spin_unlock(&ltds->ltd_lock);
1286                         break;
1287                 default:
1288                         spin_lock(&ltds->ltd_lock);
1289                         list_del_init(&ltd->ltd_layout_phase_list);
1290                         list_del_init(&ltd->ltd_layout_list);
1291                         spin_unlock(&ltds->ltd_lock);
1292                         break;
1293                 }
1294                 break;
1295         }
1296         default:
1297                 CDEBUG(D_LFSCK, "%s: layout LFSCK unexpected event: rc = %d\n",
1298                        lfsck_lfsck2name(com->lc_lfsck), lr->lr_event);
1299                 break;
1300         }
1301
1302         if (!laia->laia_shared) {
1303                 lfsck_tgt_put(ltd);
1304                 lfsck_component_put(env, com);
1305         }
1306
1307         return 0;
1308 }
1309
1310 static int lfsck_layout_master_query_others(const struct lu_env *env,
1311                                             struct lfsck_component *com)
1312 {
1313         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1314         struct lfsck_request              *lr    = &info->lti_lr;
1315         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1316         struct lfsck_instance             *lfsck = com->lc_lfsck;
1317         struct lfsck_layout_master_data   *llmd  = com->lc_data;
1318         struct ptlrpc_request_set         *set;
1319         struct lfsck_tgt_descs            *ltds;
1320         struct lfsck_tgt_desc             *ltd;
1321         struct list_head                  *head;
1322         int                                rc    = 0;
1323         int                                rc1   = 0;
1324         ENTRY;
1325
1326         set = ptlrpc_prep_set();
1327         if (set == NULL)
1328                 RETURN(-ENOMEM);
1329
1330         llmd->llmd_touch_gen++;
1331         memset(lr, 0, sizeof(*lr));
1332         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1333         lr->lr_event = LE_QUERY;
1334         lr->lr_active = LFSCK_TYPE_LAYOUT;
1335         laia->laia_com = com;
1336         laia->laia_lr = lr;
1337         laia->laia_shared = 0;
1338
1339         if (!list_empty(&llmd->llmd_mdt_phase1_list)) {
1340                 ltds = &lfsck->li_mdt_descs;
1341                 lr->lr_flags = 0;
1342                 head = &llmd->llmd_mdt_phase1_list;
1343         } else {
1344
1345 again:
1346                 ltds = &lfsck->li_ost_descs;
1347                 lr->lr_flags = LEF_TO_OST;
1348                 head = &llmd->llmd_ost_phase1_list;
1349         }
1350
1351         laia->laia_ltds = ltds;
1352         spin_lock(&ltds->ltd_lock);
1353         while (!list_empty(head)) {
1354                 ltd = list_entry(head->next,
1355                                  struct lfsck_tgt_desc,
1356                                  ltd_layout_phase_list);
1357                 if (ltd->ltd_layout_gen == llmd->llmd_touch_gen)
1358                         break;
1359
1360                 ltd->ltd_layout_gen = llmd->llmd_touch_gen;
1361                 list_del(&ltd->ltd_layout_phase_list);
1362                 list_add_tail(&ltd->ltd_layout_phase_list, head);
1363                 atomic_inc(&ltd->ltd_ref);
1364                 laia->laia_ltd = ltd;
1365                 spin_unlock(&ltds->ltd_lock);
1366                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1367                                          lfsck_layout_master_async_interpret,
1368                                          laia, LFSCK_QUERY);
1369                 if (rc != 0) {
1370                         CDEBUG(D_LFSCK, "%s: layout LFSCK fail to query %s %x: "
1371                                "rc = %d\n", lfsck_lfsck2name(lfsck),
1372                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1373                                ltd->ltd_index, rc);
1374                         lfsck_tgt_put(ltd);
1375                         rc1 = rc;
1376                 }
1377                 spin_lock(&ltds->ltd_lock);
1378         }
1379         spin_unlock(&ltds->ltd_lock);
1380
1381         rc = ptlrpc_set_wait(set);
1382         if (rc < 0) {
1383                 ptlrpc_set_destroy(set);
1384                 RETURN(rc);
1385         }
1386
1387         if (!(lr->lr_flags & LEF_TO_OST) &&
1388             list_empty(&llmd->llmd_mdt_phase1_list))
1389                 goto again;
1390
1391         ptlrpc_set_destroy(set);
1392
1393         RETURN(rc1 != 0 ? rc1 : rc);
1394 }
1395
1396 static inline bool
1397 lfsck_layout_master_to_orphan(struct lfsck_layout_master_data *llmd)
1398 {
1399         return list_empty(&llmd->llmd_mdt_phase1_list) &&
1400                (!list_empty(&llmd->llmd_ost_phase2_list) ||
1401                 list_empty(&llmd->llmd_ost_phase1_list));
1402 }
1403
1404 static int lfsck_layout_master_notify_others(const struct lu_env *env,
1405                                              struct lfsck_component *com,
1406                                              struct lfsck_request *lr)
1407 {
1408         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1409         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1410         struct lfsck_instance             *lfsck = com->lc_lfsck;
1411         struct lfsck_layout_master_data   *llmd  = com->lc_data;
1412         struct lfsck_layout               *lo    = com->lc_file_ram;
1413         struct lfsck_bookmark             *bk    = &lfsck->li_bookmark_ram;
1414         struct ptlrpc_request_set         *set;
1415         struct lfsck_tgt_descs            *ltds;
1416         struct lfsck_tgt_desc             *ltd;
1417         struct lfsck_tgt_desc             *next;
1418         struct list_head                  *head;
1419         __u32                              idx;
1420         int                                rc    = 0;
1421         ENTRY;
1422
1423         set = ptlrpc_prep_set();
1424         if (set == NULL)
1425                 RETURN(-ENOMEM);
1426
1427         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1428         lr->lr_active = LFSCK_TYPE_LAYOUT;
1429         laia->laia_com = com;
1430         laia->laia_lr = lr;
1431         laia->laia_shared = 0;
1432         switch (lr->lr_event) {
1433         case LE_START:
1434                 /* Notify OSTs firstly, then handle other MDTs if needed. */
1435                 ltds = &lfsck->li_ost_descs;
1436                 laia->laia_ltds = ltds;
1437                 down_read(&ltds->ltd_rw_sem);
1438                 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1439                         ltd = lfsck_tgt_get(ltds, idx);
1440                         LASSERT(ltd != NULL);
1441
1442                         laia->laia_ltd = ltd;
1443                         ltd->ltd_layout_done = 0;
1444                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1445                                         lfsck_layout_master_async_interpret,
1446                                         laia, LFSCK_NOTIFY);
1447                         if (rc != 0) {
1448                                 CDEBUG(D_LFSCK, "%s: layout LFSCK fail to "
1449                                        "notify %s %x for start: rc = %d\n",
1450                                        lfsck_lfsck2name(lfsck),
1451                                        (lr->lr_flags & LEF_TO_OST) ? "OST" :
1452                                        "MDT", idx, rc);
1453                                 lfsck_tgt_put(ltd);
1454                                 lo->ll_flags |= LF_INCOMPLETE;
1455                         }
1456                 }
1457                 up_read(&ltds->ltd_rw_sem);
1458
1459                 /* Sync up */
1460                 rc = ptlrpc_set_wait(set);
1461                 if (rc < 0) {
1462                         ptlrpc_set_destroy(set);
1463                         RETURN(rc);
1464                 }
1465
1466                 if (!(bk->lb_param & LPF_ALL_TGT))
1467                         break;
1468
1469                 /* link other MDT targets locallly. */
1470                 ltds = &lfsck->li_mdt_descs;
1471                 spin_lock(&ltds->ltd_lock);
1472                 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1473                         ltd = LTD_TGT(ltds, idx);
1474                         LASSERT(ltd != NULL);
1475
1476                         if (!list_empty(&ltd->ltd_layout_list))
1477                                 continue;
1478
1479                         list_add_tail(&ltd->ltd_layout_list,
1480                                       &llmd->llmd_mdt_list);
1481                         list_add_tail(&ltd->ltd_layout_phase_list,
1482                                       &llmd->llmd_mdt_phase1_list);
1483                 }
1484                 spin_unlock(&ltds->ltd_lock);
1485                 break;
1486         case LE_STOP:
1487         case LE_PHASE2_DONE:
1488         case LE_PEER_EXIT: {
1489                 /* Handle other MDTs firstly if needed, then notify the OSTs. */
1490                 if (bk->lb_param & LPF_ALL_TGT) {
1491                         head = &llmd->llmd_mdt_list;
1492                         ltds = &lfsck->li_mdt_descs;
1493                         if (lr->lr_event == LE_STOP) {
1494                                 /* unlink other MDT targets locallly. */
1495                                 spin_lock(&ltds->ltd_lock);
1496                                 list_for_each_entry_safe(ltd, next, head,
1497                                                          ltd_layout_list) {
1498                                         list_del_init(&ltd->ltd_layout_phase_list);
1499                                         list_del_init(&ltd->ltd_layout_list);
1500                                 }
1501                                 spin_unlock(&ltds->ltd_lock);
1502
1503                                 lr->lr_flags |= LEF_TO_OST;
1504                                 head = &llmd->llmd_ost_list;
1505                                 ltds = &lfsck->li_ost_descs;
1506                         } else {
1507                                 lr->lr_flags &= ~LEF_TO_OST;
1508                         }
1509                 } else {
1510                         lr->lr_flags |= LEF_TO_OST;
1511                         head = &llmd->llmd_ost_list;
1512                         ltds = &lfsck->li_ost_descs;
1513                 }
1514
1515 again:
1516                 laia->laia_ltds = ltds;
1517                 spin_lock(&ltds->ltd_lock);
1518                 while (!list_empty(head)) {
1519                         ltd = list_entry(head->next, struct lfsck_tgt_desc,
1520                                          ltd_layout_list);
1521                         if (!list_empty(&ltd->ltd_layout_phase_list))
1522                                 list_del_init(&ltd->ltd_layout_phase_list);
1523                         list_del_init(&ltd->ltd_layout_list);
1524                         atomic_inc(&ltd->ltd_ref);
1525                         laia->laia_ltd = ltd;
1526                         spin_unlock(&ltds->ltd_lock);
1527                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1528                                         lfsck_layout_master_async_interpret,
1529                                         laia, LFSCK_NOTIFY);
1530                         if (rc != 0) {
1531                                 CDEBUG(D_LFSCK, "%s: layout LFSCK fail to "
1532                                        "notify %s %x for stop/phase2_done/"
1533                                        "peer_exit: rc = %d\n",
1534                                        lfsck_lfsck2name(lfsck),
1535                                        (lr->lr_flags & LEF_TO_OST) ? "OST" :
1536                                        "MDT", ltd->ltd_index, rc);
1537                                 lfsck_tgt_put(ltd);
1538                         }
1539                         spin_lock(&ltds->ltd_lock);
1540                 }
1541                 spin_unlock(&ltds->ltd_lock);
1542
1543                 rc = ptlrpc_set_wait(set);
1544                 if (rc < 0) {
1545                         ptlrpc_set_destroy(set);
1546                         RETURN(rc);
1547                 }
1548
1549                 if (!(lr->lr_flags & LEF_TO_OST)) {
1550                         lr->lr_flags |= LEF_TO_OST;
1551                         head = &llmd->llmd_ost_list;
1552                         ltds = &lfsck->li_ost_descs;
1553                         goto again;
1554                 }
1555                 break;
1556         }
1557         case LE_PHASE1_DONE:
1558                 llmd->llmd_touch_gen++;
1559                 ltds = &lfsck->li_mdt_descs;
1560                 laia->laia_ltds = ltds;
1561                 spin_lock(&ltds->ltd_lock);
1562                 while (!list_empty(&llmd->llmd_mdt_phase1_list)) {
1563                         ltd = list_entry(llmd->llmd_mdt_phase1_list.next,
1564                                          struct lfsck_tgt_desc,
1565                                          ltd_layout_phase_list);
1566                         if (ltd->ltd_layout_gen == llmd->llmd_touch_gen)
1567                                 break;
1568
1569                         ltd->ltd_layout_gen = llmd->llmd_touch_gen;
1570                         list_del_init(&ltd->ltd_layout_phase_list);
1571                         list_add_tail(&ltd->ltd_layout_phase_list,
1572                                       &llmd->llmd_mdt_phase1_list);
1573                         atomic_inc(&ltd->ltd_ref);
1574                         laia->laia_ltd = ltd;
1575                         spin_unlock(&ltds->ltd_lock);
1576                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1577                                         lfsck_layout_master_async_interpret,
1578                                         laia, LFSCK_NOTIFY);
1579                         if (rc != 0) {
1580                                 CDEBUG(D_LFSCK, "%s: layout LFSCK fail to "
1581                                        "notify MDT %x for phase1_done: "
1582                                        "rc = %d\n", lfsck_lfsck2name(lfsck),
1583                                        ltd->ltd_index, rc);
1584                                 lfsck_tgt_put(ltd);
1585                         }
1586                         spin_lock(&ltds->ltd_lock);
1587                 }
1588                 spin_unlock(&ltds->ltd_lock);
1589                 break;
1590         default:
1591                 CDEBUG(D_LFSCK, "%s: layout LFSCK unexpected event: rc = %d\n",
1592                        lfsck_lfsck2name(lfsck), lr->lr_event);
1593                 rc = -EINVAL;
1594                 break;
1595         }
1596
1597         rc = ptlrpc_set_wait(set);
1598         ptlrpc_set_destroy(set);
1599
1600         RETURN(rc);
1601 }
1602
1603 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1604                                            struct lfsck_component *com,
1605                                            int rc)
1606 {
1607         struct lfsck_instance   *lfsck = com->lc_lfsck;
1608         struct lfsck_layout     *lo    = com->lc_file_ram;
1609         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
1610
1611         down_write(&com->lc_sem);
1612         lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1613                                 HALF_SEC - lfsck->li_time_last_checkpoint);
1614         lo->ll_time_last_checkpoint = cfs_time_current_sec();
1615         lo->ll_objs_checked_phase2 += com->lc_new_checked;
1616
1617         if (rc > 0) {
1618                 com->lc_journal = 0;
1619                 if (lo->ll_flags & LF_INCOMPLETE)
1620                         lo->ll_status = LS_PARTIAL;
1621                 else
1622                         lo->ll_status = LS_COMPLETED;
1623                 if (!(bk->lb_param & LPF_DRYRUN))
1624                         lo->ll_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
1625                 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1626                 lo->ll_success_count++;
1627         } else if (rc == 0) {
1628                 lo->ll_status = lfsck->li_status;
1629                 if (lo->ll_status == 0)
1630                         lo->ll_status = LS_STOPPED;
1631         } else {
1632                 lo->ll_status = LS_FAILED;
1633         }
1634
1635         rc = lfsck_layout_store(env, com);
1636         up_write(&com->lc_sem);
1637
1638         return rc;
1639 }
1640
1641 static int lfsck_layout_lock(const struct lu_env *env,
1642                              struct lfsck_component *com,
1643                              struct dt_object *obj,
1644                              struct lustre_handle *lh, __u64 bits)
1645 {
1646         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1647         ldlm_policy_data_t              *policy = &info->lti_policy;
1648         struct ldlm_res_id              *resid  = &info->lti_resid;
1649         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1650         __u64                            flags  = LDLM_FL_ATOMIC_CB;
1651         int                              rc;
1652
1653         LASSERT(lfsck->li_namespace != NULL);
1654
1655         memset(policy, 0, sizeof(*policy));
1656         policy->l_inodebits.bits = bits;
1657         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
1658         rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_IBITS,
1659                                     policy, LCK_EX, &flags, ldlm_blocking_ast,
1660                                     ldlm_completion_ast, NULL, NULL, 0,
1661                                     LVB_T_NONE, NULL, lh);
1662         if (rc == ELDLM_OK) {
1663                 rc = 0;
1664         } else {
1665                 memset(lh, 0, sizeof(*lh));
1666                 rc = -EIO;
1667         }
1668
1669         return rc;
1670 }
1671
1672 static void lfsck_layout_unlock(struct lustre_handle *lh)
1673 {
1674         if (lustre_handle_is_used(lh)) {
1675                 ldlm_lock_decref(lh, LCK_EX);
1676                 memset(lh, 0, sizeof(*lh));
1677         }
1678 }
1679
1680 static int lfsck_layout_trans_stop(const struct lu_env *env,
1681                                    struct dt_device *dev,
1682                                    struct thandle *handle, int result)
1683 {
1684         int rc;
1685
1686         handle->th_result = result;
1687         rc = dt_trans_stop(env, dev, handle);
1688         if (rc > 0)
1689                 rc = 0;
1690         else if (rc == 0)
1691                 rc = 1;
1692
1693         return rc;
1694 }
1695
1696 /**
1697  * Get the system default stripe size.
1698  *
1699  * \param[in] env       pointer to the thread context
1700  * \param[in] lfsck     pointer to the lfsck instance
1701  * \param[out] size     pointer to the default stripe size
1702  *
1703  * \retval              0 for success
1704  * \retval              negative error number on failure
1705  */
1706 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1707                                            struct lfsck_instance *lfsck,
1708                                            __u32 *size)
1709 {
1710         struct lov_user_md      *lum = &lfsck_env_info(env)->lti_lum;
1711         struct dt_object        *root;
1712         int                      rc;
1713
1714         root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1715         if (IS_ERR(root))
1716                 return PTR_ERR(root);
1717
1718         /* Get the default stripe size via xattr_get on the backend root. */
1719         rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1720                           XATTR_NAME_LOV, BYPASS_CAPA);
1721         if (rc > 0) {
1722                 /* The lum->lmm_stripe_size is LE mode. The *size also
1723                  * should be LE mode. So it is unnecessary to convert. */
1724                 *size = lum->lmm_stripe_size;
1725                 rc = 0;
1726         } else if (unlikely(rc == 0)) {
1727                 rc = -EINVAL;
1728         }
1729
1730         lfsck_object_put(env, root);
1731
1732         return rc;
1733 }
1734
1735 /**
1736  * \retval       +1: repaired
1737  * \retval        0: did nothing
1738  * \retval      -ve: on error
1739  */
1740 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1741                                      struct thandle *handle,
1742                                      struct dt_object *parent,
1743                                      struct lu_fid *cfid,
1744                                      struct lu_buf *buf,
1745                                      struct lov_ost_data_v1 *slot,
1746                                      int fl, __u32 ost_idx)
1747 {
1748         struct ost_id           *oi     = &lfsck_env_info(env)->lti_oi;
1749         struct lov_mds_md_v1    *lmm    = buf->lb_buf;
1750         int                      rc;
1751
1752         fid_to_ostid(cfid, oi);
1753         ostid_cpu_to_le(oi, &slot->l_ost_oi);
1754         slot->l_ost_gen = cpu_to_le32(0);
1755         slot->l_ost_idx = cpu_to_le32(ost_idx);
1756
1757         if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE) {
1758                 struct lov_ost_data_v1 *objs;
1759                 int                     i;
1760                 __u16                   count;
1761
1762                 count = le16_to_cpu(lmm->lmm_stripe_count);
1763                 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
1764                         objs = &lmm->lmm_objects[0];
1765                 else
1766                         objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1767                 for (i = 0; i < count; i++, objs++) {
1768                         if (objs != slot && lovea_slot_is_dummy(objs))
1769                                 break;
1770                 }
1771
1772                 /* If the @slot is the last dummy slot to be refilled,
1773                  * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1774                 if (i == count)
1775                         lmm->lmm_pattern &= ~cpu_to_le32(LOV_PATTERN_F_HOLE);
1776         }
1777
1778         rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV, fl, handle,
1779                           BYPASS_CAPA);
1780         if (rc == 0)
1781                 rc = 1;
1782
1783         return rc;
1784 }
1785
1786 /**
1787  * \retval       +1: repaired
1788  * \retval        0: did nothing
1789  * \retval      -ve: on error
1790  */
1791 static int lfsck_layout_extend_lovea(const struct lu_env *env,
1792                                      struct lfsck_instance *lfsck,
1793                                      struct thandle *handle,
1794                                      struct dt_object *parent,
1795                                      struct lu_fid *cfid,
1796                                      struct lu_buf *buf, int fl,
1797                                      __u32 ost_idx, __u32 ea_off, bool reset)
1798 {
1799         struct lov_mds_md_v1    *lmm    = buf->lb_buf;
1800         struct lov_ost_data_v1  *objs;
1801         int                      rc;
1802         __u16                    count;
1803         bool                     hole   = false;
1804         ENTRY;
1805
1806         if (fl == LU_XATTR_CREATE || reset) {
1807                 __u32 pattern = LOV_PATTERN_RAID0;
1808
1809                 count = ea_off + 1;
1810                 LASSERT(buf->lb_len == lov_mds_md_size(count, LOV_MAGIC_V1));
1811
1812                 if (ea_off != 0 || reset) {
1813                         pattern |= LOV_PATTERN_F_HOLE;
1814                         hole = true;
1815                 }
1816
1817                 memset(lmm, 0, buf->lb_len);
1818                 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1819                 lmm->lmm_pattern = cpu_to_le32(pattern);
1820                 fid_to_lmm_oi(lfsck_dto2fid(parent), &lmm->lmm_oi);
1821                 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1822
1823                 rc = lfsck_layout_get_def_stripesize(env, lfsck,
1824                                                      &lmm->lmm_stripe_size);
1825                 if (rc != 0)
1826                         RETURN(rc);
1827
1828                 objs = &lmm->lmm_objects[ea_off];
1829         } else {
1830                 __u32   magic = le32_to_cpu(lmm->lmm_magic);
1831                 int     gap;
1832
1833                 count = le16_to_cpu(lmm->lmm_stripe_count);
1834                 if (magic == LOV_MAGIC_V1)
1835                         objs = &lmm->lmm_objects[count];
1836                 else
1837                         objs = &((struct lov_mds_md_v3 *)lmm)->
1838                                                         lmm_objects[count];
1839
1840                 gap = ea_off - count;
1841                 if (gap >= 0)
1842                         count = ea_off + 1;
1843                 LASSERT(buf->lb_len == lov_mds_md_size(count, magic));
1844
1845                 if (gap > 0) {
1846                         memset(objs, 0, gap * sizeof(*objs));
1847                         lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
1848                         hole = true;
1849                 }
1850
1851                 lmm->lmm_layout_gen =
1852                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
1853                 objs += gap;
1854         }
1855
1856         lmm->lmm_stripe_count = cpu_to_le16(count);
1857         rc = lfsck_layout_refill_lovea(env, handle, parent, cfid, buf, objs,
1858                                        fl, ost_idx);
1859
1860         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
1861                DFID": parent "DFID", OST-index %u, stripe-index %u, fl %d, "
1862                "reset %s, %s LOV EA hole: rc = %d\n",
1863                lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1864                ost_idx, ea_off, fl, reset ? "yes" : "no",
1865                hole ? "with" : "without", rc);
1866
1867         RETURN(rc);
1868 }
1869
1870 /**
1871  * \retval       +1: repaired
1872  * \retval        0: did nothing
1873  * \retval      -ve: on error
1874  */
1875 static int lfsck_layout_update_pfid(const struct lu_env *env,
1876                                     struct lfsck_component *com,
1877                                     struct dt_object *parent,
1878                                     struct lu_fid *cfid,
1879                                     struct dt_device *cdev, __u32 ea_off)
1880 {
1881         struct filter_fid       *pfid   = &lfsck_env_info(env)->lti_new_pfid;
1882         struct dt_object        *child;
1883         struct thandle          *handle;
1884         const struct lu_fid     *tfid   = lu_object_fid(&parent->do_lu);
1885         struct lu_buf           *buf;
1886         int                      rc     = 0;
1887         ENTRY;
1888
1889         child = lfsck_object_find_by_dev(env, cdev, cfid);
1890         if (IS_ERR(child))
1891                 RETURN(PTR_ERR(child));
1892
1893         handle = dt_trans_create(env, cdev);
1894         if (IS_ERR(handle))
1895                 GOTO(out, rc = PTR_ERR(handle));
1896
1897         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
1898         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
1899         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
1900          * MDT-object's FID::f_ver, instead it is the OST-object index in its
1901          * parent MDT-object's layout EA. */
1902         pfid->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
1903         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
1904
1905         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
1906         if (rc != 0)
1907                 GOTO(stop, rc);
1908
1909         rc = dt_trans_start(env, cdev, handle);
1910         if (rc != 0)
1911                 GOTO(stop, rc);
1912
1913         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle,
1914                           BYPASS_CAPA);
1915
1916         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1917
1918 stop:
1919         dt_trans_stop(env, cdev, handle);
1920
1921 out:
1922         lu_object_put(env, &child->do_lu);
1923
1924         return rc;
1925 }
1926
1927 /**
1928  * This function will create the MDT-object with the given (partial) LOV EA.
1929  *
1930  * Under some data corruption cases, the MDT-object of the file may be lost,
1931  * but its OST-objects, or some of them are there. The layout LFSCK needs to
1932  * re-create the MDT-object with the orphan OST-object(s) information.
