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
  38 #include <lustre/lustre_idl.h>
  39 #include <lu_object.h>
  40 #include <dt_object.h>
  41 #include <lustre_linkea.h>
  42 #include <lustre_fid.h>
  43 #include <lustre_lib.h>
  44 #include <lustre_net.h>
  45 #include <lustre/lustre_user.h>
  46 #include <md_object.h>
  47 #include <obd_class.h>
  48
  49 #include "lfsck_internal.h"
  50
  51 #define LFSCK_LAYOUT_MAGIC              0xB173AE14
  52
  53 static const char lfsck_layout_name[] = "lfsck_layout";
  54
  55 struct lfsck_layout_seq {
  56         struct list_head         lls_list;
  57         __u64                    lls_seq;
  58         __u64                    lls_lastid;
  59         __u64                    lls_lastid_known;
  60         struct dt_object        *lls_lastid_obj;
  61         unsigned int             lls_dirty:1;
  62 };
  63
  64 struct lfsck_layout_slave_target {
  65         /* link into lfsck_layout_slave_data::llsd_master_list. */
  66         struct list_head        llst_list;
  67         __u64                   llst_gen;
  68         atomic_t                llst_ref;
  69         __u32                   llst_index;
  70 };
  71
  72 struct lfsck_layout_slave_data {
  73         /* list for lfsck_layout_seq */
  74         struct list_head         llsd_seq_list;
  75
  76         /* list for the masters involve layout verification. */
  77         struct list_head         llsd_master_list;
  78         spinlock_t               llsd_lock;
  79         __u64                    llsd_touch_gen;
  80 };
  81
  82 struct lfsck_layout_object {
  83         struct dt_object        *llo_obj;
  84         struct lu_attr           llo_attr;
  85         atomic_t                 llo_ref;
  86         __u16                    llo_gen;
  87 };
  88
  89 struct lfsck_layout_req {
  90         struct list_head                 llr_list;
  91         struct lfsck_layout_object      *llr_parent;
  92         struct dt_object                *llr_child;
  93         __u32                            llr_ost_idx;
  94         __u32                            llr_lov_idx; /* offset in LOV EA */
  95 };
  96
  97 struct lfsck_layout_master_data {
  98         spinlock_t              llmd_lock;
  99         struct list_head        llmd_req_list;
 100
 101         /* list for the ost targets involve layout verification. */
 102         struct list_head        llmd_ost_list;
 103
 104         /* list for the ost targets in phase1 scanning. */
 105         struct list_head        llmd_ost_phase1_list;
 106
 107         /* list for the ost targets in phase1 scanning. */
 108         struct list_head        llmd_ost_phase2_list;
 109
 110         /* list for the mdt targets involve layout verification. */
 111         struct list_head        llmd_mdt_list;
 112
 113         /* list for the mdt targets in phase1 scanning. */
 114         struct list_head        llmd_mdt_phase1_list;
 115
 116         /* list for the mdt targets in phase1 scanning. */
 117         struct list_head        llmd_mdt_phase2_list;
 118
 119         struct ptlrpc_thread    llmd_thread;
 120         __u32                   llmd_touch_gen;
 121         int                     llmd_prefetched;
 122         int                     llmd_assistant_status;
 123         int                     llmd_post_result;
 124         unsigned int            llmd_to_post:1,
 125                                 llmd_to_double_scan:1,
 126                                 llmd_in_double_scan:1,
 127                                 llmd_exit:1;
 128 };
 129
 130 struct lfsck_layout_slave_async_args {
 131         struct obd_export                *llsaa_exp;
 132         struct lfsck_component           *llsaa_com;
 133         struct lfsck_layout_slave_target *llsaa_llst;
 134 };
 135
 136 static struct lfsck_layout_object *
 137 lfsck_layout_object_init(const struct lu_env *env, struct dt_object *obj,
 138                          __u16 gen)
 139 {
 140         struct lfsck_layout_object *llo;
 141         int                         rc;
 142
 143         OBD_ALLOC_PTR(llo);
 144         if (llo == NULL)
 145                 return ERR_PTR(-ENOMEM);
 146
 147         rc = dt_attr_get(env, obj, &llo->llo_attr, BYPASS_CAPA);
 148         if (rc != 0) {
 149                 OBD_FREE_PTR(llo);
 150
 151                 return ERR_PTR(rc);
 152         }
 153
 154         lu_object_get(&obj->do_lu);
 155         llo->llo_obj = obj;
 156         /* The gen can be used to check whether some others have changed the
 157          * file layout after LFSCK pre-fetching but before real verification. */
 158         llo->llo_gen = gen;
 159         atomic_set(&llo->llo_ref, 1);
 160
 161         return llo;
 162 }
 163
 164 static inline void
 165 lfsck_layout_llst_put(struct lfsck_layout_slave_target *llst)
 166 {
 167         if (atomic_dec_and_test(&llst->llst_ref)) {
 168                 LASSERT(list_empty(&llst->llst_list));
 169
 170                 OBD_FREE_PTR(llst);
 171         }
 172 }
 173
 174 static inline int
 175 lfsck_layout_llst_add(struct lfsck_layout_slave_data *llsd, __u32 index)
 176 {
 177         struct lfsck_layout_slave_target *llst;
 178         struct lfsck_layout_slave_target *tmp;
 179         int                               rc   = 0;
 180
 181         OBD_ALLOC_PTR(llst);
 182         if (llst == NULL)
 183                 return -ENOMEM;
 184
 185         INIT_LIST_HEAD(&llst->llst_list);
 186         llst->llst_gen = 0;
 187         llst->llst_index = index;
 188         atomic_set(&llst->llst_ref, 1);
 189
 190         spin_lock(&llsd->llsd_lock);
 191         list_for_each_entry(tmp, &llsd->llsd_master_list, llst_list) {
 192                 if (tmp->llst_index == index) {
 193                         rc = -EALREADY;
 194                         break;
 195                 }
 196         }
 197         if (rc == 0)
 198                 list_add_tail(&llst->llst_list, &llsd->llsd_master_list);
 199         spin_unlock(&llsd->llsd_lock);
 200
 201         if (rc != 0)
 202                 OBD_FREE_PTR(llst);
 203
 204         return rc;
 205 }
 206
 207 static inline void
 208 lfsck_layout_llst_del(struct lfsck_layout_slave_data *llsd,
 209                       struct lfsck_layout_slave_target *llst)
 210 {
 211         bool del = false;
 212
 213         spin_lock(&llsd->llsd_lock);
 214         if (!list_empty(&llst->llst_list)) {
 215                 list_del_init(&llst->llst_list);
 216                 del = true;
 217         }
 218         spin_unlock(&llsd->llsd_lock);
 219
 220         if (del)
 221                 lfsck_layout_llst_put(llst);
 222 }
 223
 224 static inline struct lfsck_layout_slave_target *
 225 lfsck_layout_llst_find_and_del(struct lfsck_layout_slave_data *llsd,
 226                                __u32 index)
 227 {
 228         struct lfsck_layout_slave_target *llst;
 229
 230         spin_lock(&llsd->llsd_lock);
 231         list_for_each_entry(llst, &llsd->llsd_master_list, llst_list) {
 232                 if (llst->llst_index == index) {
 233                         list_del_init(&llst->llst_list);
 234                         spin_unlock(&llsd->llsd_lock);
 235
 236                         return llst;
 237                 }
 238         }
 239         spin_unlock(&llsd->llsd_lock);
 240
 241         return NULL;
 242 }
 243
 244 static inline void lfsck_layout_object_put(const struct lu_env *env,
 245                                            struct lfsck_layout_object *llo)
 246 {
 247         if (atomic_dec_and_test(&llo->llo_ref)) {
 248                 lfsck_object_put(env, llo->llo_obj);
 249                 OBD_FREE_PTR(llo);
 250         }
 251 }
 252
 253 static struct lfsck_layout_req *
 254 lfsck_layout_req_init(struct lfsck_layout_object *parent,
 255                       struct dt_object *child, __u32 ost_idx, __u32 lov_idx)
 256 {
 257         struct lfsck_layout_req *llr;
 258
 259         OBD_ALLOC_PTR(llr);
 260         if (llr == NULL)
 261                 return ERR_PTR(-ENOMEM);
 262
 263         INIT_LIST_HEAD(&llr->llr_list);
 264         atomic_inc(&parent->llo_ref);
 265         llr->llr_parent = parent;
 266         llr->llr_child = child;
 267         llr->llr_ost_idx = ost_idx;
 268         llr->llr_lov_idx = lov_idx;
 269
 270         return llr;
 271 }
 272
 273 static inline void lfsck_layout_req_fini(const struct lu_env *env,
 274                                          struct lfsck_layout_req *llr)
 275 {
 276         lu_object_put(env, &llr->llr_child->do_lu);
 277         lfsck_layout_object_put(env, llr->llr_parent);
 278         OBD_FREE_PTR(llr);
 279 }
 280
 281 static inline bool lfsck_layout_req_empty(struct lfsck_layout_master_data *llmd)
 282 {
 283         bool empty = false;
 284
 285         spin_lock(&llmd->llmd_lock);
 286         if (list_empty(&llmd->llmd_req_list))
 287                 empty = true;
 288         spin_unlock(&llmd->llmd_lock);
 289
 290         return empty;
 291 }
 292
 293 static int lfsck_layout_get_lovea(const struct lu_env *env,
 294                                   struct dt_object *obj,
 295                                   struct lu_buf *buf, ssize_t *buflen)
 296 {
 297         int rc;
 298
 299 again:
 300         rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LOV, BYPASS_CAPA);
 301         if (rc == -ERANGE) {
 302                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV,
 303                                   BYPASS_CAPA);
 304                 if (rc <= 0)
 305                         return rc;
 306
 307                 lu_buf_realloc(buf, rc);
 308                 if (buflen != NULL)
 309                         *buflen = buf->lb_len;
 310
 311                 if (buf->lb_buf == NULL)
 312                         return -ENOMEM;
 313
 314                 goto again;
 315         }
 316
 317         if (rc == -ENODATA)
 318                 rc = 0;
 319
 320         if (rc <= 0)
 321                 return rc;
 322
 323         if (unlikely(buf->lb_buf == NULL)) {
 324                 lu_buf_alloc(buf, rc);
 325                 if (buflen != NULL)
 326                         *buflen = buf->lb_len;
 327
 328                 if (buf->lb_buf == NULL)
 329                         return -ENOMEM;
 330
 331                 goto again;
 332         }
 333
 334         return rc;
 335 }
 336
 337 static int lfsck_layout_verify_header(struct lov_mds_md_v1 *lmm)
 338 {
 339         __u32 magic;
 340         __u32 patten;
 341
 342         magic = le32_to_cpu(lmm->lmm_magic);
 343         /* If magic crashed, keep it there. Sometime later, during OST-object
 344          * orphan handling, if some OST-object(s) back-point to it, it can be
 345          * verified and repaired. */
 346         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
 347                 return -EINVAL;
 348
 349         patten = le32_to_cpu(lmm->lmm_pattern);
 350         /* XXX: currently, we only support LOV_PATTERN_RAID0. */
 351         if (patten != LOV_PATTERN_RAID0)
 352                 return -EOPNOTSUPP;
 353
 354         return 0;
 355 }
 356
 357 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
 358                                    const struct lfsck_layout *src)
 359 {
 360         int i;
 361
 362         des->ll_magic = le32_to_cpu(src->ll_magic);
 363         des->ll_status = le32_to_cpu(src->ll_status);
 364         des->ll_flags = le32_to_cpu(src->ll_flags);
 365         des->ll_success_count = le32_to_cpu(src->ll_success_count);
 366         des->ll_run_time_phase1 = le32_to_cpu(src->ll_run_time_phase1);
 367         des->ll_run_time_phase2 = le32_to_cpu(src->ll_run_time_phase2);
 368         des->ll_time_last_complete = le64_to_cpu(src->ll_time_last_complete);
 369         des->ll_time_latest_start = le64_to_cpu(src->ll_time_latest_start);
 370         des->ll_time_last_checkpoint =
 371                                 le64_to_cpu(src->ll_time_last_checkpoint);
 372         des->ll_pos_latest_start = le64_to_cpu(src->ll_pos_latest_start);
 373         des->ll_pos_last_checkpoint = le64_to_cpu(src->ll_pos_last_checkpoint);
 374         des->ll_pos_first_inconsistent =
 375                         le64_to_cpu(src->ll_pos_first_inconsistent);
 376         des->ll_objs_checked_phase1 = le64_to_cpu(src->ll_objs_checked_phase1);
 377         des->ll_objs_failed_phase1 = le64_to_cpu(src->ll_objs_failed_phase1);
 378         des->ll_objs_checked_phase2 = le64_to_cpu(src->ll_objs_checked_phase2);
 379         des->ll_objs_failed_phase2 = le64_to_cpu(src->ll_objs_failed_phase2);
 380         for (i = 0; i < LLIT_MAX; i++)
 381                 des->ll_objs_repaired[i] =
 382                                 le64_to_cpu(src->ll_objs_repaired[i]);
 383         des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
 384 }
 385
 386 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
 387                                    const struct lfsck_layout *src)
 388 {
 389         int i;
 390
 391         des->ll_magic = cpu_to_le32(src->ll_magic);
 392         des->ll_status = cpu_to_le32(src->ll_status);
 393         des->ll_flags = cpu_to_le32(src->ll_flags);
 394         des->ll_success_count = cpu_to_le32(src->ll_success_count);
 395         des->ll_run_time_phase1 = cpu_to_le32(src->ll_run_time_phase1);
 396         des->ll_run_time_phase2 = cpu_to_le32(src->ll_run_time_phase2);
 397         des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
 398         des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
 399         des->ll_time_last_checkpoint =
 400                                 cpu_to_le64(src->ll_time_last_checkpoint);
 401         des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
 402         des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
 403         des->ll_pos_first_inconsistent =
 404                         cpu_to_le64(src->ll_pos_first_inconsistent);
 405         des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
 406         des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
 407         des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
 408         des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
 409         for (i = 0; i < LLIT_MAX; i++)
 410                 des->ll_objs_repaired[i] =
 411                                 cpu_to_le64(src->ll_objs_repaired[i]);
 412         des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
 413 }
 414
 415 /**
 416  * \retval +ve: the lfsck_layout is broken, the caller should reset it.
