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         __u64                   llst_gen;
  69         atomic_t                llst_ref;
  70         __u32                   llst_index;
  71 };
  72
  73 struct lfsck_layout_slave_data {
  74         /* list for lfsck_layout_seq */
  75         struct list_head         llsd_seq_list;
  76
  77         /* list for the masters involve layout verification. */
  78         struct list_head         llsd_master_list;
  79         spinlock_t               llsd_lock;
  80         __u64                    llsd_touch_gen;
  81         struct dt_object        *llsd_rb_obj;
  82         struct rb_root           llsd_rb_root;
  83         rwlock_t                 llsd_rb_lock;
  84         unsigned int             llsd_rbtree_valid:1;
  85 };
  86
  87 struct lfsck_layout_object {
  88         struct dt_object        *llo_obj;
  89         struct lu_attr           llo_attr;
  90         atomic_t                 llo_ref;
  91         __u16                    llo_gen;
  92 };
  93
  94 struct lfsck_layout_req {
  95         struct list_head                 llr_list;
  96         struct lfsck_layout_object      *llr_parent;
  97         struct dt_object                *llr_child;
  98         __u32                            llr_ost_idx;
  99         __u32                            llr_lov_idx; /* offset in LOV EA */
 100 };
 101
 102 struct lfsck_layout_master_data {
 103         spinlock_t              llmd_lock;
 104         struct list_head        llmd_req_list;
 105
 106         /* list for the ost targets involve layout verification. */
 107         struct list_head        llmd_ost_list;
 108
 109         /* list for the ost targets in phase1 scanning. */
 110         struct list_head        llmd_ost_phase1_list;
 111
 112         /* list for the ost targets in phase1 scanning. */
 113         struct list_head        llmd_ost_phase2_list;
 114
 115         /* list for the mdt targets involve layout verification. */
 116         struct list_head        llmd_mdt_list;
 117
 118         /* list for the mdt targets in phase1 scanning. */
 119         struct list_head        llmd_mdt_phase1_list;
 120
 121         /* list for the mdt targets in phase1 scanning. */
 122         struct list_head        llmd_mdt_phase2_list;
 123
 124         struct ptlrpc_thread    llmd_thread;
 125         __u32                   llmd_touch_gen;
 126         int                     llmd_prefetched;
 127         int                     llmd_assistant_status;
 128         int                     llmd_post_result;
 129         unsigned int            llmd_to_post:1,
 130                                 llmd_to_double_scan:1,
 131                                 llmd_in_double_scan:1,
 132                                 llmd_exit:1;
 133 };
 134
 135 struct lfsck_layout_slave_async_args {
 136         struct obd_export                *llsaa_exp;
 137         struct lfsck_component           *llsaa_com;
 138         struct lfsck_layout_slave_target *llsaa_llst;
 139 };
 140
 141 static struct lfsck_layout_object *
 142 lfsck_layout_object_init(const struct lu_env *env, struct dt_object *obj,
 143                          __u16 gen)
 144 {
 145         struct lfsck_layout_object *llo;
 146         int                         rc;
 147
 148         OBD_ALLOC_PTR(llo);
 149         if (llo == NULL)
 150                 return ERR_PTR(-ENOMEM);
 151
 152         rc = dt_attr_get(env, obj, &llo->llo_attr, BYPASS_CAPA);
 153         if (rc != 0) {
 154                 OBD_FREE_PTR(llo);
 155
 156                 return ERR_PTR(rc);
 157         }
 158
 159         lu_object_get(&obj->do_lu);
 160         llo->llo_obj = obj;
 161         /* The gen can be used to check whether some others have changed the
 162          * file layout after LFSCK pre-fetching but before real verification. */
 163         llo->llo_gen = gen;
 164         atomic_set(&llo->llo_ref, 1);
 165
 166         return llo;
 167 }
 168
 169 static inline void
 170 lfsck_layout_llst_put(struct lfsck_layout_slave_target *llst)
 171 {
 172         if (atomic_dec_and_test(&llst->llst_ref)) {
 173                 LASSERT(list_empty(&llst->llst_list));
 174
 175                 OBD_FREE_PTR(llst);
 176         }
 177 }
 178
 179 static inline int
 180 lfsck_layout_llst_add(struct lfsck_layout_slave_data *llsd, __u32 index)
 181 {
 182         struct lfsck_layout_slave_target *llst;
 183         struct lfsck_layout_slave_target *tmp;
 184         int                               rc   = 0;
 185
 186         OBD_ALLOC_PTR(llst);
 187         if (llst == NULL)
 188                 return -ENOMEM;
 189
 190         INIT_LIST_HEAD(&llst->llst_list);
 191         llst->llst_gen = 0;
 192         llst->llst_index = index;
 193         atomic_set(&llst->llst_ref, 1);
 194
 195         spin_lock(&llsd->llsd_lock);
 196         list_for_each_entry(tmp, &llsd->llsd_master_list, llst_list) {
 197                 if (tmp->llst_index == index) {
 198                         rc = -EALREADY;
 199                         break;
 200                 }
 201         }
 202         if (rc == 0)
 203                 list_add_tail(&llst->llst_list, &llsd->llsd_master_list);
 204         spin_unlock(&llsd->llsd_lock);
 205
 206         if (rc != 0)
 207                 OBD_FREE_PTR(llst);
 208
 209         return rc;
 210 }
 211
 212 static inline void
 213 lfsck_layout_llst_del(struct lfsck_layout_slave_data *llsd,
 214                       struct lfsck_layout_slave_target *llst)
 215 {
 216         bool del = false;
 217
 218         spin_lock(&llsd->llsd_lock);
 219         if (!list_empty(&llst->llst_list)) {
 220                 list_del_init(&llst->llst_list);
 221                 del = true;
 222         }
 223         spin_unlock(&llsd->llsd_lock);
 224
 225         if (del)
 226                 lfsck_layout_llst_put(llst);
 227 }
 228
 229 static inline struct lfsck_layout_slave_target *
 230 lfsck_layout_llst_find_and_del(struct lfsck_layout_slave_data *llsd,
 231                                __u32 index)
 232 {
 233         struct lfsck_layout_slave_target *llst;
 234
 235         spin_lock(&llsd->llsd_lock);
 236         list_for_each_entry(llst, &llsd->llsd_master_list, llst_list) {
 237                 if (llst->llst_index == index) {
 238                         list_del_init(&llst->llst_list);
 239                         spin_unlock(&llsd->llsd_lock);
 240
 241                         return llst;
 242                 }
 243         }
 244         spin_unlock(&llsd->llsd_lock);
 245
 246         return NULL;
 247 }
 248
 249 static inline void lfsck_layout_object_put(const struct lu_env *env,
 250                                            struct lfsck_layout_object *llo)
 251 {
 252         if (atomic_dec_and_test(&llo->llo_ref)) {
 253                 lfsck_object_put(env, llo->llo_obj);
 254                 OBD_FREE_PTR(llo);
 255         }
 256 }
 257
 258 static struct lfsck_layout_req *
 259 lfsck_layout_req_init(struct lfsck_layout_object *parent,
 260                       struct dt_object *child, __u32 ost_idx, __u32 lov_idx)
 261 {
 262         struct lfsck_layout_req *llr;
 263
 264         OBD_ALLOC_PTR(llr);
 265         if (llr == NULL)
 266                 return ERR_PTR(-ENOMEM);
 267
 268         INIT_LIST_HEAD(&llr->llr_list);
 269         atomic_inc(&parent->llo_ref);
 270         llr->llr_parent = parent;
 271         llr->llr_child = child;
 272         llr->llr_ost_idx = ost_idx;
 273         llr->llr_lov_idx = lov_idx;
 274
 275         return llr;
 276 }
 277
 278 static inline void lfsck_layout_req_fini(const struct lu_env *env,
 279                                          struct lfsck_layout_req *llr)
 280 {
 281         lu_object_put(env, &llr->llr_child->do_lu);
 282         lfsck_layout_object_put(env, llr->llr_parent);
 283         OBD_FREE_PTR(llr);
 284 }
 285
 286 static inline bool lfsck_layout_req_empty(struct lfsck_layout_master_data *llmd)
 287 {
 288         bool empty = false;
 289
 290         spin_lock(&llmd->llmd_lock);
 291         if (list_empty(&llmd->llmd_req_list))
 292                 empty = true;
 293         spin_unlock(&llmd->llmd_lock);
 294
 295         return empty;
 296 }
 297
 298 static int lfsck_layout_get_lovea(const struct lu_env *env,
 299                                   struct dt_object *obj,
 300                                   struct lu_buf *buf, ssize_t *buflen)
 301 {
 302         int rc;
 303
 304 again:
 305         rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LOV, BYPASS_CAPA);
 306         if (rc == -ERANGE) {
 307                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV,
 308                                   BYPASS_CAPA);
 309                 if (rc <= 0)
 310                         return rc;
 311
 312                 lu_buf_realloc(buf, rc);
 313                 if (buflen != NULL)
 314                         *buflen = buf->lb_len;
 315
 316                 if (buf->lb_buf == NULL)
 317                         return -ENOMEM;
 318
 319                 goto again;
 320         }
 321
 322         if (rc == -ENODATA)
 323                 rc = 0;
 324
 325         if (rc <= 0)
 326                 return rc;
 327
 328         if (unlikely(buf->lb_buf == NULL)) {
 329                 lu_buf_alloc(buf, rc);
 330                 if (buflen != NULL)
 331                         *buflen = buf->lb_len;
 332
 333                 if (buf->lb_buf == NULL)
 334                         return -ENOMEM;
 335
 336                 goto again;
 337         }
 338
 339         return rc;
 340 }
 341
 342 static int lfsck_layout_verify_header(struct lov_mds_md_v1 *lmm)
 343 {
 344         __u32 magic;
 345         __u32 patten;
 346
 347         magic = le32_to_cpu(lmm->lmm_magic);
 348         /* If magic crashed, keep it there. Sometime later, during OST-object
 349          * orphan handling, if some OST-object(s) back-point to it, it can be
 350          * verified and repaired. */
 351         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
 352                 return -EINVAL;
 353
 354         patten = le32_to_cpu(lmm->lmm_pattern);
 355         /* XXX: currently, we only support LOV_PATTERN_RAID0. */
 356         if (patten != LOV_PATTERN_RAID0)
 357                 return -EOPNOTSUPP;
 358
 359         return 0;
 360 }
 361
 362 #define LFSCK_RBTREE_BITMAP_SIZE        PAGE_CACHE_SIZE
 363 #define LFSCK_RBTREE_BITMAP_WIDTH       (LFSCK_RBTREE_BITMAP_SIZE << 3)
 364 #define LFSCK_RBTREE_BITMAP_MASK        (LFSCK_RBTREE_BITMAP_SIZE - 1)
 365
 366 struct lfsck_rbtree_node {
 367         struct rb_node   lrn_node;
 368         __u64            lrn_seq;
 369         __u32            lrn_first_oid;
 370         atomic_t         lrn_known_count;
 371         atomic_t         lrn_accessed_count;
 372         void            *lrn_known_bitmap;
 373         void            *lrn_accessed_bitmap;
 374 };
 375
 376 static inline int lfsck_rbtree_cmp(struct lfsck_rbtree_node *lrn,
 377                                    __u64 seq, __u32 oid)
 378 {
 379         if (seq < lrn->lrn_seq)
 380                 return -1;
 381
 382         if (seq > lrn->lrn_seq)
 383                 return 1;
 384
 385         if (oid < lrn->lrn_first_oid)
 386                 return -1;
 387
 388         if (oid >= lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH)
 389                 return 1;
 390
 391         return 0;
 392 }
 393
 394 /* The caller should hold lock. */
 395 static struct lfsck_rbtree_node *
 396 lfsck_rbtree_search(struct lfsck_layout_slave_data *llsd,
 397                     const struct lu_fid *fid)
 398 {
 399         struct rb_node           *node = llsd->llsd_rb_root.rb_node;
 400         struct lfsck_rbtree_node *lrn;
 401         int                       rc;
 402
 403         while (node != NULL) {
 404                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 405                 rc = lfsck_rbtree_cmp(lrn, fid_seq(fid), fid_oid(fid));
 406                 if (rc < 0)
 407                         node = node->rb_left;
 408                 else if (rc > 0)
 409                         node = node->rb_right;
 410                 else
 411                         return lrn;
 412         }
 413
 414         return NULL;
 415 }
 416
 417 static struct lfsck_rbtree_node *lfsck_rbtree_new(const struct lu_env *env,
 418                                                   const struct lu_fid *fid)
 419 {
 420         struct lfsck_rbtree_node *lrn;
 421
 422         OBD_ALLOC_PTR(lrn);
 423         if (lrn == NULL)
 424                 return ERR_PTR(-ENOMEM);
 425
 426         OBD_ALLOC(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 427         if (lrn->lrn_known_bitmap == NULL) {
 428                 OBD_FREE_PTR(lrn);
 429
 430                 return ERR_PTR(-ENOMEM);
 431         }
 432
 433         OBD_ALLOC(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 434         if (lrn->lrn_accessed_bitmap == NULL) {
 435                 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 436                 OBD_FREE_PTR(lrn);
 437
 438                 return ERR_PTR(-ENOMEM);
 439         }
 440
 441         rb_init_node(&lrn->lrn_node);
 442         lrn->lrn_seq = fid_seq(fid);
 443         lrn->lrn_first_oid = fid_oid(fid) & ~LFSCK_RBTREE_BITMAP_MASK;
 444         atomic_set(&lrn->lrn_known_count, 0);
 445         atomic_set(&lrn->lrn_accessed_count, 0);
 446
 447         return lrn;
 448 }
 449
 450 static void lfsck_rbtree_free(struct lfsck_rbtree_node *lrn)
 451 {
 452         OBD_FREE(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 453         OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
 454         OBD_FREE_PTR(lrn);
 455 }
 456
 457 /* The caller should hold lock. */
 458 static struct lfsck_rbtree_node *
 459 lfsck_rbtree_insert(struct lfsck_layout_slave_data *llsd,
 460                     struct lfsck_rbtree_node *lrn)
 461 {
 462         struct rb_node           **pos    = &(llsd->llsd_rb_root.rb_node);
 463         struct rb_node            *parent = NULL;
 464         struct lfsck_rbtree_node  *tmp;
 465         int                        rc;
 466
 467         while (*pos) {
 468                 parent = *pos;
 469                 tmp = rb_entry(*pos, struct lfsck_rbtree_node, lrn_node);
 470                 rc = lfsck_rbtree_cmp(tmp, lrn->lrn_seq, lrn->lrn_first_oid);
 471                 if (rc < 0)
 472                         pos = &((*pos)->rb_left);
 473                 else if (rc > 0)
 474                         pos = &((*pos)->rb_right);
 475                 else
 476                         return tmp;
 477         }
 478
 479         rb_link_node(&lrn->lrn_node, parent, pos);
 480         rb_insert_color(&lrn->lrn_node, &llsd->llsd_rb_root);
 481
 482         return lrn;
 483 }
 484
 485 static int lfsck_rbtree_setup(const struct lu_env *env,
 486                               struct lfsck_component *com)
 487 {
 488         struct lu_fid                   *fid    = &lfsck_env_info(env)->lti_fid;
 489         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 490         struct dt_device                *dev    = lfsck->li_bottom;
 491         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
 492         struct dt_object                *obj;
 493
 494         fid->f_seq = FID_SEQ_LAYOUT_RBTREE;
 495         fid->f_oid = lfsck_dev_idx(dev);
 496         fid->f_ver = 0;
 497         obj = dt_locate(env, dev, fid);
 498         if (IS_ERR(obj))
 499                 RETURN(PTR_ERR(obj));
 500
 501         /* XXX: Generate an in-RAM object to stand for the layout rbtree.
 502          *      Scanning the layout rbtree will be via the iteration over
 503          *      the object. In the future, the rbtree may be written onto
 504          *      disk with the object.
 505          *
 506          *      Mark the object to be as exist. */
 507         obj->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
 508         llsd->llsd_rb_obj = obj;
 509         llsd->llsd_rbtree_valid = 1;
 510         dev->dd_record_fid_accessed = 1;
 511
 512         return 0;
 513 }
 514
 515 static void lfsck_rbtree_cleanup(const struct lu_env *env,
 516                                  struct lfsck_component *com)
 517 {
 518         struct lfsck_instance           *lfsck = com->lc_lfsck;
 519         struct lfsck_layout_slave_data  *llsd  = com->lc_data;
 520         struct rb_node                  *node  = rb_first(&llsd->llsd_rb_root);
 521         struct rb_node                  *next;
 522         struct lfsck_rbtree_node        *lrn;
 523
 524         lfsck->li_bottom->dd_record_fid_accessed = 0;
 525         /* Invalid the rbtree, then no others will use it. */
 526         write_lock(&llsd->llsd_rb_lock);
 527         llsd->llsd_rbtree_valid = 0;
 528         write_unlock(&llsd->llsd_rb_lock);
 529
 530         while (node != NULL) {
 531                 next = rb_next(node);
 532                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
 533                 rb_erase(node, &llsd->llsd_rb_root);
 534                 lfsck_rbtree_free(lrn);
 535                 node = next;
 536         }
 537
 538         if (llsd->llsd_rb_obj != NULL) {
 539                 lu_object_put(env, &llsd->llsd_rb_obj->do_lu);
 540                 llsd->llsd_rb_obj = NULL;
 541         }
 542 }
 543
 544 static void lfsck_rbtree_update_bitmap(const struct lu_env *env,
 545                                        struct lfsck_component *com,
 546                                        const struct lu_fid *fid,
 547                                        bool accessed)
 548 {
 549         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
 550         struct lfsck_rbtree_node        *lrn;
 551         bool                             insert = false;
 552         int                              idx;
 553         int                              rc     = 0;
 554         ENTRY;
 555
 556         CDEBUG(D_LFSCK, "%s: update bitmap for "DFID"\n",
 557                lfsck_lfsck2name(com->lc_lfsck), PFID(fid));
 558
 559         if (unlikely(!fid_is_sane(fid) || fid_is_last_id(fid)))
 560                 RETURN_EXIT;
 561
 562         if (!fid_is_idif(fid) && !fid_is_norm(fid))
 563                 RETURN_EXIT;
 564
 565         read_lock(&llsd->llsd_rb_lock);
 566         if (!llsd->llsd_rbtree_valid)
 567                 GOTO(unlock, rc = 0);
 568
 569         lrn = lfsck_rbtree_search(llsd, fid);
 570         if (lrn == NULL) {
 571                 struct lfsck_rbtree_node *tmp;
 572
 573                 LASSERT(!insert);
 574
 575                 read_unlock(&llsd->llsd_rb_lock);
 576                 tmp = lfsck_rbtree_new(env, fid);
 577                 if (IS_ERR(tmp))
 578                         GOTO(out, rc = PTR_ERR(tmp));
 579
 580                 insert = true;
 581                 write_lock(&llsd->llsd_rb_lock);
 582                 if (!llsd->llsd_rbtree_valid) {
 583                         lfsck_rbtree_free(tmp);
 584                         GOTO(unlock, rc = 0);
 585                 }
 586
 587                 lrn = lfsck_rbtree_insert(llsd, tmp);
 588                 if (lrn != tmp)
 589                         lfsck_rbtree_free(tmp);
 590         }
 591
 592         idx = fid_oid(fid) & LFSCK_RBTREE_BITMAP_MASK;
 593         /* Any accessed object must be a known object. */
 594         if (!test_and_set_bit(idx, lrn->lrn_known_bitmap))
 595                 atomic_inc(&lrn->lrn_known_count);
 596         if (accessed) {
 597                 if (!test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
 598                         atomic_inc(&lrn->lrn_accessed_count);
 599         }
 600
 601         GOTO(unlock, rc = 0);
 602
 603 unlock:
 604         if (insert)
 605                 write_unlock(&llsd->llsd_rb_lock);
 606         else
 607                 read_unlock(&llsd->llsd_rb_lock);
 608 out:
 609         if (rc != 0 && accessed) {
 610                 struct lfsck_layout *lo = com->lc_file_ram;
 611
 612                 CERROR("%s: Fail to update object accessed bitmap, will cause "
 613                        "incorrect LFSCK OST-object handling, so disable it to "
 614                        "cancel orphan handling for related device. rc = %d.\n",
 615                        lfsck_lfsck2name(com->lc_lfsck), rc);
 616                 lo->ll_flags |= LF_INCOMPLETE;
 617                 lfsck_rbtree_cleanup(env, com);
 618         }
 619 }
 620
 621 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
 622                                    const struct lfsck_layout *src)
 623 {
 624         int i;
 625
 626         des->ll_magic = le32_to_cpu(src->ll_magic);
 627         des->ll_status = le32_to_cpu(src->ll_status);
 628         des->ll_flags = le32_to_cpu(src->ll_flags);
 629         des->ll_success_count = le32_to_cpu(src->ll_success_count);
 630         des->ll_run_time_phase1 = le32_to_cpu(src->ll_run_time_phase1);
 631         des->ll_run_time_phase2 = le32_to_cpu(src->ll_run_time_phase2);
 632         des->ll_time_last_complete = le64_to_cpu(src->ll_time_last_complete);
 633         des->ll_time_latest_start = le64_to_cpu(src->ll_time_latest_start);
 634         des->ll_time_last_checkpoint =
 635                                 le64_to_cpu(src->ll_time_last_checkpoint);
 636         des->ll_pos_latest_start = le64_to_cpu(src->ll_pos_latest_start);
 637         des->ll_pos_last_checkpoint = le64_to_cpu(src->ll_pos_last_checkpoint);
 638         des->ll_pos_first_inconsistent =
 639                         le64_to_cpu(src->ll_pos_first_inconsistent);
 640         des->ll_objs_checked_phase1 = le64_to_cpu(src->ll_objs_checked_phase1);
 641         des->ll_objs_failed_phase1 = le64_to_cpu(src->ll_objs_failed_phase1);
 642         des->ll_objs_checked_phase2 = le64_to_cpu(src->ll_objs_checked_phase2);
 643         des->ll_objs_failed_phase2 = le64_to_cpu(src->ll_objs_failed_phase2);
 644         for (i = 0; i < LLIT_MAX; i++)
 645                 des->ll_objs_repaired[i] =
 646                                 le64_to_cpu(src->ll_objs_repaired[i]);
 647         des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
 648 }
 649
 650 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
 651                                    const struct lfsck_layout *src)
 652 {
 653         int i;
 654
 655         des->ll_magic = cpu_to_le32(src->ll_magic);
 656         des->ll_status = cpu_to_le32(src->ll_status);
 657         des->ll_flags = cpu_to_le32(src->ll_flags);
 658         des->ll_success_count = cpu_to_le32(src->ll_success_count);
 659         des->ll_run_time_phase1 = cpu_to_le32(src->ll_run_time_phase1);
 660         des->ll_run_time_phase2 = cpu_to_le32(src->ll_run_time_phase2);
 661         des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
 662         des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
 663         des->ll_time_last_checkpoint =
 664                                 cpu_to_le64(src->ll_time_last_checkpoint);
 665         des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
 666         des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
 667         des->ll_pos_first_inconsistent =
 668                         cpu_to_le64(src->ll_pos_first_inconsistent);
 669         des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
 670         des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
 671         des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
 672         des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
 673         for (i = 0; i < LLIT_MAX; i++)
 674                 des->ll_objs_repaired[i] =
 675                                 cpu_to_le64(src->ll_objs_repaired[i]);
 676         des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
 677 }
 678
 679 /**
 680  * \retval +ve: the lfsck_layout is broken, the caller should reset it.