1933  *
1934  * On the other hand, the LFSCK may has created some OST-object for repairing
1935  * dangling LOV EA reference, but as the LFSCK processing, it may find that
1936  * the old OST-object is there and should replace the former new created OST
1937  * object. Unfortunately, some others have modified such newly created object.
1938  * To keep the data (both new and old), the LFSCK will create MDT-object with
1939  * new FID to reference the original OST-object.
1940  *
1941  * \param[in] env       pointer to the thread context
1942  * \param[in] com       pointer to the lfsck component
1943  * \param[in] ltd       pointer to target device descriptor
1944  * \param[in] rec       pointer to the record for the orphan OST-object
1945  * \param[in] cfid      pointer to FID for the orphan OST-object
1946  * \param[in] infix     additional information, such as the FID for original
1947  *                      MDT-object and the stripe offset in the LOV EA
1948  * \param[in] type      the type for describing why the orphan MDT-object is
1949  *                      created. The rules are as following:
1950  *
1951  *  type "C":           Multiple OST-objects claim the same MDT-object and the
1952  *                      same slot in the layout EA. Then the LFSCK will create
1953  *                      new MDT-object(s) to hold the conflict OST-object(s).
1954  *
1955  *  type "N":           The orphan OST-object does not know which one was the
1956  *                      real parent MDT-object, so the LFSCK uses new FID for
1957  *                      its parent MDT-object.
1958  *
1959  *  type "R":           The orphan OST-object knows its parent MDT-object FID,
1960  *                      but does not know the position (the file name) in the
1961  *                      namespace.
1962  *
1963  * The orphan name will be like:
1964  * ${FID}-${infix}-${type}-${conflict_version}
1965  *
1966  * \param[in] ea_off    the stripe offset in the LOV EA
1967  *
1968  * \retval              positive on repaired something
1969  * \retval              0 if needs to repair nothing
1970  * \retval              negative error number on failure
1971  */
1972 static int lfsck_layout_recreate_parent(const struct lu_env *env,
1973                                         struct lfsck_component *com,
1974                                         struct lfsck_tgt_desc *ltd,
1975                                         struct lu_orphan_rec *rec,
1976                                         struct lu_fid *cfid,
1977                                         const char *infix,
1978                                         const char *type,
1979                                         __u32 ea_off)
1980 {
1981         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1982         struct dt_insert_rec            *dtrec  = &info->lti_dt_rec;
1983         char                            *name   = info->lti_key;
1984         struct lu_attr                  *la     = &info->lti_la;
1985         struct dt_object_format         *dof    = &info->lti_dof;
1986         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1987         struct lu_fid                   *pfid   = &rec->lor_fid;
1988         struct lu_fid                   *tfid   = &info->lti_fid3;
1989         struct dt_device                *next   = lfsck->li_next;
1990         struct dt_object                *pobj   = NULL;
1991         struct dt_object                *cobj   = NULL;
1992         struct thandle                  *th     = NULL;
1993         struct lu_buf                   *pbuf   = NULL;
1994         struct lu_buf                   *ea_buf = &info->lti_big_buf;
1995         struct lustre_handle             lh     = { 0 };
1996         struct linkea_data               ldata  = { 0 };
1997         struct lu_buf                    linkea_buf;
1998         const struct lu_name            *pname;
1999         int                              buflen = ea_buf->lb_len;
2000         int                              idx    = 0;
2001         int                              rc     = 0;
2002         ENTRY;
2003
2004         /* Create .lustre/lost+found/MDTxxxx when needed. */
2005         if (unlikely(lfsck->li_lpf_obj == NULL)) {
2006                 rc = lfsck_create_lpf(env, lfsck);
2007                 if (rc != 0)
2008                         GOTO(log, rc);
2009         }
2010
2011         if (fid_is_zero(pfid)) {
2012                 struct filter_fid *ff = &info->lti_new_pfid;
2013
2014                 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
2015                 if (rc != 0)
2016                         RETURN(rc);
2017
2018                 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
2019                 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
2020                 /* Currently, the filter_fid::ff_parent::f_ver is not the
2021                  * real parent MDT-object's FID::f_ver, instead it is the
2022                  * OST-object index in its parent MDT-object's layout EA. */
2023                 ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
2024                 pbuf = lfsck_buf_get(env, ff, sizeof(struct filter_fid));
2025                 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
2026                 if (IS_ERR(cobj))
2027                         GOTO(log, rc = PTR_ERR(cobj));
2028         }
2029
2030         pobj = lfsck_object_find_by_dev(env, lfsck->li_bottom, pfid);
2031         if (IS_ERR(pobj))
2032                 GOTO(put, rc = PTR_ERR(pobj));
2033
2034         LASSERT(infix != NULL);
2035         LASSERT(type != NULL);
2036
2037         do {
2038                 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
2039                          type, idx++);
2040                 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
2041                                (const struct dt_key *)name, BYPASS_CAPA);
2042                 if (rc != 0 && rc != -ENOENT)
2043                         GOTO(put, rc);
2044         } while (rc == 0);
2045
2046         rc = linkea_data_new(&ldata,
2047                              &lfsck_env_info(env)->lti_linkea_buf);
2048         if (rc != 0)
2049                 GOTO(put, rc);
2050
2051         pname = lfsck_name_get_const(env, name, strlen(name));
2052         rc = linkea_add_buf(&ldata, pname, lfsck_dto2fid(lfsck->li_lpf_obj));
2053         if (rc != 0)
2054                 GOTO(put, rc);
2055
2056         memset(la, 0, sizeof(*la));
2057         la->la_uid = rec->lor_uid;
2058         la->la_gid = rec->lor_gid;
2059         la->la_mode = S_IFREG | S_IRUSR;
2060         la->la_valid = LA_MODE | LA_UID | LA_GID;
2061
2062         memset(dof, 0, sizeof(*dof));
2063         dof->dof_type = dt_mode_to_dft(S_IFREG);
2064
2065         rc = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2066         if (buflen < rc) {
2067                 lu_buf_realloc(ea_buf, rc);
2068                 buflen = ea_buf->lb_len;
2069                 if (ea_buf->lb_buf == NULL)
2070                         GOTO(put, rc = -ENOMEM);
2071         } else {
2072                 ea_buf->lb_len = rc;
2073         }
2074
2075         /* Hold update lock on the .lustre/lost+found/MDTxxxx/.
2076          *
2077          * XXX: Currently, we do not grab the PDO lock as normal create cases,
2078          *      because creating MDT-object for orphan OST-object is rare, we
2079          *      do not much care about the performance. It can be improved in
2080          *      the future when needed. */
2081         rc = lfsck_layout_lock(env, com, lfsck->li_lpf_obj, &lh,
2082                                MDS_INODELOCK_UPDATE);
2083         if (rc != 0)
2084                 GOTO(put, rc);
2085
2086         th = dt_trans_create(env, next);
2087         if (IS_ERR(th))
2088                 GOTO(unlock, rc = PTR_ERR(th));
2089
2090         /* 1a. Update OST-object's parent information remotely.
2091          *
2092          * If other subsequent modifications failed, then next LFSCK scanning
2093          * will process the OST-object as orphan again with known parent FID. */
2094         if (cobj != NULL) {
2095                 rc = dt_declare_xattr_set(env, cobj, pbuf, XATTR_NAME_FID, 0, th);
2096                 if (rc != 0)
2097                         GOTO(stop, rc);
2098         }
2099
2100         /* 2a. Create the MDT-object locally. */
2101         rc = dt_declare_create(env, pobj, la, NULL, dof, th);
2102         if (rc != 0)
2103                 GOTO(stop, rc);
2104
2105         /* 3a. Add layout EA for the MDT-object. */
2106         rc = dt_declare_xattr_set(env, pobj, ea_buf, XATTR_NAME_LOV,
2107                                   LU_XATTR_CREATE, th);
2108         if (rc != 0)
2109                 GOTO(stop, rc);
2110
2111         /* 4a. Insert the MDT-object to .lustre/lost+found/MDTxxxx/ */
2112         dtrec->rec_fid = pfid;
2113         dtrec->rec_type = S_IFREG;
2114         rc = dt_declare_insert(env, lfsck->li_lpf_obj,
2115                                (const struct dt_rec *)dtrec,
2116                                (const struct dt_key *)name, th);
2117         if (rc != 0)
2118                 GOTO(stop, rc);
2119
2120         /* 5a. insert linkEA for parent. */
2121         linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
2122         linkea_buf.lb_len = ldata.ld_leh->leh_len;
2123         rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
2124                                   XATTR_NAME_LINK, 0, th);
2125         if (rc != 0)
2126                 GOTO(stop, rc);
2127
2128         rc = dt_trans_start(env, next, th);
2129         if (rc != 0)
2130                 GOTO(stop, rc);
2131
2132         /* 1b. Update OST-object's parent information remotely. */
2133         if (cobj != NULL) {
2134                 rc = dt_xattr_set(env, cobj, pbuf, XATTR_NAME_FID, 0, th,
2135                                   BYPASS_CAPA);
2136                 if (rc != 0)
2137                         GOTO(stop, rc);
2138         }
2139
2140         dt_write_lock(env, pobj, 0);
2141         /* 2b. Create the MDT-object locally. */
2142         rc = dt_create(env, pobj, la, NULL, dof, th);
2143         if (rc == 0)
2144                 /* 3b. Add layout EA for the MDT-object. */
2145                 rc = lfsck_layout_extend_lovea(env, lfsck, th, pobj, cfid,
2146                                                ea_buf, LU_XATTR_CREATE,
2147                                                ltd->ltd_index, ea_off, false);
2148         dt_write_unlock(env, pobj);
2149         if (rc < 0)
2150                 GOTO(stop, rc);
2151
2152         /* 4b. Insert the MDT-object to .lustre/lost+found/MDTxxxx/ */
2153         rc = dt_insert(env, lfsck->li_lpf_obj, (const struct dt_rec *)dtrec,
2154                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
2155         if (rc != 0)
2156                 GOTO(stop, rc);
2157
2158         /* 5b. insert linkEA for parent. */
2159         rc = dt_xattr_set(env, pobj, &linkea_buf,
2160                           XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
2161
2162         GOTO(stop, rc);
2163
2164 stop:
2165         dt_trans_stop(env, next, th);
2166
2167 unlock:
2168         lfsck_layout_unlock(&lh);
2169
2170 put:
2171         if (cobj != NULL && !IS_ERR(cobj))
2172                 lu_object_put(env, &cobj->do_lu);
2173         if (pobj != NULL && !IS_ERR(pobj))
2174                 lu_object_put(env, &pobj->do_lu);
2175         ea_buf->lb_len = buflen;
2176
2177 log:
2178         if (rc < 0)
2179                 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
2180                        "recreate the lost MDT-object: parent "DFID
2181                        ", child "DFID", OST-index %u, stripe-index %u, "
2182                        "infix %s, type %s: rc = %d\n",
2183                        lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2184                        ltd->ltd_index, ea_off, infix, type, rc);
2185
2186         return rc >= 0 ? 1 : rc;
2187 }
2188
2189 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
2190                                                    struct lfsck_component *com,
2191                                                    const struct lu_fid *fid,
2192                                                    __u32 index)
2193 {
2194         struct lfsck_thread_info *info  = lfsck_env_info(env);
2195         struct lfsck_request     *lr    = &info->lti_lr;
2196         struct lfsck_instance    *lfsck = com->lc_lfsck;
2197         struct lfsck_tgt_desc    *ltd;
2198         struct ptlrpc_request    *req;
2199         struct lfsck_request     *tmp;
2200         struct obd_export        *exp;
2201         int                       rc    = 0;
2202         ENTRY;
2203
2204         ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
2205         if (unlikely(ltd == NULL))
2206                 RETURN(-ENXIO);
2207
2208         exp = ltd->ltd_exp;
2209         if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
2210                 GOTO(put, rc = -EOPNOTSUPP);
2211
2212         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
2213         if (req == NULL)
2214                 GOTO(put, rc = -ENOMEM);
2215
2216         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
2217         if (rc != 0) {
2218                 ptlrpc_request_free(req);
2219
2220                 GOTO(put, rc);
2221         }
2222
2223         memset(lr, 0, sizeof(*lr));
2224         lr->lr_event = LE_CONDITIONAL_DESTROY;
2225         lr->lr_active = LFSCK_TYPE_LAYOUT;
2226         lr->lr_fid = *fid;
2227
2228         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2229         *tmp = *lr;
2230         ptlrpc_request_set_replen(req);
2231
2232         rc = ptlrpc_queue_wait(req);
2233         ptlrpc_req_finished(req);
2234
2235         GOTO(put, rc);
2236
2237 put:
2238         lfsck_tgt_put(ltd);
2239
2240         return rc;
2241 }
2242
2243 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
2244                                                   struct lfsck_component *com,
2245                                                   struct lfsck_request *lr)
2246 {
2247         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2248         struct lu_attr                  *la     = &info->lti_la;
2249         ldlm_policy_data_t              *policy = &info->lti_policy;
2250         struct ldlm_res_id              *resid  = &info->lti_resid;
2251         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2252         struct dt_device                *dev    = lfsck->li_bottom;
2253         struct lu_fid                   *fid    = &lr->lr_fid;
2254         struct dt_object                *obj;
2255         struct thandle                  *th     = NULL;
2256         struct lustre_handle             lh     = { 0 };
2257         __u64                            flags  = 0;
2258         int                              rc     = 0;
2259         ENTRY;
2260
2261         obj = lfsck_object_find_by_dev(env, dev, fid);
2262         if (IS_ERR(obj))
2263                 RETURN(PTR_ERR(obj));
2264
2265         dt_read_lock(env, obj, 0);
2266         if (dt_object_exists(obj) == 0) {
2267                 dt_read_unlock(env, obj);
2268
2269                 GOTO(put, rc = -ENOENT);
2270         }
2271
2272         /* Get obj's attr without lock firstly. */
2273         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
2274         dt_read_unlock(env, obj);
2275         if (rc != 0)
2276                 GOTO(put, rc);
2277
2278         if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
2279                 GOTO(put, rc = -ETXTBSY);
2280
2281         /* Acquire extent lock on [0, EOF] to sync with all possible written. */
2282         LASSERT(lfsck->li_namespace != NULL);
2283
2284         memset(policy, 0, sizeof(*policy));
2285         policy->l_extent.end = OBD_OBJECT_EOF;
2286         ost_fid_build_resid(fid, resid);
2287         rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
2288                                     policy, LCK_EX, &flags, ldlm_blocking_ast,
2289                                     ldlm_completion_ast, NULL, NULL, 0,
2290                                     LVB_T_NONE, NULL, &lh);
2291         if (rc != ELDLM_OK)
2292                 GOTO(put, rc = -EIO);
2293
2294         dt_write_lock(env, obj, 0);
2295         /* Get obj's attr within lock again. */
2296         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
2297         if (rc != 0)
2298                 GOTO(unlock, rc);
2299
2300         if (la->la_ctime != 0)
2301                 GOTO(unlock, rc = -ETXTBSY);
2302
2303         th = dt_trans_create(env, dev);
2304         if (IS_ERR(th))
2305                 GOTO(unlock, rc = PTR_ERR(th));
2306
2307         rc = dt_declare_ref_del(env, obj, th);
2308         if (rc != 0)
2309                 GOTO(stop, rc);
2310
2311         rc = dt_declare_destroy(env, obj, th);
2312         if (rc != 0)
2313                 GOTO(stop, rc);
2314
2315         rc = dt_trans_start_local(env, dev, th);
2316         if (rc != 0)
2317                 GOTO(stop, rc);
2318
2319         rc = dt_ref_del(env, obj, th);
2320         if (rc != 0)
2321                 GOTO(stop, rc);
2322
2323         rc = dt_destroy(env, obj, th);
2324         if (rc == 0)
2325                 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2326                        "OST-object "DFID" that was created for reparing "
2327                        "dangling referenced case. But the original missed "
2328                        "OST-object is found now.\n",
2329                        lfsck_lfsck2name(lfsck), PFID(fid));
2330
2331         GOTO(stop, rc);
2332
2333 stop:
2334         dt_trans_stop(env, dev, th);
2335
2336 unlock:
2337         dt_write_unlock(env, obj);
2338         ldlm_lock_decref(&lh, LCK_EX);
2339
2340 put:
2341         lu_object_put(env, &obj->do_lu);
2342
2343         return rc;
2344 }
2345
2346 /**
2347  * Some OST-object has occupied the specified layout EA slot.
2348  * Such OST-object may be generated by the LFSCK when repair
2349  * dangling referenced MDT-object, which can be indicated by
2350  * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2351  * is true and such OST-object has not been modified yet, we
2352  * will replace it with the orphan OST-object; otherwise the
2353  * LFSCK will create new MDT-object to reference the orphan.
2354  *
2355  * \retval       +1: repaired
2356  * \retval        0: did nothing
2357  * \retval      -ve: on error
2358  */
2359 static int lfsck_layout_conflict_create(const struct lu_env *env,
2360                                         struct lfsck_component *com,
2361                                         struct lfsck_tgt_desc *ltd,
2362                                         struct lu_orphan_rec *rec,
2363                                         struct dt_object *parent,
2364                                         struct lu_fid *cfid,
2365                                         struct lu_buf *ea_buf,
2366                                         struct lov_ost_data_v1 *slot,
2367                                         __u32 ea_off, __u32 ori_len)
2368 {
2369         struct lfsck_thread_info *info          = lfsck_env_info(env);
2370         struct lu_fid            *cfid2         = &info->lti_fid2;
2371         struct ost_id            *oi            = &info->lti_oi;
2372         char                     *infix         = info->lti_tmpbuf;
2373         struct lov_mds_md_v1     *lmm           = ea_buf->lb_buf;
2374         struct dt_device         *dev           = com->lc_lfsck->li_bottom;
2375         struct thandle           *th            = NULL;
2376         struct lustre_handle      lh            = { 0 };
2377         __u32                     ost_idx2      = le32_to_cpu(slot->l_ost_idx);
2378         int                       rc            = 0;
2379         ENTRY;
2380
2381         ostid_le_to_cpu(&slot->l_ost_oi, oi);
2382         ostid_to_fid(cfid2, oi, ost_idx2);
2383
2384         /* Hold layout lock on the parent to prevent others to access. */
2385         rc = lfsck_layout_lock(env, com, parent, &lh,
2386                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
2387         if (rc != 0)
2388                 GOTO(out, rc);
2389
2390         rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2391
2392         /* If the conflict OST-obejct is not created for fixing dangling
2393          * referenced MDT-object in former LFSCK check/repair, or it has
2394          * been modified by others, then we cannot destroy it. Re-create
2395          * a new MDT-object for the orphan OST-object. */
2396         if (rc == -ETXTBSY) {
2397                 /* No need the layout lock on the original parent. */
2398                 lfsck_layout_unlock(&lh);
2399                 ea_buf->lb_len = ori_len;
2400
2401                 fid_zero(&rec->lor_fid);
2402                 snprintf(infix, LFSCK_TMPBUF_LEN, "-"DFID"-%x",
2403                          PFID(lu_object_fid(&parent->do_lu)), ea_off);
2404                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2405                                                   infix, "C", ea_off);
2406
2407                 RETURN(rc);
2408         }
2409
2410         if (rc != 0 && rc != -ENOENT)
2411                 GOTO(unlock, rc);
2412
2413         th = dt_trans_create(env, dev);
2414         if (IS_ERR(th))
2415                 GOTO(unlock, rc = PTR_ERR(th));
2416
2417         rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2418                                   LU_XATTR_REPLACE, th);
2419         if (rc != 0)
2420                 GOTO(stop, rc);
2421
2422         rc = dt_trans_start_local(env, dev, th);
2423         if (rc != 0)
2424                 GOTO(stop, rc);
2425
2426         dt_write_lock(env, parent, 0);
2427         lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2428         rc = lfsck_layout_refill_lovea(env, th, parent, cfid, ea_buf, slot,
2429                                        LU_XATTR_REPLACE, ltd->ltd_index);
2430         dt_write_unlock(env, parent);
2431
2432         GOTO(stop, rc);
2433
2434 stop:
2435         dt_trans_stop(env, dev, th);
2436
2437 unlock:
2438         lfsck_layout_unlock(&lh);
2439
2440 out:
2441         ea_buf->lb_len = ori_len;
2442
2443         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2444                "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2445                "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2446                lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2447                PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2448                ea_off, rc);
2449
2450         return rc >= 0 ? 1 : rc;
2451 }
2452
2453 /**
2454  * \retval       +1: repaired
2455  * \retval        0: did nothing
2456  * \retval      -ve: on error
2457  */
2458 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2459                                        struct lfsck_component *com,
2460                                        struct lfsck_tgt_desc *ltd,
2461                                        struct lu_orphan_rec *rec,
2462                                        struct dt_object *parent,
2463                                        struct lu_fid *cfid,
2464                                        __u32 ost_idx, __u32 ea_off)
2465 {
2466         struct lfsck_thread_info *info          = lfsck_env_info(env);
2467         struct lu_buf            *buf           = &info->lti_big_buf;
2468         struct lu_fid            *fid           = &info->lti_fid2;
2469         struct ost_id            *oi            = &info->lti_oi;
2470         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2471         struct dt_device         *dt            = lfsck->li_bottom;
2472         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2473         struct thandle            *handle       = NULL;
2474         size_t                    buflen        = buf->lb_len;
2475         size_t                    lovea_size;
2476         struct lov_mds_md_v1     *lmm;
2477         struct lov_ost_data_v1   *objs;
2478         struct lustre_handle      lh            = { 0 };
2479         __u32                     magic;
2480         int                       fl            = 0;
2481         int                       rc            = 0;
2482         int                       rc1;
2483         int                       i;
2484         __u16                     count;
2485         bool                      locked        = false;
2486         ENTRY;
2487
2488         rc = lfsck_layout_lock(env, com, parent, &lh,
2489                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
2490         if (rc != 0) {
2491                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2492                        "LOV EA for "DFID": parent "DFID", OST-index %u, "
2493                        "stripe-index %u: rc = %d\n",
2494                        lfsck_lfsck2name(lfsck), PFID(cfid),
2495                        PFID(lfsck_dto2fid(parent)), ost_idx, ea_off, rc);
2496
2497                 RETURN(rc);
2498         }
2499
2500 again:
2501         if (locked) {
2502                 dt_write_unlock(env, parent);
2503                 locked = false;
2504         }
2505
2506         if (handle != NULL) {
2507                 dt_trans_stop(env, dt, handle);
2508                 handle = NULL;
2509         }
2510
2511         if (rc < 0)
2512                 GOTO(unlock_layout, rc);
2513
2514         lovea_size = rc;
2515         if (buf->lb_len < lovea_size) {
2516                 lu_buf_realloc(buf, lovea_size);
2517                 buflen = buf->lb_len;
2518                 if (buf->lb_buf == NULL)
2519                         GOTO(unlock_layout, rc = -ENOMEM);
2520         }
2521
2522         if (!(bk->lb_param & LPF_DRYRUN)) {
2523                 handle = dt_trans_create(env, dt);
2524                 if (IS_ERR(handle))
2525                         GOTO(unlock_layout, rc = PTR_ERR(handle));
2526
2527                 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2528                                           fl, handle);
2529                 if (rc != 0)
2530                         GOTO(stop, rc);
2531
2532                 rc = dt_trans_start_local(env, dt, handle);
2533                 if (rc != 0)
2534                         GOTO(stop, rc);
2535         }
2536
2537         dt_write_lock(env, parent, 0);
2538         locked = true;
2539         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV, BYPASS_CAPA);
2540         if (rc == -ERANGE) {
2541                 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV,
2542                                   BYPASS_CAPA);
2543                 LASSERT(rc != 0);
2544                 goto again;
2545         } else if (rc == -ENODATA || rc == 0) {
2546                 lovea_size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2547                 /* If the declared is not big enough, re-try. */
2548                 if (buf->lb_len < lovea_size) {
2549                         rc = lovea_size;
2550                         goto again;
2551                 }
2552                 fl = LU_XATTR_CREATE;
2553         } else if (rc < 0) {
2554                 GOTO(unlock_parent, rc);
2555         } else if (unlikely(buf->lb_len == 0)) {
2556                 goto again;
2557         } else {
2558                 fl = LU_XATTR_REPLACE;
2559                 lovea_size = rc;
2560         }
2561
2562         if (fl == LU_XATTR_CREATE) {
2563                 if (bk->lb_param & LPF_DRYRUN)
2564                         GOTO(unlock_parent, rc = 1);
2565
2566                 LASSERT(buf->lb_len >= lovea_size);
2567
2568                 buf->lb_len = lovea_size;
2569                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2570                                                buf, fl, ost_idx, ea_off, false);
2571
2572                 GOTO(unlock_parent, rc);