 417  * \retval 0: succeed.
 418  * \retval -ve: failed cases.
 419  */
 420 static int lfsck_layout_load(const struct lu_env *env,
 421                              struct lfsck_component *com)
 422 {
 423         struct lfsck_layout             *lo     = com->lc_file_ram;
 424         const struct dt_body_operations *dbo    = com->lc_obj->do_body_ops;
 425         ssize_t                          size   = com->lc_file_size;
 426         loff_t                           pos    = 0;
 427         int                              rc;
 428
 429         rc = dbo->dbo_read(env, com->lc_obj,
 430                            lfsck_buf_get(env, com->lc_file_disk, size), &pos,
 431                            BYPASS_CAPA);
 432         if (rc == 0) {
 433                 return -ENOENT;
 434         } else if (rc < 0) {
 435                 CWARN("%s: failed to load lfsck_layout: rc = %d\n",
 436                       lfsck_lfsck2name(com->lc_lfsck), rc);
 437                 return rc;
 438         } else if (rc != size) {
 439                 CWARN("%s: crashed lfsck_layout, to be reset: rc = %d\n",
 440                       lfsck_lfsck2name(com->lc_lfsck), rc);
 441                 return 1;
 442         }
 443
 444         lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
 445         if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
 446                 CWARN("%s: invalid lfsck_layout magic %#x != %#x, "
 447                       "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
 448                       lo->ll_magic, LFSCK_LAYOUT_MAGIC);
 449                 return 1;
 450         }
 451
 452         return 0;
 453 }
 454
 455 static int lfsck_layout_store(const struct lu_env *env,
 456                               struct lfsck_component *com)
 457 {
 458         struct dt_object         *obj           = com->lc_obj;
 459         struct lfsck_instance    *lfsck         = com->lc_lfsck;
 460         struct lfsck_layout      *lo            = com->lc_file_disk;
 461         struct thandle           *handle;
 462         ssize_t                   size          = com->lc_file_size;
 463         loff_t                    pos           = 0;
 464         int                       rc;
 465         ENTRY;
 466
 467         lfsck_layout_cpu_to_le(lo, com->lc_file_ram);
 468         handle = dt_trans_create(env, lfsck->li_bottom);
 469         if (IS_ERR(handle)) {
 470                 rc = PTR_ERR(handle);
 471                 CERROR("%s: fail to create trans for storing lfsck_layout: "
 472                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
 473                 RETURN(rc);
 474         }
 475
 476         rc = dt_declare_record_write(env, obj, size, pos, handle);
 477         if (rc != 0) {
 478                 CERROR("%s: fail to declare trans for storing lfsck_layout(1): "
 479                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
 480                 GOTO(out, rc);
 481         }
 482
 483         rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
 484         if (rc != 0) {
 485                 CERROR("%s: fail to start trans for storing lfsck_layout: "
 486                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
 487                 GOTO(out, rc);
 488         }
 489
 490         rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos,
 491                              handle);
 492         if (rc != 0)
 493                 CERROR("%s: fail to store lfsck_layout(1): size = %d, "
 494                        "rc = %d\n", lfsck_lfsck2name(lfsck), (int)size, rc);
 495
 496         GOTO(out, rc);
 497
 498 out:
 499         dt_trans_stop(env, lfsck->li_bottom, handle);
 500
 501         return rc;
 502 }
 503
 504 static int lfsck_layout_init(const struct lu_env *env,
 505                              struct lfsck_component *com)
 506 {
 507         struct lfsck_layout *lo = com->lc_file_ram;
 508         int rc;
 509
 510         memset(lo, 0, com->lc_file_size);
 511         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
 512         lo->ll_status = LS_INIT;
 513         down_write(&com->lc_sem);
 514         rc = lfsck_layout_store(env, com);
 515         up_write(&com->lc_sem);
 516
 517         return rc;
 518 }
 519
 520 static int fid_is_for_ostobj(const struct lu_env *env, struct dt_device *dt,
 521                              struct dt_object *obj, const struct lu_fid *fid)
 522 {
 523         struct seq_server_site  *ss     = lu_site2seq(dt->dd_lu_dev.ld_site);
 524         struct lu_seq_range      range  = { 0 };
 525         struct lustre_mdt_attrs *lma;
 526         int                      rc;
 527
 528         fld_range_set_any(&range);
 529         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), &range);
 530         if (rc == 0) {
 531                 if (fld_range_is_ost(&range))
 532                         return 1;
 533
 534                 return 0;
 535         }
 536
 537         lma = &lfsck_env_info(env)->lti_lma;
 538         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, lma, sizeof(*lma)),
 539                           XATTR_NAME_LMA, BYPASS_CAPA);
 540         if (rc == sizeof(*lma)) {
 541                 lustre_lma_swab(lma);
 542
 543                 return lma->lma_compat & LMAC_FID_ON_OST ? 1 : 0;
 544         }
 545
 546         rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID, BYPASS_CAPA);
 547
 548         return rc > 0;
 549 }
 550
 551 static struct lfsck_layout_seq *
 552 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
 553 {
 554         struct lfsck_layout_seq *lls;
 555
 556         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
 557                 if (lls->lls_seq == seq)
 558                         return lls;
 559
 560                 if (lls->lls_seq > seq)
 561                         return NULL;
 562         }
 563
 564         return NULL;
 565 }
 566
 567 static void
 568 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
 569                         struct lfsck_layout_seq *lls)
 570 {
 571         struct lfsck_layout_seq *tmp;
 572         struct list_head        *pos = &llsd->llsd_seq_list;
 573
 574         list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
 575                 if (lls->lls_seq < tmp->lls_seq) {
 576                         pos = &tmp->lls_list;
 577                         break;
 578                 }
 579         }
 580         list_add_tail(&lls->lls_list, pos);
 581 }
 582
 583 static int
 584 lfsck_layout_lastid_create(const struct lu_env *env,
 585                            struct lfsck_instance *lfsck,
 586                            struct dt_object *obj)
 587 {
 588         struct lfsck_thread_info *info   = lfsck_env_info(env);
 589         struct lu_attr           *la     = &info->lti_la;
 590         struct dt_object_format  *dof    = &info->lti_dof;
 591         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
 592         struct dt_device         *dt     = lfsck->li_bottom;
 593         struct thandle           *th;
 594         __u64                     lastid = 0;
 595         loff_t                    pos    = 0;
 596         int                       rc;
 597         ENTRY;
 598
 599         CDEBUG(D_LFSCK, "To create LAST_ID for <seq> "LPX64"\n",
 600                fid_seq(lfsck_dto2fid(obj)));
 601
 602         if (bk->lb_param & LPF_DRYRUN)
 603                 return 0;
 604
 605         memset(la, 0, sizeof(*la));
 606         la->la_mode = S_IFREG |  S_IRUGO | S_IWUSR;
 607         la->la_valid = LA_MODE | LA_UID | LA_GID;
 608         dof->dof_type = dt_mode_to_dft(S_IFREG);
 609
 610         th = dt_trans_create(env, dt);
 611         if (IS_ERR(th))
 612                 RETURN(rc = PTR_ERR(th));
 613
 614         rc = dt_declare_create(env, obj, la, NULL, dof, th);
 615         if (rc != 0)
 616                 GOTO(stop, rc);
 617
 618         rc = dt_declare_record_write(env, obj, sizeof(lastid), pos, th);
 619         if (rc != 0)
 620                 GOTO(stop, rc);
 621
 622         rc = dt_trans_start_local(env, dt, th);
 623         if (rc != 0)
 624                 GOTO(stop, rc);
 625
 626         dt_write_lock(env, obj, 0);
 627         if (likely(!dt_object_exists(obj))) {
 628                 rc = dt_create(env, obj, la, NULL, dof, th);
 629                 if (rc == 0)
 630                         rc = dt_record_write(env, obj,
 631                                 lfsck_buf_get(env, &lastid, sizeof(lastid)),
 632                                 &pos, th);
 633         }
 634         dt_write_unlock(env, obj);
 635
 636         GOTO(stop, rc);
 637
 638 stop:
 639         dt_trans_stop(env, dt, th);
 640
 641         return rc;
 642 }
 643
 644 static int
 645 lfsck_layout_lastid_reload(const struct lu_env *env,
 646                            struct lfsck_component *com,
 647                            struct lfsck_layout_seq *lls)
 648 {
 649         __u64   lastid;
 650         loff_t  pos     = 0;
 651         int     rc;
 652
 653         dt_read_lock(env, lls->lls_lastid_obj, 0);
 654         rc = dt_record_read(env, lls->lls_lastid_obj,
 655                             lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
 656         dt_read_unlock(env, lls->lls_lastid_obj);
 657         if (unlikely(rc != 0))
 658                 return rc;
 659
 660         lastid = le64_to_cpu(lastid);
 661         if (lastid < lls->lls_lastid_known) {
 662                 struct lfsck_instance   *lfsck  = com->lc_lfsck;
 663                 struct lfsck_layout     *lo     = com->lc_file_ram;
 664
 665                 lls->lls_lastid = lls->lls_lastid_known;
 666                 lls->lls_dirty = 1;
 667                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
 668                         LASSERT(lfsck->li_out_notify != NULL);
 669
 670                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
 671                                              LE_LASTID_REBUILDING);
 672                         lo->ll_flags |= LF_CRASHED_LASTID;
 673                 }
 674         } else if (lastid >= lls->lls_lastid) {
 675                 lls->lls_lastid = lastid;
 676                 lls->lls_dirty = 0;
 677         }
 678
 679         return 0;
 680 }
 681
 682 static int
 683 lfsck_layout_lastid_store(const struct lu_env *env,
 684                           struct lfsck_component *com)
 685 {
 686         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 687         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
 688         struct dt_device                *dt     = lfsck->li_bottom;
 689         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
 690         struct lfsck_layout_seq         *lls;
 691         struct thandle                  *th;
 692         __u64                            lastid;
 693         int                              rc     = 0;
 694         int                              rc1    = 0;
 695
 696         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
 697                 loff_t pos = 0;
 698
 699                 /* XXX: Add the code back if we really found related
 700                  *      inconsistent cases in the future. */
 701 #if 0
 702                 if (!lls->lls_dirty) {
 703                         /* In OFD, before the pre-creation, the LAST_ID
 704                          * file will be updated firstly, which may hide
 705                          * some potential crashed cases. For example:
 706                          *
 707                          * The old obj1's ID is higher than old LAST_ID
 708                          * but lower than the new LAST_ID, but the LFSCK
 709                          * have not touch the obj1 until the OFD updated
 710                          * the LAST_ID. So the LFSCK does not regard it
 711                          * as crashed case. But when OFD does not create
 712                          * successfully, it will set the LAST_ID as the
 713                          * real created objects' ID, then LFSCK needs to
 714                          * found related inconsistency. */
 715                         rc = lfsck_layout_lastid_reload(env, com, lls);
 716                         if (likely(!lls->lls_dirty))
 717                                 continue;
 718                 }
 719 #endif
 720
 721                 CDEBUG(D_LFSCK, "To sync the LAST_ID for <seq> "LPX64
 722                        " as <oid> "LPU64"\n", lls->lls_seq, lls->lls_lastid);
 723
 724                 if (bk->lb_param & LPF_DRYRUN) {
 725                         lls->lls_dirty = 0;
 726                         continue;
 727                 }
 728
 729                 th = dt_trans_create(env, dt);
 730                 if (IS_ERR(th)) {
 731                         rc1 = PTR_ERR(th);
 732                         CERROR("%s: (1) failed to store "LPX64": rc = %d\n",
 733                                lfsck_lfsck2name(com->lc_lfsck),
 734                                lls->lls_seq, rc1);
 735                         continue;
 736                 }
 737
 738                 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
 739                                              sizeof(lastid), pos, th);
 740                 if (rc != 0)
 741                         goto stop;
 742
 743                 rc = dt_trans_start_local(env, dt, th);
 744                 if (rc != 0)
 745                         goto stop;
 746
 747                 lastid = cpu_to_le64(lls->lls_lastid);
 748                 dt_write_lock(env, lls->lls_lastid_obj, 0);
 749                 rc = dt_record_write(env, lls->lls_lastid_obj,
 750                                      lfsck_buf_get(env, &lastid,
 751                                      sizeof(lastid)), &pos, th);
 752                 dt_write_unlock(env, lls->lls_lastid_obj);
 753                 if (rc == 0)
 754                         lls->lls_dirty = 0;
 755
 756 stop:
 757                 dt_trans_stop(env, dt, th);
 758                 if (rc != 0) {
 759                         rc1 = rc;
 760                         CERROR("%s: (2) failed to store "LPX64": rc = %d\n",
 761                                lfsck_lfsck2name(com->lc_lfsck),
 762                                lls->lls_seq, rc1);
 763                 }
 764         }
 765
 766         return rc1;
 767 }
 768
 769 static int
 770 lfsck_layout_lastid_load(const struct lu_env *env,
 771                          struct lfsck_component *com,
 772                          struct lfsck_layout_seq *lls)
 773 {
 774         struct lfsck_instance   *lfsck  = com->lc_lfsck;
 775         struct lfsck_layout     *lo     = com->lc_file_ram;
 776         struct lu_fid           *fid    = &lfsck_env_info(env)->lti_fid;
 777         struct dt_object        *obj;
 778         loff_t                   pos    = 0;
 779         int                      rc;
 780         ENTRY;
 781
 782         lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck->li_bottom));