 681  * \retval 0: succeed.
 682  * \retval -ve: failed cases.
 683  */
 684 static int lfsck_layout_load(const struct lu_env *env,
 685                              struct lfsck_component *com)
 686 {
 687         struct lfsck_layout             *lo     = com->lc_file_ram;
 688         const struct dt_body_operations *dbo    = com->lc_obj->do_body_ops;
 689         ssize_t                          size   = com->lc_file_size;
 690         loff_t                           pos    = 0;
 691         int                              rc;
 692
 693         rc = dbo->dbo_read(env, com->lc_obj,
 694                            lfsck_buf_get(env, com->lc_file_disk, size), &pos,
 695                            BYPASS_CAPA);
 696         if (rc == 0) {
 697                 return -ENOENT;
 698         } else if (rc < 0) {
 699                 CWARN("%s: failed to load lfsck_layout: rc = %d\n",
 700                       lfsck_lfsck2name(com->lc_lfsck), rc);
 701                 return rc;
 702         } else if (rc != size) {
 703                 CWARN("%s: crashed lfsck_layout, to be reset: rc = %d\n",
 704                       lfsck_lfsck2name(com->lc_lfsck), rc);
 705                 return 1;
 706         }
 707
 708         lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
 709         if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
 710                 CWARN("%s: invalid lfsck_layout magic %#x != %#x, "
 711                       "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
 712                       lo->ll_magic, LFSCK_LAYOUT_MAGIC);
 713                 return 1;
 714         }
 715
 716         return 0;
 717 }
 718
 719 static int lfsck_layout_store(const struct lu_env *env,
 720                               struct lfsck_component *com)
 721 {
 722         struct dt_object         *obj           = com->lc_obj;
 723         struct lfsck_instance    *lfsck         = com->lc_lfsck;
 724         struct lfsck_layout      *lo            = com->lc_file_disk;
 725         struct thandle           *handle;
 726         ssize_t                   size          = com->lc_file_size;
 727         loff_t                    pos           = 0;
 728         int                       rc;
 729         ENTRY;
 730
 731         lfsck_layout_cpu_to_le(lo, com->lc_file_ram);
 732         handle = dt_trans_create(env, lfsck->li_bottom);
 733         if (IS_ERR(handle)) {
 734                 rc = PTR_ERR(handle);
 735                 CERROR("%s: fail to create trans for storing lfsck_layout: "
 736                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
 737                 RETURN(rc);
 738         }
 739
 740         rc = dt_declare_record_write(env, obj, size, pos, handle);
 741         if (rc != 0) {
 742                 CERROR("%s: fail to declare trans for storing lfsck_layout(1): "
 743                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
 744                 GOTO(out, rc);
 745         }
 746
 747         rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
 748         if (rc != 0) {
 749                 CERROR("%s: fail to start trans for storing lfsck_layout: "
 750                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
 751                 GOTO(out, rc);
 752         }
 753
 754         rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos,
 755                              handle);
 756         if (rc != 0)
 757                 CERROR("%s: fail to store lfsck_layout(1): size = %d, "
 758                        "rc = %d\n", lfsck_lfsck2name(lfsck), (int)size, rc);
 759
 760         GOTO(out, rc);
 761
 762 out:
 763         dt_trans_stop(env, lfsck->li_bottom, handle);
 764
 765         return rc;
 766 }
 767
 768 static int lfsck_layout_init(const struct lu_env *env,
 769                              struct lfsck_component *com)
 770 {
 771         struct lfsck_layout *lo = com->lc_file_ram;
 772         int rc;
 773
 774         memset(lo, 0, com->lc_file_size);
 775         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
 776         lo->ll_status = LS_INIT;
 777         down_write(&com->lc_sem);
 778         rc = lfsck_layout_store(env, com);
 779         up_write(&com->lc_sem);
 780
 781         return rc;
 782 }
 783
 784 static int fid_is_for_ostobj(const struct lu_env *env, struct dt_device *dt,
 785                              struct dt_object *obj, const struct lu_fid *fid)
 786 {
 787         struct seq_server_site  *ss     = lu_site2seq(dt->dd_lu_dev.ld_site);
 788         struct lu_seq_range      range  = { 0 };
 789         struct lustre_mdt_attrs *lma;
 790         int                      rc;
 791
 792         fld_range_set_any(&range);
 793         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), &range);
 794         if (rc == 0) {
 795                 if (fld_range_is_ost(&range))
 796                         return 1;
 797
 798                 return 0;
 799         }
 800
 801         lma = &lfsck_env_info(env)->lti_lma;
 802         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, lma, sizeof(*lma)),
 803                           XATTR_NAME_LMA, BYPASS_CAPA);
 804         if (rc == sizeof(*lma)) {
 805                 lustre_lma_swab(lma);
 806
 807                 return lma->lma_compat & LMAC_FID_ON_OST ? 1 : 0;
 808         }
 809
 810         rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID, BYPASS_CAPA);
 811
 812         return rc > 0;
 813 }
 814
 815 static struct lfsck_layout_seq *
 816 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
 817 {
 818         struct lfsck_layout_seq *lls;
 819
 820         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
 821                 if (lls->lls_seq == seq)
 822                         return lls;
 823
 824                 if (lls->lls_seq > seq)
 825                         return NULL;
 826         }
 827
 828         return NULL;
 829 }
 830
 831 static void
 832 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
 833                         struct lfsck_layout_seq *lls)
 834 {
 835         struct lfsck_layout_seq *tmp;
 836         struct list_head        *pos = &llsd->llsd_seq_list;
 837
 838         list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
 839                 if (lls->lls_seq < tmp->lls_seq) {
 840                         pos = &tmp->lls_list;
 841                         break;
 842                 }
 843         }
 844         list_add_tail(&lls->lls_list, pos);
 845 }
 846
 847 static int
 848 lfsck_layout_lastid_create(const struct lu_env *env,
 849                            struct lfsck_instance *lfsck,
 850                            struct dt_object *obj)
 851 {
 852         struct lfsck_thread_info *info   = lfsck_env_info(env);
 853         struct lu_attr           *la     = &info->lti_la;
 854         struct dt_object_format  *dof    = &info->lti_dof;
 855         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
 856         struct dt_device         *dt     = lfsck->li_bottom;
 857         struct thandle           *th;
 858         __u64                     lastid = 0;
 859         loff_t                    pos    = 0;
 860         int                       rc;
 861         ENTRY;
 862
 863         CDEBUG(D_LFSCK, "To create LAST_ID for <seq> "LPX64"\n",
 864                fid_seq(lfsck_dto2fid(obj)));
 865
 866         if (bk->lb_param & LPF_DRYRUN)
 867                 return 0;
 868
 869         memset(la, 0, sizeof(*la));
 870         la->la_mode = S_IFREG |  S_IRUGO | S_IWUSR;
 871         la->la_valid = LA_MODE | LA_UID | LA_GID;
 872         dof->dof_type = dt_mode_to_dft(S_IFREG);
 873
 874         th = dt_trans_create(env, dt);
 875         if (IS_ERR(th))
 876                 RETURN(rc = PTR_ERR(th));
 877
 878         rc = dt_declare_create(env, obj, la, NULL, dof, th);
 879         if (rc != 0)
 880                 GOTO(stop, rc);
 881
 882         rc = dt_declare_record_write(env, obj, sizeof(lastid), pos, th);
 883         if (rc != 0)
 884                 GOTO(stop, rc);
 885
 886         rc = dt_trans_start_local(env, dt, th);
 887         if (rc != 0)
 888                 GOTO(stop, rc);
 889
 890         dt_write_lock(env, obj, 0);
 891         if (likely(!dt_object_exists(obj))) {
 892                 rc = dt_create(env, obj, la, NULL, dof, th);
 893                 if (rc == 0)
 894                         rc = dt_record_write(env, obj,
 895                                 lfsck_buf_get(env, &lastid, sizeof(lastid)),
 896                                 &pos, th);
 897         }
 898         dt_write_unlock(env, obj);
 899
 900         GOTO(stop, rc);
 901
 902 stop:
 903         dt_trans_stop(env, dt, th);
 904
 905         return rc;
 906 }
 907
 908 static int
 909 lfsck_layout_lastid_reload(const struct lu_env *env,
 910                            struct lfsck_component *com,
 911                            struct lfsck_layout_seq *lls)
 912 {
 913         __u64   lastid;
 914         loff_t  pos     = 0;
 915         int     rc;
 916
 917         dt_read_lock(env, lls->lls_lastid_obj, 0);
 918         rc = dt_record_read(env, lls->lls_lastid_obj,
 919                             lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
 920         dt_read_unlock(env, lls->lls_lastid_obj);
 921         if (unlikely(rc != 0))
 922                 return rc;
 923
 924         lastid = le64_to_cpu(lastid);
 925         if (lastid < lls->lls_lastid_known) {
 926                 struct lfsck_instance   *lfsck  = com->lc_lfsck;
 927                 struct lfsck_layout     *lo     = com->lc_file_ram;
 928
 929                 lls->lls_lastid = lls->lls_lastid_known;
 930                 lls->lls_dirty = 1;
 931                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
 932                         LASSERT(lfsck->li_out_notify != NULL);
 933
 934                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
 935                                              LE_LASTID_REBUILDING);
 936                         lo->ll_flags |= LF_CRASHED_LASTID;
 937                 }
 938         } else if (lastid >= lls->lls_lastid) {
 939                 lls->lls_lastid = lastid;
 940                 lls->lls_dirty = 0;
 941         }
 942
 943         return 0;
 944 }
 945
 946 static int
 947 lfsck_layout_lastid_store(const struct lu_env *env,
 948                           struct lfsck_component *com)
 949 {
 950         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 951         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
 952         struct dt_device                *dt     = lfsck->li_bottom;
 953         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
 954         struct lfsck_layout_seq         *lls;
 955         struct thandle                  *th;
 956         __u64                            lastid;
 957         int                              rc     = 0;
 958         int                              rc1    = 0;
 959
 960         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
 961                 loff_t pos = 0;
 962
 963                 /* XXX: Add the code back if we really found related
 964                  *      inconsistent cases in the future. */
 965 #if 0
 966                 if (!lls->lls_dirty) {
 967                         /* In OFD, before the pre-creation, the LAST_ID
 968                          * file will be updated firstly, which may hide
 969                          * some potential crashed cases. For example:
 970                          *
 971                          * The old obj1's ID is higher than old LAST_ID
 972                          * but lower than the new LAST_ID, but the LFSCK
 973                          * have not touch the obj1 until the OFD updated
 974                          * the LAST_ID. So the LFSCK does not regard it
 975                          * as crashed case. But when OFD does not create
 976                          * successfully, it will set the LAST_ID as the
 977                          * real created objects' ID, then LFSCK needs to
 978                          * found related inconsistency. */
 979                         rc = lfsck_layout_lastid_reload(env, com, lls);
 980                         if (likely(!lls->lls_dirty))
 981                                 continue;
 982                 }
 983 #endif
 984
 985                 CDEBUG(D_LFSCK, "To sync the LAST_ID for <seq> "LPX64
 986                        " as <oid> "LPU64"\n", lls->lls_seq, lls->lls_lastid);
 987
 988                 if (bk->lb_param & LPF_DRYRUN) {
 989                         lls->lls_dirty = 0;
 990                         continue;
 991                 }
 992
 993                 th = dt_trans_create(env, dt);
 994                 if (IS_ERR(th)) {
 995                         rc1 = PTR_ERR(th);
 996                         CERROR("%s: (1) failed to store "LPX64": rc = %d\n",
 997                                lfsck_lfsck2name(com->lc_lfsck),
 998                                lls->lls_seq, rc1);
 999                         continue;
1000                 }
1001
1002                 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1003                                              sizeof(lastid), pos, th);
1004                 if (rc != 0)
1005                         goto stop;
1006
1007                 rc = dt_trans_start_local(env, dt, th);
1008                 if (rc != 0)
1009                         goto stop;
1010
1011                 lastid = cpu_to_le64(lls->lls_lastid);
1012                 dt_write_lock(env, lls->lls_lastid_obj, 0);
1013                 rc = dt_record_write(env, lls->lls_lastid_obj,
1014                                      lfsck_buf_get(env, &lastid,
1015                                      sizeof(lastid)), &pos, th);
1016                 dt_write_unlock(env, lls->lls_lastid_obj);
1017                 if (rc == 0)
1018                         lls->lls_dirty = 0;
1019
1020 stop:
1021                 dt_trans_stop(env, dt, th);
1022                 if (rc != 0) {
1023                         rc1 = rc;
1024                         CERROR("%s: (2) failed to store "LPX64": rc = %d\n",
1025                                lfsck_lfsck2name(com->lc_lfsck),
1026                                lls->lls_seq, rc1);
1027                 }
1028         }
1029
1030         return rc1;
1031 }
1032
1033 static int
1034 lfsck_layout_lastid_load(const struct lu_env *env,
1035                          struct lfsck_component *com,
1036                          struct lfsck_layout_seq *lls)
1037 {
1038         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1039         struct lfsck_layout     *lo     = com->lc_file_ram;
1040         struct lu_fid           *fid    = &lfsck_env_info(env)->lti_fid;
1041         struct dt_object        *obj;
1042         loff_t                   pos    = 0;
1043         int                      rc;
1044         ENTRY;
1045
1046         lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck->li_bottom));
1047         obj = dt_locate(env, lfsck->li_bottom, fid);
1048         if (IS_ERR(obj))
1049                 RETURN(PTR_ERR(obj));
1050
1051         /* LAST_ID crashed, to be rebuilt */
1052         if (!dt_object_exists(obj)) {
1053                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1054                         LASSERT(lfsck->li_out_notify != NULL);
1055
1056                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1057                                              LE_LASTID_REBUILDING);
1058                         lo->ll_flags |= LF_CRASHED_LASTID;
1059
1060                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1061                             cfs_fail_val > 0) {
1062                                 struct l_wait_info lwi = LWI_TIMEOUT(
1063                                                 cfs_time_seconds(cfs_fail_val),
1064                                                 NULL, NULL);
1065
1066                                 up_write(&com->lc_sem);
1067                                 l_wait_event(lfsck->li_thread.t_ctl_waitq,
1068                                              !thread_is_running(&lfsck->li_thread),
1069                                              &lwi);
1070                                 down_write(&com->lc_sem);
1071                         }
1072                 }
1073
1074                 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1075         } else {
1076                 dt_read_lock(env, obj, 0);
1077                 rc = dt_read(env, obj,
1078                         lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1079                         &pos);
1080                 dt_read_unlock(env, obj);
1081                 if (rc != 0 && rc != sizeof(__u64))
1082                         GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1083
1084                 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1085                         LASSERT(lfsck->li_out_notify != NULL);
1086
1087                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1088                                              LE_LASTID_REBUILDING);
1089                         lo->ll_flags |= LF_CRASHED_LASTID;
1090                 }
1091
1092                 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1093                 rc = 0;
1094         }
1095
1096         GOTO(out, rc);
1097
1098 out:
1099         if (rc != 0)
1100                 lfsck_object_put(env, obj);
1101         else
1102                 lls->lls_lastid_obj = obj;
1103
1104         return rc;
1105 }
1106
1107 static int lfsck_layout_master_async_interpret(const struct lu_env *env,
1108                                                struct ptlrpc_request *req,
1109                                                void *args, int rc)
1110 {
1111         struct lfsck_async_interpret_args *laia = args;
1112         struct lfsck_component            *com  = laia->laia_com;
1113         struct lfsck_layout_master_data   *llmd = com->lc_data;
1114         struct lfsck_tgt_descs            *ltds = laia->laia_ltds;
1115         struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
1116         struct lfsck_request              *lr   = laia->laia_lr;
1117
1118         switch (lr->lr_event) {
1119         case LE_START:
1120                 if (rc != 0) {
1121                         struct lfsck_layout *lo = com->lc_file_ram;
1122
1123                         CERROR("%s: fail to notify %s %x for layout start: "
1124                                "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
1125                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1126                                ltd->ltd_index, rc);
1127                         lo->ll_flags |= LF_INCOMPLETE;
1128                         break;
1129                 }
1130
1131                 spin_lock(&ltds->ltd_lock);
1132                 if (ltd->ltd_dead || ltd->ltd_layout_done) {
1133                         spin_unlock(&ltds->ltd_lock);
1134                         break;
1135                 }
1136
1137                 if (lr->lr_flags & LEF_TO_OST) {
1138                         if (list_empty(&ltd->ltd_layout_list))
1139                                 list_add_tail(&ltd->ltd_layout_list,
1140                                               &llmd->llmd_ost_list);
1141                         if (list_empty(&ltd->ltd_layout_phase_list))
1142                                 list_add_tail(&ltd->ltd_layout_phase_list,
1143                                               &llmd->llmd_ost_phase1_list);
1144                 } else {
1145                         if (list_empty(&ltd->ltd_layout_list))
1146                                 list_add_tail(&ltd->ltd_layout_list,
1147                                               &llmd->llmd_mdt_list);
1148                         if (list_empty(&ltd->ltd_layout_phase_list))
1149                                 list_add_tail(&ltd->ltd_layout_phase_list,
1150                                               &llmd->llmd_mdt_phase1_list);
1151                 }
1152                 spin_unlock(&ltds->ltd_lock);
1153                 break;
1154         case LE_STOP:
1155         case LE_PHASE1_DONE:
1156         case LE_PHASE2_DONE:
1157         case LE_PEER_EXIT:
1158                 if (rc != 0 && rc != -EALREADY)
1159                         CWARN("%s: fail to notify %s %x for layout: "
1160                               "event = %d, rc = %d\n",
1161                               lfsck_lfsck2name(com->lc_lfsck),
1162                               (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1163                               ltd->ltd_index, lr->lr_event, rc);
1164                 break;
1165         case LE_QUERY: {
1166                 struct lfsck_reply *reply;
1167
1168                 if (rc != 0) {
1169                         spin_lock(&ltds->ltd_lock);
1170                         list_del_init(&ltd->ltd_layout_phase_list);
1171                         list_del_init(&ltd->ltd_layout_list);
1172                         spin_unlock(&ltds->ltd_lock);
1173                         break;
1174                 }
1175
1176                 reply = req_capsule_server_get(&req->rq_pill,
1177                                                &RMF_LFSCK_REPLY);
1178                 if (reply == NULL) {
1179                         rc = -EPROTO;
1180                         CERROR("%s: invalid return value: rc = %d\n",
1181                                lfsck_lfsck2name(com->lc_lfsck), rc);
1182                         spin_lock(&ltds->ltd_lock);
1183                         list_del_init(&ltd->ltd_layout_phase_list);
1184                         list_del_init(&ltd->ltd_layout_list);