2573         }
2574
2575         lmm = buf->lb_buf;
2576         rc1 = lfsck_layout_verify_header(lmm);
2577
2578         /* If the LOV EA crashed, the rebuild it. */
2579         if (rc1 == -EINVAL) {
2580                 if (bk->lb_param & LPF_DRYRUN)
2581                         GOTO(unlock_parent, rc = 1);
2582
2583                 LASSERT(buf->lb_len >= lovea_size);
2584
2585                 buf->lb_len = lovea_size;
2586                 memset(lmm, 0, buf->lb_len);
2587                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2588                                                buf, fl, ost_idx, ea_off, true);
2589
2590                 GOTO(unlock_parent, rc);
2591         }
2592
2593         /* For other unknown magic/pattern, keep the current LOV EA. */
2594         if (rc1 != 0)
2595                 GOTO(unlock_parent, rc = rc1);
2596
2597         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2598          * been verified in lfsck_layout_verify_header() already. If some
2599          * new magic introduced in the future, then layout LFSCK needs to
2600          * be updated also. */
2601         magic = le32_to_cpu(lmm->lmm_magic);
2602         if (magic == LOV_MAGIC_V1) {
2603                 objs = &lmm->lmm_objects[0];
2604         } else {
2605                 LASSERT(magic == LOV_MAGIC_V3);
2606                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2607         }
2608
2609         count = le16_to_cpu(lmm->lmm_stripe_count);
2610         if (count == 0)
2611                 GOTO(unlock_parent, rc = -EINVAL);
2612         LASSERT(count > 0);
2613
2614         /* Exceed the current end of MDT-object layout EA. Then extend it. */
2615         if (count <= ea_off) {
2616                 if (bk->lb_param & LPF_DRYRUN)
2617                         GOTO(unlock_parent, rc = 1);
2618
2619                 lovea_size = lov_mds_md_size(ea_off + 1, magic);
2620                 /* If the declared is not big enough, re-try. */
2621                 if (buf->lb_len < lovea_size) {
2622                         rc = lovea_size;
2623                         goto again;
2624                 }
2625                 buf->lb_len = lovea_size;
2626                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2627                                                buf, fl, ost_idx, ea_off, false);
2628
2629                 GOTO(unlock_parent, rc);
2630         }
2631
2632         LASSERTF(rc > 0, "invalid rc = %d\n", rc);
2633
2634         buf->lb_len = lovea_size;
2635         for (i = 0; i < count; i++, objs++) {
2636                 /* The MDT-object was created via lfsck_layout_recover_create()
2637                  * by others before, and we fill the dummy layout EA. */
2638                 if (lovea_slot_is_dummy(objs)) {
2639                         if (i != ea_off)
2640                                 continue;
2641
2642                         if (bk->lb_param & LPF_DRYRUN)
2643                                 GOTO(unlock_parent, rc = 1);
2644
2645                         lmm->lmm_layout_gen =
2646                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2647                         rc = lfsck_layout_refill_lovea(env, handle, parent,
2648                                                        cfid, buf, objs, fl,
2649                                                        ost_idx);
2650
2651                         CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
2652                                "dummy layout slot for "DFID": parent "DFID
2653                                ", OST-index %u, stripe-index %u: rc = %d\n",
2654                                lfsck_lfsck2name(lfsck), PFID(cfid),
2655                                PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
2656
2657                         GOTO(unlock_parent, rc);
2658                 }
2659
2660                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2661                 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
2662                 /* It should be rare case, the slot is there, but the LFSCK
2663                  * does not handle it during the first-phase cycle scanning. */
2664                 if (unlikely(lu_fid_eq(fid, cfid))) {
2665                         if (i == ea_off) {
2666                                 GOTO(unlock_parent, rc = 0);
2667                         } else {
2668                                 /* Rare case that the OST-object index
2669                                  * does not match the parent MDT-object
2670                                  * layout EA. We trust the later one. */
2671                                 if (bk->lb_param & LPF_DRYRUN)
2672                                         GOTO(unlock_parent, rc = 1);
2673
2674                                 dt_write_unlock(env, parent);
2675                                 if (handle != NULL)
2676                                         dt_trans_stop(env, dt, handle);
2677                                 lfsck_layout_unlock(&lh);
2678                                 buf->lb_len = buflen;
2679                                 rc = lfsck_layout_update_pfid(env, com, parent,
2680                                                         cfid, ltd->ltd_tgt, i);
2681
2682                                 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
2683                                        "updated OST-object's pfid for "DFID
2684                                        ": parent "DFID", OST-index %u, "
2685                                        "stripe-index %u: rc = %d\n",
2686                                        lfsck_lfsck2name(lfsck), PFID(cfid),
2687                                        PFID(lfsck_dto2fid(parent)),
2688                                        ltd->ltd_index, i, rc);
2689
2690                                 RETURN(rc);
2691                         }
2692                 }
2693         }
2694
2695         /* The MDT-object exists, but related layout EA slot is occupied
2696          * by others. */
2697         if (bk->lb_param & LPF_DRYRUN)
2698                 GOTO(unlock_parent, rc = 1);
2699
2700         dt_write_unlock(env, parent);
2701         if (handle != NULL)
2702                 dt_trans_stop(env, dt, handle);
2703         lfsck_layout_unlock(&lh);
2704         if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
2705                 objs = &lmm->lmm_objects[ea_off];
2706         else
2707                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
2708         rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
2709                                           buf, objs, ea_off, buflen);
2710
2711         RETURN(rc);
2712
2713 unlock_parent:
2714         if (locked)
2715                 dt_write_unlock(env, parent);
2716
2717 stop:
2718         if (handle != NULL)
2719                 dt_trans_stop(env, dt, handle);
2720
2721 unlock_layout:
2722         lfsck_layout_unlock(&lh);
2723         buf->lb_len = buflen;
2724
2725         return rc;
2726 }
2727
2728 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
2729                                         struct lfsck_component *com,
2730                                         struct lfsck_tgt_desc *ltd,
2731                                         struct lu_orphan_rec *rec,
2732                                         struct lu_fid *cfid)
2733 {
2734         struct lfsck_layout     *lo     = com->lc_file_ram;
2735         struct lu_fid           *pfid   = &rec->lor_fid;
2736         struct dt_object        *parent = NULL;
2737         __u32                    ea_off = pfid->f_stripe_idx;
2738         int                      rc     = 0;
2739         ENTRY;
2740
2741         if (!fid_is_sane(cfid))
2742                 GOTO(out, rc = -EINVAL);
2743
2744         if (fid_is_zero(pfid)) {
2745                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2746                                                   "", "N", ea_off);
2747                 GOTO(out, rc);
2748         }
2749
2750         pfid->f_ver = 0;
2751         if (!fid_is_sane(pfid))
2752                 GOTO(out, rc = -EINVAL);
2753
2754         parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
2755         if (IS_ERR(parent))
2756                 GOTO(out, rc = PTR_ERR(parent));
2757
2758         if (unlikely(dt_object_remote(parent) != 0))
2759                 GOTO(put, rc = -EXDEV);
2760
2761         if (dt_object_exists(parent) == 0) {
2762                 lu_object_put(env, &parent->do_lu);
2763                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2764                                                   "", "R", ea_off);
2765                 GOTO(out, rc);
2766         }
2767
2768         if (!S_ISREG(lu_object_attr(&parent->do_lu)))
2769                 GOTO(put, rc = -EISDIR);
2770
2771         rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
2772                                          ltd->ltd_index, ea_off);
2773
2774         GOTO(put, rc);
2775
2776 put:
2777         if (rc <= 0)
2778                 lu_object_put(env, &parent->do_lu);
2779         else
2780                 /* The layout EA is changed, need to be reloaded next time. */
2781                 lu_object_put_nocache(env, &parent->do_lu);
2782
2783 out:
2784         down_write(&com->lc_sem);
2785         com->lc_new_scanned++;
2786         com->lc_new_checked++;
2787         if (rc > 0) {
2788                 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
2789                 rc = 0;
2790         } else if (rc < 0) {
2791                 lo->ll_objs_failed_phase2++;
2792         }
2793         up_write(&com->lc_sem);
2794
2795         return rc;
2796 }
2797
2798 static int lfsck_layout_scan_orphan(const struct lu_env *env,
2799                                     struct lfsck_component *com,
2800                                     struct lfsck_tgt_desc *ltd)
2801 {
2802         struct lfsck_layout             *lo     = com->lc_file_ram;
2803         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2804         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
2805         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2806         struct ost_id                   *oi     = &info->lti_oi;
2807         struct lu_fid                   *fid    = &info->lti_fid;
2808         struct dt_object                *obj;
2809         const struct dt_it_ops          *iops;
2810         struct dt_it                    *di;
2811         int                              rc     = 0;
2812         ENTRY;
2813
2814         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
2815                "scanning for OST%04x\n",
2816                lfsck_lfsck2name(lfsck), ltd->ltd_index);
2817
2818         ostid_set_seq(oi, FID_SEQ_IDIF);
2819         ostid_set_id(oi, 0);
2820         ostid_to_fid(fid, oi, ltd->ltd_index);
2821         obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
2822         if (unlikely(IS_ERR(obj)))
2823                 GOTO(log, rc = PTR_ERR(obj));
2824
2825         rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_orphan_features);
2826         if (rc != 0)
2827                 GOTO(put, rc);
2828
2829         iops = &obj->do_index_ops->dio_it;
2830         di = iops->init(env, obj, 0, BYPASS_CAPA);
2831         if (IS_ERR(di))
2832                 GOTO(put, rc = PTR_ERR(di));
2833
2834         rc = iops->load(env, di, 0);
2835         if (rc == -ESRCH) {
2836                 /* -ESRCH means that the orphan OST-objects rbtree has been
2837                  * cleanup because of the OSS server restart or other errors. */
2838                 lo->ll_flags |= LF_INCOMPLETE;
2839                 GOTO(fini, rc);
2840         }
2841
2842         if (rc == 0)
2843                 rc = iops->next(env, di);
2844         else if (rc > 0)
2845                 rc = 0;
2846
2847         if (rc < 0)
2848                 GOTO(fini, rc);
2849
2850         if (rc > 0)
2851                 GOTO(fini, rc = 0);
2852
2853         do {
2854                 struct dt_key           *key;
2855                 struct lu_orphan_rec    *rec = &info->lti_rec;
2856
2857                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
2858                     cfs_fail_val > 0) {
2859                         struct ptlrpc_thread    *thread = &lfsck->li_thread;
2860                         struct l_wait_info       lwi;
2861
2862                         lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
2863                                           NULL, NULL);
2864                         l_wait_event(thread->t_ctl_waitq,
2865                                      !thread_is_running(thread),
2866                                      &lwi);
2867                 }
2868
2869                 key = iops->key(env, di);
2870                 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
2871                 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
2872                 if (rc == 0)
2873                         rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
2874                                         &com->lc_fid_latest_scanned_phase2);
2875                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2876                         GOTO(fini, rc);
2877
2878                 lfsck_control_speed_by_self(com);
2879                 do {
2880                         rc = iops->next(env, di);
2881                 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
2882         } while (rc == 0);
2883
2884         GOTO(fini, rc);
2885
2886 fini:
2887         iops->put(env, di);
2888         iops->fini(env, di);
2889 put:
2890         lu_object_put(env, &obj->do_lu);
2891
2892 log:
2893         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
2894                "scanning for OST%04x: rc = %d\n",
2895                lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
2896
2897         return rc > 0 ? 0 : rc;
2898 }
2899
2900 /* For the MDT-object with dangling reference, we need to repare the
2901  * inconsistency according to the LFSCK sponsor's requirement:
2902  *
2903  * 1) Keep the inconsistency there and report the inconsistency case,
2904  *    then give the chance to the application to find related issues,
2905  *    and the users can make the decision about how to handle it with
2906  *    more human knownledge. (by default)
2907  *
2908  * 2) Re-create the missed OST-object with the FID/owner information. */
2909 static int lfsck_layout_repair_dangling(const struct lu_env *env,
2910                                         struct lfsck_component *com,
2911                                         struct lfsck_layout_req *llr,
2912                                         const struct lu_attr *pla)
2913 {
2914         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2915         struct filter_fid               *pfid   = &info->lti_new_pfid;
2916         struct dt_allocation_hint       *hint   = &info->lti_hint;
2917         struct lu_attr                  *cla    = &info->lti_la2;
2918         struct dt_object                *parent = llr->llr_parent->llo_obj;
2919         struct dt_object                *child  = llr->llr_child;
2920         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
2921         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
2922         struct thandle                  *handle;
2923         struct lu_buf                   *buf;
2924         struct lustre_handle             lh     = { 0 };
2925         int                              rc;
2926         bool                             create;
2927         ENTRY;
2928
2929         if (com->lc_lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ)
2930                 create = true;
2931         else
2932                 create = false;
2933
2934         if (!create)
2935                 GOTO(log, rc = 1);
2936
2937         memset(cla, 0, sizeof(*cla));
2938         cla->la_uid = pla->la_uid;
2939         cla->la_gid = pla->la_gid;
2940         cla->la_mode = S_IFREG | 0666;
2941         cla->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
2942                         LA_ATIME | LA_MTIME | LA_CTIME;
2943
2944         rc = lfsck_layout_lock(env, com, parent, &lh,
2945                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
2946         if (rc != 0)
2947                 GOTO(log, rc);
2948
2949         handle = dt_trans_create(env, dev);
2950         if (IS_ERR(handle))
2951                 GOTO(unlock1, rc = PTR_ERR(handle));
2952
2953         hint->dah_parent = NULL;
2954         hint->dah_mode = 0;
2955         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
2956         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
2957         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2958          * MDT-object's FID::f_ver, instead it is the OST-object index in its
2959          * parent MDT-object's layout EA. */
2960         pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
2961         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
2962
2963         rc = dt_declare_create(env, child, cla, hint, NULL, handle);
2964         if (rc != 0)
2965                 GOTO(stop, rc);
2966
2967         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
2968                                   LU_XATTR_CREATE, handle);
2969         if (rc != 0)
2970                 GOTO(stop, rc);
2971
2972         rc = dt_trans_start(env, dev, handle);
2973         if (rc != 0)
2974                 GOTO(stop, rc);
2975
2976         dt_read_lock(env, parent, 0);
2977         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
2978                 GOTO(unlock2, rc = 1);
2979
2980         rc = dt_create(env, child, cla, hint, NULL, handle);
2981         if (rc != 0)
2982                 GOTO(unlock2, rc);
2983
2984         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
2985                           handle, BYPASS_CAPA);
2986
2987         GOTO(unlock2, rc);
2988
2989 unlock2:
2990         dt_read_unlock(env, parent);
2991
2992 stop:
2993         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2994
2995 unlock1:
2996         lfsck_layout_unlock(&lh);
2997
2998 log:
2999         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found dangling "
3000                "reference for: parent "DFID", child "DFID", OST-index %u, "
3001                "stripe-index %u, owner %u/%u. %s: rc = %d\n",
3002                lfsck_lfsck2name(com->lc_lfsck), PFID(lfsck_dto2fid(parent)),
3003                PFID(lfsck_dto2fid(child)), llr->llr_ost_idx,
3004                llr->llr_lov_idx, pla->la_uid, pla->la_gid,
3005                create ? "Create the lost OST-object as required" :
3006                         "Keep the MDT-object there by default", rc);
3007
3008         return rc;
3009 }
3010
3011 /* If the OST-object does not recognize the MDT-object as its parent, and
3012  * there is no other MDT-object claims as its parent, then just trust the
3013  * given MDT-object as its parent. So update the OST-object filter_fid. */
3014 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
3015                                               struct lfsck_component *com,
3016                                               struct lfsck_layout_req *llr,
3017                                               const struct lu_attr *pla)
3018 {
3019         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3020         struct filter_fid               *pfid   = &info->lti_new_pfid;
3021         struct lu_attr                  *tla    = &info->lti_la3;
3022         struct dt_object                *parent = llr->llr_parent->llo_obj;
3023         struct dt_object                *child  = llr->llr_child;
3024         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
3025         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
3026         struct thandle                  *handle;
3027         struct lu_buf                   *buf;
3028         struct lustre_handle             lh     = { 0 };
3029         int                              rc;
3030         ENTRY;
3031
3032         rc = lfsck_layout_lock(env, com, parent, &lh,
3033                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
3034         if (rc != 0)
3035                 GOTO(log, rc);
3036
3037         handle = dt_trans_create(env, dev);
3038         if (IS_ERR(handle))
3039                 GOTO(unlock1, rc = PTR_ERR(handle));
3040
3041         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
3042         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
3043         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3044          * MDT-object's FID::f_ver, instead it is the OST-object index in its
3045          * parent MDT-object's layout EA. */
3046         pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
3047         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
3048
3049         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3050         if (rc != 0)
3051                 GOTO(stop, rc);
3052
3053         tla->la_valid = LA_UID | LA_GID;
3054         tla->la_uid = pla->la_uid;
3055         tla->la_gid = pla->la_gid;
3056         rc = dt_declare_attr_set(env, child, tla, handle);
3057         if (rc != 0)
3058                 GOTO(stop, rc);
3059
3060         rc = dt_trans_start(env, dev, handle);
3061         if (rc != 0)
3062                 GOTO(stop, rc);
3063
3064         dt_write_lock(env, parent, 0);
3065         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
3066                 GOTO(unlock2, rc = 1);
3067
3068         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle,
3069                           BYPASS_CAPA);
3070         if (rc != 0)
3071                 GOTO(unlock2, rc);
3072
3073         /* Get the latest parent's owner. */
3074         rc = dt_attr_get(env, parent, tla, BYPASS_CAPA);
3075         if (rc != 0)
3076                 GOTO(unlock2, rc);
3077
3078         tla->la_valid = LA_UID | LA_GID;
3079         rc = dt_attr_set(env, child, tla, handle, BYPASS_CAPA);
3080
3081         GOTO(unlock2, rc);
3082
3083 unlock2:
3084         dt_write_unlock(env, parent);
3085
3086 stop:
3087         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3088
3089 unlock1:
3090         lfsck_layout_unlock(&lh);
3091
3092 log:
3093         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired unmatched "
3094                "MDT-OST pair for: parent "DFID", child "DFID", OST-index %u, "
3095                "stripe-index %u, owner %u/%u: rc = %d\n",
3096                lfsck_lfsck2name(com->lc_lfsck), PFID(lfsck_dto2fid(parent)),
3097                PFID(lfsck_dto2fid(child)), llr->llr_ost_idx, llr->llr_lov_idx,
3098                pla->la_uid, pla->la_gid, rc);
3099
3100         return rc;
3101 }
3102
3103 /* If there are more than one MDT-objects claim as the OST-object's parent,
3104  * and the OST-object only recognizes one of them, then we need to generate
3105  * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
3106 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
3107                                                    struct lfsck_component *com,
3108                                                    struct lfsck_layout_req *llr,
3109                                                    struct lu_attr *la,
3110                                                    struct lu_buf *buf)
3111 {
3112         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3113         struct dt_allocation_hint       *hint   = &info->lti_hint;
3114         struct dt_object_format         *dof    = &info->lti_dof;
3115         struct dt_device                *pdev   = com->lc_lfsck->li_next;
3116         struct ost_id                   *oi     = &info->lti_oi;
3117         struct dt_object                *parent = llr->llr_parent->llo_obj;
3118         struct dt_device                *cdev   = lfsck_obj2dt_dev(llr->llr_child);
3119         struct dt_object                *child  = NULL;
3120         struct lu_device                *d      = &cdev->dd_lu_dev;
3121         struct lu_object                *o      = NULL;
3122         struct thandle                  *handle;
3123         struct lov_mds_md_v1            *lmm;
3124         struct lov_ost_data_v1          *objs;
3125         struct lustre_handle             lh     = { 0 };
3126         __u32                            magic;
3127         int                              rc;
3128         ENTRY;
3129
3130         rc = lfsck_layout_lock(env, com, parent, &lh,
3131                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
3132         if (rc != 0)
3133                 GOTO(log, rc);
3134
3135         handle = dt_trans_create(env, pdev);
3136         if (IS_ERR(handle))
3137                 GOTO(unlock1, rc = PTR_ERR(handle));
3138
3139         o = lu_object_anon(env, d, NULL);
3140         if (IS_ERR(o))
3141                 GOTO(stop, rc = PTR_ERR(o));
3142
3143         child = container_of(o, struct dt_object, do_lu);
3144         o = lu_object_locate(o->lo_header, d->ld_type);
3145         if (unlikely(o == NULL))
3146                 GOTO(stop, rc = -EINVAL);
3147
3148         child = container_of(o, struct dt_object, do_lu);
3149         la->la_valid = LA_UID | LA_GID;
3150         hint->dah_parent = NULL;
3151         hint->dah_mode = 0;
3152         dof->dof_type = DFT_REGULAR;
3153         rc = dt_declare_create(env, child, la, NULL, NULL, handle);
3154         if (rc != 0)
3155                 GOTO(stop, rc);
3156
3157         rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3158                                   LU_XATTR_REPLACE, handle);
3159         if (rc != 0)
3160                 GOTO(stop, rc);
3161
3162         rc = dt_trans_start(env, pdev, handle);
3163         if (rc != 0)
3164                 GOTO(stop, rc);
3165
3166         dt_write_lock(env, parent, 0);
3167         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
3168                 GOTO(unlock2, rc = 0);
3169
3170         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV, BYPASS_CAPA);
3171         if (unlikely(rc == 0 || rc == -ENODATA || rc == -ERANGE))
3172                 GOTO(unlock2, rc = 0);
3173
3174         lmm = buf->lb_buf;
3175         /* Someone change layout during the LFSCK, no need to repair then. */
3176         if (le16_to_cpu(lmm->lmm_layout_gen) != llr->llr_parent->llo_gen)
3177                 GOTO(unlock2, rc = 0);
3178
3179         rc = dt_create(env, child, la, hint, dof, handle);
3180         if (rc != 0)
3181                 GOTO(unlock2, rc);
3182
3183         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
3184          * been verified in lfsck_layout_verify_header() already. If some
3185          * new magic introduced in the future, then layout LFSCK needs to
3186          * be updated also. */
3187         magic = le32_to_cpu(lmm->lmm_magic);
3188         if (magic == LOV_MAGIC_V1) {
3189                 objs = &lmm->lmm_objects[0];
3190         } else {
3191                 LASSERT(magic == LOV_MAGIC_V3);
3192                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3193         }
3194
3195         lmm->lmm_layout_gen = cpu_to_le16(llr->llr_parent->llo_gen + 1);
3196         fid_to_ostid(lu_object_fid(&child->do_lu), oi);
3197         ostid_cpu_to_le(oi, &objs[llr->llr_lov_idx].l_ost_oi);
3198         objs[llr->llr_lov_idx].l_ost_gen = cpu_to_le32(0);
3199         objs[llr->llr_lov_idx].l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
3200         rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3201                           LU_XATTR_REPLACE, handle, BYPASS_CAPA);
3202
3203         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
3204
3205 unlock2:
3206         dt_write_unlock(env, parent);
3207
3208 stop:
3209         if (child != NULL)
3210                 lu_object_put(env, &child->do_lu);
3211
3212         dt_trans_stop(env, pdev, handle);
3213
3214 unlock1:
3215         lfsck_layout_unlock(&lh);
