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_linkea.h>
  43 #include <lustre_fid.h>
  44 #include <lustre_lib.h>
  45 #include <lustre_net.h>
  46 #include <lustre/lustre_user.h>
  47 #include <md_object.h>
  48 #include <obd_class.h>
  49
  50 #include "lfsck_internal.h"
  51
  52 #define LFSCK_LAYOUT_MAGIC              0xB173AE14
  53
  54 static const char lfsck_layout_name[] = "lfsck_layout";
  55
  56 struct lfsck_layout_seq {
  57         struct list_head         lls_list;
  58         __u64                    lls_seq;
  59         __u64                    lls_lastid;
  60         __u64                    lls_lastid_known;
  61         struct dt_object        *lls_lastid_obj;
  62         unsigned int             lls_dirty:1;
  63 };
  64
  65 struct lfsck_layout_slave_target {
  66         /* link into lfsck_layout_slave_data::llsd_master_list. */
  67         struct list_head        llst_list;
  68         /* The position for next record in the rbtree for iteration. */
  69         struct lu_fid           llst_fid;
  70         /* Dummy hash for iteration against the rbtree. */
  71         __u64                   llst_hash;
  72         __u64                   llst_gen;
  73         atomic_t                llst_ref;
  74         __u32                   llst_index;
  75 };
  76
  77 struct lfsck_layout_slave_data {
  78         /* list for lfsck_layout_seq */
  79         struct list_head         llsd_seq_list;
  80
  81         /* list for the masters involve layout verification. */
  82         struct list_head         llsd_master_list;
  83         spinlock_t               llsd_lock;
  84         __u64                    llsd_touch_gen;
  85         struct dt_object        *llsd_rb_obj;
  86         struct rb_root           llsd_rb_root;
  87         rwlock_t                 llsd_rb_lock;
  88         unsigned int             llsd_rbtree_valid:1;
  89 };
  90
  91 struct lfsck_layout_object {
  92         struct dt_object        *llo_obj;
  93         struct lu_attr           llo_attr;
  94         atomic_t                 llo_ref;
  95         __u16                    llo_gen;
  96 };
  97
  98 struct lfsck_layout_req {
  99         struct list_head                 llr_list;
 100         struct lfsck_layout_object      *llr_parent;
 101         struct dt_object                *llr_child;
 102         __u32                            llr_ost_idx;
 103         __u32                            llr_lov_idx; /* offset in LOV EA */
 104 };
 105
 106 struct lfsck_layout_master_data {
 107         spinlock_t              llmd_lock;
 108         struct list_head        llmd_req_list;
 109
 110         /* list for the ost targets involve layout verification. */
 111         struct list_head        llmd_ost_list;
 112
 113         /* list for the ost targets in phase1 scanning. */
 114         struct list_head        llmd_ost_phase1_list;
 115
 116         /* list for the ost targets in phase1 scanning. */
 117         struct list_head        llmd_ost_phase2_list;
 118
 119         /* list for the mdt targets involve layout verification. */
 120         struct list_head        llmd_mdt_list;
 121
 122         /* list for the mdt targets in phase1 scanning. */
 123         struct list_head        llmd_mdt_phase1_list;
 124
 125         /* list for the mdt targets in phase1 scanning. */
 126         struct list_head        llmd_mdt_phase2_list;
 127
 128         struct ptlrpc_thread    llmd_thread;
 129         __u32                   llmd_touch_gen;
 130         int                     llmd_prefetched;
 131         int                     llmd_assistant_status;
 132         int                     llmd_post_result;
 133         unsigned int            llmd_to_post:1,
 134                                 llmd_to_double_scan:1,
 135                                 llmd_in_double_scan:1,
 136                                 llmd_exit:1;
 137 };
 138
 139 struct lfsck_layout_slave_async_args {
 140         struct obd_export                *llsaa_exp;
 141         struct lfsck_component           *llsaa_com;
 142         struct lfsck_layout_slave_target *llsaa_llst;
 143 };
 144
 145 static struct lfsck_layout_object *
 146 lfsck_layout_object_init(const struct lu_env *env, struct dt_object *obj,
 147                          __u16 gen)
 148 {
 149         struct lfsck_layout_object *llo;
 150         int                         rc;
 151
 152         OBD_ALLOC_PTR(llo);
 153         if (llo == NULL)
 154                 return ERR_PTR(-ENOMEM);
 155
 156         rc = dt_attr_get(env, obj, &llo->llo_attr, BYPASS_CAPA);
 157         if (rc != 0) {
 158                 OBD_FREE_PTR(llo);
 159
 160                 return ERR_PTR(rc);
 161         }
 162
 163         lu_object_get(&obj->do_lu);
 164         llo->llo_obj = obj;
 165         /* The gen can be used to check whether some others have changed the
 166          * file layout after LFSCK pre-fetching but before real verification. */
 167         llo->llo_gen = gen;
 168         atomic_set(&llo->llo_ref, 1);
 169
 170         return llo;
 171 }
 172
 173 static inline void
 174 lfsck_layout_llst_put(struct lfsck_layout_slave_target *llst)
 175 {
 176         if (atomic_dec_and_test(&llst->llst_ref)) {
 177                 LASSERT(list_empty(&llst->llst_list));
 178
 179                 OBD_FREE_PTR(llst);
 180         }
 181 }
 182
 183 static inline int
 184 lfsck_layout_llst_add(struct lfsck_layout_slave_data *llsd, __u32 index)
 185 {
 186         struct lfsck_layout_slave_target *llst;
 187         struct lfsck_layout_slave_target *tmp;
 188         int                               rc   = 0;
 189
 190         OBD_ALLOC_PTR(llst);
 191         if (llst == NULL)
 192                 return -ENOMEM;
 193
 194         INIT_LIST_HEAD(&llst->llst_list);
 195         llst->llst_gen = 0;
 196         llst->llst_index = index;
 197         atomic_set(&llst->llst_ref, 1);
 198
 199         spin_lock(&llsd->llsd_lock);
 200         list_for_each_entry(tmp, &llsd->llsd_master_list, llst_list) {
 201                 if (tmp->llst_index == index) {
 202                         rc = -EALREADY;
 203                         break;
 204                 }
 205         }
 206         if (rc == 0)
 207                 list_add_tail(&llst->llst_list, &llsd->llsd_master_list);
 208         spin_unlock(&llsd->llsd_lock);
 209
 210         if (rc != 0)
 211                 OBD_FREE_PTR(llst);
 212
 213         return rc;
 214 }
 215
 216 static inline void
 217 lfsck_layout_llst_del(struct lfsck_layout_slave_data *llsd,
 218                       struct lfsck_layout_slave_target *llst)
 219 {
 220         bool del = false;
 221
 222         spin_lock(&llsd->llsd_lock);
 223         if (!list_empty(&llst->llst_list)) {
 224                 list_del_init(&llst->llst_list);
 225                 del = true;
 226         }
 227         spin_unlock(&llsd->llsd_lock);
 228
 229         if (del)
 230                 lfsck_layout_llst_put(llst);
 231 }
 232
 233 static inline struct lfsck_layout_slave_target *
 234 lfsck_layout_llst_find_and_del(struct lfsck_layout_slave_data *llsd,
 235                                __u32 index, bool unlink)
 236 {
 237         struct lfsck_layout_slave_target *llst;
 238
 239         spin_lock(&llsd->llsd_lock);
 240         list_for_each_entry(llst, &llsd->llsd_master_list, llst_list) {
 241                 if (llst->llst_index == index) {
 242                         if (unlink)
 243                                 list_del_init(&llst->llst_list);
 244                         else
 245                                 atomic_inc(&llst->llst_ref);
 246                         spin_unlock(&llsd->llsd_lock);
 247
 248                         return llst;
 249                 }
 250         }
 251         spin_unlock(&llsd->llsd_lock);
 252
 253         return NULL;
 254 }
 255
 256 static inline void lfsck_layout_object_put(const struct lu_env *env,
 257                                            struct lfsck_layout_object *llo)
 258 {
 259         if (atomic_dec_and_test(&llo->llo_ref)) {
 260                 lfsck_object_put(env, llo->llo_obj);
 261                 OBD_FREE_PTR(llo);
 262         }
 263 }
 264
 265 static struct lfsck_layout_req *
 266 lfsck_layout_req_init(struct lfsck_layout_object *parent,
 267                       struct dt_object *child, __u32 ost_idx, __u32 lov_idx)
 268 {
 269         struct lfsck_layout_req *llr;
 270
 271         OBD_ALLOC_PTR(llr);
 272         if (llr == NULL)
 273                 return ERR_PTR(-ENOMEM);
 274
 275         INIT_LIST_HEAD(&llr->llr_list);
 276         atomic_inc(&parent->llo_ref);
 277         llr->llr_parent = parent;
 278         llr->llr_child = child;
 279         llr->llr_ost_idx = ost_idx;
 280         llr->llr_lov_idx = lov_idx;
 281
 282         return llr;
 283 }
 284
 285 static inline void lfsck_layout_req_fini(const struct lu_env *env,
 286                                          struct lfsck_layout_req *llr)
 287 {
 288         lu_object_put(env, &llr->llr_child->do_lu);
 289         lfsck_layout_object_put(env, llr->llr_parent);
 290         OBD_FREE_PTR(llr);
 291 }
 292
 293 static inline bool lfsck_layout_req_empty(struct lfsck_layout_master_data *llmd)
 294 {
 295         bool empty = false;
 296
 297         spin_lock(&llmd->llmd_lock);
 298         if (list_empty(&llmd->llmd_req_list))
 299                 empty = true;
 300         spin_unlock(&llmd->llmd_lock);
 301
 302         return empty;
 303 }
 304
 305 static int lfsck_layout_get_lovea(const struct lu_env *env,
 306                                   struct dt_object *obj,
 307                                   struct lu_buf *buf, ssize_t *buflen)
 308 {
 309         int rc;
 310
 311 again:
 312         rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LOV, BYPASS_CAPA);
 313         if (rc == -ERANGE) {
 314                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV,
 315                                   BYPASS_CAPA);
 316                 if (rc <= 0)
 317                         return rc;
 318
 319                 lu_buf_realloc(buf, rc);
 320                 if (buflen != NULL)
 321                         *buflen = buf->lb_len;
 322
 323                 if (buf->lb_buf == NULL)
 324                         return -ENOMEM;
 325
 326                 goto again;
 327         }
 328
 329         if (rc == -ENODATA)
 330                 rc = 0;
 331
 332         if (rc <= 0)
 333                 return rc;
 334
 335         if (unlikely(buf->lb_buf == NULL)) {
 336                 lu_buf_alloc(buf, rc);
 337                 if (buflen != NULL)
 338                         *buflen = buf->lb_len;
 339
 340                 if (buf->lb_buf == NULL)
 341                         return -ENOMEM;
 342
 343                 goto again;
 344         }
 345
 346         return rc;
 347 }
 348
 349 static int lfsck_layout_verify_header(struct lov_mds_md_v1 *lmm)
 350 {
 351         __u32 magic;
 352         __u32 pattern;
 353
 354         magic = le32_to_cpu(lmm->lmm_magic);
 355         /* If magic crashed, keep it there. Sometime later, during OST-object
 356          * orphan handling, if some OST-object(s) back-point to it, it can be
 357          * verified and repaired. */
 358         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) {
 359                 struct ost_id   oi;
 360                 int             rc;
 361
 362                 lmm_oi_cpu_to_le(&oi, &lmm->lmm_oi);
 363                 if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
 364                         rc = -EOPNOTSUPP;
 365                 else
 366                         rc = -EINVAL;
 367
 368                 CDEBUG(D_LFSCK, "%s LOV EA magic %u on "DOSTID"\n",
 369                        rc == -EINVAL ? "Unknown" : "Unsupported",
 370                        magic, POSTID(&oi));
 371
 372                 return rc;
 373         }
 374
 375         pattern = le32_to_cpu(lmm->lmm_pattern);
 376         /* XXX: currently, we only support LOV_PATTERN_RAID0. */
 377         if (lov_pattern(pattern) != LOV_PATTERN_RAID0) {
 378                 struct ost_id oi;
 379
 380                 lmm_oi_cpu_to_le(&oi, &lmm->lmm_oi);
 381                 CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u on "DOSTID"\n",
 382                        pattern, POSTID(&oi));
 383
 384                 return -EOPNOTSUPP;
 385         }
 386
 387         return 0;
 388 }
 389
 390 #define LFSCK_RBTREE_BITMAP_SIZE        PAGE_CACHE_SIZE
 391 #define LFSCK_RBTREE_BITMAP_WIDTH       (LFSCK_RBTREE_BITMAP_SIZE << 3)
 392 #define LFSCK_RBTREE_BITMAP_MASK        (LFSCK_RBTREE_BITMAP_WIDTH - 1)
 393
 394 struct lfsck_rbtree_node {
 395         struct rb_node   lrn_node;
 396         __u64            lrn_seq;
 397         __u32            lrn_first_oid;
 398         atomic_t         lrn_known_count;
 399         atomic_t         lrn_accessed_count;
 400         void            *lrn_known_bitmap;
 401         void            *lrn_accessed_bitmap;
 402 };
 403
 404 static inline int lfsck_rbtree_cmp(struct lfsck_rbtree_node *lrn,
 405                                    __u64 seq, __u32 oid)
 406 {
 407         if (seq < lrn->lrn_seq)
 408                 return -1;
 409
 410         if (seq > lrn->lrn_seq)
 411                 return 1;
 412
 413         if (oid < lrn->lrn_first_oid)
 414                 return -1;
 415
 416         if (oid - lrn->lrn_first_oid >= LFSCK_RBTREE_BITMAP_WIDTH)
 417                 return 1;
 418
 419         return 0;
 420 }
 421
 422 /* The caller should hold llsd->llsd_rb_lock. */
 423 static struct lfsck_rbtree_node *
 424 lfsck_rbtree_search(struct lfsck_layout_slave_data *llsd,
 425                     const struct lu_fid *fid, bool *exact)
 426 {
 427         struct rb_node           *node  = llsd->llsd_rb_root.rb_node;
 428         struct rb_node           *prev  = NULL;
 429         struct lfsck_rbtree_node *lrn   = NULL;
 430         int                       rc    = 0;
 431
 432         if (exact != NULL)
 433                 *exact = true;
 434
 435         while (node != NULL) {
 436                 prev = node;
 437                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 438                 rc = lfsck_rbtree_cmp(lrn, fid_seq(fid), fid_oid(fid));
 439                 if (rc < 0)
 440                         node = node->rb_left;
 441                 else if (rc > 0)
 442                         node = node->rb_right;
 443                 else
 444                         return lrn;
 445         }
 446
 447         if (exact == NULL)
 448                 return NULL;
 449
 450         /* If there is no exactly matched one, then to the next valid one. */
 451         *exact = false;
 452
 453         /* The rbtree is empty. */
 454         if (rc == 0)
 455                 return NULL;
 456
 457         if (rc < 0)
 458                 return lrn;
 459
 460         node = rb_next(prev);
 461
 462         /* The end of the rbtree. */
 463         if (node == NULL)
 464                 return NULL;
 465
 466         lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 467
 468         return lrn;
 469 }
 470
 471 static struct lfsck_rbtree_node *lfsck_rbtree_new(const struct lu_env *env,
 472                                                   const struct lu_fid *fid)
 473 {
 474         struct lfsck_rbtree_node *lrn;
 475
 476         OBD_ALLOC_PTR(lrn);
 477         if (lrn == NULL)
 478                 return ERR_PTR(-ENOMEM);
 479
 480         OBD_ALLOC(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 481         if (lrn->lrn_known_bitmap == NULL) {
 482                 OBD_FREE_PTR(lrn);
 483
 484                 return ERR_PTR(-ENOMEM);
 485         }
 486
 487         OBD_ALLOC(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 488         if (lrn->lrn_accessed_bitmap == NULL) {
 489                 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 490                 OBD_FREE_PTR(lrn);
 491
 492                 return ERR_PTR(-ENOMEM);
 493         }
 494
 495         rb_init_node(&lrn->lrn_node);
 496         lrn->lrn_seq = fid_seq(fid);
 497         lrn->lrn_first_oid = fid_oid(fid) & ~LFSCK_RBTREE_BITMAP_MASK;
 498         atomic_set(&lrn->lrn_known_count, 0);
 499         atomic_set(&lrn->lrn_accessed_count, 0);
 500
 501         return lrn;
 502 }
 503
 504 static void lfsck_rbtree_free(struct lfsck_rbtree_node *lrn)
 505 {
 506         OBD_FREE(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 507         OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 508         OBD_FREE_PTR(lrn);
 509 }
 510
 511 /* The caller should hold lock. */
 512 static struct lfsck_rbtree_node *
 513 lfsck_rbtree_insert(struct lfsck_layout_slave_data *llsd,
 514                     struct lfsck_rbtree_node *lrn)
 515 {
 516         struct rb_node           **pos    = &llsd->llsd_rb_root.rb_node;
 517         struct rb_node            *parent = NULL;
 518         struct lfsck_rbtree_node  *tmp;
 519         int                        rc;
 520
 521         while (*pos != NULL) {
 522                 parent = *pos;
 523                 tmp = rb_entry(parent, struct lfsck_rbtree_node, lrn_node);
 524                 rc = lfsck_rbtree_cmp(tmp, lrn->lrn_seq, lrn->lrn_first_oid);
 525                 if (rc < 0)
 526                         pos = &(*pos)->rb_left;
 527                 else if (rc > 0)
 528                         pos = &(*pos)->rb_right;
 529                 else
 530                         return tmp;
 531         }
 532
 533         rb_link_node(&lrn->lrn_node, parent, pos);
 534         rb_insert_color(&lrn->lrn_node, &llsd->llsd_rb_root);
 535
 536         return lrn;
 537 }
 538
 539 extern const struct dt_index_operations lfsck_orphan_index_ops;
 540
 541 static int lfsck_rbtree_setup(const struct lu_env *env,
 542                               struct lfsck_component *com)
 543 {
 544         struct lu_fid                   *fid    = &lfsck_env_info(env)->lti_fid;
 545         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 546         struct dt_device                *dev    = lfsck->li_bottom;
 547         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
 548         struct dt_object                *obj;
 549
 550         fid->f_seq = FID_SEQ_LAYOUT_RBTREE;
 551         fid->f_oid = lfsck_dev_idx(dev);
 552         fid->f_ver = 0;
 553         obj = dt_locate(env, dev, fid);
 554         if (IS_ERR(obj))
 555                 RETURN(PTR_ERR(obj));
 556
 557         /* Generate an in-RAM object to stand for the layout rbtree.
 558          * Scanning the layout rbtree will be via the iteration over
 559          * the object. In the future, the rbtree may be written onto
 560          * disk with the object.
 561          *
 562          * Mark the object to be as exist. */
 563         obj->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 564         obj->do_index_ops = &lfsck_orphan_index_ops;
 565         llsd->llsd_rb_obj = obj;
 566         llsd->llsd_rbtree_valid = 1;
 567         dev->dd_record_fid_accessed = 1;
 568
 569         return 0;
 570 }
 571
 572 static void lfsck_rbtree_cleanup(const struct lu_env *env,
 573                                  struct lfsck_component *com)
 574 {
 575         struct lfsck_instance           *lfsck = com->lc_lfsck;
 576         struct lfsck_layout_slave_data  *llsd  = com->lc_data;
 577         struct rb_node                  *node  = rb_first(&llsd->llsd_rb_root);
 578         struct rb_node                  *next;
 579         struct lfsck_rbtree_node        *lrn;
 580
 581         lfsck->li_bottom->dd_record_fid_accessed = 0;
 582         /* Invalid the rbtree, then no others will use it. */
 583         write_lock(&llsd->llsd_rb_lock);
 584         llsd->llsd_rbtree_valid = 0;
 585         write_unlock(&llsd->llsd_rb_lock);
 586
 587         while (node != NULL) {
 588                 next = rb_next(node);
 589                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 590                 rb_erase(node, &llsd->llsd_rb_root);
 591                 lfsck_rbtree_free(lrn);
 592                 node = next;
 593         }
 594
 595         if (llsd->llsd_rb_obj != NULL) {
 596                 lu_object_put(env, &llsd->llsd_rb_obj->do_lu);
 597                 llsd->llsd_rb_obj = NULL;
 598         }
 599 }
 600
 601 static void lfsck_rbtree_update_bitmap(const struct lu_env *env,
 602                                        struct lfsck_component *com,
 603                                        const struct lu_fid *fid,
 604                                        bool accessed)
 605 {
 606         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
 607         struct lfsck_rbtree_node        *lrn;
 608         bool                             insert = false;
 609         int                              idx;
 610         int                              rc     = 0;
 611         ENTRY;
 612
 613         CDEBUG(D_LFSCK, "%s: update bitmap for "DFID"\n",
 614                lfsck_lfsck2name(com->lc_lfsck), PFID(fid));
 615
 616         if (unlikely(!fid_is_sane(fid) || fid_is_last_id(fid)))
 617                 RETURN_EXIT;
 618
 619         if (!fid_is_idif(fid) && !fid_is_norm(fid))
 620                 RETURN_EXIT;
 621
 622         read_lock(&llsd->llsd_rb_lock);
 623         if (!llsd->llsd_rbtree_valid)
 624                 GOTO(unlock, rc = 0);
 625
 626         lrn = lfsck_rbtree_search(llsd, fid, NULL);
 627         if (lrn == NULL) {
 628                 struct lfsck_rbtree_node *tmp;
 629
 630                 LASSERT(!insert);
 631
 632                 read_unlock(&llsd->llsd_rb_lock);
 633                 tmp = lfsck_rbtree_new(env, fid);
 634                 if (IS_ERR(tmp))
 635                         GOTO(out, rc = PTR_ERR(tmp));
 636
 637                 insert = true;
 638                 write_lock(&llsd->llsd_rb_lock);
 639                 if (!llsd->llsd_rbtree_valid) {
 640                         lfsck_rbtree_free(tmp);
 641                         GOTO(unlock, rc = 0);
 642                 }
 643
 644                 lrn = lfsck_rbtree_insert(llsd, tmp);
 645                 if (lrn != tmp)
 646                         lfsck_rbtree_free(tmp);
 647         }
 648
 649         idx = fid_oid(fid) & LFSCK_RBTREE_BITMAP_MASK;
 650         /* Any accessed object must be a known object. */
 651         if (!test_and_set_bit(idx, lrn->lrn_known_bitmap))
 652                 atomic_inc(&lrn->lrn_known_count);
 653         if (accessed && !test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
 654                 atomic_inc(&lrn->lrn_accessed_count);
 655
 656         GOTO(unlock, rc = 0);
 657
 658 unlock:
 659         if (insert)
 660                 write_unlock(&llsd->llsd_rb_lock);
 661         else
 662                 read_unlock(&llsd->llsd_rb_lock);
 663 out:
 664         if (rc != 0 && accessed) {
 665                 struct lfsck_layout *lo = com->lc_file_ram;
 666
 667                 CERROR("%s: Fail to update object accessed bitmap, will cause "
 668                        "incorrect LFSCK OST-object handling, so disable it to "
 669                        "cancel orphan handling for related device. rc = %d.\n",
 670                        lfsck_lfsck2name(com->lc_lfsck), rc);
 671                 lo->ll_flags |= LF_INCOMPLETE;
 672                 lfsck_rbtree_cleanup(env, com);
 673         }
 674 }
 675
 676 static inline bool is_dummy_lov_ost_data(struct lov_ost_data_v1 *obj)
 677 {
 678         if (fid_is_zero(&obj->l_ost_oi.oi_fid) &&
 679             obj->l_ost_gen == 0 && obj->l_ost_idx == 0)
 680                 return true;
 681
 682         return false;
 683 }
 684
 685 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
 686                                    const struct lfsck_layout *src)
 687 {
 688         int i;
 689
 690         des->ll_magic = le32_to_cpu(src->ll_magic);
 691         des->ll_status = le32_to_cpu(src->ll_status);
 692         des->ll_flags = le32_to_cpu(src->ll_flags);
 693         des->ll_success_count = le32_to_cpu(src->ll_success_count);
 694         des->ll_run_time_phase1 = le32_to_cpu(src->ll_run_time_phase1);
 695         des->ll_run_time_phase2 = le32_to_cpu(src->ll_run_time_phase2);
 696         des->ll_time_last_complete = le64_to_cpu(src->ll_time_last_complete);
 697         des->ll_time_latest_start = le64_to_cpu(src->ll_time_latest_start);
 698         des->ll_time_last_checkpoint =
 699                                 le64_to_cpu(src->ll_time_last_checkpoint);
 700         des->ll_pos_latest_start = le64_to_cpu(src->ll_pos_latest_start);
 701         des->ll_pos_last_checkpoint = le64_to_cpu(src->ll_pos_last_checkpoint);
 702         des->ll_pos_first_inconsistent =
 703                         le64_to_cpu(src->ll_pos_first_inconsistent);
 704         des->ll_objs_checked_phase1 = le64_to_cpu(src->ll_objs_checked_phase1);
 705         des->ll_objs_failed_phase1 = le64_to_cpu(src->ll_objs_failed_phase1);
 706         des->ll_objs_checked_phase2 = le64_to_cpu(src->ll_objs_checked_phase2);
 707         des->ll_objs_failed_phase2 = le64_to_cpu(src->ll_objs_failed_phase2);
 708         for (i = 0; i < LLIT_MAX; i++)
 709                 des->ll_objs_repaired[i] =
 710                                 le64_to_cpu(src->ll_objs_repaired[i]);
 711         des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
 712 }
 713
 714 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
 715                                    const struct lfsck_layout *src)
 716 {
 717         int i;
 718
 719         des->ll_magic = cpu_to_le32(src->ll_magic);
 720         des->ll_status = cpu_to_le32(src->ll_status);
 721         des->ll_flags = cpu_to_le32(src->ll_flags);
 722         des->ll_success_count = cpu_to_le32(src->ll_success_count);
 723         des->ll_run_time_phase1 = cpu_to_le32(src->ll_run_time_phase1);
 724         des->ll_run_time_phase2 = cpu_to_le32(src->ll_run_time_phase2);
 725         des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
 726         des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
 727         des->ll_time_last_checkpoint =
 728                                 cpu_to_le64(src->ll_time_last_checkpoint);
 729         des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
 730         des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
 731         des->ll_pos_first_inconsistent =
 732                         cpu_to_le64(src->ll_pos_first_inconsistent);
 733         des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
 734         des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
 735         des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
 736         des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
 737         for (i = 0; i < LLIT_MAX; i++)
 738                 des->ll_objs_repaired[i] =
 739                                 cpu_to_le64(src->ll_objs_repaired[i]);
 740         des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
 741 }
 742
 743 /**
 744  * \retval +ve: the lfsck_layout is broken, the caller should reset it.
 745  * \retval 0: succeed.
 746  * \retval -ve: failed cases.