1185                         spin_unlock(&ltds->ltd_lock);
1186                         break;
1187                 }
1188
1189                 switch (reply->lr_status) {
1190                 case LS_SCANNING_PHASE1:
1191                         break;
1192                 case LS_SCANNING_PHASE2:
1193                         spin_lock(&ltds->ltd_lock);
1194                         list_del_init(&ltd->ltd_layout_phase_list);
1195                         if (ltd->ltd_dead || ltd->ltd_layout_done) {
1196                                 spin_unlock(&ltds->ltd_lock);
1197                                 break;
1198                         }
1199
1200                         if (lr->lr_flags & LEF_TO_OST)
1201                                 list_add_tail(&ltd->ltd_layout_phase_list,
1202                                               &llmd->llmd_ost_phase2_list);
1203                         else
1204                                 list_add_tail(&ltd->ltd_layout_phase_list,
1205                                               &llmd->llmd_mdt_phase2_list);
1206                         spin_unlock(&ltds->ltd_lock);
1207                         break;
1208                 default:
1209                         spin_lock(&ltds->ltd_lock);
1210                         list_del_init(&ltd->ltd_layout_phase_list);
1211                         list_del_init(&ltd->ltd_layout_list);
1212                         spin_unlock(&ltds->ltd_lock);
1213                         break;
1214                 }
1215                 break;
1216         }
1217         default:
1218                 CERROR("%s: unexpected event: rc = %d\n",
1219                        lfsck_lfsck2name(com->lc_lfsck), lr->lr_event);
1220                 break;
1221         }
1222
1223         if (!laia->laia_shared) {
1224                 lfsck_tgt_put(ltd);
1225                 lfsck_component_put(env, com);
1226         }
1227
1228         return 0;
1229 }
1230
1231 static int lfsck_layout_master_query_others(const struct lu_env *env,
1232                                             struct lfsck_component *com)
1233 {
1234         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1235         struct lfsck_request              *lr    = &info->lti_lr;
1236         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1237         struct lfsck_instance             *lfsck = com->lc_lfsck;
1238         struct lfsck_layout_master_data   *llmd  = com->lc_data;
1239         struct ptlrpc_request_set         *set;
1240         struct lfsck_tgt_descs            *ltds;
1241         struct lfsck_tgt_desc             *ltd;
1242         struct list_head                  *head;
1243         int                                rc    = 0;
1244         int                                rc1   = 0;
1245         ENTRY;
1246
1247         set = ptlrpc_prep_set();
1248         if (set == NULL)
1249                 RETURN(-ENOMEM);
1250
1251         llmd->llmd_touch_gen++;
1252         memset(lr, 0, sizeof(*lr));
1253         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1254         lr->lr_event = LE_QUERY;
1255         lr->lr_active = LT_LAYOUT;
1256         laia->laia_com = com;
1257         laia->laia_lr = lr;
1258         laia->laia_shared = 0;
1259
1260         if (!list_empty(&llmd->llmd_mdt_phase1_list)) {
1261                 ltds = &lfsck->li_mdt_descs;
1262                 lr->lr_flags = 0;
1263                 head = &llmd->llmd_mdt_phase1_list;
1264         } else {
1265
1266 again:
1267                 ltds = &lfsck->li_ost_descs;
1268                 lr->lr_flags = LEF_TO_OST;
1269                 head = &llmd->llmd_ost_phase1_list;
1270         }
1271
1272         laia->laia_ltds = ltds;
1273         spin_lock(&ltds->ltd_lock);
1274         while (!list_empty(head)) {
1275                 ltd = list_entry(head->next,
1276                                  struct lfsck_tgt_desc,
1277                                  ltd_layout_phase_list);
1278                 if (ltd->ltd_layout_gen == llmd->llmd_touch_gen)
1279                         break;
1280
1281                 ltd->ltd_layout_gen = llmd->llmd_touch_gen;
1282                 list_del(&ltd->ltd_layout_phase_list);
1283                 list_add_tail(&ltd->ltd_layout_phase_list, head);
1284                 atomic_inc(&ltd->ltd_ref);
1285                 laia->laia_ltd = ltd;
1286                 spin_unlock(&ltds->ltd_lock);
1287                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1288                                          lfsck_layout_master_async_interpret,
1289                                          laia, LFSCK_QUERY);
1290                 if (rc != 0) {
1291                         CERROR("%s: fail to query %s %x for layout: rc = %d\n",
1292                                lfsck_lfsck2name(lfsck),
1293                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1294                                ltd->ltd_index, rc);
1295                         lfsck_tgt_put(ltd);
1296                         rc1 = rc;
1297                 }
1298                 spin_lock(&ltds->ltd_lock);
1299         }
1300         spin_unlock(&ltds->ltd_lock);
1301
1302         rc = ptlrpc_set_wait(set);
1303         if (rc < 0) {
1304                 ptlrpc_set_destroy(set);
1305                 RETURN(rc);
1306         }
1307
1308         if (!(lr->lr_flags & LEF_TO_OST) &&
1309             list_empty(&llmd->llmd_mdt_phase1_list))
1310                 goto again;
1311
1312         ptlrpc_set_destroy(set);
1313
1314         RETURN(rc1 != 0 ? rc1 : rc);
1315 }
1316
1317 static inline bool
1318 lfsck_layout_master_to_orphan(struct lfsck_layout_master_data *llmd)
1319 {
1320         return list_empty(&llmd->llmd_mdt_phase1_list) &&
1321                (!list_empty(&llmd->llmd_ost_phase2_list) ||
1322                 list_empty(&llmd->llmd_ost_phase1_list));
1323 }
1324
1325 static int lfsck_layout_master_notify_others(const struct lu_env *env,
1326                                              struct lfsck_component *com,
1327                                              struct lfsck_request *lr)
1328 {
1329         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1330         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1331         struct lfsck_instance             *lfsck = com->lc_lfsck;
1332         struct lfsck_layout_master_data   *llmd  = com->lc_data;
1333         struct lfsck_layout               *lo    = com->lc_file_ram;
1334         struct lfsck_bookmark             *bk    = &lfsck->li_bookmark_ram;
1335         struct ptlrpc_request_set         *set;
1336         struct lfsck_tgt_descs            *ltds;
1337         struct lfsck_tgt_desc             *ltd;
1338         struct lfsck_tgt_desc             *next;
1339         struct list_head                  *head;
1340         __u32                              idx;
1341         int                                rc    = 0;
1342         ENTRY;
1343
1344         set = ptlrpc_prep_set();
1345         if (set == NULL)
1346                 RETURN(-ENOMEM);
1347
1348         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1349         lr->lr_active = LT_LAYOUT;
1350         laia->laia_com = com;
1351         laia->laia_lr = lr;
1352         laia->laia_shared = 0;
1353         switch (lr->lr_event) {
1354         case LE_START:
1355                 /* Notify OSTs firstly, then handle other MDTs if needed. */
1356                 ltds = &lfsck->li_ost_descs;
1357                 laia->laia_ltds = ltds;
1358                 down_read(&ltds->ltd_rw_sem);
1359                 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1360                         ltd = lfsck_tgt_get(ltds, idx);
1361                         LASSERT(ltd != NULL);
1362
1363                         laia->laia_ltd = ltd;
1364                         ltd->ltd_layout_done = 0;
1365                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1366                                         lfsck_layout_master_async_interpret,
1367                                         laia, LFSCK_NOTIFY);
1368                         if (rc != 0) {
1369                                 CERROR("%s: fail to notify %s %x for layout "
1370                                        "start: rc = %d\n",
1371                                        lfsck_lfsck2name(lfsck),
1372                                        (lr->lr_flags & LEF_TO_OST) ? "OST" :
1373                                        "MDT", idx, rc);
1374                                 lfsck_tgt_put(ltd);
1375                                 lo->ll_flags |= LF_INCOMPLETE;
1376                         }
1377                 }
1378                 up_read(&ltds->ltd_rw_sem);
1379
1380                 /* Sync up */
1381                 rc = ptlrpc_set_wait(set);
1382                 if (rc < 0) {
1383                         ptlrpc_set_destroy(set);
1384                         RETURN(rc);
1385                 }
1386
1387                 if (!(bk->lb_param & LPF_ALL_TGT))
1388                         break;
1389
1390                 /* link other MDT targets locallly. */
1391                 spin_lock(&ltds->ltd_lock);
1392                 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1393                         ltd = LTD_TGT(ltds, idx);
1394                         LASSERT(ltd != NULL);
1395
1396                         if (!list_empty(&ltd->ltd_layout_list))
1397                                 continue;
1398
1399                         list_add_tail(&ltd->ltd_layout_list,
1400                                       &llmd->llmd_mdt_list);
1401                         list_add_tail(&ltd->ltd_layout_phase_list,
1402                                       &llmd->llmd_mdt_phase1_list);
1403                 }
1404                 spin_unlock(&ltds->ltd_lock);
1405                 break;
1406         case LE_STOP:
1407         case LE_PHASE2_DONE:
1408         case LE_PEER_EXIT: {
1409                 /* Handle other MDTs firstly if needed, then notify the OSTs. */
1410                 if (bk->lb_param & LPF_ALL_TGT) {
1411                         head = &llmd->llmd_mdt_list;
1412                         ltds = &lfsck->li_mdt_descs;
1413                         if (lr->lr_event == LE_STOP) {
1414                                 /* unlink other MDT targets locallly. */
1415                                 spin_lock(&ltds->ltd_lock);
1416                                 list_for_each_entry_safe(ltd, next, head,
1417                                                          ltd_layout_list) {
1418                                         list_del_init(&ltd->ltd_layout_phase_list);
1419                                         list_del_init(&ltd->ltd_layout_list);
1420                                 }
1421                                 spin_unlock(&ltds->ltd_lock);
1422
1423                                 lr->lr_flags |= LEF_TO_OST;
1424                                 head = &llmd->llmd_ost_list;
1425                                 ltds = &lfsck->li_ost_descs;
1426                         } else {
1427                                 lr->lr_flags &= ~LEF_TO_OST;
1428                         }
1429                 } else {
1430                         lr->lr_flags |= LEF_TO_OST;
1431                         head = &llmd->llmd_ost_list;
1432                         ltds = &lfsck->li_ost_descs;
1433                 }
1434
1435 again:
1436                 laia->laia_ltds = ltds;
1437                 spin_lock(&ltds->ltd_lock);
1438                 while (!list_empty(head)) {
1439                         ltd = list_entry(head->next, struct lfsck_tgt_desc,
1440                                          ltd_layout_list);
1441                         if (!list_empty(&ltd->ltd_layout_phase_list))
1442                                 list_del_init(&ltd->ltd_layout_phase_list);
1443                         list_del_init(&ltd->ltd_layout_list);
1444                         atomic_inc(&ltd->ltd_ref);
1445                         laia->laia_ltd = ltd;
1446                         spin_unlock(&ltds->ltd_lock);
1447                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1448                                         lfsck_layout_master_async_interpret,
1449                                         laia, LFSCK_NOTIFY);
1450                         if (rc != 0) {
1451                                 CERROR("%s: fail to notify %s %x for layout "
1452                                        "stop/phase2: rc = %d\n",
1453                                        lfsck_lfsck2name(lfsck),
1454                                        (lr->lr_flags & LEF_TO_OST) ? "OST" :
1455                                        "MDT", ltd->ltd_index, rc);
1456                                 lfsck_tgt_put(ltd);
1457                         }
1458                         spin_lock(&ltds->ltd_lock);
1459                 }
1460                 spin_unlock(&ltds->ltd_lock);
1461
1462                 rc = ptlrpc_set_wait(set);
1463                 if (rc < 0) {
1464                         ptlrpc_set_destroy(set);
1465                         RETURN(rc);
1466                 }
1467
1468                 if (!(lr->lr_flags & LEF_TO_OST)) {
1469                         lr->lr_flags |= LEF_TO_OST;
1470                         head = &llmd->llmd_ost_list;
1471                         ltds = &lfsck->li_ost_descs;
1472                         goto again;
1473                 }
1474                 break;
1475         }
1476         case LE_PHASE1_DONE:
1477                 llmd->llmd_touch_gen++;
1478                 ltds = &lfsck->li_mdt_descs;
1479                 laia->laia_ltds = ltds;
1480                 spin_lock(&ltds->ltd_lock);
1481                 while (!list_empty(&llmd->llmd_mdt_phase1_list)) {
1482                         ltd = list_entry(llmd->llmd_mdt_phase1_list.next,
1483                                          struct lfsck_tgt_desc,
1484                                          ltd_layout_phase_list);
1485                         if (ltd->ltd_layout_gen == llmd->llmd_touch_gen)
1486                                 break;
1487
1488                         ltd->ltd_layout_gen = llmd->llmd_touch_gen;
1489                         list_del_init(&ltd->ltd_layout_phase_list);
1490                         list_add_tail(&ltd->ltd_layout_phase_list,
1491                                       &llmd->llmd_mdt_phase1_list);
1492                         atomic_inc(&ltd->ltd_ref);
1493                         laia->laia_ltd = ltd;
1494                         spin_unlock(&ltds->ltd_lock);
1495                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1496                                         lfsck_layout_master_async_interpret,
1497                                         laia, LFSCK_NOTIFY);
1498                         if (rc != 0) {
1499                                 CERROR("%s: fail to notify MDT %x for layout "
1500                                        "phase1 done: rc = %d\n",
1501                                        lfsck_lfsck2name(lfsck),
1502                                        ltd->ltd_index, rc);
1503                                 lfsck_tgt_put(ltd);
1504                         }
1505                         spin_lock(&ltds->ltd_lock);
1506                 }
1507                 spin_unlock(&ltds->ltd_lock);
1508                 break;
1509         default:
1510                 CERROR("%s: unexpected LFSCK event: rc = %d\n",
1511                        lfsck_lfsck2name(lfsck), lr->lr_event);
1512                 rc = -EINVAL;
1513                 break;
1514         }
1515
1516         rc = ptlrpc_set_wait(set);
1517         ptlrpc_set_destroy(set);
1518
1519         RETURN(rc);
1520 }
1521
1522 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1523                                            struct lfsck_component *com,
1524                                            int rc)
1525 {
1526         struct lfsck_instance   *lfsck = com->lc_lfsck;
1527         struct lfsck_layout     *lo    = com->lc_file_ram;
1528         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
1529
1530         down_write(&com->lc_sem);
1531
1532         lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1533                                 HALF_SEC - lfsck->li_time_last_checkpoint);
1534         lo->ll_time_last_checkpoint = cfs_time_current_sec();
1535         lo->ll_objs_checked_phase2 += com->lc_new_checked;
1536
1537         if (rc > 0) {
1538                 com->lc_journal = 0;
1539                 if (lo->ll_flags & LF_INCOMPLETE)
1540                         lo->ll_status = LS_PARTIAL;
1541                 else
1542                         lo->ll_status = LS_COMPLETED;
1543                 if (!(bk->lb_param & LPF_DRYRUN))
1544                         lo->ll_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
1545                 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1546                 lo->ll_success_count++;
1547         } else if (rc == 0) {
1548                 lo->ll_status = lfsck->li_status;
1549                 if (lo->ll_status == 0)
1550                         lo->ll_status = LS_STOPPED;
1551         } else {
1552                 lo->ll_status = LS_FAILED;
1553         }
1554
1555         if (lo->ll_status != LS_PAUSED) {
1556                 spin_lock(&lfsck->li_lock);
1557                 list_del_init(&com->lc_link);
1558                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
1559                 spin_unlock(&lfsck->li_lock);
1560         }
1561
1562         rc = lfsck_layout_store(env, com);
1563
1564         up_write(&com->lc_sem);
1565
1566         return rc;
1567 }
1568
1569 static int lfsck_layout_lock(const struct lu_env *env,
1570                              struct lfsck_component *com,
1571                              struct dt_object *obj,
1572                              struct lustre_handle *lh, __u64 bits)
1573 {
1574         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1575         ldlm_policy_data_t              *policy = &info->lti_policy;
1576         struct ldlm_res_id              *resid  = &info->lti_resid;
1577         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1578         __u64                            flags  = LDLM_FL_ATOMIC_CB;
1579         int                              rc;
1580
1581         LASSERT(lfsck->li_namespace != NULL);
1582
1583         memset(policy, 0, sizeof(*policy));
1584         policy->l_inodebits.bits = bits;
1585         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
1586         rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_IBITS,
1587                                     policy, LCK_EX, &flags, ldlm_blocking_ast,
1588                                     ldlm_completion_ast, NULL, NULL, 0,
1589                                     LVB_T_NONE, NULL, lh);
1590         if (rc == ELDLM_OK) {
1591                 rc = 0;
1592         } else {
1593                 memset(lh, 0, sizeof(*lh));
1594                 rc = -EIO;
1595         }
1596
1597         return rc;
1598 }
1599
1600 static void lfsck_layout_unlock(struct lustre_handle *lh)
1601 {
1602         if (lustre_handle_is_used(lh)) {
1603                 ldlm_lock_decref(lh, LCK_EX);
1604                 memset(lh, 0, sizeof(*lh));
1605         }
1606 }
1607
1608 static int lfsck_layout_trans_stop(const struct lu_env *env,
1609                                    struct dt_device *dev,
1610                                    struct thandle *handle, int result)
1611 {
1612         int rc;
1613
1614         handle->th_result = result;
1615         rc = dt_trans_stop(env, dev, handle);
1616         if (rc > 0)
1617                 rc = 0;
1618         else if (rc == 0)
1619                 rc = 1;
1620
1621         return rc;
1622 }
1623
1624 static int lfsck_layout_scan_orphan(const struct lu_env *env,
1625                                     struct lfsck_component *com,
1626                                     struct lfsck_tgt_desc *ltd)
1627 {
1628         /* XXX: To be extended in other patch. */
1629
1630         return 0;
1631 }
1632
1633 /* For the MDT-object with dangling reference, we need to re-create
1634  * the missed OST-object with the known FID/owner information. */
1635 static int lfsck_layout_recreate_ostobj(const struct lu_env *env,
1636                                         struct lfsck_component *com,
1637                                         struct lfsck_layout_req *llr,
1638                                         struct lu_attr *la)
1639 {
1640         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1641         struct filter_fid               *pfid   = &info->lti_new_pfid;
1642         struct dt_allocation_hint       *hint   = &info->lti_hint;
1643         struct dt_object                *parent = llr->llr_parent->llo_obj;
1644         struct dt_object                *child  = llr->llr_child;
1645         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
1646         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
1647         struct thandle                  *handle;
1648         struct lu_buf                   *buf;
1649         struct lustre_handle             lh     = { 0 };
1650         int                              rc;
1651         ENTRY;
1652
1653         CDEBUG(D_LFSCK, "Repair dangling reference for: parent "DFID
1654                ", child "DFID", OST-index %u, stripe-index %u, owner %u:%u\n",
1655                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
1656                llr->llr_ost_idx, llr->llr_lov_idx, la->la_uid, la->la_gid);
1657
1658         rc = lfsck_layout_lock(env, com, parent, &lh,
1659                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
1660         if (rc != 0)
1661                 RETURN(rc);