3216
3217 log:
3218         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired multiple "
3219                "references for: parent "DFID", OST-index %u, stripe-index %u, "
3220                "owner %u/%u: rc = %d\n",
3221                lfsck_lfsck2name(com->lc_lfsck), PFID(lfsck_dto2fid(parent)),
3222                llr->llr_ost_idx, llr->llr_lov_idx, la->la_uid, la->la_gid, rc);
3223
3224         return rc;
3225 }
3226
3227 /* If the MDT-object and the OST-object have different owner information,
3228  * then trust the MDT-object, because the normal chown/chgrp handle order
3229  * is from MDT to OST, and it is possible that some chown/chgrp operation
3230  * is partly done. */
3231 static int lfsck_layout_repair_owner(const struct lu_env *env,
3232                                      struct lfsck_component *com,
3233                                      struct lfsck_layout_req *llr,
3234                                      struct lu_attr *pla)
3235 {
3236         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3237         struct lu_attr                  *tla    = &info->lti_la3;
3238         struct dt_object                *parent = llr->llr_parent->llo_obj;
3239         struct dt_object                *child  = llr->llr_child;
3240         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
3241         struct thandle                  *handle;
3242         int                              rc;
3243         ENTRY;
3244
3245         handle = dt_trans_create(env, dev);
3246         if (IS_ERR(handle))
3247                 GOTO(log, rc = PTR_ERR(handle));
3248
3249         tla->la_uid = pla->la_uid;
3250         tla->la_gid = pla->la_gid;
3251         tla->la_valid = LA_UID | LA_GID;
3252         rc = dt_declare_attr_set(env, child, tla, handle);
3253         if (rc != 0)
3254                 GOTO(stop, rc);
3255
3256         rc = dt_trans_start(env, dev, handle);
3257         if (rc != 0)
3258                 GOTO(stop, rc);
3259
3260         /* Use the dt_object lock to serialize with destroy and attr_set. */
3261         dt_read_lock(env, parent, 0);
3262         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
3263                 GOTO(unlock, rc = 1);
3264
3265         /* Get the latest parent's owner. */
3266         rc = dt_attr_get(env, parent, tla, BYPASS_CAPA);
3267         if (rc != 0)
3268                 GOTO(unlock, rc);
3269
3270         /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
3271         if (unlikely(tla->la_uid != pla->la_uid ||
3272                      tla->la_gid != pla->la_gid))
3273                 GOTO(unlock, rc = 1);
3274
3275         tla->la_valid = LA_UID | LA_GID;
3276         rc = dt_attr_set(env, child, tla, handle, BYPASS_CAPA);
3277
3278         GOTO(unlock, rc);
3279
3280 unlock:
3281         dt_read_unlock(env, parent);
3282
3283 stop:
3284         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3285
3286 log:
3287         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired inconsistent "
3288                "file owner for: parent "DFID", child "DFID", OST-index %u, "
3289                "stripe-index %u, owner %u/%u: rc = %d\n",
3290                lfsck_lfsck2name(com->lc_lfsck), PFID(lfsck_dto2fid(parent)),
3291                PFID(lfsck_dto2fid(child)), llr->llr_ost_idx, llr->llr_lov_idx,
3292                pla->la_uid, pla->la_gid, rc);
3293
3294         return rc;
3295 }
3296
3297 /* Check whether the OST-object correctly back points to the
3298  * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
3299 static int lfsck_layout_check_parent(const struct lu_env *env,
3300                                      struct lfsck_component *com,
3301                                      struct dt_object *parent,
3302                                      const struct lu_fid *pfid,
3303                                      const struct lu_fid *cfid,
3304                                      const struct lu_attr *pla,
3305                                      const struct lu_attr *cla,
3306                                      struct lfsck_layout_req *llr,
3307                                      struct lu_buf *lov_ea, __u32 idx)
3308 {
3309         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3310         struct lu_buf                   *buf    = &info->lti_big_buf;
3311         struct dt_object                *tobj;
3312         struct lov_mds_md_v1            *lmm;
3313         struct lov_ost_data_v1          *objs;
3314         int                              rc;
3315         int                              i;
3316         __u32                            magic;
3317         __u16                            count;
3318         ENTRY;
3319
3320         if (fid_is_zero(pfid)) {
3321                 /* client never wrote. */
3322                 if (cla->la_size == 0 && cla->la_blocks == 0) {
3323                         if (unlikely(cla->la_uid != pla->la_uid ||
3324                                      cla->la_gid != pla->la_gid))
3325                                 RETURN (LLIT_INCONSISTENT_OWNER);
3326
3327                         RETURN(0);
3328                 }
3329
3330                 RETURN(LLIT_UNMATCHED_PAIR);
3331         }
3332
3333         if (unlikely(!fid_is_sane(pfid)))
3334                 RETURN(LLIT_UNMATCHED_PAIR);
3335
3336         if (lu_fid_eq(pfid, lu_object_fid(&parent->do_lu))) {
3337                 if (llr->llr_lov_idx == idx)
3338                         RETURN(0);
3339
3340                 RETURN(LLIT_UNMATCHED_PAIR);
3341         }
3342
3343         tobj = lfsck_object_find(env, com->lc_lfsck, pfid);
3344         if (tobj == NULL)
3345                 RETURN(LLIT_UNMATCHED_PAIR);
3346
3347         if (IS_ERR(tobj))
3348                 RETURN(PTR_ERR(tobj));
3349
3350         if (!dt_object_exists(tobj))
3351                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3352
3353         /* Load the tobj's layout EA, in spite of it is a local MDT-object or
3354          * remote one on another MDT. Then check whether the given OST-object
3355          * is in such layout. If yes, it is multiple referenced, otherwise it
3356          * is unmatched referenced case. */
3357         rc = lfsck_layout_get_lovea(env, tobj, buf, NULL);
3358         if (rc == 0)
3359                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3360
3361         if (rc < 0)
3362                 GOTO(out, rc);
3363
3364         lmm = buf->lb_buf;
3365         magic = le32_to_cpu(lmm->lmm_magic);
3366         if (magic == LOV_MAGIC_V1) {
3367                 objs = &lmm->lmm_objects[0];
3368         } else {
3369                 LASSERT(magic == LOV_MAGIC_V3);
3370                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3371         }
3372
3373         count = le16_to_cpu(lmm->lmm_stripe_count);
3374         for (i = 0; i < count; i++, objs++) {
3375                 struct lu_fid           *tfid   = &info->lti_fid2;
3376                 struct ost_id           *oi     = &info->lti_oi;
3377
3378                 if (lovea_slot_is_dummy(objs))
3379                         continue;
3380
3381                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3382                 ostid_to_fid(tfid, oi, le32_to_cpu(objs->l_ost_idx));
3383                 if (lu_fid_eq(cfid, tfid)) {
3384                         *lov_ea = *buf;
3385
3386                         GOTO(out, rc = LLIT_MULTIPLE_REFERENCED);
3387                 }
3388         }
3389
3390         GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3391
3392 out:
3393         lfsck_object_put(env, tobj);
3394
3395         return rc;
3396 }
3397
3398 static int lfsck_layout_assistant_handle_one(const struct lu_env *env,
3399                                              struct lfsck_component *com,
3400                                              struct lfsck_layout_req *llr)
3401 {
3402         struct lfsck_layout                  *lo     = com->lc_file_ram;
3403         struct lfsck_thread_info             *info   = lfsck_env_info(env);
3404         struct filter_fid_old                *pea    = &info->lti_old_pfid;
3405         struct lu_fid                        *pfid   = &info->lti_fid;
3406         struct lu_buf                        *buf    = NULL;
3407         struct dt_object                     *parent = llr->llr_parent->llo_obj;
3408         struct dt_object                     *child  = llr->llr_child;
3409         struct lu_attr                       *pla    = &info->lti_la;
3410         struct lu_attr                       *cla    = &info->lti_la2;
3411         struct lfsck_instance                *lfsck  = com->lc_lfsck;
3412         struct lfsck_bookmark                *bk     = &lfsck->li_bookmark_ram;
3413         enum lfsck_layout_inconsistency_type  type   = LLIT_NONE;
3414         __u32                                 idx    = 0;
3415         int                                   rc;
3416         ENTRY;
3417
3418         rc = dt_attr_get(env, parent, pla, BYPASS_CAPA);
3419         if (rc != 0) {
3420                 if (lu_object_is_dying(parent->do_lu.lo_header))
3421                         RETURN(0);
3422
3423                 GOTO(out, rc);
3424         }
3425
3426         rc = dt_attr_get(env, child, cla, BYPASS_CAPA);
3427         if (rc == -ENOENT) {
3428                 if (lu_object_is_dying(parent->do_lu.lo_header))
3429                         RETURN(0);
3430
3431                 type = LLIT_DANGLING;
3432                 goto repair;
3433         }
3434
3435         if (rc != 0)
3436                 GOTO(out, rc);
3437
3438         buf = lfsck_buf_get(env, pea, sizeof(struct filter_fid_old));
3439         rc= dt_xattr_get(env, child, buf, XATTR_NAME_FID, BYPASS_CAPA);
3440         if (unlikely(rc >= 0 && rc != sizeof(struct filter_fid_old) &&
3441                      rc != sizeof(struct filter_fid))) {
3442                 type = LLIT_UNMATCHED_PAIR;
3443                 goto repair;
3444         }
3445
3446         if (rc < 0 && rc != -ENODATA)
3447                 GOTO(out, rc);
3448
3449         if (rc == -ENODATA) {
3450                 fid_zero(pfid);
3451         } else {
3452                 fid_le_to_cpu(pfid, &pea->ff_parent);
3453                 /* Currently, the filter_fid::ff_parent::f_ver is not the
3454                  * real parent MDT-object's FID::f_ver, instead it is the
3455                  * OST-object index in its parent MDT-object's layout EA. */
3456                 idx = pfid->f_stripe_idx;
3457                 pfid->f_ver = 0;
3458         }
3459
3460         rc = lfsck_layout_check_parent(env, com, parent, pfid,
3461                                        lu_object_fid(&child->do_lu),
3462                                        pla, cla, llr, buf, idx);
3463         if (rc > 0) {
3464                 type = rc;
3465                 goto repair;
3466         }
3467
3468         if (rc < 0)
3469                 GOTO(out, rc);
3470
3471         if (unlikely(cla->la_uid != pla->la_uid ||
3472                      cla->la_gid != pla->la_gid)) {
3473                 type = LLIT_INCONSISTENT_OWNER;
3474                 goto repair;
3475         }
3476
3477 repair:
3478         if (bk->lb_param & LPF_DRYRUN) {
3479                 if (type != LLIT_NONE)
3480                         GOTO(out, rc = 1);
3481                 else
3482                         GOTO(out, rc = 0);
3483         }
3484
3485         switch (type) {
3486         case LLIT_DANGLING:
3487                 rc = lfsck_layout_repair_dangling(env, com, llr, pla);
3488                 break;
3489         case LLIT_UNMATCHED_PAIR:
3490                 rc = lfsck_layout_repair_unmatched_pair(env, com, llr, pla);
3491                 break;
3492         case LLIT_MULTIPLE_REFERENCED:
3493                 rc = lfsck_layout_repair_multiple_references(env, com, llr,
3494                                                              pla, buf);
3495                 break;
3496         case LLIT_INCONSISTENT_OWNER:
3497                 rc = lfsck_layout_repair_owner(env, com, llr, pla);
3498                 break;
3499         default:
3500                 rc = 0;
3501                 break;
3502         }
3503
3504         GOTO(out, rc);
3505
3506 out:
3507         down_write(&com->lc_sem);
3508         if (rc < 0) {
3509                 struct lfsck_layout_master_data *llmd = com->lc_data;
3510
3511                 if (unlikely(llmd->llmd_exit)) {
3512                         rc = 0;
3513                 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
3514                            rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
3515                            rc == -EHOSTUNREACH) {
3516                         /* If cannot touch the target server,
3517                          * mark the LFSCK as INCOMPLETE. */
3518                         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
3519                                "talk with OST %x: rc = %d\n",
3520                                lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
3521                         lo->ll_flags |= LF_INCOMPLETE;
3522                         lo->ll_objs_skipped++;
3523                         rc = 0;
3524                 } else {
3525                         lfsck_layout_record_failure(env, lfsck, lo);
3526                 }
3527         } else if (rc > 0) {
3528                 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
3529                          "unknown type = %d\n", type);
3530
3531                 lo->ll_objs_repaired[type - 1]++;
3532                 if (bk->lb_param & LPF_DRYRUN &&
3533                     unlikely(lo->ll_pos_first_inconsistent == 0))
3534                         lo->ll_pos_first_inconsistent =
3535                         lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
3536                                                         lfsck->li_di_oit);
3537         }
3538         up_write(&com->lc_sem);
3539
3540         return rc;
3541 }
3542
3543 static int lfsck_layout_assistant(void *args)
3544 {
3545         struct lfsck_thread_args        *lta     = args;
3546         struct lu_env                   *env     = &lta->lta_env;
3547         struct lfsck_component          *com     = lta->lta_com;
3548         struct lfsck_instance           *lfsck   = lta->lta_lfsck;
3549         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
3550         struct lfsck_position           *pos     = &com->lc_pos_start;
3551         struct lfsck_thread_info        *info    = lfsck_env_info(env);
3552         struct lfsck_request            *lr      = &info->lti_lr;
3553         struct lfsck_layout_master_data *llmd    = com->lc_data;
3554         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
3555         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
3556         struct lfsck_layout_req         *llr;
3557         struct l_wait_info               lwi     = { 0 };
3558         int                              rc      = 0;
3559         int                              rc1     = 0;
3560         ENTRY;
3561
3562         memset(lr, 0, sizeof(*lr));
3563         lr->lr_event = LE_START;
3564         lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
3565                        LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ;
3566         lr->lr_speed = bk->lb_speed_limit;
3567         lr->lr_version = bk->lb_version;
3568         lr->lr_param = bk->lb_param;
3569         lr->lr_async_windows = bk->lb_async_windows;
3570         lr->lr_flags = LEF_TO_OST;
3571         if (pos->lp_oit_cookie <= 1)
3572                 lr->lr_param |= LPF_RESET;
3573
3574         rc = lfsck_layout_master_notify_others(env, com, lr);
3575         if (rc != 0) {
3576                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to notify "
3577                        "others for LFSCK start: rc = %d\n",
3578                        lfsck_lfsck2name(lfsck), rc);
3579                 GOTO(fini, rc);
3580         }
3581
3582         spin_lock(&llmd->llmd_lock);
3583         thread_set_flags(athread, SVC_RUNNING);
3584         spin_unlock(&llmd->llmd_lock);
3585         wake_up_all(&mthread->t_ctl_waitq);
3586
3587         while (1) {
3588                 while (!list_empty(&llmd->llmd_req_list)) {
3589                         bool wakeup = false;
3590
3591                         if (unlikely(llmd->llmd_exit ||
3592                                      !thread_is_running(mthread)))
3593                                 GOTO(cleanup1, rc = llmd->llmd_post_result);
3594
3595                         llr = list_entry(llmd->llmd_req_list.next,
3596                                          struct lfsck_layout_req,
3597                                          llr_list);
3598                         /* Only the lfsck_layout_assistant thread itself can
3599                          * remove the "llr" from the head of the list, LFSCK
3600                          * engine thread only inserts other new "lld" at the
3601                          * end of the list. So it is safe to handle current
3602                          * "llr" without the spin_lock. */
3603                         rc = lfsck_layout_assistant_handle_one(env, com, llr);
3604                         spin_lock(&llmd->llmd_lock);
3605                         list_del_init(&llr->llr_list);
3606                         llmd->llmd_prefetched--;
3607                         /* Wake up the main engine thread only when the list
3608                          * is empty or half of the prefetched items have been
3609                          * handled to avoid too frequent thread schedule. */
3610                         if (llmd->llmd_prefetched == 0 ||
3611                             (bk->lb_async_windows != 0 &&
3612                              bk->lb_async_windows / 2 ==
3613                              llmd->llmd_prefetched))
3614                                 wakeup = true;
3615                         spin_unlock(&llmd->llmd_lock);
3616                         if (wakeup)
3617                                 wake_up_all(&mthread->t_ctl_waitq);
3618
3619                         lfsck_layout_req_fini(env, llr);
3620                         if (rc < 0 && bk->lb_param & LPF_FAILOUT)
3621                                 GOTO(cleanup1, rc);
3622                 }
3623
3624                 l_wait_event(athread->t_ctl_waitq,
3625                              !lfsck_layout_req_empty(llmd) ||
3626                              llmd->llmd_exit ||
3627                              llmd->llmd_to_post ||
3628                              llmd->llmd_to_double_scan,
3629                              &lwi);
3630
3631                 if (unlikely(llmd->llmd_exit))
3632                         GOTO(cleanup1, rc = llmd->llmd_post_result);
3633
3634                 if (!list_empty(&llmd->llmd_req_list))
3635                         continue;
3636
3637                 if (llmd->llmd_to_post) {
3638                         llmd->llmd_to_post = 0;
3639                         LASSERT(llmd->llmd_post_result > 0);
3640
3641                         memset(lr, 0, sizeof(*lr));
3642                         lr->lr_event = LE_PHASE1_DONE;
3643                         lr->lr_status = llmd->llmd_post_result;
3644                         rc = lfsck_layout_master_notify_others(env, com, lr);
3645                         if (rc != 0)
3646                                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant "
3647                                        "failed to notify others for LFSCK "
3648                                        "post: rc = %d\n",
3649                                        lfsck_lfsck2name(lfsck), rc);
3650
3651                         /* Wakeup the master engine to go ahead. */
3652                         wake_up_all(&mthread->t_ctl_waitq);
3653                 }
3654
3655                 if (llmd->llmd_to_double_scan) {
3656                         llmd->llmd_to_double_scan = 0;
3657                         atomic_inc(&lfsck->li_double_scan_count);
3658                         llmd->llmd_in_double_scan = 1;
3659                         wake_up_all(&mthread->t_ctl_waitq);
3660
3661                         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant phase2 "
3662                                "scan start\n", lfsck_lfsck2name(lfsck));
3663
3664                         com->lc_new_checked = 0;
3665                         com->lc_new_scanned = 0;
3666                         com->lc_time_last_checkpoint = cfs_time_current();
3667                         com->lc_time_next_checkpoint =
3668                                 com->lc_time_last_checkpoint +
3669                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
3670
3671                         /* flush all async updating before handling orphan. */
3672                         dt_sync(env, lfsck->li_next);
3673
3674                         while (llmd->llmd_in_double_scan) {
3675                                 struct lfsck_tgt_descs  *ltds =
3676                                                         &lfsck->li_ost_descs;
3677                                 struct lfsck_tgt_desc   *ltd;
3678
3679                                 rc = lfsck_layout_master_query_others(env, com);
3680                                 if (lfsck_layout_master_to_orphan(llmd))
3681                                         goto orphan;
3682
3683                                 if (rc < 0)
3684                                         GOTO(cleanup2, rc);
3685
3686                                 /* Pull LFSCK status on related targets once
3687                                  * per 30 seconds if we are not notified. */
3688                                 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
3689                                                            cfs_time_seconds(1),
3690                                                            NULL, NULL);
3691                                 rc = l_wait_event(athread->t_ctl_waitq,
3692                                         lfsck_layout_master_to_orphan(llmd) ||
3693                                         llmd->llmd_exit ||
3694                                         !thread_is_running(mthread),
3695                                         &lwi);
3696
3697                                 if (unlikely(llmd->llmd_exit ||
3698                                              !thread_is_running(mthread)))
3699                                         GOTO(cleanup2, rc = 0);
3700
3701                                 if (rc == -ETIMEDOUT)
3702                                         continue;
3703
3704                                 if (rc < 0)
3705                                         GOTO(cleanup2, rc);
3706
3707 orphan:
3708                                 spin_lock(&ltds->ltd_lock);
3709                                 while (!list_empty(
3710                                                 &llmd->llmd_ost_phase2_list)) {
3711                                         ltd = list_entry(
3712                                               llmd->llmd_ost_phase2_list.next,
3713                                               struct lfsck_tgt_desc,
3714                                               ltd_layout_phase_list);
3715                                         list_del_init(
3716                                                 &ltd->ltd_layout_phase_list);
3717                                         spin_unlock(&ltds->ltd_lock);
3718
3719                                         if (bk->lb_param & LPF_ALL_TGT) {
3720                                                 rc = lfsck_layout_scan_orphan(
3721                                                                 env, com, ltd);
3722                                                 if (rc != 0 &&
3723                                                     bk->lb_param & LPF_FAILOUT)
3724                                                         GOTO(cleanup2, rc);
3725                                         }
3726
3727                                         if (unlikely(llmd->llmd_exit ||
3728                                                 !thread_is_running(mthread)))
3729                                                 GOTO(cleanup2, rc = 0);
3730
3731                                         spin_lock(&ltds->ltd_lock);
3732                                 }
3733
3734                                 if (list_empty(&llmd->llmd_ost_phase1_list)) {
3735                                         spin_unlock(&ltds->ltd_lock);
3736                                         GOTO(cleanup2, rc = 1);
3737                                 }
3738                                 spin_unlock(&ltds->ltd_lock);
3739                         }
3740                 }
3741         }
3742
3743 cleanup1:
3744         /* Cleanup the unfinished requests. */
3745         spin_lock(&llmd->llmd_lock);
3746         if (rc < 0)
3747                 llmd->llmd_assistant_status = rc;
3748
3749         while (!list_empty(&llmd->llmd_req_list)) {
3750                 llr = list_entry(llmd->llmd_req_list.next,
3751                                  struct lfsck_layout_req,
3752                                  llr_list);
3753                 list_del_init(&llr->llr_list);
3754                 llmd->llmd_prefetched--;
3755                 spin_unlock(&llmd->llmd_lock);
3756                 lfsck_layout_req_fini(env, llr);
3757                 spin_lock(&llmd->llmd_lock);
3758         }
3759         spin_unlock(&llmd->llmd_lock);
3760
3761         LASSERTF(llmd->llmd_prefetched == 0, "unmatched prefeteched objs %d\n",
3762                  llmd->llmd_prefetched);
3763
3764 cleanup2:
3765         memset(lr, 0, sizeof(*lr));
3766         if (rc > 0) {
3767                 lr->lr_event = LE_PHASE2_DONE;
3768                 lr->lr_status = rc;
3769         } else if (rc == 0) {
3770                 if (lfsck->li_flags & LPF_ALL_TGT) {
3771                         lr->lr_event = LE_STOP;
3772                         lr->lr_status = LS_STOPPED;
3773                 } else {
3774                         lr->lr_event = LE_PEER_EXIT;
3775                         switch (lfsck->li_status) {
3776                         case LS_PAUSED:
3777                         case LS_CO_PAUSED:
3778                                 lr->lr_status = LS_CO_PAUSED;
3779                                 break;
3780                         case LS_STOPPED:
3781                         case LS_CO_STOPPED:
3782                                 lr->lr_status = LS_CO_STOPPED;
3783                                 break;
3784                         default:
3785                                 CDEBUG(D_LFSCK, "%s: unknown status: rc = %d\n",
3786                                        lfsck_lfsck2name(lfsck),
3787                                        lfsck->li_status);
3788                                 lr->lr_status = LS_CO_FAILED;
3789                                 break;
3790                         }
3791                 }
3792         } else {
3793                 if (lfsck->li_flags & LPF_ALL_TGT) {
3794                         lr->lr_event = LE_STOP;
3795                         lr->lr_status = LS_FAILED;
3796                 } else {
3797                         lr->lr_event = LE_PEER_EXIT;
3798                         lr->lr_status = LS_CO_FAILED;
3799                 }
3800         }
3801
3802         rc1 = lfsck_layout_master_notify_others(env, com, lr);
3803         if (rc1 != 0) {
3804                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to "
3805                        "notify others for LFSCK quit: rc = %d\n",
3806                        lfsck_lfsck2name(lfsck), rc1);
3807                 rc = rc1;
3808         }
3809
3810         /* flush all async updating before exit. */
3811         dt_sync(env, lfsck->li_next);
3812
3813         /* Under force exit case, some requests may be just freed without
3814          * verification, those objects should be re-handled when next run.