 747  */
 748 static int lfsck_layout_load(const struct lu_env *env,
 749                              struct lfsck_component *com)
 750 {
 751         struct lfsck_layout             *lo     = com->lc_file_ram;
 752         const struct dt_body_operations *dbo    = com->lc_obj->do_body_ops;
 753         ssize_t                          size   = com->lc_file_size;
 754         loff_t                           pos    = 0;
 755         int                              rc;
 756
 757         rc = dbo->dbo_read(env, com->lc_obj,
 758                            lfsck_buf_get(env, com->lc_file_disk, size), &pos,
 759                            BYPASS_CAPA);
 760         if (rc == 0) {
 761                 return -ENOENT;
 762         } else if (rc < 0) {
 763                 CWARN("%s: failed to load lfsck_layout: rc = %d\n",
 764                       lfsck_lfsck2name(com->lc_lfsck), rc);
 765                 return rc;
 766         } else if (rc != size) {
 767                 CWARN("%s: crashed lfsck_layout, to be reset: rc = %d\n",
 768                       lfsck_lfsck2name(com->lc_lfsck), rc);
 769                 return 1;
 770         }
 771
 772         lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
 773         if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
 774                 CWARN("%s: invalid lfsck_layout magic %#x != %#x, "
 775                       "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
 776                       lo->ll_magic, LFSCK_LAYOUT_MAGIC);
 777                 return 1;
 778         }
 779
 780         return 0;
 781 }
 782
 783 static int lfsck_layout_store(const struct lu_env *env,
 784                               struct lfsck_component *com)
 785 {
 786         struct dt_object         *obj           = com->lc_obj;
 787         struct lfsck_instance    *lfsck         = com->lc_lfsck;
 788         struct lfsck_layout      *lo            = com->lc_file_disk;
 789         struct thandle           *handle;
 790         ssize_t                   size          = com->lc_file_size;
 791         loff_t                    pos           = 0;
 792         int                       rc;
 793         ENTRY;
 794
 795         lfsck_layout_cpu_to_le(lo, com->lc_file_ram);
 796         handle = dt_trans_create(env, lfsck->li_bottom);
 797         if (IS_ERR(handle)) {
 798                 rc = PTR_ERR(handle);
 799                 CERROR("%s: fail to create trans for storing lfsck_layout: "
 800                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
 801                 RETURN(rc);
 802         }
 803
 804         rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
 805                                      pos, handle);
 806         if (rc != 0) {
 807                 CERROR("%s: fail to declare trans for storing lfsck_layout(1): "
 808                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
 809                 GOTO(out, rc);
 810         }
 811
 812         rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
 813         if (rc != 0) {
 814                 CERROR("%s: fail to start trans for storing lfsck_layout: "
 815                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
 816                 GOTO(out, rc);
 817         }
 818
 819         rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos,
 820                              handle);
 821         if (rc != 0)
 822                 CERROR("%s: fail to store lfsck_layout(1): size = %d, "
 823                        "rc = %d\n", lfsck_lfsck2name(lfsck), (int)size, rc);
 824
 825         GOTO(out, rc);
 826
 827 out:
 828         dt_trans_stop(env, lfsck->li_bottom, handle);
 829
 830         return rc;
 831 }
 832
 833 static int lfsck_layout_init(const struct lu_env *env,
 834                              struct lfsck_component *com)
 835 {
 836         struct lfsck_layout *lo = com->lc_file_ram;
 837         int rc;
 838
 839         memset(lo, 0, com->lc_file_size);
 840         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
 841         lo->ll_status = LS_INIT;
 842         down_write(&com->lc_sem);
 843         rc = lfsck_layout_store(env, com);
 844         up_write(&com->lc_sem);
 845
 846         return rc;
 847 }
 848
 849 static int fid_is_for_ostobj(const struct lu_env *env, struct dt_device *dt,
 850                              struct dt_object *obj, const struct lu_fid *fid)
 851 {
 852         struct seq_server_site  *ss     = lu_site2seq(dt->dd_lu_dev.ld_site);
 853         struct lu_seq_range      range  = { 0 };
 854         struct lustre_mdt_attrs *lma;
 855         int                      rc;
 856
 857         fld_range_set_any(&range);
 858         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), &range);
 859         if (rc == 0) {
 860                 if (fld_range_is_ost(&range))
 861                         return 1;
 862
 863                 return 0;
 864         }
 865
 866         lma = &lfsck_env_info(env)->lti_lma;
 867         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, lma, sizeof(*lma)),
 868                           XATTR_NAME_LMA, BYPASS_CAPA);
 869         if (rc == sizeof(*lma)) {
 870                 lustre_lma_swab(lma);
 871
 872                 return lma->lma_compat & LMAC_FID_ON_OST ? 1 : 0;
 873         }
 874
 875         rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID, BYPASS_CAPA);
 876
 877         return rc > 0;
 878 }
 879
 880 static struct lfsck_layout_seq *
 881 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
 882 {
 883         struct lfsck_layout_seq *lls;
 884
 885         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
 886                 if (lls->lls_seq == seq)
 887                         return lls;
 888
 889                 if (lls->lls_seq > seq)
 890                         return NULL;
 891         }
 892
 893         return NULL;
 894 }
 895
 896 static void
 897 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
 898                         struct lfsck_layout_seq *lls)
 899 {
 900         struct lfsck_layout_seq *tmp;
 901         struct list_head        *pos = &llsd->llsd_seq_list;
 902
 903         list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
 904                 if (lls->lls_seq < tmp->lls_seq) {
 905                         pos = &tmp->lls_list;
 906                         break;
 907                 }
 908         }
 909         list_add_tail(&lls->lls_list, pos);
 910 }
 911
 912 static int
 913 lfsck_layout_lastid_create(const struct lu_env *env,
 914                            struct lfsck_instance *lfsck,
 915                            struct dt_object *obj)
 916 {
 917         struct lfsck_thread_info *info   = lfsck_env_info(env);
 918         struct lu_attr           *la     = &info->lti_la;
 919         struct dt_object_format  *dof    = &info->lti_dof;
 920         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
 921         struct dt_device         *dt     = lfsck->li_bottom;
 922         struct thandle           *th;
 923         __u64                     lastid = 0;
 924         loff_t                    pos    = 0;
 925         int                       rc;
 926         ENTRY;
 927
 928         CDEBUG(D_LFSCK, "To create LAST_ID for <seq> "LPX64"\n",
 929                fid_seq(lfsck_dto2fid(obj)));
 930
 931         if (bk->lb_param & LPF_DRYRUN)
 932                 return 0;
 933
 934         memset(la, 0, sizeof(*la));
 935         la->la_mode = S_IFREG |  S_IRUGO | S_IWUSR;
 936         la->la_valid = LA_MODE | LA_UID | LA_GID;
 937         dof->dof_type = dt_mode_to_dft(S_IFREG);
 938
 939         th = dt_trans_create(env, dt);
 940         if (IS_ERR(th))
 941                 RETURN(rc = PTR_ERR(th));
 942
 943         rc = dt_declare_create(env, obj, la, NULL, dof, th);
 944         if (rc != 0)
 945                 GOTO(stop, rc);
 946
 947         rc = dt_declare_record_write(env, obj,
 948                                      lfsck_buf_get(env, &lastid,
 949                                                    sizeof(lastid)),
 950                                      pos, th);
 951         if (rc != 0)
 952                 GOTO(stop, rc);
 953
 954         rc = dt_trans_start_local(env, dt, th);
 955         if (rc != 0)
 956                 GOTO(stop, rc);
 957
 958         dt_write_lock(env, obj, 0);
 959         if (likely(!dt_object_exists(obj))) {
 960                 rc = dt_create(env, obj, la, NULL, dof, th);
 961                 if (rc == 0)
 962                         rc = dt_record_write(env, obj,
 963                                 lfsck_buf_get(env, &lastid, sizeof(lastid)),
 964                                 &pos, th);
 965         }
 966         dt_write_unlock(env, obj);
 967
 968         GOTO(stop, rc);
 969
 970 stop:
 971         dt_trans_stop(env, dt, th);
 972
 973         return rc;
 974 }
 975
 976 static int
 977 lfsck_layout_lastid_reload(const struct lu_env *env,
 978                            struct lfsck_component *com,
 979                            struct lfsck_layout_seq *lls)
 980 {
 981         __u64   lastid;
 982         loff_t  pos     = 0;
 983         int     rc;
 984
 985         dt_read_lock(env, lls->lls_lastid_obj, 0);
 986         rc = dt_record_read(env, lls->lls_lastid_obj,
 987                             lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
 988         dt_read_unlock(env, lls->lls_lastid_obj);
 989         if (unlikely(rc != 0))
 990                 return rc;
 991
 992         lastid = le64_to_cpu(lastid);
 993         if (lastid < lls->lls_lastid_known) {
 994                 struct lfsck_instance   *lfsck  = com->lc_lfsck;
 995                 struct lfsck_layout     *lo     = com->lc_file_ram;
 996
 997                 lls->lls_lastid = lls->lls_lastid_known;
 998                 lls->lls_dirty = 1;
 999                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1000                         LASSERT(lfsck->li_out_notify != NULL);
1001
1002                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1003                                              LE_LASTID_REBUILDING);
1004                         lo->ll_flags |= LF_CRASHED_LASTID;
1005                 }
1006         } else if (lastid >= lls->lls_lastid) {
1007                 lls->lls_lastid = lastid;
1008                 lls->lls_dirty = 0;
1009         }
1010
1011         return 0;
1012 }
1013
1014 static int
1015 lfsck_layout_lastid_store(const struct lu_env *env,
1016                           struct lfsck_component *com)
1017 {
1018         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1019         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
1020         struct dt_device                *dt     = lfsck->li_bottom;
1021         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
1022         struct lfsck_layout_seq         *lls;
1023         struct thandle                  *th;
1024         __u64                            lastid;
1025         int                              rc     = 0;
1026         int                              rc1    = 0;
1027
1028         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1029                 loff_t pos = 0;
1030
1031                 /* XXX: Add the code back if we really found related
1032                  *      inconsistent cases in the future. */
1033 #if 0
1034                 if (!lls->lls_dirty) {
1035                         /* In OFD, before the pre-creation, the LAST_ID
1036                          * file will be updated firstly, which may hide
1037                          * some potential crashed cases. For example:
1038                          *
1039                          * The old obj1's ID is higher than old LAST_ID
1040                          * but lower than the new LAST_ID, but the LFSCK
1041                          * have not touch the obj1 until the OFD updated
1042                          * the LAST_ID. So the LFSCK does not regard it
1043                          * as crashed case. But when OFD does not create
1044                          * successfully, it will set the LAST_ID as the
1045                          * real created objects' ID, then LFSCK needs to
1046                          * found related inconsistency. */
1047                         rc = lfsck_layout_lastid_reload(env, com, lls);
1048                         if (likely(!lls->lls_dirty))
1049                                 continue;
1050                 }
1051 #endif
1052
1053                 CDEBUG(D_LFSCK, "To sync the LAST_ID for <seq> "LPX64
1054                        " as <oid> "LPU64"\n", lls->lls_seq, lls->lls_lastid);
1055
1056                 if (bk->lb_param & LPF_DRYRUN) {
1057                         lls->lls_dirty = 0;
1058                         continue;
1059                 }
1060
1061                 th = dt_trans_create(env, dt);
1062                 if (IS_ERR(th)) {
1063                         rc1 = PTR_ERR(th);
1064                         CERROR("%s: (1) failed to store "LPX64": rc = %d\n",
1065                                lfsck_lfsck2name(com->lc_lfsck),
1066                                lls->lls_seq, rc1);
1067                         continue;
1068                 }
1069
1070                 lastid = cpu_to_le64(lls->lls_lastid);
1071                 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1072                                              lfsck_buf_get(env, &lastid,
1073                                                            sizeof(lastid)),
1074                                              pos, th);
1075                 if (rc != 0)
1076                         goto stop;
1077
1078                 rc = dt_trans_start_local(env, dt, th);
1079                 if (rc != 0)
1080                         goto stop;
1081
1082                 dt_write_lock(env, lls->lls_lastid_obj, 0);
1083                 rc = dt_record_write(env, lls->lls_lastid_obj,
1084                                      lfsck_buf_get(env, &lastid,
1085                                      sizeof(lastid)), &pos, th);
1086                 dt_write_unlock(env, lls->lls_lastid_obj);
1087                 if (rc == 0)
1088                         lls->lls_dirty = 0;
1089
1090 stop:
1091                 dt_trans_stop(env, dt, th);
1092                 if (rc != 0) {
1093                         rc1 = rc;
1094                         CERROR("%s: (2) failed to store "LPX64": rc = %d\n",
1095                                lfsck_lfsck2name(com->lc_lfsck),
1096                                lls->lls_seq, rc1);
1097                 }
1098         }
1099
1100         return rc1;
1101 }
1102
1103 static int
1104 lfsck_layout_lastid_load(const struct lu_env *env,
1105                          struct lfsck_component *com,
1106                          struct lfsck_layout_seq *lls)
1107 {
1108         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1109         struct lfsck_layout     *lo     = com->lc_file_ram;
1110         struct lu_fid           *fid    = &lfsck_env_info(env)->lti_fid;
1111         struct dt_object        *obj;
1112         loff_t                   pos    = 0;
1113         int                      rc;
1114         ENTRY;
1115
1116         lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck->li_bottom));
1117         obj = dt_locate(env, lfsck->li_bottom, fid);
1118         if (IS_ERR(obj))
1119                 RETURN(PTR_ERR(obj));
1120
1121         /* LAST_ID crashed, to be rebuilt */
1122         if (!dt_object_exists(obj)) {
1123                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1124                         LASSERT(lfsck->li_out_notify != NULL);
1125
1126                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1127                                              LE_LASTID_REBUILDING);
1128                         lo->ll_flags |= LF_CRASHED_LASTID;
1129
1130                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1131                             cfs_fail_val > 0) {
1132                                 struct l_wait_info lwi = LWI_TIMEOUT(
1133                                                 cfs_time_seconds(cfs_fail_val),
1134                                                 NULL, NULL);
1135
1136                                 up_write(&com->lc_sem);
1137                                 l_wait_event(lfsck->li_thread.t_ctl_waitq,
1138                                              !thread_is_running(&lfsck->li_thread),
1139                                              &lwi);
1140                                 down_write(&com->lc_sem);
1141                         }
1142                 }
1143
1144                 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1145         } else {
1146                 dt_read_lock(env, obj, 0);
1147                 rc = dt_read(env, obj,
1148                         lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1149                         &pos);
1150                 dt_read_unlock(env, obj);
1151                 if (rc != 0 && rc != sizeof(__u64))
1152                         GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1153
1154                 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1155                         LASSERT(lfsck->li_out_notify != NULL);
1156
1157                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1158                                              LE_LASTID_REBUILDING);
1159                         lo->ll_flags |= LF_CRASHED_LASTID;
1160                 }
1161
1162                 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1163                 rc = 0;
1164         }
1165
1166         GOTO(out, rc);
1167
1168 out:
1169         if (rc != 0)
1170                 lfsck_object_put(env, obj);
1171         else
1172                 lls->lls_lastid_obj = obj;
1173
1174         return rc;
1175 }
1176
1177 static void lfsck_layout_record_failure(const struct lu_env *env,
1178                                                  struct lfsck_instance *lfsck,
1179                                                  struct lfsck_layout *lo)
1180 {
1181         lo->ll_objs_failed_phase1++;
1182         if (unlikely(lo->ll_pos_first_inconsistent == 0))
1183                 lo->ll_pos_first_inconsistent =
1184                         lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
1185                                                         lfsck->li_di_oit);
1186 }
1187
1188 static int lfsck_layout_master_async_interpret(const struct lu_env *env,
1189                                                struct ptlrpc_request *req,
1190                                                void *args, int rc)
1191 {
1192         struct lfsck_async_interpret_args *laia = args;
1193         struct lfsck_component            *com  = laia->laia_com;
1194         struct lfsck_layout_master_data   *llmd = com->lc_data;
1195         struct lfsck_tgt_descs            *ltds = laia->laia_ltds;
1196         struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
1197         struct lfsck_request              *lr   = laia->laia_lr;
1198
1199         switch (lr->lr_event) {
1200         case LE_START:
1201                 if (rc != 0) {
1202                         struct lfsck_layout *lo = com->lc_file_ram;
1203
1204                         CERROR("%s: fail to notify %s %x for layout start: "
1205                                "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
1206                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1207                                ltd->ltd_index, rc);
1208                         lo->ll_flags |= LF_INCOMPLETE;
1209                         break;
1210                 }
1211
1212                 spin_lock(&ltds->ltd_lock);
1213                 if (ltd->ltd_dead || ltd->ltd_layout_done) {
1214                         spin_unlock(&ltds->ltd_lock);
1215                         break;
1216                 }
1217
1218                 if (lr->lr_flags & LEF_TO_OST) {
1219                         if (list_empty(&ltd->ltd_layout_list))
1220                                 list_add_tail(&ltd->ltd_layout_list,
1221                                               &llmd->llmd_ost_list);
1222                         if (list_empty(&ltd->ltd_layout_phase_list))
1223                                 list_add_tail(&ltd->ltd_layout_phase_list,
1224                                               &llmd->llmd_ost_phase1_list);
1225                 } else {
1226                         if (list_empty(&ltd->ltd_layout_list))
1227                                 list_add_tail(&ltd->ltd_layout_list,
1228                                               &llmd->llmd_mdt_list);
1229                         if (list_empty(&ltd->ltd_layout_phase_list))
1230                                 list_add_tail(&ltd->ltd_layout_phase_list,
1231                                               &llmd->llmd_mdt_phase1_list);
1232                 }
1233                 spin_unlock(&ltds->ltd_lock);
1234                 break;
1235         case LE_STOP:
1236         case LE_PHASE1_DONE:
1237         case LE_PHASE2_DONE:
1238         case LE_PEER_EXIT:
1239                 if (rc != 0 && rc != -EALREADY)
1240                         CWARN("%s: fail to notify %s %x for layout: "
1241                               "event = %d, rc = %d\n",
1242                               lfsck_lfsck2name(com->lc_lfsck),
1243                               (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1244                               ltd->ltd_index, lr->lr_event, rc);
1245                 break;
1246         case LE_QUERY: {
1247                 struct lfsck_reply *reply;
1248
1249                 if (rc != 0) {
1250                         spin_lock(&ltds->ltd_lock);
1251                         list_del_init(&ltd->ltd_layout_phase_list);
1252                         list_del_init(&ltd->ltd_layout_list);
1253                         spin_unlock(&ltds->ltd_lock);
1254                         break;
1255                 }
1256
1257                 reply = req_capsule_server_get(&req->rq_pill,
1258                                                &RMF_LFSCK_REPLY);
1259                 if (reply == NULL) {
1260                         rc = -EPROTO;
1261                         CERROR("%s: invalid return value: rc = %d\n",
1262                                lfsck_lfsck2name(com->lc_lfsck), rc);
1263                         spin_lock(&ltds->ltd_lock);
1264                         list_del_init(&ltd->ltd_layout_phase_list);
1265                         list_del_init(&ltd->ltd_layout_list);
1266                         spin_unlock(&ltds->ltd_lock);
1267                         break;
1268                 }
1269
1270                 switch (reply->lr_status) {
1271                 case LS_SCANNING_PHASE1:
1272                         break;
1273                 case LS_SCANNING_PHASE2:
1274                         spin_lock(&ltds->ltd_lock);
1275                         list_del_init(&ltd->ltd_layout_phase_list);
1276                         if (ltd->ltd_dead || ltd->ltd_layout_done) {
1277                                 spin_unlock(&ltds->ltd_lock);
1278                                 break;
1279                         }
1280
1281                         if (lr->lr_flags & LEF_TO_OST)
1282                                 list_add_tail(&ltd->ltd_layout_phase_list,
1283                                               &llmd->llmd_ost_phase2_list);
1284                         else
1285                                 list_add_tail(&ltd->ltd_layout_phase_list,
1286                                               &llmd->llmd_mdt_phase2_list);
1287                         spin_unlock(&ltds->ltd_lock);
1288                         break;
1289                 default:
1290                         spin_lock(&ltds->ltd_lock);
1291                         list_del_init(&ltd->ltd_layout_phase_list);
1292                         list_del_init(&ltd->ltd_layout_list);
1293                         spin_unlock(&ltds->ltd_lock);
1294                         break;
1295                 }
1296                 break;
1297         }
1298         default:
1299                 CERROR("%s: unexpected event: rc = %d\n",
1300                        lfsck_lfsck2name(com->lc_lfsck), lr->lr_event);
1301                 break;
1302         }
1303
1304         if (!laia->laia_shared) {
1305                 lfsck_tgt_put(ltd);
1306                 lfsck_component_put(env, com);
1307         }
1308
1309         return 0;
1310 }
1311
1312 static int lfsck_layout_master_query_others(const struct lu_env *env,
1313                                             struct lfsck_component *com)
1314 {
1315         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1316         struct lfsck_request              *lr    = &info->lti_lr;
1317         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1318         struct lfsck_instance             *lfsck = com->lc_lfsck;
1319         struct lfsck_layout_master_data   *llmd  = com->lc_data;
1320         struct ptlrpc_request_set         *set;
1321         struct lfsck_tgt_descs            *ltds;
1322         struct lfsck_tgt_desc             *ltd;
1323         struct list_head                  *head;
1324         int                                rc    = 0;
1325         int                                rc1   = 0;
1326         ENTRY;
1327
1328         set = ptlrpc_prep_set();
1329         if (set == NULL)
1330                 RETURN(-ENOMEM);
1331
1332         llmd->llmd_touch_gen++;
1333         memset(lr, 0, sizeof(*lr));
1334         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1335         lr->lr_event = LE_QUERY;
1336         lr->lr_active = LT_LAYOUT;
1337         laia->laia_com = com;
1338         laia->laia_lr = lr;
1339         laia->laia_shared = 0;
1340
1341         if (!list_empty(&llmd->llmd_mdt_phase1_list)) {
1342                 ltds = &lfsck->li_mdt_descs;
1343                 lr->lr_flags = 0;
1344                 head = &llmd->llmd_mdt_phase1_list;
1345         } else {
1346
1347 again:
1348                 ltds = &lfsck->li_ost_descs;
1349                 lr->lr_flags = LEF_TO_OST;
1350                 head = &llmd->llmd_ost_phase1_list;
1351         }
1352
1353         laia->laia_ltds = ltds;
1354         spin_lock(&ltds->ltd_lock);
1355         while (!list_empty(head)) {
1356                 ltd = list_entry(head->next,
1357                                  struct lfsck_tgt_desc,
1358                                  ltd_layout_phase_list);
1359                 if (ltd->ltd_layout_gen == llmd->llmd_touch_gen)
1360                         break;
1361
1362                 ltd->ltd_layout_gen = llmd->llmd_touch_gen;
1363                 list_del(&ltd->ltd_layout_phase_list);
1364                 list_add_tail(&ltd->ltd_layout_phase_list, head);
1365                 atomic_inc(&ltd->ltd_ref);
1366                 laia->laia_ltd = ltd;
1367                 spin_unlock(&ltds->ltd_lock);
1368                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1369                                          lfsck_layout_master_async_interpret,
1370                                          laia, LFSCK_QUERY);
1371                 if (rc != 0) {
1372                         CERROR("%s: fail to query %s %x for layout: rc = %d\n",
1373                                lfsck_lfsck2name(lfsck),
1374                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1375                                ltd->ltd_index, rc);
1376                         lfsck_tgt_put(ltd);
1377                         rc1 = rc;
1378                 }
1379                 spin_lock(&ltds->ltd_lock);
1380         }
1381         spin_unlock(&ltds->ltd_lock);
1382
1383         rc = ptlrpc_set_wait(set);
1384         if (rc < 0) {
1385                 ptlrpc_set_destroy(set);
1386                 RETURN(rc);
1387         }
1388
1389         if (!(lr->lr_flags & LEF_TO_OST) &&
1390             list_empty(&llmd->llmd_mdt_phase1_list))
1391                 goto again;
1392
1393         ptlrpc_set_destroy(set);
1394
1395         RETURN(rc1 != 0 ? rc1 : rc);
1396 }
1397
1398 static inline bool
1399 lfsck_layout_master_to_orphan(struct lfsck_layout_master_data *llmd)
1400 {
1401         return list_empty(&llmd->llmd_mdt_phase1_list) &&
1402                (!list_empty(&llmd->llmd_ost_phase2_list) ||
1403                 list_empty(&llmd->llmd_ost_phase1_list));
1404 }
1405
1406 static int lfsck_layout_master_notify_others(const struct lu_env *env,
1407                                              struct lfsck_component *com,
1408                                              struct lfsck_request *lr)
1409 {
1410         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1411         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1412         struct lfsck_instance             *lfsck = com->lc_lfsck;
1413         struct lfsck_layout_master_data   *llmd  = com->lc_data;
1414         struct lfsck_layout               *lo    = com->lc_file_ram;
1415         struct lfsck_bookmark             *bk    = &lfsck->li_bookmark_ram;
1416         struct ptlrpc_request_set         *set;
1417         struct lfsck_tgt_descs            *ltds;
1418         struct lfsck_tgt_desc             *ltd;
1419         struct lfsck_tgt_desc             *next;
1420         struct list_head                  *head;
1421         __u32                              idx;
1422         int                                rc    = 0;
1423         ENTRY;
1424
1425         set = ptlrpc_prep_set();
1426         if (set == NULL)
1427                 RETURN(-ENOMEM);
1428
1429         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1430         lr->lr_active = LT_LAYOUT;
1431         laia->laia_com = com;
1432         laia->laia_lr = lr;
1433         laia->laia_shared = 0;
1434         switch (lr->lr_event) {
1435         case LE_START:
1436                 /* Notify OSTs firstly, then handle other MDTs if needed. */
1437                 ltds = &lfsck->li_ost_descs;
1438                 laia->laia_ltds = ltds;
1439                 down_read(&ltds->ltd_rw_sem);
1440                 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1441                         ltd = lfsck_tgt_get(ltds, idx);
1442                         LASSERT(ltd != NULL);
1443
1444                         laia->laia_ltd = ltd;
1445                         ltd->ltd_layout_done = 0;
1446                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1447                                         lfsck_layout_master_async_interpret,
1448                                         laia, LFSCK_NOTIFY);
1449                         if (rc != 0) {
1450                                 CERROR("%s: fail to notify %s %x for layout "
1451                                        "start: rc = %d\n",
1452                                        lfsck_lfsck2name(lfsck),
1453                                        (lr->lr_flags & LEF_TO_OST) ? "OST" :
1454                                        "MDT", idx, rc);
1455                                 lfsck_tgt_put(ltd);
1456                                 lo->ll_flags |= LF_INCOMPLETE;
1457                         }
1458                 }
1459                 up_read(&ltds->ltd_rw_sem);
1460
1461                 /* Sync up */
1462                 rc = ptlrpc_set_wait(set);
1463                 if (rc < 0) {
1464                         ptlrpc_set_destroy(set);
1465                         RETURN(rc);
1466                 }
1467
1468                 if (!(bk->lb_param & LPF_ALL_TGT))
1469                         break;
1470
1471                 /* link other MDT targets locallly. */
1472                 spin_lock(&ltds->ltd_lock);
1473                 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1474                         ltd = LTD_TGT(ltds, idx);
1475                         LASSERT(ltd != NULL);
1476
1477                         if (!list_empty(&ltd->ltd_layout_list))
1478                                 continue;
1479
1480                         list_add_tail(&ltd->ltd_layout_list,
1481                                       &llmd->llmd_mdt_list);
1482                         list_add_tail(&ltd->ltd_layout_phase_list,
1483                                       &llmd->llmd_mdt_phase1_list);
1484                 }
1485                 spin_unlock(&ltds->ltd_lock);