 783         obj = dt_locate(env, lfsck->li_bottom, fid);
 784         if (IS_ERR(obj))
 785                 RETURN(PTR_ERR(obj));
 786
 787         /* LAST_ID crashed, to be rebuilt */
 788         if (!dt_object_exists(obj)) {
 789                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
 790                         LASSERT(lfsck->li_out_notify != NULL);
 791
 792                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
 793                                              LE_LASTID_REBUILDING);
 794                         lo->ll_flags |= LF_CRASHED_LASTID;
 795
 796                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
 797                             cfs_fail_val > 0) {
 798                                 struct l_wait_info lwi = LWI_TIMEOUT(
 799                                                 cfs_time_seconds(cfs_fail_val),
 800                                                 NULL, NULL);
 801
 802                                 up_write(&com->lc_sem);
 803                                 l_wait_event(lfsck->li_thread.t_ctl_waitq,
 804                                              !thread_is_running(&lfsck->li_thread),
 805                                              &lwi);
 806                                 down_write(&com->lc_sem);
 807                         }
 808                 }
 809
 810                 rc = lfsck_layout_lastid_create(env, lfsck, obj);
 811         } else {
 812                 dt_read_lock(env, obj, 0);
 813                 rc = dt_read(env, obj,
 814                         lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
 815                         &pos);
 816                 dt_read_unlock(env, obj);
 817                 if (rc != 0 && rc != sizeof(__u64))
 818                         GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
 819
 820                 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
 821                         LASSERT(lfsck->li_out_notify != NULL);
 822
 823                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
 824                                              LE_LASTID_REBUILDING);
 825                         lo->ll_flags |= LF_CRASHED_LASTID;
 826                 }
 827
 828                 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
 829                 rc = 0;
 830         }
 831
 832         GOTO(out, rc);
 833
 834 out:
 835         if (rc != 0)
 836                 lfsck_object_put(env, obj);
 837         else
 838                 lls->lls_lastid_obj = obj;
 839
 840         return rc;
 841 }
 842
 843 static int lfsck_layout_master_async_interpret(const struct lu_env *env,
 844                                                struct ptlrpc_request *req,
 845                                                void *args, int rc)
 846 {
 847         struct lfsck_async_interpret_args *laia = args;
 848         struct lfsck_component            *com  = laia->laia_com;
 849         struct lfsck_layout_master_data   *llmd = com->lc_data;
 850         struct lfsck_tgt_descs            *ltds = laia->laia_ltds;
 851         struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
 852         struct lfsck_request              *lr   = laia->laia_lr;
 853
 854         switch (lr->lr_event) {
 855         case LE_START:
 856                 if (rc != 0) {
 857                         struct lfsck_layout *lo = com->lc_file_ram;
 858
 859                         lo->ll_flags |= LF_INCOMPLETE;
 860                         lfsck_tgt_put(ltd);
 861                         break;
 862                 }
 863
 864                 spin_lock(&ltds->ltd_lock);
 865                 if (ltd->ltd_dead || ltd->ltd_layout_done) {
 866                         spin_unlock(&ltds->ltd_lock);
 867                         lfsck_tgt_put(ltd);
 868                         break;
 869                 }
 870
 871                 if (lr->lr_flags & LEF_TO_OST) {
 872                         if (list_empty(&ltd->ltd_layout_list))
 873                                 list_add_tail(&ltd->ltd_layout_list,
 874                                               &llmd->llmd_ost_list);
 875                         if (list_empty(&ltd->ltd_layout_phase_list))
 876                                 list_add_tail(&ltd->ltd_layout_phase_list,
 877                                               &llmd->llmd_ost_phase1_list);
 878                 } else {
 879                         if (list_empty(&ltd->ltd_layout_list))
 880                                 list_add_tail(&ltd->ltd_layout_list,
 881                                               &llmd->llmd_mdt_list);
 882                         if (list_empty(&ltd->ltd_layout_phase_list))
 883                                 list_add_tail(&ltd->ltd_layout_phase_list,
 884                                               &llmd->llmd_mdt_phase1_list);
 885                 }
 886                 spin_unlock(&ltds->ltd_lock);
 887                 lfsck_tgt_put(ltd);
 888                 break;
 889         case LE_STOP:
 890         case LE_PHASE1_DONE:
 891         case LE_PHASE2_DONE:
 892                 if (rc != 0)
 893                         CERROR("%s: fail to notify %s %x for layout: "
 894                                "event = %d, rc = %d\n",
 895                                lfsck_lfsck2name(com->lc_lfsck),
 896                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
 897                                ltd->ltd_index, lr->lr_event, rc);
 898                 break;
 899         case LE_QUERY: {
 900                 struct lfsck_reply *reply;
 901
 902                 if (rc != 0) {
 903                         spin_lock(&ltds->ltd_lock);
 904                         list_del_init(&ltd->ltd_layout_phase_list);
 905                         list_del_init(&ltd->ltd_layout_list);
 906                         spin_unlock(&ltds->ltd_lock);
 907                         lfsck_tgt_put(ltd);
 908                         break;
 909                 }
 910
 911                 reply = req_capsule_server_get(&req->rq_pill,
 912                                                &RMF_LFSCK_REPLY);
 913                 if (reply == NULL) {
 914                         rc = -EPROTO;
 915                         CERROR("%s: invalid return value: rc = %d\n",
 916                                lfsck_lfsck2name(com->lc_lfsck), rc);
 917                         spin_lock(&ltds->ltd_lock);
 918                         list_del_init(&ltd->ltd_layout_phase_list);
 919                         list_del_init(&ltd->ltd_layout_list);
 920                         spin_unlock(&ltds->ltd_lock);
 921                         lfsck_tgt_put(ltd);
 922                         break;
 923                 }
 924
 925                 switch (reply->lr_status) {
 926                 case LS_SCANNING_PHASE1:
 927                         break;
 928                 case LS_SCANNING_PHASE2:
 929                         spin_lock(&ltds->ltd_lock);
 930                         list_del_init(&ltd->ltd_layout_phase_list);
 931                         if (ltd->ltd_dead || ltd->ltd_layout_done) {
 932                                 spin_unlock(&ltds->ltd_lock);
 933                                 break;
 934                         }
 935
 936                         if (lr->lr_flags & LEF_TO_OST)
 937                                 list_add_tail(&ltd->ltd_layout_phase_list,
 938                                               &llmd->llmd_ost_phase2_list);
 939                         else
 940                                 list_add_tail(&ltd->ltd_layout_phase_list,
 941                                               &llmd->llmd_mdt_phase2_list);
 942                         spin_unlock(&ltds->ltd_lock);
 943                         break;
 944                 default:
 945                         spin_lock(&ltds->ltd_lock);
 946                         list_del_init(&ltd->ltd_layout_phase_list);
 947                         list_del_init(&ltd->ltd_layout_list);
 948                         spin_unlock(&ltds->ltd_lock);
 949                         break;
 950                 }
 951                 lfsck_tgt_put(ltd);
 952                 break;
 953         }
 954         default:
 955                 CERROR("%s: unexpected event: rc = %d\n",
 956                        lfsck_lfsck2name(com->lc_lfsck), lr->lr_event);
 957                 break;
 958         }
 959
 960         lfsck_component_put(env, com);
 961
 962         return 0;
 963 }
 964
 965 static int lfsck_layout_master_query_others(const struct lu_env *env,
 966                                             struct lfsck_component *com)
 967 {
 968         struct lfsck_thread_info          *info  = lfsck_env_info(env);
 969         struct lfsck_request              *lr    = &info->lti_lr;
 970         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
 971         struct lfsck_instance             *lfsck = com->lc_lfsck;
 972         struct lfsck_layout_master_data   *llmd  = com->lc_data;
 973         struct ptlrpc_request_set         *set;
 974         struct lfsck_tgt_descs            *ltds;
 975         struct lfsck_tgt_desc             *ltd;
 976         struct list_head                  *head;
 977         __u32                              cnt   = 0;
 978         int                                rc    = 0;
 979         int                                rc1   = 0;
 980         ENTRY;
 981
 982         set = ptlrpc_prep_set();
 983         if (set == NULL)
 984                 RETURN(-ENOMEM);
 985
 986         llmd->llmd_touch_gen++;
 987         memset(lr, 0, sizeof(*lr));
 988         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
 989         lr->lr_event = LE_QUERY;
 990         lr->lr_active = LT_LAYOUT;
 991         laia->laia_com = com;
 992         laia->laia_lr = lr;
 993
 994         if (!list_empty(&llmd->llmd_mdt_phase1_list)) {
 995                 ltds = &lfsck->li_mdt_descs;
 996                 lr->lr_flags = 0;
 997                 head = &llmd->llmd_mdt_phase1_list;
 998         } else {
 999
1000 again:
1001                 ltds = &lfsck->li_ost_descs;
1002                 lr->lr_flags = LEF_TO_OST;
1003                 head = &llmd->llmd_ost_phase1_list;
1004         }
1005
1006         laia->laia_ltds = ltds;
1007         spin_lock(&ltds->ltd_lock);
1008         while (!list_empty(head)) {
1009                 ltd = list_entry(head->next,
1010                                  struct lfsck_tgt_desc,
1011                                  ltd_layout_phase_list);
1012                 if (ltd->ltd_layout_gen == llmd->llmd_touch_gen)
1013                         break;
1014
1015                 ltd->ltd_layout_gen = llmd->llmd_touch_gen;
1016                 list_del(&ltd->ltd_layout_phase_list);
1017                 list_add_tail(&ltd->ltd_layout_phase_list, head);
1018                 atomic_inc(&ltd->ltd_ref);
1019                 laia->laia_ltd = ltd;
1020                 spin_unlock(&ltds->ltd_lock);
1021                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1022                                          lfsck_layout_master_async_interpret,
1023                                          laia, LFSCK_QUERY);
1024                 if (rc != 0) {
1025                         CERROR("%s: fail to query %s %x for layout: rc = %d\n",
1026                                lfsck_lfsck2name(lfsck),
1027                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1028                                ltd->ltd_index, rc);
1029                         lfsck_tgt_put(ltd);
1030                         rc1 = rc;
1031                 } else {
1032                         cnt++;
1033                 }
1034                 spin_lock(&ltds->ltd_lock);
1035         }
1036         spin_unlock(&ltds->ltd_lock);
1037
1038         if (cnt > 0) {
1039                 rc = ptlrpc_set_wait(set);
1040                 if (rc < 0) {
1041                         ptlrpc_set_destroy(set);
1042                         RETURN(rc);
1043                 }
1044                 cnt = 0;
1045         }
1046
1047         if (!(lr->lr_flags & LEF_TO_OST) &&
1048             list_empty(&llmd->llmd_mdt_phase1_list))
1049                 goto again;
1050
1051         ptlrpc_set_destroy(set);
1052
1053         RETURN(rc1 != 0 ? rc1 : rc);
1054 }
1055
1056 static inline bool
1057 lfsck_layout_master_to_orphan(struct lfsck_layout_master_data *llmd)
1058 {
1059         return list_empty(&llmd->llmd_mdt_phase1_list) &&
1060                (!list_empty(&llmd->llmd_ost_phase2_list) ||
1061                 list_empty(&llmd->llmd_ost_phase1_list));
1062 }
1063
1064 static int lfsck_layout_master_notify_others(const struct lu_env *env,
1065                                              struct lfsck_component *com,
1066                                              struct lfsck_request *lr,
1067                                              __u32 flags)
1068 {
1069         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1070         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1071         struct lfsck_instance             *lfsck = com->lc_lfsck;
1072         struct lfsck_layout_master_data   *llmd  = com->lc_data;
1073         struct lfsck_layout               *lo    = com->lc_file_ram;
1074         struct ptlrpc_request_set         *set;
1075         struct lfsck_tgt_descs            *ltds;
1076         struct lfsck_tgt_desc             *ltd;
1077         struct lfsck_tgt_desc             *next;
1078         struct list_head                  *head;
1079         __u32                              idx;
1080         __u32                              cnt   = 0;
1081         int                                rc    = 0;
1082         ENTRY;
1083
1084         set = ptlrpc_prep_set();
1085         if (set == NULL)
1086                 RETURN(-ENOMEM);
1087
1088         lr->lr_active = LT_LAYOUT;
1089         laia->laia_com = com;
1090         laia->laia_lr = lr;
1091         lr->lr_flags = 0;
1092         switch (lr->lr_event) {
1093         case LE_START:
1094                 /* Notify OSTs firstly, then other MDTs if needed. */
1095                 lr->lr_flags |= LEF_TO_OST;
1096                 ltds = &lfsck->li_ost_descs;
1097
1098 lable1:
1099                 laia->laia_ltds = ltds;
1100                 down_read(&ltds->ltd_rw_sem);
1101                 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1102                         ltd = lfsck_tgt_get(ltds, idx);
1103                         LASSERT(ltd != NULL);
1104
1105                         laia->laia_ltd = ltd;
1106                         ltd->ltd_layout_done = 0;
1107                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1108                                         lfsck_layout_master_async_interpret,
1109                                         laia, LFSCK_NOTIFY);
1110                         if (rc != 0) {
1111                                 CERROR("%s: fail to notify %s %x for layout "