1662
1663         handle = dt_trans_create(env, dev);
1664         if (IS_ERR(handle))
1665                 GOTO(unlock1, rc = PTR_ERR(handle));
1666
1667         hint->dah_parent = NULL;
1668         hint->dah_mode = 0;
1669         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
1670         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
1671         pfid->ff_parent.f_ver = cpu_to_le32(llr->llr_lov_idx);
1672         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
1673
1674         rc = dt_declare_create(env, child, la, hint, NULL, handle);
1675         if (rc != 0)
1676                 GOTO(stop, rc);
1677
1678         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
1679                                   LU_XATTR_CREATE, handle);
1680         if (rc != 0)
1681                 GOTO(stop, rc);
1682
1683         rc = dt_trans_start(env, dev, handle);
1684         if (rc != 0)
1685                 GOTO(stop, rc);
1686
1687         dt_read_lock(env, parent, 0);
1688         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
1689                 GOTO(unlock2, rc = 1);
1690
1691         rc = dt_create(env, child, la, hint, NULL, handle);
1692         if (rc != 0)
1693                 GOTO(unlock2, rc);
1694
1695         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
1696                           handle, BYPASS_CAPA);
1697
1698         GOTO(unlock2, rc);
1699
1700 unlock2:
1701         dt_read_unlock(env, parent);
1702
1703 stop:
1704         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
1705
1706 unlock1:
1707         lfsck_layout_unlock(&lh);
1708
1709         return rc;
1710 }
1711
1712 /* If the OST-object does not recognize the MDT-object as its parent, and
1713  * there is no other MDT-object claims as its parent, then just trust the
1714  * given MDT-object as its parent. So update the OST-object filter_fid. */
1715 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
1716                                               struct lfsck_component *com,
1717                                               struct lfsck_layout_req *llr,
1718                                               const struct lu_attr *pla)
1719 {
1720         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1721         struct filter_fid               *pfid   = &info->lti_new_pfid;
1722         struct lu_attr                  *tla    = &info->lti_la3;
1723         struct dt_object                *parent = llr->llr_parent->llo_obj;
1724         struct dt_object                *child  = llr->llr_child;
1725         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
1726         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
1727         struct thandle                  *handle;
1728         struct lu_buf                   *buf;
1729         struct lustre_handle             lh     = { 0 };
1730         int                              rc;
1731         ENTRY;
1732
1733         CDEBUG(D_LFSCK, "Repair unmatched MDT-OST pair for: parent "DFID
1734                ", child "DFID", OST-index %u, stripe-index %u, owner %u:%u\n",
1735                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
1736                llr->llr_ost_idx, llr->llr_lov_idx, pla->la_uid, pla->la_gid);
1737
1738         rc = lfsck_layout_lock(env, com, parent, &lh,
1739                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
1740         if (rc != 0)
1741                 RETURN(rc);
1742
1743         handle = dt_trans_create(env, dev);
1744         if (IS_ERR(handle))
1745                 GOTO(unlock1, rc = PTR_ERR(handle));
1746
1747         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
1748         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
1749         /* The ff_parent->f_ver is not the real parent fid->f_ver. Instead,
1750          * it is the OST-object index in the parent MDT-object layout. */
1751         pfid->ff_parent.f_ver = cpu_to_le32(llr->llr_lov_idx);
1752         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
1753
1754         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
1755         if (rc != 0)
1756                 GOTO(stop, rc);
1757
1758         tla->la_valid = LA_UID | LA_GID;
1759         tla->la_uid = pla->la_uid;
1760         tla->la_gid = pla->la_gid;
1761         rc = dt_declare_attr_set(env, child, tla, handle);
1762         if (rc != 0)
1763                 GOTO(stop, rc);
1764
1765         rc = dt_trans_start(env, dev, handle);
1766         if (rc != 0)
1767                 GOTO(stop, rc);
1768
1769         dt_write_lock(env, parent, 0);
1770         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
1771                 GOTO(unlock2, rc = 1);
1772
1773         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle,
1774                           BYPASS_CAPA);
1775         if (rc != 0)
1776                 GOTO(unlock2, rc);
1777
1778         /* Get the latest parent's owner. */
1779         rc = dt_attr_get(env, parent, tla, BYPASS_CAPA);
1780         if (rc != 0)
1781                 GOTO(unlock2, rc);
1782
1783         tla->la_valid = LA_UID | LA_GID;
1784         rc = dt_attr_set(env, child, tla, handle, BYPASS_CAPA);
1785
1786         GOTO(unlock2, rc);
1787
1788 unlock2:
1789         dt_write_unlock(env, parent);
1790
1791 stop:
1792         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
1793
1794 unlock1:
1795         lfsck_layout_unlock(&lh);
1796
1797         return rc;
1798 }
1799
1800 /* If there are more than one MDT-objects claim as the OST-object's parent,
1801  * and the OST-object only recognizes one of them, then we need to generate
1802  * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
1803 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
1804                                                    struct lfsck_component *com,
1805                                                    struct lfsck_layout_req *llr,
1806                                                    struct lu_attr *la,
1807                                                    struct lu_buf *buf)
1808 {
1809         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1810         struct dt_allocation_hint       *hint   = &info->lti_hint;
1811         struct dt_object_format         *dof    = &info->lti_dof;
1812         struct dt_device                *pdev   = com->lc_lfsck->li_next;
1813         struct ost_id                   *oi     = &info->lti_oi;
1814         struct dt_object                *parent = llr->llr_parent->llo_obj;
1815         struct dt_device                *cdev   = lfsck_obj2dt_dev(llr->llr_child);
1816         struct dt_object                *child  = NULL;
1817         struct lu_device                *d      = &cdev->dd_lu_dev;
1818         struct lu_object                *o      = NULL;
1819         struct thandle                  *handle;
1820         struct lov_mds_md_v1            *lmm;
1821         struct lov_ost_data_v1          *objs;
1822         struct lustre_handle             lh     = { 0 };
1823         __u32                            magic;
1824         int                              rc;
1825         ENTRY;
1826
1827         CDEBUG(D_LFSCK, "Repair multiple references for: parent "DFID
1828                ", OST-index %u, stripe-index %u, owner %u:%u\n",
1829                PFID(lfsck_dto2fid(parent)), llr->llr_ost_idx,
1830                llr->llr_lov_idx, la->la_uid, la->la_gid);
1831
1832         rc = lfsck_layout_lock(env, com, parent, &lh,
1833                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
1834         if (rc != 0)
1835                 RETURN(rc);
1836
1837         handle = dt_trans_create(env, pdev);
1838         if (IS_ERR(handle))
1839                 GOTO(unlock1, rc = PTR_ERR(handle));
1840
1841         o = lu_object_anon(env, d, NULL);
1842         if (IS_ERR(o))
1843                 GOTO(stop, rc = PTR_ERR(o));
1844
1845         child = container_of(o, struct dt_object, do_lu);
1846         o = lu_object_locate(o->lo_header, d->ld_type);
1847         if (unlikely(o == NULL))
1848                 GOTO(stop, rc = -EINVAL);
1849
1850         child = container_of(o, struct dt_object, do_lu);
1851         la->la_valid = LA_UID | LA_GID;
1852         hint->dah_parent = NULL;
1853         hint->dah_mode = 0;
1854         dof->dof_type = DFT_REGULAR;
1855         rc = dt_declare_create(env, child, la, NULL, NULL, handle);
1856         if (rc != 0)
1857                 GOTO(stop, rc);
1858
1859         rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
1860                                   LU_XATTR_REPLACE, handle);
1861         if (rc != 0)
1862                 GOTO(stop, rc);
1863
1864         rc = dt_trans_start(env, pdev, handle);
1865         if (rc != 0)
1866                 GOTO(stop, rc);
1867
1868         dt_write_lock(env, parent, 0);
1869         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
1870                 GOTO(unlock2, rc = 0);
1871
1872         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV, BYPASS_CAPA);
1873         if (unlikely(rc == 0 || rc == -ENODATA || rc == -ERANGE))
1874                 GOTO(unlock2, rc = 0);
1875
1876         lmm = buf->lb_buf;
1877         rc = lfsck_layout_verify_header(lmm);
1878         if (rc != 0)
1879                 GOTO(unlock2, rc);
1880
1881         /* Someone change layout during the LFSCK, no need to repair then. */
1882         if (le16_to_cpu(lmm->lmm_layout_gen) != llr->llr_parent->llo_gen)
1883                 GOTO(unlock2, rc = 0);
1884
1885         rc = dt_create(env, child, la, hint, dof, handle);
1886         if (rc != 0)
1887                 GOTO(unlock2, rc);
1888
1889         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
1890          * been verified in lfsck_layout_verify_header() already. If some
1891          * new magic introduced in the future, then layout LFSCK needs to
1892          * be updated also. */
1893         magic = le32_to_cpu(lmm->lmm_magic);
1894         if (magic == LOV_MAGIC_V1) {
1895                 objs = &(lmm->lmm_objects[0]);
1896         } else {
1897                 LASSERT(magic == LOV_MAGIC_V3);
1898                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1899         }
1900
1901         lmm->lmm_layout_gen = cpu_to_le16(llr->llr_parent->llo_gen + 1);
1902         fid_to_ostid(lu_object_fid(&child->do_lu), oi);
1903         ostid_cpu_to_le(oi, &objs[llr->llr_lov_idx].l_ost_oi);
1904         objs[llr->llr_lov_idx].l_ost_gen = cpu_to_le32(0);
1905         objs[llr->llr_lov_idx].l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
1906         rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV,
1907                           LU_XATTR_REPLACE, handle, BYPASS_CAPA);
1908
1909         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1910
1911 unlock2:
1912         dt_write_unlock(env, parent);
1913
1914 stop:
1915         if (child != NULL)
1916                 lu_object_put(env, &child->do_lu);
1917
1918         dt_trans_stop(env, pdev, handle);
1919
1920 unlock1:
1921         lfsck_layout_unlock(&lh);
1922
1923         return rc;
1924 }
1925
1926 /* If the MDT-object and the OST-object have different owner information,
1927  * then trust the MDT-object, because the normal chown/chgrp handle order
1928  * is from MDT to OST, and it is possible that some chown/chgrp operation
1929  * is partly done. */
1930 static int lfsck_layout_repair_owner(const struct lu_env *env,
1931                                      struct lfsck_component *com,
1932                                      struct lfsck_layout_req *llr,
1933                                      struct lu_attr *pla)
1934 {
1935         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1936         struct lu_attr                  *tla    = &info->lti_la3;
1937         struct dt_object                *parent = llr->llr_parent->llo_obj;
1938         struct dt_object                *child  = llr->llr_child;
1939         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
1940         struct thandle                  *handle;
1941         int                              rc;
1942         ENTRY;
1943
1944         CDEBUG(D_LFSCK, "Repair inconsistent file owner for: parent "DFID
1945                ", child "DFID", OST-index %u, stripe-index %u, owner %u:%u\n",
1946                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
1947                llr->llr_ost_idx, llr->llr_lov_idx, pla->la_uid, pla->la_gid);
1948
1949         handle = dt_trans_create(env, dev);
1950         if (IS_ERR(handle))
1951                 RETURN(PTR_ERR(handle));
1952
1953         tla->la_uid = pla->la_uid;
1954         tla->la_gid = pla->la_gid;
1955         tla->la_valid = LA_UID | LA_GID;
1956         rc = dt_declare_attr_set(env, child, tla, handle);
1957         if (rc != 0)
1958                 GOTO(stop, rc);
1959
1960         rc = dt_trans_start(env, dev, handle);
1961         if (rc != 0)
1962                 GOTO(stop, rc);
1963
1964         /* Use the dt_object lock to serialize with destroy and attr_set. */
1965         dt_read_lock(env, parent, 0);
1966         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
1967                 GOTO(unlock, rc = 1);
1968
1969         /* Get the latest parent's owner. */
1970         rc = dt_attr_get(env, parent, tla, BYPASS_CAPA);
1971         if (rc != 0) {
1972                 CWARN("%s: fail to get the latest parent's ("DFID") owner, "
1973                       "not sure whether some others chown/chgrp during the "
1974                       "LFSCK: rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
1975                       PFID(lfsck_dto2fid(parent)), rc);
1976
1977                 GOTO(unlock, rc);
1978         }
1979
1980         /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
1981         if (unlikely(tla->la_uid != pla->la_uid ||
1982                      tla->la_gid != pla->la_gid))
1983                 GOTO(unlock, rc = 1);
1984
1985         tla->la_valid = LA_UID | LA_GID;
1986         rc = dt_attr_set(env, child, tla, handle, BYPASS_CAPA);
1987
1988         GOTO(unlock, rc);
1989
1990 unlock:
1991         dt_read_unlock(env, parent);
1992
1993 stop:
1994         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
1995
1996         return rc;
1997 }
1998
1999 /* Check whether the OST-object correctly back points to the
2000  * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
2001 static int lfsck_layout_check_parent(const struct lu_env *env,
2002                                      struct lfsck_component *com,
2003                                      struct dt_object *parent,
2004                                      const struct lu_fid *pfid,
2005                                      const struct lu_fid *cfid,
2006                                      const struct lu_attr *pla,
2007                                      const struct lu_attr *cla,
2008                                      struct lfsck_layout_req *llr,
2009                                      struct lu_buf *lov_ea, __u32 idx)
2010 {
2011         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2012         struct lu_buf                   *buf    = &info->lti_big_buf;
2013         struct dt_object                *tobj;
2014         struct lov_mds_md_v1            *lmm;
2015         struct lov_ost_data_v1          *objs;
2016         int                              rc;
2017         int                              i;
2018         __u32                            magic;
2019         __u16                            count;
2020         ENTRY;
2021
2022         if (fid_is_zero(pfid)) {
2023                 /* client never wrote. */
2024                 if (cla->la_size == 0 && cla->la_blocks == 0) {
2025                         if (unlikely(cla->la_uid != pla->la_uid ||
2026                                      cla->la_gid != pla->la_gid))
2027                                 RETURN (LLIT_INCONSISTENT_OWNER);
2028
2029                         RETURN(0);
2030                 }
2031
2032                 RETURN(LLIT_UNMATCHED_PAIR);
2033         }
2034
2035         if (unlikely(!fid_is_sane(pfid)))
2036                 RETURN(LLIT_UNMATCHED_PAIR);
2037
2038         if (lu_fid_eq(pfid, lu_object_fid(&parent->do_lu))) {
2039                 if (llr->llr_lov_idx == idx)
2040                         RETURN(0);
2041
2042                 RETURN(LLIT_UNMATCHED_PAIR);
2043         }
2044
2045         tobj = lfsck_object_find(env, com->lc_lfsck, pfid);
2046         if (tobj == NULL)
2047                 RETURN(LLIT_UNMATCHED_PAIR);
2048
2049         if (IS_ERR(tobj))
2050                 RETURN(PTR_ERR(tobj));
2051
2052         if (!dt_object_exists(tobj))
2053                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
2054
2055         /* Load the tobj's layout EA, in spite of it is a local MDT-object or
2056          * remote one on another MDT. Then check whether the given OST-object
2057          * is in such layout. If yes, it is multiple referenced, otherwise it
2058          * is unmatched referenced case. */
2059         rc = lfsck_layout_get_lovea(env, tobj, buf, NULL);
2060         if (rc == 0)
2061                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
2062
2063         if (rc < 0)
2064                 GOTO(out, rc);
2065
2066         lmm = buf->lb_buf;
2067         rc = lfsck_layout_verify_header(lmm);
2068         if (rc != 0)
2069                 GOTO(out, rc);
2070
2071         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2072          * been verified in lfsck_layout_verify_header() already. If some
2073          * new magic introduced in the future, then layout LFSCK needs to
2074          * be updated also. */
2075         magic = le32_to_cpu(lmm->lmm_magic);
2076         if (magic == LOV_MAGIC_V1) {
2077                 objs = &(lmm->lmm_objects[0]);
2078         } else {
2079                 LASSERT(magic == LOV_MAGIC_V3);
2080                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2081         }
2082
2083         count = le16_to_cpu(lmm->lmm_stripe_count);
2084         for (i = 0; i < count; i++, objs++) {
2085                 struct lu_fid           *tfid   = &info->lti_fid2;
2086                 struct ost_id           *oi     = &info->lti_oi;
2087
2088                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2089                 ostid_to_fid(tfid, oi, le32_to_cpu(objs->l_ost_idx));
2090                 if (lu_fid_eq(cfid, tfid)) {
2091                         *lov_ea = *buf;
2092
2093                         GOTO(out, rc = LLIT_MULTIPLE_REFERENCED);
2094                 }
2095         }
2096
2097         GOTO(out, rc = LLIT_UNMATCHED_PAIR);
2098
2099 out:
2100         lfsck_object_put(env, tobj);
2101
2102         return rc;
2103 }
2104
2105 static int lfsck_layout_assistant_handle_one(const struct lu_env *env,
2106                                              struct lfsck_component *com,
2107                                              struct lfsck_layout_req *llr)
2108 {
2109         struct lfsck_layout                  *lo     = com->lc_file_ram;
2110         struct lfsck_thread_info             *info   = lfsck_env_info(env);
2111         struct filter_fid_old                *pea    = &info->lti_old_pfid;
2112         struct lu_fid                        *pfid   = &info->lti_fid;
2113         struct lu_buf                        *buf    = NULL;
2114         struct dt_object                     *parent = llr->llr_parent->llo_obj;
2115         struct dt_object                     *child  = llr->llr_child;
2116         struct lu_attr                       *pla    = &info->lti_la;
2117         struct lu_attr                       *cla    = &info->lti_la2;
2118         struct lfsck_instance                *lfsck  = com->lc_lfsck;
2119         struct lfsck_bookmark                *bk     = &lfsck->li_bookmark_ram;
2120         enum lfsck_layout_inconsistency_type  type   = LLIT_NONE;
2121         __u32                                 idx    = 0;
2122         int                                   rc;
2123         ENTRY;
2124
2125         rc = dt_attr_get(env, parent, pla, BYPASS_CAPA);
2126         if (rc != 0) {
2127                 if (lu_object_is_dying(parent->do_lu.lo_header))
2128                         RETURN(0);
2129
2130                 GOTO(out, rc);
2131         }
2132
2133         rc = dt_attr_get(env, child, cla, BYPASS_CAPA);
2134         if (rc == -ENOENT) {
2135                 if (lu_object_is_dying(parent->do_lu.lo_header))
2136                         RETURN(0);
2137
2138                 type = LLIT_DANGLING;
2139                 goto repair;
2140         }
2141
2142         if (rc != 0)
2143                 GOTO(out, rc);
2144
2145         buf = lfsck_buf_get(env, pea, sizeof(struct filter_fid_old));
2146         rc= dt_xattr_get(env, child, buf, XATTR_NAME_FID, BYPASS_CAPA);
2147         if (unlikely(rc >= 0 && rc != sizeof(struct filter_fid_old) &&
2148                      rc != sizeof(struct filter_fid))) {
2149                 type = LLIT_UNMATCHED_PAIR;
2150                 goto repair;
2151         }
2152
2153         if (rc < 0 && rc != -ENODATA)
2154                 GOTO(out, rc);
2155
2156         if (rc == -ENODATA) {
2157                 fid_zero(pfid);
2158         } else {
2159                 fid_le_to_cpu(pfid, &pea->ff_parent);
2160                 /* OST-object does not save parent FID::f_ver, instead,
2161                  * the OST-object index in the parent MDT-object layout
2162                  * EA reuses the pfid->f_ver. */
2163                 idx = pfid->f_ver;
2164                 pfid->f_ver = 0;
2165         }
2166
2167         rc = lfsck_layout_check_parent(env, com, parent, pfid,