1486                 break;
1487         case LE_STOP:
1488         case LE_PHASE2_DONE:
1489         case LE_PEER_EXIT: {
1490                 /* Handle other MDTs firstly if needed, then notify the OSTs. */
1491                 if (bk->lb_param & LPF_ALL_TGT) {
1492                         head = &llmd->llmd_mdt_list;
1493                         ltds = &lfsck->li_mdt_descs;
1494                         if (lr->lr_event == LE_STOP) {
1495                                 /* unlink other MDT targets locallly. */
1496                                 spin_lock(&ltds->ltd_lock);
1497                                 list_for_each_entry_safe(ltd, next, head,
1498                                                          ltd_layout_list) {
1499                                         list_del_init(&ltd->ltd_layout_phase_list);
1500                                         list_del_init(&ltd->ltd_layout_list);
1501                                 }
1502                                 spin_unlock(&ltds->ltd_lock);
1503
1504                                 lr->lr_flags |= LEF_TO_OST;
1505                                 head = &llmd->llmd_ost_list;
1506                                 ltds = &lfsck->li_ost_descs;
1507                         } else {
1508                                 lr->lr_flags &= ~LEF_TO_OST;
1509                         }
1510                 } else {
1511                         lr->lr_flags |= LEF_TO_OST;
1512                         head = &llmd->llmd_ost_list;
1513                         ltds = &lfsck->li_ost_descs;
1514                 }
1515
1516 again:
1517                 laia->laia_ltds = ltds;
1518                 spin_lock(&ltds->ltd_lock);
1519                 while (!list_empty(head)) {
1520                         ltd = list_entry(head->next, struct lfsck_tgt_desc,
1521                                          ltd_layout_list);
1522                         if (!list_empty(&ltd->ltd_layout_phase_list))
1523                                 list_del_init(&ltd->ltd_layout_phase_list);
1524                         list_del_init(&ltd->ltd_layout_list);
1525                         atomic_inc(&ltd->ltd_ref);
1526                         laia->laia_ltd = ltd;
1527                         spin_unlock(&ltds->ltd_lock);
1528                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1529                                         lfsck_layout_master_async_interpret,
1530                                         laia, LFSCK_NOTIFY);
1531                         if (rc != 0) {
1532                                 CERROR("%s: fail to notify %s %x for layout "
1533                                        "stop/phase2: rc = %d\n",
1534                                        lfsck_lfsck2name(lfsck),
1535                                        (lr->lr_flags & LEF_TO_OST) ? "OST" :
1536                                        "MDT", ltd->ltd_index, rc);
1537                                 lfsck_tgt_put(ltd);
1538                         }
1539                         spin_lock(&ltds->ltd_lock);
1540                 }
1541                 spin_unlock(&ltds->ltd_lock);
1542
1543                 rc = ptlrpc_set_wait(set);
1544                 if (rc < 0) {
1545                         ptlrpc_set_destroy(set);
1546                         RETURN(rc);
1547                 }
1548
1549                 if (!(lr->lr_flags & LEF_TO_OST)) {
1550                         lr->lr_flags |= LEF_TO_OST;
1551                         head = &llmd->llmd_ost_list;
1552                         ltds = &lfsck->li_ost_descs;
1553                         goto again;
1554                 }
1555                 break;
1556         }
1557         case LE_PHASE1_DONE:
1558                 llmd->llmd_touch_gen++;
1559                 ltds = &lfsck->li_mdt_descs;
1560                 laia->laia_ltds = ltds;
1561                 spin_lock(&ltds->ltd_lock);
1562                 while (!list_empty(&llmd->llmd_mdt_phase1_list)) {
1563                         ltd = list_entry(llmd->llmd_mdt_phase1_list.next,
1564                                          struct lfsck_tgt_desc,
1565                                          ltd_layout_phase_list);
1566                         if (ltd->ltd_layout_gen == llmd->llmd_touch_gen)
1567                                 break;
1568
1569                         ltd->ltd_layout_gen = llmd->llmd_touch_gen;
1570                         list_del_init(&ltd->ltd_layout_phase_list);
1571                         list_add_tail(&ltd->ltd_layout_phase_list,
1572                                       &llmd->llmd_mdt_phase1_list);
1573                         atomic_inc(&ltd->ltd_ref);
1574                         laia->laia_ltd = ltd;
1575                         spin_unlock(&ltds->ltd_lock);
1576                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1577                                         lfsck_layout_master_async_interpret,
1578                                         laia, LFSCK_NOTIFY);
1579                         if (rc != 0) {
1580                                 CERROR("%s: fail to notify MDT %x for layout "
1581                                        "phase1 done: rc = %d\n",
1582                                        lfsck_lfsck2name(lfsck),
1583                                        ltd->ltd_index, rc);
1584                                 lfsck_tgt_put(ltd);
1585                         }
1586                         spin_lock(&ltds->ltd_lock);
1587                 }
1588                 spin_unlock(&ltds->ltd_lock);
1589                 break;
1590         default:
1591                 CERROR("%s: unexpected LFSCK event: rc = %d\n",
1592                        lfsck_lfsck2name(lfsck), lr->lr_event);
1593                 rc = -EINVAL;
1594                 break;
1595         }
1596
1597         rc = ptlrpc_set_wait(set);
1598         ptlrpc_set_destroy(set);
1599
1600         RETURN(rc);
1601 }
1602
1603 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1604                                            struct lfsck_component *com,
1605                                            int rc)
1606 {
1607         struct lfsck_instance   *lfsck = com->lc_lfsck;
1608         struct lfsck_layout     *lo    = com->lc_file_ram;
1609         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
1610
1611         down_write(&com->lc_sem);
1612         lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1613                                 HALF_SEC - lfsck->li_time_last_checkpoint);
1614         lo->ll_time_last_checkpoint = cfs_time_current_sec();
1615         lo->ll_objs_checked_phase2 += com->lc_new_checked;
1616
1617         if (rc > 0) {
1618                 com->lc_journal = 0;
1619                 if (lo->ll_flags & LF_INCOMPLETE)
1620                         lo->ll_status = LS_PARTIAL;
1621                 else
1622                         lo->ll_status = LS_COMPLETED;
1623                 if (!(bk->lb_param & LPF_DRYRUN))
1624                         lo->ll_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
1625                 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1626                 lo->ll_success_count++;
1627         } else if (rc == 0) {
1628                 lo->ll_status = lfsck->li_status;
1629                 if (lo->ll_status == 0)
1630                         lo->ll_status = LS_STOPPED;
1631         } else {
1632                 lo->ll_status = LS_FAILED;
1633         }
1634
1635         rc = lfsck_layout_store(env, com);
1636         up_write(&com->lc_sem);
1637
1638         return rc;
1639 }
1640
1641 static int lfsck_layout_lock(const struct lu_env *env,
1642                              struct lfsck_component *com,
1643                              struct dt_object *obj,
1644                              struct lustre_handle *lh, __u64 bits)
1645 {
1646         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1647         ldlm_policy_data_t              *policy = &info->lti_policy;
1648         struct ldlm_res_id              *resid  = &info->lti_resid;
1649         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1650         __u64                            flags  = LDLM_FL_ATOMIC_CB;
1651         int                              rc;
1652
1653         LASSERT(lfsck->li_namespace != NULL);
1654
1655         memset(policy, 0, sizeof(*policy));
1656         policy->l_inodebits.bits = bits;
1657         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
1658         rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_IBITS,
1659                                     policy, LCK_EX, &flags, ldlm_blocking_ast,
1660                                     ldlm_completion_ast, NULL, NULL, 0,
1661                                     LVB_T_NONE, NULL, lh);
1662         if (rc == ELDLM_OK) {
1663                 rc = 0;
1664         } else {
1665                 memset(lh, 0, sizeof(*lh));
1666                 rc = -EIO;
1667         }
1668
1669         return rc;
1670 }
1671
1672 static void lfsck_layout_unlock(struct lustre_handle *lh)
1673 {
1674         if (lustre_handle_is_used(lh)) {
1675                 ldlm_lock_decref(lh, LCK_EX);
1676                 memset(lh, 0, sizeof(*lh));
1677         }
1678 }
1679
1680 static int lfsck_layout_trans_stop(const struct lu_env *env,
1681                                    struct dt_device *dev,
1682                                    struct thandle *handle, int result)
1683 {
1684         int rc;
1685
1686         handle->th_result = result;
1687         rc = dt_trans_stop(env, dev, handle);
1688         if (rc > 0)
1689                 rc = 0;
1690         else if (rc == 0)
1691                 rc = 1;
1692
1693         return rc;
1694 }
1695
1696 /**
1697  * \retval       +1: repaired
1698  * \retval        0: did nothing
1699  * \retval      -ve: on error
1700  */
1701 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1702                                      struct thandle *handle,
1703                                      struct dt_object *parent,
1704                                      struct lu_fid *cfid,
1705                                      struct lu_buf *buf,
1706                                      struct lov_ost_data_v1 *slot,
1707                                      int fl, __u32 ost_idx)
1708 {
1709         struct ost_id   *oi     = &lfsck_env_info(env)->lti_oi;
1710         int              rc;
1711
1712         fid_to_ostid(cfid, oi);
1713         ostid_cpu_to_le(oi, &slot->l_ost_oi);
1714         slot->l_ost_gen = cpu_to_le32(0);
1715         slot->l_ost_idx = cpu_to_le32(ost_idx);
1716         rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV, fl, handle,
1717                           BYPASS_CAPA);
1718         if (rc == 0)
1719                 rc = 1;
1720
1721         return rc;
1722 }
1723
1724 /**
1725  * \retval       +1: repaired
1726  * \retval        0: did nothing
1727  * \retval      -ve: on error
1728  */
1729 static int lfsck_layout_extend_lovea(const struct lu_env *env,
1730                                      struct thandle *handle,
1731                                      struct dt_object *parent,
1732                                      struct lu_fid *cfid,
1733                                      struct lu_buf *buf, int fl,
1734                                      __u32 ost_idx, __u32 ea_off, bool reset)
1735 {
1736         struct lov_mds_md_v1    *lmm    = buf->lb_buf;
1737         struct lov_ost_data_v1  *objs;
1738         int                      rc;
1739         ENTRY;
1740
1741         if (fl == LU_XATTR_CREATE || reset) {
1742                 LASSERT(buf->lb_len == lov_mds_md_size(ea_off + 1,
1743                                                        LOV_MAGIC_V1));
1744
1745                 memset(lmm, 0, buf->lb_len);
1746                 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1747                 /* XXX: currently, we only support LOV_PATTERN_RAID0. */
1748                 lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1749                 fid_to_lmm_oi(lfsck_dto2fid(parent), &lmm->lmm_oi);
1750                 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1751                 /* XXX: We cannot know the stripe size,
1752                  *      then use the default value (1 MB). */
1753                 lmm->lmm_stripe_size =
1754                         cpu_to_le32(LOV_DESC_STRIPE_SIZE_DEFAULT);
1755                 objs = &(lmm->lmm_objects[ea_off]);
1756         } else {
1757                 __u16   count = le16_to_cpu(lmm->lmm_stripe_count);
1758                 int     gap   = ea_off - count;
1759                 __u32   magic = le32_to_cpu(lmm->lmm_magic);
1760
1761                 /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3
1762                  * which has been verified in lfsck_layout_verify_header()
1763                  * already. If some new magic introduced in the future,
1764                  * then layout LFSCK needs to be updated also. */
1765                 if (magic == LOV_MAGIC_V1) {
1766                         objs = &(lmm->lmm_objects[count]);
1767                 } else {
1768                         LASSERT(magic == LOV_MAGIC_V3);
1769                         objs = &((struct lov_mds_md_v3 *)lmm)->
1770                                                         lmm_objects[count];
1771                 }
1772
1773                 if (gap > 0)
1774                         memset(objs, 0, gap * sizeof(*objs));
1775                 lmm->lmm_layout_gen =
1776                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
1777                 objs += gap;
1778
1779                 LASSERT(buf->lb_len == lov_mds_md_size(ea_off + 1, magic));
1780         }
1781
1782         lmm->lmm_stripe_count = cpu_to_le16(ea_off + 1);
1783         rc = lfsck_layout_refill_lovea(env, handle, parent, cfid, buf, objs,
1784                                        fl, ost_idx);
1785
1786         RETURN(rc);
1787 }
1788
1789 /**
1790  * \retval       +1: repaired
1791  * \retval        0: did nothing
1792  * \retval      -ve: on error
1793  */
1794 static int lfsck_layout_update_pfid(const struct lu_env *env,
1795                                     struct lfsck_component *com,
1796                                     struct dt_object *parent,
1797                                     struct lu_fid *cfid,
1798                                     struct dt_device *cdev, __u32 ea_off)
1799 {
1800         struct filter_fid       *pfid   = &lfsck_env_info(env)->lti_new_pfid;
1801         struct dt_object        *child;
1802         struct thandle          *handle;
1803         const struct lu_fid     *tfid   = lu_object_fid(&parent->do_lu);
1804         struct lu_buf           *buf;
1805         int                      rc     = 0;
1806         ENTRY;
1807
1808         child = lfsck_object_find_by_dev(env, cdev, cfid);
1809         if (IS_ERR(child))
1810                 RETURN(PTR_ERR(child));
1811
1812         handle = dt_trans_create(env, cdev);
1813         if (IS_ERR(handle))
1814                 GOTO(out, rc = PTR_ERR(handle));
1815
1816         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
1817         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
1818         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
1819          * MDT-object's FID::f_ver, instead it is the OST-object index in its
1820          * parent MDT-object's layout EA. */
1821         pfid->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
1822         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
1823
1824         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
1825         if (rc != 0)
1826                 GOTO(stop, rc);
1827
1828         rc = dt_trans_start(env, cdev, handle);
1829         if (rc != 0)
1830                 GOTO(stop, rc);
1831
1832         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle,
1833                           BYPASS_CAPA);
1834
1835         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1836
1837 stop:
1838         dt_trans_stop(env, cdev, handle);
1839
1840 out:
1841         lu_object_put(env, &child->do_lu);
1842
1843         return rc;
1844 }
1845
1846 /**
1847  * \retval       +1: repaired
1848  * \retval        0: did nothing
1849  * \retval      -ve: on error
1850  */
1851 static int lfsck_layout_recreate_parent(const struct lu_env *env,
1852                                         struct lfsck_component *com,
1853                                         struct lfsck_tgt_desc *ltd,
1854                                         struct lu_orphan_rec *rec,
1855                                         struct lu_fid *cfid,
1856                                         const char *prefix,
1857                                         const char *postfix,
1858                                         __u32 ea_off)
1859 {
1860         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1861         char                            *name   = info->lti_key;
1862         struct lu_attr                  *la     = &info->lti_la;
1863         struct dt_object_format         *dof    = &info->lti_dof;
1864         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1865         struct lu_fid                   *pfid   = &rec->lor_fid;
1866         struct lu_fid                   *tfid   = &info->lti_fid3;
1867         struct dt_device                *next   = lfsck->li_next;
1868         struct dt_object                *pobj   = NULL;
1869         struct dt_object                *cobj   = NULL;
1870         struct thandle                  *th     = NULL;
1871         struct lu_buf                   *pbuf   = NULL;
1872         struct lu_buf                   *ea_buf = &info->lti_big_buf;
1873         struct lustre_handle             lh     = { 0 };
1874         int                              buflen = ea_buf->lb_len;
1875         int                              idx    = 0;
1876         int                              rc     = 0;
1877         ENTRY;
1878
1879         /* Create .lustre/lost+found/MDTxxxx when needed. */
1880         if (unlikely(lfsck->li_lpf_obj == NULL)) {
1881                 rc = lfsck_create_lpf(env, lfsck);
1882                 if (rc != 0)
1883                         RETURN(rc);
1884         }
1885
1886         if (fid_is_zero(pfid)) {
1887                 struct filter_fid *ff = &info->lti_new_pfid;
1888
1889                 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
1890                 if (rc != 0)
1891                         RETURN(rc);
1892
1893                 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
1894                 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
1895                 /* Currently, the filter_fid::ff_parent::f_ver is not the
1896                  * real parent MDT-object's FID::f_ver, instead it is the
1897                  * OST-object index in its parent MDT-object's layout EA. */
1898                 ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
1899                 pbuf = lfsck_buf_get(env, ff, sizeof(struct filter_fid));
1900                 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
1901                 if (IS_ERR(cobj))
1902                         RETURN(PTR_ERR(cobj));
1903         }
1904
1905         CDEBUG(D_LFSCK, "Re-create the lost MDT-object: parent "
1906                DFID", child "DFID", OST-index %u, stripe-index %u, "
1907                "prefix %s, postfix %s\n",
1908                PFID(pfid), PFID(cfid), ltd->ltd_index, ea_off, prefix, postfix);
1909
1910         pobj = lfsck_object_find_by_dev(env, lfsck->li_bottom, pfid);
1911         if (IS_ERR(pobj))
1912                 GOTO(put, rc = PTR_ERR(pobj));
1913
1914         LASSERT(prefix != NULL);
1915         LASSERT(postfix != NULL);
1916
1917         /** name rules:
1918          *
1919          *  1. Use the MDT-object's FID as the name with prefix and postfix.
1920          *
1921          *  1.1 prefix "C-":    More than one OST-objects claim the same
1922          *                      MDT-object and the same slot in the layout EA.
1923          *                      It may be created for dangling referenced MDT
1924          *                      object or may be not.
1925          *  1.2 prefix "N-":    The orphan OST-object does not know which one
1926          *                      is the real parent, so the LFSCK assign a new
1927          *                      FID as its parent.
1928          *  1.3 prefix "R-":    The orphan OST-object know its parent FID but
1929          *                      does not know the position in the namespace.
1930          *
1931          *  2. If there is name conflict, append more index for new name. */
1932         sprintf(name, "%s"DFID"%s", prefix, PFID(pfid), postfix);
1933         do {
1934                 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1935                                (const struct dt_key *)name, BYPASS_CAPA);
1936                 if (rc != 0 && rc != -ENOENT)
1937                         GOTO(put, rc);
1938
1939                 if (unlikely(rc == 0)) {
1940                         CWARN("%s: The name %s under lost+found has been used "
1941                               "by the "DFID". Try to increase the FID version "
1942                               "for the new file name.\n",
1943                               lfsck_lfsck2name(lfsck), name, PFID(tfid));
1944                         sprintf(name, "%s"DFID"%s-%d", prefix, PFID(pfid),
1945                                 postfix, ++idx);
1946                 }
1947         } while (rc == 0);
1948
1949         memset(la, 0, sizeof(*la));
1950         la->la_uid = rec->lor_uid;
1951         la->la_gid = rec->lor_gid;
1952         la->la_mode = S_IFREG | S_IRUSR | S_IWUSR;
1953         la->la_valid = LA_MODE | LA_UID | LA_GID;
1954
1955         memset(dof, 0, sizeof(*dof));
1956         dof->dof_type = dt_mode_to_dft(S_IFREG);
1957
1958         rc = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
1959         if (buflen < rc) {
1960                 lu_buf_realloc(ea_buf, rc);
1961                 buflen = ea_buf->lb_len;
1962                 if (ea_buf->lb_buf == NULL)
1963                         GOTO(put, rc = -ENOMEM);
1964         } else {
1965                 ea_buf->lb_len = rc;
1966         }
1967
1968         /* Hold update lock on the .lustre/lost+found/MDTxxxx/.
1969          *
1970          * XXX: Currently, we do not grab the PDO lock as normal create cases,
1971          *      because creating MDT-object for orphan OST-object is rare, we
1972          *      do not much care about the performance. It can be improved in
1973          *      the future when needed. */
1974         rc = lfsck_layout_lock(env, com, lfsck->li_lpf_obj, &lh,
1975                                MDS_INODELOCK_UPDATE);
1976         if (rc != 0)
1977                 GOTO(put, rc);
1978
1979         th = dt_trans_create(env, next);
1980         if (IS_ERR(th))
1981                 GOTO(unlock, rc = PTR_ERR(th));
1982
1983         /* 1a. Update OST-object's parent information remotely.
1984          *
1985          * If other subsequent modifications failed, then next LFSCK scanning
1986          * will process the OST-object as orphan again with known parent FID. */
1987         if (cobj != NULL) {
1988                 rc = dt_declare_xattr_set(env, cobj, pbuf, XATTR_NAME_FID, 0, th);
1989                 if (rc != 0)
1990                         GOTO(stop, rc);
1991         }
1992
1993         /* 2a. Create the MDT-object locally. */
1994         rc = dt_declare_create(env, pobj, la, NULL, dof, th);
1995         if (rc != 0)
1996                 GOTO(stop, rc);
1997
1998         /* 3a. Add layout EA for the MDT-object. */
1999         rc = dt_declare_xattr_set(env, pobj, ea_buf, XATTR_NAME_LOV,
2000                                   LU_XATTR_CREATE, th);
2001         if (rc != 0)
2002                 GOTO(stop, rc);
2003
2004         /* 4a. Insert the MDT-object to .lustre/lost+found/MDTxxxx/ */
2005         rc = dt_declare_insert(env, lfsck->li_lpf_obj,
2006                                (const struct dt_rec *)pfid,
2007                                (const struct dt_key *)name, th);
2008         if (rc != 0)
2009                 GOTO(stop, rc);
2010
2011         rc = dt_trans_start(env, next, th);
2012         if (rc != 0)
2013                 GOTO(stop, rc);
2014
2015         /* 1b. Update OST-object's parent information remotely. */
2016         if (cobj != NULL) {
2017                 rc = dt_xattr_set(env, cobj, pbuf, XATTR_NAME_FID, 0, th,
2018                                   BYPASS_CAPA);
2019                 if (rc != 0)
2020                         GOTO(stop, rc);
2021         }
2022
2023         dt_write_lock(env, pobj, 0);
2024         /* 2b. Create the MDT-object locally. */
2025         rc = dt_create(env, pobj, la, NULL, dof, th);
2026         if (rc == 0)
2027                 /* 3b. Add layout EA for the MDT-object. */
2028                 rc = lfsck_layout_extend_lovea(env, th, pobj, cfid, ea_buf,
2029                                                LU_XATTR_CREATE, ltd->ltd_index,
2030                                                ea_off, false);
2031         dt_write_unlock(env, pobj);
2032         if (rc < 0)
2033                 GOTO(stop, rc);
2034
2035         /* 4b. Insert the MDT-object to .lustre/lost+found/MDTxxxx/ */
2036         rc = dt_insert(env, lfsck->li_lpf_obj,
2037                        (const struct dt_rec *)pfid,
2038                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
2039
2040         GOTO(stop, rc);
2041
2042 stop:
2043         dt_trans_stop(env, next, th);
2044
2045 unlock:
2046         lfsck_layout_unlock(&lh);
2047
2048 put:
2049         if (cobj != NULL && !IS_ERR(cobj))
2050                 lu_object_put(env, &cobj->do_lu);
2051         if (pobj != NULL && !IS_ERR(pobj))
2052                 lu_object_put(env, &pobj->do_lu);
2053         ea_buf->lb_len = buflen;
2054
2055         return rc >= 0 ? 1 : rc;
2056 }
2057
2058 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
2059                                                    struct lfsck_component *com,
2060                                                    const struct lu_fid *fid,
2061                                                    __u32 index)
2062 {
2063         struct lfsck_thread_info *info  = lfsck_env_info(env);
2064         struct lfsck_request     *lr    = &info->lti_lr;
2065         struct lfsck_instance    *lfsck = com->lc_lfsck;
2066         struct lfsck_tgt_desc    *ltd;
2067         struct ptlrpc_request    *req;
2068         struct lfsck_request     *tmp;
2069         struct obd_export        *exp;
2070         int                       rc    = 0;
2071         ENTRY;
2072
2073         ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
2074         if (unlikely(ltd == NULL))
2075                 RETURN(-ENXIO);
2076
2077         exp = ltd->ltd_exp;
2078         if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
2079                 GOTO(put, rc = -EOPNOTSUPP);
2080
2081         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
2082         if (req == NULL)
2083                 GOTO(put, rc = -ENOMEM);
2084
2085         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
2086         if (rc != 0) {
2087                 ptlrpc_request_free(req);
2088
2089                 GOTO(put, rc);
2090         }
2091
2092         memset(lr, 0, sizeof(*lr));
2093         lr->lr_event = LE_CONDITIONAL_DESTROY;
2094         lr->lr_active = LT_LAYOUT;
2095         lr->lr_fid = *fid;
2096
2097         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2098         *tmp = *lr;
2099         ptlrpc_request_set_replen(req);
2100
2101         rc = ptlrpc_queue_wait(req);
2102         ptlrpc_req_finished(req);
2103
2104         GOTO(put, rc);
2105
2106 put:
2107         lfsck_tgt_put(ltd);
2108
2109         return rc;
2110 }
2111
2112 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
2113                                                   struct lfsck_component *com,
2114                                                   struct lfsck_request *lr)
2115 {
2116         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2117         struct lu_attr                  *la     = &info->lti_la;
2118         ldlm_policy_data_t              *policy = &info->lti_policy;
2119         struct ldlm_res_id              *resid  = &info->lti_resid;
2120         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2121         struct dt_device                *dev    = lfsck->li_bottom;
2122         struct lu_fid                   *fid    = &lr->lr_fid;
2123         struct dt_object                *obj;
2124         struct thandle                  *th     = NULL;
2125         struct lustre_handle             lh     = { 0 };
2126         __u64                            flags  = 0;
2127         int                              rc     = 0;
2128         ENTRY;
2129
2130         obj = lfsck_object_find_by_dev(env, dev, fid);
2131         if (IS_ERR(obj))
2132                 RETURN(PTR_ERR(obj));
2133
2134         dt_read_lock(env, obj, 0);
2135         if (dt_object_exists(obj) == 0) {
2136                 dt_read_unlock(env, obj);
2137
2138                 GOTO(put, rc = -ENOENT);
2139         }
2140
2141         /* Get obj's attr without lock firstly. */
2142         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
2143         dt_read_unlock(env, obj);
2144         if (rc != 0)
2145                 GOTO(put, rc);
2146
2147         if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
2148                 GOTO(put, rc = -ETXTBSY);
2149
2150         /* Acquire extent lock on [0, EOF] to sync with all possible written. */
2151         LASSERT(lfsck->li_namespace != NULL);
2152
2153         memset(policy, 0, sizeof(*policy));
2154         policy->l_extent.end = OBD_OBJECT_EOF;
2155         ost_fid_build_resid(fid, resid);
2156         rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
2157                                     policy, LCK_EX, &flags, ldlm_blocking_ast,
2158                                     ldlm_completion_ast, NULL, NULL, 0,
2159                                     LVB_T_NONE, NULL, &lh);
2160         if (rc != ELDLM_OK)
2161                 GOTO(put, rc = -EIO);
2162
2163         dt_write_lock(env, obj, 0);
2164         /* Get obj's attr within lock again. */
2165         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
2166         if (rc != 0)
2167                 GOTO(unlock, rc);
2168
2169         if (la->la_ctime != 0)
2170                 GOTO(unlock, rc = -ETXTBSY);
2171
2172         th = dt_trans_create(env, dev);
2173         if (IS_ERR(th))
2174                 GOTO(unlock, rc = PTR_ERR(th));
2175
2176         rc = dt_declare_ref_del(env, obj, th);
2177         if (rc != 0)
2178                 GOTO(stop, rc);
2179
2180         rc = dt_declare_destroy(env, obj, th);
2181         if (rc != 0)
2182                 GOTO(stop, rc);
2183
2184         rc = dt_trans_start_local(env, dev, th);
2185         if (rc != 0)
2186                 GOTO(stop, rc);
2187
2188         rc = dt_ref_del(env, obj, th);
2189         if (rc != 0)
2190                 GOTO(stop, rc);
2191
2192         rc = dt_destroy(env, obj, th);
2193         if (rc == 0)
2194                 CDEBUG(D_LFSCK, "Destroy the empty OST-object "DFID" which "
2195                        "was created for reparing dangling referenced case. "
2196                        "But the original missed OST-object is found now.\n",
2197                        PFID(fid));
2198
2199         GOTO(stop, rc);
2200
2201 stop:
2202         dt_trans_stop(env, dev, th);
2203
2204 unlock:
2205         dt_write_unlock(env, obj);
2206         ldlm_lock_decref(&lh, LCK_EX);
2207
2208 put:
2209         lu_object_put(env, &obj->do_lu);
2210
2211         return rc;
2212 }
2213
2214 /**
2215  * Some OST-object has occupied the specified layout EA slot.
2216  * Such OST-object may be generated by the LFSCK when repair
2217  * dangling referenced MDT-object, which can be indicated by
2218  * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2219  * is true and such OST-object has not been modified yet, we
2220  * will replace it with the orphan OST-object; otherwise the
2221  * LFSCK will create new MDT-object to reference the orphan.