1112                                        "start: rc = %d\n",
1113                                        lfsck_lfsck2name(lfsck),
1114                                        (lr->lr_flags & LEF_TO_OST) ? "OST" :
1115                                        "MDT", idx, rc);
1116                                 lfsck_tgt_put(ltd);
1117                                 lo->ll_flags |= LF_INCOMPLETE;
1118                         } else {
1119                                 cnt++;
1120                         }
1121                 }
1122                 up_read(&ltds->ltd_rw_sem);
1123
1124                 /* Sync up */
1125                 if (cnt > 0) {
1126                         rc = ptlrpc_set_wait(set);
1127                         if (rc < 0) {
1128                                 ptlrpc_set_destroy(set);
1129                                 RETURN(rc);
1130                         }
1131                         cnt = 0;
1132                 }
1133
1134                 if (!(flags & LPF_ALL_MDT))
1135                         break;
1136
1137                 ltds = &lfsck->li_mdt_descs;
1138                 /* The sponsor broadcasts the request to other MDTs. */
1139                 if (flags & LPF_BROADCAST) {
1140                         flags &= ~LPF_ALL_MDT;
1141                         lr->lr_flags &= ~LEF_TO_OST;
1142                         goto lable1;
1143                 }
1144
1145                 /* non-sponsors link other MDT targets locallly. */
1146                 spin_lock(&ltds->ltd_lock);
1147                 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1148                         ltd = LTD_TGT(ltds, idx);
1149                         LASSERT(ltd != NULL);
1150
1151                         if (!list_empty(&ltd->ltd_layout_list))
1152                                 continue;
1153
1154                         list_add_tail(&ltd->ltd_layout_list,
1155                                       &llmd->llmd_mdt_list);
1156                         list_add_tail(&ltd->ltd_layout_phase_list,
1157                                       &llmd->llmd_mdt_phase1_list);
1158                 }
1159                 spin_unlock(&ltds->ltd_lock);
1160
1161                 break;
1162         case LE_STOP:
1163                 if (flags & LPF_BROADCAST)
1164                         lr->lr_flags |= LEF_FORCE_STOP;
1165         case LE_PHASE2_DONE:
1166                 /* Notify other MDTs if needed, then the OSTs. */
1167                 if (flags & LPF_ALL_MDT) {
1168                         /* The sponsor broadcasts the request to other MDTs. */
1169                         if (flags & LPF_BROADCAST) {
1170                                 lr->lr_flags &= ~LEF_TO_OST;
1171                                 head = &llmd->llmd_mdt_list;
1172                                 ltds = &lfsck->li_mdt_descs;
1173                                 goto lable3;
1174                         }
1175
1176                         /* non-sponsors unlink other MDT targets locallly. */
1177                         ltds = &lfsck->li_mdt_descs;
1178                         spin_lock(&ltds->ltd_lock);
1179                         list_for_each_entry_safe(ltd, next,
1180                                                  &llmd->llmd_mdt_list,
1181                                                  ltd_layout_list) {
1182                                 list_del_init(&ltd->ltd_layout_phase_list);
1183                                 list_del_init(&ltd->ltd_layout_list);
1184                         }
1185                         spin_unlock(&ltds->ltd_lock);
1186                 }
1187
1188 lable2:
1189                 lr->lr_flags |= LEF_TO_OST;
1190                 head = &llmd->llmd_ost_list;
1191                 ltds = &lfsck->li_ost_descs;
1192
1193 lable3:
1194                 laia->laia_ltds = ltds;
1195                 spin_lock(&ltds->ltd_lock);
1196                 while (!list_empty(head)) {
1197                         ltd = list_entry(head->next, struct lfsck_tgt_desc,
1198                                          ltd_layout_list);
1199                         if (!list_empty(&ltd->ltd_layout_phase_list))
1200                                 list_del_init(&ltd->ltd_layout_phase_list);
1201                         list_del_init(&ltd->ltd_layout_list);
1202                         laia->laia_ltd = ltd;
1203                         spin_unlock(&ltds->ltd_lock);
1204                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1205                                         lfsck_layout_master_async_interpret,
1206                                         laia, LFSCK_NOTIFY);
1207                         if (rc != 0)
1208                                 CERROR("%s: fail to notify %s %x for layout "
1209                                        "stop/phase2: rc = %d\n",
1210                                        lfsck_lfsck2name(lfsck),
1211                                        (lr->lr_flags & LEF_TO_OST) ? "OST" :
1212                                        "MDT", ltd->ltd_index, rc);
1213                         else
1214                                 cnt++;
1215                         spin_lock(&ltds->ltd_lock);
1216                 }
1217                 spin_unlock(&ltds->ltd_lock);
1218
1219                 if (!(flags & LPF_BROADCAST))
1220                         break;
1221
1222                 /* Sync up */
1223                 if (cnt > 0) {
1224                         rc = ptlrpc_set_wait(set);
1225                         if (rc < 0) {
1226                                 ptlrpc_set_destroy(set);
1227                                 RETURN(rc);
1228                         }
1229                         cnt = 0;
1230                 }
1231
1232                 flags &= ~LPF_BROADCAST;
1233                 goto lable2;
1234         case LE_PHASE1_DONE:
1235                 llmd->llmd_touch_gen++;
1236                 lr->lr_flags &= ~LEF_TO_OST;
1237                 ltds = &lfsck->li_mdt_descs;
1238                 laia->laia_ltds = ltds;
1239                 spin_lock(&ltds->ltd_lock);
1240                 while (!list_empty(&llmd->llmd_mdt_phase1_list)) {
1241                         ltd = list_entry(llmd->llmd_mdt_phase1_list.next,
1242                                          struct lfsck_tgt_desc,
1243                                          ltd_layout_phase_list);
1244                         if (ltd->ltd_layout_gen == llmd->llmd_touch_gen)
1245                                 break;
1246
1247                         ltd->ltd_layout_gen = llmd->llmd_touch_gen;
1248                         list_del_init(&ltd->ltd_layout_phase_list);
1249                         list_add_tail(&ltd->ltd_layout_phase_list,
1250                                       &llmd->llmd_mdt_phase1_list);
1251                         laia->laia_ltd = ltd;
1252                         spin_unlock(&ltds->ltd_lock);
1253                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1254                                         lfsck_layout_master_async_interpret,
1255                                         laia, LFSCK_NOTIFY);
1256                         if (rc != 0)
1257                                 CERROR("%s: fail to notify MDT %x for layout "
1258                                        "phase1 done: rc = %d\n",
1259                                        lfsck_lfsck2name(lfsck),
1260                                        ltd->ltd_index, rc);
1261                         else
1262                                 cnt++;
1263                         spin_lock(&ltds->ltd_lock);
1264                 }
1265                 spin_unlock(&ltds->ltd_lock);
1266                 break;
1267         default:
1268                 CERROR("%s: unexpected LFSCK event: rc = %d\n",
1269                        lfsck_lfsck2name(lfsck), lr->lr_event);
1270                 rc = -EINVAL;
1271                 break;
1272         }
1273
1274         if (cnt > 0)
1275                 rc = ptlrpc_set_wait(set);
1276         ptlrpc_set_destroy(set);
1277
1278         if (rc == 0 && lr->lr_event == LE_START &&
1279             list_empty(&llmd->llmd_ost_list))
1280                 rc = -ENODEV;
1281
1282         RETURN(rc);
1283 }
1284
1285 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1286                                            struct lfsck_component *com,
1287                                            int rc)
1288 {
1289         struct lfsck_instance   *lfsck = com->lc_lfsck;
1290         struct lfsck_layout     *lo    = com->lc_file_ram;
1291         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
1292
1293         down_write(&com->lc_sem);
1294
1295         lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1296                                 HALF_SEC - lfsck->li_time_last_checkpoint);
1297         lo->ll_time_last_checkpoint = cfs_time_current_sec();
1298         lo->ll_objs_checked_phase2 += com->lc_new_checked;
1299
1300         if (rc > 0) {
1301                 com->lc_journal = 0;
1302                 if (lo->ll_flags & LF_INCOMPLETE)
1303                         lo->ll_status = LS_PARTIAL;
1304                 else
1305                         lo->ll_status = LS_COMPLETED;
1306                 if (!(bk->lb_param & LPF_DRYRUN))
1307                         lo->ll_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
1308                 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1309                 lo->ll_success_count++;
1310         } else if (rc == 0) {
1311                 lo->ll_status = lfsck->li_status;
1312                 if (lo->ll_status == 0)
1313                         lo->ll_status = LS_STOPPED;
1314         } else {
1315                 lo->ll_status = LS_FAILED;
1316         }
1317
1318         if (lo->ll_status != LS_PAUSED) {
1319                 spin_lock(&lfsck->li_lock);
1320                 list_del_init(&com->lc_link);
1321                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
1322                 spin_unlock(&lfsck->li_lock);
1323         }
1324
1325         rc = lfsck_layout_store(env, com);
1326
1327         up_write(&com->lc_sem);
1328
1329         return rc;
1330 }
1331
1332 static int lfsck_layout_lock(const struct lu_env *env,
1333                              struct lfsck_component *com,
1334                              struct dt_object *obj,
1335                              struct lustre_handle *lh, __u64 bits)
1336 {
1337         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1338         ldlm_policy_data_t              *policy = &info->lti_policy;
1339         struct ldlm_res_id              *resid  = &info->lti_resid;
1340         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1341         __u64                            flags  = LDLM_FL_ATOMIC_CB;
1342         int                              rc;
1343
1344         LASSERT(lfsck->li_namespace != NULL);
1345
1346         memset(policy, 0, sizeof(*policy));
1347         policy->l_inodebits.bits = bits;
1348         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
1349         rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_IBITS,
1350                                     policy, LCK_EX, &flags, ldlm_blocking_ast,
1351                                     ldlm_completion_ast, NULL, NULL, 0,
1352                                     LVB_T_NONE, NULL, lh);
1353         if (rc == ELDLM_OK) {
1354                 rc = 0;
1355         } else {
1356                 memset(lh, 0, sizeof(*lh));
1357                 rc = -EIO;
1358         }
1359
1360         return rc;
1361 }
1362
1363 static void lfsck_layout_unlock(struct lustre_handle *lh)
1364 {
1365         if (lustre_handle_is_used(lh)) {
1366                 ldlm_lock_decref(lh, LCK_EX);
1367                 memset(lh, 0, sizeof(*lh));
1368         }
1369 }
1370
1371 static int lfsck_layout_trans_stop(const struct lu_env *env,
1372                                    struct dt_device *dev,
1373                                    struct thandle *handle, int result)
1374 {
1375         int rc;
1376
1377         handle->th_result = result;
1378         rc = dt_trans_stop(env, dev, handle);
1379         if (rc > 0)
1380                 rc = 0;
1381         else if (rc == 0)
1382                 rc = 1;
1383
1384         return rc;
1385 }
1386
1387 static int lfsck_layout_scan_orphan(const struct lu_env *env,
1388                                     struct lfsck_component *com,
1389                                     struct lfsck_tgt_desc *ltd)
1390 {
1391         /* XXX: To be extended in other patch. */
1392
1393         return 0;
1394 }
1395
1396 /* For the MDT-object with dangling reference, we need to re-create
1397  * the missed OST-object with the known FID/owner information. */
1398 static int lfsck_layout_recreate_ostobj(const struct lu_env *env,
1399                                         struct lfsck_component *com,
1400                                         struct lfsck_layout_req *llr,
1401                                         struct lu_attr *la)
1402 {
1403         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1404         struct filter_fid               *pfid   = &info->lti_new_pfid;
1405         struct dt_allocation_hint       *hint   = &info->lti_hint;
1406         struct dt_object                *parent = llr->llr_parent->llo_obj;
1407         struct dt_object                *child  = llr->llr_child;
1408         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
1409         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
1410         struct thandle                  *handle;
1411         struct lu_buf                   *buf;
1412         struct lustre_handle             lh     = { 0 };
1413         int                              rc;
1414         ENTRY;
1415
1416         CDEBUG(D_LFSCK, "Repair dangling reference for: parent "DFID
1417                ", child "DFID", OST-index %u, stripe-index %u, owner %u:%u\n",
1418                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
1419                llr->llr_ost_idx, llr->llr_lov_idx, la->la_uid, la->la_gid);
1420
1421         rc = lfsck_layout_lock(env, com, parent, &lh,
1422                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
1423         if (rc != 0)
1424                 RETURN(rc);
1425
1426         handle = dt_trans_create(env, dev);
1427         if (IS_ERR(handle))
1428                 GOTO(unlock1, rc = PTR_ERR(handle));
1429
1430         hint->dah_parent = NULL;
1431         hint->dah_mode = 0;
1432         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
1433         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
1434         pfid->ff_parent.f_ver = cpu_to_le32(llr->llr_lov_idx);
1435         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
1436
1437         rc = dt_declare_create(env, child, la, hint, NULL, handle);
1438         if (rc != 0)
1439                 GOTO(stop, rc);
1440
1441         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