2168                                        lu_object_fid(&child->do_lu),
2169                                        pla, cla, llr, buf, idx);
2170         if (rc > 0) {
2171                 type = rc;
2172                 goto repair;
2173         }
2174
2175         if (rc < 0)
2176                 GOTO(out, rc);
2177
2178         if (unlikely(cla->la_uid != pla->la_uid ||
2179                      cla->la_gid != pla->la_gid)) {
2180                 type = LLIT_INCONSISTENT_OWNER;
2181                 goto repair;
2182         }
2183
2184 repair:
2185         if (bk->lb_param & LPF_DRYRUN) {
2186                 if (type != LLIT_NONE)
2187                         GOTO(out, rc = 1);
2188                 else
2189                         GOTO(out, rc = 0);
2190         }
2191
2192         switch (type) {
2193         case LLIT_DANGLING:
2194                 memset(cla, 0, sizeof(*cla));
2195                 cla->la_uid = pla->la_uid;
2196                 cla->la_gid = pla->la_gid;
2197                 cla->la_mode = S_IFREG | 0666;
2198                 cla->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
2199                                 LA_ATIME | LA_MTIME | LA_CTIME;
2200                 rc = lfsck_layout_recreate_ostobj(env, com, llr, cla);
2201                 break;
2202         case LLIT_UNMATCHED_PAIR:
2203                 rc = lfsck_layout_repair_unmatched_pair(env, com, llr, pla);
2204                 break;
2205         case LLIT_MULTIPLE_REFERENCED:
2206                 rc = lfsck_layout_repair_multiple_references(env, com, llr,
2207                                                              pla, buf);
2208                 break;
2209         case LLIT_INCONSISTENT_OWNER:
2210                 rc = lfsck_layout_repair_owner(env, com, llr, pla);
2211                 break;
2212         default:
2213                 rc = 0;
2214                 break;
2215         }
2216
2217         GOTO(out, rc);
2218
2219 out:
2220         down_write(&com->lc_sem);
2221         if (rc < 0) {
2222                 /* If cannot touch the target server,
2223                  * mark the LFSCK as INCOMPLETE. */
2224                 if (rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -ETIMEDOUT ||
2225                     rc == -EHOSTDOWN || rc == -EHOSTUNREACH) {
2226                         CERROR("%s: Fail to take with OST %x: rc = %d.\n",
2227                                lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
2228                         lo->ll_flags |= LF_INCOMPLETE;
2229                         lo->ll_objs_skipped++;
2230                         rc = 0;
2231                 } else {
2232                         lo->ll_objs_failed_phase1++;
2233                 }
2234         } else if (rc > 0) {
2235                 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
2236                          "unknown type = %d\n", type);
2237
2238                 lo->ll_objs_repaired[type - 1]++;
2239         }
2240         up_write(&com->lc_sem);
2241
2242         return rc;
2243 }
2244
2245 static int lfsck_layout_assistant(void *args)
2246 {
2247         struct lfsck_thread_args        *lta     = args;
2248         struct lu_env                   *env     = &lta->lta_env;
2249         struct lfsck_component          *com     = lta->lta_com;
2250         struct lfsck_instance           *lfsck   = lta->lta_lfsck;
2251         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
2252         struct lfsck_position           *pos     = &com->lc_pos_start;
2253         struct lfsck_thread_info        *info    = lfsck_env_info(env);
2254         struct lfsck_request            *lr      = &info->lti_lr;
2255         struct lfsck_layout_master_data *llmd    = com->lc_data;
2256         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
2257         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
2258         struct lfsck_layout_req         *llr;
2259         struct l_wait_info               lwi     = { 0 };
2260         int                              rc      = 0;
2261         int                              rc1     = 0;
2262         ENTRY;
2263
2264         memset(lr, 0, sizeof(*lr));
2265         lr->lr_event = LE_START;
2266         lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
2267                        LSV_ASYNC_WINDOWS;
2268         lr->lr_speed = bk->lb_speed_limit;
2269         lr->lr_version = bk->lb_version;
2270         lr->lr_param = bk->lb_param;
2271         lr->lr_async_windows = bk->lb_async_windows;
2272         lr->lr_flags = LEF_TO_OST;
2273         if (pos->lp_oit_cookie <= 1)
2274                 lr->lr_param |= LPF_RESET;
2275
2276         rc = lfsck_layout_master_notify_others(env, com, lr);
2277         if (rc != 0) {
2278                 CERROR("%s: fail to notify others for layout start: rc = %d\n",
2279                        lfsck_lfsck2name(lfsck), rc);
2280                 GOTO(fini, rc);
2281         }
2282
2283         spin_lock(&llmd->llmd_lock);
2284         thread_set_flags(athread, SVC_RUNNING);
2285         spin_unlock(&llmd->llmd_lock);
2286         wake_up_all(&mthread->t_ctl_waitq);
2287
2288         while (1) {
2289                 while (!list_empty(&llmd->llmd_req_list)) {
2290                         bool wakeup = false;
2291
2292                         if (unlikely(llmd->llmd_exit))
2293                                 GOTO(cleanup1, rc = llmd->llmd_post_result);
2294
2295                         llr = list_entry(llmd->llmd_req_list.next,
2296                                          struct lfsck_layout_req,
2297                                          llr_list);
2298                         /* Only the lfsck_layout_assistant thread itself can
2299                          * remove the "llr" from the head of the list, LFSCK
2300                          * engine thread only inserts other new "lld" at the
2301                          * end of the list. So it is safe to handle current
2302                          * "llr" without the spin_lock. */
2303                         rc = lfsck_layout_assistant_handle_one(env, com, llr);
2304                         spin_lock(&llmd->llmd_lock);
2305                         list_del_init(&llr->llr_list);
2306                         if (bk->lb_async_windows != 0 &&
2307                             llmd->llmd_prefetched >= bk->lb_async_windows)
2308                                 wakeup = true;
2309
2310                         llmd->llmd_prefetched--;
2311                         spin_unlock(&llmd->llmd_lock);
2312                         if (wakeup)
2313                                 wake_up_all(&mthread->t_ctl_waitq);
2314
2315                         lfsck_layout_req_fini(env, llr);
2316                         if (rc < 0 && bk->lb_param & LPF_FAILOUT)
2317                                 GOTO(cleanup1, rc);
2318                 }
2319
2320                 /* Wakeup the master engine if it is waiting in checkpoint. */
2321                 wake_up_all(&mthread->t_ctl_waitq);
2322
2323                 l_wait_event(athread->t_ctl_waitq,
2324                              !lfsck_layout_req_empty(llmd) ||
2325                              llmd->llmd_exit ||
2326                              llmd->llmd_to_post ||
2327                              llmd->llmd_to_double_scan,
2328                              &lwi);
2329
2330                 if (unlikely(llmd->llmd_exit))
2331                         GOTO(cleanup1, rc = llmd->llmd_post_result);
2332
2333                 if (!list_empty(&llmd->llmd_req_list))
2334                         continue;
2335
2336                 if (llmd->llmd_to_post) {
2337                         llmd->llmd_to_post = 0;
2338                         LASSERT(llmd->llmd_post_result > 0);
2339
2340                         memset(lr, 0, sizeof(*lr));
2341                         lr->lr_event = LE_PHASE1_DONE;
2342                         lr->lr_status = llmd->llmd_post_result;
2343                         rc = lfsck_layout_master_notify_others(env, com, lr);
2344                         if (rc != 0)
2345                                 CERROR("%s: failed to notify others "
2346                                        "for layout post: rc = %d\n",
2347                                        lfsck_lfsck2name(lfsck), rc);
2348
2349                         /* Wakeup the master engine to go ahead. */
2350                         wake_up_all(&mthread->t_ctl_waitq);
2351                 }
2352
2353                 if (llmd->llmd_to_double_scan) {
2354                         llmd->llmd_to_double_scan = 0;
2355                         atomic_inc(&lfsck->li_double_scan_count);
2356                         llmd->llmd_in_double_scan = 1;
2357                         wake_up_all(&mthread->t_ctl_waitq);
2358
2359                         while (llmd->llmd_in_double_scan) {
2360                                 struct lfsck_tgt_descs  *ltds =
2361                                                         &lfsck->li_ost_descs;
2362                                 struct lfsck_tgt_desc   *ltd;
2363
2364                                 rc = lfsck_layout_master_query_others(env, com);
2365                                 if (lfsck_layout_master_to_orphan(llmd))
2366                                         goto orphan;
2367
2368                                 if (rc < 0)
2369                                         GOTO(cleanup2, rc);
2370
2371                                 /* Pull LFSCK status on related targets once
2372                                  * per 30 seconds if we are not notified. */
2373                                 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
2374                                                            cfs_time_seconds(1),
2375                                                            NULL, NULL);
2376                                 rc = l_wait_event(athread->t_ctl_waitq,
2377                                         lfsck_layout_master_to_orphan(llmd) ||
2378                                         llmd->llmd_exit ||
2379                                         !thread_is_running(mthread),
2380                                         &lwi);
2381
2382                                 if (unlikely(llmd->llmd_exit ||
2383                                              !thread_is_running(mthread)))
2384                                         GOTO(cleanup2, rc = 0);
2385
2386                                 if (rc == -ETIMEDOUT)
2387                                         continue;
2388
2389                                 if (rc < 0)
2390                                         GOTO(cleanup2, rc);
2391
2392 orphan:
2393                                 spin_lock(&ltds->ltd_lock);
2394                                 while (!list_empty(
2395                                                 &llmd->llmd_ost_phase2_list)) {
2396                                         ltd = list_entry(
2397                                               llmd->llmd_ost_phase2_list.next,
2398                                               struct lfsck_tgt_desc,
2399                                               ltd_layout_phase_list);
2400                                         list_del_init(
2401                                                 &ltd->ltd_layout_phase_list);
2402                                         spin_unlock(&ltds->ltd_lock);
2403
2404                                         rc = lfsck_layout_scan_orphan(env, com,
2405                                                                       ltd);
2406                                         if (rc != 0 &&
2407                                             bk->lb_param & LPF_FAILOUT)
2408                                                 GOTO(cleanup2, rc);
2409
2410                                         if (unlikely(llmd->llmd_exit ||
2411                                                 !thread_is_running(mthread)))
2412                                                 GOTO(cleanup2, rc = 0);
2413
2414                                         spin_lock(&ltds->ltd_lock);
2415                                 }
2416
2417                                 if (list_empty(&llmd->llmd_ost_phase1_list)) {
2418                                         spin_unlock(&ltds->ltd_lock);
2419                                         GOTO(cleanup2, rc = 1);
2420                                 }
2421                                 spin_unlock(&ltds->ltd_lock);
2422                         }
2423                 }
2424         }
2425
2426 cleanup1:
2427         /* Cleanup the unfinished requests. */
2428         spin_lock(&llmd->llmd_lock);
2429         if (rc < 0)
2430                 llmd->llmd_assistant_status = rc;
2431
2432         while (!list_empty(&llmd->llmd_req_list)) {
2433                 llr = list_entry(llmd->llmd_req_list.next,
2434                                  struct lfsck_layout_req,
2435                                  llr_list);
2436                 list_del_init(&llr->llr_list);
2437                 llmd->llmd_prefetched--;
2438                 spin_unlock(&llmd->llmd_lock);
2439                 lfsck_layout_req_fini(env, llr);
2440                 spin_lock(&llmd->llmd_lock);
2441         }
2442         spin_unlock(&llmd->llmd_lock);
2443
2444         LASSERTF(llmd->llmd_prefetched == 0, "unmatched prefeteched objs %d\n",
2445                  llmd->llmd_prefetched);
2446
2447 cleanup2:
2448         memset(lr, 0, sizeof(*lr));
2449         if (rc > 0) {
2450                 lr->lr_event = LE_PHASE2_DONE;
2451                 lr->lr_status = rc;
2452         } else if (rc == 0) {
2453                 if (lfsck->li_flags & LPF_ALL_TGT) {
2454                         lr->lr_event = LE_STOP;
2455                         lr->lr_status = LS_STOPPED;
2456                 } else {
2457                         lr->lr_event = LE_PEER_EXIT;
2458                         switch (lfsck->li_status) {
2459                         case LS_PAUSED:
2460                         case LS_CO_PAUSED:
2461                                 lr->lr_status = LS_CO_PAUSED;
2462                                 break;
2463                         case LS_STOPPED:
2464                         case LS_CO_STOPPED:
2465                                 lr->lr_status = LS_CO_STOPPED;
2466                                 break;
2467                         default:
2468                                 CERROR("%s: unknown status: rc = %d\n",
2469                                        lfsck_lfsck2name(lfsck),
2470                                        lfsck->li_status);
2471                                 lr->lr_status = LS_CO_FAILED;
2472                                 break;
2473                         }
2474                 }
2475         } else {
2476                 if (lfsck->li_flags & LPF_ALL_TGT) {
2477                         lr->lr_event = LE_STOP;
2478                         lr->lr_status = LS_FAILED;
2479                 } else {
2480                         lr->lr_event = LE_PEER_EXIT;
2481                         lr->lr_status = LS_CO_FAILED;
2482                 }
2483         }
2484
2485         rc1 = lfsck_layout_master_notify_others(env, com, lr);
2486         if (rc1 != 0) {
2487                 CERROR("%s: failed to notify others for layout quit: rc = %d\n",
2488                        lfsck_lfsck2name(lfsck), rc1);
2489                 rc = rc1;
2490         }
2491
2492         /* Under force exit case, some requests may be just freed without
2493          * verification, those objects should be re-handled when next run.
2494          * So not update the on-disk tracing file under such case. */
2495         if (!llmd->llmd_exit)
2496                 rc1 = lfsck_layout_double_scan_result(env, com, rc);
2497
2498 fini:
2499         if (llmd->llmd_in_double_scan)
2500                 atomic_dec(&lfsck->li_double_scan_count);
2501
2502         spin_lock(&llmd->llmd_lock);
2503         llmd->llmd_assistant_status = (rc1 != 0 ? rc1 : rc);
2504         thread_set_flags(athread, SVC_STOPPED);
2505         wake_up_all(&mthread->t_ctl_waitq);
2506         spin_unlock(&llmd->llmd_lock);
2507         lfsck_thread_args_fini(lta);
2508
2509         return rc;
2510 }
2511
2512 static int
2513 lfsck_layout_slave_async_interpret(const struct lu_env *env,
2514                                    struct ptlrpc_request *req,
2515                                    void *args, int rc)
2516 {
2517         struct lfsck_layout_slave_async_args *llsaa = args;
2518         struct obd_export                    *exp   = llsaa->llsaa_exp;
2519         struct lfsck_component               *com   = llsaa->llsaa_com;
2520         struct lfsck_layout_slave_target     *llst  = llsaa->llsaa_llst;
2521         struct lfsck_layout_slave_data       *llsd  = com->lc_data;
2522         bool                                  done  = false;
2523
2524         if (rc != 0) {
2525                 /* It is quite probably caused by target crash,
2526                  * to make the LFSCK can go ahead, assume that
2527                  * the target finished the LFSCK prcoessing. */
2528                 done = true;
2529         } else {
2530                 struct lfsck_reply *lr;
2531
2532                 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
2533                 if (lr->lr_status != LS_SCANNING_PHASE1 &&
2534                     lr->lr_status != LS_SCANNING_PHASE2)
2535                         done = true;
2536         }
2537         if (done)
2538                 lfsck_layout_llst_del(llsd, llst);
2539         lfsck_layout_llst_put(llst);
2540         lfsck_component_put(env, com);
2541         class_export_put(exp);
2542
2543         return 0;
2544 }
2545
2546 static int lfsck_layout_async_query(const struct lu_env *env,
2547                                     struct lfsck_component *com,
2548                                     struct obd_export *exp,
2549                                     struct lfsck_layout_slave_target *llst,
2550                                     struct lfsck_request *lr,
2551                                     struct ptlrpc_request_set *set)
2552 {
2553         struct lfsck_layout_slave_async_args *llsaa;
2554         struct ptlrpc_request                *req;
2555         struct lfsck_request                 *tmp;
2556         int                                   rc;
2557         ENTRY;
2558
2559         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
2560         if (req == NULL)
2561                 RETURN(-ENOMEM);
2562
2563         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
2564         if (rc != 0) {
2565                 ptlrpc_request_free(req);
2566                 RETURN(rc);
2567         }
2568
2569         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2570         *tmp = *lr;
2571         ptlrpc_request_set_replen(req);
2572
2573         llsaa = ptlrpc_req_async_args(req);
2574         llsaa->llsaa_exp = exp;
2575         llsaa->llsaa_com = lfsck_component_get(com);
2576         llsaa->llsaa_llst = llst;
2577         req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
2578         ptlrpc_set_add_req(set, req);
2579
2580         RETURN(0);
2581 }
2582
2583 static int lfsck_layout_async_notify(const struct lu_env *env,
2584                                      struct obd_export *exp,
2585                                      struct lfsck_request *lr,
2586                                      struct ptlrpc_request_set *set)
2587 {
2588         struct ptlrpc_request   *req;
2589         struct lfsck_request    *tmp;
2590         int                      rc;
2591         ENTRY;
2592
2593         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
2594         if (req == NULL)
2595                 RETURN(-ENOMEM);
2596
2597         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
2598         if (rc != 0) {
2599                 ptlrpc_request_free(req);
2600                 RETURN(rc);
2601         }
2602
2603         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2604         *tmp = *lr;
2605         ptlrpc_request_set_replen(req);
2606         ptlrpc_set_add_req(set, req);
2607
2608         RETURN(0);
2609 }
2610
2611 static int
2612 lfsck_layout_slave_query_master(const struct lu_env *env,
2613                                 struct lfsck_component *com)
2614 {
2615         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
2616         struct lfsck_instance            *lfsck = com->lc_lfsck;
2617         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
2618         struct lfsck_layout_slave_target *llst;
2619         struct obd_export                *exp;
2620         struct ptlrpc_request_set        *set;
2621         int                               rc    = 0;
2622         int                               rc1   = 0;
2623         ENTRY;
2624
2625         set = ptlrpc_prep_set();
2626         if (set == NULL)
2627                 RETURN(-ENOMEM);
2628
2629         memset(lr, 0, sizeof(*lr));
2630         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
2631         lr->lr_event = LE_QUERY;
2632         lr->lr_active = LT_LAYOUT;
2633
2634         llsd->llsd_touch_gen++;
2635         spin_lock(&llsd->llsd_lock);
2636         while (!list_empty(&llsd->llsd_master_list)) {
2637                 llst = list_entry(llsd->llsd_master_list.next,
2638                                   struct lfsck_layout_slave_target,
2639                                   llst_list);
2640                 if (llst->llst_gen == llsd->llsd_touch_gen)
2641                         break;
2642
2643                 llst->llst_gen = llsd->llsd_touch_gen;
2644                 list_del(&llst->llst_list);
2645                 list_add_tail(&llst->llst_list,
2646                               &llsd->llsd_master_list);
