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