1442                                   LU_XATTR_CREATE, handle);
1443         if (rc != 0)
1444                 GOTO(stop, rc);
1445
1446         rc = dt_trans_start(env, dev, handle);
1447         if (rc != 0)
1448                 GOTO(stop, rc);
1449
1450         dt_read_lock(env, parent, 0);
1451         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
1452                 GOTO(unlock2, rc = 1);
1453
1454         rc = dt_create(env, child, la, hint, NULL, handle);
1455         if (rc != 0)
1456                 GOTO(unlock2, rc);
1457
1458         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
1459                           handle, BYPASS_CAPA);
1460
1461         GOTO(unlock2, rc);
1462
1463 unlock2:
1464         dt_read_unlock(env, parent);
1465
1466 stop:
1467         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
1468
1469 unlock1:
1470         lfsck_layout_unlock(&lh);
1471
1472         return rc;
1473 }
1474
1475 /* If the OST-object does not recognize the MDT-object as its parent, and
1476  * there is no other MDT-object claims as its parent, then just trust the
1477  * given MDT-object as its parent. So update the OST-object filter_fid. */
1478 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
1479                                               struct lfsck_component *com,
1480                                               struct lfsck_layout_req *llr,
1481                                               const struct lu_attr *pla)
1482 {
1483         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1484         struct filter_fid               *pfid   = &info->lti_new_pfid;
1485         struct lu_attr                  *tla    = &info->lti_la3;
1486         struct dt_object                *parent = llr->llr_parent->llo_obj;
1487         struct dt_object                *child  = llr->llr_child;
1488         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
1489         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
1490         struct thandle                  *handle;
1491         struct lu_buf                   *buf;
1492         struct lustre_handle             lh     = { 0 };
1493         int                              rc;
1494         ENTRY;
1495
1496         CDEBUG(D_LFSCK, "Repair unmatched MDT-OST pair for: parent "DFID
1497                ", child "DFID", OST-index %u, stripe-index %u, owner %u:%u\n",
1498                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
1499                llr->llr_ost_idx, llr->llr_lov_idx, pla->la_uid, pla->la_gid);
1500
1501         rc = lfsck_layout_lock(env, com, parent, &lh,
1502                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
1503         if (rc != 0)
1504                 RETURN(rc);
1505
1506         handle = dt_trans_create(env, dev);
1507         if (IS_ERR(handle))
1508                 GOTO(unlock1, rc = PTR_ERR(handle));
1509
1510         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
1511         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
1512         /* The ff_parent->f_ver is not the real parent fid->f_ver. Instead,
1513          * it is the OST-object index in the parent MDT-object layout. */
1514         pfid->ff_parent.f_ver = cpu_to_le32(llr->llr_lov_idx);
1515         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
1516
1517         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
1518         if (rc != 0)
1519                 GOTO(stop, rc);
1520
1521         tla->la_valid = LA_UID | LA_GID;
1522         tla->la_uid = pla->la_uid;
1523         tla->la_gid = pla->la_gid;
1524         rc = dt_declare_attr_set(env, child, tla, handle);
1525         if (rc != 0)
1526                 GOTO(stop, rc);
1527
1528         rc = dt_trans_start(env, dev, handle);
1529         if (rc != 0)
1530                 GOTO(stop, rc);
1531
1532         dt_write_lock(env, parent, 0);
1533         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
1534                 GOTO(unlock2, rc = 1);
1535
1536         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle,
1537                           BYPASS_CAPA);
1538         if (rc != 0)
1539                 GOTO(unlock2, rc);
1540
1541         /* Get the latest parent's owner. */
1542         rc = dt_attr_get(env, parent, tla, BYPASS_CAPA);
1543         if (rc != 0)
1544                 GOTO(unlock2, rc);
1545
1546         tla->la_valid = LA_UID | LA_GID;
1547         rc = dt_attr_set(env, child, tla, handle, BYPASS_CAPA);
1548
1549         GOTO(unlock2, rc);
1550
1551 unlock2:
1552         dt_write_unlock(env, parent);
1553
1554 stop:
1555         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
1556
1557 unlock1:
1558         lfsck_layout_unlock(&lh);
1559
1560         return rc;
1561 }
1562
1563 /* If there are more than one MDT-objects claim as the OST-object's parent,
1564  * and the OST-object only recognizes one of them, then we need to generate
1565  * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
1566 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
1567                                                    struct lfsck_component *com,
1568                                                    struct lfsck_layout_req *llr,
1569                                                    struct lu_attr *la,
1570                                                    struct lu_buf *buf)
1571 {
1572         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1573         struct dt_allocation_hint       *hint   = &info->lti_hint;
1574         struct dt_object_format         *dof    = &info->lti_dof;
1575         struct dt_device                *pdev   = com->lc_lfsck->li_next;
1576         struct ost_id                   *oi     = &info->lti_oi;
1577         struct dt_object                *parent = llr->llr_parent->llo_obj;
1578         struct dt_device                *cdev   = lfsck_obj2dt_dev(llr->llr_child);
1579         struct dt_object                *child  = NULL;
1580         struct lu_device                *d      = &cdev->dd_lu_dev;
1581         struct lu_object                *o      = NULL;
1582         struct thandle                  *handle;
1583         struct lov_mds_md_v1            *lmm;
1584         struct lov_ost_data_v1          *objs;
1585         struct lustre_handle             lh     = { 0 };
1586         __u32                            magic;
1587         int                              rc;
1588         ENTRY;
1589
1590         CDEBUG(D_LFSCK, "Repair multiple references for: parent "DFID
1591                ", OST-index %u, stripe-index %u, owner %u:%u\n",
1592                PFID(lfsck_dto2fid(parent)), llr->llr_ost_idx,
1593                llr->llr_lov_idx, la->la_uid, la->la_gid);
1594
1595         rc = lfsck_layout_lock(env, com, parent, &lh,
1596                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
1597         if (rc != 0)
1598                 RETURN(rc);
1599
1600         handle = dt_trans_create(env, pdev);
1601         if (IS_ERR(handle))
1602                 GOTO(unlock1, rc = PTR_ERR(handle));
1603
1604         o = lu_object_anon(env, d, NULL);
1605         if (IS_ERR(o))
1606                 GOTO(stop, rc = PTR_ERR(o));
1607
1608         child = container_of(o, struct dt_object, do_lu);
1609         o = lu_object_locate(o->lo_header, d->ld_type);
1610         if (unlikely(o == NULL))
1611                 GOTO(stop, rc = -EINVAL);
1612
1613         child = container_of(o, struct dt_object, do_lu);
1614         la->la_valid = LA_UID | LA_GID;
1615         hint->dah_parent = NULL;
1616         hint->dah_mode = 0;
1617         dof->dof_type = DFT_REGULAR;
1618         rc = dt_declare_create(env, child, la, NULL, NULL, handle);
1619         if (rc != 0)
1620                 GOTO(stop, rc);
1621
1622         rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
1623                                   LU_XATTR_REPLACE, handle);
1624         if (rc != 0)
1625                 GOTO(stop, rc);
1626
1627         rc = dt_trans_start(env, pdev, handle);
1628         if (rc != 0)
1629                 GOTO(stop, rc);
1630
1631         dt_write_lock(env, parent, 0);
1632         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
1633                 GOTO(unlock2, rc = 0);
1634
1635         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV, BYPASS_CAPA);
1636         if (unlikely(rc == 0 || rc == -ENODATA || rc == -ERANGE))
1637                 GOTO(unlock2, rc = 0);
1638
1639         lmm = buf->lb_buf;
1640         rc = lfsck_layout_verify_header(lmm);
1641         if (rc != 0)
1642                 GOTO(unlock2, rc);
1643
1644         /* Someone change layout during the LFSCK, no need to repair then. */
1645         if (le16_to_cpu(lmm->lmm_layout_gen) != llr->llr_parent->llo_gen)
1646                 GOTO(unlock2, rc = 0);
1647
1648         rc = dt_create(env, child, la, hint, dof, handle);
1649         if (rc != 0)
1650                 GOTO(unlock2, rc);
1651
1652         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
1653          * been verified in lfsck_layout_verify_header() already. If some
1654          * new magic introduced in the future, then layout LFSCK needs to
1655          * be updated also. */
1656         magic = le32_to_cpu(lmm->lmm_magic);
1657         if (magic == LOV_MAGIC_V1) {
1658                 objs = &(lmm->lmm_objects[0]);
1659         } else {
1660                 LASSERT(magic == LOV_MAGIC_V3);
1661                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1662         }
1663
1664         lmm->lmm_layout_gen = cpu_to_le16(llr->llr_parent->llo_gen + 1);
1665         fid_to_ostid(lu_object_fid(&child->do_lu), oi);
1666         ostid_cpu_to_le(oi, &objs[llr->llr_lov_idx].l_ost_oi);
1667         objs[llr->llr_lov_idx].l_ost_gen = cpu_to_le32(0);
1668         objs[llr->llr_lov_idx].l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
1669         rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV,
1670                           LU_XATTR_REPLACE, handle, BYPASS_CAPA);
1671
1672         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1673
1674 unlock2:
1675         dt_write_unlock(env, parent);
1676
1677 stop:
1678         if (child != NULL)
1679                 lu_object_put(env, &child->do_lu);
1680
1681         dt_trans_stop(env, pdev, handle);
1682
1683 unlock1:
1684         lfsck_layout_unlock(&lh);
1685
1686         return rc;
1687 }
1688
1689 /* If the MDT-object and the OST-object have different owner information,
1690  * then trust the MDT-object, because the normal chown/chgrp handle order
1691  * is from MDT to OST, and it is possible that some chown/chgrp operation
1692  * is partly done. */
1693 static int lfsck_layout_repair_owner(const struct lu_env *env,
1694                                      struct lfsck_component *com,
1695                                      struct lfsck_layout_req *llr,
1696                                      struct lu_attr *pla)
1697 {
1698         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1699         struct lu_attr                  *tla    = &info->lti_la3;
1700         struct dt_object                *parent = llr->llr_parent->llo_obj;
1701         struct dt_object                *child  = llr->llr_child;
1702         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
1703         struct thandle                  *handle;
1704         int                              rc;
1705         ENTRY;
1706
1707         CDEBUG(D_LFSCK, "Repair inconsistent file owner for: parent "DFID
1708                ", child "DFID", OST-index %u, stripe-index %u, owner %u:%u\n",
1709                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
1710                llr->llr_ost_idx, llr->llr_lov_idx, pla->la_uid, pla->la_gid);
1711
1712         handle = dt_trans_create(env, dev);
1713         if (IS_ERR(handle))
1714                 RETURN(PTR_ERR(handle));
1715
1716         tla->la_uid = pla->la_uid;
1717         tla->la_gid = pla->la_gid;
1718         tla->la_valid = LA_UID | LA_GID;
1719         rc = dt_declare_attr_set(env, child, tla, handle);
1720         if (rc != 0)
1721                 GOTO(stop, rc);
1722
1723         rc = dt_trans_start(env, dev, handle);
1724         if (rc != 0)
1725                 GOTO(stop, rc);
1726
1727         /* Use the dt_object lock to serialize with destroy and attr_set. */
1728         dt_read_lock(env, parent, 0);
1729         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
1730                 GOTO(unlock, rc = 1);
1731
1732         /* Get the latest parent's owner. */
1733         rc = dt_attr_get(env, parent, tla, BYPASS_CAPA);
1734         if (rc != 0) {
1735                 CWARN("%s: fail to get the latest parent's ("DFID") owner, "
1736                       "not sure whether some others chown/chgrp during the "
1737                       "LFSCK: rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
1738                       PFID(lfsck_dto2fid(parent)), rc);
1739
1740                 GOTO(unlock, rc);
1741         }
1742
1743         /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
1744         if (unlikely(tla->la_uid != pla->la_uid ||
1745                      tla->la_gid != pla->la_gid))
1746                 GOTO(unlock, rc = 1);
1747
1748         tla->la_valid = LA_UID | LA_GID;
1749         rc = dt_attr_set(env, child, tla, handle, BYPASS_CAPA);
1750
1751         GOTO(unlock, rc);
1752
1753 unlock:
1754         dt_read_unlock(env, parent);
1755
1756 stop:
1757         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
1758
1759         return rc;
1760 }
1761
1762 /* Check whether the OST-object correctly back points to the
1763  * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
1764 static int lfsck_layout_check_parent(const struct lu_env *env,
1765                                      struct lfsck_component *com,
1766                                      struct dt_object *parent,
1767                                      const struct lu_fid *pfid,
1768                                      const struct lu_fid *cfid,
1769                                      const struct lu_attr *pla,
1770                                      const struct lu_attr *cla,
1771                                      struct lfsck_layout_req *llr,
1772                                      struct lu_buf *lov_ea, __u32 idx)
1773 {
1774         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1775         struct lu_buf                   *buf    = &info->lti_big_buf;
1776         struct dt_object                *tobj;
1777         struct lov_mds_md_v1            *lmm;
1778         struct lov_ost_data_v1          *objs;
1779         int                              rc;
1780         int                              i;
1781         __u32                            magic;
1782         __u16                            count;
1783         ENTRY;
1784
1785         if (fid_is_zero(pfid)) {
1786                 /* client never wrote. */
1787                 if (cla->la_size == 0 && cla->la_blocks == 0) {