3815          * So not update the on-disk tracing file under such case. */
3816         if (llmd->llmd_in_double_scan) {
3817                 struct lfsck_layout *lo = com->lc_file_ram;
3818
3819                 if (!llmd->llmd_exit)
3820                         rc1 = lfsck_layout_double_scan_result(env, com, rc);
3821
3822                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant phase2 scan "
3823                        "finished, status %d: rc = %d\n",
3824                        lfsck_lfsck2name(lfsck), lo->ll_status, rc1);
3825         }
3826
3827 fini:
3828         if (llmd->llmd_in_double_scan)
3829                 atomic_dec(&lfsck->li_double_scan_count);
3830
3831         spin_lock(&llmd->llmd_lock);
3832         llmd->llmd_assistant_status = (rc1 != 0 ? rc1 : rc);
3833         thread_set_flags(athread, SVC_STOPPED);
3834         wake_up_all(&mthread->t_ctl_waitq);
3835         spin_unlock(&llmd->llmd_lock);
3836         lfsck_thread_args_fini(lta);
3837
3838         return rc;
3839 }
3840
3841 static int
3842 lfsck_layout_slave_async_interpret(const struct lu_env *env,
3843                                    struct ptlrpc_request *req,
3844                                    void *args, int rc)
3845 {
3846         struct lfsck_layout_slave_async_args *llsaa = args;
3847         struct obd_export                    *exp   = llsaa->llsaa_exp;
3848         struct lfsck_component               *com   = llsaa->llsaa_com;
3849         struct lfsck_layout_slave_target     *llst  = llsaa->llsaa_llst;
3850         struct lfsck_layout_slave_data       *llsd  = com->lc_data;
3851         struct lfsck_reply                   *lr    = NULL;
3852         bool                                  done  = false;
3853
3854         if (rc != 0) {
3855                 /* It is quite probably caused by target crash,
3856                  * to make the LFSCK can go ahead, assume that
3857                  * the target finished the LFSCK prcoessing. */
3858                 done = true;
3859         } else {
3860                 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
3861                 if (lr->lr_status != LS_SCANNING_PHASE1 &&
3862                     lr->lr_status != LS_SCANNING_PHASE2)
3863                         done = true;
3864         }
3865
3866         if (done) {
3867                 CDEBUG(D_LFSCK, "%s: layout LFSCK slave gets the MDT %x "
3868                        "status %d\n", lfsck_lfsck2name(com->lc_lfsck),
3869                        llst->llst_index, lr != NULL ? lr->lr_status : rc);
3870
3871                 lfsck_layout_llst_del(llsd, llst);
3872         }
3873
3874         lfsck_layout_llst_put(llst);
3875         lfsck_component_put(env, com);
3876         class_export_put(exp);
3877
3878         return 0;
3879 }
3880
3881 static int lfsck_layout_async_query(const struct lu_env *env,
3882                                     struct lfsck_component *com,
3883                                     struct obd_export *exp,
3884                                     struct lfsck_layout_slave_target *llst,
3885                                     struct lfsck_request *lr,
3886                                     struct ptlrpc_request_set *set)
3887 {
3888         struct lfsck_layout_slave_async_args *llsaa;
3889         struct ptlrpc_request                *req;
3890         struct lfsck_request                 *tmp;
3891         int                                   rc;
3892         ENTRY;
3893
3894         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
3895         if (req == NULL)
3896                 RETURN(-ENOMEM);
3897
3898         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
3899         if (rc != 0) {
3900                 ptlrpc_request_free(req);
3901                 RETURN(rc);
3902         }
3903
3904         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3905         *tmp = *lr;
3906         ptlrpc_request_set_replen(req);
3907
3908         llsaa = ptlrpc_req_async_args(req);
3909         llsaa->llsaa_exp = exp;
3910         llsaa->llsaa_com = lfsck_component_get(com);
3911         llsaa->llsaa_llst = llst;
3912         req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
3913         ptlrpc_set_add_req(set, req);
3914
3915         RETURN(0);
3916 }
3917
3918 static int lfsck_layout_async_notify(const struct lu_env *env,
3919                                      struct obd_export *exp,
3920                                      struct lfsck_request *lr,
3921                                      struct ptlrpc_request_set *set)
3922 {
3923         struct ptlrpc_request   *req;
3924         struct lfsck_request    *tmp;
3925         int                      rc;
3926         ENTRY;
3927
3928         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
3929         if (req == NULL)
3930                 RETURN(-ENOMEM);
3931
3932         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
3933         if (rc != 0) {
3934                 ptlrpc_request_free(req);
3935                 RETURN(rc);
3936         }
3937
3938         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3939         *tmp = *lr;
3940         ptlrpc_request_set_replen(req);
3941         ptlrpc_set_add_req(set, req);
3942
3943         RETURN(0);
3944 }
3945
3946 static int
3947 lfsck_layout_slave_query_master(const struct lu_env *env,
3948                                 struct lfsck_component *com)
3949 {
3950         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
3951         struct lfsck_instance            *lfsck = com->lc_lfsck;
3952         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3953         struct lfsck_layout_slave_target *llst;
3954         struct obd_export                *exp;
3955         struct ptlrpc_request_set        *set;
3956         int                               rc    = 0;
3957         int                               rc1   = 0;
3958         ENTRY;
3959
3960         set = ptlrpc_prep_set();
3961         if (set == NULL)
3962                 GOTO(log, rc = -ENOMEM);
3963
3964         memset(lr, 0, sizeof(*lr));
3965         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
3966         lr->lr_event = LE_QUERY;
3967         lr->lr_active = LFSCK_TYPE_LAYOUT;
3968
3969         llsd->llsd_touch_gen++;
3970         spin_lock(&llsd->llsd_lock);
3971         while (!list_empty(&llsd->llsd_master_list)) {
3972                 llst = list_entry(llsd->llsd_master_list.next,
3973                                   struct lfsck_layout_slave_target,
3974                                   llst_list);
3975                 if (llst->llst_gen == llsd->llsd_touch_gen)
3976                         break;
3977
3978                 llst->llst_gen = llsd->llsd_touch_gen;
3979                 list_del(&llst->llst_list);
3980                 list_add_tail(&llst->llst_list,
3981                               &llsd->llsd_master_list);
3982                 atomic_inc(&llst->llst_ref);
3983                 spin_unlock(&llsd->llsd_lock);
3984
3985                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
3986                                                llst->llst_index);
3987                 if (exp == NULL) {
3988                         lfsck_layout_llst_del(llsd, llst);
3989                         lfsck_layout_llst_put(llst);
3990                         spin_lock(&llsd->llsd_lock);
3991                         continue;
3992                 }
3993
3994                 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
3995                 if (rc != 0) {
3996                         CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
3997                                "query %s for layout: rc = %d\n",
3998                                lfsck_lfsck2name(lfsck),
3999                                exp->exp_obd->obd_name, rc);
4000
4001                         rc1 = rc;
4002                         lfsck_layout_llst_put(llst);
4003                         class_export_put(exp);
4004                 }
4005                 spin_lock(&llsd->llsd_lock);
4006         }
4007         spin_unlock(&llsd->llsd_lock);
4008
4009         rc = ptlrpc_set_wait(set);
4010         ptlrpc_set_destroy(set);
4011
4012         GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
4013
4014 log:
4015         CDEBUG(D_LFSCK, "%s: layout LFSCK slave queries master: rc = %d\n",
4016                lfsck_lfsck2name(com->lc_lfsck), rc);
4017
4018         return rc;
4019 }
4020
4021 static void
4022 lfsck_layout_slave_notify_master(const struct lu_env *env,
4023                                  struct lfsck_component *com,
4024                                  enum lfsck_events event, int result)
4025 {
4026         struct lfsck_instance            *lfsck = com->lc_lfsck;
4027         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
4028         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
4029         struct lfsck_layout_slave_target *llst;
4030         struct obd_export                *exp;
4031         struct ptlrpc_request_set        *set;
4032         int                               rc;
4033         ENTRY;
4034
4035         CDEBUG(D_LFSCK, "%s: layout LFSCK slave notifies master\n",
4036                lfsck_lfsck2name(com->lc_lfsck));
4037
4038         set = ptlrpc_prep_set();
4039         if (set == NULL)
4040                 RETURN_EXIT;
4041
4042         memset(lr, 0, sizeof(*lr));
4043         lr->lr_event = event;
4044         lr->lr_flags = LEF_FROM_OST;
4045         lr->lr_status = result;
4046         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
4047         lr->lr_active = LFSCK_TYPE_LAYOUT;
4048         llsd->llsd_touch_gen++;
4049         spin_lock(&llsd->llsd_lock);
4050         while (!list_empty(&llsd->llsd_master_list)) {
4051                 llst = list_entry(llsd->llsd_master_list.next,
4052                                   struct lfsck_layout_slave_target,
4053                                   llst_list);
4054                 if (llst->llst_gen == llsd->llsd_touch_gen)
4055                         break;
4056
4057                 llst->llst_gen = llsd->llsd_touch_gen;
4058                 list_del(&llst->llst_list);
4059                 list_add_tail(&llst->llst_list,
4060                               &llsd->llsd_master_list);
4061                 atomic_inc(&llst->llst_ref);
4062                 spin_unlock(&llsd->llsd_lock);
4063
4064                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4065                                                llst->llst_index);
4066                 if (exp == NULL) {
4067                         lfsck_layout_llst_del(llsd, llst);
4068                         lfsck_layout_llst_put(llst);
4069                         spin_lock(&llsd->llsd_lock);
4070                         continue;
4071                 }
4072
4073                 rc = lfsck_layout_async_notify(env, exp, lr, set);
4074                 if (rc != 0)
4075                         CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4076                                "notify %s for layout: rc = %d\n",
4077                                lfsck_lfsck2name(lfsck),
4078                                exp->exp_obd->obd_name, rc);
4079
4080                 lfsck_layout_llst_put(llst);
4081                 class_export_put(exp);
4082                 spin_lock(&llsd->llsd_lock);
4083         }
4084         spin_unlock(&llsd->llsd_lock);
4085
4086         ptlrpc_set_wait(set);
4087         ptlrpc_set_destroy(set);
4088
4089         RETURN_EXIT;
4090 }
4091
4092 /*
4093  * \ret -ENODATA: unrecognized stripe
4094  * \ret = 0     : recognized stripe
4095  * \ret < 0     : other failures
4096  */
4097 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
4098                                            struct lfsck_component *com,
4099                                            struct lu_fid *cfid,
4100                                            struct lu_fid *pfid)
4101 {
4102         struct lfsck_thread_info        *info   = lfsck_env_info(env);
4103         struct lu_buf                   *buf    = &info->lti_big_buf;
4104         struct ost_id                   *oi     = &info->lti_oi;
4105         struct dt_object                *obj;
4106         struct lov_mds_md_v1            *lmm;
4107         struct lov_ost_data_v1          *objs;
4108         __u32                            idx    = pfid->f_stripe_idx;
4109         __u32                            magic;
4110         int                              rc     = 0;
4111         int                              i;
4112         __u16                            count;
4113         ENTRY;
4114
4115         pfid->f_ver = 0;
4116         obj = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
4117         if (IS_ERR(obj))
4118                 RETURN(PTR_ERR(obj));
4119
4120         dt_read_lock(env, obj, 0);
4121         if (unlikely(!dt_object_exists(obj)))
4122                 GOTO(unlock, rc = -ENOENT);
4123
4124         rc = lfsck_layout_get_lovea(env, obj, buf, NULL);
4125         if (rc < 0)
4126                 GOTO(unlock, rc);
4127
4128         if (rc == 0)
4129                 GOTO(unlock, rc = -ENODATA);
4130
4131         lmm = buf->lb_buf;
4132         rc = lfsck_layout_verify_header(lmm);
4133         if (rc != 0)
4134                 GOTO(unlock, rc);
4135
4136         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
4137          * been verified in lfsck_layout_verify_header() already. If some
4138          * new magic introduced in the future, then layout LFSCK needs to
4139          * be updated also. */
4140         magic = le32_to_cpu(lmm->lmm_magic);
4141         if (magic == LOV_MAGIC_V1) {
4142                 objs = &lmm->lmm_objects[0];
4143         } else {
4144                 LASSERT(magic == LOV_MAGIC_V3);
4145                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4146         }
4147
4148         fid_to_ostid(cfid, oi);
4149         count = le16_to_cpu(lmm->lmm_stripe_count);
4150         for (i = 0; i < count; i++, objs++) {
4151                 struct ost_id oi2;
4152
4153                 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
4154                 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
4155                         GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
4156         }
4157
4158         GOTO(unlock, rc = -ENODATA);
4159
4160 unlock:
4161         dt_read_unlock(env, obj);
4162         lu_object_put(env, &obj->do_lu);
4163
4164         return rc;
4165 }
4166
4167 /*
4168  * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
4169  * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
4170  * layout EA from MDT to OST. On one hand, the OST no need to understand
4171  * the layout EA structure; on the other hand, it may cause trouble when
4172  * transfer large layout EA from MDT to OST via normal OUT RPC.
4173  *
4174  * \ret > 0: unrecognized stripe
4175  * \ret = 0: recognized stripe
4176  * \ret < 0: other failures
4177  */
4178 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
4179                                           struct lfsck_component *com,
4180                                           struct lu_fid *cfid,
4181                                           struct lu_fid *pfid)
4182 {
4183         struct lfsck_instance    *lfsck  = com->lc_lfsck;
4184         struct obd_device        *obd    = lfsck->li_obd;
4185         struct seq_server_site   *ss     =
4186                         lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
4187         struct obd_export        *exp    = NULL;
4188         struct ptlrpc_request    *req    = NULL;
4189         struct lfsck_request     *lr;
4190         struct lu_seq_range       range  = { 0 };
4191         int                       rc     = 0;
4192         ENTRY;
4193
4194         if (unlikely(fid_is_idif(pfid)))
4195                 RETURN(1);
4196
4197         fld_range_set_any(&range);
4198         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), &range);
4199         if (rc != 0)
4200                 RETURN(rc == -ENOENT ? 1 : rc);
4201
4202         if (unlikely(!fld_range_is_mdt(&range)))
4203                 RETURN(1);
4204
4205         exp = lustre_find_lwp_by_index(obd->obd_name, range.lsr_index);
4206         if (unlikely(exp == NULL))
4207                 RETURN(1);
4208
4209         if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
4210                 GOTO(out, rc = -EOPNOTSUPP);
4211
4212         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4213         if (req == NULL)
4214                 GOTO(out, rc = -ENOMEM);
4215
4216         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4217         if (rc != 0) {
4218                 ptlrpc_request_free(req);
4219
4220                 GOTO(out, rc);
4221         }
4222
4223         lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4224         memset(lr, 0, sizeof(*lr));
4225         lr->lr_event = LE_PAIRS_VERIFY;
4226         lr->lr_active = LFSCK_TYPE_LAYOUT;
4227         lr->lr_fid = *cfid; /* OST-object itself FID. */
4228         lr->lr_fid2 = *pfid; /* The claimed parent FID. */
4229
4230         ptlrpc_request_set_replen(req);
4231         rc = ptlrpc_queue_wait(req);
4232         ptlrpc_req_finished(req);
4233
4234         if (rc == -ENOENT || rc == -ENODATA)
4235                 rc = 1;
4236
4237         GOTO(out, rc);
4238
4239 out:
4240         if (exp != NULL)
4241                 class_export_put(exp);
4242
4243         return rc;
4244 }
4245
4246 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
4247                                           struct lfsck_component *com,
4248                                           struct lfsck_request *lr)
4249 {
4250         struct lfsck_thread_info        *info   = lfsck_env_info(env);
4251         struct filter_fid               *ff     = &info->lti_new_pfid;
4252         struct lu_buf                   *buf;
4253         struct dt_device                *dev    = com->lc_lfsck->li_bottom;
4254         struct dt_object                *obj;
4255         struct thandle                  *th     = NULL;
4256         int                              rc     = 0;
4257         ENTRY;
4258
4259         obj = lfsck_object_find_by_dev(env, dev, &lr->lr_fid);
4260         if (IS_ERR(obj))
4261                 GOTO(log, rc = PTR_ERR(obj));
4262
4263         fid_cpu_to_le(&ff->ff_parent, &lr->lr_fid2);
4264         buf = lfsck_buf_get(env, ff, sizeof(*ff));
4265         dt_write_lock(env, obj, 0);
4266         if (unlikely(!dt_object_exists(obj)))
4267                 GOTO(unlock, rc = 0);
4268
4269         th = dt_trans_create(env, dev);
4270         if (IS_ERR(th))
4271                 GOTO(unlock, rc = PTR_ERR(th));
4272
4273         rc = dt_declare_xattr_set(env, obj, buf, XATTR_NAME_FID, 0, th);
4274         if (rc != 0)
4275                 GOTO(stop, rc);
4276
4277         rc = dt_trans_start_local(env, dev, th);
4278         if (rc != 0)
4279                 GOTO(stop, rc);
4280
4281         rc = dt_xattr_set(env, obj, buf, XATTR_NAME_FID, 0, th, BYPASS_CAPA);
4282
4283         GOTO(stop, rc);
4284
4285 stop:
4286         dt_trans_stop(env, dev, th);
4287
4288 unlock:
4289         dt_write_unlock(env, obj);
4290         lu_object_put(env, &obj->do_lu);
4291
4292 log:
4293         CDEBUG(D_LFSCK, "%s: layout LFSCK slave repaired pfid for "DFID
4294                ", parent "DFID": rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
4295                PFID(&lr->lr_fid), PFID(&lr->lr_fid2), rc);
4296
4297         return rc;
4298 }
4299
4300 /* layout APIs */
4301
4302 static int lfsck_layout_reset(const struct lu_env *env,
4303                               struct lfsck_component *com, bool init)
4304 {
4305         struct lfsck_layout     *lo    = com->lc_file_ram;
4306         int                      rc;
4307
4308         down_write(&com->lc_sem);
4309         if (init) {
4310                 memset(lo, 0, com->lc_file_size);
4311         } else {
4312                 __u32 count = lo->ll_success_count;
4313                 __u64 last_time = lo->ll_time_last_complete;
4314
4315                 memset(lo, 0, com->lc_file_size);
4316                 lo->ll_success_count = count;
4317                 lo->ll_time_last_complete = last_time;
4318         }
4319
4320         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
4321         lo->ll_status = LS_INIT;
4322
4323         rc = lfsck_layout_store(env, com);
4324         up_write(&com->lc_sem);
4325
4326         CDEBUG(D_LFSCK, "%s: layout LFSCK reset: rc = %d\n",
4327                lfsck_lfsck2name(com->lc_lfsck), rc);
4328
4329         return rc;
4330 }
4331
4332 static void lfsck_layout_fail(const struct lu_env *env,
4333                               struct lfsck_component *com, bool new_checked)
4334 {
4335         struct lfsck_layout *lo = com->lc_file_ram;
4336
4337         down_write(&com->lc_sem);
4338         if (new_checked)
4339                 com->lc_new_checked++;
4340         lfsck_layout_record_failure(env, com->lc_lfsck, lo);
4341         up_write(&com->lc_sem);
4342 }
4343
4344 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
4345                                           struct lfsck_component *com, bool init)
4346 {
4347         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4348         struct lfsck_layout             *lo      = com->lc_file_ram;
4349         struct lfsck_layout_master_data *llmd    = com->lc_data;
4350         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4351         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
4352         struct l_wait_info               lwi     = { 0 };
4353         int                              rc;
4354
4355         if (com->lc_new_checked == 0 && !init)
4356                 return 0;
4357
4358         l_wait_event(mthread->t_ctl_waitq,
4359                      list_empty(&llmd->llmd_req_list) ||
4360                      !thread_is_running(mthread) ||
4361                      thread_is_stopped(athread),
4362                      &lwi);
4363
4364         if (!thread_is_running(mthread) || thread_is_stopped(athread))
4365                 return 0;
4366
4367         down_write(&com->lc_sem);
4368         if (init) {
4369                 lo->ll_pos_latest_start = lfsck->li_pos_current.lp_oit_cookie;
4370         } else {
4371                 lo->ll_pos_last_checkpoint =
4372                                         lfsck->li_pos_current.lp_oit_cookie;
4373                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4374                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4375                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4376                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4377                 com->lc_new_checked = 0;
4378         }
4379
4380         rc = lfsck_layout_store(env, com);
4381         up_write(&com->lc_sem);
4382
4383         CDEBUG(D_LFSCK, "%s: layout LFSCK master checkpoint at the pos ["
4384                LPU64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
4385                lfsck->li_pos_current.lp_oit_cookie, rc);
4386
4387         return rc;
4388 }
4389
4390 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
4391                                          struct lfsck_component *com, bool init)
4392 {
4393         struct lfsck_instance   *lfsck = com->lc_lfsck;
4394         struct lfsck_layout     *lo    = com->lc_file_ram;
4395         int                      rc;
4396
4397         if (com->lc_new_checked == 0 && !init)
4398                 return 0;
4399
4400         down_write(&com->lc_sem);
4401         if (init) {
4402                 lo->ll_pos_latest_start = lfsck->li_pos_current.lp_oit_cookie;
4403         } else {
4404                 lo->ll_pos_last_checkpoint =
4405                                         lfsck->li_pos_current.lp_oit_cookie;
4406                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4407                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4408                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4409                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4410                 com->lc_new_checked = 0;
4411         }
4412
4413         rc = lfsck_layout_store(env, com);
4414         up_write(&com->lc_sem);
4415
4416         CDEBUG(D_LFSCK, "%s: layout LFSCK slave checkpoint at the pos ["
4417                LPU64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
4418                lfsck->li_pos_current.lp_oit_cookie, rc);
4419
4420         return rc;
4421 }
4422
4423 static int lfsck_layout_prep(const struct lu_env *env,
4424                              struct lfsck_component *com,
4425                              struct lfsck_start *start)
4426 {
4427         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4428         struct lfsck_layout     *lo     = com->lc_file_ram;
4429         struct lfsck_position   *pos    = &com->lc_pos_start;
4430
4431         fid_zero(&pos->lp_dir_parent);
4432         pos->lp_dir_cookie = 0;
4433         if (lo->ll_status == LS_COMPLETED ||
4434             lo->ll_status == LS_PARTIAL ||
4435             /* To handle orphan, must scan from the beginning. */
4436             (start != NULL && start->ls_flags & LPF_ORPHAN)) {
4437                 int rc;
4438
4439                 rc = lfsck_layout_reset(env, com, false);
4440                 if (rc == 0)
4441                         rc = lfsck_set_param(env, lfsck, start, true);
4442
4443                 if (rc != 0) {
4444                         CDEBUG(D_LFSCK, "%s: layout LFSCK prep failed: "
4445                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4446
4447                         return rc;
4448                 }
4449         }
4450
4451         down_write(&com->lc_sem);
4452         lo->ll_time_latest_start = cfs_time_current_sec();
4453         spin_lock(&lfsck->li_lock);
4454         if (lo->ll_flags & LF_SCANNED_ONCE) {
4455                 if (!lfsck->li_drop_dryrun ||
4456                     lo->ll_pos_first_inconsistent == 0) {
4457                         lo->ll_status = LS_SCANNING_PHASE2;
4458                         list_del_init(&com->lc_link);
4459                         list_add_tail(&com->lc_link,
4460                                       &lfsck->li_list_double_scan);
4461                         pos->lp_oit_cookie = 0;
4462                 } else {
4463                         int i;
4464
4465                         lo->ll_status = LS_SCANNING_PHASE1;
4466                         lo->ll_run_time_phase1 = 0;
4467                         lo->ll_run_time_phase2 = 0;
4468                         lo->ll_objs_checked_phase1 = 0;
4469                         lo->ll_objs_checked_phase2 = 0;
4470                         lo->ll_objs_failed_phase1 = 0;
4471                         lo->ll_objs_failed_phase2 = 0;
4472                         for (i = 0; i < LLIT_MAX; i++)
4473                                 lo->ll_objs_repaired[i] = 0;
4474
4475                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4476                         fid_zero(&com->lc_fid_latest_scanned_phase2);
4477                 }
4478         } else {
4479                 lo->ll_status = LS_SCANNING_PHASE1;
4480                 if (!lfsck->li_drop_dryrun ||
4481                     lo->ll_pos_first_inconsistent == 0)
4482                         pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
4483                 else
4484                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4485         }
4486         spin_unlock(&lfsck->li_lock);
4487         up_write(&com->lc_sem);
4488
4489         return 0;
4490 }
4491
4492 static int lfsck_layout_slave_prep(const struct lu_env *env,
4493                                    struct lfsck_component *com,
4494                                    struct lfsck_start_param *lsp)
4495 {
4496         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4497         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4498         struct lfsck_layout             *lo     = com->lc_file_ram;
4499         struct lfsck_start              *start  = lsp->lsp_start;
4500         int                              rc;
4501
4502         rc = lfsck_layout_prep(env, com, start);
4503         if (rc != 0)
4504                 return rc;
4505
4506         if (lo->ll_flags & LF_CRASHED_LASTID &&
4507             list_empty(&llsd->llsd_master_list)) {
4508                 LASSERT(lfsck->li_out_notify != NULL);
4509
4510                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4511                                      LE_LASTID_REBUILDING);
4512         }
4513
4514         if (!lsp->lsp_index_valid)
4515                 return 0;
4516
4517         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
4518         if (rc == 0 && start != NULL && start->ls_flags & LPF_ORPHAN) {
4519                 LASSERT(!llsd->llsd_rbtree_valid);
4520
4521                 write_lock(&llsd->llsd_rb_lock);
4522                 rc = lfsck_rbtree_setup(env, com);
4523                 write_unlock(&llsd->llsd_rb_lock);
4524         }
4525
4526         CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
4527                LPU64"]\n", lfsck_lfsck2name(lfsck),
4528                com->lc_pos_start.lp_oit_cookie);
4529
4530         return rc;
4531 }
4532
4533 static int lfsck_layout_master_prep(const struct lu_env *env,
4534                                     struct lfsck_component *com,
4535                                     struct lfsck_start_param *lsp)
4536 {
4537         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4538         struct lfsck_layout_master_data *llmd    = com->lc_data;
4539         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4540         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
4541         struct lfsck_thread_args        *lta;
4542         struct task_struct              *task;
4543         int                              rc;
4544         ENTRY;
4545
4546         rc = lfsck_layout_prep(env, com, lsp->lsp_start);
4547         if (rc != 0)
4548                 RETURN(rc);
4549
4550         llmd->llmd_assistant_status = 0;
4551         llmd->llmd_post_result = 0;
4552         llmd->llmd_to_post = 0;
4553         llmd->llmd_to_double_scan = 0;
4554         llmd->llmd_in_double_scan = 0;
4555         llmd->llmd_exit = 0;
4556         thread_set_flags(athread, 0);
4557
4558         lta = lfsck_thread_args_init(lfsck, com, lsp);
4559         if (IS_ERR(lta))
4560                 RETURN(PTR_ERR(lta));
4561
4562         task = kthread_run(lfsck_layout_assistant, lta, "lfsck_layout");
4563         if (IS_ERR(task)) {
4564                 rc = PTR_ERR(task);
4565                 CERROR("%s: cannot start LFSCK layout assistant thread: "
4566                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4567                 lfsck_thread_args_fini(lta);
4568         } else {
4569                 struct l_wait_info lwi = { 0 };
4570
4571                 l_wait_event(mthread->t_ctl_waitq,
4572                              thread_is_running(athread) ||
4573                              thread_is_stopped(athread),
4574                              &lwi);
4575                 if (unlikely(!thread_is_running(athread)))
4576                         rc = llmd->llmd_assistant_status;
4577                 else
4578                         rc = 0;
4579         }
4580
4581         CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
4582                LPU64"\n", lfsck_lfsck2name(lfsck),
4583                com->lc_pos_start.lp_oit_cookie);
4584
4585         RETURN(rc);
4586 }
4587
4588 /* Pre-fetch the attribute for each stripe in the given layout EA. */
4589 static int lfsck_layout_scan_stripes(const struct lu_env *env,
4590                                      struct lfsck_component *com,
4591                                      struct dt_object *parent,
4592                                      struct lov_mds_md_v1 *lmm)
4593 {
4594         struct lfsck_thread_info        *info    = lfsck_env_info(env);
4595         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4596         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
4597         struct lfsck_layout             *lo      = com->lc_file_ram;
4598         struct lfsck_layout_master_data *llmd    = com->lc_data;
4599         struct lfsck_layout_object      *llo     = NULL;
4600         struct lov_ost_data_v1          *objs;
4601         struct lfsck_tgt_descs          *ltds    = &lfsck->li_ost_descs;
4602         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4603         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
4604                 struct l_wait_info       lwi     = { 0 };
4605         struct lu_buf                   *buf;
4606         int                              rc      = 0;
4607         int                              i;
4608         __u32                            magic;
4609         __u16                            count;
4610         __u16                            gen;
4611         ENTRY;
4612
4613         buf = lfsck_buf_get(env, &info->lti_old_pfid,
4614                             sizeof(struct filter_fid_old));
4615         count = le16_to_cpu(lmm->lmm_stripe_count);
4616         gen = le16_to_cpu(lmm->lmm_layout_gen);
4617         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
4618          * been verified in lfsck_layout_verify_header() already. If some
4619          * new magic introduced in the future, then layout LFSCK needs to
4620          * be updated also. */
4621         magic = le32_to_cpu(lmm->lmm_magic);
4622         if (magic == LOV_MAGIC_V1) {
4623                 objs = &lmm->lmm_objects[0];
4624         } else {
4625                 LASSERT(magic == LOV_MAGIC_V3);
4626                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4627         }
4628
4629         for (i = 0; i < count; i++, objs++) {
4630                 struct lu_fid           *fid    = &info->lti_fid;
4631                 struct ost_id           *oi     = &info->lti_oi;
4632                 struct lfsck_layout_req *llr;
4633                 struct lfsck_tgt_desc   *tgt    = NULL;
4634                 struct dt_object        *cobj   = NULL;
4635                 __u32                    index  =
4636                                         le32_to_cpu(objs->l_ost_idx);
4637                 bool                     wakeup = false;
4638
4639                 if (unlikely(lovea_slot_is_dummy(objs)))
4640                         continue;
4641
4642                 l_wait_event(mthread->t_ctl_waitq,
4643                              bk->lb_async_windows == 0 ||
4644                              llmd->llmd_prefetched < bk->lb_async_windows ||
4645                              !thread_is_running(mthread) ||
4646                              thread_is_stopped(athread),
4647                              &lwi);
4648
4649                 if (unlikely(!thread_is_running(mthread)) ||
4650                              thread_is_stopped(athread))
4651                         GOTO(out, rc = 0);
4652
4653                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4654                 ostid_to_fid(fid, oi, index);
4655                 tgt = lfsck_tgt_get(ltds, index);
4656                 if (unlikely(tgt == NULL)) {
4657                         CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
4658                                "did not join the layout LFSCK\n",
4659                                lfsck_lfsck2name(lfsck), index);
4660                         lo->ll_flags |= LF_INCOMPLETE;
4661                         goto next;
4662                 }
4663
4664                 cobj = lfsck_object_find_by_dev(env, tgt->ltd_tgt, fid);
4665                 if (IS_ERR(cobj)) {
4666                         rc = PTR_ERR(cobj);
4667                         goto next;
4668                 }
4669
4670                 rc = dt_declare_attr_get(env, cobj, BYPASS_CAPA);
4671                 if (rc != 0)
4672                         goto next;
4673
4674                 rc = dt_declare_xattr_get(env, cobj, buf, XATTR_NAME_FID,
4675                                           BYPASS_CAPA);
4676                 if (rc != 0)
4677                         goto next;
4678
4679                 if (llo == NULL) {
4680                         llo = lfsck_layout_object_init(env, parent, gen);
4681                         if (IS_ERR(llo)) {
4682                                 rc = PTR_ERR(llo);
4683                                 goto next;
4684                         }
4685                 }
4686
4687                 llr = lfsck_layout_req_init(llo, cobj, index, i);
4688                 if (IS_ERR(llr)) {
4689                         rc = PTR_ERR(llr);
4690                         goto next;
4691                 }
4692
4693                 cobj = NULL;
4694                 spin_lock(&llmd->llmd_lock);
4695                 if (llmd->llmd_assistant_status < 0) {
4696                         spin_unlock(&llmd->llmd_lock);
4697                         lfsck_layout_req_fini(env, llr);
4698                         lfsck_tgt_put(tgt);
4699                         RETURN(llmd->llmd_assistant_status);
4700                 }
4701
4702                 list_add_tail(&llr->llr_list, &llmd->llmd_req_list);
4703                 if (llmd->llmd_prefetched == 0)
4704                         wakeup = true;
4705
4706                 llmd->llmd_prefetched++;
4707                 spin_unlock(&llmd->llmd_lock);
4708                 if (wakeup)
4709                         wake_up_all(&athread->t_ctl_waitq);
4710
4711 next:
4712                 down_write(&com->lc_sem);
4713                 com->lc_new_checked++;
4714                 if (rc < 0)
4715                         lfsck_layout_record_failure(env, lfsck, lo);
4716                 up_write(&com->lc_sem);
4717
4718                 if (cobj != NULL && !IS_ERR(cobj))
4719                         lu_object_put(env, &cobj->do_lu);
4720
4721                 if (likely(tgt != NULL))
4722                         lfsck_tgt_put(tgt);
4723
4724                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4725                         GOTO(out, rc);
4726         }
4727
4728         GOTO(out, rc = 0);
4729
4730 out:
4731         if (llo != NULL && !IS_ERR(llo))
4732                 lfsck_layout_object_put(env, llo);
4733
4734         return rc;
4735 }
4736
4737 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
4738  * the OST-object's attribute and generate an structure lfsck_layout_req on the
4739  * list ::llmd_req_list.
