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