2647                 atomic_inc(&llst->llst_ref);
2648                 spin_unlock(&llsd->llsd_lock);
2649
2650                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
2651                                                llst->llst_index);
2652                 if (exp == NULL) {
2653                         lfsck_layout_llst_del(llsd, llst);
2654                         lfsck_layout_llst_put(llst);
2655                         spin_lock(&llsd->llsd_lock);
2656                         continue;
2657                 }
2658
2659                 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
2660                 if (rc != 0) {
2661                         CERROR("%s: slave fail to query %s for layout: "
2662                                "rc = %d\n", lfsck_lfsck2name(lfsck),
2663                                exp->exp_obd->obd_name, rc);
2664                         rc1 = rc;
2665                         lfsck_layout_llst_put(llst);
2666                         class_export_put(exp);
2667                 }
2668                 spin_lock(&llsd->llsd_lock);
2669         }
2670         spin_unlock(&llsd->llsd_lock);
2671
2672         rc = ptlrpc_set_wait(set);
2673         ptlrpc_set_destroy(set);
2674
2675         RETURN(rc1 != 0 ? rc1 : rc);
2676 }
2677
2678 static void
2679 lfsck_layout_slave_notify_master(const struct lu_env *env,
2680                                  struct lfsck_component *com,
2681                                  enum lfsck_events event, int result)
2682 {
2683         struct lfsck_instance            *lfsck = com->lc_lfsck;
2684         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
2685         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
2686         struct lfsck_layout_slave_target *llst;
2687         struct obd_export                *exp;
2688         struct ptlrpc_request_set        *set;
2689         int                               rc;
2690         ENTRY;
2691
2692         set = ptlrpc_prep_set();
2693         if (set == NULL)
2694                 RETURN_EXIT;
2695
2696         memset(lr, 0, sizeof(*lr));
2697         lr->lr_event = event;
2698         lr->lr_flags = LEF_FROM_OST;
2699         lr->lr_status = result;
2700         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
2701         lr->lr_active = LT_LAYOUT;
2702         llsd->llsd_touch_gen++;
2703         spin_lock(&llsd->llsd_lock);
2704         while (!list_empty(&llsd->llsd_master_list)) {
2705                 llst = list_entry(llsd->llsd_master_list.next,
2706                                   struct lfsck_layout_slave_target,
2707                                   llst_list);
2708                 if (llst->llst_gen == llsd->llsd_touch_gen)
2709                         break;
2710
2711                 llst->llst_gen = llsd->llsd_touch_gen;
2712                 list_del(&llst->llst_list);
2713                 list_add_tail(&llst->llst_list,
2714                               &llsd->llsd_master_list);
2715                 atomic_inc(&llst->llst_ref);
2716                 spin_unlock(&llsd->llsd_lock);
2717
2718                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
2719                                                llst->llst_index);
2720                 if (exp == NULL) {
2721                         lfsck_layout_llst_del(llsd, llst);
2722                         lfsck_layout_llst_put(llst);
2723                         spin_lock(&llsd->llsd_lock);
2724                         continue;
2725                 }
2726
2727                 rc = lfsck_layout_async_notify(env, exp, lr, set);
2728                 if (rc != 0)
2729                         CERROR("%s: slave fail to notify %s for layout: "
2730                                "rc = %d\n", lfsck_lfsck2name(lfsck),
2731                                exp->exp_obd->obd_name, rc);
2732                 lfsck_layout_llst_put(llst);
2733                 class_export_put(exp);
2734                 spin_lock(&llsd->llsd_lock);
2735         }
2736         spin_unlock(&llsd->llsd_lock);
2737
2738         ptlrpc_set_wait(set);
2739         ptlrpc_set_destroy(set);
2740
2741         RETURN_EXIT;
2742 }
2743
2744 /* layout APIs */
2745
2746 static int lfsck_layout_reset(const struct lu_env *env,
2747                               struct lfsck_component *com, bool init)
2748 {
2749         struct lfsck_layout     *lo    = com->lc_file_ram;
2750         int                      rc;
2751
2752         down_write(&com->lc_sem);
2753         if (init) {
2754                 memset(lo, 0, com->lc_file_size);
2755         } else {
2756                 __u32 count = lo->ll_success_count;
2757                 __u64 last_time = lo->ll_time_last_complete;
2758
2759                 memset(lo, 0, com->lc_file_size);
2760                 lo->ll_success_count = count;
2761                 lo->ll_time_last_complete = last_time;
2762         }
2763
2764         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
2765         lo->ll_status = LS_INIT;
2766
2767         rc = lfsck_layout_store(env, com);
2768         up_write(&com->lc_sem);
2769
2770         return rc;
2771 }
2772
2773 static void lfsck_layout_fail(const struct lu_env *env,
2774                               struct lfsck_component *com, bool new_checked)
2775 {
2776         struct lfsck_layout *lo = com->lc_file_ram;
2777
2778         down_write(&com->lc_sem);
2779         if (new_checked)
2780                 com->lc_new_checked++;
2781         lo->ll_objs_failed_phase1++;
2782         if (lo->ll_pos_first_inconsistent == 0) {
2783                 struct lfsck_instance *lfsck = com->lc_lfsck;
2784
2785                 lo->ll_pos_first_inconsistent =
2786                         lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
2787                                                         lfsck->li_di_oit);
2788         }
2789         up_write(&com->lc_sem);
2790 }
2791
2792 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
2793                                           struct lfsck_component *com, bool init)
2794 {
2795         struct lfsck_instance           *lfsck   = com->lc_lfsck;
2796         struct lfsck_layout             *lo      = com->lc_file_ram;
2797         struct lfsck_layout_master_data *llmd    = com->lc_data;
2798         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
2799         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
2800         struct l_wait_info               lwi     = { 0 };
2801         int                              rc;
2802
2803         if (com->lc_new_checked == 0 && !init)
2804                 return 0;
2805
2806         l_wait_event(mthread->t_ctl_waitq,
2807                      list_empty(&llmd->llmd_req_list) ||
2808                      !thread_is_running(mthread) ||
2809                      thread_is_stopped(athread),
2810                      &lwi);
2811
2812         if (!thread_is_running(mthread) || thread_is_stopped(athread))
2813                 return 0;
2814
2815         down_write(&com->lc_sem);
2816         if (init) {
2817                 lo->ll_pos_latest_start = lfsck->li_pos_current.lp_oit_cookie;
2818         } else {
2819                 lo->ll_pos_last_checkpoint =
2820                                         lfsck->li_pos_current.lp_oit_cookie;
2821                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
2822                                 HALF_SEC - lfsck->li_time_last_checkpoint);
2823                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
2824                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
2825                 com->lc_new_checked = 0;
2826         }
2827
2828         rc = lfsck_layout_store(env, com);
2829         up_write(&com->lc_sem);
2830
2831         return rc;
2832 }
2833
2834 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
2835                                          struct lfsck_component *com, bool init)
2836 {
2837         struct lfsck_instance   *lfsck = com->lc_lfsck;
2838         struct lfsck_layout     *lo    = com->lc_file_ram;
2839         int                      rc;
2840
2841         if (com->lc_new_checked == 0 && !init)
2842                 return 0;
2843
2844         down_write(&com->lc_sem);
2845
2846         if (init) {
2847                 lo->ll_pos_latest_start = lfsck->li_pos_current.lp_oit_cookie;
2848         } else {
2849                 lo->ll_pos_last_checkpoint =
2850                                         lfsck->li_pos_current.lp_oit_cookie;
2851                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
2852                                 HALF_SEC - lfsck->li_time_last_checkpoint);
2853                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
2854                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
2855                 com->lc_new_checked = 0;
2856         }
2857
2858         rc = lfsck_layout_store(env, com);
2859
2860         up_write(&com->lc_sem);
2861
2862         return rc;
2863 }
2864
2865 static int lfsck_layout_prep(const struct lu_env *env,
2866                              struct lfsck_component *com,
2867                              struct lfsck_start *start)
2868 {
2869         struct lfsck_instance   *lfsck  = com->lc_lfsck;
2870         struct lfsck_layout     *lo     = com->lc_file_ram;
2871         struct lfsck_position   *pos    = &com->lc_pos_start;
2872
2873         fid_zero(&pos->lp_dir_parent);
2874         pos->lp_dir_cookie = 0;
2875         if (lo->ll_status == LS_COMPLETED ||
2876             lo->ll_status == LS_PARTIAL ||
2877             /* To handle orphan, must scan from the beginning. */
2878             (start != NULL && start->ls_flags & LPF_ORPHAN)) {
2879                 int rc;
2880
2881                 rc = lfsck_layout_reset(env, com, false);
2882                 if (rc != 0)
2883                         return rc;
2884         }
2885
2886         down_write(&com->lc_sem);
2887         lo->ll_time_latest_start = cfs_time_current_sec();
2888         spin_lock(&lfsck->li_lock);
2889         if (lo->ll_flags & LF_SCANNED_ONCE) {
2890                 if (!lfsck->li_drop_dryrun ||
2891                     lo->ll_pos_first_inconsistent == 0) {
2892                         lo->ll_status = LS_SCANNING_PHASE2;
2893                         list_del_init(&com->lc_link);
2894                         list_add_tail(&com->lc_link,
2895                                       &lfsck->li_list_double_scan);
2896                         pos->lp_oit_cookie = 0;
2897                 } else {
2898                         int i;
2899
2900                         lo->ll_status = LS_SCANNING_PHASE1;
2901                         lo->ll_run_time_phase1 = 0;
2902                         lo->ll_run_time_phase2 = 0;
2903                         lo->ll_objs_checked_phase1 = 0;
2904                         lo->ll_objs_checked_phase2 = 0;
2905                         lo->ll_objs_failed_phase1 = 0;
2906                         lo->ll_objs_failed_phase2 = 0;
2907                         for (i = 0; i < LLIT_MAX; i++)
2908                                 lo->ll_objs_repaired[i] = 0;
2909
2910                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
2911                 }
2912         } else {
2913                 lo->ll_status = LS_SCANNING_PHASE1;
2914                 if (!lfsck->li_drop_dryrun ||
2915                     lo->ll_pos_first_inconsistent == 0)
2916                         pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
2917                 else
2918                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
2919         }
2920         spin_unlock(&lfsck->li_lock);
2921         up_write(&com->lc_sem);
2922
2923         return 0;
2924 }
2925
2926 static int lfsck_layout_slave_prep(const struct lu_env *env,
2927                                    struct lfsck_component *com,
2928                                    struct lfsck_start_param *lsp)
2929 {
2930         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
2931         struct lfsck_start              *start  = lsp->lsp_start;
2932         int                              rc;
2933
2934         rc = lfsck_layout_prep(env, com, start);
2935         if (rc != 0 || !lsp->lsp_index_valid)
2936                 return rc;
2937
2938         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
2939         if (rc == 0 && start != NULL && start->ls_flags & LPF_ORPHAN) {
2940                 LASSERT(!llsd->llsd_rbtree_valid);
2941
2942                 write_lock(&llsd->llsd_rb_lock);
2943                 rc = lfsck_rbtree_setup(env, com);
2944                 write_unlock(&llsd->llsd_rb_lock);
2945         }
2946
2947         return rc;
2948 }
2949
2950 static int lfsck_layout_master_prep(const struct lu_env *env,
2951                                     struct lfsck_component *com,
2952                                     struct lfsck_start_param *lsp)
2953 {
2954         struct lfsck_instance           *lfsck   = com->lc_lfsck;
2955         struct lfsck_layout_master_data *llmd    = com->lc_data;
2956         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
2957         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
2958         struct lfsck_thread_args        *lta;
2959         long                             rc;
2960         ENTRY;
2961
2962         rc = lfsck_layout_prep(env, com, lsp->lsp_start);
2963         if (rc != 0)
2964                 RETURN(rc);
2965
2966         llmd->llmd_assistant_status = 0;
2967         llmd->llmd_post_result = 0;
2968         llmd->llmd_to_post = 0;
2969         llmd->llmd_to_double_scan = 0;
2970         llmd->llmd_in_double_scan = 0;
2971         llmd->llmd_exit = 0;
2972         thread_set_flags(athread, 0);
2973
2974         lta = lfsck_thread_args_init(lfsck, com, lsp);
2975         if (IS_ERR(lta))
2976                 RETURN(PTR_ERR(lta));
2977
2978         rc = PTR_ERR(kthread_run(lfsck_layout_assistant, lta, "lfsck_layout"));
2979         if (IS_ERR_VALUE(rc)) {
2980                 CERROR("%s: Cannot start LFSCK layout assistant thread: "
2981                        "rc = %ld\n", lfsck_lfsck2name(lfsck), rc);
2982                 lfsck_thread_args_fini(lta);
2983         } else {
2984                 struct l_wait_info lwi = { 0 };
2985
2986                 l_wait_event(mthread->t_ctl_waitq,
2987                              thread_is_running(athread) ||
2988                              thread_is_stopped(athread),
2989                              &lwi);
2990                 if (unlikely(!thread_is_running(athread)))
2991                         rc = llmd->llmd_assistant_status;
2992                 else
2993                         rc = 0;
2994         }
2995
2996         RETURN(rc);
2997 }
2998
2999 /* Pre-fetch the attribute for each stripe in the given layout EA. */
3000 static int lfsck_layout_scan_stripes(const struct lu_env *env,
3001                                      struct lfsck_component *com,
3002                                      struct dt_object *parent,
3003                                      struct lov_mds_md_v1 *lmm)
3004 {
3005         struct lfsck_thread_info        *info    = lfsck_env_info(env);
3006         struct lfsck_instance           *lfsck   = com->lc_lfsck;
3007         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
3008         struct lfsck_layout             *lo      = com->lc_file_ram;
3009         struct lfsck_layout_master_data *llmd    = com->lc_data;
3010         struct lfsck_layout_object      *llo     = NULL;
3011         struct lov_ost_data_v1          *objs;
3012         struct lfsck_tgt_descs          *ltds    = &lfsck->li_ost_descs;
3013         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
3014         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
3015                 struct l_wait_info       lwi     = { 0 };
3016         struct lu_buf                   *buf;
3017         int                              rc      = 0;
3018         int                              i;
3019         __u32                            magic;
3020         __u16                            count;
3021         __u16                            gen;
3022         ENTRY;
3023
3024         buf = lfsck_buf_get(env, &info->lti_old_pfid,
3025                             sizeof(struct filter_fid_old));
3026         count = le16_to_cpu(lmm->lmm_stripe_count);
3027         gen = le16_to_cpu(lmm->lmm_layout_gen);
3028         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
3029          * been verified in lfsck_layout_verify_header() already. If some
3030          * new magic introduced in the future, then layout LFSCK needs to
3031          * be updated also. */
3032         magic = le32_to_cpu(lmm->lmm_magic);
3033         if (magic == LOV_MAGIC_V1) {
3034                 objs = &(lmm->lmm_objects[0]);
3035         } else {
3036                 LASSERT(magic == LOV_MAGIC_V3);
3037                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3038         }
3039
3040         for (i = 0; i < count; i++, objs++) {
3041                 struct lu_fid           *fid    = &info->lti_fid;
3042                 struct ost_id           *oi     = &info->lti_oi;
3043                 struct lfsck_layout_req *llr;
3044                 struct lfsck_tgt_desc   *tgt    = NULL;
3045                 struct dt_object        *cobj   = NULL;
3046                 __u32                    index  =
3047                                         le32_to_cpu(objs->l_ost_idx);
3048                 bool                     wakeup = false;
3049
3050                 l_wait_event(mthread->t_ctl_waitq,
3051                              bk->lb_async_windows == 0 ||
3052                              llmd->llmd_prefetched < bk->lb_async_windows ||
3053                              !thread_is_running(mthread) ||
3054                              thread_is_stopped(athread),
3055                              &lwi);
3056
3057                 if (unlikely(!thread_is_running(mthread)) ||
3058                              thread_is_stopped(athread))
3059                         GOTO(out, rc = 0);
3060
3061                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3062                 ostid_to_fid(fid, oi, index);
3063                 tgt = lfsck_tgt_get(ltds, index);
3064                 if (unlikely(tgt == NULL)) {
3065                         CERROR("%s: Cannot talk with OST %x which is not join "
3066                                "the layout LFSCK.\n",
3067                                lfsck_lfsck2name(lfsck), index);
3068                         lo->ll_flags |= LF_INCOMPLETE;
3069                         goto next;
3070                 }
3071
3072                 cobj = lfsck_object_find_by_dev(env, tgt->ltd_tgt, fid);
3073                 if (IS_ERR(cobj)) {
3074                         rc = PTR_ERR(cobj);
3075                         goto next;
3076                 }
3077
3078                 rc = dt_declare_attr_get(env, cobj, BYPASS_CAPA);
3079                 if (rc != 0)
3080                         goto next;
3081
3082                 rc = dt_declare_xattr_get(env, cobj, buf, XATTR_NAME_FID,
3083                                           BYPASS_CAPA);
3084                 if (rc != 0)
3085                         goto next;
3086
3087                 if (llo == NULL) {
3088                         llo = lfsck_layout_object_init(env, parent, gen);
3089                         if (IS_ERR(llo)) {
3090                                 rc = PTR_ERR(llo);
3091                                 goto next;
3092                         }
3093                 }
3094
3095                 llr = lfsck_layout_req_init(llo, cobj, index, i);
3096                 if (IS_ERR(llr)) {
3097                         rc = PTR_ERR(llr);
3098                         goto next;
3099                 }
3100
3101                 cobj = NULL;
3102                 spin_lock(&llmd->llmd_lock);
3103                 if (llmd->llmd_assistant_status < 0) {
3104                         spin_unlock(&llmd->llmd_lock);
3105                         lfsck_layout_req_fini(env, llr);
3106                         lfsck_tgt_put(tgt);
3107                         RETURN(llmd->llmd_assistant_status);
3108                 }
3109
3110                 list_add_tail(&llr->llr_list, &llmd->llmd_req_list);
3111                 if (llmd->llmd_prefetched == 0)
3112                         wakeup = true;
3113
3114                 llmd->llmd_prefetched++;
3115                 spin_unlock(&llmd->llmd_lock);
3116                 if (wakeup)
3117                         wake_up_all(&athread->t_ctl_waitq);
3118
3119 next:
3120                 down_write(&com->lc_sem);
3121                 com->lc_new_checked++;
3122                 if (rc < 0)
3123                         lo->ll_objs_failed_phase1++;
3124                 up_write(&com->lc_sem);
3125
3126                 if (cobj != NULL && !IS_ERR(cobj))
3127                         lu_object_put(env, &cobj->do_lu);
3128
3129                 if (likely(tgt != NULL))
3130                         lfsck_tgt_put(tgt);
3131
3132                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
3133                         GOTO(out, rc);
3134         }
3135
3136         GOTO(out, rc = 0);
3137
3138 out:
3139         if (llo != NULL && !IS_ERR(llo))
3140                 lfsck_layout_object_put(env, llo);
3141
3142         return rc;
3143 }
3144
3145 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
3146  * the OST-object's attribute and generate an structure lfsck_layout_req on the
3147  * list ::llmd_req_list.
3148  *
3149  * For each request on above list, the lfsck_layout_assistant thread compares
3150  * the OST side attribute with local attribute, if inconsistent, then repair it.