4740  *
4741  * For each request on above list, the lfsck_layout_assistant thread compares
4742  * the OST side attribute with local attribute, if inconsistent, then repair it.
4743  *
4744  * All above processing is async mode with pipeline. */
4745 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
4746                                         struct lfsck_component *com,
4747                                         struct dt_object *obj)
4748 {
4749         struct lfsck_thread_info        *info   = lfsck_env_info(env);
4750         struct ost_id                   *oi     = &info->lti_oi;
4751         struct lfsck_layout             *lo     = com->lc_file_ram;
4752         struct lfsck_layout_master_data *llmd   = com->lc_data;
4753         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4754         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
4755         struct thandle                  *handle = NULL;
4756         struct lu_buf                   *buf    = &info->lti_big_buf;
4757         struct lov_mds_md_v1            *lmm    = NULL;
4758         struct dt_device                *dev    = lfsck->li_bottom;
4759         struct lustre_handle             lh     = { 0 };
4760         ssize_t                          buflen = buf->lb_len;
4761         int                              rc     = 0;
4762         bool                             locked = false;
4763         bool                             stripe = false;
4764         bool                             bad_oi = false;
4765         ENTRY;
4766
4767         if (!S_ISREG(lfsck_object_type(obj)))
4768                 GOTO(out, rc = 0);
4769
4770         if (llmd->llmd_assistant_status < 0)
4771                 GOTO(out, rc = -ESRCH);
4772
4773         fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
4774         lmm_oi_cpu_to_le(oi, oi);
4775         dt_read_lock(env, obj, 0);
4776         locked = true;
4777
4778 again:
4779         rc = lfsck_layout_get_lovea(env, obj, buf, &buflen);
4780         if (rc <= 0)
4781                 GOTO(out, rc);
4782
4783         buf->lb_len = rc;
4784         lmm = buf->lb_buf;
4785         rc = lfsck_layout_verify_header(lmm);
4786         /* If the LOV EA crashed, then it is possible to be rebuilt later
4787          * when handle orphan OST-objects. */
4788         if (rc != 0)
4789                 GOTO(out, rc);
4790
4791         if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0)
4792                 GOTO(out, stripe = true);
4793
4794         /* Inconsistent lmm_oi, should be repaired. */
4795         bad_oi = true;
4796
4797         if (bk->lb_param & LPF_DRYRUN) {
4798                 down_write(&com->lc_sem);
4799                 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4800                 up_write(&com->lc_sem);
4801
4802                 GOTO(out, stripe = true);
4803         }
4804
4805         if (!lustre_handle_is_used(&lh)) {
4806                 dt_read_unlock(env, obj);
4807                 locked = false;
4808                 buf->lb_len = buflen;
4809                 rc = lfsck_layout_lock(env, com, obj, &lh,
4810                                        MDS_INODELOCK_LAYOUT |
4811                                        MDS_INODELOCK_XATTR);
4812                 if (rc != 0)
4813                         GOTO(out, rc);
4814
4815                 handle = dt_trans_create(env, dev);
4816                 if (IS_ERR(handle))
4817                         GOTO(out, rc = PTR_ERR(handle));
4818
4819                 rc = dt_declare_xattr_set(env, obj, buf, XATTR_NAME_LOV,
4820                                           LU_XATTR_REPLACE, handle);
4821                 if (rc != 0)
4822                         GOTO(out, rc);
4823
4824                 rc = dt_trans_start_local(env, dev, handle);
4825                 if (rc != 0)
4826                         GOTO(out, rc);
4827
4828                 dt_write_lock(env, obj, 0);
4829                 locked = true;
4830
4831                 goto again;
4832         }
4833
4834         lmm->lmm_oi = *oi;
4835         rc = dt_xattr_set(env, obj, buf, XATTR_NAME_LOV,
4836                           LU_XATTR_REPLACE, handle, BYPASS_CAPA);
4837         if (rc != 0)
4838                 GOTO(out, rc);
4839
4840         down_write(&com->lc_sem);
4841         lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4842         up_write(&com->lc_sem);
4843
4844         GOTO(out, stripe = true);
4845
4846 out:
4847         if (locked) {
4848                 if (lustre_handle_is_used(&lh))
4849                         dt_write_unlock(env, obj);
4850                 else
4851                         dt_read_unlock(env, obj);
4852         }
4853
4854         if (handle != NULL && !IS_ERR(handle))
4855                 dt_trans_stop(env, dev, handle);
4856
4857         lfsck_layout_unlock(&lh);
4858
4859         if (bad_oi)
4860                 CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
4861                        DFID": rc = %d\n", lfsck_lfsck2name(lfsck),
4862                        bk->lb_param & LPF_DRYRUN ? "found" : "repaired",
4863                        PFID(lfsck_dto2fid(obj)), rc);
4864
4865         if (stripe) {
4866                 rc = lfsck_layout_scan_stripes(env, com, obj, lmm);
4867         } else {
4868                 down_write(&com->lc_sem);
4869                 com->lc_new_checked++;
4870                 if (rc < 0)
4871                         lfsck_layout_record_failure(env, lfsck, lo);
4872                 up_write(&com->lc_sem);
4873         }
4874         buf->lb_len = buflen;
4875
4876         return rc;
4877 }
4878
4879 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
4880                                        struct lfsck_component *com,
4881                                        struct dt_object *obj)
4882 {
4883         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4884         struct lfsck_layout             *lo     = com->lc_file_ram;
4885         const struct lu_fid             *fid    = lfsck_dto2fid(obj);
4886         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4887         struct lfsck_layout_seq         *lls;
4888         __u64                            seq;
4889         __u64                            oid;
4890         int                              rc;
4891         ENTRY;
4892
4893         LASSERT(llsd != NULL);
4894
4895         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
4896             cfs_fail_val == lfsck_dev_idx(lfsck->li_bottom)) {
4897                 struct l_wait_info       lwi = LWI_TIMEOUT(cfs_time_seconds(1),
4898                                                            NULL, NULL);
4899                 struct ptlrpc_thread    *thread = &lfsck->li_thread;
4900
4901                 l_wait_event(thread->t_ctl_waitq,
4902                              !thread_is_running(thread),
4903                              &lwi);
4904         }
4905
4906         lfsck_rbtree_update_bitmap(env, com, fid, false);
4907
4908         down_write(&com->lc_sem);
4909         if (fid_is_idif(fid))
4910                 seq = 0;
4911         else if (!fid_is_norm(fid) ||
4912                  !fid_is_for_ostobj(env, lfsck->li_next, obj, fid))
4913                 GOTO(unlock, rc = 0);
4914         else
4915                 seq = fid_seq(fid);
4916         com->lc_new_checked++;
4917
4918         lls = lfsck_layout_seq_lookup(llsd, seq);
4919         if (lls == NULL) {
4920                 OBD_ALLOC_PTR(lls);
4921                 if (unlikely(lls == NULL))
4922                         GOTO(unlock, rc = -ENOMEM);
4923
4924                 INIT_LIST_HEAD(&lls->lls_list);
4925                 lls->lls_seq = seq;
4926                 rc = lfsck_layout_lastid_load(env, com, lls);
4927                 if (rc != 0) {
4928                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
4929                               "load LAST_ID for "LPX64": rc = %d\n",
4930                               lfsck_lfsck2name(com->lc_lfsck), seq, rc);
4931                         lo->ll_objs_failed_phase1++;
4932                         OBD_FREE_PTR(lls);
4933                         GOTO(unlock, rc);
4934                 }
4935
4936                 lfsck_layout_seq_insert(llsd, lls);
4937         }
4938
4939         if (unlikely(fid_is_last_id(fid)))
4940                 GOTO(unlock, rc = 0);
4941
4942         oid = fid_oid(fid);
4943         if (oid > lls->lls_lastid_known)
4944                 lls->lls_lastid_known = oid;
4945
4946         if (oid > lls->lls_lastid) {
4947                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
4948                         /* OFD may create new objects during LFSCK scanning. */
4949                         rc = lfsck_layout_lastid_reload(env, com, lls);
4950                         if (unlikely(rc != 0))
4951                                 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
4952                                       "reload LAST_ID for "LPX64": rc = %d\n",
4953                                       lfsck_lfsck2name(com->lc_lfsck),
4954                                       lls->lls_seq, rc);
4955                         if (oid <= lls->lls_lastid)
4956                                 GOTO(unlock, rc = 0);
4957
4958                         LASSERT(lfsck->li_out_notify != NULL);
4959
4960                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4961                                              LE_LASTID_REBUILDING);
4962                         lo->ll_flags |= LF_CRASHED_LASTID;
4963                 }
4964
4965                 lls->lls_lastid = oid;
4966                 lls->lls_dirty = 1;
4967         }
4968
4969         GOTO(unlock, rc = 0);
4970
4971 unlock:
4972         up_write(&com->lc_sem);
4973
4974         return rc;
4975 }
4976
4977 static int lfsck_layout_exec_dir(const struct lu_env *env,
4978                                  struct lfsck_component *com,
4979                                  struct dt_object *obj,
4980                                  struct lu_dirent *ent)
4981 {
4982         return 0;
4983 }
4984
4985 static int lfsck_layout_master_post(const struct lu_env *env,
4986                                     struct lfsck_component *com,
4987                                     int result, bool init)
4988 {
4989         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4990         struct lfsck_layout             *lo      = com->lc_file_ram;
4991         struct lfsck_layout_master_data *llmd    = com->lc_data;
4992         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4993         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
4994         struct l_wait_info               lwi     = { 0 };
4995         int                              rc;
4996         ENTRY;
4997
4998
4999         llmd->llmd_post_result = result;
5000         llmd->llmd_to_post = 1;
5001         if (llmd->llmd_post_result <= 0)
5002                 llmd->llmd_exit = 1;
5003
5004         wake_up_all(&athread->t_ctl_waitq);
5005         l_wait_event(mthread->t_ctl_waitq,
5006                      (result > 0 && list_empty(&llmd->llmd_req_list)) ||
5007                      thread_is_stopped(athread),
5008                      &lwi);
5009
5010         if (llmd->llmd_assistant_status < 0)
5011                 result = llmd->llmd_assistant_status;
5012
5013         down_write(&com->lc_sem);
5014         spin_lock(&lfsck->li_lock);
5015         /* When LFSCK failed, there may be some prefetched objects those are
5016          * not been processed yet, we do not know the exactly position, then
5017          * just restart from last check-point next time. */
5018         if (!init && !llmd->llmd_exit)
5019                 lo->ll_pos_last_checkpoint =
5020                                         lfsck->li_pos_current.lp_oit_cookie;
5021
5022         if (result > 0) {
5023                 lo->ll_status = LS_SCANNING_PHASE2;
5024                 lo->ll_flags |= LF_SCANNED_ONCE;
5025                 lo->ll_flags &= ~LF_UPGRADE;
5026                 list_del_init(&com->lc_link);
5027                 list_add_tail(&com->lc_link, &lfsck->li_list_double_scan);
5028         } else if (result == 0) {
5029                 lo->ll_status = lfsck->li_status;
5030                 if (lo->ll_status == 0)
5031                         lo->ll_status = LS_STOPPED;
5032                 if (lo->ll_status != LS_PAUSED) {
5033                         list_del_init(&com->lc_link);
5034                         list_add_tail(&com->lc_link, &lfsck->li_list_idle);
5035                 }
5036         } else {
5037                 lo->ll_status = LS_FAILED;
5038                 list_del_init(&com->lc_link);
5039                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
5040         }
5041         spin_unlock(&lfsck->li_lock);
5042
5043         if (!init) {
5044                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
5045                                 HALF_SEC - lfsck->li_time_last_checkpoint);
5046                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
5047                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5048                 com->lc_new_checked = 0;
5049         }
5050
5051         rc = lfsck_layout_store(env, com);
5052         up_write(&com->lc_sem);
5053
5054         CDEBUG(D_LFSCK, "%s: layout LFSCK master post done: rc = %d\n",
5055                lfsck_lfsck2name(lfsck), rc);
5056
5057         RETURN(rc);
5058 }
5059
5060 static int lfsck_layout_slave_post(const struct lu_env *env,
5061                                    struct lfsck_component *com,
5062                                    int result, bool init)
5063 {
5064         struct lfsck_instance   *lfsck = com->lc_lfsck;
5065         struct lfsck_layout     *lo    = com->lc_file_ram;
5066         int                      rc;
5067         bool                     done  = false;
5068
5069         rc = lfsck_layout_lastid_store(env, com);
5070         if (rc != 0)
5071                 result = rc;
5072
5073         LASSERT(lfsck->li_out_notify != NULL);
5074
5075         down_write(&com->lc_sem);
5076         spin_lock(&lfsck->li_lock);
5077         if (!init)
5078                 lo->ll_pos_last_checkpoint =
5079                                         lfsck->li_pos_current.lp_oit_cookie;
5080         if (result > 0) {
5081                 lo->ll_status = LS_SCANNING_PHASE2;
5082                 lo->ll_flags |= LF_SCANNED_ONCE;
5083                 if (lo->ll_flags & LF_CRASHED_LASTID) {
5084                         done = true;
5085                         lo->ll_flags &= ~LF_CRASHED_LASTID;
5086                 }
5087                 lo->ll_flags &= ~LF_UPGRADE;
5088                 list_del_init(&com->lc_link);
5089                 list_add_tail(&com->lc_link, &lfsck->li_list_double_scan);
5090         } else if (result == 0) {
5091                 lo->ll_status = lfsck->li_status;
5092                 if (lo->ll_status == 0)
5093                         lo->ll_status = LS_STOPPED;
5094                 if (lo->ll_status != LS_PAUSED) {
5095                         list_del_init(&com->lc_link);
5096                         list_add_tail(&com->lc_link, &lfsck->li_list_idle);
5097                 }
5098         } else {
5099                 lo->ll_status = LS_FAILED;
5100                 list_del_init(&com->lc_link);
5101                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
5102         }
5103         spin_unlock(&lfsck->li_lock);
5104
5105         if (done)
5106                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5107                                      LE_LASTID_REBUILT);
5108
5109         if (!init) {
5110                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
5111                                 HALF_SEC - lfsck->li_time_last_checkpoint);
5112                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
5113                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5114                 com->lc_new_checked = 0;
5115         }
5116
5117         rc = lfsck_layout_store(env, com);
5118         up_write(&com->lc_sem);
5119
5120         lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
5121
5122         if (result <= 0)
5123                 lfsck_rbtree_cleanup(env, com);
5124
5125         CDEBUG(D_LFSCK, "%s: layout LFSCK slave post done: rc = %d\n",
5126                lfsck_lfsck2name(lfsck), rc);
5127
5128         return rc;
5129 }
5130
5131 static int lfsck_layout_dump(const struct lu_env *env,
5132                              struct lfsck_component *com, struct seq_file *m)
5133 {
5134         struct lfsck_instance   *lfsck = com->lc_lfsck;
5135         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
5136         struct lfsck_layout     *lo    = com->lc_file_ram;
5137         int                      rc;
5138
5139         down_read(&com->lc_sem);
5140         seq_printf(m, "name: lfsck_layout\n"
5141                       "magic: %#x\n"
5142                       "version: %d\n"
5143                       "status: %s\n",
5144                       lo->ll_magic,
5145                       bk->lb_version,
5146                       lfsck_status2names(lo->ll_status));
5147
5148         rc = lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
5149         if (rc < 0)
5150                 goto out;
5151
5152         rc = lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
5153         if (rc < 0)
5154                 goto out;
5155
5156         rc = lfsck_time_dump(m, lo->ll_time_last_complete,
5157                              "time_since_last_completed");
5158         if (rc < 0)
5159                 goto out;
5160
5161         rc = lfsck_time_dump(m, lo->ll_time_latest_start,
5162                              "time_since_latest_start");
5163         if (rc < 0)
5164                 goto out;
5165
5166         rc = lfsck_time_dump(m, lo->ll_time_last_checkpoint,
5167                              "time_since_last_checkpoint");
5168         if (rc < 0)
5169                 goto out;
5170
5171         seq_printf(m, "latest_start_position: "LPU64"\n"
5172                       "last_checkpoint_position: "LPU64"\n"
5173                       "first_failure_position: "LPU64"\n",
5174                       lo->ll_pos_latest_start,
5175                       lo->ll_pos_last_checkpoint,
5176                       lo->ll_pos_first_inconsistent);
5177
5178         seq_printf(m, "success_count: %u\n"
5179                       "repaired_dangling: "LPU64"\n"
5180                       "repaired_unmatched_pair: "LPU64"\n"
5181                       "repaired_multiple_referenced: "LPU64"\n"
5182                       "repaired_orphan: "LPU64"\n"
5183                       "repaired_inconsistent_owner: "LPU64"\n"
5184                       "repaired_others: "LPU64"\n"
5185                       "skipped: "LPU64"\n"
5186                       "failed_phase1: "LPU64"\n"
5187                       "failed_phase2: "LPU64"\n",
5188                       lo->ll_success_count,
5189                       lo->ll_objs_repaired[LLIT_DANGLING - 1],
5190                       lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
5191                       lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
5192                       lo->ll_objs_repaired[LLIT_ORPHAN - 1],
5193                       lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
5194                       lo->ll_objs_repaired[LLIT_OTHERS - 1],
5195                       lo->ll_objs_skipped,
5196                       lo->ll_objs_failed_phase1,
5197                       lo->ll_objs_failed_phase2);
5198
5199         if (lo->ll_status == LS_SCANNING_PHASE1) {
5200                 __u64 pos;
5201                 const struct dt_it_ops *iops;
5202                 cfs_duration_t duration = cfs_time_current() -
5203                                           lfsck->li_time_last_checkpoint;
5204                 __u64 checked = lo->ll_objs_checked_phase1 +
5205                                 com->lc_new_checked;
5206                 __u64 speed = checked;
5207                 __u64 new_checked = com->lc_new_checked * HZ;
5208                 __u32 rtime = lo->ll_run_time_phase1 +
5209                               cfs_duration_sec(duration + HALF_SEC);
5210
5211                 if (duration != 0)
5212                         do_div(new_checked, duration);
5213                 if (rtime != 0)
5214                         do_div(speed, rtime);
5215                 seq_printf(m, "checked_phase1: "LPU64"\n"
5216                               "checked_phase2: "LPU64"\n"
5217                               "run_time_phase1: %u seconds\n"
5218                               "run_time_phase2: %u seconds\n"
5219                               "average_speed_phase1: "LPU64" items/sec\n"
5220                               "average_speed_phase2: N/A\n"
5221                               "real-time_speed_phase1: "LPU64" items/sec\n"
5222                               "real-time_speed_phase2: N/A\n",
5223                               checked,
5224                               lo->ll_objs_checked_phase2,
5225                               rtime,
5226                               lo->ll_run_time_phase2,
5227                               speed,
5228                               new_checked);
5229
5230                 LASSERT(lfsck->li_di_oit != NULL);
5231
5232                 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
5233
5234                 /* The low layer otable-based iteration position may NOT
5235                  * exactly match the layout-based directory traversal
5236                  * cookie. Generally, it is not a serious issue. But the
5237                  * caller should NOT make assumption on that. */
5238                 pos = iops->store(env, lfsck->li_di_oit);
5239                 if (!lfsck->li_current_oit_processed)
5240                         pos--;
5241                 seq_printf(m, "current_position: "LPU64"\n", pos);
5242
5243         } else if (lo->ll_status == LS_SCANNING_PHASE2) {
5244                 cfs_duration_t duration = cfs_time_current() -
5245                                           lfsck->li_time_last_checkpoint;
5246                 __u64 checked = lo->ll_objs_checked_phase2 +
5247                                 com->lc_new_checked;
5248                 __u64 speed1 = lo->ll_objs_checked_phase1;
5249                 __u64 speed2 = checked;
5250                 __u64 new_checked = com->lc_new_checked * HZ;
5251                 __u32 rtime = lo->ll_run_time_phase2 +
5252                               cfs_duration_sec(duration + HALF_SEC);
5253
5254                 if (duration != 0)
5255                         do_div(new_checked, duration);
5256                 if (lo->ll_run_time_phase1 != 0)
5257                         do_div(speed1, lo->ll_run_time_phase1);
5258                 if (rtime != 0)
5259                         do_div(speed2, rtime);
5260                 rc = seq_printf(m, "checked_phase1: "LPU64"\n"
5261                                 "checked_phase2: "LPU64"\n"
5262                                 "run_time_phase1: %u seconds\n"
5263                                 "run_time_phase2: %u seconds\n"
5264                                 "average_speed_phase1: "LPU64" items/sec\n"
5265                                 "average_speed_phase2: "LPU64" items/sec\n"
5266                                 "real-time_speed_phase1: N/A\n"
5267                                 "real-time_speed_phase2: "LPU64" items/sec\n"
5268                                 "current_position: "DFID"\n",
5269                                 lo->ll_objs_checked_phase1,
5270                                 checked,
5271                                 lo->ll_run_time_phase1,
5272                                 rtime,
5273                                 speed1,
5274                                 speed2,
5275                                 new_checked,
5276                                 PFID(&com->lc_fid_latest_scanned_phase2));
5277                 if (rc <= 0)
5278                         goto out;
5279
5280         } else {
5281                 __u64 speed1 = lo->ll_objs_checked_phase1;
5282                 __u64 speed2 = lo->ll_objs_checked_phase2;
5283
5284                 if (lo->ll_run_time_phase1 != 0)
5285                         do_div(speed1, lo->ll_run_time_phase1);
5286                 if (lo->ll_run_time_phase2 != 0)
5287                         do_div(speed2, lo->ll_run_time_phase2);
5288                 seq_printf(m, "checked_phase1: "LPU64"\n"
5289                            "checked_phase2: "LPU64"\n"
5290                            "run_time_phase1: %u seconds\n"
5291                            "run_time_phase2: %u seconds\n"
5292                            "average_speed_phase1: "LPU64" items/sec\n"
5293                            "average_speed_phase2: "LPU64" objs/sec\n"
5294                            "real-time_speed_phase1: N/A\n"
5295                            "real-time_speed_phase2: N/A\n"
5296                            "current_position: N/A\n",
5297                            lo->ll_objs_checked_phase1,
5298                            lo->ll_objs_checked_phase2,
5299                            lo->ll_run_time_phase1,
5300                            lo->ll_run_time_phase2,
5301                            speed1,
5302                            speed2);
5303         }
5304 out:
5305         up_read(&com->lc_sem);
5306
5307         return rc;
5308 }
5309
5310 static int lfsck_layout_master_double_scan(const struct lu_env *env,
5311                                            struct lfsck_component *com)
5312 {
5313         struct lfsck_layout_master_data *llmd    = com->lc_data;
5314         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
5315         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
5316         struct lfsck_layout             *lo      = com->lc_file_ram;
5317         struct l_wait_info               lwi     = { 0 };
5318
5319         if (unlikely(lo->ll_status != LS_SCANNING_PHASE2))
5320                 return 0;
5321
5322         llmd->llmd_to_double_scan = 1;
5323         wake_up_all(&athread->t_ctl_waitq);
5324         l_wait_event(mthread->t_ctl_waitq,