2222  *
2223  * \retval       +1: repaired
2224  * \retval        0: did nothing
2225  * \retval      -ve: on error
2226  */
2227 static int lfsck_layout_conflict_create(const struct lu_env *env,
2228                                         struct lfsck_component *com,
2229                                         struct lfsck_tgt_desc *ltd,
2230                                         struct lu_orphan_rec *rec,
2231                                         struct dt_object *parent,
2232                                         struct lu_fid *cfid,
2233                                         struct lu_buf *ea_buf,
2234                                         struct lov_ost_data_v1 *slot,
2235                                         __u32 ea_off, __u32 ori_len)
2236 {
2237         struct lfsck_thread_info *info          = lfsck_env_info(env);
2238         struct lu_fid            *cfid2         = &info->lti_fid2;
2239         struct ost_id            *oi            = &info->lti_oi;
2240         char                     *postfix       = info->lti_tmpbuf;
2241         struct lov_mds_md_v1     *lmm           = ea_buf->lb_buf;
2242         struct dt_device         *dev           = com->lc_lfsck->li_bottom;
2243         struct thandle           *th            = NULL;
2244         struct lustre_handle      lh            = { 0 };
2245         __u32                     ost_idx2      = le32_to_cpu(slot->l_ost_idx);
2246         int                       rc            = 0;
2247         ENTRY;
2248
2249         ostid_le_to_cpu(&slot->l_ost_oi, oi);
2250         ostid_to_fid(cfid2, oi, ost_idx2);
2251
2252         CDEBUG(D_LFSCK, "Handle layout EA conflict: parent "DFID
2253                ", cur-child "DFID" on the OST %u, orphan-child "
2254                DFID" on the OST %u, stripe-index %u\n",
2255                PFID(lfsck_dto2fid(parent)), PFID(cfid2), ost_idx2,
2256                PFID(cfid), ltd->ltd_index, ea_off);
2257
2258         /* Hold layout lock on the parent to prevent others to access. */
2259         rc = lfsck_layout_lock(env, com, parent, &lh,
2260                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
2261         if (rc != 0)
2262                 GOTO(out, rc);
2263
2264         rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2265
2266         /* If the conflict OST-obejct is not created for fixing dangling
2267          * referenced MDT-object in former LFSCK check/repair, or it has
2268          * been modified by others, then we cannot destroy it. Re-create
2269          * a new MDT-object for the orphan OST-object. */
2270         if (rc == -ETXTBSY) {
2271                 /* No need the layout lock on the original parent. */
2272                 lfsck_layout_unlock(&lh);
2273                 ea_buf->lb_len = ori_len;
2274
2275                 fid_zero(&rec->lor_fid);
2276                 snprintf(postfix, LFSCK_TMPBUF_LEN, "-"DFID"-%x",
2277                          PFID(lu_object_fid(&parent->do_lu)), ea_off);
2278                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2279                                                   "C-", postfix, ea_off);
2280
2281                 RETURN(rc);
2282         }
2283
2284         if (rc != 0 && rc != -ENOENT)
2285                 GOTO(unlock, rc);
2286
2287         th = dt_trans_create(env, dev);
2288         if (IS_ERR(th))
2289                 GOTO(unlock, rc = PTR_ERR(th));
2290
2291         rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2292                                   LU_XATTR_REPLACE, th);
2293         if (rc != 0)
2294                 GOTO(stop, rc);
2295
2296         rc = dt_trans_start_local(env, dev, th);
2297         if (rc != 0)
2298                 GOTO(stop, rc);
2299
2300         dt_write_lock(env, parent, 0);
2301         lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2302         rc = lfsck_layout_refill_lovea(env, th, parent, cfid, ea_buf, slot,
2303                                        LU_XATTR_REPLACE, ltd->ltd_index);
2304         dt_write_unlock(env, parent);
2305
2306         GOTO(stop, rc);
2307
2308 stop:
2309         dt_trans_stop(env, dev, th);
2310
2311 unlock:
2312         lfsck_layout_unlock(&lh);
2313
2314 out:
2315         ea_buf->lb_len = ori_len;
2316
2317         return rc >= 0 ? 1 : rc;
2318 }
2319
2320 /**
2321  * \retval       +1: repaired
2322  * \retval        0: did nothing
2323  * \retval      -ve: on error
2324  */
2325 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2326                                        struct lfsck_component *com,
2327                                        struct lfsck_tgt_desc *ltd,
2328                                        struct lu_orphan_rec *rec,
2329                                        struct dt_object *parent,
2330                                        struct lu_fid *cfid,
2331                                        __u32 ost_idx, __u32 ea_off)
2332 {
2333         struct lfsck_thread_info *info          = lfsck_env_info(env);
2334         struct lu_buf            *buf           = &info->lti_big_buf;
2335         struct lu_fid            *fid           = &info->lti_fid2;
2336         struct ost_id            *oi            = &info->lti_oi;
2337         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2338         struct dt_device         *dt            = lfsck->li_bottom;
2339         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2340         struct thandle            *handle       = NULL;
2341         size_t                    buflen        = buf->lb_len;
2342         struct lov_mds_md_v1     *lmm;
2343         struct lov_ost_data_v1   *objs;
2344         struct lustre_handle      lh            = { 0 };
2345         __u32                     magic;
2346         int                       fl            = 0;
2347         int                       rc            = 0;
2348         int                       rc1;
2349         int                       i;
2350         __u16                     count;
2351         bool                      locked        = false;
2352         ENTRY;
2353
2354         CDEBUG(D_LFSCK, "Re-create the crashed layout EA: parent "
2355                DFID", child "DFID", OST-index %u, stripe-index %u\n",
2356                PFID(lfsck_dto2fid(parent)), PFID(cfid), ost_idx, ea_off);
2357
2358         rc = lfsck_layout_lock(env, com, parent, &lh,
2359                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
2360         if (rc != 0)
2361                 RETURN(rc);
2362
2363 again:
2364         if (locked) {
2365                 dt_write_unlock(env, parent);
2366                 locked = false;
2367         }
2368
2369         if (handle != NULL) {
2370                 dt_trans_stop(env, dt, handle);
2371                 handle = NULL;
2372         }
2373
2374         if (rc < 0)
2375                 GOTO(unlock_layout, rc);
2376
2377         if (buf->lb_len < rc) {
2378                 lu_buf_realloc(buf, rc);
2379                 buflen = buf->lb_len;
2380                 if (buf->lb_buf == NULL)
2381                         GOTO(unlock_layout, rc = -ENOMEM);
2382         }
2383
2384         if (!(bk->lb_param & LPF_DRYRUN)) {
2385                 handle = dt_trans_create(env, dt);
2386                 if (IS_ERR(handle))
2387                         GOTO(unlock_layout, rc = PTR_ERR(handle));
2388
2389                 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2390                                           fl, handle);
2391                 if (rc != 0)
2392                         GOTO(stop, rc);
2393
2394                 rc = dt_trans_start_local(env, dt, handle);
2395                 if (rc != 0)
2396                         GOTO(stop, rc);
2397         }
2398
2399         dt_write_lock(env, parent, 0);
2400         locked = true;
2401         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV, BYPASS_CAPA);
2402         if (rc == -ERANGE) {
2403                 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV,
2404                                   BYPASS_CAPA);
2405                 LASSERT(rc != 0);
2406                 goto again;
2407         } else if (rc == -ENODATA || rc == 0) {
2408                 rc = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2409                 /* If the declared is not big enough, re-try. */
2410                 if (buf->lb_len < rc)
2411                         goto again;
2412
2413                 fl = LU_XATTR_CREATE;
2414         } else if (rc < 0) {
2415                 GOTO(unlock_parent, rc);
2416         } else if (unlikely(buf->lb_len == 0)) {
2417                 goto again;
2418         } else {
2419                 fl = LU_XATTR_REPLACE;
2420         }
2421
2422         if (fl == LU_XATTR_CREATE) {
2423                 if (bk->lb_param & LPF_DRYRUN)
2424                         GOTO(unlock_parent, rc = 1);
2425
2426                 LASSERT(buf->lb_len >= rc);
2427
2428                 buf->lb_len = rc;
2429                 rc = lfsck_layout_extend_lovea(env, handle, parent, cfid, buf,
2430                                                fl, ost_idx, ea_off, false);
2431
2432                 GOTO(unlock_parent, rc);
2433         }
2434
2435         lmm = buf->lb_buf;
2436         rc1 = lfsck_layout_verify_header(lmm);
2437
2438         /* If the LOV EA crashed, the rebuild it. */
2439         if (rc1 == -EINVAL) {
2440                 if (bk->lb_param & LPF_DRYRUN)
2441                         GOTO(unlock_parent, rc = 1);
2442
2443                 LASSERT(buf->lb_len >= rc);
2444
2445                 buf->lb_len = rc;
2446                 memset(lmm, 0, buf->lb_len);
2447                 rc = lfsck_layout_extend_lovea(env, handle, parent, cfid, buf,
2448                                                fl, ost_idx, ea_off, true);
2449
2450                 GOTO(unlock_parent, rc);
2451         }
2452
2453         /* For other unknown magic/pattern, keep the current LOV EA. */
2454         if (rc1 != 0)
2455                 GOTO(unlock_parent, rc = rc1);
2456
2457         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2458          * been verified in lfsck_layout_verify_header() already. If some
2459          * new magic introduced in the future, then layout LFSCK needs to
2460          * be updated also. */
2461         magic = le32_to_cpu(lmm->lmm_magic);
2462         if (magic == LOV_MAGIC_V1) {
2463                 objs = &(lmm->lmm_objects[0]);
2464         } else {
2465                 LASSERT(magic == LOV_MAGIC_V3);
2466                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2467         }
2468
2469         count = le16_to_cpu(lmm->lmm_stripe_count);
2470         if (count == 0)
2471                 GOTO(unlock_parent, rc = -EINVAL);
2472         LASSERT(count > 0);
2473
2474         /* Exceed the current end of MDT-object layout EA. Then extend it. */
2475         if (count <= ea_off) {
2476                 if (bk->lb_param & LPF_DRYRUN)
2477                         GOTO(unlock_parent, rc = 1);
2478
2479                 rc = lov_mds_md_size(ea_off + 1, magic);
2480                 /* If the declared is not big enough, re-try. */
2481                 if (buf->lb_len < rc)
2482                         goto again;
2483
2484                 buf->lb_len = rc;
2485                 rc = lfsck_layout_extend_lovea(env, handle, parent, cfid, buf,
2486                                                fl, ost_idx, ea_off, false);
2487                 GOTO(unlock_parent, rc);
2488         }
2489
2490         LASSERTF(rc > 0, "invalid rc = %d\n", rc);
2491
2492         buf->lb_len = rc;
2493         for (i = 0; i < count; i++, objs++) {
2494                 /* The MDT-object was created via lfsck_layout_recover_create()
2495                  * by others before, and we fill the dummy layout EA. */
2496                 if (is_dummy_lov_ost_data(objs)) {
2497                         if (i != ea_off)
2498                                 continue;
2499
2500                         if (bk->lb_param & LPF_DRYRUN)
2501                                 GOTO(unlock_parent, rc = 1);
2502
2503                         lmm->lmm_layout_gen =
2504                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2505                         rc = lfsck_layout_refill_lovea(env, handle, parent,
2506                                                        cfid, buf, objs, fl,
2507                                                        ost_idx);
2508                         GOTO(unlock_parent, rc);
2509                 }
2510
2511                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2512                 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
2513                 /* It should be rare case, the slot is there, but the LFSCK
2514                  * does not handle it during the first-phase cycle scanning. */
2515                 if (unlikely(lu_fid_eq(fid, cfid))) {
2516                         if (i == ea_off) {
2517                                 GOTO(unlock_parent, rc = 0);
2518                         } else {
2519                                 /* Rare case that the OST-object index
2520                                  * does not match the parent MDT-object
2521                                  * layout EA. We trust the later one. */
2522                                 if (bk->lb_param & LPF_DRYRUN)
2523                                         GOTO(unlock_parent, rc = 1);
2524
2525                                 dt_write_unlock(env, parent);
2526                                 if (handle != NULL)
2527                                         dt_trans_stop(env, dt, handle);
2528                                 lfsck_layout_unlock(&lh);
2529                                 buf->lb_len = buflen;
2530                                 rc = lfsck_layout_update_pfid(env, com, parent,
2531                                                         cfid, ltd->ltd_tgt, i);
2532
2533                                 RETURN(rc);
2534                         }
2535                 }
2536         }
2537
2538         /* The MDT-object exists, but related layout EA slot is occupied
2539          * by others. */
2540         if (bk->lb_param & LPF_DRYRUN)
2541                 GOTO(unlock_parent, rc = 1);
2542
2543         dt_write_unlock(env, parent);
2544         if (handle != NULL)
2545                 dt_trans_stop(env, dt, handle);
2546         lfsck_layout_unlock(&lh);
2547         if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
2548                 objs = &(lmm->lmm_objects[ea_off]);
2549         else
2550                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
2551         rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
2552                                           buf, objs, ea_off, buflen);
2553
2554         RETURN(rc);
2555
2556 unlock_parent:
2557         if (locked)
2558                 dt_write_unlock(env, parent);
2559
2560 stop:
2561         if (handle != NULL)
2562                 dt_trans_stop(env, dt, handle);
2563
2564 unlock_layout:
2565         lfsck_layout_unlock(&lh);
2566         buf->lb_len = buflen;
2567
2568         return rc;
2569 }
2570
2571 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
2572                                         struct lfsck_component *com,
2573                                         struct lfsck_tgt_desc *ltd,
2574                                         struct lu_orphan_rec *rec,
2575                                         struct lu_fid *cfid)
2576 {
2577         struct lfsck_layout     *lo     = com->lc_file_ram;
2578         struct lu_fid           *pfid   = &rec->lor_fid;
2579         struct dt_object        *parent = NULL;
2580         __u32                    ea_off = pfid->f_stripe_idx;
2581         int                      rc     = 0;
2582         ENTRY;
2583
2584         if (!fid_is_sane(cfid))
2585                 GOTO(out, rc = -EINVAL);
2586
2587         if (fid_is_zero(pfid)) {
2588                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2589                                                   "N-", "", ea_off);
2590                 GOTO(out, rc);
2591         }
2592
2593         pfid->f_ver = 0;
2594         if (!fid_is_sane(pfid))
2595                 GOTO(out, rc = -EINVAL);
2596
2597         parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
2598         if (IS_ERR(parent))
2599                 GOTO(out, rc = PTR_ERR(parent));
2600
2601         if (unlikely(dt_object_remote(parent) != 0))
2602                 GOTO(put, rc = -EXDEV);
2603
2604         if (dt_object_exists(parent) == 0) {
2605                 lu_object_put(env, &parent->do_lu);
2606                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2607                                                   "R-", "", ea_off);
2608                 GOTO(out, rc);
2609         }
2610
2611         if (!S_ISREG(lu_object_attr(&parent->do_lu)))
2612                 GOTO(put, rc = -EISDIR);
2613
2614         rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
2615                                          ltd->ltd_index, ea_off);
2616
2617         GOTO(put, rc);
2618
2619 put:
2620         if (rc <= 0)
2621                 lu_object_put(env, &parent->do_lu);
2622         else
2623                 /* The layout EA is changed, need to be reloaded next time. */
2624                 lu_object_put_nocache(env, &parent->do_lu);
2625
2626 out:
2627         down_write(&com->lc_sem);
2628         com->lc_new_scanned++;
2629         com->lc_new_checked++;
2630         if (rc > 0) {
2631                 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
2632                 rc = 0;
2633         } else if (rc < 0) {
2634                 lo->ll_objs_failed_phase2++;
2635         }
2636         up_write(&com->lc_sem);
2637
2638         return rc;
2639 }
2640
2641 static int lfsck_layout_scan_orphan(const struct lu_env *env,
2642                                     struct lfsck_component *com,
2643                                     struct lfsck_tgt_desc *ltd)
2644 {
2645         struct lfsck_layout             *lo     = com->lc_file_ram;
2646         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2647         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
2648         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2649         struct ost_id                   *oi     = &info->lti_oi;
2650         struct lu_fid                   *fid    = &info->lti_fid;
2651         struct dt_object                *obj;
2652         const struct dt_it_ops          *iops;
2653         struct dt_it                    *di;
2654         int                              rc     = 0;
2655         ENTRY;
2656
2657         CDEBUG(D_LFSCK, "%s: start the orphan scanning for OST%04x\n",
2658                lfsck_lfsck2name(lfsck), ltd->ltd_index);
2659
2660         ostid_set_seq(oi, FID_SEQ_IDIF);
2661         ostid_set_id(oi, 0);
2662         ostid_to_fid(fid, oi, ltd->ltd_index);
2663         obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
2664         if (unlikely(IS_ERR(obj)))
2665                 RETURN(PTR_ERR(obj));
2666
2667         rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_orphan_features);
2668         if (rc != 0)
2669                 GOTO(put, rc);
2670
2671         iops = &obj->do_index_ops->dio_it;
2672         di = iops->init(env, obj, 0, BYPASS_CAPA);
2673         if (IS_ERR(di))
2674                 GOTO(put, rc = PTR_ERR(di));
2675
2676         rc = iops->load(env, di, 0);
2677         if (rc == -ESRCH) {
2678                 /* -ESRCH means that the orphan OST-objects rbtree has been
2679                  * cleanup because of the OSS server restart or other errors. */
2680                 lo->ll_flags |= LF_INCOMPLETE;
2681                 GOTO(fini, rc);
2682         }
2683
2684         if (rc == 0)
2685                 rc = iops->next(env, di);
2686         else if (rc > 0)
2687                 rc = 0;
2688
2689         if (rc < 0)
2690                 GOTO(fini, rc);
2691
2692         if (rc > 0)
2693                 GOTO(fini, rc = 0);
2694
2695         do {
2696                 struct dt_key           *key;
2697                 struct lu_orphan_rec    *rec = &info->lti_rec;
2698
2699                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
2700                     cfs_fail_val > 0) {
2701                         struct ptlrpc_thread    *thread = &lfsck->li_thread;
2702                         struct l_wait_info       lwi;
2703
2704                         lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
2705                                           NULL, NULL);
2706                         l_wait_event(thread->t_ctl_waitq,
2707                                      !thread_is_running(thread),
2708                                      &lwi);
2709                 }
2710
2711                 key = iops->key(env, di);
2712                 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
2713                 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
2714                 if (rc == 0)
2715                         rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
2716                                         &com->lc_fid_latest_scanned_phase2);
2717                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2718                         GOTO(fini, rc);
2719
2720                 lfsck_control_speed_by_self(com);
2721                 do {
2722                         rc = iops->next(env, di);
2723                 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
2724         } while (rc == 0);
2725
2726         GOTO(fini, rc);
2727
2728 fini:
2729         iops->put(env, di);
2730         iops->fini(env, di);
2731 put:
2732         lu_object_put(env, &obj->do_lu);
2733
2734         CDEBUG(D_LFSCK, "%s: finish the orphan scanning for OST%04x, rc = %d\n",
2735                lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
2736
2737         return rc > 0 ? 0 : rc;
2738 }
2739
2740 /* For the MDT-object with dangling reference, we need to repare the
2741  * inconsistency according to the LFSCK sponsor's requirement:
2742  *
2743  * 1) Keep the inconsistency there and report the inconsistency case,
2744  *    then give the chance to the application to find related issues,
2745  *    and the users can make the decision about how to handle it with
2746  *    more human knownledge. (by default)
2747  *
2748  * 2) Re-create the missed OST-object with the FID/owner information. */
2749 static int lfsck_layout_repair_dangling(const struct lu_env *env,
2750                                         struct lfsck_component *com,
2751                                         struct lfsck_layout_req *llr,
2752                                         const struct lu_attr *pla)
2753 {
2754         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2755         struct filter_fid               *pfid   = &info->lti_new_pfid;
2756         struct dt_allocation_hint       *hint   = &info->lti_hint;
2757         struct lu_attr                  *cla    = &info->lti_la2;
2758         struct dt_object                *parent = llr->llr_parent->llo_obj;
2759         struct dt_object                *child  = llr->llr_child;
2760         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
2761         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
2762         struct thandle                  *handle;
2763         struct lu_buf                   *buf;
2764         struct lustre_handle             lh     = { 0 };
2765         int                              rc;
2766         bool                             create;
2767         ENTRY;
2768
2769         if (com->lc_lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ)
2770                 create = true;
2771         else
2772                 create = false;
2773
2774         CDEBUG(D_LFSCK, "Found dangling reference for: parent "DFID
2775                ", child "DFID", OST-index %u, stripe-index %u, owner %u:%u. %s",
2776                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
2777                llr->llr_ost_idx, llr->llr_lov_idx, pla->la_uid, pla->la_gid,
2778                create ? "Create the lost OST-object as required.\n" :
2779                         "Keep the MDT-object there by default.\n");
2780
2781         if (!create)
2782                 RETURN(1);
2783
2784         memset(cla, 0, sizeof(*cla));
2785         cla->la_uid = pla->la_uid;
2786         cla->la_gid = pla->la_gid;
2787         cla->la_mode = S_IFREG | 0666;
2788         cla->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
2789                         LA_ATIME | LA_MTIME | LA_CTIME;
2790
2791         rc = lfsck_layout_lock(env, com, parent, &lh,
2792                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
2793         if (rc != 0)
2794                 RETURN(rc);
2795
2796         handle = dt_trans_create(env, dev);
2797         if (IS_ERR(handle))
2798                 GOTO(unlock1, rc = PTR_ERR(handle));
2799
2800         hint->dah_parent = NULL;
2801         hint->dah_mode = 0;
2802         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
2803         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
2804         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2805          * MDT-object's FID::f_ver, instead it is the OST-object index in its
2806          * parent MDT-object's layout EA. */
2807         pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
2808         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
2809
2810         rc = dt_declare_create(env, child, cla, hint, NULL, handle);
2811         if (rc != 0)
2812                 GOTO(stop, rc);
2813
2814         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
2815                                   LU_XATTR_CREATE, handle);
2816         if (rc != 0)
2817                 GOTO(stop, rc);
2818
2819         rc = dt_trans_start(env, dev, handle);
2820         if (rc != 0)
2821                 GOTO(stop, rc);
2822
2823         dt_read_lock(env, parent, 0);
2824         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
2825                 GOTO(unlock2, rc = 1);
2826
2827         rc = dt_create(env, child, cla, hint, NULL, handle);
2828         if (rc != 0)
2829                 GOTO(unlock2, rc);
2830
2831         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
2832                           handle, BYPASS_CAPA);
2833
2834         GOTO(unlock2, rc);
2835
2836 unlock2:
2837         dt_read_unlock(env, parent);
2838
2839 stop:
2840         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2841
2842 unlock1:
2843         lfsck_layout_unlock(&lh);
2844
2845         return rc;
2846 }
2847
2848 /* If the OST-object does not recognize the MDT-object as its parent, and
2849  * there is no other MDT-object claims as its parent, then just trust the
2850  * given MDT-object as its parent. So update the OST-object filter_fid. */
2851 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
2852                                               struct lfsck_component *com,
2853                                               struct lfsck_layout_req *llr,
2854                                               const struct lu_attr *pla)
2855 {
2856         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2857         struct filter_fid               *pfid   = &info->lti_new_pfid;
2858         struct lu_attr                  *tla    = &info->lti_la3;
2859         struct dt_object                *parent = llr->llr_parent->llo_obj;
2860         struct dt_object                *child  = llr->llr_child;
2861         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
2862         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
2863         struct thandle                  *handle;
2864         struct lu_buf                   *buf;
2865         struct lustre_handle             lh     = { 0 };
2866         int                              rc;
2867         ENTRY;
2868
2869         CDEBUG(D_LFSCK, "Repair unmatched MDT-OST pair for: parent "DFID
2870                ", child "DFID", OST-index %u, stripe-index %u, owner %u:%u\n",
2871                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
2872                llr->llr_ost_idx, llr->llr_lov_idx, pla->la_uid, pla->la_gid);
2873
2874         rc = lfsck_layout_lock(env, com, parent, &lh,
2875                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
2876         if (rc != 0)
2877                 RETURN(rc);
2878
2879         handle = dt_trans_create(env, dev);
2880         if (IS_ERR(handle))
2881                 GOTO(unlock1, rc = PTR_ERR(handle));
2882
2883         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
2884         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
2885         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2886          * MDT-object's FID::f_ver, instead it is the OST-object index in its
2887          * parent MDT-object's layout EA. */
2888         pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
2889         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
2890
2891         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
2892         if (rc != 0)
2893                 GOTO(stop, rc);
2894
2895         tla->la_valid = LA_UID | LA_GID;
2896         tla->la_uid = pla->la_uid;
2897         tla->la_gid = pla->la_gid;
2898         rc = dt_declare_attr_set(env, child, tla, handle);
2899         if (rc != 0)
2900                 GOTO(stop, rc);
2901
2902         rc = dt_trans_start(env, dev, handle);
2903         if (rc != 0)
2904                 GOTO(stop, rc);
2905
2906         dt_write_lock(env, parent, 0);
2907         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
2908                 GOTO(unlock2, rc = 1);
2909
2910         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle,
2911                           BYPASS_CAPA);
2912         if (rc != 0)
2913                 GOTO(unlock2, rc);
2914
2915         /* Get the latest parent's owner. */
2916         rc = dt_attr_get(env, parent, tla, BYPASS_CAPA);
2917         if (rc != 0)
2918                 GOTO(unlock2, rc);
2919
2920         tla->la_valid = LA_UID | LA_GID;
2921         rc = dt_attr_set(env, child, tla, handle, BYPASS_CAPA);
2922
2923         GOTO(unlock2, rc);
2924
2925 unlock2:
2926         dt_write_unlock(env, parent);
2927
2928 stop:
2929         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2930
2931 unlock1:
2932         lfsck_layout_unlock(&lh);
2933
2934         return rc;
2935 }
2936
2937 /* If there are more than one MDT-objects claim as the OST-object's parent,
2938  * and the OST-object only recognizes one of them, then we need to generate
2939  * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
2940 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
2941                                                    struct lfsck_component *com,
2942                                                    struct lfsck_layout_req *llr,
2943                                                    struct lu_attr *la,
2944                                                    struct lu_buf *buf)
2945 {
2946         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2947         struct dt_allocation_hint       *hint   = &info->lti_hint;
2948         struct dt_object_format         *dof    = &info->lti_dof;
2949         struct dt_device                *pdev   = com->lc_lfsck->li_next;
2950         struct ost_id                   *oi     = &info->lti_oi;
2951         struct dt_object                *parent = llr->llr_parent->llo_obj;
2952         struct dt_device                *cdev   = lfsck_obj2dt_dev(llr->llr_child);
2953         struct dt_object                *child  = NULL;
2954         struct lu_device                *d      = &cdev->dd_lu_dev;
2955         struct lu_object                *o      = NULL;
2956         struct thandle                  *handle;
2957         struct lov_mds_md_v1            *lmm;
2958         struct lov_ost_data_v1          *objs;
2959         struct lustre_handle             lh     = { 0 };
2960         __u32                            magic;
2961         int                              rc;
2962         ENTRY;
2963
2964         CDEBUG(D_LFSCK, "Repair multiple references for: parent "DFID
2965                ", OST-index %u, stripe-index %u, owner %u:%u\n",
2966                PFID(lfsck_dto2fid(parent)), llr->llr_ost_idx,
2967                llr->llr_lov_idx, la->la_uid, la->la_gid);
2968
2969         rc = lfsck_layout_lock(env, com, parent, &lh,
2970                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
2971         if (rc != 0)
2972                 RETURN(rc);
2973
2974         handle = dt_trans_create(env, pdev);
2975         if (IS_ERR(handle))
2976                 GOTO(unlock1, rc = PTR_ERR(handle));
2977
2978         o = lu_object_anon(env, d, NULL);
2979         if (IS_ERR(o))
2980                 GOTO(stop, rc = PTR_ERR(o));
2981
2982         child = container_of(o, struct dt_object, do_lu);
2983         o = lu_object_locate(o->lo_header, d->ld_type);
2984         if (unlikely(o == NULL))
2985                 GOTO(stop, rc = -EINVAL);
2986
2987         child = container_of(o, struct dt_object, do_lu);
2988         la->la_valid = LA_UID | LA_GID;
2989         hint->dah_parent = NULL;
2990         hint->dah_mode = 0;
2991         dof->dof_type = DFT_REGULAR;
2992         rc = dt_declare_create(env, child, la, NULL, NULL, handle);
2993         if (rc != 0)
2994                 GOTO(stop, rc);
2995
2996         rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2997                                   LU_XATTR_REPLACE, handle);
2998         if (rc != 0)
2999                 GOTO(stop, rc);
3000
3001         rc = dt_trans_start(env, pdev, handle);
3002         if (rc != 0)
3003                 GOTO(stop, rc);
3004
3005         dt_write_lock(env, parent, 0);
3006         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
3007                 GOTO(unlock2, rc = 0);
3008
3009         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV, BYPASS_CAPA);
3010         if (unlikely(rc == 0 || rc == -ENODATA || rc == -ERANGE))
3011                 GOTO(unlock2, rc = 0);
3012
3013         lmm = buf->lb_buf;
3014         /* Someone change layout during the LFSCK, no need to repair then. */
3015         if (le16_to_cpu(lmm->lmm_layout_gen) != llr->llr_parent->llo_gen)
3016                 GOTO(unlock2, rc = 0);
3017
3018         rc = dt_create(env, child, la, hint, dof, handle);
3019         if (rc != 0)
3020                 GOTO(unlock2, rc);
3021
3022         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
3023          * been verified in lfsck_layout_verify_header() already. If some
3024          * new magic introduced in the future, then layout LFSCK needs to
3025          * be updated also. */
3026         magic = le32_to_cpu(lmm->lmm_magic);
3027         if (magic == LOV_MAGIC_V1) {
3028                 objs = &(lmm->lmm_objects[0]);
3029         } else {
3030                 LASSERT(magic == LOV_MAGIC_V3);
3031                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3032         }
3033
3034         lmm->lmm_layout_gen = cpu_to_le16(llr->llr_parent->llo_gen + 1);
3035         fid_to_ostid(lu_object_fid(&child->do_lu), oi);
3036         ostid_cpu_to_le(oi, &objs[llr->llr_lov_idx].l_ost_oi);
3037         objs[llr->llr_lov_idx].l_ost_gen = cpu_to_le32(0);
3038         objs[llr->llr_lov_idx].l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
3039         rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3040                           LU_XATTR_REPLACE, handle, BYPASS_CAPA);
3041
3042         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
3043
3044 unlock2:
3045         dt_write_unlock(env, parent);
3046
3047 stop:
3048         if (child != NULL)
3049                 lu_object_put(env, &child->do_lu);
3050
3051         dt_trans_stop(env, pdev, handle);
3052
3053 unlock1:
3054         lfsck_layout_unlock(&lh);
3055
3056         return rc;
3057 }
3058
3059 /* If the MDT-object and the OST-object have different owner information,
3060  * then trust the MDT-object, because the normal chown/chgrp handle order
3061  * is from MDT to OST, and it is possible that some chown/chgrp operation
3062  * is partly done. */
3063 static int lfsck_layout_repair_owner(const struct lu_env *env,
3064                                      struct lfsck_component *com,
3065                                      struct lfsck_layout_req *llr,
3066                                      struct lu_attr *pla)
3067 {
3068         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3069         struct lu_attr                  *tla    = &info->lti_la3;
3070         struct dt_object                *parent = llr->llr_parent->llo_obj;
3071         struct dt_object                *child  = llr->llr_child;
3072         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
3073         struct thandle                  *handle;
3074         int                              rc;
3075         ENTRY;
3076
3077         CDEBUG(D_LFSCK, "Repair inconsistent file owner for: parent "DFID
3078                ", child "DFID", OST-index %u, stripe-index %u, owner %u:%u\n",
3079                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
3080                llr->llr_ost_idx, llr->llr_lov_idx, pla->la_uid, pla->la_gid);
3081
3082         handle = dt_trans_create(env, dev);
3083         if (IS_ERR(handle))
3084                 RETURN(PTR_ERR(handle));
3085
3086         tla->la_uid = pla->la_uid;
3087         tla->la_gid = pla->la_gid;
3088         tla->la_valid = LA_UID | LA_GID;
3089         rc = dt_declare_attr_set(env, child, tla, handle);
3090         if (rc != 0)
3091                 GOTO(stop, rc);
3092
3093         rc = dt_trans_start(env, dev, handle);
3094         if (rc != 0)
3095                 GOTO(stop, rc);
3096
3097         /* Use the dt_object lock to serialize with destroy and attr_set. */
3098         dt_read_lock(env, parent, 0);
3099         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
3100                 GOTO(unlock, rc = 1);
3101
3102         /* Get the latest parent's owner. */
3103         rc = dt_attr_get(env, parent, tla, BYPASS_CAPA);
3104         if (rc != 0) {
3105                 CWARN("%s: fail to get the latest parent's ("DFID") owner, "
3106                       "not sure whether some others chown/chgrp during the "