3151  *
3152  * All above processing is async mode with pipeline. */
3153 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
3154                                         struct lfsck_component *com,
3155                                         struct dt_object *obj)
3156 {
3157         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3158         struct ost_id                   *oi     = &info->lti_oi;
3159         struct lfsck_layout             *lo     = com->lc_file_ram;
3160         struct lfsck_layout_master_data *llmd   = com->lc_data;
3161         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3162         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
3163         struct thandle                  *handle = NULL;
3164         struct lu_buf                   *buf    = &info->lti_big_buf;
3165         struct lov_mds_md_v1            *lmm    = NULL;
3166         struct dt_device                *dev    = lfsck->li_bottom;
3167         struct lustre_handle             lh     = { 0 };
3168         ssize_t                          buflen = buf->lb_len;
3169         int                              rc     = 0;
3170         bool                             locked = false;
3171         bool                             stripe = false;
3172         ENTRY;
3173
3174         if (!S_ISREG(lfsck_object_type(obj)))
3175                 GOTO(out, rc = 0);
3176
3177         if (llmd->llmd_assistant_status < 0)
3178                 GOTO(out, rc = -ESRCH);
3179
3180         fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
3181         lmm_oi_cpu_to_le(oi, oi);
3182         dt_read_lock(env, obj, 0);
3183         locked = true;
3184
3185 again:
3186         rc = lfsck_layout_get_lovea(env, obj, buf, &buflen);
3187         if (rc <= 0)
3188                 GOTO(out, rc);
3189
3190         buf->lb_len = rc;
3191         lmm = buf->lb_buf;
3192         rc = lfsck_layout_verify_header(lmm);
3193         if (rc != 0)
3194                 GOTO(out, rc);
3195
3196         if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0)
3197                 GOTO(out, stripe = true);
3198
3199         /* Inconsistent lmm_oi, should be repaired. */
3200         CDEBUG(D_LFSCK, "Repair bad lmm_oi for "DFID"\n",
3201                PFID(lfsck_dto2fid(obj)));
3202
3203         if (bk->lb_param & LPF_DRYRUN) {
3204                 down_write(&com->lc_sem);
3205                 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
3206                 up_write(&com->lc_sem);
3207
3208                 GOTO(out, stripe = true);
3209         }
3210
3211         if (!lustre_handle_is_used(&lh)) {
3212                 dt_read_unlock(env, obj);
3213                 locked = false;
3214                 buf->lb_len = buflen;
3215                 rc = lfsck_layout_lock(env, com, obj, &lh,
3216                                        MDS_INODELOCK_LAYOUT |
3217                                        MDS_INODELOCK_XATTR);
3218                 if (rc != 0)
3219                         GOTO(out, rc);
3220
3221                 handle = dt_trans_create(env, dev);
3222                 if (IS_ERR(handle))
3223                         GOTO(out, rc = PTR_ERR(handle));
3224
3225                 rc = dt_declare_xattr_set(env, obj, buf, XATTR_NAME_LOV,
3226                                           LU_XATTR_REPLACE, handle);
3227                 if (rc != 0)
3228                         GOTO(out, rc);
3229
3230                 rc = dt_trans_start_local(env, dev, handle);
3231                 if (rc != 0)
3232                         GOTO(out, rc);
3233
3234                 dt_write_lock(env, obj, 0);
3235                 locked = true;
3236
3237                 goto again;
3238         }
3239
3240         lmm->lmm_oi = *oi;
3241         rc = dt_xattr_set(env, obj, buf, XATTR_NAME_LOV,
3242                           LU_XATTR_REPLACE, handle, BYPASS_CAPA);
3243         if (rc != 0)
3244                 GOTO(out, rc);
3245
3246         down_write(&com->lc_sem);
3247         lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
3248         up_write(&com->lc_sem);
3249
3250         GOTO(out, stripe = true);
3251
3252 out:
3253         if (locked) {
3254                 if (lustre_handle_is_used(&lh))
3255                         dt_write_unlock(env, obj);
3256                 else
3257                         dt_read_unlock(env, obj);
3258         }
3259
3260         if (handle != NULL && !IS_ERR(handle))
3261                 dt_trans_stop(env, dev, handle);
3262
3263         lfsck_layout_unlock(&lh);
3264         if (stripe) {
3265                 rc = lfsck_layout_scan_stripes(env, com, obj, lmm);
3266         } else {
3267                 down_write(&com->lc_sem);
3268                 com->lc_new_checked++;
3269                 if (rc < 0)
3270                         lo->ll_objs_failed_phase1++;
3271                 up_write(&com->lc_sem);
3272         }
3273         buf->lb_len = buflen;
3274
3275         return rc;
3276 }
3277
3278 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
3279                                        struct lfsck_component *com,
3280                                        struct dt_object *obj)
3281 {
3282         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3283         struct lfsck_layout             *lo     = com->lc_file_ram;
3284         const struct lu_fid             *fid    = lfsck_dto2fid(obj);
3285         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
3286         struct lfsck_layout_seq         *lls;
3287         __u64                            seq;
3288         __u64                            oid;
3289         int                              rc;
3290         ENTRY;
3291
3292         LASSERT(llsd != NULL);
3293
3294         lfsck_rbtree_update_bitmap(env, com, fid, false);
3295
3296         down_write(&com->lc_sem);
3297         if (fid_is_idif(fid))
3298                 seq = 0;
3299         else if (!fid_is_norm(fid) ||
3300                  !fid_is_for_ostobj(env, lfsck->li_next, obj, fid))
3301                 GOTO(unlock, rc = 0);
3302         else
3303                 seq = fid_seq(fid);
3304         com->lc_new_checked++;
3305
3306         lls = lfsck_layout_seq_lookup(llsd, seq);
3307         if (lls == NULL) {
3308                 OBD_ALLOC_PTR(lls);
3309                 if (unlikely(lls == NULL))
3310                         GOTO(unlock, rc = -ENOMEM);
3311
3312                 INIT_LIST_HEAD(&lls->lls_list);
3313                 lls->lls_seq = seq;
3314                 rc = lfsck_layout_lastid_load(env, com, lls);
3315                 if (rc != 0) {
3316                         lo->ll_objs_failed_phase1++;
3317                         OBD_FREE_PTR(lls);
3318                         GOTO(unlock, rc);
3319                 }
3320
3321                 lfsck_layout_seq_insert(llsd, lls);
3322         }
3323
3324         if (unlikely(fid_is_last_id(fid)))
3325                 GOTO(unlock, rc = 0);
3326
3327         oid = fid_oid(fid);
3328         if (oid > lls->lls_lastid_known)
3329                 lls->lls_lastid_known = oid;
3330
3331         if (oid > lls->lls_lastid) {
3332                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
3333                         /* OFD may create new objects during LFSCK scanning. */
3334                         rc = lfsck_layout_lastid_reload(env, com, lls);
3335                         if (unlikely(rc != 0))
3336                                 CWARN("%s: failed to reload LAST_ID for "LPX64
3337                                       ": rc = %d\n",
3338                                       lfsck_lfsck2name(com->lc_lfsck),
3339                                       lls->lls_seq, rc);
3340                         if (oid <= lls->lls_lastid)
3341                                 GOTO(unlock, rc = 0);
3342
3343                         LASSERT(lfsck->li_out_notify != NULL);
3344
3345                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
3346                                              LE_LASTID_REBUILDING);
3347                         lo->ll_flags |= LF_CRASHED_LASTID;
3348                 }
3349
3350                 lls->lls_lastid = oid;
3351                 lls->lls_dirty = 1;
3352         }
3353
3354         GOTO(unlock, rc = 0);
3355
3356 unlock:
3357         up_write(&com->lc_sem);
3358
3359         return rc;
3360 }
3361
3362 static int lfsck_layout_exec_dir(const struct lu_env *env,
3363                                  struct lfsck_component *com,
3364                                  struct dt_object *obj,
3365                                  struct lu_dirent *ent)
3366 {
3367         return 0;
3368 }
3369
3370 static int lfsck_layout_master_post(const struct lu_env *env,
3371                                     struct lfsck_component *com,
3372                                     int result, bool init)
3373 {
3374         struct lfsck_instance           *lfsck   = com->lc_lfsck;
3375         struct lfsck_layout             *lo      = com->lc_file_ram;
3376         struct lfsck_layout_master_data *llmd    = com->lc_data;
3377         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
3378         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
3379         struct l_wait_info               lwi     = { 0 };
3380         int                              rc;
3381         ENTRY;
3382
3383
3384         llmd->llmd_post_result = result;
3385         llmd->llmd_to_post = 1;
3386         if (llmd->llmd_post_result <= 0)
3387                 llmd->llmd_exit = 1;
3388
3389         wake_up_all(&athread->t_ctl_waitq);
3390         l_wait_event(mthread->t_ctl_waitq,
3391                      (result > 0 && list_empty(&llmd->llmd_req_list)) ||
3392                      thread_is_stopped(athread),
3393                      &lwi);
3394
3395         if (llmd->llmd_assistant_status < 0)
3396                 result = llmd->llmd_assistant_status;
3397
3398         down_write(&com->lc_sem);
3399         spin_lock(&lfsck->li_lock);
3400         /* When LFSCK failed, there may be some prefetched objects those are
3401          * not been processed yet, we do not know the exactly position, then
3402          * just restart from last check-point next time. */
3403         if (!init && !llmd->llmd_exit)
3404                 lo->ll_pos_last_checkpoint =
3405                                         lfsck->li_pos_current.lp_oit_cookie;
3406
3407         if (result > 0) {
3408                 lo->ll_status = LS_SCANNING_PHASE2;
3409                 lo->ll_flags |= LF_SCANNED_ONCE;
3410                 lo->ll_flags &= ~LF_UPGRADE;
3411                 list_del_init(&com->lc_link);
3412                 list_add_tail(&com->lc_link, &lfsck->li_list_double_scan);
3413         } else if (result == 0) {
3414                 lo->ll_status = lfsck->li_status;
3415                 if (lo->ll_status == 0)
3416                         lo->ll_status = LS_STOPPED;
3417                 if (lo->ll_status != LS_PAUSED) {
3418                         list_del_init(&com->lc_link);
3419                         list_add_tail(&com->lc_link, &lfsck->li_list_idle);
3420                 }
3421         } else {
3422                 lo->ll_status = LS_FAILED;
3423                 list_del_init(&com->lc_link);
3424                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
3425         }
3426         spin_unlock(&lfsck->li_lock);
3427
3428         if (!init) {
3429                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3430                                 HALF_SEC - lfsck->li_time_last_checkpoint);
3431                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
3432                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
3433                 com->lc_new_checked = 0;
3434         }
3435
3436         rc = lfsck_layout_store(env, com);
3437         up_write(&com->lc_sem);
3438
3439         RETURN(rc);
3440 }
3441
3442 static int lfsck_layout_slave_post(const struct lu_env *env,
3443                                    struct lfsck_component *com,
3444                                    int result, bool init)
3445 {
3446         struct lfsck_instance   *lfsck = com->lc_lfsck;
3447         struct lfsck_layout     *lo    = com->lc_file_ram;
3448         int                      rc;
3449         bool                     done  = false;
3450
3451         rc = lfsck_layout_lastid_store(env, com);
3452         if (rc != 0)
3453                 result = rc;
3454
3455         LASSERT(lfsck->li_out_notify != NULL);
3456
3457         down_write(&com->lc_sem);
3458
3459         spin_lock(&lfsck->li_lock);
3460         if (!init)
3461                 lo->ll_pos_last_checkpoint =
3462                                         lfsck->li_pos_current.lp_oit_cookie;
3463         if (result > 0) {
3464                 lo->ll_status = LS_SCANNING_PHASE2;
3465                 lo->ll_flags |= LF_SCANNED_ONCE;
3466                 if (lo->ll_flags & LF_CRASHED_LASTID) {
3467                         done = true;
3468                         lo->ll_flags &= ~LF_CRASHED_LASTID;
3469                 }
3470                 lo->ll_flags &= ~LF_UPGRADE;
3471                 list_del_init(&com->lc_link);
3472                 list_add_tail(&com->lc_link, &lfsck->li_list_double_scan);
3473         } else if (result == 0) {
3474                 lo->ll_status = lfsck->li_status;
3475                 if (lo->ll_status == 0)
3476                         lo->ll_status = LS_STOPPED;
3477                 if (lo->ll_status != LS_PAUSED) {
3478                         list_del_init(&com->lc_link);
3479                         list_add_tail(&com->lc_link, &lfsck->li_list_idle);
3480                 }
3481         } else {
3482                 lo->ll_status = LS_FAILED;
3483                 list_del_init(&com->lc_link);
3484                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
3485         }
3486         spin_unlock(&lfsck->li_lock);
3487
3488         if (done)
3489                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
3490                                      LE_LASTID_REBUILT);
3491
3492         if (!init) {
3493                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3494                                 HALF_SEC - lfsck->li_time_last_checkpoint);
3495                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
3496                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
3497                 com->lc_new_checked = 0;
3498         }
3499
3500         rc = lfsck_layout_store(env, com);
3501
3502         up_write(&com->lc_sem);
3503
3504         lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
3505
3506         if (result <= 0)
3507                 lfsck_rbtree_cleanup(env, com);
3508
3509         return rc;
3510 }
3511
3512 static int lfsck_layout_dump(const struct lu_env *env,
3513                              struct lfsck_component *com, char *buf, int len)
3514 {
3515         struct lfsck_instance   *lfsck = com->lc_lfsck;
3516         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
3517         struct lfsck_layout     *lo    = com->lc_file_ram;
3518         int                      save  = len;
3519         int                      ret   = -ENOSPC;
3520         int                      rc;
3521
3522         down_read(&com->lc_sem);
3523         rc = snprintf(buf, len,
3524                       "name: lfsck_layout\n"
3525                       "magic: %#x\n"
3526                       "version: %d\n"
3527                       "status: %s\n",
3528                       lo->ll_magic,
3529                       bk->lb_version,
3530                       lfsck_status2names(lo->ll_status));
3531         if (rc <= 0)
3532                 goto out;
3533
3534         buf += rc;
3535         len -= rc;
3536         rc = lfsck_bits_dump(&buf, &len, lo->ll_flags, lfsck_flags_names,
3537                              "flags");
3538         if (rc < 0)
3539                 goto out;
3540
3541         rc = lfsck_bits_dump(&buf, &len, bk->lb_param, lfsck_param_names,
3542                              "param");
3543         if (rc < 0)
3544                 goto out;
3545
3546         rc = lfsck_time_dump(&buf, &len, lo->ll_time_last_complete,
3547                              "time_since_last_completed");
3548         if (rc < 0)
3549                 goto out;
3550
3551         rc = lfsck_time_dump(&buf, &len, lo->ll_time_latest_start,
3552                              "time_since_latest_start");
3553         if (rc < 0)
3554                 goto out;
3555
3556         rc = lfsck_time_dump(&buf, &len, lo->ll_time_last_checkpoint,
3557                              "time_since_last_checkpoint");
3558         if (rc < 0)
3559                 goto out;
3560
3561         rc = snprintf(buf, len,
3562                       "latest_start_position: "LPU64"\n"
3563                       "last_checkpoint_position: "LPU64"\n"
3564                       "first_failure_position: "LPU64"\n",
3565                       lo->ll_pos_latest_start,
3566                       lo->ll_pos_last_checkpoint,
3567                       lo->ll_pos_first_inconsistent);
3568         if (rc <= 0)
3569                 goto out;
3570
3571         buf += rc;
3572         len -= rc;
3573
3574         rc = snprintf(buf, len,
3575                       "success_count: %u\n"
3576                       "repaired_dangling: "LPU64"\n"
3577                       "repaired_unmatched_pair: "LPU64"\n"
3578                       "repaired_multiple_referenced: "LPU64"\n"
3579                       "repaired_orphan: "LPU64"\n"
3580                       "repaired_inconsistent_owner: "LPU64"\n"
3581                       "repaired_others: "LPU64"\n"
3582                       "skipped: "LPU64"\n"
3583                       "failed_phase1: "LPU64"\n"
3584                       "failed_phase2: "LPU64"\n",
3585                       lo->ll_success_count,
3586                       lo->ll_objs_repaired[LLIT_DANGLING - 1],
3587                       lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
3588                       lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
3589                       lo->ll_objs_repaired[LLIT_ORPHAN - 1],
3590                       lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
3591                       lo->ll_objs_repaired[LLIT_OTHERS - 1],
3592                       lo->ll_objs_skipped,
3593                       lo->ll_objs_failed_phase1,
3594                       lo->ll_objs_failed_phase2);
3595         if (rc <= 0)
3596                 goto out;
3597
3598         buf += rc;
3599         len -= rc;
3600
3601         if (lo->ll_status == LS_SCANNING_PHASE1) {
3602                 __u64 pos;
3603                 const struct dt_it_ops *iops;
3604                 cfs_duration_t duration = cfs_time_current() -
3605                                           lfsck->li_time_last_checkpoint;
3606                 __u64 checked = lo->ll_objs_checked_phase1 + com->lc_new_checked;
3607                 __u64 speed = checked;
3608                 __u64 new_checked = com->lc_new_checked * HZ;
3609                 __u32 rtime = lo->ll_run_time_phase1 +
3610                               cfs_duration_sec(duration + HALF_SEC);
3611
3612                 if (duration != 0)
3613                         do_div(new_checked, duration);
3614                 if (rtime != 0)
3615                         do_div(speed, rtime);
3616                 rc = snprintf(buf, len,
3617                               "checked_phase1: "LPU64"\n"
3618                               "checked_phase2: "LPU64"\n"
3619                               "run_time_phase1: %u seconds\n"
3620                               "run_time_phase2: %u seconds\n"
3621                               "average_speed_phase1: "LPU64" items/sec\n"
3622                               "average_speed_phase2: N/A\n"
3623                               "real-time_speed_phase1: "LPU64" items/sec\n"
3624                               "real-time_speed_phase2: N/A\n",
3625                               checked,
3626                               lo->ll_objs_checked_phase2,
3627                               rtime,
3628                               lo->ll_run_time_phase2,
3629                               speed,
3630                               new_checked);
3631                 if (rc <= 0)
3632                         goto out;
3633
3634                 buf += rc;
3635                 len -= rc;
3636
3637                 LASSERT(lfsck->li_di_oit != NULL);
3638
3639                 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
3640
3641                 /* The low layer otable-based iteration position may NOT
3642                  * exactly match the layout-based directory traversal
3643                  * cookie. Generally, it is not a serious issue. But the
3644                  * caller should NOT make assumption on that. */
3645                 pos = iops->store(env, lfsck->li_di_oit);
3646                 if (!lfsck->li_current_oit_processed)
3647                         pos--;
3648                 rc = snprintf(buf, len, "current_position: "LPU64"\n", pos);
3649                 if (rc <= 0)
3650                         goto out;
3651
3652                 buf += rc;
3653                 len -= rc;
3654         } else {
3655                 /* XXX: LS_SCANNING_PHASE2 will be handled in the future. */
3656                 __u64 speed1 = lo->ll_objs_checked_phase1;
3657                 __u64 speed2 = lo->ll_objs_checked_phase2;
3658
3659                 if (lo->ll_run_time_phase1 != 0)
3660                         do_div(speed1, lo->ll_run_time_phase1);
3661                 if (lo->ll_run_time_phase2 != 0)
3662                         do_div(speed2, lo->ll_run_time_phase2);
3663                 rc = snprintf(buf, len,
3664                               "checked_phase1: "LPU64"\n"
3665                               "checked_phase2: "LPU64"\n"
3666                               "run_time_phase1: %u seconds\n"
3667                               "run_time_phase2: %u seconds\n"
3668                               "average_speed_phase1: "LPU64" items/sec\n"
3669                               "average_speed_phase2: "LPU64" objs/sec\n"
3670                               "real-time_speed_phase1: N/A\n"
3671                               "real-time_speed_phase2: N/A\n"
3672                               "current_position: N/A\n",
3673                               lo->ll_objs_checked_phase1,
3674                               lo->ll_objs_checked_phase2,
3675                               lo->ll_run_time_phase1,
3676                               lo->ll_run_time_phase2,
3677                               speed1,
3678                               speed2);
3679                 if (rc <= 0)
3680                         goto out;
3681
3682                 buf += rc;
3683                 len -= rc;
3684         }
3685         ret = save - len;
3686
3687 out:
3688         up_read(&com->lc_sem);
3689
3690         return ret;
3691 }
3692
3693 static int lfsck_layout_master_double_scan(const struct lu_env *env,
3694                                            struct lfsck_component *com)
3695 {
3696         struct lfsck_layout_master_data *llmd    = com->lc_data;
3697         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
3698         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
3699         struct lfsck_layout             *lo      = com->lc_file_ram;
3700         struct l_wait_info               lwi     = { 0 };
3701
3702         if (unlikely(lo->ll_status != LS_SCANNING_PHASE2))
3703                 return 0;
3704
3705         llmd->llmd_to_double_scan = 1;
3706         wake_up_all(&athread->t_ctl_waitq);
3707         l_wait_event(mthread->t_ctl_waitq,
3708                      llmd->llmd_in_double_scan ||
3709                      thread_is_stopped(athread),
3710                      &lwi);
3711         if (llmd->llmd_assistant_status < 0)
3712                 return llmd->llmd_assistant_status;
3713
3714         return 0;
3715 }
3716
3717 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