5325                      llmd->llmd_in_double_scan ||
5326                      thread_is_stopped(athread),
5327                      &lwi);
5328         if (llmd->llmd_assistant_status < 0)
5329                 return llmd->llmd_assistant_status;
5330
5331         return 0;
5332 }
5333
5334 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
5335                                           struct lfsck_component *com)
5336 {
5337         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5338         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
5339         struct lfsck_layout             *lo     = com->lc_file_ram;
5340         struct ptlrpc_thread            *thread = &lfsck->li_thread;
5341         int                              rc;
5342         ENTRY;
5343
5344         if (unlikely(lo->ll_status != LS_SCANNING_PHASE2)) {
5345                 lfsck_rbtree_cleanup(env, com);
5346                 lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE, 0);
5347                 RETURN(0);
5348         }
5349
5350         CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan start\n",
5351                lfsck_lfsck2name(lfsck));
5352
5353         atomic_inc(&lfsck->li_double_scan_count);
5354
5355         com->lc_new_checked = 0;
5356         com->lc_new_scanned = 0;
5357         com->lc_time_last_checkpoint = cfs_time_current();
5358         com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
5359                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
5360
5361         while (1) {
5362                 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(30),
5363                                                      NULL, NULL);
5364
5365                 rc = lfsck_layout_slave_query_master(env, com);
5366                 if (list_empty(&llsd->llsd_master_list)) {
5367                         if (unlikely(!thread_is_running(thread)))
5368                                 rc = 0;
5369                         else
5370                                 rc = 1;
5371
5372                         GOTO(done, rc);
5373                 }
5374
5375                 if (rc < 0)
5376                         GOTO(done, rc);
5377
5378                 rc = l_wait_event(thread->t_ctl_waitq,
5379                                   !thread_is_running(thread) ||
5380                                   list_empty(&llsd->llsd_master_list),
5381                                   &lwi);
5382                 if (unlikely(!thread_is_running(thread)))
5383                         GOTO(done, rc = 0);
5384
5385                 if (rc == -ETIMEDOUT)
5386                         continue;
5387
5388                 GOTO(done, rc = (rc < 0 ? rc : 1));
5389         }
5390
5391 done:
5392         rc = lfsck_layout_double_scan_result(env, com, rc);
5393
5394         lfsck_rbtree_cleanup(env, com);
5395         lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE, rc);
5396         if (atomic_dec_and_test(&lfsck->li_double_scan_count))
5397                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5398
5399         CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan finished, "
5400                "status %d: rc = %d\n",
5401                lfsck_lfsck2name(lfsck), lo->ll_status, rc);
5402
5403         return rc;
5404 }
5405
5406 static void lfsck_layout_master_data_release(const struct lu_env *env,
5407                                              struct lfsck_component *com)
5408 {
5409         struct lfsck_layout_master_data *llmd   = com->lc_data;
5410         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5411         struct lfsck_tgt_descs          *ltds;
5412         struct lfsck_tgt_desc           *ltd;
5413         struct lfsck_tgt_desc           *next;
5414
5415         LASSERT(llmd != NULL);
5416         LASSERT(thread_is_init(&llmd->llmd_thread) ||
5417                 thread_is_stopped(&llmd->llmd_thread));
5418         LASSERT(list_empty(&llmd->llmd_req_list));
5419
5420         com->lc_data = NULL;
5421
5422         ltds = &lfsck->li_ost_descs;
5423         spin_lock(&ltds->ltd_lock);
5424         list_for_each_entry_safe(ltd, next, &llmd->llmd_ost_phase1_list,
5425                                  ltd_layout_phase_list) {
5426                 list_del_init(&ltd->ltd_layout_phase_list);
5427         }
5428         list_for_each_entry_safe(ltd, next, &llmd->llmd_ost_phase2_list,
5429                                  ltd_layout_phase_list) {
5430                 list_del_init(&ltd->ltd_layout_phase_list);
5431         }
5432         list_for_each_entry_safe(ltd, next, &llmd->llmd_ost_list,
5433                                  ltd_layout_list) {
5434                 list_del_init(&ltd->ltd_layout_list);
5435         }
5436         spin_unlock(&ltds->ltd_lock);
5437
5438         ltds = &lfsck->li_mdt_descs;
5439         spin_lock(&ltds->ltd_lock);
5440         list_for_each_entry_safe(ltd, next, &llmd->llmd_mdt_phase1_list,
5441                                  ltd_layout_phase_list) {
5442                 list_del_init(&ltd->ltd_layout_phase_list);
5443         }
5444         list_for_each_entry_safe(ltd, next, &llmd->llmd_mdt_phase2_list,
5445                                  ltd_layout_phase_list) {
5446                 list_del_init(&ltd->ltd_layout_phase_list);
5447         }
5448         list_for_each_entry_safe(ltd, next, &llmd->llmd_mdt_list,
5449                                  ltd_layout_list) {
5450                 list_del_init(&ltd->ltd_layout_list);
5451         }
5452         spin_unlock(&ltds->ltd_lock);
5453
5454         OBD_FREE_PTR(llmd);
5455 }
5456
5457 static void lfsck_layout_slave_data_release(const struct lu_env *env,
5458                                             struct lfsck_component *com)
5459 {
5460         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5461         struct lfsck_layout_seq          *lls;
5462         struct lfsck_layout_seq          *next;
5463         struct lfsck_layout_slave_target *llst;
5464         struct lfsck_layout_slave_target *tmp;
5465
5466         LASSERT(llsd != NULL);
5467
5468         list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
5469                                      lls_list) {
5470                 list_del_init(&lls->lls_list);
5471                 lfsck_object_put(env, lls->lls_lastid_obj);
5472                 OBD_FREE_PTR(lls);
5473         }
5474
5475         list_for_each_entry_safe(llst, tmp, &llsd->llsd_master_list,
5476                                  llst_list) {
5477                 list_del_init(&llst->llst_list);
5478                 OBD_FREE_PTR(llst);
5479         }
5480
5481         lfsck_rbtree_cleanup(env, com);
5482         com->lc_data = NULL;
5483         OBD_FREE_PTR(llsd);
5484 }
5485
5486 static void lfsck_layout_master_quit(const struct lu_env *env,
5487                                      struct lfsck_component *com)
5488 {
5489         struct lfsck_layout_master_data *llmd    = com->lc_data;
5490         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
5491         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
5492         struct l_wait_info               lwi     = { 0 };
5493
5494         llmd->llmd_exit = 1;
5495         wake_up_all(&athread->t_ctl_waitq);
5496         l_wait_event(mthread->t_ctl_waitq,
5497                      thread_is_init(athread) ||
5498                      thread_is_stopped(athread),
5499                      &lwi);
5500 }
5501
5502 static void lfsck_layout_slave_quit(const struct lu_env *env,
5503                                     struct lfsck_component *com)
5504 {
5505         lfsck_rbtree_cleanup(env, com);
5506 }
5507
5508 static int lfsck_layout_master_in_notify(const struct lu_env *env,
5509                                          struct lfsck_component *com,
5510                                          struct lfsck_request *lr)
5511 {
5512         struct lfsck_instance           *lfsck = com->lc_lfsck;
5513         struct lfsck_layout             *lo    = com->lc_file_ram;
5514         struct lfsck_layout_master_data *llmd  = com->lc_data;
5515         struct lfsck_tgt_descs          *ltds;
5516         struct lfsck_tgt_desc           *ltd;
5517         bool                             fail  = false;
5518         ENTRY;
5519
5520         if (lr->lr_event == LE_PAIRS_VERIFY) {
5521                 int rc;
5522
5523                 rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
5524                                                      &lr->lr_fid2);
5525
5526                 RETURN(rc);
5527         }
5528
5529         CDEBUG(D_LFSCK, "%s: layout LFSCK master handle notify %u "
5530                "from %s %x, status %d\n", lfsck_lfsck2name(lfsck),
5531                lr->lr_event, (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
5532                lr->lr_index, lr->lr_status);
5533
5534         if (lr->lr_event != LE_PHASE1_DONE &&
5535             lr->lr_event != LE_PHASE2_DONE &&
5536             lr->lr_event != LE_PEER_EXIT)
5537                 RETURN(-EINVAL);
5538
5539         if (lr->lr_flags & LEF_FROM_OST)
5540                 ltds = &lfsck->li_ost_descs;
5541         else
5542                 ltds = &lfsck->li_mdt_descs;
5543         spin_lock(&ltds->ltd_lock);
5544         ltd = LTD_TGT(ltds, lr->lr_index);
5545         if (ltd == NULL) {
5546                 spin_unlock(&ltds->ltd_lock);
5547
5548                 RETURN(-ENXIO);
5549         }
5550
5551         list_del_init(&ltd->ltd_layout_phase_list);
5552         switch (lr->lr_event) {
5553         case LE_PHASE1_DONE:
5554                 if (lr->lr_status <= 0) {
5555                         ltd->ltd_layout_done = 1;
5556                         list_del_init(&ltd->ltd_layout_list);
5557                         lo->ll_flags |= LF_INCOMPLETE;
5558                         fail = true;
5559                         break;
5560                 }
5561
5562                 if (lr->lr_flags & LEF_FROM_OST) {
5563                         if (list_empty(&ltd->ltd_layout_list))
5564                                 list_add_tail(&ltd->ltd_layout_list,
5565                                               &llmd->llmd_ost_list);
5566                         list_add_tail(&ltd->ltd_layout_phase_list,
5567                                       &llmd->llmd_ost_phase2_list);
5568                 } else {
5569                         if (list_empty(&ltd->ltd_layout_list))
5570                                 list_add_tail(&ltd->ltd_layout_list,
5571                                               &llmd->llmd_mdt_list);
5572                         list_add_tail(&ltd->ltd_layout_phase_list,
5573                                       &llmd->llmd_mdt_phase2_list);
5574                 }
5575                 break;
5576         case LE_PHASE2_DONE:
5577                 ltd->ltd_layout_done = 1;
5578                 list_del_init(&ltd->ltd_layout_list);
5579                 break;
5580         case LE_PEER_EXIT:
5581                 fail = true;
5582                 ltd->ltd_layout_done = 1;
5583                 list_del_init(&ltd->ltd_layout_list);
5584                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT))
5585                         lo->ll_flags |= LF_INCOMPLETE;
5586                 break;
5587         default:
5588                 break;
5589         }
5590         spin_unlock(&ltds->ltd_lock);
5591
5592         if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
5593                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5594
5595                 memset(stop, 0, sizeof(*stop));
5596                 stop->ls_status = lr->lr_status;
5597                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5598                 lfsck_stop(env, lfsck->li_bottom, stop);
5599         } else if (lfsck_layout_master_to_orphan(llmd)) {
5600                 wake_up_all(&llmd->llmd_thread.t_ctl_waitq);
5601         }
5602
5603         RETURN(0);
5604 }
5605
5606 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
5607                                         struct lfsck_component *com,
5608                                         struct lfsck_request *lr)
5609 {
5610         struct lfsck_instance            *lfsck = com->lc_lfsck;
5611         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5612         struct lfsck_layout_slave_target *llst;
5613         int                               rc;
5614         ENTRY;
5615
5616         switch (lr->lr_event) {
5617         case LE_FID_ACCESSED:
5618                 lfsck_rbtree_update_bitmap(env, com, &lr->lr_fid, true);
5619                 RETURN(0);
5620         case LE_CONDITIONAL_DESTROY:
5621                 rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
5622                 RETURN(rc);
5623         case LE_PAIRS_VERIFY: {
5624                 lr->lr_status = LPVS_INIT;
5625                 /* Firstly, if the MDT-object which is claimed via OST-object
5626                  * local stored PFID xattr recognizes the OST-object, then it
5627                  * must be that the client given PFID is wrong. */
5628                 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5629                                                     &lr->lr_fid3);
5630                 if (rc <= 0)
5631                         RETURN(0);
5632
5633                 lr->lr_status = LPVS_INCONSISTENT;
5634                 /* The OST-object local stored PFID xattr is stale. We need to
5635                  * check whether the MDT-object that is claimed via the client
5636                  * given PFID information recognizes the OST-object or not. If
5637                  * matches, then need to update the OST-object's PFID xattr. */
5638                 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5639                                                     &lr->lr_fid2);
5640                 /* For rc < 0 case:
5641                  * We are not sure whether the client given PFID information
5642                  * is correct or not, do nothing to avoid improper fixing.
5643                  *
5644                  * For rc > 0 case:
5645                  * The client given PFID information is also invalid, we can
5646                  * NOT fix the OST-object inconsistency.
5647                  */
5648                 if (rc != 0)
5649                         RETURN(rc);
5650
5651                 lr->lr_status = LPVS_INCONSISTENT_TOFIX;
5652                 rc = lfsck_layout_slave_repair_pfid(env, com, lr);
5653
5654                 RETURN(rc);
5655         }
5656         case LE_PHASE2_DONE:
5657         case LE_PEER_EXIT:
5658                 CDEBUG(D_LFSCK, "%s: layout LFSCK slave handle notify %u "
5659                        "from MDT %x, status %d\n", lfsck_lfsck2name(lfsck),
5660                        lr->lr_event, lr->lr_index, lr->lr_status);
5661                 break;
5662         default:
5663                 RETURN(-EINVAL);
5664         }
5665
5666         llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
5667         if (llst == NULL)
5668                 RETURN(-ENXIO);
5669
5670         lfsck_layout_llst_put(llst);
5671         if (list_empty(&llsd->llsd_master_list))
5672                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5673
5674         if (lr->lr_event == LE_PEER_EXIT &&
5675             lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
5676                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5677
5678                 memset(stop, 0, sizeof(*stop));
5679                 stop->ls_status = lr->lr_status;
5680                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5681                 lfsck_stop(env, lfsck->li_bottom, stop);
5682         }
5683
5684         RETURN(0);
5685 }
5686
5687 static int lfsck_layout_query(const struct lu_env *env,
5688                               struct lfsck_component *com)
5689 {
5690         struct lfsck_layout *lo = com->lc_file_ram;
5691
5692         return lo->ll_status;
5693 }
5694
5695 static int lfsck_layout_master_stop_notify(const struct lu_env *env,
5696                                            struct lfsck_component *com,
5697                                            struct lfsck_tgt_descs *ltds,
5698                                            struct lfsck_tgt_desc *ltd,
5699                                            struct ptlrpc_request_set *set)
5700 {
5701         struct lfsck_thread_info          *info  = lfsck_env_info(env);
5702         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
5703         struct lfsck_request              *lr    = &info->lti_lr;
5704         struct lfsck_instance             *lfsck = com->lc_lfsck;
5705         int                                rc;
5706
5707         spin_lock(&ltds->ltd_lock);
5708         if (list_empty(&ltd->ltd_layout_list)) {
5709                 LASSERT(list_empty(&ltd->ltd_layout_phase_list));
5710                 spin_unlock(&ltds->ltd_lock);
5711
5712                 return 0;
5713         }
5714
5715         list_del_init(&ltd->ltd_layout_phase_list);
5716         list_del_init(&ltd->ltd_layout_list);
5717         spin_unlock(&ltds->ltd_lock);
5718
5719         memset(lr, 0, sizeof(*lr));
5720         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
5721         lr->lr_event = LE_PEER_EXIT;
5722         lr->lr_active = LFSCK_TYPE_LAYOUT;
5723         lr->lr_status = LS_CO_PAUSED;
5724         if (ltds == &lfsck->li_ost_descs)
5725                 lr->lr_flags = LEF_TO_OST;
5726
5727         laia->laia_com = com;
5728         laia->laia_ltds = ltds;
5729         atomic_inc(&ltd->ltd_ref);
5730         laia->laia_ltd = ltd;
5731         laia->laia_lr = lr;
5732         laia->laia_shared = 0;
5733
5734         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
5735                                  lfsck_layout_master_async_interpret,
5736                                  laia, LFSCK_NOTIFY);
5737         if (rc != 0) {
5738                 CDEBUG(D_LFSCK, "%s: layout LFSCK fail to notify %s %x "
5739                        "for co-stop: rc = %d\n",
5740                        lfsck_lfsck2name(lfsck),
5741                        (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
5742                        ltd->ltd_index, rc);
5743                 lfsck_tgt_put(ltd);
5744         }
5745
5746         return rc;
5747 }
5748
5749 /* with lfsck::li_lock held */
5750 static int lfsck_layout_slave_join(const struct lu_env *env,
5751                                    struct lfsck_component *com,
5752                                    struct lfsck_start_param *lsp)
5753 {
5754         struct lfsck_instance            *lfsck = com->lc_lfsck;
5755         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5756         struct lfsck_layout_slave_target *llst;
5757         struct lfsck_start               *start = lsp->lsp_start;
5758         int                               rc    = 0;
5759         ENTRY;
5760
5761         if (start == NULL || !(start->ls_flags & LPF_ORPHAN))
5762                 RETURN(0);
5763
5764         if (!lsp->lsp_index_valid)
5765                 RETURN(-EINVAL);
5766
5767         /* If someone is running the LFSCK without orphan handling,
5768          * it will not maintain the object accessing rbtree. So we
5769          * cannot join it for orphan handling. */
5770         if (!llsd->llsd_rbtree_valid)
5771                 RETURN(-EBUSY);
5772
5773         spin_unlock(&lfsck->li_lock);
5774         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5775         spin_lock(&lfsck->li_lock);
5776         if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
5777                 spin_unlock(&lfsck->li_lock);
5778                 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
5779                                                       true);
5780                 if (llst != NULL)
5781                         lfsck_layout_llst_put(llst);
5782                 spin_lock(&lfsck->li_lock);
5783                 rc = -EAGAIN;
5784         }
5785
5786         RETURN(rc);
5787 }
5788
5789 static struct lfsck_operations lfsck_layout_master_ops = {
5790         .lfsck_reset            = lfsck_layout_reset,
5791         .lfsck_fail             = lfsck_layout_fail,
5792         .lfsck_checkpoint       = lfsck_layout_master_checkpoint,
5793         .lfsck_prep             = lfsck_layout_master_prep,
5794         .lfsck_exec_oit         = lfsck_layout_master_exec_oit,
5795         .lfsck_exec_dir         = lfsck_layout_exec_dir,
5796         .lfsck_post             = lfsck_layout_master_post,
5797         .lfsck_interpret        = lfsck_layout_master_async_interpret,
5798         .lfsck_dump             = lfsck_layout_dump,
5799         .lfsck_double_scan      = lfsck_layout_master_double_scan,
5800         .lfsck_data_release     = lfsck_layout_master_data_release,
5801         .lfsck_quit             = lfsck_layout_master_quit,
5802         .lfsck_in_notify        = lfsck_layout_master_in_notify,
5803         .lfsck_query            = lfsck_layout_query,
5804         .lfsck_stop_notify      = lfsck_layout_master_stop_notify,
5805 };
5806
5807 static struct lfsck_operations lfsck_layout_slave_ops = {
5808         .lfsck_reset            = lfsck_layout_reset,
5809         .lfsck_fail             = lfsck_layout_fail,
5810         .lfsck_checkpoint       = lfsck_layout_slave_checkpoint,
5811         .lfsck_prep             = lfsck_layout_slave_prep,
5812         .lfsck_exec_oit         = lfsck_layout_slave_exec_oit,
5813         .lfsck_exec_dir         = lfsck_layout_exec_dir,
5814         .lfsck_post             = lfsck_layout_slave_post,
5815         .lfsck_dump             = lfsck_layout_dump,
5816         .lfsck_double_scan      = lfsck_layout_slave_double_scan,
5817         .lfsck_data_release     = lfsck_layout_slave_data_release,
5818         .lfsck_quit             = lfsck_layout_slave_quit,
5819         .lfsck_in_notify        = lfsck_layout_slave_in_notify,
5820         .lfsck_query            = lfsck_layout_query,
5821         .lfsck_join             = lfsck_layout_slave_join,
5822 };
5823
5824 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
5825 {
5826         struct lfsck_component  *com;
5827         struct lfsck_layout     *lo;
5828         struct dt_object        *root = NULL;
5829         struct dt_object        *obj;
5830         int                      rc;
5831         ENTRY;
5832
5833         OBD_ALLOC_PTR(com);
5834         if (com == NULL)
5835                 RETURN(-ENOMEM);
5836
5837         INIT_LIST_HEAD(&com->lc_link);
5838         INIT_LIST_HEAD(&com->lc_link_dir);
5839         init_rwsem(&com->lc_sem);
5840         atomic_set(&com->lc_ref, 1);
5841         com->lc_lfsck = lfsck;
5842         com->lc_type = LFSCK_TYPE_LAYOUT;
5843         if (lfsck->li_master) {
5844                 struct lfsck_layout_master_data *llmd;
5845
5846                 com->lc_ops = &lfsck_layout_master_ops;
5847                 OBD_ALLOC_PTR(llmd);
5848                 if (llmd == NULL)
5849                         GOTO(out, rc = -ENOMEM);
5850
5851                 INIT_LIST_HEAD(&llmd->llmd_req_list);
5852                 spin_lock_init(&llmd->llmd_lock);
5853                 INIT_LIST_HEAD(&llmd->llmd_ost_list);
5854                 INIT_LIST_HEAD(&llmd->llmd_ost_phase1_list);
5855                 INIT_LIST_HEAD(&llmd->llmd_ost_phase2_list);
5856                 INIT_LIST_HEAD(&llmd->llmd_mdt_list);
5857                 INIT_LIST_HEAD(&llmd->llmd_mdt_phase1_list);
5858                 INIT_LIST_HEAD(&llmd->llmd_mdt_phase2_list);
5859                 init_waitqueue_head(&llmd->llmd_thread.t_ctl_waitq);
5860                 com->lc_data = llmd;
5861         } else {
5862                 struct lfsck_layout_slave_data *llsd;
5863
5864                 com->lc_ops = &lfsck_layout_slave_ops;
5865                 OBD_ALLOC_PTR(llsd);
5866                 if (llsd == NULL)
5867                         GOTO(out, rc = -ENOMEM);
5868
5869                 INIT_LIST_HEAD(&llsd->llsd_seq_list);
5870                 INIT_LIST_HEAD(&llsd->llsd_master_list);
5871                 spin_lock_init(&llsd->llsd_lock);
5872                 llsd->llsd_rb_root = RB_ROOT;
5873                 rwlock_init(&llsd->llsd_rb_lock);
5874                 com->lc_data = llsd;
5875         }
5876         com->lc_file_size = sizeof(*lo);
5877         OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
5878         if (com->lc_file_ram == NULL)
5879                 GOTO(out, rc = -ENOMEM);
5880
5881         OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
5882         if (com->lc_file_disk == NULL)
5883                 GOTO(out, rc = -ENOMEM);
5884
5885         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
5886         if (IS_ERR(root))
5887                 GOTO(out, rc = PTR_ERR(root));
5888
5889         if (unlikely(!dt_try_as_dir(env, root)))
5890                 GOTO(out, rc = -ENOTDIR);
5891
5892         obj = local_file_find_or_create(env, lfsck->li_los, root,
5893                                         lfsck_layout_name,
5894                                         S_IFREG | S_IRUGO | S_IWUSR);
5895         if (IS_ERR(obj))
5896                 GOTO(out, rc = PTR_ERR(obj));
5897
5898         com->lc_obj = obj;
5899         rc = lfsck_layout_load(env, com);
5900         if (rc > 0)
5901                 rc = lfsck_layout_reset(env, com, true);
5902         else if (rc == -ENOENT)
5903                 rc = lfsck_layout_init(env, com);
5904
5905         if (rc != 0)
5906                 GOTO(out, rc);
5907
5908         lo = com->lc_file_ram;
5909         switch (lo->ll_status) {
5910         case LS_INIT:
5911         case LS_COMPLETED:
5912         case LS_FAILED:
5913         case LS_STOPPED:
5914         case LS_PARTIAL:
5915                 spin_lock(&lfsck->li_lock);
5916                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
5917                 spin_unlock(&lfsck->li_lock);
5918                 break;
5919         default:
5920                 CERROR("%s: unknown lfsck_layout status %d\n",
5921                        lfsck_lfsck2name(lfsck), lo->ll_status);
5922                 /* fall through */
5923         case LS_SCANNING_PHASE1:
5924         case LS_SCANNING_PHASE2:
5925                 /* No need to store the status to disk right now.