3107                       "LFSCK: rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
3108                       PFID(lfsck_dto2fid(parent)), rc);
3109
3110                 GOTO(unlock, rc);
3111         }
3112
3113         /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
3114         if (unlikely(tla->la_uid != pla->la_uid ||
3115                      tla->la_gid != pla->la_gid))
3116                 GOTO(unlock, rc = 1);
3117
3118         tla->la_valid = LA_UID | LA_GID;
3119         rc = dt_attr_set(env, child, tla, handle, BYPASS_CAPA);
3120
3121         GOTO(unlock, rc);
3122
3123 unlock:
3124         dt_read_unlock(env, parent);
3125
3126 stop:
3127         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3128
3129         return rc;
3130 }
3131
3132 /* Check whether the OST-object correctly back points to the
3133  * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
3134 static int lfsck_layout_check_parent(const struct lu_env *env,
3135                                      struct lfsck_component *com,
3136                                      struct dt_object *parent,
3137                                      const struct lu_fid *pfid,
3138                                      const struct lu_fid *cfid,
3139                                      const struct lu_attr *pla,
3140                                      const struct lu_attr *cla,
3141                                      struct lfsck_layout_req *llr,
3142                                      struct lu_buf *lov_ea, __u32 idx)
3143 {
3144         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3145         struct lu_buf                   *buf    = &info->lti_big_buf;
3146         struct dt_object                *tobj;
3147         struct lov_mds_md_v1            *lmm;
3148         struct lov_ost_data_v1          *objs;
3149         int                              rc;
3150         int                              i;
3151         __u32                            magic;
3152         __u16                            count;
3153         ENTRY;
3154
3155         if (fid_is_zero(pfid)) {
3156                 /* client never wrote. */
3157                 if (cla->la_size == 0 && cla->la_blocks == 0) {
3158                         if (unlikely(cla->la_uid != pla->la_uid ||
3159                                      cla->la_gid != pla->la_gid))
3160                                 RETURN (LLIT_INCONSISTENT_OWNER);
3161
3162                         RETURN(0);
3163                 }
3164
3165                 RETURN(LLIT_UNMATCHED_PAIR);
3166         }
3167
3168         if (unlikely(!fid_is_sane(pfid)))
3169                 RETURN(LLIT_UNMATCHED_PAIR);
3170
3171         if (lu_fid_eq(pfid, lu_object_fid(&parent->do_lu))) {
3172                 if (llr->llr_lov_idx == idx)
3173                         RETURN(0);
3174
3175                 RETURN(LLIT_UNMATCHED_PAIR);
3176         }
3177
3178         tobj = lfsck_object_find(env, com->lc_lfsck, pfid);
3179         if (tobj == NULL)
3180                 RETURN(LLIT_UNMATCHED_PAIR);
3181
3182         if (IS_ERR(tobj))
3183                 RETURN(PTR_ERR(tobj));
3184
3185         if (!dt_object_exists(tobj))
3186                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3187
3188         /* Load the tobj's layout EA, in spite of it is a local MDT-object or
3189          * remote one on another MDT. Then check whether the given OST-object
3190          * is in such layout. If yes, it is multiple referenced, otherwise it
3191          * is unmatched referenced case. */
3192         rc = lfsck_layout_get_lovea(env, tobj, buf, NULL);
3193         if (rc == 0)
3194                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3195
3196         if (rc < 0)
3197                 GOTO(out, rc);
3198
3199         lmm = buf->lb_buf;
3200         magic = le32_to_cpu(lmm->lmm_magic);
3201         if (magic == LOV_MAGIC_V1) {
3202                 objs = &(lmm->lmm_objects[0]);
3203         } else {
3204                 LASSERT(magic == LOV_MAGIC_V3);
3205                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3206         }
3207
3208         count = le16_to_cpu(lmm->lmm_stripe_count);
3209         for (i = 0; i < count; i++, objs++) {
3210                 struct lu_fid           *tfid   = &info->lti_fid2;
3211                 struct ost_id           *oi     = &info->lti_oi;
3212
3213                 if (is_dummy_lov_ost_data(objs))
3214                         continue;
3215
3216                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3217                 ostid_to_fid(tfid, oi, le32_to_cpu(objs->l_ost_idx));
3218                 if (lu_fid_eq(cfid, tfid)) {
3219                         *lov_ea = *buf;
3220
3221                         GOTO(out, rc = LLIT_MULTIPLE_REFERENCED);
3222                 }
3223         }
3224
3225         GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3226
3227 out:
3228         lfsck_object_put(env, tobj);
3229
3230         return rc;
3231 }
3232
3233 static int lfsck_layout_assistant_handle_one(const struct lu_env *env,
3234                                              struct lfsck_component *com,
3235                                              struct lfsck_layout_req *llr)
3236 {
3237         struct lfsck_layout                  *lo     = com->lc_file_ram;
3238         struct lfsck_thread_info             *info   = lfsck_env_info(env);
3239         struct filter_fid_old                *pea    = &info->lti_old_pfid;
3240         struct lu_fid                        *pfid   = &info->lti_fid;
3241         struct lu_buf                        *buf    = NULL;
3242         struct dt_object                     *parent = llr->llr_parent->llo_obj;
3243         struct dt_object                     *child  = llr->llr_child;
3244         struct lu_attr                       *pla    = &info->lti_la;
3245         struct lu_attr                       *cla    = &info->lti_la2;
3246         struct lfsck_instance                *lfsck  = com->lc_lfsck;
3247         struct lfsck_bookmark                *bk     = &lfsck->li_bookmark_ram;
3248         enum lfsck_layout_inconsistency_type  type   = LLIT_NONE;
3249         __u32                                 idx    = 0;
3250         int                                   rc;
3251         ENTRY;
3252
3253         rc = dt_attr_get(env, parent, pla, BYPASS_CAPA);
3254         if (rc != 0) {
3255                 if (lu_object_is_dying(parent->do_lu.lo_header))
3256                         RETURN(0);
3257
3258                 GOTO(out, rc);
3259         }
3260
3261         rc = dt_attr_get(env, child, cla, BYPASS_CAPA);
3262         if (rc == -ENOENT) {
3263                 if (lu_object_is_dying(parent->do_lu.lo_header))
3264                         RETURN(0);
3265
3266                 type = LLIT_DANGLING;
3267                 goto repair;
3268         }
3269
3270         if (rc != 0)
3271                 GOTO(out, rc);
3272
3273         buf = lfsck_buf_get(env, pea, sizeof(struct filter_fid_old));
3274         rc= dt_xattr_get(env, child, buf, XATTR_NAME_FID, BYPASS_CAPA);
3275         if (unlikely(rc >= 0 && rc != sizeof(struct filter_fid_old) &&
3276                      rc != sizeof(struct filter_fid))) {
3277                 type = LLIT_UNMATCHED_PAIR;
3278                 goto repair;
3279         }
3280
3281         if (rc < 0 && rc != -ENODATA)
3282                 GOTO(out, rc);
3283
3284         if (rc == -ENODATA) {
3285                 fid_zero(pfid);
3286         } else {
3287                 fid_le_to_cpu(pfid, &pea->ff_parent);
3288                 /* Currently, the filter_fid::ff_parent::f_ver is not the
3289                  * real parent MDT-object's FID::f_ver, instead it is the
3290                  * OST-object index in its parent MDT-object's layout EA. */
3291                 idx = pfid->f_stripe_idx;
3292                 pfid->f_ver = 0;
3293         }
3294
3295         rc = lfsck_layout_check_parent(env, com, parent, pfid,
3296                                        lu_object_fid(&child->do_lu),
3297                                        pla, cla, llr, buf, idx);
3298         if (rc > 0) {
3299                 type = rc;
3300                 goto repair;
3301         }
3302
3303         if (rc < 0)
3304                 GOTO(out, rc);
3305
3306         if (unlikely(cla->la_uid != pla->la_uid ||
3307                      cla->la_gid != pla->la_gid)) {
3308                 type = LLIT_INCONSISTENT_OWNER;
3309                 goto repair;
3310         }
3311
3312 repair:
3313         if (bk->lb_param & LPF_DRYRUN) {
3314                 if (type != LLIT_NONE)
3315                         GOTO(out, rc = 1);
3316                 else
3317                         GOTO(out, rc = 0);
3318         }
3319
3320         switch (type) {
3321         case LLIT_DANGLING:
3322                 rc = lfsck_layout_repair_dangling(env, com, llr, pla);
3323                 break;
3324         case LLIT_UNMATCHED_PAIR:
3325                 rc = lfsck_layout_repair_unmatched_pair(env, com, llr, pla);
3326                 break;
3327         case LLIT_MULTIPLE_REFERENCED:
3328                 rc = lfsck_layout_repair_multiple_references(env, com, llr,
3329                                                              pla, buf);
3330                 break;
3331         case LLIT_INCONSISTENT_OWNER:
3332                 rc = lfsck_layout_repair_owner(env, com, llr, pla);
3333                 break;
3334         default:
3335                 rc = 0;
3336                 break;
3337         }
3338
3339         GOTO(out, rc);
3340
3341 out:
3342         down_write(&com->lc_sem);
3343         if (rc < 0) {
3344                 struct lfsck_layout_master_data *llmd = com->lc_data;
3345
3346                 if (unlikely(llmd->llmd_exit)) {
3347                         rc = 0;
3348                 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
3349                            rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
3350                            rc == -EHOSTUNREACH) {
3351                         /* If cannot touch the target server,
3352                          * mark the LFSCK as INCOMPLETE. */
3353                         CERROR("%s: Fail to talk with OST %x: rc = %d.\n",
3354                                lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
3355                         lo->ll_flags |= LF_INCOMPLETE;
3356                         lo->ll_objs_skipped++;
3357                         rc = 0;
3358                 } else {
3359                         lfsck_layout_record_failure(env, lfsck, lo);
3360                 }
3361         } else if (rc > 0) {
3362                 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
3363                          "unknown type = %d\n", type);
3364
3365                 lo->ll_objs_repaired[type - 1]++;
3366                 if (bk->lb_param & LPF_DRYRUN &&
3367                     unlikely(lo->ll_pos_first_inconsistent == 0))
3368                         lo->ll_pos_first_inconsistent =
3369                         lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
3370                                                         lfsck->li_di_oit);
3371         }
3372         up_write(&com->lc_sem);
3373
3374         return rc;
3375 }
3376
3377 static int lfsck_layout_assistant(void *args)
3378 {
3379         struct lfsck_thread_args        *lta     = args;
3380         struct lu_env                   *env     = &lta->lta_env;
3381         struct lfsck_component          *com     = lta->lta_com;
3382         struct lfsck_instance           *lfsck   = lta->lta_lfsck;
3383         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
3384         struct lfsck_position           *pos     = &com->lc_pos_start;
3385         struct lfsck_thread_info        *info    = lfsck_env_info(env);
3386         struct lfsck_request            *lr      = &info->lti_lr;
3387         struct lfsck_layout_master_data *llmd    = com->lc_data;
3388         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
3389         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
3390         struct lfsck_layout_req         *llr;
3391         struct l_wait_info               lwi     = { 0 };
3392         int                              rc      = 0;
3393         int                              rc1     = 0;
3394         ENTRY;
3395
3396         memset(lr, 0, sizeof(*lr));
3397         lr->lr_event = LE_START;
3398         lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
3399                        LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ;
3400         lr->lr_speed = bk->lb_speed_limit;
3401         lr->lr_version = bk->lb_version;
3402         lr->lr_param = bk->lb_param;
3403         lr->lr_async_windows = bk->lb_async_windows;
3404         lr->lr_flags = LEF_TO_OST;
3405         if (pos->lp_oit_cookie <= 1)
3406                 lr->lr_param |= LPF_RESET;
3407
3408         rc = lfsck_layout_master_notify_others(env, com, lr);
3409         if (rc != 0) {
3410                 CERROR("%s: fail to notify others for layout start: rc = %d\n",
3411                        lfsck_lfsck2name(lfsck), rc);
3412                 GOTO(fini, rc);
3413         }
3414
3415         spin_lock(&llmd->llmd_lock);
3416         thread_set_flags(athread, SVC_RUNNING);
3417         spin_unlock(&llmd->llmd_lock);
3418         wake_up_all(&mthread->t_ctl_waitq);
3419
3420         while (1) {
3421                 while (!list_empty(&llmd->llmd_req_list)) {
3422                         bool wakeup = false;
3423
3424                         if (unlikely(llmd->llmd_exit ||
3425                                      !thread_is_running(mthread)))
3426                                 GOTO(cleanup1, rc = llmd->llmd_post_result);
3427
3428                         llr = list_entry(llmd->llmd_req_list.next,
3429                                          struct lfsck_layout_req,
3430                                          llr_list);
3431                         /* Only the lfsck_layout_assistant thread itself can
3432                          * remove the "llr" from the head of the list, LFSCK
3433                          * engine thread only inserts other new "lld" at the
3434                          * end of the list. So it is safe to handle current
3435                          * "llr" without the spin_lock. */
3436                         rc = lfsck_layout_assistant_handle_one(env, com, llr);
3437                         spin_lock(&llmd->llmd_lock);
3438                         list_del_init(&llr->llr_list);
3439                         llmd->llmd_prefetched--;
3440                         /* Wake up the main engine thread only when the list
3441                          * is empty or half of the prefetched items have been
3442                          * handled to avoid too frequent thread schedule. */
3443                         if (llmd->llmd_prefetched == 0 ||
3444                             (bk->lb_async_windows != 0 &&
3445                              bk->lb_async_windows / 2 ==
3446                              llmd->llmd_prefetched))
3447                                 wakeup = true;
3448                         spin_unlock(&llmd->llmd_lock);
3449                         if (wakeup)
3450                                 wake_up_all(&mthread->t_ctl_waitq);
3451
3452                         lfsck_layout_req_fini(env, llr);
3453                         if (rc < 0 && bk->lb_param & LPF_FAILOUT)
3454                                 GOTO(cleanup1, rc);
3455                 }
3456
3457                 l_wait_event(athread->t_ctl_waitq,
3458                              !lfsck_layout_req_empty(llmd) ||
3459                              llmd->llmd_exit ||
3460                              llmd->llmd_to_post ||
3461                              llmd->llmd_to_double_scan,
3462                              &lwi);
3463
3464                 if (unlikely(llmd->llmd_exit))
3465                         GOTO(cleanup1, rc = llmd->llmd_post_result);
3466
3467                 if (!list_empty(&llmd->llmd_req_list))
3468                         continue;
3469
3470                 if (llmd->llmd_to_post) {
3471                         llmd->llmd_to_post = 0;
3472                         LASSERT(llmd->llmd_post_result > 0);
3473
3474                         memset(lr, 0, sizeof(*lr));
3475                         lr->lr_event = LE_PHASE1_DONE;
3476                         lr->lr_status = llmd->llmd_post_result;
3477                         rc = lfsck_layout_master_notify_others(env, com, lr);
3478                         if (rc != 0)
3479                                 CERROR("%s: failed to notify others "
3480                                        "for layout post: rc = %d\n",
3481                                        lfsck_lfsck2name(lfsck), rc);
3482
3483                         /* Wakeup the master engine to go ahead. */
3484                         wake_up_all(&mthread->t_ctl_waitq);
3485                 }
3486
3487                 if (llmd->llmd_to_double_scan) {
3488                         llmd->llmd_to_double_scan = 0;
3489                         atomic_inc(&lfsck->li_double_scan_count);
3490                         llmd->llmd_in_double_scan = 1;
3491                         wake_up_all(&mthread->t_ctl_waitq);
3492
3493                         com->lc_new_checked = 0;
3494                         com->lc_new_scanned = 0;
3495                         com->lc_time_last_checkpoint = cfs_time_current();
3496                         com->lc_time_next_checkpoint =
3497                                 com->lc_time_last_checkpoint +
3498                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
3499
3500                         while (llmd->llmd_in_double_scan) {
3501                                 struct lfsck_tgt_descs  *ltds =
3502                                                         &lfsck->li_ost_descs;
3503                                 struct lfsck_tgt_desc   *ltd;
3504
3505                                 rc = lfsck_layout_master_query_others(env, com);
3506                                 if (lfsck_layout_master_to_orphan(llmd))
3507                                         goto orphan;
3508
3509                                 if (rc < 0)
3510                                         GOTO(cleanup2, rc);
3511
3512                                 /* Pull LFSCK status on related targets once
3513                                  * per 30 seconds if we are not notified. */
3514                                 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
3515                                                            cfs_time_seconds(1),
3516                                                            NULL, NULL);
3517                                 rc = l_wait_event(athread->t_ctl_waitq,
3518                                         lfsck_layout_master_to_orphan(llmd) ||
3519                                         llmd->llmd_exit ||
3520                                         !thread_is_running(mthread),
3521                                         &lwi);
3522
3523                                 if (unlikely(llmd->llmd_exit ||
3524                                              !thread_is_running(mthread)))
3525                                         GOTO(cleanup2, rc = 0);
3526
3527                                 if (rc == -ETIMEDOUT)
3528                                         continue;
3529
3530                                 if (rc < 0)
3531                                         GOTO(cleanup2, rc);
3532
3533 orphan:
3534                                 spin_lock(&ltds->ltd_lock);
3535                                 while (!list_empty(
3536                                                 &llmd->llmd_ost_phase2_list)) {
3537                                         ltd = list_entry(
3538                                               llmd->llmd_ost_phase2_list.next,
3539                                               struct lfsck_tgt_desc,
3540                                               ltd_layout_phase_list);
3541                                         list_del_init(
3542                                                 &ltd->ltd_layout_phase_list);
3543                                         spin_unlock(&ltds->ltd_lock);
3544
3545                                         if (bk->lb_param & LPF_ALL_TGT) {
3546                                                 rc = lfsck_layout_scan_orphan(
3547                                                                 env, com, ltd);
3548                                                 if (rc != 0 &&
3549                                                     bk->lb_param & LPF_FAILOUT)
3550                                                         GOTO(cleanup2, rc);
3551                                         }
3552
3553                                         if (unlikely(llmd->llmd_exit ||
3554                                                 !thread_is_running(mthread)))
3555                                                 GOTO(cleanup2, rc = 0);
3556
3557                                         spin_lock(&ltds->ltd_lock);
3558                                 }
3559
3560                                 if (list_empty(&llmd->llmd_ost_phase1_list)) {
3561                                         spin_unlock(&ltds->ltd_lock);
3562                                         GOTO(cleanup2, rc = 1);
3563                                 }
3564                                 spin_unlock(&ltds->ltd_lock);
3565                         }
3566                 }
3567         }
3568
3569 cleanup1:
3570         /* Cleanup the unfinished requests. */
3571         spin_lock(&llmd->llmd_lock);
3572         if (rc < 0)
3573                 llmd->llmd_assistant_status = rc;
3574
3575         while (!list_empty(&llmd->llmd_req_list)) {
3576                 llr = list_entry(llmd->llmd_req_list.next,
3577                                  struct lfsck_layout_req,
3578                                  llr_list);
3579                 list_del_init(&llr->llr_list);
3580                 llmd->llmd_prefetched--;
3581                 spin_unlock(&llmd->llmd_lock);
3582                 lfsck_layout_req_fini(env, llr);
3583                 spin_lock(&llmd->llmd_lock);
3584         }
3585         spin_unlock(&llmd->llmd_lock);
3586
3587         LASSERTF(llmd->llmd_prefetched == 0, "unmatched prefeteched objs %d\n",
3588                  llmd->llmd_prefetched);
3589
3590 cleanup2:
3591         memset(lr, 0, sizeof(*lr));
3592         if (rc > 0) {
3593                 lr->lr_event = LE_PHASE2_DONE;
3594                 lr->lr_status = rc;
3595         } else if (rc == 0) {
3596                 if (lfsck->li_flags & LPF_ALL_TGT) {
3597                         lr->lr_event = LE_STOP;
3598                         lr->lr_status = LS_STOPPED;
3599                 } else {
3600                         lr->lr_event = LE_PEER_EXIT;
3601                         switch (lfsck->li_status) {
3602                         case LS_PAUSED:
3603                         case LS_CO_PAUSED:
3604                                 lr->lr_status = LS_CO_PAUSED;
3605                                 break;
3606                         case LS_STOPPED:
3607                         case LS_CO_STOPPED:
3608                                 lr->lr_status = LS_CO_STOPPED;
3609                                 break;
3610                         default:
3611                                 CERROR("%s: unknown status: rc = %d\n",
3612                                        lfsck_lfsck2name(lfsck),
3613                                        lfsck->li_status);
3614                                 lr->lr_status = LS_CO_FAILED;
3615                                 break;
3616                         }
3617                 }
3618         } else {
3619                 if (lfsck->li_flags & LPF_ALL_TGT) {
3620                         lr->lr_event = LE_STOP;
3621                         lr->lr_status = LS_FAILED;
3622                 } else {
3623                         lr->lr_event = LE_PEER_EXIT;
3624                         lr->lr_status = LS_CO_FAILED;
3625                 }
3626         }
3627
3628         rc1 = lfsck_layout_master_notify_others(env, com, lr);
3629         if (rc1 != 0) {
3630                 CERROR("%s: failed to notify others for layout quit: rc = %d\n",
3631                        lfsck_lfsck2name(lfsck), rc1);
3632                 rc = rc1;
3633         }
3634
3635         /* Under force exit case, some requests may be just freed without
3636          * verification, those objects should be re-handled when next run.
3637          * So not update the on-disk tracing file under such case. */
3638         if (llmd->llmd_in_double_scan && !llmd->llmd_exit)
3639                 rc1 = lfsck_layout_double_scan_result(env, com, rc);
3640
3641 fini:
3642         if (llmd->llmd_in_double_scan)
3643                 atomic_dec(&lfsck->li_double_scan_count);
3644
3645         spin_lock(&llmd->llmd_lock);
3646         llmd->llmd_assistant_status = (rc1 != 0 ? rc1 : rc);
3647         thread_set_flags(athread, SVC_STOPPED);
3648         wake_up_all(&mthread->t_ctl_waitq);
3649         spin_unlock(&llmd->llmd_lock);
3650         lfsck_thread_args_fini(lta);
3651
3652         return rc;
3653 }
3654
3655 static int
3656 lfsck_layout_slave_async_interpret(const struct lu_env *env,
3657                                    struct ptlrpc_request *req,
3658                                    void *args, int rc)
3659 {
3660         struct lfsck_layout_slave_async_args *llsaa = args;
3661         struct obd_export                    *exp   = llsaa->llsaa_exp;
3662         struct lfsck_component               *com   = llsaa->llsaa_com;
3663         struct lfsck_layout_slave_target     *llst  = llsaa->llsaa_llst;
3664         struct lfsck_layout_slave_data       *llsd  = com->lc_data;
3665         bool                                  done  = false;
3666
3667         if (rc != 0) {
3668                 /* It is quite probably caused by target crash,
3669                  * to make the LFSCK can go ahead, assume that
3670                  * the target finished the LFSCK prcoessing. */
3671                 done = true;
3672         } else {
3673                 struct lfsck_reply *lr;
3674
3675                 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
3676                 if (lr->lr_status != LS_SCANNING_PHASE1 &&
3677                     lr->lr_status != LS_SCANNING_PHASE2)
3678                         done = true;
3679         }
3680         if (done)
3681                 lfsck_layout_llst_del(llsd, llst);
3682         lfsck_layout_llst_put(llst);
3683         lfsck_component_put(env, com);
3684         class_export_put(exp);
3685
3686         return 0;
3687 }
3688
3689 static int lfsck_layout_async_query(const struct lu_env *env,
3690                                     struct lfsck_component *com,
3691                                     struct obd_export *exp,
3692                                     struct lfsck_layout_slave_target *llst,
3693                                     struct lfsck_request *lr,
3694                                     struct ptlrpc_request_set *set)
3695 {
3696         struct lfsck_layout_slave_async_args *llsaa;
3697         struct ptlrpc_request                *req;
3698         struct lfsck_request                 *tmp;
3699         int                                   rc;
3700         ENTRY;
3701
3702         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
3703         if (req == NULL)
3704                 RETURN(-ENOMEM);
3705
3706         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
3707         if (rc != 0) {
3708                 ptlrpc_request_free(req);
3709                 RETURN(rc);
3710         }
3711
3712         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3713         *tmp = *lr;
3714         ptlrpc_request_set_replen(req);
3715
3716         llsaa = ptlrpc_req_async_args(req);
3717         llsaa->llsaa_exp = exp;
3718         llsaa->llsaa_com = lfsck_component_get(com);
3719         llsaa->llsaa_llst = llst;
3720         req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
3721         ptlrpc_set_add_req(set, req);
3722
3723         RETURN(0);
3724 }
3725
3726 static int lfsck_layout_async_notify(const struct lu_env *env,
3727                                      struct obd_export *exp,
3728                                      struct lfsck_request *lr,
3729                                      struct ptlrpc_request_set *set)
3730 {
3731         struct ptlrpc_request   *req;
3732         struct lfsck_request    *tmp;
3733         int                      rc;
3734         ENTRY;
3735
3736         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
3737         if (req == NULL)
3738                 RETURN(-ENOMEM);
3739
3740         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
3741         if (rc != 0) {
3742                 ptlrpc_request_free(req);
3743                 RETURN(rc);
3744         }
3745
3746         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3747         *tmp = *lr;
3748         ptlrpc_request_set_replen(req);
3749         ptlrpc_set_add_req(set, req);
3750
3751         RETURN(0);
3752 }
3753
3754 static int
3755 lfsck_layout_slave_query_master(const struct lu_env *env,
3756                                 struct lfsck_component *com)
3757 {
3758         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
3759         struct lfsck_instance            *lfsck = com->lc_lfsck;
3760         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3761         struct lfsck_layout_slave_target *llst;
3762         struct obd_export                *exp;
3763         struct ptlrpc_request_set        *set;
3764         int                               rc    = 0;
3765         int                               rc1   = 0;
3766         ENTRY;
3767
3768         set = ptlrpc_prep_set();
3769         if (set == NULL)
3770                 RETURN(-ENOMEM);
3771
3772         memset(lr, 0, sizeof(*lr));
3773         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
3774         lr->lr_event = LE_QUERY;
3775         lr->lr_active = LT_LAYOUT;
3776
3777         llsd->llsd_touch_gen++;
3778         spin_lock(&llsd->llsd_lock);
3779         while (!list_empty(&llsd->llsd_master_list)) {
3780                 llst = list_entry(llsd->llsd_master_list.next,
3781                                   struct lfsck_layout_slave_target,
3782                                   llst_list);
3783                 if (llst->llst_gen == llsd->llsd_touch_gen)
3784                         break;
3785
3786                 llst->llst_gen = llsd->llsd_touch_gen;
3787                 list_del(&llst->llst_list);
3788                 list_add_tail(&llst->llst_list,
3789                               &llsd->llsd_master_list);
3790                 atomic_inc(&llst->llst_ref);
3791                 spin_unlock(&llsd->llsd_lock);
3792
3793                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
3794                                                llst->llst_index);
3795                 if (exp == NULL) {
3796                         lfsck_layout_llst_del(llsd, llst);
3797                         lfsck_layout_llst_put(llst);
3798                         spin_lock(&llsd->llsd_lock);
3799                         continue;
3800                 }
3801
3802                 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
3803                 if (rc != 0) {
3804                         CERROR("%s: slave fail to query %s for layout: "
3805                                "rc = %d\n", lfsck_lfsck2name(lfsck),
3806                                exp->exp_obd->obd_name, rc);
3807                         rc1 = rc;
3808                         lfsck_layout_llst_put(llst);
3809                         class_export_put(exp);
3810                 }
3811                 spin_lock(&llsd->llsd_lock);
3812         }
3813         spin_unlock(&llsd->llsd_lock);
3814
3815         rc = ptlrpc_set_wait(set);
3816         ptlrpc_set_destroy(set);
3817
3818         RETURN(rc1 != 0 ? rc1 : rc);
3819 }
3820
3821 static void
3822 lfsck_layout_slave_notify_master(const struct lu_env *env,
3823                                  struct lfsck_component *com,
3824                                  enum lfsck_events event, int result)
3825 {
3826         struct lfsck_instance            *lfsck = com->lc_lfsck;
3827         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3828         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
3829         struct lfsck_layout_slave_target *llst;
3830         struct obd_export                *exp;
3831         struct ptlrpc_request_set        *set;
3832         int                               rc;
3833         ENTRY;
3834
3835         set = ptlrpc_prep_set();
3836         if (set == NULL)
3837                 RETURN_EXIT;
3838
3839         memset(lr, 0, sizeof(*lr));
3840         lr->lr_event = event;
3841         lr->lr_flags = LEF_FROM_OST;
3842         lr->lr_status = result;
3843         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
3844         lr->lr_active = LT_LAYOUT;
3845         llsd->llsd_touch_gen++;
3846         spin_lock(&llsd->llsd_lock);
3847         while (!list_empty(&llsd->llsd_master_list)) {
3848                 llst = list_entry(llsd->llsd_master_list.next,
3849                                   struct lfsck_layout_slave_target,
3850                                   llst_list);
3851                 if (llst->llst_gen == llsd->llsd_touch_gen)
3852                         break;
3853
3854                 llst->llst_gen = llsd->llsd_touch_gen;
3855                 list_del(&llst->llst_list);
3856                 list_add_tail(&llst->llst_list,
3857                               &llsd->llsd_master_list);
3858                 atomic_inc(&llst->llst_ref);
3859                 spin_unlock(&llsd->llsd_lock);
3860
3861                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
3862                                                llst->llst_index);
3863                 if (exp == NULL) {
3864                         lfsck_layout_llst_del(llsd, llst);
3865                         lfsck_layout_llst_put(llst);
3866                         spin_lock(&llsd->llsd_lock);
3867                         continue;
3868                 }
3869
3870                 rc = lfsck_layout_async_notify(env, exp, lr, set);
3871                 if (rc != 0)
3872                         CERROR("%s: slave fail to notify %s for layout: "
3873                                "rc = %d\n", lfsck_lfsck2name(lfsck),
3874                                exp->exp_obd->obd_name, rc);
3875                 lfsck_layout_llst_put(llst);
3876                 class_export_put(exp);
3877                 spin_lock(&llsd->llsd_lock);
3878         }
3879         spin_unlock(&llsd->llsd_lock);
3880
3881         ptlrpc_set_wait(set);
3882         ptlrpc_set_destroy(set);
3883
3884         RETURN_EXIT;
3885 }
3886
3887 /*
3888  * \ret -ENODATA: unrecognized stripe
3889  * \ret = 0     : recognized stripe
3890  * \ret < 0     : other failures
3891  */
3892 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
3893                                            struct lfsck_component *com,
3894                                            struct lu_fid *cfid,
3895                                            struct lu_fid *pfid)
3896 {
3897         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3898         struct lu_buf                   *buf    = &info->lti_big_buf;
3899         struct ost_id                   *oi     = &info->lti_oi;
3900         struct dt_object                *obj;
3901         struct lov_mds_md_v1            *lmm;
3902         struct lov_ost_data_v1          *objs;
3903         __u32                            idx    = pfid->f_stripe_idx;
3904         __u32                            magic;
3905         int                              rc     = 0;
3906         int                              i;
3907         __u16                            count;
3908         ENTRY;
3909
3910         pfid->f_ver = 0;
3911         obj = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
3912         if (IS_ERR(obj))
3913                 RETURN(PTR_ERR(obj));
3914
3915         dt_read_lock(env, obj, 0);
3916         if (unlikely(!dt_object_exists(obj)))
3917                 GOTO(unlock, rc = -ENOENT);
3918
3919         rc = lfsck_layout_get_lovea(env, obj, buf, NULL);
3920         if (rc < 0)
3921                 GOTO(unlock, rc);
3922
3923         if (rc == 0)
3924                 GOTO(unlock, rc = -ENODATA);
3925
3926         lmm = buf->lb_buf;
3927         rc = lfsck_layout_verify_header(lmm);
3928         if (rc != 0)
3929                 GOTO(unlock, rc);
3930
3931         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
3932          * been verified in lfsck_layout_verify_header() already. If some
3933          * new magic introduced in the future, then layout LFSCK needs to
3934          * be updated also. */
3935         magic = le32_to_cpu(lmm->lmm_magic);
3936         if (magic == LOV_MAGIC_V1) {
3937                 objs = &(lmm->lmm_objects[0]);
3938         } else {
3939                 LASSERT(magic == LOV_MAGIC_V3);
3940                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3941         }
3942
3943         fid_to_ostid(cfid, oi);
3944         count = le16_to_cpu(lmm->lmm_stripe_count);
3945         for (i = 0; i < count; i++, objs++) {
3946                 struct ost_id oi2;
3947
3948                 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
3949                 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
3950                         GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
3951         }
3952
3953         GOTO(unlock, rc = -ENODATA);
3954
3955 unlock:
3956         dt_read_unlock(env, obj);
3957         lu_object_put(env, &obj->do_lu);
3958
3959         return rc;
3960 }
3961
3962 /*
3963  * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
3964  * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
3965  * layout EA from MDT to OST. On one hand, the OST no need to understand
3966  * the layout EA structure; on the other hand, it may cause trouble when
3967  * transfer large layout EA from MDT to OST via normal OUT RPC.