3718                                           struct lfsck_component *com)
3719 {
3720         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3721         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
3722         struct lfsck_layout             *lo     = com->lc_file_ram;
3723         struct ptlrpc_thread            *thread = &lfsck->li_thread;
3724         int                              rc;
3725         ENTRY;
3726
3727         if (unlikely(lo->ll_status != LS_SCANNING_PHASE2)) {
3728                 lfsck_rbtree_cleanup(env, com);
3729                 lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE, 0);
3730                 RETURN(0);
3731         }
3732
3733         atomic_inc(&lfsck->li_double_scan_count);
3734
3735         com->lc_new_checked = 0;
3736         com->lc_new_scanned = 0;
3737         com->lc_time_last_checkpoint = cfs_time_current();
3738         com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
3739                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
3740
3741         while (1) {
3742                 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(30),
3743                                                      NULL, NULL);
3744
3745                 rc = lfsck_layout_slave_query_master(env, com);
3746                 if (list_empty(&llsd->llsd_master_list)) {
3747                         if (unlikely(!thread_is_running(thread)))
3748                                 rc = 0;
3749                         else
3750                                 rc = 1;
3751
3752                         GOTO(done, rc);
3753                 }
3754
3755                 if (rc < 0)
3756                         GOTO(done, rc);
3757
3758                 rc = l_wait_event(thread->t_ctl_waitq,
3759                                   !thread_is_running(thread) ||
3760                                   list_empty(&llsd->llsd_master_list),
3761                                   &lwi);
3762                 if (unlikely(!thread_is_running(thread)))
3763                         GOTO(done, rc = 0);
3764
3765                 if (rc == -ETIMEDOUT)
3766                         continue;
3767
3768                 GOTO(done, rc = (rc < 0 ? rc : 1));
3769         }
3770
3771 done:
3772         rc = lfsck_layout_double_scan_result(env, com, rc);
3773
3774         lfsck_rbtree_cleanup(env, com);
3775         lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE, rc);
3776         if (atomic_dec_and_test(&lfsck->li_double_scan_count))
3777                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
3778
3779         return rc;
3780 }
3781
3782 static void lfsck_layout_master_data_release(const struct lu_env *env,
3783                                              struct lfsck_component *com)
3784 {
3785         struct lfsck_layout_master_data *llmd   = com->lc_data;
3786         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3787         struct lfsck_tgt_descs          *ltds;
3788         struct lfsck_tgt_desc           *ltd;
3789         struct lfsck_tgt_desc           *next;
3790
3791         LASSERT(llmd != NULL);
3792         LASSERT(thread_is_init(&llmd->llmd_thread) ||
3793                 thread_is_stopped(&llmd->llmd_thread));
3794         LASSERT(list_empty(&llmd->llmd_req_list));
3795
3796         com->lc_data = NULL;
3797
3798         ltds = &lfsck->li_ost_descs;
3799         spin_lock(&ltds->ltd_lock);
3800         list_for_each_entry_safe(ltd, next, &llmd->llmd_ost_phase1_list,
3801                                  ltd_layout_phase_list) {
3802                 list_del_init(&ltd->ltd_layout_phase_list);
3803         }
3804         list_for_each_entry_safe(ltd, next, &llmd->llmd_ost_phase2_list,
3805                                  ltd_layout_phase_list) {
3806                 list_del_init(&ltd->ltd_layout_phase_list);
3807         }
3808         list_for_each_entry_safe(ltd, next, &llmd->llmd_ost_list,
3809                                  ltd_layout_list) {
3810                 list_del_init(&ltd->ltd_layout_list);
3811         }
3812         list_for_each_entry_safe(ltd, next, &llmd->llmd_mdt_phase1_list,
3813                                  ltd_layout_phase_list) {
3814                 list_del_init(&ltd->ltd_layout_phase_list);
3815         }
3816         list_for_each_entry_safe(ltd, next, &llmd->llmd_mdt_phase2_list,
3817                                  ltd_layout_phase_list) {
3818                 list_del_init(&ltd->ltd_layout_phase_list);
3819         }
3820         list_for_each_entry_safe(ltd, next, &llmd->llmd_mdt_list,
3821                                  ltd_layout_list) {
3822                 list_del_init(&ltd->ltd_layout_list);
3823         }
3824         spin_unlock(&ltds->ltd_lock);
3825
3826         OBD_FREE_PTR(llmd);
3827 }
3828
3829 static void lfsck_layout_slave_data_release(const struct lu_env *env,
3830                                             struct lfsck_component *com)
3831 {
3832         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3833         struct lfsck_layout_seq          *lls;
3834         struct lfsck_layout_seq          *next;
3835         struct lfsck_layout_slave_target *llst;
3836         struct lfsck_layout_slave_target *tmp;
3837
3838         LASSERT(llsd != NULL);
3839
3840         list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
3841                                      lls_list) {
3842                 list_del_init(&lls->lls_list);
3843                 lfsck_object_put(env, lls->lls_lastid_obj);
3844                 OBD_FREE_PTR(lls);
3845         }
3846
3847         list_for_each_entry_safe(llst, tmp, &llsd->llsd_master_list,
3848                                  llst_list) {
3849                 list_del_init(&llst->llst_list);
3850                 OBD_FREE_PTR(llst);
3851         }
3852
3853         lfsck_rbtree_cleanup(env, com);
3854         com->lc_data = NULL;
3855         OBD_FREE_PTR(llsd);
3856 }
3857
3858 static void lfsck_layout_master_quit(const struct lu_env *env,
3859                                      struct lfsck_component *com)
3860 {
3861         struct lfsck_layout_master_data *llmd    = com->lc_data;
3862         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
3863         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
3864         struct l_wait_info               lwi     = { 0 };
3865
3866         llmd->llmd_exit = 1;
3867         wake_up_all(&athread->t_ctl_waitq);
3868         l_wait_event(mthread->t_ctl_waitq,
3869                      thread_is_init(athread) ||
3870                      thread_is_stopped(athread),
3871                      &lwi);
3872 }
3873
3874 static void lfsck_layout_slave_quit(const struct lu_env *env,
3875                                     struct lfsck_component *com)
3876 {
3877         lfsck_rbtree_cleanup(env, com);
3878 }
3879
3880 static int lfsck_layout_master_in_notify(const struct lu_env *env,
3881                                          struct lfsck_component *com,
3882                                          struct lfsck_request *lr)
3883 {
3884         struct lfsck_instance           *lfsck = com->lc_lfsck;
3885         struct lfsck_layout             *lo    = com->lc_file_ram;
3886         struct lfsck_layout_master_data *llmd  = com->lc_data;
3887         struct lfsck_tgt_descs          *ltds;
3888         struct lfsck_tgt_desc           *ltd;
3889         bool                             fail  = false;
3890         ENTRY;
3891
3892         if (lr->lr_event != LE_PHASE1_DONE &&
3893             lr->lr_event != LE_PHASE2_DONE &&
3894             lr->lr_event != LE_PEER_EXIT)
3895                 RETURN(-EINVAL);
3896
3897         if (lr->lr_flags & LEF_FROM_OST)
3898                 ltds = &lfsck->li_ost_descs;
3899         else
3900                 ltds = &lfsck->li_mdt_descs;
3901         spin_lock(&ltds->ltd_lock);
3902         ltd = LTD_TGT(ltds, lr->lr_index);
3903         if (ltd == NULL) {
3904                 spin_unlock(&ltds->ltd_lock);
3905
3906                 RETURN(-ENODEV);
3907         }
3908
3909         list_del_init(&ltd->ltd_layout_phase_list);
3910         switch (lr->lr_event) {
3911         case LE_PHASE1_DONE:
3912                 if (lr->lr_status <= 0) {
3913                         ltd->ltd_layout_done = 1;
3914                         list_del_init(&ltd->ltd_layout_list);
3915                         CWARN("%s: %s %x failed/stopped at phase1: rc = %d.\n",
3916                               lfsck_lfsck2name(lfsck),
3917                               (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
3918                               ltd->ltd_index, lr->lr_status);
3919                         lo->ll_flags |= LF_INCOMPLETE;
3920                         fail = true;
3921                         break;
3922                 }
3923
3924                 if (lr->lr_flags & LEF_FROM_OST) {
3925                         if (list_empty(&ltd->ltd_layout_list))
3926                                 list_add_tail(&ltd->ltd_layout_list,
3927                                               &llmd->llmd_ost_list);
3928                         list_add_tail(&ltd->ltd_layout_phase_list,
3929                                       &llmd->llmd_ost_phase2_list);
3930                 } else {
3931                         if (list_empty(&ltd->ltd_layout_list))
3932                                 list_add_tail(&ltd->ltd_layout_list,
3933                                               &llmd->llmd_mdt_list);
3934                         list_add_tail(&ltd->ltd_layout_phase_list,
3935                                       &llmd->llmd_mdt_phase2_list);
3936                 }
3937                 break;
3938         case LE_PHASE2_DONE:
3939                 ltd->ltd_layout_done = 1;
3940                 list_del_init(&ltd->ltd_layout_list);
3941                 break;
3942         case LE_PEER_EXIT:
3943                 fail = true;
3944                 ltd->ltd_layout_done = 1;
3945                 list_del_init(&ltd->ltd_layout_list);
3946                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
3947                         CWARN("%s: the peer %s %x exit layout LFSCK.\n",
3948                               lfsck_lfsck2name(lfsck),
3949                               (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
3950                               ltd->ltd_index);
3951                         lo->ll_flags |= LF_INCOMPLETE;
3952                 }
3953                 break;
3954         default:
3955                 break;
3956         }
3957         spin_unlock(&ltds->ltd_lock);
3958
3959         if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
3960                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
3961
3962                 memset(stop, 0, sizeof(*stop));
3963                 stop->ls_status = lr->lr_status;
3964                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
3965                 lfsck_stop(env, lfsck->li_bottom, stop);
3966         } else if (lfsck_layout_master_to_orphan(llmd)) {
3967                 wake_up_all(&llmd->llmd_thread.t_ctl_waitq);
3968         }
3969
3970         RETURN(0);
3971 }
3972
3973 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
3974                                         struct lfsck_component *com,
3975                                         struct lfsck_request *lr)
3976 {
3977         struct lfsck_instance            *lfsck = com->lc_lfsck;
3978         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3979         struct lfsck_layout_slave_target *llst;
3980         ENTRY;
3981
3982         if (lr->lr_event == LE_FID_ACCESSED) {
3983                 lfsck_rbtree_update_bitmap(env, com, &lr->lr_fid, true);
3984
3985                 RETURN(0);
3986         }
3987
3988         if (lr->lr_event != LE_PHASE2_DONE && lr->lr_event != LE_PEER_EXIT)
3989                 RETURN(-EINVAL);
3990
3991         llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index);
3992         if (llst == NULL)
3993                 RETURN(-ENODEV);
3994
3995         lfsck_layout_llst_put(llst);
3996         if (list_empty(&llsd->llsd_master_list))
3997                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
3998
3999         if (lr->lr_event == LE_PEER_EXIT &&
4000             lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
4001                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
4002
4003                 memset(stop, 0, sizeof(*stop));
4004                 stop->ls_status = lr->lr_status;
4005                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
4006                 lfsck_stop(env, lfsck->li_bottom, stop);
4007         }
4008
4009         RETURN(0);
4010 }
4011
4012 static int lfsck_layout_query(const struct lu_env *env,
4013                               struct lfsck_component *com)
4014 {
4015         struct lfsck_layout *lo = com->lc_file_ram;
4016
4017         return lo->ll_status;
4018 }
4019
4020 static int lfsck_layout_master_stop_notify(const struct lu_env *env,
4021                                            struct lfsck_component *com,
4022                                            struct lfsck_tgt_descs *ltds,
4023                                            struct lfsck_tgt_desc *ltd,
4024                                            struct ptlrpc_request_set *set)
4025 {
4026         struct lfsck_thread_info          *info  = lfsck_env_info(env);
4027         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
4028         struct lfsck_request              *lr    = &info->lti_lr;
4029         struct lfsck_instance             *lfsck = com->lc_lfsck;
4030         int                                rc;
4031
4032         spin_lock(&ltds->ltd_lock);
4033         if (list_empty(&ltd->ltd_layout_list)) {
4034                 LASSERT(list_empty(&ltd->ltd_layout_phase_list));
4035                 spin_unlock(&ltds->ltd_lock);
4036
4037                 return 0;
4038         }
4039
4040         list_del_init(&ltd->ltd_layout_phase_list);
4041         list_del_init(&ltd->ltd_layout_list);
4042         spin_unlock(&ltds->ltd_lock);
4043
4044         memset(lr, 0, sizeof(*lr));
4045         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
4046         lr->lr_event = LE_PEER_EXIT;
4047         lr->lr_active = LT_LAYOUT;
4048         lr->lr_status = LS_CO_PAUSED;
4049         if (ltds == &lfsck->li_ost_descs)
4050                 lr->lr_flags = LEF_TO_OST;
4051
4052         laia->laia_com = com;
4053         laia->laia_ltds = ltds;
4054         atomic_inc(&ltd->ltd_ref);
4055         laia->laia_ltd = ltd;
4056         laia->laia_lr = lr;
4057         laia->laia_shared = 0;
4058
4059         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
4060                                  lfsck_layout_master_async_interpret,
4061                                  laia, LFSCK_NOTIFY);
4062         if (rc != 0) {
4063                 CERROR("%s: Fail to notify %s %x for co-stop: rc = %d\n",
4064                        lfsck_lfsck2name(lfsck),
4065                        (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
4066                        ltd->ltd_index, rc);
4067                 lfsck_tgt_put(ltd);
4068         }
4069
4070         return rc;
4071 }
4072
4073 /* with lfsck::li_lock held */
4074 static int lfsck_layout_slave_join(const struct lu_env *env,
4075                                    struct lfsck_component *com,
4076                                    struct lfsck_start_param *lsp)
4077 {
4078         struct lfsck_instance            *lfsck = com->lc_lfsck;
4079         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
4080         struct lfsck_layout_slave_target *llst;
4081         struct lfsck_start               *start = lsp->lsp_start;
4082         int                               rc    = 0;
4083         ENTRY;
4084
4085         if (!lsp->lsp_index_valid || start == NULL ||
4086             !(start->ls_flags & LPF_ALL_TGT) ||
4087             !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT))
4088                 RETURN(-EALREADY);
4089
4090         spin_unlock(&lfsck->li_lock);
4091         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
4092         spin_lock(&lfsck->li_lock);
4093         if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
4094                 spin_unlock(&lfsck->li_lock);
4095                 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index);
4096                 if (llst != NULL)
4097                         lfsck_layout_llst_put(llst);
4098                 spin_lock(&lfsck->li_lock);
4099                 rc = -EAGAIN;
4100         }
4101
4102         RETURN(rc);
4103 }
4104
4105 static struct lfsck_operations lfsck_layout_master_ops = {
4106         .lfsck_reset            = lfsck_layout_reset,
4107         .lfsck_fail             = lfsck_layout_fail,
4108         .lfsck_checkpoint       = lfsck_layout_master_checkpoint,
4109         .lfsck_prep             = lfsck_layout_master_prep,
4110         .lfsck_exec_oit         = lfsck_layout_master_exec_oit,
4111         .lfsck_exec_dir         = lfsck_layout_exec_dir,
4112         .lfsck_post             = lfsck_layout_master_post,
4113         .lfsck_interpret        = lfsck_layout_master_async_interpret,
4114         .lfsck_dump             = lfsck_layout_dump,
4115         .lfsck_double_scan      = lfsck_layout_master_double_scan,
4116         .lfsck_data_release     = lfsck_layout_master_data_release,
4117         .lfsck_quit             = lfsck_layout_master_quit,
4118         .lfsck_in_notify        = lfsck_layout_master_in_notify,
4119         .lfsck_query            = lfsck_layout_query,
4120         .lfsck_stop_notify      = lfsck_layout_master_stop_notify,
4121 };
4122
4123 static struct lfsck_operations lfsck_layout_slave_ops = {
4124         .lfsck_reset            = lfsck_layout_reset,
4125         .lfsck_fail             = lfsck_layout_fail,
4126         .lfsck_checkpoint       = lfsck_layout_slave_checkpoint,
4127         .lfsck_prep             = lfsck_layout_slave_prep,
4128         .lfsck_exec_oit         = lfsck_layout_slave_exec_oit,
4129         .lfsck_exec_dir         = lfsck_layout_exec_dir,
4130         .lfsck_post             = lfsck_layout_slave_post,
4131         .lfsck_dump             = lfsck_layout_dump,
4132         .lfsck_double_scan      = lfsck_layout_slave_double_scan,
4133         .lfsck_data_release     = lfsck_layout_slave_data_release,
4134         .lfsck_quit             = lfsck_layout_slave_quit,
4135         .lfsck_in_notify        = lfsck_layout_slave_in_notify,
4136         .lfsck_query            = lfsck_layout_query,
4137         .lfsck_join             = lfsck_layout_slave_join,
4138 };
4139
4140 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
4141 {
4142         struct lfsck_component  *com;
4143         struct lfsck_layout     *lo;
4144         struct dt_object        *root = NULL;
4145         struct dt_object        *obj;
4146         int                      rc;
4147         ENTRY;
4148
4149         OBD_ALLOC_PTR(com);
4150         if (com == NULL)
4151                 RETURN(-ENOMEM);
4152
4153         INIT_LIST_HEAD(&com->lc_link);
4154         INIT_LIST_HEAD(&com->lc_link_dir);
4155         init_rwsem(&com->lc_sem);
4156         atomic_set(&com->lc_ref, 1);
4157         com->lc_lfsck = lfsck;
4158         com->lc_type = LT_LAYOUT;
4159         if (lfsck->li_master) {
4160                 struct lfsck_layout_master_data *llmd;
4161
4162                 com->lc_ops = &lfsck_layout_master_ops;
4163                 OBD_ALLOC_PTR(llmd);
4164                 if (llmd == NULL)
4165                         GOTO(out, rc = -ENOMEM);
4166
4167                 INIT_LIST_HEAD(&llmd->llmd_req_list);
4168                 spin_lock_init(&llmd->llmd_lock);
4169                 INIT_LIST_HEAD(&llmd->llmd_ost_list);
4170                 INIT_LIST_HEAD(&llmd->llmd_ost_phase1_list);
4171                 INIT_LIST_HEAD(&llmd->llmd_ost_phase2_list);
4172                 INIT_LIST_HEAD(&llmd->llmd_mdt_list);
4173                 INIT_LIST_HEAD(&llmd->llmd_mdt_phase1_list);
4174                 INIT_LIST_HEAD(&llmd->llmd_mdt_phase2_list);
4175                 init_waitqueue_head(&llmd->llmd_thread.t_ctl_waitq);
4176                 com->lc_data = llmd;
4177         } else {
4178                 struct lfsck_layout_slave_data *llsd;
4179
4180                 com->lc_ops = &lfsck_layout_slave_ops;
4181                 OBD_ALLOC_PTR(llsd);
4182                 if (llsd == NULL)
4183                         GOTO(out, rc = -ENOMEM);
4184
4185                 INIT_LIST_HEAD(&llsd->llsd_seq_list);
4186                 INIT_LIST_HEAD(&llsd->llsd_master_list);
4187                 spin_lock_init(&llsd->llsd_lock);
4188                 llsd->llsd_rb_root = RB_ROOT;
4189                 rwlock_init(&llsd->llsd_rb_lock);
4190                 com->lc_data = llsd;
4191         }
4192         com->lc_file_size = sizeof(*lo);
4193         OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
4194         if (com->lc_file_ram == NULL)
4195                 GOTO(out, rc = -ENOMEM);
4196
4197         OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
4198         if (com->lc_file_disk == NULL)
4199                 GOTO(out, rc = -ENOMEM);
4200
4201         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
4202         if (IS_ERR(root))
4203                 GOTO(out, rc = PTR_ERR(root));
4204
4205         if (unlikely(!dt_try_as_dir(env, root)))
4206                 GOTO(out, rc = -ENOTDIR);
4207
4208         obj = local_file_find_or_create(env, lfsck->li_los, root,
4209                                         lfsck_layout_name,
4210                                         S_IFREG | S_IRUGO | S_IWUSR);
4211         if (IS_ERR(obj))
4212                 GOTO(out, rc = PTR_ERR(obj));
4213
4214         com->lc_obj = obj;
4215         rc = lfsck_layout_load(env, com);
4216         if (rc > 0)
4217                 rc = lfsck_layout_reset(env, com, true);
4218         else if (rc == -ENOENT)
4219                 rc = lfsck_layout_init(env, com);
4220
4221         if (rc != 0)
4222                 GOTO(out, rc);
4223
4224         lo = com->lc_file_ram;
4225         switch (lo->ll_status) {
4226         case LS_INIT:
4227         case LS_COMPLETED:
4228         case LS_FAILED:
4229         case LS_STOPPED:
4230         case LS_PARTIAL:
4231                 spin_lock(&lfsck->li_lock);
4232                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
4233                 spin_unlock(&lfsck->li_lock);
4234                 break;
4235         default:
4236                 CERROR("%s: unknown lfsck_layout status: rc = %u\n",
4237                        lfsck_lfsck2name(lfsck), lo->ll_status);
4238                 /* fall through */
4239         case LS_SCANNING_PHASE1:
4240         case LS_SCANNING_PHASE2:
4241                 /* No need to store the status to disk right now.
4242                  * If the system crashed before the status stored,
4243                  * it will be loaded back when next time. */
4244                 lo->ll_status = LS_CRASHED;
4245                 lo->ll_flags |= LF_INCOMPLETE;
4246                 /* fall through */
4247         case LS_PAUSED:
4248         case LS_CRASHED:
4249         case LS_CO_FAILED:
4250         case LS_CO_STOPPED:
4251         case LS_CO_PAUSED:
4252                 spin_lock(&lfsck->li_lock);
4253                 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
4254                 spin_unlock(&lfsck->li_lock);
4255                 break;
4256         }
4257
4258         if (lo->ll_flags & LF_CRASHED_LASTID) {
4259                 LASSERT(lfsck->li_out_notify != NULL);
4260
4261                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4262                                      LE_LASTID_REBUILDING);
4263         }
4264
4265         GOTO(out, rc = 0);
4266
4267 out:
4268         if (root != NULL && !IS_ERR(root))
4269                 lu_object_put(env, &root->do_lu);
4270
4271         if (rc != 0)
4272                 lfsck_component_cleanup(env, com);
4273
4274         return rc;
4275 }