1788                         if (unlikely(cla->la_uid != pla->la_uid ||
1789                                      cla->la_gid != pla->la_gid))
1790                                 RETURN (LLIT_INCONSISTENT_OWNER);
1791
1792                         RETURN(0);
1793                 }
1794
1795                 RETURN(LLIT_UNMATCHED_PAIR);
1796         }
1797
1798         if (unlikely(!fid_is_sane(pfid)))
1799                 RETURN(LLIT_UNMATCHED_PAIR);
1800
1801         if (lu_fid_eq(pfid, lu_object_fid(&parent->do_lu))) {
1802                 if (llr->llr_lov_idx == idx)
1803                         RETURN(0);
1804
1805                 RETURN(LLIT_UNMATCHED_PAIR);
1806         }
1807
1808         tobj = lfsck_object_find(env, com->lc_lfsck, pfid);
1809         if (tobj == NULL)
1810                 RETURN(LLIT_UNMATCHED_PAIR);
1811
1812         if (IS_ERR(tobj))
1813                 RETURN(PTR_ERR(tobj));
1814
1815         if (!dt_object_exists(tobj))
1816                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
1817
1818         /* Load the tobj's layout EA, in spite of it is a local MDT-object or
1819          * remote one on another MDT. Then check whether the given OST-object
1820          * is in such layout. If yes, it is multiple referenced, otherwise it
1821          * is unmatched referenced case. */
1822         rc = lfsck_layout_get_lovea(env, tobj, buf, NULL);
1823         if (rc == 0)
1824                 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
1825
1826         if (rc < 0)
1827                 GOTO(out, rc);
1828
1829         lmm = buf->lb_buf;
1830         rc = lfsck_layout_verify_header(lmm);
1831         if (rc != 0)
1832                 GOTO(out, rc);
1833
1834         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
1835          * been verified in lfsck_layout_verify_header() already. If some
1836          * new magic introduced in the future, then layout LFSCK needs to
1837          * be updated also. */
1838         magic = le32_to_cpu(lmm->lmm_magic);
1839         if (magic == LOV_MAGIC_V1) {
1840                 objs = &(lmm->lmm_objects[0]);
1841         } else {
1842                 LASSERT(magic == LOV_MAGIC_V3);
1843                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1844         }
1845
1846         count = le16_to_cpu(lmm->lmm_stripe_count);
1847         for (i = 0; i < count; i++, objs++) {
1848                 struct lu_fid           *tfid   = &info->lti_fid2;
1849                 struct ost_id           *oi     = &info->lti_oi;
1850
1851                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1852                 ostid_to_fid(tfid, oi, le32_to_cpu(objs->l_ost_idx));
1853                 if (lu_fid_eq(cfid, tfid)) {
1854                         *lov_ea = *buf;
1855
1856                         GOTO(out, rc = LLIT_MULTIPLE_REFERENCED);
1857                 }
1858         }
1859
1860         GOTO(out, rc = LLIT_UNMATCHED_PAIR);
1861
1862 out:
1863         lfsck_object_put(env, tobj);
1864
1865         return rc;
1866 }
1867
1868 static int lfsck_layout_assistant_handle_one(const struct lu_env *env,
1869                                              struct lfsck_component *com,
1870                                              struct lfsck_layout_req *llr)
1871 {
1872         struct lfsck_layout                  *lo     = com->lc_file_ram;
1873         struct lfsck_thread_info             *info   = lfsck_env_info(env);
1874         struct filter_fid_old                *pea    = &info->lti_old_pfid;
1875         struct lu_fid                        *pfid   = &info->lti_fid;
1876         struct lu_buf                        *buf    = NULL;
1877         struct dt_object                     *parent = llr->llr_parent->llo_obj;
1878         struct dt_object                     *child  = llr->llr_child;
1879         struct lu_attr                       *pla    = &info->lti_la;
1880         struct lu_attr                       *cla    = &info->lti_la2;
1881         struct lfsck_instance                *lfsck  = com->lc_lfsck;
1882         struct lfsck_bookmark                *bk     = &lfsck->li_bookmark_ram;
1883         enum lfsck_layout_inconsistency_type  type   = LLIT_NONE;
1884         __u32                                 idx    = 0;
1885         int                                   rc;
1886         ENTRY;
1887
1888         rc = dt_attr_get(env, parent, pla, BYPASS_CAPA);
1889         if (rc != 0) {
1890                 if (lu_object_is_dying(parent->do_lu.lo_header))
1891                         RETURN(0);
1892
1893                 GOTO(out, rc);
1894         }
1895
1896         rc = dt_attr_get(env, child, cla, BYPASS_CAPA);
1897         if (rc == -ENOENT) {
1898                 if (lu_object_is_dying(parent->do_lu.lo_header))
1899                         RETURN(0);
1900
1901                 type = LLIT_DANGLING;
1902                 goto repair;
1903         }
1904
1905         if (rc != 0)
1906                 GOTO(out, rc);
1907
1908         buf = lfsck_buf_get(env, pea, sizeof(struct filter_fid_old));
1909         rc= dt_xattr_get(env, child, buf, XATTR_NAME_FID, BYPASS_CAPA);
1910         if (unlikely(rc >= 0 && rc != sizeof(struct filter_fid_old) &&
1911                      rc != sizeof(struct filter_fid))) {
1912                 type = LLIT_UNMATCHED_PAIR;
1913                 goto repair;
1914         }
1915
1916         if (rc < 0 && rc != -ENODATA)
1917                 GOTO(out, rc);
1918
1919         if (rc == -ENODATA) {
1920                 fid_zero(pfid);
1921         } else {
1922                 fid_le_to_cpu(pfid, &pea->ff_parent);
1923                 /* OST-object does not save parent FID::f_ver, instead,
1924                  * the OST-object index in the parent MDT-object layout
1925                  * EA reuses the pfid->f_ver. */
1926                 idx = pfid->f_ver;
1927                 pfid->f_ver = 0;
1928         }
1929
1930         rc = lfsck_layout_check_parent(env, com, parent, pfid,
1931                                        lu_object_fid(&child->do_lu),
1932                                        pla, cla, llr, buf, idx);
1933         if (rc > 0) {
1934                 type = rc;
1935                 goto repair;
1936         }
1937
1938         if (rc < 0)
1939                 GOTO(out, rc);
1940
1941         if (unlikely(cla->la_uid != pla->la_uid ||
1942                      cla->la_gid != pla->la_gid)) {
1943                 type = LLIT_INCONSISTENT_OWNER;
1944                 goto repair;
1945         }
1946
1947 repair:
1948         if (bk->lb_param & LPF_DRYRUN) {
1949                 if (type != LLIT_NONE)
1950                         GOTO(out, rc = 1);
1951                 else
1952                         GOTO(out, rc = 0);
1953         }
1954
1955         switch (type) {
1956         case LLIT_DANGLING:
1957                 memset(cla, 0, sizeof(*cla));
1958                 cla->la_uid = pla->la_uid;
1959                 cla->la_gid = pla->la_gid;
1960                 cla->la_mode = S_IFREG | 0666;
1961                 cla->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1962                                 LA_ATIME | LA_MTIME | LA_CTIME;
1963                 rc = lfsck_layout_recreate_ostobj(env, com, llr, cla);
1964                 break;
1965         case LLIT_UNMATCHED_PAIR:
1966                 rc = lfsck_layout_repair_unmatched_pair(env, com, llr, pla);
1967                 break;
1968         case LLIT_MULTIPLE_REFERENCED:
1969                 rc = lfsck_layout_repair_multiple_references(env, com, llr,
1970                                                              pla, buf);
1971                 break;
1972         case LLIT_INCONSISTENT_OWNER:
1973                 rc = lfsck_layout_repair_owner(env, com, llr, pla);
1974                 break;
1975         default:
1976                 rc = 0;
1977                 break;
1978         }
1979
1980         GOTO(out, rc);
1981
1982 out:
1983         down_write(&com->lc_sem);
1984         if (rc < 0) {
1985                 /* If cannot touch the target server,
1986                  * mark the LFSCK as INCOMPLETE. */
1987                 if (rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -ETIMEDOUT ||
1988                     rc == -EHOSTDOWN || rc == -EHOSTUNREACH) {
1989                         lo->ll_flags |= LF_INCOMPLETE;
1990                         lo->ll_objs_skipped++;
1991                         rc = 0;
1992                 } else {
1993                         lo->ll_objs_failed_phase1++;
1994                 }
1995         } else if (rc > 0) {
1996                 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
1997                          "unknown type = %d\n", type);
1998
1999                 lo->ll_objs_repaired[type - 1]++;
2000         }
2001         up_write(&com->lc_sem);
2002
2003         return rc;
2004 }
2005
2006 static int lfsck_layout_assistant(void *args)
2007 {
2008         struct lfsck_thread_args        *lta     = args;
2009         struct lu_env                   *env     = &lta->lta_env;
2010         struct lfsck_component          *com     = lta->lta_com;
2011         struct lfsck_instance           *lfsck   = lta->lta_lfsck;
2012         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
2013         struct lfsck_position           *pos     = &com->lc_pos_start;
2014         struct lfsck_thread_info        *info    = lfsck_env_info(env);
2015         struct lfsck_request            *lr      = &info->lti_lr;
2016         struct lfsck_layout_master_data *llmd    = com->lc_data;
2017         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
2018         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
2019         struct lfsck_layout_req         *llr;
2020         struct l_wait_info               lwi     = { 0 };
2021         int                              rc      = 0;
2022         int                              rc1     = 0;
2023         __u32                            flags;
2024         ENTRY;
2025
2026         if (lta->lta_lsp->lsp_start != NULL)
2027                 flags  = lta->lta_lsp->lsp_start->ls_flags;
2028         else
2029                 flags = bk->lb_param;
2030         memset(lr, 0, sizeof(*lr));
2031         lr->lr_event = LE_START;
2032         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
2033         lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
2034                        LSV_ASYNC_WINDOWS;
2035         lr->lr_speed = bk->lb_speed_limit;
2036         lr->lr_version = bk->lb_version;
2037         lr->lr_param = bk->lb_param;
2038         lr->lr_async_windows = bk->lb_async_windows;
2039         if (pos->lp_oit_cookie <= 1)
2040                 lr->lr_param |= LPF_RESET;
2041
2042         rc = lfsck_layout_master_notify_others(env, com, lr, flags);
2043         if (rc != 0) {
2044                 CERROR("%s: fail to notify others for layout start: rc = %d\n",
2045                        lfsck_lfsck2name(lfsck), rc);
2046                 GOTO(fini, rc);
2047         }
2048
2049         spin_lock(&llmd->llmd_lock);
2050         thread_set_flags(athread, SVC_RUNNING);
2051         spin_unlock(&llmd->llmd_lock);
2052         wake_up_all(&mthread->t_ctl_waitq);
2053
2054         while (1) {
2055                 while (!list_empty(&llmd->llmd_req_list)) {
2056                         bool wakeup = false;
2057
2058                         if (unlikely(llmd->llmd_exit))
2059                                 GOTO(cleanup1, rc = llmd->llmd_post_result);
2060
2061                         llr = list_entry(llmd->llmd_req_list.next,
2062                                          struct lfsck_layout_req,
2063                                          llr_list);
2064                         /* Only the lfsck_layout_assistant thread itself can
2065                          * remove the "llr" from the head of the list, LFSCK
2066                          * engine thread only inserts other new "lld" at the
2067                          * end of the list. So it is safe to handle current
2068                          * "llr" without the spin_lock. */
2069                         rc = lfsck_layout_assistant_handle_one(env, com, llr);
2070                         spin_lock(&llmd->llmd_lock);
2071                         list_del_init(&llr->llr_list);
2072                         if (bk->lb_async_windows != 0 &&
2073                             llmd->llmd_prefetched >= bk->lb_async_windows)
2074                                 wakeup = true;
2075
2076                         llmd->llmd_prefetched--;
2077                         spin_unlock(&llmd->llmd_lock);
2078                         if (wakeup)
2079                                 wake_up_all(&mthread->t_ctl_waitq);
2080
2081                         lfsck_layout_req_fini(env, llr);
2082                         if (rc < 0 && bk->lb_param & LPF_FAILOUT)
2083                                 GOTO(cleanup1, rc);
2084                 }
2085
2086                 /* Wakeup the master engine if it is waiting in checkpoint. */
2087                 wake_up_all(&mthread->t_ctl_waitq);
2088
2089                 l_wait_event(athread->t_ctl_waitq,
2090                              !lfsck_layout_req_empty(llmd) ||
2091                              llmd->llmd_exit ||
2092                              llmd->llmd_to_post ||
2093                              llmd->llmd_to_double_scan,
2094                              &lwi);
2095
2096                 if (unlikely(llmd->llmd_exit))
2097                         GOTO(cleanup1, rc = llmd->llmd_post_result);
2098
2099                 if (!list_empty(&llmd->llmd_req_list))
2100                         continue;
2101
2102                 if (llmd->llmd_to_post) {
2103                         llmd->llmd_to_post = 0;
2104                         LASSERT(llmd->llmd_post_result > 0);
2105
2106                         memset(lr, 0, sizeof(*lr));
2107                         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
2108                         lr->lr_event = LE_PHASE1_DONE;
2109                         lr->lr_status = llmd->llmd_post_result;
2110                         rc = lfsck_layout_master_notify_others(env, com, lr, 0);
2111                         if (rc != 0)
2112                                 CERROR("%s: failed to notify others "
2113                                        "for layout post: rc = %d\n",
2114                                        lfsck_lfsck2name(lfsck), rc);
2115
2116                         /* Wakeup the master engine to go ahead. */
2117                         wake_up_all(&mthread->t_ctl_waitq);
2118                 }
2119
2120                 if (llmd->llmd_to_double_scan) {
2121                         llmd->llmd_to_double_scan = 0;
2122                         atomic_inc(&lfsck->li_double_scan_count);
2123                         llmd->llmd_in_double_scan = 1;
2124                         wake_up_all(&mthread->t_ctl_waitq);
2125
2126                         while (llmd->llmd_in_double_scan) {