3968  *
3969  * \ret > 0: unrecognized stripe
3970  * \ret = 0: recognized stripe
3971  * \ret < 0: other failures
3972  */
3973 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
3974                                           struct lfsck_component *com,
3975                                           struct lu_fid *cfid,
3976                                           struct lu_fid *pfid)
3977 {
3978         struct lfsck_instance    *lfsck  = com->lc_lfsck;
3979         struct obd_device        *obd    = lfsck->li_obd;
3980         struct seq_server_site   *ss     =
3981                         lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
3982         struct obd_export        *exp    = NULL;
3983         struct ptlrpc_request    *req    = NULL;
3984         struct lfsck_request     *lr;
3985         struct lu_seq_range       range  = { 0 };
3986         int                       rc     = 0;
3987         ENTRY;
3988
3989         if (unlikely(fid_is_idif(pfid)))
3990                 RETURN(1);
3991
3992         fld_range_set_any(&range);
3993         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), &range);
3994         if (rc != 0)
3995                 RETURN(rc == -ENOENT ? 1 : rc);
3996
3997         if (unlikely(!fld_range_is_mdt(&range)))
3998                 RETURN(1);
3999
4000         exp = lustre_find_lwp_by_index(obd->obd_name, range.lsr_index);
4001         if (unlikely(exp == NULL))
4002                 RETURN(1);
4003
4004         if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
4005                 GOTO(out, rc = -EOPNOTSUPP);
4006
4007         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4008         if (req == NULL)
4009                 GOTO(out, rc = -ENOMEM);
4010
4011         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4012         if (rc != 0) {
4013                 ptlrpc_request_free(req);
4014
4015                 GOTO(out, rc);
4016         }
4017
4018         lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4019         memset(lr, 0, sizeof(*lr));
4020         lr->lr_event = LE_PAIRS_VERIFY;
4021         lr->lr_active = LT_LAYOUT;
4022         lr->lr_fid = *cfid; /* OST-object itself FID. */
4023         lr->lr_fid2 = *pfid; /* The claimed parent FID. */
4024
4025         ptlrpc_request_set_replen(req);
4026         rc = ptlrpc_queue_wait(req);
4027         ptlrpc_req_finished(req);
4028
4029         if (rc == -ENOENT || rc == -ENODATA)
4030                 rc = 1;
4031
4032         GOTO(out, rc);
4033
4034 out:
4035         if (exp != NULL)
4036                 class_export_put(exp);
4037
4038         return rc;
4039 }
4040
4041 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
4042                                           struct lfsck_component *com,
4043                                           struct lfsck_request *lr)
4044 {
4045         struct lfsck_thread_info        *info   = lfsck_env_info(env);
4046         struct filter_fid               *ff     = &info->lti_new_pfid;
4047         struct lu_buf                   *buf;
4048         struct dt_device                *dev    = com->lc_lfsck->li_bottom;
4049         struct dt_object                *obj;
4050         struct thandle                  *th     = NULL;
4051         int                              rc     = 0;
4052         ENTRY;
4053
4054         obj = lfsck_object_find_by_dev(env, dev, &lr->lr_fid);
4055         if (IS_ERR(obj))
4056                 RETURN(PTR_ERR(obj));
4057
4058         fid_cpu_to_le(&ff->ff_parent, &lr->lr_fid2);
4059         buf = lfsck_buf_get(env, ff, sizeof(*ff));
4060         dt_write_lock(env, obj, 0);
4061         if (unlikely(!dt_object_exists(obj)))
4062                 GOTO(unlock, rc = 0);
4063
4064         th = dt_trans_create(env, dev);
4065         if (IS_ERR(th))
4066                 GOTO(unlock, rc = PTR_ERR(th));
4067
4068         rc = dt_declare_xattr_set(env, obj, buf, XATTR_NAME_FID, 0, th);
4069         if (rc != 0)
4070                 GOTO(stop, rc);
4071
4072         rc = dt_trans_start_local(env, dev, th);
4073         if (rc != 0)
4074                 GOTO(stop, rc);
4075
4076         rc = dt_xattr_set(env, obj, buf, XATTR_NAME_FID, 0, th, BYPASS_CAPA);
4077
4078         GOTO(stop, rc);
4079
4080 stop:
4081         dt_trans_stop(env, dev, th);
4082
4083 unlock:
4084         dt_write_unlock(env, obj);
4085         lu_object_put(env, &obj->do_lu);
4086
4087         return rc;
4088 }
4089
4090 /* layout APIs */
4091
4092 static int lfsck_layout_reset(const struct lu_env *env,
4093                               struct lfsck_component *com, bool init)
4094 {
4095         struct lfsck_layout     *lo    = com->lc_file_ram;
4096         int                      rc;
4097
4098         down_write(&com->lc_sem);
4099         if (init) {
4100                 memset(lo, 0, com->lc_file_size);
4101         } else {
4102                 __u32 count = lo->ll_success_count;
4103                 __u64 last_time = lo->ll_time_last_complete;
4104
4105                 memset(lo, 0, com->lc_file_size);
4106                 lo->ll_success_count = count;
4107                 lo->ll_time_last_complete = last_time;
4108         }
4109
4110         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
4111         lo->ll_status = LS_INIT;
4112
4113         rc = lfsck_layout_store(env, com);
4114         up_write(&com->lc_sem);
4115
4116         return rc;
4117 }
4118
4119 static void lfsck_layout_fail(const struct lu_env *env,
4120                               struct lfsck_component *com, bool new_checked)
4121 {
4122         struct lfsck_layout *lo = com->lc_file_ram;
4123
4124         down_write(&com->lc_sem);
4125         if (new_checked)
4126                 com->lc_new_checked++;
4127         lfsck_layout_record_failure(env, com->lc_lfsck, lo);
4128         up_write(&com->lc_sem);
4129 }
4130
4131 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
4132                                           struct lfsck_component *com, bool init)
4133 {
4134         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4135         struct lfsck_layout             *lo      = com->lc_file_ram;
4136         struct lfsck_layout_master_data *llmd    = com->lc_data;
4137         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4138         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
4139         struct l_wait_info               lwi     = { 0 };
4140         int                              rc;
4141
4142         if (com->lc_new_checked == 0 && !init)
4143                 return 0;
4144
4145         l_wait_event(mthread->t_ctl_waitq,
4146                      list_empty(&llmd->llmd_req_list) ||
4147                      !thread_is_running(mthread) ||
4148                      thread_is_stopped(athread),
4149                      &lwi);
4150
4151         if (!thread_is_running(mthread) || thread_is_stopped(athread))
4152                 return 0;
4153
4154         down_write(&com->lc_sem);
4155         if (init) {
4156                 lo->ll_pos_latest_start = lfsck->li_pos_current.lp_oit_cookie;
4157         } else {
4158                 lo->ll_pos_last_checkpoint =
4159                                         lfsck->li_pos_current.lp_oit_cookie;
4160                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4161                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4162                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4163                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4164                 com->lc_new_checked = 0;
4165         }
4166
4167         rc = lfsck_layout_store(env, com);
4168         up_write(&com->lc_sem);
4169
4170         return rc;
4171 }
4172
4173 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
4174                                          struct lfsck_component *com, bool init)
4175 {
4176         struct lfsck_instance   *lfsck = com->lc_lfsck;
4177         struct lfsck_layout     *lo    = com->lc_file_ram;
4178         int                      rc;
4179
4180         if (com->lc_new_checked == 0 && !init)
4181                 return 0;
4182
4183         down_write(&com->lc_sem);
4184
4185         if (init) {
4186                 lo->ll_pos_latest_start = lfsck->li_pos_current.lp_oit_cookie;
4187         } else {
4188                 lo->ll_pos_last_checkpoint =
4189                                         lfsck->li_pos_current.lp_oit_cookie;
4190                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4191                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4192                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4193                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4194                 com->lc_new_checked = 0;
4195         }
4196
4197         rc = lfsck_layout_store(env, com);
4198
4199         up_write(&com->lc_sem);
4200
4201         return rc;
4202 }
4203
4204 static int lfsck_layout_prep(const struct lu_env *env,
4205                              struct lfsck_component *com,
4206                              struct lfsck_start *start)
4207 {
4208         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4209         struct lfsck_layout     *lo     = com->lc_file_ram;
4210         struct lfsck_position   *pos    = &com->lc_pos_start;
4211
4212         fid_zero(&pos->lp_dir_parent);
4213         pos->lp_dir_cookie = 0;
4214         if (lo->ll_status == LS_COMPLETED ||
4215             lo->ll_status == LS_PARTIAL ||
4216             /* To handle orphan, must scan from the beginning. */
4217             (start != NULL && start->ls_flags & LPF_ORPHAN)) {
4218                 int rc;
4219
4220                 rc = lfsck_layout_reset(env, com, false);
4221                 if (rc == 0)
4222                         rc = lfsck_set_param(env, lfsck, start, true);
4223
4224                 if (rc != 0)
4225                         return rc;
4226         }
4227
4228         down_write(&com->lc_sem);
4229         lo->ll_time_latest_start = cfs_time_current_sec();
4230         spin_lock(&lfsck->li_lock);
4231         if (lo->ll_flags & LF_SCANNED_ONCE) {
4232                 if (!lfsck->li_drop_dryrun ||
4233                     lo->ll_pos_first_inconsistent == 0) {
4234                         lo->ll_status = LS_SCANNING_PHASE2;
4235                         list_del_init(&com->lc_link);
4236                         list_add_tail(&com->lc_link,
4237                                       &lfsck->li_list_double_scan);
4238                         pos->lp_oit_cookie = 0;
4239                 } else {
4240                         int i;
4241
4242                         lo->ll_status = LS_SCANNING_PHASE1;
4243                         lo->ll_run_time_phase1 = 0;
4244                         lo->ll_run_time_phase2 = 0;
4245                         lo->ll_objs_checked_phase1 = 0;
4246                         lo->ll_objs_checked_phase2 = 0;
4247                         lo->ll_objs_failed_phase1 = 0;
4248                         lo->ll_objs_failed_phase2 = 0;
4249                         for (i = 0; i < LLIT_MAX; i++)
4250                                 lo->ll_objs_repaired[i] = 0;
4251
4252                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4253                         fid_zero(&com->lc_fid_latest_scanned_phase2);
4254                 }
4255         } else {
4256                 lo->ll_status = LS_SCANNING_PHASE1;
4257                 if (!lfsck->li_drop_dryrun ||
4258                     lo->ll_pos_first_inconsistent == 0)
4259                         pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
4260                 else
4261                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4262         }
4263         spin_unlock(&lfsck->li_lock);
4264         up_write(&com->lc_sem);
4265
4266         return 0;
4267 }
4268
4269 static int lfsck_layout_slave_prep(const struct lu_env *env,
4270                                    struct lfsck_component *com,
4271                                    struct lfsck_start_param *lsp)
4272 {
4273         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4274         struct lfsck_start              *start  = lsp->lsp_start;
4275         int                              rc;
4276
4277         rc = lfsck_layout_prep(env, com, start);
4278         if (rc != 0 || !lsp->lsp_index_valid)
4279                 return rc;
4280
4281         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
4282         if (rc == 0 && start != NULL && start->ls_flags & LPF_ORPHAN) {
4283                 LASSERT(!llsd->llsd_rbtree_valid);
4284
4285                 write_lock(&llsd->llsd_rb_lock);
4286                 rc = lfsck_rbtree_setup(env, com);
4287                 write_unlock(&llsd->llsd_rb_lock);
4288         }
4289
4290         return rc;
4291 }
4292
4293 static int lfsck_layout_master_prep(const struct lu_env *env,
4294                                     struct lfsck_component *com,
4295                                     struct lfsck_start_param *lsp)
4296 {
4297         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4298         struct lfsck_layout_master_data *llmd    = com->lc_data;
4299         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4300         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
4301         struct lfsck_thread_args        *lta;
4302         long                             rc;
4303         ENTRY;
4304
4305         rc = lfsck_layout_prep(env, com, lsp->lsp_start);
4306         if (rc != 0)
4307                 RETURN(rc);
4308
4309         llmd->llmd_assistant_status = 0;
4310         llmd->llmd_post_result = 0;
4311         llmd->llmd_to_post = 0;
4312         llmd->llmd_to_double_scan = 0;
4313         llmd->llmd_in_double_scan = 0;
4314         llmd->llmd_exit = 0;
4315         thread_set_flags(athread, 0);
4316
4317         lta = lfsck_thread_args_init(lfsck, com, lsp);
4318         if (IS_ERR(lta))
4319                 RETURN(PTR_ERR(lta));
4320
4321         rc = PTR_ERR(kthread_run(lfsck_layout_assistant, lta, "lfsck_layout"));
4322         if (IS_ERR_VALUE(rc)) {
4323                 CERROR("%s: Cannot start LFSCK layout assistant thread: "
4324                        "rc = %ld\n", lfsck_lfsck2name(lfsck), rc);
4325                 lfsck_thread_args_fini(lta);
4326         } else {
4327                 struct l_wait_info lwi = { 0 };
4328
4329                 l_wait_event(mthread->t_ctl_waitq,
4330                              thread_is_running(athread) ||
4331                              thread_is_stopped(athread),
4332                              &lwi);
4333                 if (unlikely(!thread_is_running(athread)))
4334                         rc = llmd->llmd_assistant_status;
4335                 else
4336                         rc = 0;
4337         }
4338
4339         RETURN(rc);
4340 }
4341
4342 /* Pre-fetch the attribute for each stripe in the given layout EA. */
4343 static int lfsck_layout_scan_stripes(const struct lu_env *env,
4344                                      struct lfsck_component *com,
4345                                      struct dt_object *parent,
4346                                      struct lov_mds_md_v1 *lmm)
4347 {
4348         struct lfsck_thread_info        *info    = lfsck_env_info(env);
4349         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4350         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
4351         struct lfsck_layout             *lo      = com->lc_file_ram;
4352         struct lfsck_layout_master_data *llmd    = com->lc_data;
4353         struct lfsck_layout_object      *llo     = NULL;
4354         struct lov_ost_data_v1          *objs;
4355         struct lfsck_tgt_descs          *ltds    = &lfsck->li_ost_descs;
4356         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4357         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
4358                 struct l_wait_info       lwi     = { 0 };
4359         struct lu_buf                   *buf;
4360         int                              rc      = 0;
4361         int                              i;
4362         __u32                            magic;
4363         __u16                            count;
4364         __u16                            gen;
4365         ENTRY;
4366
4367         buf = lfsck_buf_get(env, &info->lti_old_pfid,
4368                             sizeof(struct filter_fid_old));
4369         count = le16_to_cpu(lmm->lmm_stripe_count);
4370         gen = le16_to_cpu(lmm->lmm_layout_gen);
4371         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
4372          * been verified in lfsck_layout_verify_header() already. If some
4373          * new magic introduced in the future, then layout LFSCK needs to
4374          * be updated also. */
4375         magic = le32_to_cpu(lmm->lmm_magic);
4376         if (magic == LOV_MAGIC_V1) {
4377                 objs = &(lmm->lmm_objects[0]);
4378         } else {
4379                 LASSERT(magic == LOV_MAGIC_V3);
4380                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4381         }
4382
4383         for (i = 0; i < count; i++, objs++) {
4384                 struct lu_fid           *fid    = &info->lti_fid;
4385                 struct ost_id           *oi     = &info->lti_oi;
4386                 struct lfsck_layout_req *llr;
4387                 struct lfsck_tgt_desc   *tgt    = NULL;
4388                 struct dt_object        *cobj   = NULL;
4389                 __u32                    index  =
4390                                         le32_to_cpu(objs->l_ost_idx);
4391                 bool                     wakeup = false;
4392
4393                 if (is_dummy_lov_ost_data(objs))
4394                         continue;
4395
4396                 l_wait_event(mthread->t_ctl_waitq,
4397                              bk->lb_async_windows == 0 ||
4398                              llmd->llmd_prefetched < bk->lb_async_windows ||
4399                              !thread_is_running(mthread) ||
4400                              thread_is_stopped(athread),
4401                              &lwi);
4402
4403                 if (unlikely(!thread_is_running(mthread)) ||
4404                              thread_is_stopped(athread))
4405                         GOTO(out, rc = 0);
4406
4407                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4408                 ostid_to_fid(fid, oi, index);
4409                 tgt = lfsck_tgt_get(ltds, index);
4410                 if (unlikely(tgt == NULL)) {
4411                         CERROR("%s: Cannot talk with OST %x which did not join "
4412                                "the layout LFSCK.\n",
4413                                lfsck_lfsck2name(lfsck), index);
4414                         lo->ll_flags |= LF_INCOMPLETE;
4415                         goto next;
4416                 }
4417
4418                 cobj = lfsck_object_find_by_dev(env, tgt->ltd_tgt, fid);
4419                 if (IS_ERR(cobj)) {
4420                         rc = PTR_ERR(cobj);
4421                         goto next;
4422                 }
4423
4424                 rc = dt_declare_attr_get(env, cobj, BYPASS_CAPA);
4425                 if (rc != 0)
4426                         goto next;
4427
4428                 rc = dt_declare_xattr_get(env, cobj, buf, XATTR_NAME_FID,
4429                                           BYPASS_CAPA);
4430                 if (rc != 0)
4431                         goto next;
4432
4433                 if (llo == NULL) {
4434                         llo = lfsck_layout_object_init(env, parent, gen);
4435                         if (IS_ERR(llo)) {
4436                                 rc = PTR_ERR(llo);
4437                                 goto next;
4438                         }
4439                 }
4440
4441                 llr = lfsck_layout_req_init(llo, cobj, index, i);
4442                 if (IS_ERR(llr)) {
4443                         rc = PTR_ERR(llr);
4444                         goto next;
4445                 }
4446
4447                 cobj = NULL;
4448                 spin_lock(&llmd->llmd_lock);
4449                 if (llmd->llmd_assistant_status < 0) {
4450                         spin_unlock(&llmd->llmd_lock);
4451                         lfsck_layout_req_fini(env, llr);
4452                         lfsck_tgt_put(tgt);
4453                         RETURN(llmd->llmd_assistant_status);
4454                 }
4455
4456                 list_add_tail(&llr->llr_list, &llmd->llmd_req_list);
4457                 if (llmd->llmd_prefetched == 0)
4458                         wakeup = true;
4459
4460                 llmd->llmd_prefetched++;
4461                 spin_unlock(&llmd->llmd_lock);
4462                 if (wakeup)
4463                         wake_up_all(&athread->t_ctl_waitq);
4464
4465 next:
4466                 down_write(&com->lc_sem);
4467                 com->lc_new_checked++;
4468                 if (rc < 0)
4469                         lfsck_layout_record_failure(env, lfsck, lo);
4470                 up_write(&com->lc_sem);
4471
4472                 if (cobj != NULL && !IS_ERR(cobj))
4473                         lu_object_put(env, &cobj->do_lu);
4474
4475                 if (likely(tgt != NULL))
4476                         lfsck_tgt_put(tgt);
4477
4478                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4479                         GOTO(out, rc);
4480         }
4481
4482         GOTO(out, rc = 0);
4483
4484 out:
4485         if (llo != NULL && !IS_ERR(llo))
4486                 lfsck_layout_object_put(env, llo);
4487
4488         return rc;
4489 }
4490
4491 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
4492  * the OST-object's attribute and generate an structure lfsck_layout_req on the
4493  * list ::llmd_req_list.
4494  *
4495  * For each request on above list, the lfsck_layout_assistant thread compares
4496  * the OST side attribute with local attribute, if inconsistent, then repair it.