5926                  * If the system crashed before the status stored,
5927                  * it will be loaded back when next time. */
5928                 lo->ll_status = LS_CRASHED;
5929                 lo->ll_flags |= LF_INCOMPLETE;
5930                 /* fall through */
5931         case LS_PAUSED:
5932         case LS_CRASHED:
5933         case LS_CO_FAILED:
5934         case LS_CO_STOPPED:
5935         case LS_CO_PAUSED:
5936                 spin_lock(&lfsck->li_lock);
5937                 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
5938                 spin_unlock(&lfsck->li_lock);
5939                 break;
5940         }
5941
5942         if (lo->ll_flags & LF_CRASHED_LASTID) {
5943                 LASSERT(lfsck->li_out_notify != NULL);
5944
5945                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5946                                      LE_LASTID_REBUILDING);
5947         }
5948
5949         GOTO(out, rc = 0);
5950
5951 out:
5952         if (root != NULL && !IS_ERR(root))
5953                 lu_object_put(env, &root->do_lu);
5954
5955         if (rc != 0) {
5956                 lfsck_component_cleanup(env, com);
5957                 CERROR("%s: fail to init layout LFSCK component: rc = %d\n",
5958                        lfsck_lfsck2name(lfsck), rc);
5959         }
5960
5961         return rc;
5962 }
5963
5964 struct lfsck_orphan_it {
5965         struct lfsck_component           *loi_com;
5966         struct lfsck_rbtree_node         *loi_lrn;
5967         struct lfsck_layout_slave_target *loi_llst;
5968         struct lu_fid                     loi_key;
5969         struct lu_orphan_rec              loi_rec;
5970         __u64                             loi_hash;
5971         unsigned int                      loi_over:1;
5972 };
5973
5974 static int lfsck_fid_match_idx(const struct lu_env *env,
5975                                struct lfsck_instance *lfsck,
5976                                const struct lu_fid *fid, int idx)
5977 {
5978         struct seq_server_site  *ss;
5979         struct lu_server_fld    *sf;
5980         struct lu_seq_range      range  = { 0 };
5981         int                      rc;
5982
5983         /* All abnormal cases will be returned to MDT0. */
5984         if (!fid_is_norm(fid)) {
5985                 if (idx == 0)
5986                         return 1;
5987
5988                 return 0;
5989         }
5990
5991         ss = lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
5992         if (unlikely(ss == NULL))
5993                 return -ENOTCONN;
5994
5995         sf = ss->ss_server_fld;
5996         LASSERT(sf != NULL);
5997
5998         fld_range_set_any(&range);
5999         rc = fld_server_lookup(env, sf, fid_seq(fid), &range);
6000         if (rc != 0)
6001                 return rc;
6002
6003         if (!fld_range_is_mdt(&range))
6004                 return -EINVAL;
6005
6006         if (range.lsr_index == idx)
6007                 return 1;
6008
6009         return 0;
6010 }
6011
6012 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
6013                                         struct dt_device *dev,
6014                                         struct dt_object *obj)
6015 {
6016         struct thandle *handle;
6017         int             rc;
6018         ENTRY;
6019
6020         handle = dt_trans_create(env, dev);
6021         if (IS_ERR(handle))
6022                 RETURN_EXIT;
6023
6024         rc = dt_declare_ref_del(env, obj, handle);
6025         if (rc != 0)
6026                 GOTO(stop, rc);
6027
6028         rc = dt_declare_destroy(env, obj, handle);
6029         if (rc != 0)
6030                 GOTO(stop, rc);
6031
6032         rc = dt_trans_start_local(env, dev, handle);
6033         if (rc != 0)
6034                 GOTO(stop, rc);
6035
6036         dt_write_lock(env, obj, 0);
6037         rc = dt_ref_del(env, obj, handle);
6038         if (rc == 0)
6039                 rc = dt_destroy(env, obj, handle);
6040         dt_write_unlock(env, obj);
6041
6042         GOTO(stop, rc);
6043
6044 stop:
6045         dt_trans_stop(env, dev, handle);
6046
6047         CDEBUG(D_LFSCK, "destroy orphan OST-object "DFID": rc = %d\n",
6048                PFID(lfsck_dto2fid(obj)), rc);
6049
6050         RETURN_EXIT;
6051 }
6052
6053 static int lfsck_orphan_index_lookup(const struct lu_env *env,
6054                                      struct dt_object *dt,
6055                                      struct dt_rec *rec,
6056                                      const struct dt_key *key,
6057                                      struct lustre_capa *capa)
6058 {
6059         return -EOPNOTSUPP;
6060 }
6061
6062 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
6063                                              struct dt_object *dt,
6064                                              const struct dt_rec *rec,
6065                                              const struct dt_key *key,
6066                                              struct thandle *handle)
6067 {
6068         return -EOPNOTSUPP;
6069 }
6070
6071 static int lfsck_orphan_index_insert(const struct lu_env *env,
6072                                      struct dt_object *dt,
6073                                      const struct dt_rec *rec,
6074                                      const struct dt_key *key,
6075                                      struct thandle *handle,
6076                                      struct lustre_capa *capa,
6077                                      int ignore_quota)
6078 {
6079         return -EOPNOTSUPP;
6080 }
6081
6082 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
6083                                              struct dt_object *dt,
6084                                              const struct dt_key *key,
6085                                              struct thandle *handle)
6086 {
6087         return -EOPNOTSUPP;
6088 }
6089
6090 static int lfsck_orphan_index_delete(const struct lu_env *env,
6091                                      struct dt_object *dt,
6092                                      const struct dt_key *key,
6093                                      struct thandle *handle,
6094                                      struct lustre_capa *capa)
6095 {
6096         return -EOPNOTSUPP;
6097 }
6098
6099 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
6100                                           struct dt_object *dt,
6101                                           __u32 attr,
6102                                           struct lustre_capa *capa)
6103 {
6104         struct dt_device                *dev    = lu2dt_dev(dt->do_lu.lo_dev);
6105         struct lfsck_instance           *lfsck;
6106         struct lfsck_component          *com    = NULL;
6107         struct lfsck_layout_slave_data  *llsd;
6108         struct lfsck_orphan_it          *it     = NULL;
6109         int                              rc     = 0;
6110         ENTRY;
6111
6112         lfsck = lfsck_instance_find(dev, true, false);
6113         if (unlikely(lfsck == NULL))
6114                 RETURN(ERR_PTR(-ENXIO));
6115
6116         com = lfsck_component_find(lfsck, LFSCK_TYPE_LAYOUT);
6117         if (unlikely(com == NULL))
6118                 GOTO(out, rc = -ENOENT);
6119
6120         llsd = com->lc_data;
6121         if (!llsd->llsd_rbtree_valid)
6122                 GOTO(out, rc = -ESRCH);
6123
6124         OBD_ALLOC_PTR(it);
6125         if (it == NULL)
6126                 GOTO(out, rc = -ENOMEM);
6127
6128         it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
6129         if (it->loi_llst == NULL)
6130                 GOTO(out, rc = -ENXIO);
6131
6132         if (dev->dd_record_fid_accessed) {
6133                 /* The first iteration against the rbtree, scan the whole rbtree
6134                  * to remove the nodes which do NOT need to be handled. */
6135                 write_lock(&llsd->llsd_rb_lock);
6136                 if (dev->dd_record_fid_accessed) {
6137                         struct rb_node                  *node;
6138                         struct rb_node                  *next;
6139                         struct lfsck_rbtree_node        *lrn;
6140
6141                         /* No need to record the fid accessing anymore. */
6142                         dev->dd_record_fid_accessed = 0;
6143
6144                         node = rb_first(&llsd->llsd_rb_root);
6145                         while (node != NULL) {
6146                                 next = rb_next(node);
6147                                 lrn = rb_entry(node, struct lfsck_rbtree_node,
6148                                                lrn_node);
6149                                 if (atomic_read(&lrn->lrn_known_count) <=
6150                                     atomic_read(&lrn->lrn_accessed_count)) {
6151                                         rb_erase(node, &llsd->llsd_rb_root);
6152                                         lfsck_rbtree_free(lrn);
6153                                 }
6154                                 node = next;
6155                         }
6156                 }
6157                 write_unlock(&llsd->llsd_rb_lock);
6158         }
6159
6160         /* read lock the rbtree when init, and unlock when fini */
6161         read_lock(&llsd->llsd_rb_lock);
6162         it->loi_com = com;
6163         com = NULL;
6164
6165         GOTO(out, rc = 0);
6166
6167 out:
6168         if (com != NULL)
6169                 lfsck_component_put(env, com);
6170
6171         CDEBUG(D_LFSCK, "%s: init the orphan iteration: rc = %d\n",
6172                lfsck_lfsck2name(lfsck), rc);
6173
6174         lfsck_instance_put(env, lfsck);
6175         if (rc != 0) {
6176                 if (it != NULL)
6177                         OBD_FREE_PTR(it);
6178
6179                 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
6180         }
6181
6182         return (struct dt_it *)it;
6183 }
6184
6185 static void lfsck_orphan_it_fini(const struct lu_env *env,
6186                                  struct dt_it *di)
6187 {
6188         struct lfsck_orphan_it           *it    = (struct lfsck_orphan_it *)di;
6189         struct lfsck_component           *com   = it->loi_com;
6190         struct lfsck_layout_slave_data   *llsd;
6191         struct lfsck_layout_slave_target *llst;
6192
6193         if (com != NULL) {
6194                 CDEBUG(D_LFSCK, "%s: fini the orphan iteration\n",
6195                        lfsck_lfsck2name(com->lc_lfsck));
6196
6197                 llsd = com->lc_data;
6198                 read_unlock(&llsd->llsd_rb_lock);
6199                 llst = it->loi_llst;
6200                 LASSERT(llst != NULL);
6201
6202                 /* Save the key and hash for iterate next. */
6203                 llst->llst_fid = it->loi_key;
6204                 llst->llst_hash = it->loi_hash;
6205                 lfsck_layout_llst_put(llst);
6206                 lfsck_component_put(env, com);
6207         }
6208         OBD_FREE_PTR(it);
6209 }
6210
6211 /**
6212  * \retval       +1: the iteration finished
6213  * \retval        0: on success, not finished
6214  * \retval      -ve: on error
6215  */
6216 static int lfsck_orphan_it_next(const struct lu_env *env,
6217                                 struct dt_it *di)
6218 {
6219         struct lfsck_thread_info        *info   = lfsck_env_info(env);
6220         struct filter_fid_old           *pfid   = &info->lti_old_pfid;
6221         struct lu_attr                  *la     = &info->lti_la;
6222         struct lfsck_orphan_it          *it     = (struct lfsck_orphan_it *)di;
6223         struct lu_fid                   *key    = &it->loi_key;
6224         struct lu_orphan_rec            *rec    = &it->loi_rec;
6225         struct lfsck_component          *com    = it->loi_com;
6226         struct lfsck_instance           *lfsck  = com->lc_lfsck;
6227         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
6228         struct dt_object                *obj;
6229         struct lfsck_rbtree_node        *lrn;
6230         int                              pos;
6231         int                              rc;
6232         __u32                            save;
6233         __u32                            idx    = it->loi_llst->llst_index;
6234         bool                             exact  = false;
6235         ENTRY;
6236
6237         if (it->loi_over)
6238                 RETURN(1);
6239
6240 again0:
6241         lrn = it->loi_lrn;
6242         if (lrn == NULL) {
6243                 lrn = lfsck_rbtree_search(llsd, key, &exact);
6244                 if (lrn == NULL) {
6245                         it->loi_over = 1;
6246                         RETURN(1);
6247                 }
6248
6249                 it->loi_lrn = lrn;
6250                 if (!exact) {
6251                         key->f_seq = lrn->lrn_seq;
6252                         key->f_oid = lrn->lrn_first_oid;
6253                         key->f_ver = 0;
6254                 }
6255         } else {
6256                 key->f_oid++;
6257                 if (unlikely(key->f_oid == 0)) {
6258                         key->f_seq++;
6259                         it->loi_lrn = NULL;
6260                         goto again0;
6261                 }
6262
6263                 if (key->f_oid >=
6264                     lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
6265                         it->loi_lrn = NULL;
6266                         goto again0;
6267                 }
6268         }
6269
6270         if (unlikely(atomic_read(&lrn->lrn_known_count) <=
6271                      atomic_read(&lrn->lrn_accessed_count))) {
6272                 struct rb_node *next = rb_next(&lrn->lrn_node);
6273
6274                 while (next != NULL) {
6275                         lrn = rb_entry(next, struct lfsck_rbtree_node,
6276                                        lrn_node);
6277                         if (atomic_read(&lrn->lrn_known_count) >
6278                             atomic_read(&lrn->lrn_accessed_count))
6279                                 break;
6280                         next = rb_next(next);
6281                 }
6282
6283                 if (next == NULL) {
6284                         it->loi_over = 1;
6285                         RETURN(1);
6286                 }
6287
6288                 it->loi_lrn = lrn;
6289                 key->f_seq = lrn->lrn_seq;
6290                 key->f_oid = lrn->lrn_first_oid;
6291                 key->f_ver = 0;
6292         }
6293
6294         pos = key->f_oid - lrn->lrn_first_oid;
6295
6296 again1:
6297         pos = find_next_bit(lrn->lrn_known_bitmap,
6298                             LFSCK_RBTREE_BITMAP_WIDTH, pos);
6299         if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
6300                 key->f_oid = lrn->lrn_first_oid + pos;
6301                 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
6302                         key->f_seq++;
6303                         key->f_oid = 0;
6304                 }
6305                 it->loi_lrn = NULL;
6306                 goto again0;
6307         }
6308
6309         if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
6310                 pos++;
6311                 goto again1;
6312         }
6313
6314         key->f_oid = lrn->lrn_first_oid + pos;
6315         obj = lfsck_object_find(env, lfsck, key);
6316         if (IS_ERR(obj)) {
6317                 rc = PTR_ERR(obj);
6318                 if (rc == -ENOENT) {
6319                         pos++;
6320                         goto again1;
6321                 }
6322                 RETURN(rc);
6323         }
6324
6325         dt_read_lock(env, obj, 0);
6326         if (!dt_object_exists(obj)) {
6327                 dt_read_unlock(env, obj);
6328                 lfsck_object_put(env, obj);
6329                 pos++;
6330                 goto again1;
6331         }
6332
6333         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
6334         if (rc != 0)
6335                 GOTO(out, rc);
6336
6337         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, pfid, sizeof(*pfid)),
6338                           XATTR_NAME_FID, BYPASS_CAPA);
6339         if (rc == -ENODATA) {
6340                 /* For the pre-created OST-object, update the bitmap to avoid
6341                  * others LFSCK (second phase) iteration to touch it again. */
6342                 if (la->la_ctime == 0) {
6343                         if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
6344                                 atomic_inc(&lrn->lrn_accessed_count);
6345
6346                         /* For the race between repairing dangling referenced
6347                          * MDT-object and unlink the file, it may left orphan
6348                          * OST-object there. Destroy it now! */
6349                         if (unlikely(!(la->la_mode & S_ISUID))) {
6350                                 dt_read_unlock(env, obj);
6351                                 lfsck_layout_destroy_orphan(env,
6352                                                             lfsck->li_bottom,
6353                                                             obj);
6354                                 lfsck_object_put(env, obj);
6355                                 pos++;
6356                                 goto again1;
6357                         }
6358                 } else if (idx == 0) {
6359                         /* If the orphan OST-object has no parent information,
6360                          * regard it as referenced by the MDT-object on MDT0. */
6361                         fid_zero(&rec->lor_fid);
6362                         rec->lor_uid = la->la_uid;
6363                         rec->lor_gid = la->la_gid;
6364                         GOTO(out, rc = 0);
6365                 }
6366
6367                 dt_read_unlock(env, obj);
6368                 lfsck_object_put(env, obj);
6369                 pos++;
6370                 goto again1;
6371         }
6372
6373         if (rc < 0)
6374                 GOTO(out, rc);
6375
6376         if (rc != sizeof(struct filter_fid) &&
6377             rc != sizeof(struct filter_fid_old))
6378                 GOTO(out, rc = -EINVAL);
6379
6380         fid_le_to_cpu(&rec->lor_fid, &pfid->ff_parent);
6381         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
6382          * MDT-object's FID::f_ver, instead it is the OST-object index in its
6383          * parent MDT-object's layout EA. */
6384         save = rec->lor_fid.f_stripe_idx;
6385         rec->lor_fid.f_ver = 0;
6386         rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_fid, idx);
6387         /* If the orphan OST-object does not claim the MDT, then next.
6388          *
6389          * If we do not know whether it matches or not, then return it
6390          * to the MDT for further check. */
6391         if (rc == 0) {
6392                 dt_read_unlock(env, obj);
6393                 lfsck_object_put(env, obj);
6394                 pos++;
6395                 goto again1;
6396         }
6397
6398         rec->lor_fid.f_stripe_idx = save;
6399         rec->lor_uid = la->la_uid;
6400         rec->lor_gid = la->la_gid;
6401
6402         CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u\n",
6403                lfsck_lfsck2name(com->lc_lfsck), PFID(key), PFID(&rec->lor_fid),
6404                rec->lor_uid, rec->lor_gid);
6405
6406         GOTO(out, rc = 0);
6407
6408 out:
6409         dt_read_unlock(env, obj);
6410         lfsck_object_put(env, obj);
6411         if (rc == 0)
6412                 it->loi_hash++;
6413
6414         return rc;
6415 }
6416
6417 /**
6418  * \retval       +1: locate to the exactly position
6419  * \retval        0: cannot locate to the exactly position,
6420  *                   call next() to move to a valid position.
6421  * \retval      -ve: on error
6422  */
6423 static int lfsck_orphan_it_get(const struct lu_env *env,
6424                                struct dt_it *di,
6425                                const struct dt_key *key)
6426 {
6427         struct lfsck_orphan_it  *it   = (struct lfsck_orphan_it *)di;
6428         int                      rc;
6429
6430         it->loi_key = *(struct lu_fid *)key;
6431         rc = lfsck_orphan_it_next(env, di);
6432         if (rc == 1)
6433                 return 0;
6434
6435         if (rc == 0)
6436                 return 1;
6437
6438         return rc;
6439 }
6440
6441 static void lfsck_orphan_it_put(const struct lu_env *env,
6442                                 struct dt_it *di)
6443 {
6444 }
6445
6446 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
6447                                           const struct dt_it *di)
6448 {
6449         struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6450
6451         return (struct dt_key *)&it->loi_key;
6452 }
6453
6454 static int lfsck_orphan_it_key_size(const struct lu_env *env,
6455                                     const struct dt_it *di)
6456 {
6457         return sizeof(struct lu_fid);
6458 }
6459
6460 static int lfsck_orphan_it_rec(const struct lu_env *env,
6461                                const struct dt_it *di,
6462                                struct dt_rec *rec,
6463                                __u32 attr)
6464 {
6465         struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6466
6467         *(struct lu_orphan_rec *)rec = it->loi_rec;
6468
6469         return 0;
6470 }
6471
6472 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
6473                                    const struct dt_it *di)
6474 {
6475         struct lfsck_orphan_it  *it   = (struct lfsck_orphan_it *)di;
6476
6477         return it->loi_hash;
6478 }
6479
6480 /**
6481  * \retval       +1: locate to the exactly position
6482  * \retval        0: cannot locate to the exactly position,
6483  *                   call next() to move to a valid position.
6484  * \retval      -ve: on error
6485  */
6486 static int lfsck_orphan_it_load(const struct lu_env *env,
6487                                 const struct dt_it *di,
6488                                 __u64 hash)
6489 {
6490         struct lfsck_orphan_it           *it   = (struct lfsck_orphan_it *)di;
6491         struct lfsck_layout_slave_target *llst = it->loi_llst;
6492         int                               rc;
6493
6494         LASSERT(llst != NULL);
6495
6496         if (hash != llst->llst_hash) {
6497                 CDEBUG(D_LFSCK, "%s: the given hash "LPU64" for orphan "
6498                        "iteration does not match the one when fini "
6499                        LPU64", to be reset.\n",
6500                        lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
6501                        llst->llst_hash);
6502                 fid_zero(&llst->llst_fid);
6503                 llst->llst_hash = 0;
6504         }
6505
6506         it->loi_key = llst->llst_fid;
6507         it->loi_hash = llst->llst_hash;
6508         rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
6509         if (rc == 1)
6510                 return 0;
6511
6512         if (rc == 0)
6513                 return 1;
6514
6515         return rc;
6516 }
6517
6518 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
6519                                    const struct dt_it *di,
6520                                    void *key_rec)
6521 {
6522         return 0;
6523 }
6524
6525 const struct dt_index_operations lfsck_orphan_index_ops = {
6526         .dio_lookup             = lfsck_orphan_index_lookup,
6527         .dio_declare_insert     = lfsck_orphan_index_declare_insert,
6528         .dio_insert             = lfsck_orphan_index_insert,
6529         .dio_declare_delete     = lfsck_orphan_index_declare_delete,
6530         .dio_delete             = lfsck_orphan_index_delete,
6531         .dio_it = {
6532                 .init           = lfsck_orphan_it_init,
6533                 .fini           = lfsck_orphan_it_fini,
6534                 .get            = lfsck_orphan_it_get,
6535                 .put            = lfsck_orphan_it_put,
6536                 .next           = lfsck_orphan_it_next,
6537                 .key            = lfsck_orphan_it_key,
6538                 .key_size       = lfsck_orphan_it_key_size,
6539                 .rec            = lfsck_orphan_it_rec,
6540                 .store          = lfsck_orphan_it_store,
6541                 .load           = lfsck_orphan_it_load,
6542                 .key_rec        = lfsck_orphan_it_key_rec,
6543         }
6544 };