2127                                 struct lfsck_tgt_descs  *ltds =
2128                                                         &lfsck->li_ost_descs;
2129                                 struct lfsck_tgt_desc   *ltd;
2130
2131                                 rc = lfsck_layout_master_query_others(env, com);
2132                                 if (lfsck_layout_master_to_orphan(llmd))
2133                                         goto orphan;
2134
2135                                 if (rc < 0)
2136                                         GOTO(cleanup2, rc);
2137
2138                                 /* Pull LFSCK status on related targets once
2139                                  * per 30 seconds if we are not notified. */
2140                                 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
2141                                                            cfs_time_seconds(1),
2142                                                            NULL, NULL);
2143                                 rc = l_wait_event(athread->t_ctl_waitq,
2144                                         lfsck_layout_master_to_orphan(llmd) ||
2145                                         llmd->llmd_exit ||
2146                                         !thread_is_running(mthread),
2147                                         &lwi);
2148
2149                                 if (unlikely(llmd->llmd_exit ||
2150                                              !thread_is_running(mthread)))
2151                                         GOTO(cleanup2, rc = 0);
2152
2153                                 if (rc == -ETIMEDOUT)
2154                                         continue;
2155
2156                                 if (rc < 0)
2157                                         GOTO(cleanup2, rc);
2158
2159 orphan:
2160                                 spin_lock(&ltds->ltd_lock);
2161                                 while (!list_empty(
2162                                                 &llmd->llmd_ost_phase2_list)) {
2163                                         ltd = list_entry(
2164                                               llmd->llmd_ost_phase2_list.next,
2165                                               struct lfsck_tgt_desc,
2166                                               ltd_layout_phase_list);
2167                                         list_del_init(
2168                                                 &ltd->ltd_layout_phase_list);
2169                                         spin_unlock(&ltds->ltd_lock);
2170
2171                                         rc = lfsck_layout_scan_orphan(env, com,
2172                                                                       ltd);
2173                                         if (rc != 0 &&
2174                                             bk->lb_param & LPF_FAILOUT)
2175                                                 GOTO(cleanup2, rc);
2176
2177                                         if (unlikely(llmd->llmd_exit ||
2178                                                 !thread_is_running(mthread)))
2179                                                 GOTO(cleanup2, rc = 0);
2180
2181                                         spin_lock(&ltds->ltd_lock);
2182                                 }
2183
2184                                 if (list_empty(&llmd->llmd_ost_phase1_list)) {
2185                                         spin_unlock(&ltds->ltd_lock);
2186                                         GOTO(cleanup2, rc = 1);
2187                                 }
2188                                 spin_unlock(&ltds->ltd_lock);
2189                         }
2190                 }
2191         }
2192
2193 cleanup1:
2194         /* Cleanup the unfinished requests. */
2195         spin_lock(&llmd->llmd_lock);
2196         if (rc < 0)
2197                 llmd->llmd_assistant_status = rc;
2198
2199         while (!list_empty(&llmd->llmd_req_list)) {
2200                 llr = list_entry(llmd->llmd_req_list.next,
2201                                  struct lfsck_layout_req,
2202                                  llr_list);
2203                 list_del_init(&llr->llr_list);
2204                 llmd->llmd_prefetched--;
2205                 spin_unlock(&llmd->llmd_lock);
2206                 lfsck_layout_req_fini(env, llr);
2207                 spin_lock(&llmd->llmd_lock);
2208         }
2209         spin_unlock(&llmd->llmd_lock);
2210
2211         LASSERTF(llmd->llmd_prefetched == 0, "unmatched prefeteched objs %d\n",
2212                  llmd->llmd_prefetched);
2213
2214 cleanup2:
2215         memset(lr, 0, sizeof(*lr));
2216         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
2217         if (rc > 0) {
2218                 lr->lr_event = LE_PHASE2_DONE;
2219                 flags = 0;
2220                 lr->lr_status = rc;
2221         } else if (rc == 0) {
2222                 lr->lr_event = LE_STOP;
2223                 if (lfsck->li_status == LS_PAUSED ||
2224                     lfsck->li_status == LS_CO_PAUSED) {
2225                         flags = 0;
2226                         lr->lr_status = LS_CO_PAUSED;
2227                 } else if (lfsck->li_status == LS_STOPPED ||
2228                          lfsck->li_status == LS_CO_STOPPED) {
2229                         flags = lfsck->li_flags;
2230                         if (flags & LPF_BROADCAST)
2231                                 lr->lr_status = LS_STOPPED;
2232                         else
2233                                 lr->lr_status = LS_CO_STOPPED;
2234                 } else {
2235                         LBUG();
2236                 }
2237         } else {
2238                 lr->lr_event = LE_STOP;
2239                 flags = 0;
2240                 lr->lr_status = LS_CO_FAILED;
2241         }
2242
2243         rc1 = lfsck_layout_master_notify_others(env, com, lr, flags);
2244         if (rc1 != 0) {
2245                 CERROR("%s: failed to notify others for layout quit: rc = %d\n",
2246                        lfsck_lfsck2name(lfsck), rc1);
2247                 rc = rc1;
2248         }
2249
2250         /* Under force exit case, some requests may be just freed without
2251          * verification, those objects should be re-handled when next run.
2252          * So not update the on-disk tracing file under such case. */
2253         if (!llmd->llmd_exit)
2254                 rc1 = lfsck_layout_double_scan_result(env, com, rc);
2255
2256 fini:
2257         if (llmd->llmd_in_double_scan)
2258                 atomic_dec(&lfsck->li_double_scan_count);
2259
2260         spin_lock(&llmd->llmd_lock);
2261         llmd->llmd_assistant_status = (rc1 != 0 ? rc1 : rc);
2262         thread_set_flags(athread, SVC_STOPPED);
2263         wake_up_all(&mthread->t_ctl_waitq);
2264         spin_unlock(&llmd->llmd_lock);
2265         lfsck_thread_args_fini(lta);
2266
2267         return rc;
2268 }
2269
2270 static int
2271 lfsck_layout_slave_async_interpret(const struct lu_env *env,
2272                                    struct ptlrpc_request *req,
2273                                    void *args, int rc)
2274 {
2275         struct lfsck_layout_slave_async_args *llsaa = args;
2276         struct obd_export                    *exp   = llsaa->llsaa_exp;
2277         struct lfsck_component               *com   = llsaa->llsaa_com;
2278         struct lfsck_layout_slave_target     *llst  = llsaa->llsaa_llst;
2279         struct lfsck_layout_slave_data       *llsd  = com->lc_data;
2280         bool                                  done  = false;
2281
2282         if (rc != 0) {
2283                 /* It is quite probably caused by target crash,
2284                  * to make the LFSCK can go ahead, assume that
2285                  * the target finished the LFSCK prcoessing. */
2286                 done = true;
2287         } else {
2288                 struct lfsck_reply *lr;
2289
2290                 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
2291                 if (lr->lr_status != LS_SCANNING_PHASE1 &&
2292                     lr->lr_status != LS_SCANNING_PHASE2)
2293                         done = true;
2294         }
2295         if (done)
2296                 lfsck_layout_llst_del(llsd, llst);
2297         lfsck_layout_llst_put(llst);
2298         lfsck_component_put(env, com);
2299         class_export_put(exp);
2300
2301         return 0;
2302 }
2303
2304 static int lfsck_layout_async_query(const struct lu_env *env,
2305                                     struct lfsck_component *com,
2306                                     struct obd_export *exp,
2307                                     struct lfsck_layout_slave_target *llst,
2308                                     struct lfsck_request *lr,
2309                                     struct ptlrpc_request_set *set)
2310 {
2311         struct lfsck_layout_slave_async_args *llsaa;
2312         struct ptlrpc_request                *req;
2313         struct lfsck_request                 *tmp;
2314         int                                   rc;
2315         ENTRY;
2316
2317         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
2318         if (req == NULL)
2319                 RETURN(-ENOMEM);
2320
2321         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
2322         if (rc != 0) {
2323                 ptlrpc_request_free(req);
2324                 RETURN(rc);
2325         }
2326
2327         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2328         *tmp = *lr;
2329         ptlrpc_request_set_replen(req);
2330
2331         llsaa = ptlrpc_req_async_args(req);
2332         llsaa->llsaa_exp = exp;
2333         llsaa->llsaa_com = lfsck_component_get(com);
2334         llsaa->llsaa_llst = llst;
2335         req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
2336         ptlrpc_set_add_req(set, req);
2337
2338         RETURN(0);
2339 }
2340
2341 static int lfsck_layout_async_notify(const struct lu_env *env,
2342                                      struct obd_export *exp,
2343                                      struct lfsck_request *lr,
2344                                      struct ptlrpc_request_set *set)
2345 {
2346         struct ptlrpc_request   *req;
2347         struct lfsck_request    *tmp;
2348         int                      rc;
2349         ENTRY;
2350
2351         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
2352         if (req == NULL)
2353                 RETURN(-ENOMEM);
2354
2355         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
2356         if (rc != 0) {
2357                 ptlrpc_request_free(req);
2358                 RETURN(rc);
2359         }
2360
2361         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2362         *tmp = *lr;
2363         ptlrpc_request_set_replen(req);
2364         ptlrpc_set_add_req(set, req);
2365
2366         RETURN(0);
2367 }
2368
2369 static int
2370 lfsck_layout_slave_query_master(const struct lu_env *env,
2371                                 struct lfsck_component *com)
2372 {
2373         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
2374         struct lfsck_instance            *lfsck = com->lc_lfsck;
2375         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
2376         struct lfsck_layout_slave_target *llst;
2377         struct obd_export                *exp;
2378         struct ptlrpc_request_set        *set;
2379         int                               cnt   = 0;
2380         int                               rc    = 0;
2381         int                               rc1   = 0;
2382         ENTRY;
2383
2384         set = ptlrpc_prep_set();
2385         if (set == NULL)
2386                 RETURN(-ENOMEM);
2387
2388         memset(lr, 0, sizeof(*lr));
2389         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
2390         lr->lr_event = LE_QUERY;
2391         lr->lr_active = LT_LAYOUT;
2392
2393         llsd->llsd_touch_gen++;
2394         spin_lock(&llsd->llsd_lock);
2395         while (!list_empty(&llsd->llsd_master_list)) {
2396                 llst = list_entry(llsd->llsd_master_list.next,
2397                                   struct lfsck_layout_slave_target,
2398                                   llst_list);
2399                 if (llst->llst_gen == llsd->llsd_touch_gen)
2400                         break;
2401
2402                 llst->llst_gen = llsd->llsd_touch_gen;
2403                 list_del(&llst->llst_list);
2404                 list_add_tail(&llst->llst_list,
2405                               &llsd->llsd_master_list);
2406                 atomic_inc(&llst->llst_ref);
2407                 spin_unlock(&llsd->llsd_lock);
2408
2409                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
2410                                                llst->llst_index);
2411                 if (exp == NULL) {
2412                         lfsck_layout_llst_del(llsd, llst);
2413                         lfsck_layout_llst_put(llst);
2414                         spin_lock(&llsd->llsd_lock);
2415                         continue;
2416                 }
2417
2418                 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
2419                 if (rc != 0) {
2420                         CERROR("%s: slave fail to query %s for layout: "
2421                                "rc = %d\n", lfsck_lfsck2name(lfsck),
2422                                exp->exp_obd->obd_name, rc);
2423                         rc1 = rc;
2424                         lfsck_layout_llst_put(llst);
2425                         class_export_put(exp);
2426                 } else {
2427                         cnt++;
2428                 }
2429                 spin_lock(&llsd->llsd_lock);
2430         }
2431         spin_unlock(&llsd->llsd_lock);
2432
2433         if (cnt > 0)
2434                 rc = ptlrpc_set_wait(set);
2435         ptlrpc_set_destroy(set);
2436
2437         RETURN(rc1 != 0 ? rc1 : rc);
2438 }
2439
2440 static void
2441 lfsck_layout_slave_notify_master(const struct lu_env *env,
2442                                  struct lfsck_component *com,
2443                                  enum lfsck_events event, int result)
2444 {
2445         struct lfsck_instance            *lfsck = com->lc_lfsck;
2446         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
2447         struct lfsck_request             *lr    = &lfsck_env_info(env)->lti_lr;
2448         struct lfsck_layout_slave_target *llst;
2449         struct obd_export                *exp;
2450         struct ptlrpc_request_set        *set;
2451         int                               cnt   = 0;
2452         int                               rc;
2453         ENTRY;
2454
2455         set = ptlrpc_prep_set();
2456         if (set == NULL)
2457                 RETURN_EXIT;
2458
2459         memset(lr, 0, sizeof(*lr));
2460         lr->lr_event = event;
2461         lr->lr_flags = LEF_FROM_OST;
2462         lr->lr_status = result;
2463         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