4497  *
4498  * All above processing is async mode with pipeline. */
4499 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
4500                                         struct lfsck_component *com,
4501                                         struct dt_object *obj)
4502 {
4503         struct lfsck_thread_info        *info   = lfsck_env_info(env);
4504         struct ost_id                   *oi     = &info->lti_oi;
4505         struct lfsck_layout             *lo     = com->lc_file_ram;
4506         struct lfsck_layout_master_data *llmd   = com->lc_data;
4507         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4508         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
4509         struct thandle                  *handle = NULL;
4510         struct lu_buf                   *buf    = &info->lti_big_buf;
4511         struct lov_mds_md_v1            *lmm    = NULL;
4512         struct dt_device                *dev    = lfsck->li_bottom;
4513         struct lustre_handle             lh     = { 0 };
4514         ssize_t                          buflen = buf->lb_len;
4515         int                              rc     = 0;
4516         bool                             locked = false;
4517         bool                             stripe = false;
4518         ENTRY;
4519
4520         if (!S_ISREG(lfsck_object_type(obj)))
4521                 GOTO(out, rc = 0);
4522
4523         if (llmd->llmd_assistant_status < 0)
4524                 GOTO(out, rc = -ESRCH);
4525
4526         fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
4527         lmm_oi_cpu_to_le(oi, oi);
4528         dt_read_lock(env, obj, 0);
4529         locked = true;
4530
4531 again:
4532         rc = lfsck_layout_get_lovea(env, obj, buf, &buflen);
4533         if (rc <= 0)
4534                 GOTO(out, rc);
4535
4536         buf->lb_len = rc;
4537         lmm = buf->lb_buf;
4538         rc = lfsck_layout_verify_header(lmm);
4539         /* If the LOV EA crashed, then it is possible to be rebuilt later
4540          * when handle orphan OST-objects. */
4541         if (rc != 0)
4542                 GOTO(out, rc);
4543
4544         if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0)
4545                 GOTO(out, stripe = true);
4546
4547         /* Inconsistent lmm_oi, should be repaired. */
4548         CDEBUG(D_LFSCK, "Repair bad lmm_oi for "DFID"\n",
4549                PFID(lfsck_dto2fid(obj)));
4550
4551         if (bk->lb_param & LPF_DRYRUN) {
4552                 down_write(&com->lc_sem);
4553                 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4554                 up_write(&com->lc_sem);
4555
4556                 GOTO(out, stripe = true);
4557         }
4558
4559         if (!lustre_handle_is_used(&lh)) {
4560                 dt_read_unlock(env, obj);
4561                 locked = false;
4562                 buf->lb_len = buflen;
4563                 rc = lfsck_layout_lock(env, com, obj, &lh,
4564                                        MDS_INODELOCK_LAYOUT |
4565                                        MDS_INODELOCK_XATTR);
4566                 if (rc != 0)
4567                         GOTO(out, rc);
4568
4569                 handle = dt_trans_create(env, dev);
4570                 if (IS_ERR(handle))
4571                         GOTO(out, rc = PTR_ERR(handle));
4572
4573                 rc = dt_declare_xattr_set(env, obj, buf, XATTR_NAME_LOV,
4574                                           LU_XATTR_REPLACE, handle);
4575                 if (rc != 0)
4576                         GOTO(out, rc);
4577
4578                 rc = dt_trans_start_local(env, dev, handle);
4579                 if (rc != 0)
4580                         GOTO(out, rc);
4581
4582                 dt_write_lock(env, obj, 0);
4583                 locked = true;
4584
4585                 goto again;
4586         }
4587
4588         lmm->lmm_oi = *oi;
4589         rc = dt_xattr_set(env, obj, buf, XATTR_NAME_LOV,
4590                           LU_XATTR_REPLACE, handle, BYPASS_CAPA);
4591         if (rc != 0)
4592                 GOTO(out, rc);
4593
4594         down_write(&com->lc_sem);
4595         lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4596         up_write(&com->lc_sem);
4597
4598         GOTO(out, stripe = true);
4599
4600 out:
4601         if (locked) {
4602                 if (lustre_handle_is_used(&lh))
4603                         dt_write_unlock(env, obj);
4604                 else
4605                         dt_read_unlock(env, obj);
4606         }
4607
4608         if (handle != NULL && !IS_ERR(handle))
4609                 dt_trans_stop(env, dev, handle);
4610
4611         lfsck_layout_unlock(&lh);
4612         if (stripe) {
4613                 rc = lfsck_layout_scan_stripes(env, com, obj, lmm);
4614         } else {
4615                 down_write(&com->lc_sem);
4616                 com->lc_new_checked++;
4617                 if (rc < 0)
4618                         lfsck_layout_record_failure(env, lfsck, lo);
4619                 up_write(&com->lc_sem);
4620         }
4621         buf->lb_len = buflen;
4622
4623         return rc;
4624 }
4625
4626 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
4627                                        struct lfsck_component *com,
4628                                        struct dt_object *obj)
4629 {
4630         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4631         struct lfsck_layout             *lo     = com->lc_file_ram;
4632         const struct lu_fid             *fid    = lfsck_dto2fid(obj);
4633         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
4634         struct lfsck_layout_seq         *lls;
4635         __u64                            seq;
4636         __u64                            oid;
4637         int                              rc;
4638         ENTRY;
4639
4640         LASSERT(llsd != NULL);
4641
4642         lfsck_rbtree_update_bitmap(env, com, fid, false);
4643
4644         down_write(&com->lc_sem);
4645         if (fid_is_idif(fid))
4646                 seq = 0;
4647         else if (!fid_is_norm(fid) ||
4648                  !fid_is_for_ostobj(env, lfsck->li_next, obj, fid))
4649                 GOTO(unlock, rc = 0);
4650         else
4651                 seq = fid_seq(fid);
4652         com->lc_new_checked++;
4653
4654         lls = lfsck_layout_seq_lookup(llsd, seq);
4655         if (lls == NULL) {
4656                 OBD_ALLOC_PTR(lls);
4657                 if (unlikely(lls == NULL))
4658                         GOTO(unlock, rc = -ENOMEM);
4659
4660                 INIT_LIST_HEAD(&lls->lls_list);
4661                 lls->lls_seq = seq;
4662                 rc = lfsck_layout_lastid_load(env, com, lls);
4663                 if (rc != 0) {
4664                         lo->ll_objs_failed_phase1++;
4665                         OBD_FREE_PTR(lls);
4666                         GOTO(unlock, rc);
4667                 }
4668
4669                 lfsck_layout_seq_insert(llsd, lls);
4670         }
4671
4672         if (unlikely(fid_is_last_id(fid)))
4673                 GOTO(unlock, rc = 0);
4674
4675         oid = fid_oid(fid);
4676         if (oid > lls->lls_lastid_known)
4677                 lls->lls_lastid_known = oid;
4678
4679         if (oid > lls->lls_lastid) {
4680                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
4681                         /* OFD may create new objects during LFSCK scanning. */
4682                         rc = lfsck_layout_lastid_reload(env, com, lls);
4683                         if (unlikely(rc != 0))
4684                                 CWARN("%s: failed to reload LAST_ID for "LPX64
4685                                       ": rc = %d\n",
4686                                       lfsck_lfsck2name(com->lc_lfsck),
4687                                       lls->lls_seq, rc);
4688                         if (oid <= lls->lls_lastid)
4689                                 GOTO(unlock, rc = 0);
4690
4691                         LASSERT(lfsck->li_out_notify != NULL);
4692
4693                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4694                                              LE_LASTID_REBUILDING);
4695                         lo->ll_flags |= LF_CRASHED_LASTID;
4696                 }
4697
4698                 lls->lls_lastid = oid;
4699                 lls->lls_dirty = 1;
4700         }
4701
4702         GOTO(unlock, rc = 0);
4703
4704 unlock:
4705         up_write(&com->lc_sem);
4706
4707         return rc;
4708 }
4709
4710 static int lfsck_layout_exec_dir(const struct lu_env *env,
4711                                  struct lfsck_component *com,
4712                                  struct dt_object *obj,
4713                                  struct lu_dirent *ent)
4714 {
4715         return 0;
4716 }
4717
4718 static int lfsck_layout_master_post(const struct lu_env *env,
4719                                     struct lfsck_component *com,
4720                                     int result, bool init)
4721 {
4722         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4723         struct lfsck_layout             *lo      = com->lc_file_ram;
4724         struct lfsck_layout_master_data *llmd    = com->lc_data;
4725         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4726         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
4727         struct l_wait_info               lwi     = { 0 };
4728         int                              rc;
4729         ENTRY;
4730
4731
4732         llmd->llmd_post_result = result;
4733         llmd->llmd_to_post = 1;
4734         if (llmd->llmd_post_result <= 0)
4735                 llmd->llmd_exit = 1;
4736
4737         wake_up_all(&athread->t_ctl_waitq);
4738         l_wait_event(mthread->t_ctl_waitq,
4739                      (result > 0 && list_empty(&llmd->llmd_req_list)) ||
4740                      thread_is_stopped(athread),
4741                      &lwi);
4742
4743         if (llmd->llmd_assistant_status < 0)
4744                 result = llmd->llmd_assistant_status;
4745
4746         down_write(&com->lc_sem);
4747         spin_lock(&lfsck->li_lock);
4748         /* When LFSCK failed, there may be some prefetched objects those are
4749          * not been processed yet, we do not know the exactly position, then
4750          * just restart from last check-point next time. */
4751         if (!init && !llmd->llmd_exit)
4752                 lo->ll_pos_last_checkpoint =
4753                                         lfsck->li_pos_current.lp_oit_cookie;
4754
4755         if (result > 0) {
4756                 lo->ll_status = LS_SCANNING_PHASE2;
4757                 lo->ll_flags |= LF_SCANNED_ONCE;
4758                 lo->ll_flags &= ~LF_UPGRADE;
4759                 list_del_init(&com->lc_link);
4760                 list_add_tail(&com->lc_link, &lfsck->li_list_double_scan);
4761         } else if (result == 0) {
4762                 lo->ll_status = lfsck->li_status;
4763                 if (lo->ll_status == 0)
4764                         lo->ll_status = LS_STOPPED;
4765                 if (lo->ll_status != LS_PAUSED) {
4766                         list_del_init(&com->lc_link);
4767                         list_add_tail(&com->lc_link, &lfsck->li_list_idle);
4768                 }
4769         } else {
4770                 lo->ll_status = LS_FAILED;
4771                 list_del_init(&com->lc_link);
4772                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
4773         }
4774         spin_unlock(&lfsck->li_lock);
4775
4776         if (!init) {
4777                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4778                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4779                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4780                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4781                 com->lc_new_checked = 0;
4782         }
4783
4784         rc = lfsck_layout_store(env, com);
4785         up_write(&com->lc_sem);
4786
4787         RETURN(rc);
4788 }
4789
4790 static int lfsck_layout_slave_post(const struct lu_env *env,
4791                                    struct lfsck_component *com,
4792                                    int result, bool init)
4793 {
4794         struct lfsck_instance   *lfsck = com->lc_lfsck;
4795         struct lfsck_layout     *lo    = com->lc_file_ram;
4796         int                      rc;
4797         bool                     done  = false;
4798
4799         rc = lfsck_layout_lastid_store(env, com);
4800         if (rc != 0)
4801                 result = rc;
4802
4803         LASSERT(lfsck->li_out_notify != NULL);
4804
4805         down_write(&com->lc_sem);
4806
4807         spin_lock(&lfsck->li_lock);
4808         if (!init)
4809                 lo->ll_pos_last_checkpoint =
4810                                         lfsck->li_pos_current.lp_oit_cookie;
4811         if (result > 0) {
4812                 lo->ll_status = LS_SCANNING_PHASE2;
4813                 lo->ll_flags |= LF_SCANNED_ONCE;
4814                 if (lo->ll_flags & LF_CRASHED_LASTID) {
4815                         done = true;
4816                         lo->ll_flags &= ~LF_CRASHED_LASTID;
4817                 }
4818                 lo->ll_flags &= ~LF_UPGRADE;
4819                 list_del_init(&com->lc_link);
4820                 list_add_tail(&com->lc_link, &lfsck->li_list_double_scan);
4821         } else if (result == 0) {
4822                 lo->ll_status = lfsck->li_status;
4823                 if (lo->ll_status == 0)
4824                         lo->ll_status = LS_STOPPED;
4825                 if (lo->ll_status != LS_PAUSED) {
4826                         list_del_init(&com->lc_link);
4827                         list_add_tail(&com->lc_link, &lfsck->li_list_idle);
4828                 }
4829         } else {
4830                 lo->ll_status = LS_FAILED;
4831                 list_del_init(&com->lc_link);
4832                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
4833         }
4834         spin_unlock(&lfsck->li_lock);
4835
4836         if (done)
4837                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4838                                      LE_LASTID_REBUILT);
4839
4840         if (!init) {
4841                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4842                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4843                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4844                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4845                 com->lc_new_checked = 0;
4846         }
4847
4848         rc = lfsck_layout_store(env, com);
4849
4850         up_write(&com->lc_sem);
4851
4852         lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
4853
4854         if (result <= 0)
4855                 lfsck_rbtree_cleanup(env, com);
4856
4857         return rc;
4858 }
4859
4860 static int lfsck_layout_dump(const struct lu_env *env,
4861                              struct lfsck_component *com, char *buf, int len)
4862 {
4863         struct lfsck_instance   *lfsck = com->lc_lfsck;
4864         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
4865         struct lfsck_layout     *lo    = com->lc_file_ram;
4866         int                      save  = len;
4867         int                      ret   = -ENOSPC;
4868         int                      rc;
4869
4870         down_read(&com->lc_sem);
4871         rc = snprintf(buf, len,
4872                       "name: lfsck_layout\n"
4873                       "magic: %#x\n"
4874                       "version: %d\n"
4875                       "status: %s\n",
4876                       lo->ll_magic,
4877                       bk->lb_version,
4878                       lfsck_status2names(lo->ll_status));
4879         if (rc <= 0)
4880                 goto out;
4881
4882         buf += rc;
4883         len -= rc;
4884         rc = lfsck_bits_dump(&buf, &len, lo->ll_flags, lfsck_flags_names,
4885                              "flags");
4886         if (rc < 0)
4887                 goto out;
4888
4889         rc = lfsck_bits_dump(&buf, &len, bk->lb_param, lfsck_param_names,
4890                              "param");
4891         if (rc < 0)
4892                 goto out;
4893
4894         rc = lfsck_time_dump(&buf, &len, lo->ll_time_last_complete,
4895                              "time_since_last_completed");
4896         if (rc < 0)
4897                 goto out;
4898
4899         rc = lfsck_time_dump(&buf, &len, lo->ll_time_latest_start,
4900                              "time_since_latest_start");
4901         if (rc < 0)
4902                 goto out;
4903
4904         rc = lfsck_time_dump(&buf, &len, lo->ll_time_last_checkpoint,
4905                              "time_since_last_checkpoint");
4906         if (rc < 0)
4907                 goto out;
4908
4909         rc = snprintf(buf, len,
4910                       "latest_start_position: "LPU64"\n"
4911                       "last_checkpoint_position: "LPU64"\n"
4912                       "first_failure_position: "LPU64"\n",
4913                       lo->ll_pos_latest_start,
4914                       lo->ll_pos_last_checkpoint,
4915                       lo->ll_pos_first_inconsistent);
4916         if (rc <= 0)
4917                 goto out;
4918
4919         buf += rc;
4920         len -= rc;
4921
4922         rc = snprintf(buf, len,
4923                       "success_count: %u\n"
4924                       "repaired_dangling: "LPU64"\n"
4925                       "repaired_unmatched_pair: "LPU64"\n"
4926                       "repaired_multiple_referenced: "LPU64"\n"
4927                       "repaired_orphan: "LPU64"\n"
4928                       "repaired_inconsistent_owner: "LPU64"\n"
4929                       "repaired_others: "LPU64"\n"
4930                       "skipped: "LPU64"\n"
4931                       "failed_phase1: "LPU64"\n"
4932                       "failed_phase2: "LPU64"\n",
4933                       lo->ll_success_count,
4934                       lo->ll_objs_repaired[LLIT_DANGLING - 1],
4935                       lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
4936                       lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
4937                       lo->ll_objs_repaired[LLIT_ORPHAN - 1],
4938                       lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
4939                       lo->ll_objs_repaired[LLIT_OTHERS - 1],
4940                       lo->ll_objs_skipped,
4941                       lo->ll_objs_failed_phase1,
4942                       lo->ll_objs_failed_phase2);
4943         if (rc <= 0)
4944                 goto out;
4945
4946         buf += rc;
4947         len -= rc;
4948
4949         if (lo->ll_status == LS_SCANNING_PHASE1) {
4950                 __u64 pos;
4951                 const struct dt_it_ops *iops;
4952                 cfs_duration_t duration = cfs_time_current() -
4953                                           lfsck->li_time_last_checkpoint;
4954                 __u64 checked = lo->ll_objs_checked_phase1 +
4955                                 com->lc_new_checked;
4956                 __u64 speed = checked;
4957                 __u64 new_checked = com->lc_new_checked * HZ;
4958                 __u32 rtime = lo->ll_run_time_phase1 +
4959                               cfs_duration_sec(duration + HALF_SEC);
4960
4961                 if (duration != 0)
4962                         do_div(new_checked, duration);
4963                 if (rtime != 0)
4964                         do_div(speed, rtime);
4965                 rc = snprintf(buf, len,
4966                               "checked_phase1: "LPU64"\n"
4967                               "checked_phase2: "LPU64"\n"
4968                               "run_time_phase1: %u seconds\n"
4969                               "run_time_phase2: %u seconds\n"
4970                               "average_speed_phase1: "LPU64" items/sec\n"
4971                               "average_speed_phase2: N/A\n"
4972                               "real-time_speed_phase1: "LPU64" items/sec\n"
4973                               "real-time_speed_phase2: N/A\n",
4974                               checked,
4975                               lo->ll_objs_checked_phase2,
4976                               rtime,
4977                               lo->ll_run_time_phase2,
4978                               speed,
4979                               new_checked);
4980                 if (rc <= 0)
4981                         goto out;
4982
4983                 buf += rc;
4984                 len -= rc;
4985
4986                 LASSERT(lfsck->li_di_oit != NULL);
4987
4988                 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
4989
4990                 /* The low layer otable-based iteration position may NOT
4991                  * exactly match the layout-based directory traversal
4992                  * cookie. Generally, it is not a serious issue. But the
4993                  * caller should NOT make assumption on that. */
4994                 pos = iops->store(env, lfsck->li_di_oit);
4995                 if (!lfsck->li_current_oit_processed)
4996                         pos--;
4997                 rc = snprintf(buf, len, "current_position: "LPU64"\n", pos);
4998                 if (rc <= 0)
4999                         goto out;
5000
5001                 buf += rc;
5002                 len -= rc;
5003         } else if (lo->ll_status == LS_SCANNING_PHASE2) {
5004                 cfs_duration_t duration = cfs_time_current() -
5005                                           lfsck->li_time_last_checkpoint;
5006                 __u64 checked = lo->ll_objs_checked_phase2 +
5007                                 com->lc_new_checked;
5008                 __u64 speed1 = lo->ll_objs_checked_phase1;
5009                 __u64 speed2 = checked;
5010                 __u64 new_checked = com->lc_new_checked * HZ;
5011                 __u32 rtime = lo->ll_run_time_phase2 +
5012                               cfs_duration_sec(duration + HALF_SEC);
5013
5014                 if (duration != 0)
5015                         do_div(new_checked, duration);
5016                 if (lo->ll_run_time_phase1 != 0)
5017                         do_div(speed1, lo->ll_run_time_phase1);
5018                 if (rtime != 0)
5019                         do_div(speed2, rtime);
5020                 rc = snprintf(buf, len,
5021                               "checked_phase1: "LPU64"\n"
5022                               "checked_phase2: "LPU64"\n"
5023                               "run_time_phase1: %u seconds\n"
5024                               "run_time_phase2: %u seconds\n"
5025                               "average_speed_phase1: "LPU64" items/sec\n"
5026                               "average_speed_phase2: "LPU64" items/sec\n"
5027                               "real-time_speed_phase1: N/A\n"
5028                               "real-time_speed_phase2: "LPU64" items/sec\n"
5029                               "current_position: "DFID"\n",
5030                               lo->ll_objs_checked_phase1,
5031                               checked,
5032                               lo->ll_run_time_phase1,
5033                               rtime,
5034                               speed1,
5035                               speed2,
5036                               new_checked,
5037                               PFID(&com->lc_fid_latest_scanned_phase2));
5038                 if (rc <= 0)
5039                         goto out;
5040
5041                 buf += rc;
5042                 len -= rc;
5043         } else {
5044                 __u64 speed1 = lo->ll_objs_checked_phase1;
5045                 __u64 speed2 = lo->ll_objs_checked_phase2;
5046
5047                 if (lo->ll_run_time_phase1 != 0)
5048                         do_div(speed1, lo->ll_run_time_phase1);
5049                 if (lo->ll_run_time_phase2 != 0)
5050                         do_div(speed2, lo->ll_run_time_phase2);
5051                 rc = snprintf(buf, len,
5052                               "checked_phase1: "LPU64"\n"
5053                               "checked_phase2: "LPU64"\n"
5054                               "run_time_phase1: %u seconds\n"
5055                               "run_time_phase2: %u seconds\n"
5056                               "average_speed_phase1: "LPU64" items/sec\n"
5057                               "average_speed_phase2: "LPU64" objs/sec\n"
5058                               "real-time_speed_phase1: N/A\n"
5059                               "real-time_speed_phase2: N/A\n"
5060                               "current_position: N/A\n",
5061                               lo->ll_objs_checked_phase1,
5062                               lo->ll_objs_checked_phase2,
5063                               lo->ll_run_time_phase1,
5064                               lo->ll_run_time_phase2,
5065                               speed1,
5066                               speed2);
5067                 if (rc <= 0)
5068                         goto out;
5069
5070                 buf += rc;
5071                 len -= rc;
5072         }
5073         ret = save - len;
5074
5075 out:
5076         up_read(&com->lc_sem);
5077
5078         return ret;
5079 }
5080
5081 static int lfsck_layout_master_double_scan(const struct lu_env *env,
5082                                            struct lfsck_component *com)
5083 {
5084         struct lfsck_layout_master_data *llmd    = com->lc_data;
5085         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
5086         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
5087         struct lfsck_layout             *lo      = com->lc_file_ram;
5088         struct l_wait_info               lwi     = { 0 };
5089
5090         if (unlikely(lo->ll_status != LS_SCANNING_PHASE2))
5091                 return 0;
5092
5093         llmd->llmd_to_double_scan = 1;
5094         wake_up_all(&athread->t_ctl_waitq);
5095         l_wait_event(mthread->t_ctl_waitq,
5096                      llmd->llmd_in_double_scan ||
5097                      thread_is_stopped(athread),
5098                      &lwi);
5099         if (llmd->llmd_assistant_status < 0)
5100                 return llmd->llmd_assistant_status;
5101
5102         return 0;
5103 }
5104
5105 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
5106                                           struct lfsck_component *com)
5107 {
5108         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5109         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
5110         struct lfsck_layout             *lo     = com->lc_file_ram;
5111         struct ptlrpc_thread            *thread = &lfsck->li_thread;
5112         int                              rc;
5113         ENTRY;
5114
5115         if (unlikely(lo->ll_status != LS_SCANNING_PHASE2)) {
5116                 lfsck_rbtree_cleanup(env, com);
5117                 lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE, 0);
5118                 RETURN(0);
5119         }
5120
5121         atomic_inc(&lfsck->li_double_scan_count);
5122
5123         com->lc_new_checked = 0;
5124         com->lc_new_scanned = 0;
5125         com->lc_time_last_checkpoint = cfs_time_current();
5126         com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
5127                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
5128
5129         while (1) {
5130                 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(30),
5131                                                      NULL, NULL);
5132
5133                 rc = lfsck_layout_slave_query_master(env, com);
5134                 if (list_empty(&llsd->llsd_master_list)) {
5135                         if (unlikely(!thread_is_running(thread)))
5136                                 rc = 0;
5137                         else
5138                                 rc = 1;
5139
5140                         GOTO(done, rc);
5141                 }
5142
5143                 if (rc < 0)
5144                         GOTO(done, rc);
5145
5146                 rc = l_wait_event(thread->t_ctl_waitq,
5147                                   !thread_is_running(thread) ||
5148                                   list_empty(&llsd->llsd_master_list),
5149                                   &lwi);
5150                 if (unlikely(!thread_is_running(thread)))
5151                         GOTO(done, rc = 0);
5152
5153                 if (rc == -ETIMEDOUT)
5154                         continue;
5155
5156                 GOTO(done, rc = (rc < 0 ? rc : 1));
5157         }
5158
5159 done:
5160         rc = lfsck_layout_double_scan_result(env, com, rc);
5161
5162         lfsck_rbtree_cleanup(env, com);
5163         lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE, rc);
5164         if (atomic_dec_and_test(&lfsck->li_double_scan_count))
5165                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5166
5167         return rc;
5168 }
5169
5170 static void lfsck_layout_master_data_release(const struct lu_env *env,
5171                                              struct lfsck_component *com)
5172 {
5173         struct lfsck_layout_master_data *llmd   = com->lc_data;
5174         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5175         struct lfsck_tgt_descs          *ltds;
5176         struct lfsck_tgt_desc           *ltd;
5177         struct lfsck_tgt_desc           *next;
5178
5179         LASSERT(llmd != NULL);
5180         LASSERT(thread_is_init(&llmd->llmd_thread) ||
5181                 thread_is_stopped(&llmd->llmd_thread));
5182         LASSERT(list_empty(&llmd->llmd_req_list));
5183
5184         com->lc_data = NULL;
5185
5186         ltds = &lfsck->li_ost_descs;
5187         spin_lock(&ltds->ltd_lock);
5188         list_for_each_entry_safe(ltd, next, &llmd->llmd_ost_phase1_list,
5189                                  ltd_layout_phase_list) {
5190                 list_del_init(&ltd->ltd_layout_phase_list);
5191         }
5192         list_for_each_entry_safe(ltd, next, &llmd->llmd_ost_phase2_list,
5193                                  ltd_layout_phase_list) {
5194                 list_del_init(&ltd->ltd_layout_phase_list);
5195         }
5196         list_for_each_entry_safe(ltd, next, &llmd->llmd_ost_list,
5197                                  ltd_layout_list) {
5198                 list_del_init(&ltd->ltd_layout_list);
5199         }
5200         list_for_each_entry_safe(ltd, next, &llmd->llmd_mdt_phase1_list,
5201                                  ltd_layout_phase_list) {
5202                 list_del_init(&ltd->ltd_layout_phase_list);
5203         }
5204         list_for_each_entry_safe(ltd, next, &llmd->llmd_mdt_phase2_list,
5205                                  ltd_layout_phase_list) {
5206                 list_del_init(&ltd->ltd_layout_phase_list);
5207         }
5208         list_for_each_entry_safe(ltd, next, &llmd->llmd_mdt_list,
5209                                  ltd_layout_list) {
5210                 list_del_init(&ltd->ltd_layout_list);
5211         }
5212         spin_unlock(&ltds->ltd_lock);
5213
5214         OBD_FREE_PTR(llmd);
5215 }
5216
5217 static void lfsck_layout_slave_data_release(const struct lu_env *env,
5218                                             struct lfsck_component *com)
5219 {
5220         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5221         struct lfsck_layout_seq          *lls;
5222         struct lfsck_layout_seq          *next;
5223         struct lfsck_layout_slave_target *llst;
5224         struct lfsck_layout_slave_target *tmp;
5225
5226         LASSERT(llsd != NULL);
5227
5228         list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
5229                                      lls_list) {
5230                 list_del_init(&lls->lls_list);
5231                 lfsck_object_put(env, lls->lls_lastid_obj);
5232                 OBD_FREE_PTR(lls);
5233         }
5234
5235         list_for_each_entry_safe(llst, tmp, &llsd->llsd_master_list,
5236                                  llst_list) {
5237                 list_del_init(&llst->llst_list);
5238                 OBD_FREE_PTR(llst);
5239         }
5240
5241         lfsck_rbtree_cleanup(env, com);
5242         com->lc_data = NULL;
5243         OBD_FREE_PTR(llsd);
5244 }
5245
5246 static void lfsck_layout_master_quit(const struct lu_env *env,
5247                                      struct lfsck_component *com)
5248 {
5249         struct lfsck_layout_master_data *llmd    = com->lc_data;
5250         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
5251         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
5252         struct l_wait_info               lwi     = { 0 };
5253
5254         llmd->llmd_exit = 1;
5255         wake_up_all(&athread->t_ctl_waitq);
5256         l_wait_event(mthread->t_ctl_waitq,
5257                      thread_is_init(athread) ||
5258                      thread_is_stopped(athread),
5259                      &lwi);
5260 }
5261
5262 static void lfsck_layout_slave_quit(const struct lu_env *env,
5263                                     struct lfsck_component *com)
5264 {
5265         lfsck_rbtree_cleanup(env, com);
5266 }
5267
5268 static int lfsck_layout_master_in_notify(const struct lu_env *env,
5269                                          struct lfsck_component *com,
5270                                          struct lfsck_request *lr)
5271 {
5272         struct lfsck_instance           *lfsck = com->lc_lfsck;
5273         struct lfsck_layout             *lo    = com->lc_file_ram;
5274         struct lfsck_layout_master_data *llmd  = com->lc_data;
5275         struct lfsck_tgt_descs          *ltds;
5276         struct lfsck_tgt_desc           *ltd;
5277         bool                             fail  = false;
5278         ENTRY;
5279
5280         if (lr->lr_event == LE_PAIRS_VERIFY) {
5281                 int rc;
5282
5283                 rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
5284                                                      &lr->lr_fid2);
5285
5286                 RETURN(rc);
5287         }
5288
5289         if (lr->lr_event != LE_PHASE1_DONE &&
5290             lr->lr_event != LE_PHASE2_DONE &&
5291             lr->lr_event != LE_PEER_EXIT)
5292                 RETURN(-EINVAL);
5293
5294         if (lr->lr_flags & LEF_FROM_OST)
5295                 ltds = &lfsck->li_ost_descs;
5296         else
5297                 ltds = &lfsck->li_mdt_descs;
5298         spin_lock(&ltds->ltd_lock);
5299         ltd = LTD_TGT(ltds, lr->lr_index);
5300         if (ltd == NULL) {
5301                 spin_unlock(&ltds->ltd_lock);
5302
5303                 RETURN(-ENXIO);
5304         }
5305
5306         list_del_init(&ltd->ltd_layout_phase_list);
5307         switch (lr->lr_event) {
5308         case LE_PHASE1_DONE:
5309                 if (lr->lr_status <= 0) {
5310                         ltd->ltd_layout_done = 1;
5311                         list_del_init(&ltd->ltd_layout_list);
5312                         CWARN("%s: %s %x failed/stopped at phase1: rc = %d.\n",
5313                               lfsck_lfsck2name(lfsck),
5314                               (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
5315                               ltd->ltd_index, lr->lr_status);
5316                         lo->ll_flags |= LF_INCOMPLETE;
5317                         fail = true;
5318                         break;
5319                 }
5320
5321                 if (lr->lr_flags & LEF_FROM_OST) {
5322                         if (list_empty(&ltd->ltd_layout_list))
5323                                 list_add_tail(&ltd->ltd_layout_list,
5324                                               &llmd->llmd_ost_list);
5325                         list_add_tail(&ltd->ltd_layout_phase_list,
5326                                       &llmd->llmd_ost_phase2_list);
5327                 } else {
5328                         if (list_empty(&ltd->ltd_layout_list))
5329                                 list_add_tail(&ltd->ltd_layout_list,
5330                                               &llmd->llmd_mdt_list);
5331                         list_add_tail(&ltd->ltd_layout_phase_list,
5332                                       &llmd->llmd_mdt_phase2_list);
5333                 }
5334                 break;
5335         case LE_PHASE2_DONE:
5336                 ltd->ltd_layout_done = 1;
5337                 list_del_init(&ltd->ltd_layout_list);
5338                 break;
5339         case LE_PEER_EXIT:
5340                 fail = true;
5341                 ltd->ltd_layout_done = 1;
5342                 list_del_init(&ltd->ltd_layout_list);
5343                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
5344                         CWARN("%s: the peer %s %x exit layout LFSCK.\n",
5345                               lfsck_lfsck2name(lfsck),
5346                               (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
5347                               ltd->ltd_index);
5348                         lo->ll_flags |= LF_INCOMPLETE;
5349                 }
5350                 break;
5351         default:
5352                 break;
5353         }
5354         spin_unlock(&ltds->ltd_lock);
5355
5356         if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
5357                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5358
5359                 memset(stop, 0, sizeof(*stop));
5360                 stop->ls_status = lr->lr_status;
5361                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5362                 lfsck_stop(env, lfsck->li_bottom, stop);
5363         } else if (lfsck_layout_master_to_orphan(llmd)) {
5364                 wake_up_all(&llmd->llmd_thread.t_ctl_waitq);
5365         }
5366
5367         RETURN(0);
5368 }
5369
5370 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
5371                                         struct lfsck_component *com,
5372                                         struct lfsck_request *lr)
5373 {
5374         struct lfsck_instance            *lfsck = com->lc_lfsck;
5375         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5376         struct lfsck_layout_slave_target *llst;
5377         int                               rc;
5378         ENTRY;
5379
5380         switch (lr->lr_event) {
5381         case LE_FID_ACCESSED:
5382                 lfsck_rbtree_update_bitmap(env, com, &lr->lr_fid, true);
5383                 RETURN(0);
5384         case LE_CONDITIONAL_DESTROY:
5385                 rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
5386                 RETURN(rc);
5387         case LE_PAIRS_VERIFY: {
5388                 lr->lr_status = LPVS_INIT;
5389                 /* Firstly, if the MDT-object which is claimed via OST-object
5390                  * local stored PFID xattr recognizes the OST-object, then it
5391                  * must be that the client given PFID is wrong. */
5392                 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5393                                                     &lr->lr_fid3);
5394                 if (rc <= 0)
5395                         RETURN(0);
5396
5397                 lr->lr_status = LPVS_INCONSISTENT;
5398                 /* The OST-object local stored PFID xattr is stale. We need to
5399                  * check whether the MDT-object that is claimed via the client
5400                  * given PFID information recognizes the OST-object or not. If
5401                  * matches, then need to update the OST-object's PFID xattr. */
5402                 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5403                                                     &lr->lr_fid2);
5404                 /* For rc < 0 case:
5405                  * We are not sure whether the client given PFID information
5406                  * is correct or not, do nothing to avoid improper fixing.
5407                  *
5408                  * For rc > 0 case:
5409                  * The client given PFID information is also invalid, we can
5410                  * NOT fix the OST-object inconsistency.
5411                  */
5412                 if (rc != 0)
5413                         RETURN(rc);
5414
5415                 lr->lr_status = LPVS_INCONSISTENT_TOFIX;
5416                 rc = lfsck_layout_slave_repair_pfid(env, com, lr);
5417
5418                 RETURN(rc);
5419         }
5420         case LE_PHASE2_DONE:
5421         case LE_PEER_EXIT:
5422                 break;
5423         default:
5424                 RETURN(-EINVAL);
5425         }
5426
5427         llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
5428         if (llst == NULL)
5429                 RETURN(-ENXIO);
5430
5431         lfsck_layout_llst_put(llst);
5432         if (list_empty(&llsd->llsd_master_list))
5433                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5434
5435         if (lr->lr_event == LE_PEER_EXIT &&
5436             lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
5437                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5438
5439                 memset(stop, 0, sizeof(*stop));
5440                 stop->ls_status = lr->lr_status;
5441                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5442                 lfsck_stop(env, lfsck->li_bottom, stop);
5443         }
5444
5445         RETURN(0);
5446 }
5447
5448 static int lfsck_layout_query(const struct lu_env *env,
5449                               struct lfsck_component *com)
5450 {
5451         struct lfsck_layout *lo = com->lc_file_ram;
5452
5453         return lo->ll_status;
5454 }
5455
5456 static int lfsck_layout_master_stop_notify(const struct lu_env *env,
5457                                            struct lfsck_component *com,
5458                                            struct lfsck_tgt_descs *ltds,
5459                                            struct lfsck_tgt_desc *ltd,
5460                                            struct ptlrpc_request_set *set)
5461 {
5462         struct lfsck_thread_info          *info  = lfsck_env_info(env);
5463         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
5464         struct lfsck_request              *lr    = &info->lti_lr;
5465         struct lfsck_instance             *lfsck = com->lc_lfsck;
5466         int                                rc;
5467
5468         spin_lock(&ltds->ltd_lock);
5469         if (list_empty(&ltd->ltd_layout_list)) {
5470                 LASSERT(list_empty(&ltd->ltd_layout_phase_list));
5471                 spin_unlock(&ltds->ltd_lock);
5472
5473                 return 0;
5474         }
5475
5476         list_del_init(&ltd->ltd_layout_phase_list);
5477         list_del_init(&ltd->ltd_layout_list);
5478         spin_unlock(&ltds->ltd_lock);
5479
5480         memset(lr, 0, sizeof(*lr));
5481         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
5482         lr->lr_event = LE_PEER_EXIT;
5483         lr->lr_active = LT_LAYOUT;
5484         lr->lr_status = LS_CO_PAUSED;
5485         if (ltds == &lfsck->li_ost_descs)
5486                 lr->lr_flags = LEF_TO_OST;
5487
5488         laia->laia_com = com;
5489         laia->laia_ltds = ltds;
5490         atomic_inc(&ltd->ltd_ref);
5491         laia->laia_ltd = ltd;
5492         laia->laia_lr = lr;
5493         laia->laia_shared = 0;
5494
5495         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
5496                                  lfsck_layout_master_async_interpret,
5497                                  laia, LFSCK_NOTIFY);
5498         if (rc != 0) {
5499                 CERROR("%s: Fail to notify %s %x for co-stop: rc = %d\n",
5500                        lfsck_lfsck2name(lfsck),
5501                        (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
5502                        ltd->ltd_index, rc);
5503                 lfsck_tgt_put(ltd);
5504         }
5505
5506         return rc;
5507 }
5508
5509 /* with lfsck::li_lock held */
5510 static int lfsck_layout_slave_join(const struct lu_env *env,
5511                                    struct lfsck_component *com,
5512                                    struct lfsck_start_param *lsp)
5513 {
5514         struct lfsck_instance            *lfsck = com->lc_lfsck;
5515         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
5516         struct lfsck_layout_slave_target *llst;
5517         struct lfsck_start               *start = lsp->lsp_start;
5518         int                               rc    = 0;
5519         ENTRY;
5520
5521         if (!lsp->lsp_index_valid || start == NULL ||
5522             !(start->ls_flags & LPF_ALL_TGT) ||
5523             !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT))
5524                 RETURN(-EALREADY);
5525
5526         spin_unlock(&lfsck->li_lock);
5527         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5528         spin_lock(&lfsck->li_lock);
5529         if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
5530                 spin_unlock(&lfsck->li_lock);
5531                 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
5532                                                       true);
5533                 if (llst != NULL)
5534                         lfsck_layout_llst_put(llst);
5535                 spin_lock(&lfsck->li_lock);
5536                 rc = -EAGAIN;
5537         }
5538
5539         RETURN(rc);
5540 }
5541
5542 static struct lfsck_operations lfsck_layout_master_ops = {
5543         .lfsck_reset            = lfsck_layout_reset,
5544         .lfsck_fail             = lfsck_layout_fail,
5545         .lfsck_checkpoint       = lfsck_layout_master_checkpoint,
5546         .lfsck_prep             = lfsck_layout_master_prep,
5547         .lfsck_exec_oit         = lfsck_layout_master_exec_oit,
5548         .lfsck_exec_dir         = lfsck_layout_exec_dir,
5549         .lfsck_post             = lfsck_layout_master_post,
5550         .lfsck_interpret        = lfsck_layout_master_async_interpret,
5551         .lfsck_dump             = lfsck_layout_dump,
5552         .lfsck_double_scan      = lfsck_layout_master_double_scan,
5553         .lfsck_data_release     = lfsck_layout_master_data_release,
5554         .lfsck_quit             = lfsck_layout_master_quit,
5555         .lfsck_in_notify        = lfsck_layout_master_in_notify,
5556         .lfsck_query            = lfsck_layout_query,
5557         .lfsck_stop_notify      = lfsck_layout_master_stop_notify,
5558 };
5559
5560 static struct lfsck_operations lfsck_layout_slave_ops = {
5561         .lfsck_reset            = lfsck_layout_reset,
5562         .lfsck_fail             = lfsck_layout_fail,
5563         .lfsck_checkpoint       = lfsck_layout_slave_checkpoint,
5564         .lfsck_prep             = lfsck_layout_slave_prep,
5565         .lfsck_exec_oit         = lfsck_layout_slave_exec_oit,
5566         .lfsck_exec_dir         = lfsck_layout_exec_dir,
5567         .lfsck_post             = lfsck_layout_slave_post,
5568         .lfsck_dump             = lfsck_layout_dump,
5569         .lfsck_double_scan      = lfsck_layout_slave_double_scan,
5570         .lfsck_data_release     = lfsck_layout_slave_data_release,
5571         .lfsck_quit             = lfsck_layout_slave_quit,
5572         .lfsck_in_notify        = lfsck_layout_slave_in_notify,
5573         .lfsck_query            = lfsck_layout_query,
5574         .lfsck_join             = lfsck_layout_slave_join,
5575 };
5576
5577 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
5578 {
5579         struct lfsck_component  *com;
5580         struct lfsck_layout     *lo;
5581         struct dt_object        *root = NULL;
5582         struct dt_object        *obj;
5583         int                      rc;
5584         ENTRY;
5585
5586         OBD_ALLOC_PTR(com);
5587         if (com == NULL)
5588                 RETURN(-ENOMEM);
5589
5590         INIT_LIST_HEAD(&com->lc_link);
5591         INIT_LIST_HEAD(&com->lc_link_dir);
5592         init_rwsem(&com->lc_sem);
5593         atomic_set(&com->lc_ref, 1);
5594         com->lc_lfsck = lfsck;
5595         com->lc_type = LT_LAYOUT;
5596         if (lfsck->li_master) {
5597                 struct lfsck_layout_master_data *llmd;
5598
5599                 com->lc_ops = &lfsck_layout_master_ops;
5600                 OBD_ALLOC_PTR(llmd);
5601                 if (llmd == NULL)
5602                         GOTO(out, rc = -ENOMEM);
5603
5604                 INIT_LIST_HEAD(&llmd->llmd_req_list);
5605                 spin_lock_init(&llmd->llmd_lock);
5606                 INIT_LIST_HEAD(&llmd->llmd_ost_list);
5607                 INIT_LIST_HEAD(&llmd->llmd_ost_phase1_list);
5608                 INIT_LIST_HEAD(&llmd->llmd_ost_phase2_list);
5609                 INIT_LIST_HEAD(&llmd->llmd_mdt_list);
5610                 INIT_LIST_HEAD(&llmd->llmd_mdt_phase1_list);
5611                 INIT_LIST_HEAD(&llmd->llmd_mdt_phase2_list);
5612                 init_waitqueue_head(&llmd->llmd_thread.t_ctl_waitq);
5613                 com->lc_data = llmd;
5614         } else {
5615                 struct lfsck_layout_slave_data *llsd;
5616
5617                 com->lc_ops = &lfsck_layout_slave_ops;
5618                 OBD_ALLOC_PTR(llsd);
5619                 if (llsd == NULL)
5620                         GOTO(out, rc = -ENOMEM);
5621
5622                 INIT_LIST_HEAD(&llsd->llsd_seq_list);
5623                 INIT_LIST_HEAD(&llsd->llsd_master_list);
5624                 spin_lock_init(&llsd->llsd_lock);
5625                 llsd->llsd_rb_root = RB_ROOT;
5626                 rwlock_init(&llsd->llsd_rb_lock);
5627                 com->lc_data = llsd;
5628         }
5629         com->lc_file_size = sizeof(*lo);
5630         OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
5631         if (com->lc_file_ram == NULL)
5632                 GOTO(out, rc = -ENOMEM);
5633
5634         OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
5635         if (com->lc_file_disk == NULL)
5636                 GOTO(out, rc = -ENOMEM);
5637
5638         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
5639         if (IS_ERR(root))
5640                 GOTO(out, rc = PTR_ERR(root));
5641
5642         if (unlikely(!dt_try_as_dir(env, root)))
5643                 GOTO(out, rc = -ENOTDIR);
5644
5645         obj = local_file_find_or_create(env, lfsck->li_los, root,
5646                                         lfsck_layout_name,
5647                                         S_IFREG | S_IRUGO | S_IWUSR);
5648         if (IS_ERR(obj))
5649                 GOTO(out, rc = PTR_ERR(obj));
5650
5651         com->lc_obj = obj;
5652         rc = lfsck_layout_load(env, com);
5653         if (rc > 0)
5654                 rc = lfsck_layout_reset(env, com, true);
5655         else if (rc == -ENOENT)
5656                 rc = lfsck_layout_init(env, com);
5657
5658         if (rc != 0)
5659                 GOTO(out, rc);
5660
5661         lo = com->lc_file_ram;
5662         switch (lo->ll_status) {
5663         case LS_INIT:
5664         case LS_COMPLETED:
5665         case LS_FAILED:
5666         case LS_STOPPED:
5667         case LS_PARTIAL:
5668                 spin_lock(&lfsck->li_lock);
5669                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
5670                 spin_unlock(&lfsck->li_lock);
5671                 break;
5672         default:
5673                 CERROR("%s: unknown lfsck_layout status: rc = %u\n",
5674                        lfsck_lfsck2name(lfsck), lo->ll_status);
5675                 /* fall through */
5676         case LS_SCANNING_PHASE1:
5677         case LS_SCANNING_PHASE2:
5678                 /* No need to store the status to disk right now.
5679                  * If the system crashed before the status stored,
5680                  * it will be loaded back when next time. */
5681                 lo->ll_status = LS_CRASHED;
5682                 lo->ll_flags |= LF_INCOMPLETE;
5683                 /* fall through */
5684         case LS_PAUSED:
5685         case LS_CRASHED:
5686         case LS_CO_FAILED:
5687         case LS_CO_STOPPED:
5688         case LS_CO_PAUSED:
5689                 spin_lock(&lfsck->li_lock);
5690                 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
5691                 spin_unlock(&lfsck->li_lock);
5692                 break;
5693         }
5694
5695         if (lo->ll_flags & LF_CRASHED_LASTID) {
5696                 LASSERT(lfsck->li_out_notify != NULL);
5697
5698                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5699                                      LE_LASTID_REBUILDING);
5700         }
5701
5702         GOTO(out, rc = 0);
5703
5704 out:
5705         if (root != NULL && !IS_ERR(root))
5706                 lu_object_put(env, &root->do_lu);
5707
5708         if (rc != 0)
5709                 lfsck_component_cleanup(env, com);
5710
5711         return rc;
5712 }
5713
5714 struct lfsck_orphan_it {
5715         struct lfsck_component           *loi_com;
5716         struct lfsck_rbtree_node         *loi_lrn;
5717         struct lfsck_layout_slave_target *loi_llst;
5718         struct lu_fid                     loi_key;
5719         struct lu_orphan_rec              loi_rec;
5720         __u64                             loi_hash;
5721         unsigned int                      loi_over:1;
5722 };
5723
5724 static int lfsck_fid_match_idx(const struct lu_env *env,
5725                                struct lfsck_instance *lfsck,
5726                                const struct lu_fid *fid, int idx)
5727 {
5728         struct seq_server_site  *ss;
5729         struct lu_server_fld    *sf;
5730         struct lu_seq_range      range  = { 0 };
5731         int                      rc;
5732
5733         /* All abnormal cases will be returned to MDT0. */
5734         if (!fid_is_norm(fid)) {
5735                 if (idx == 0)
5736                         return 1;
5737
5738                 return 0;
5739         }
5740
5741         ss = lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
5742         if (unlikely(ss == NULL))
5743                 return -ENOTCONN;
5744
5745         sf = ss->ss_server_fld;
5746         LASSERT(sf != NULL);
5747
5748         fld_range_set_any(&range);
5749         rc = fld_server_lookup(env, sf, fid_seq(fid), &range);
5750         if (rc != 0)
5751                 return rc;
5752
5753         if (!fld_range_is_mdt(&range))
5754                 return -EINVAL;
5755
5756         if (range.lsr_index == idx)
5757                 return 1;
5758
5759         return 0;
5760 }
5761
5762 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
5763                                         struct dt_device *dev,
5764                                         struct dt_object *obj)
5765 {
5766         struct thandle *handle;
5767         int             rc;
5768         ENTRY;
5769
5770         handle = dt_trans_create(env, dev);
5771         if (IS_ERR(handle))
5772                 RETURN_EXIT;
5773
5774         rc = dt_declare_ref_del(env, obj, handle);
5775         if (rc != 0)
5776                 GOTO(stop, rc);
5777
5778         rc = dt_declare_destroy(env, obj, handle);
5779         if (rc != 0)
5780                 GOTO(stop, rc);
5781
5782         rc = dt_trans_start_local(env, dev, handle);
5783         if (rc != 0)
5784                 GOTO(stop, rc);
5785
5786         dt_write_lock(env, obj, 0);
5787         rc = dt_ref_del(env, obj, handle);
5788         if (rc == 0)
5789                 rc = dt_destroy(env, obj, handle);
5790         dt_write_unlock(env, obj);
5791
5792         GOTO(stop, rc);
5793
5794 stop:
5795         dt_trans_stop(env, dev, handle);
5796
5797         RETURN_EXIT;
5798 }
5799
5800 static int lfsck_orphan_index_lookup(const struct lu_env *env,
5801                                      struct dt_object *dt,
5802                                      struct dt_rec *rec,
5803                                      const struct dt_key *key,
5804                                      struct lustre_capa *capa)
5805 {
5806         return -EOPNOTSUPP;
5807 }
5808
5809 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
5810                                              struct dt_object *dt,
5811                                              const struct dt_rec *rec,
5812                                              const struct dt_key *key,
5813                                              struct thandle *handle)
5814 {
5815         return -EOPNOTSUPP;
5816 }
5817
5818 static int lfsck_orphan_index_insert(const struct lu_env *env,
5819                                      struct dt_object *dt,
5820                                      const struct dt_rec *rec,
5821                                      const struct dt_key *key,
5822                                      struct thandle *handle,
5823                                      struct lustre_capa *capa,
5824                                      int ignore_quota)
5825 {
5826         return -EOPNOTSUPP;
5827 }
5828
5829 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
5830                                              struct dt_object *dt,
5831                                              const struct dt_key *key,
5832                                              struct thandle *handle)
5833 {
5834         return -EOPNOTSUPP;
5835 }
5836
5837 static int lfsck_orphan_index_delete(const struct lu_env *env,
5838                                      struct dt_object *dt,
5839                                      const struct dt_key *key,
5840                                      struct thandle *handle,
5841                                      struct lustre_capa *capa)
5842 {
5843         return -EOPNOTSUPP;
5844 }
5845
5846 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
5847                                           struct dt_object *dt,
5848                                           __u32 attr,
5849                                           struct lustre_capa *capa)
5850 {
5851         struct dt_device                *dev    = lu2dt_dev(dt->do_lu.lo_dev);
5852         struct lfsck_instance           *lfsck;
5853         struct lfsck_component          *com    = NULL;
5854         struct lfsck_layout_slave_data  *llsd;
5855         struct lfsck_orphan_it          *it     = NULL;
5856         int                              rc     = 0;
5857         ENTRY;
5858
5859         lfsck = lfsck_instance_find(dev, true, false);
5860         if (unlikely(lfsck == NULL))
5861                 RETURN(ERR_PTR(-ENXIO));
5862
5863         com = lfsck_component_find(lfsck, LT_LAYOUT);
5864         if (unlikely(com == NULL))
5865                 GOTO(out, rc = -ENOENT);
5866
5867         llsd = com->lc_data;
5868         if (!llsd->llsd_rbtree_valid)
5869                 GOTO(out, rc = -ESRCH);
5870
5871         OBD_ALLOC_PTR(it);
5872         if (it == NULL)
5873                 GOTO(out, rc = -ENOMEM);
5874
5875         it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
5876         if (it->loi_llst == NULL)
5877                 GOTO(out, rc = -ENXIO);
5878
5879         if (dev->dd_record_fid_accessed) {
5880                 /* The first iteration against the rbtree, scan the whole rbtree
5881                  * to remove the nodes which do NOT need to be handled. */
5882                 write_lock(&llsd->llsd_rb_lock);
5883                 if (dev->dd_record_fid_accessed) {
5884                         struct rb_node                  *node;
5885                         struct rb_node                  *next;
5886                         struct lfsck_rbtree_node        *lrn;
5887
5888                         /* No need to record the fid accessing anymore. */
5889                         dev->dd_record_fid_accessed = 0;
5890
5891                         node = rb_first(&llsd->llsd_rb_root);
5892                         while (node != NULL) {
5893                                 next = rb_next(node);
5894                                 lrn = rb_entry(node, struct lfsck_rbtree_node,
5895                                                lrn_node);
5896                                 if (atomic_read(&lrn->lrn_known_count) <=
5897                                     atomic_read(&lrn->lrn_accessed_count)) {
5898                                         rb_erase(node, &llsd->llsd_rb_root);
5899                                         lfsck_rbtree_free(lrn);
5900                                 }
5901                                 node = next;
5902                         }
5903                 }
5904                 write_unlock(&llsd->llsd_rb_lock);
5905         }
5906
5907         /* read lock the rbtree when init, and unlock when fini */
5908         read_lock(&llsd->llsd_rb_lock);
5909         it->loi_com = com;
5910         com = NULL;
5911
5912         GOTO(out, rc = 0);
5913
5914 out:
5915         if (com != NULL)
5916                 lfsck_component_put(env, com);
5917         lfsck_instance_put(env, lfsck);
5918         if (rc != 0) {
5919                 if (it != NULL)
5920                         OBD_FREE_PTR(it);
5921
5922                 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
5923         }
5924
5925         return (struct dt_it *)it;
5926 }
5927
5928 static void lfsck_orphan_it_fini(const struct lu_env *env,
5929                                  struct dt_it *di)
5930 {
5931         struct lfsck_orphan_it           *it    = (struct lfsck_orphan_it *)di;
5932         struct lfsck_component           *com   = it->loi_com;
5933         struct lfsck_layout_slave_data   *llsd;
5934         struct lfsck_layout_slave_target *llst;
5935
5936         if (com != NULL) {
5937                 llsd = com->lc_data;
5938                 read_unlock(&llsd->llsd_rb_lock);
5939                 llst = it->loi_llst;
5940                 LASSERT(llst != NULL);
5941
5942                 /* Save the key and hash for iterate next. */
5943                 llst->llst_fid = it->loi_key;
5944                 llst->llst_hash = it->loi_hash;
5945                 lfsck_layout_llst_put(llst);
5946                 lfsck_component_put(env, com);
5947         }
5948         OBD_FREE_PTR(it);
5949 }
5950
5951 /**
5952  * \retval       +1: the iteration finished
5953  * \retval        0: on success, not finished
5954  * \retval      -ve: on error
5955  */
5956 static int lfsck_orphan_it_next(const struct lu_env *env,
5957                                 struct dt_it *di)
5958 {
5959         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5960         struct filter_fid_old           *pfid   = &info->lti_old_pfid;
5961         struct lu_attr                  *la     = &info->lti_la;
5962         struct lfsck_orphan_it          *it     = (struct lfsck_orphan_it *)di;
5963         struct lu_fid                   *key    = &it->loi_key;
5964         struct lu_orphan_rec            *rec    = &it->loi_rec;
5965         struct lfsck_component          *com    = it->loi_com;
5966         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5967         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
5968         struct dt_object                *obj;
5969         struct lfsck_rbtree_node        *lrn;
5970         int                              pos;
5971         int                              rc;
5972         __u32                            save;
5973         __u32                            idx    = it->loi_llst->llst_index;
5974         bool                             exact  = false;
5975         ENTRY;
5976
5977         if (it->loi_over)
5978                 RETURN(1);
5979
5980 again0:
5981         lrn = it->loi_lrn;
5982         if (lrn == NULL) {
5983                 lrn = lfsck_rbtree_search(llsd, key, &exact);
5984                 if (lrn == NULL) {
5985                         it->loi_over = 1;
5986                         RETURN(1);
5987                 }
5988
5989                 it->loi_lrn = lrn;
5990                 if (!exact) {
5991                         key->f_seq = lrn->lrn_seq;
5992                         key->f_oid = lrn->lrn_first_oid;
5993                         key->f_ver = 0;
5994                 }
5995         } else {
5996                 key->f_oid++;
5997                 if (unlikely(key->f_oid == 0)) {
5998                         key->f_seq++;
5999                         it->loi_lrn = NULL;
6000                         goto again0;
6001                 }
6002
6003                 if (key->f_oid >=
6004                     lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
6005                         it->loi_lrn = NULL;
6006                         goto again0;
6007                 }
6008         }
6009
6010         if (unlikely(atomic_read(&lrn->lrn_known_count) <=
6011                      atomic_read(&lrn->lrn_accessed_count))) {
6012                 struct rb_node *next = rb_next(&lrn->lrn_node);
6013
6014                 while (next != NULL) {
6015                         lrn = rb_entry(next, struct lfsck_rbtree_node,
6016                                        lrn_node);
6017                         if (atomic_read(&lrn->lrn_known_count) >
6018                             atomic_read(&lrn->lrn_accessed_count))
6019                                 break;
6020                         next = rb_next(next);
6021                 }
6022
6023                 if (next == NULL) {
6024                         it->loi_over = 1;
6025                         RETURN(1);
6026                 }
6027
6028                 it->loi_lrn = lrn;
6029                 key->f_seq = lrn->lrn_seq;
6030                 key->f_oid = lrn->lrn_first_oid;
6031                 key->f_ver = 0;
6032         }
6033
6034         pos = key->f_oid - lrn->lrn_first_oid;
6035
6036 again1:
6037         pos = find_next_bit(lrn->lrn_known_bitmap,
6038                             LFSCK_RBTREE_BITMAP_WIDTH, pos);
6039         if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
6040                 key->f_oid = lrn->lrn_first_oid + pos;
6041                 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
6042                         key->f_seq++;
6043                         key->f_oid = 0;
6044                 }
6045                 it->loi_lrn = NULL;
6046                 goto again0;
6047         }
6048
6049         if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
6050                 pos++;
6051                 goto again1;
6052         }
6053
6054         key->f_oid = lrn->lrn_first_oid + pos;
6055         obj = lfsck_object_find(env, lfsck, key);
6056         if (IS_ERR(obj)) {
6057                 rc = PTR_ERR(obj);
6058                 if (rc == -ENOENT) {
6059                         pos++;
6060                         goto again1;
6061                 }
6062                 RETURN(rc);
6063         }
6064
6065         dt_read_lock(env, obj, 0);
6066         if (!dt_object_exists(obj)) {
6067                 dt_read_unlock(env, obj);
6068                 lfsck_object_put(env, obj);
6069                 pos++;
6070                 goto again1;
6071         }
6072
6073         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
6074         if (rc != 0)
6075                 GOTO(out, rc);
6076
6077         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, pfid, sizeof(*pfid)),
6078                           XATTR_NAME_FID, BYPASS_CAPA);
6079         if (rc == -ENODATA) {
6080                 /* For the pre-created OST-object, update the bitmap to avoid
6081                  * others LFSCK (second phase) iteration to touch it again. */
6082                 if (la->la_ctime == 0) {
6083                         if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
6084                                 atomic_inc(&lrn->lrn_accessed_count);
6085
6086                         /* For the race between repairing dangling referenced
6087                          * MDT-object and unlink the file, it may left orphan
6088                          * OST-object there. Destroy it now! */
6089                         if (unlikely(!(la->la_mode & S_ISUID))) {
6090                                 dt_read_unlock(env, obj);
6091                                 lfsck_layout_destroy_orphan(env,
6092                                                             lfsck->li_bottom,
6093                                                             obj);
6094                                 lfsck_object_put(env, obj);
6095                                 pos++;
6096                                 goto again1;
6097                         }
6098                 } else if (idx == 0) {
6099                         /* If the orphan OST-object has no parent information,
6100                          * regard it as referenced by the MDT-object on MDT0. */
6101                         fid_zero(&rec->lor_fid);
6102                         rec->lor_uid = la->la_uid;
6103                         rec->lor_gid = la->la_gid;
6104                         GOTO(out, rc = 0);
6105                 }
6106
6107                 dt_read_unlock(env, obj);
6108                 lfsck_object_put(env, obj);
6109                 pos++;
6110                 goto again1;
6111         }
6112
6113         if (rc < 0)
6114                 GOTO(out, rc);
6115
6116         if (rc != sizeof(struct filter_fid) &&
6117             rc != sizeof(struct filter_fid_old))
6118                 GOTO(out, rc = -EINVAL);
6119
6120         fid_le_to_cpu(&rec->lor_fid, &pfid->ff_parent);
6121         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
6122          * MDT-object's FID::f_ver, instead it is the OST-object index in its
6123          * parent MDT-object's layout EA. */
6124         save = rec->lor_fid.f_stripe_idx;
6125         rec->lor_fid.f_ver = 0;
6126         rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_fid, idx);
6127         /* If the orphan OST-object does not claim the MDT, then next.
6128          *
6129          * If we do not know whether it matches or not, then return it
6130          * to the MDT for further check. */
6131         if (rc == 0) {
6132                 dt_read_unlock(env, obj);
6133                 lfsck_object_put(env, obj);
6134                 pos++;
6135                 goto again1;
6136         }
6137
6138         rec->lor_fid.f_stripe_idx = save;
6139         rec->lor_uid = la->la_uid;
6140         rec->lor_gid = la->la_gid;
6141
6142         CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u\n",
6143                lfsck_lfsck2name(com->lc_lfsck), PFID(key), PFID(&rec->lor_fid),
6144                rec->lor_uid, rec->lor_gid);
6145
6146         GOTO(out, rc = 0);
6147
6148 out:
6149         dt_read_unlock(env, obj);
6150         lfsck_object_put(env, obj);
6151         if (rc == 0)
6152                 it->loi_hash++;
6153
6154         return rc;
6155 }
6156
6157 /**
6158  * \retval       +1: locate to the exactly position
6159  * \retval        0: cannot locate to the exactly position,
6160  *                   call next() to move to a valid position.
6161  * \retval      -ve: on error
6162  */
6163 static int lfsck_orphan_it_get(const struct lu_env *env,
6164                                struct dt_it *di,
6165                                const struct dt_key *key)
6166 {
6167         struct lfsck_orphan_it  *it   = (struct lfsck_orphan_it *)di;
6168         int                      rc;
6169
6170         it->loi_key = *(struct lu_fid *)key;
6171         rc = lfsck_orphan_it_next(env, di);
6172         if (rc == 1)
6173                 return 0;
6174
6175         if (rc == 0)
6176                 return 1;
6177
6178         return rc;
6179 }
6180
6181 static void lfsck_orphan_it_put(const struct lu_env *env,
6182                                 struct dt_it *di)
6183 {
6184 }
6185
6186 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
6187                                           const struct dt_it *di)
6188 {
6189         struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6190
6191         return (struct dt_key *)&it->loi_key;
6192 }
6193
6194 static int lfsck_orphan_it_key_size(const struct lu_env *env,
6195                                     const struct dt_it *di)
6196 {
6197         return sizeof(struct lu_fid);
6198 }
6199
6200 static int lfsck_orphan_it_rec(const struct lu_env *env,
6201                                const struct dt_it *di,
6202                                struct dt_rec *rec,
6203                                __u32 attr)
6204 {
6205         struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6206
6207         *(struct lu_orphan_rec *)rec = it->loi_rec;
6208
6209         return 0;
6210 }
6211
6212 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
6213                                    const struct dt_it *di)
6214 {
6215         struct lfsck_orphan_it  *it   = (struct lfsck_orphan_it *)di;
6216
6217         return it->loi_hash;
6218 }
6219
6220 /**
6221  * \retval       +1: locate to the exactly position
6222  * \retval        0: cannot locate to the exactly position,
6223  *                   call next() to move to a valid position.
6224  * \retval      -ve: on error
6225  */
6226 static int lfsck_orphan_it_load(const struct lu_env *env,
6227                                 const struct dt_it *di,
6228                                 __u64 hash)
6229 {
6230         struct lfsck_orphan_it           *it   = (struct lfsck_orphan_it *)di;
6231         struct lfsck_layout_slave_target *llst = it->loi_llst;
6232         int                               rc;
6233
6234         LASSERT(llst != NULL);
6235
6236         if (hash != llst->llst_hash) {
6237                 CWARN("%s: the given hash "LPU64" for orphan iteration does "
6238                       "not match the one when fini "LPU64", to be reset.\n",
6239                       lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
6240                       llst->llst_hash);
6241                 fid_zero(&llst->llst_fid);
6242                 llst->llst_hash = 0;
6243         }
6244
6245         it->loi_key = llst->llst_fid;
6246         it->loi_hash = llst->llst_hash;
6247         rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
6248         if (rc == 1)
6249                 return 0;
6250
6251         if (rc == 0)
6252                 return 1;
6253
6254         return rc;
6255 }
6256
6257 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
6258                                    const struct dt_it *di,
6259                                    void *key_rec)
6260 {
6261         return 0;
6262 }
6263
6264 const struct dt_index_operations lfsck_orphan_index_ops = {
6265         .dio_lookup             = lfsck_orphan_index_lookup,
6266         .dio_declare_insert     = lfsck_orphan_index_declare_insert,
6267         .dio_insert             = lfsck_orphan_index_insert,
6268         .dio_declare_delete     = lfsck_orphan_index_declare_delete,
6269         .dio_delete             = lfsck_orphan_index_delete,
6270         .dio_it = {
6271                 .init           = lfsck_orphan_it_init,
6272                 .fini           = lfsck_orphan_it_fini,
6273                 .get            = lfsck_orphan_it_get,
6274                 .put            = lfsck_orphan_it_put,
6275                 .next           = lfsck_orphan_it_next,
6276                 .key            = lfsck_orphan_it_key,
6277                 .key_size       = lfsck_orphan_it_key_size,
6278                 .rec            = lfsck_orphan_it_rec,
6279                 .store          = lfsck_orphan_it_store,
6280                 .load           = lfsck_orphan_it_load,
6281                 .key_rec        = lfsck_orphan_it_key_rec,
6282         }
6283 };