2464         lr->lr_active = LT_LAYOUT;
2465         llsd->llsd_touch_gen++;
2466         spin_lock(&llsd->llsd_lock);
2467         while (!list_empty(&llsd->llsd_master_list)) {
2468                 llst = list_entry(llsd->llsd_master_list.next,
2469                                   struct lfsck_layout_slave_target,
2470                                   llst_list);
2471                 if (llst->llst_gen == llsd->llsd_touch_gen)
2472                         break;
2473
2474                 llst->llst_gen = llsd->llsd_touch_gen;
2475                 list_del(&llst->llst_list);
2476                 list_add_tail(&llst->llst_list,
2477                               &llsd->llsd_master_list);
2478                 atomic_inc(&llst->llst_ref);
2479                 spin_unlock(&llsd->llsd_lock);
2480
2481                 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
2482                                                llst->llst_index);
2483                 if (exp == NULL) {
2484                         lfsck_layout_llst_del(llsd, llst);
2485                         lfsck_layout_llst_put(llst);
2486                         spin_lock(&llsd->llsd_lock);
2487                         continue;
2488                 }
2489
2490                 rc = lfsck_layout_async_notify(env, exp, lr, set);
2491                 if (rc != 0)
2492                         CERROR("%s: slave fail to notify %s for layout: "
2493                                "rc = %d\n", lfsck_lfsck2name(lfsck),
2494                                exp->exp_obd->obd_name, rc);
2495                 else
2496                         cnt++;
2497                 lfsck_layout_llst_put(llst);
2498                 class_export_put(exp);
2499                 spin_lock(&llsd->llsd_lock);
2500         }
2501         spin_unlock(&llsd->llsd_lock);
2502
2503         if (cnt > 0)
2504                 rc = ptlrpc_set_wait(set);
2505
2506         ptlrpc_set_destroy(set);
2507
2508         RETURN_EXIT;
2509 }
2510
2511 /* layout APIs */
2512
2513 static int lfsck_layout_reset(const struct lu_env *env,
2514                               struct lfsck_component *com, bool init)
2515 {
2516         struct lfsck_layout     *lo    = com->lc_file_ram;
2517         int                      rc;
2518
2519         down_write(&com->lc_sem);
2520         if (init) {
2521                 memset(lo, 0, com->lc_file_size);
2522         } else {
2523                 __u32 count = lo->ll_success_count;
2524                 __u64 last_time = lo->ll_time_last_complete;
2525
2526                 memset(lo, 0, com->lc_file_size);
2527                 lo->ll_success_count = count;
2528                 lo->ll_time_last_complete = last_time;
2529         }
2530
2531         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
2532         lo->ll_status = LS_INIT;
2533
2534         rc = lfsck_layout_store(env, com);
2535         up_write(&com->lc_sem);
2536
2537         return rc;
2538 }
2539
2540 static void lfsck_layout_fail(const struct lu_env *env,
2541                               struct lfsck_component *com, bool new_checked)
2542 {
2543         struct lfsck_layout *lo = com->lc_file_ram;
2544
2545         down_write(&com->lc_sem);
2546         if (new_checked)
2547                 com->lc_new_checked++;
2548         lo->ll_objs_failed_phase1++;
2549         if (lo->ll_pos_first_inconsistent == 0) {
2550                 struct lfsck_instance *lfsck = com->lc_lfsck;
2551
2552                 lo->ll_pos_first_inconsistent =
2553                         lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
2554                                                         lfsck->li_di_oit);
2555         }
2556         up_write(&com->lc_sem);
2557 }
2558
2559 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
2560                                           struct lfsck_component *com, bool init)
2561 {
2562         struct lfsck_instance           *lfsck   = com->lc_lfsck;
2563         struct lfsck_layout             *lo      = com->lc_file_ram;
2564         struct lfsck_layout_master_data *llmd    = com->lc_data;
2565         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
2566         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
2567         struct l_wait_info               lwi     = { 0 };
2568         int                              rc;
2569
2570         if (com->lc_new_checked == 0 && !init)
2571                 return 0;
2572
2573         l_wait_event(mthread->t_ctl_waitq,
2574                      list_empty(&llmd->llmd_req_list) ||
2575                      !thread_is_running(mthread) ||
2576                      thread_is_stopped(athread),
2577                      &lwi);
2578
2579         if (!thread_is_running(mthread) || thread_is_stopped(athread))
2580                 return 0;
2581
2582         down_write(&com->lc_sem);
2583         if (init) {
2584                 lo->ll_pos_latest_start = lfsck->li_pos_current.lp_oit_cookie;
2585         } else {
2586                 lo->ll_pos_last_checkpoint =
2587                                         lfsck->li_pos_current.lp_oit_cookie;
2588                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
2589                                 HALF_SEC - lfsck->li_time_last_checkpoint);
2590                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
2591                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
2592                 com->lc_new_checked = 0;
2593         }
2594
2595         rc = lfsck_layout_store(env, com);
2596         up_write(&com->lc_sem);
2597
2598         return rc;
2599 }
2600
2601 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
2602                                          struct lfsck_component *com, bool init)
2603 {
2604         struct lfsck_instance   *lfsck = com->lc_lfsck;
2605         struct lfsck_layout     *lo    = com->lc_file_ram;
2606         int                      rc;
2607
2608         if (com->lc_new_checked == 0 && !init)
2609                 return 0;
2610
2611         down_write(&com->lc_sem);
2612
2613         if (init) {
2614                 lo->ll_pos_latest_start = lfsck->li_pos_current.lp_oit_cookie;
2615         } else {
2616                 lo->ll_pos_last_checkpoint =
2617                                         lfsck->li_pos_current.lp_oit_cookie;
2618                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
2619                                 HALF_SEC - lfsck->li_time_last_checkpoint);
2620                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
2621                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
2622                 com->lc_new_checked = 0;
2623         }
2624
2625         rc = lfsck_layout_store(env, com);
2626
2627         up_write(&com->lc_sem);
2628
2629         return rc;
2630 }
2631
2632 static int lfsck_layout_prep(const struct lu_env *env,
2633                              struct lfsck_component *com)
2634 {
2635         struct lfsck_instance   *lfsck  = com->lc_lfsck;
2636         struct lfsck_layout     *lo     = com->lc_file_ram;
2637         struct lfsck_position   *pos    = &com->lc_pos_start;
2638
2639         fid_zero(&pos->lp_dir_parent);
2640         pos->lp_dir_cookie = 0;
2641         if (lo->ll_status == LS_COMPLETED ||
2642             lo->ll_status == LS_PARTIAL) {
2643                 int rc;
2644
2645                 rc = lfsck_layout_reset(env, com, false);
2646                 if (rc != 0)
2647                         return rc;
2648         }
2649
2650         down_write(&com->lc_sem);
2651
2652         lo->ll_time_latest_start = cfs_time_current_sec();
2653
2654         spin_lock(&lfsck->li_lock);
2655         if (lo->ll_flags & LF_SCANNED_ONCE) {
2656                 if (!lfsck->li_drop_dryrun ||
2657                     lo->ll_pos_first_inconsistent == 0) {
2658                         lo->ll_status = LS_SCANNING_PHASE2;
2659                         list_del_init(&com->lc_link);
2660                         list_add_tail(&com->lc_link,
2661                                       &lfsck->li_list_double_scan);
2662                         pos->lp_oit_cookie = 0;
2663                 } else {
2664                         int i;
2665
2666                         lo->ll_status = LS_SCANNING_PHASE1;
2667                         lo->ll_run_time_phase1 = 0;
2668                         lo->ll_run_time_phase2 = 0;
2669                         lo->ll_objs_checked_phase1 = 0;
2670                         lo->ll_objs_checked_phase2 = 0;
2671                         lo->ll_objs_failed_phase1 = 0;
2672                         lo->ll_objs_failed_phase2 = 0;
2673                         for (i = 0; i < LLIT_MAX; i++)
2674                                 lo->ll_objs_repaired[i] = 0;
2675
2676                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
2677                 }
2678         } else {
2679                 lo->ll_status = LS_SCANNING_PHASE1;
2680                 if (!lfsck->li_drop_dryrun ||
2681                     lo->ll_pos_first_inconsistent == 0)
2682                         pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
2683                 else
2684                         pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
2685         }
2686         spin_unlock(&lfsck->li_lock);
2687
2688         up_write(&com->lc_sem);
2689
2690         return 0;
2691 }
2692
2693 static int lfsck_layout_slave_prep(const struct lu_env *env,
2694                                    struct lfsck_component *com,
2695                                    struct lfsck_start_param *lsp)
2696 {
2697         struct lfsck_layout             *lo     = com->lc_file_ram;
2698         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
2699         int                              rc;
2700
2701         /* XXX: For a new scanning, generate OST-objects
2702          *      bitmap for orphan detection. */
2703
2704         rc = lfsck_layout_prep(env, com);
2705         if (rc != 0 || lo->ll_status != LS_SCANNING_PHASE1 ||
2706             !lsp->lsp_index_valid)
2707                 return rc;
2708
2709         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
2710
2711         return rc;
2712 }
2713
2714 static int lfsck_layout_master_prep(const struct lu_env *env,
2715                                     struct lfsck_component *com,
2716                                     struct lfsck_start_param *lsp)
2717 {
2718         struct lfsck_instance           *lfsck   = com->lc_lfsck;
2719         struct lfsck_layout_master_data *llmd    = com->lc_data;
2720         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
2721         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
2722         struct lfsck_thread_args        *lta;
2723         long                             rc;
2724         ENTRY;
2725
2726         rc = lfsck_layout_prep(env, com);
2727         if (rc != 0)
2728                 RETURN(rc);
2729
2730         llmd->llmd_assistant_status = 0;
2731         llmd->llmd_post_result = 0;
2732         llmd->llmd_to_post = 0;
2733         llmd->llmd_to_double_scan = 0;
2734         llmd->llmd_in_double_scan = 0;
2735         llmd->llmd_exit = 0;
2736         thread_set_flags(athread, 0);
2737
2738         lta = lfsck_thread_args_init(lfsck, com, lsp);
2739         if (IS_ERR(lta))
2740                 RETURN(PTR_ERR(lta));
2741
2742         rc = PTR_ERR(kthread_run(lfsck_layout_assistant, lta, "lfsck_layout"));
2743         if (IS_ERR_VALUE(rc)) {
2744                 CERROR("%s: Cannot start LFSCK layout assistant thread: "
2745                        "rc = %ld\n", lfsck_lfsck2name(lfsck), rc);
2746                 lfsck_thread_args_fini(lta);
2747         } else {
2748                 struct l_wait_info lwi = { 0 };
2749
2750                 l_wait_event(mthread->t_ctl_waitq,
2751                              thread_is_running(athread) ||
2752                              thread_is_stopped(athread),
2753                              &lwi);
2754                 if (unlikely(!thread_is_running(athread)))
2755                         rc = llmd->llmd_assistant_status;
2756                 else
2757                         rc = 0;
2758         }
2759
2760         RETURN(rc);
2761 }
2762
2763 /* Pre-fetch the attribute for each stripe in the given layout EA. */
2764 static int lfsck_layout_scan_stripes(const struct lu_env *env,
2765                                      struct lfsck_component *com,
2766                                      struct dt_object *parent,
2767                                      struct lov_mds_md_v1 *lmm)
2768 {
2769         struct lfsck_thread_info        *info    = lfsck_env_info(env);
2770         struct lfsck_instance           *lfsck   = com->lc_lfsck;
2771         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
2772         struct lfsck_layout             *lo      = com->lc_file_ram;
2773         struct lfsck_layout_master_data *llmd    = com->lc_data;
2774         struct lfsck_layout_object      *llo     = NULL;
2775         struct lov_ost_data_v1          *objs;
2776         struct lfsck_tgt_descs          *ltds    = &lfsck->li_ost_descs;
2777         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
2778         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
2779                 struct l_wait_info       lwi     = { 0 };
2780         struct lu_buf                   *buf;
2781         int                              rc      = 0;
2782         int                              i;
2783         __u32                            magic;
2784         __u16                            count;
2785         __u16                            gen;
2786         ENTRY;
2787
2788         buf = lfsck_buf_get(env, &info->lti_old_pfid,
2789                             sizeof(struct filter_fid_old));
2790         count = le16_to_cpu(lmm->lmm_stripe_count);
2791         gen = le16_to_cpu(lmm->lmm_layout_gen);
2792         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2793          * been verified in lfsck_layout_verify_header() already. If some
2794          * new magic introduced in the future, then layout LFSCK needs to
2795          * be updated also. */
2796         magic = le32_to_cpu(lmm->lmm_magic);
2797         if (magic == LOV_MAGIC_V1) {
2798                 objs = &(lmm->lmm_objects[0]);
2799         } else {
2800                 LASSERT(magic == LOV_MAGIC_V3);
2801                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2802         }
2803
2804         for (i = 0; i < count; i++, objs++) {
2805                 struct lu_fid           *fid    = &info->lti_fid;
2806                 struct ost_id           *oi     = &info->lti_oi;
2807                 struct lfsck_layout_req *llr;
2808                 struct lfsck_tgt_desc   *tgt    = NULL;
2809                 struct dt_object        *cobj   = NULL;
2810                 __u32                    index  =
2811                                         le32_to_cpu(objs->l_ost_idx);
2812                 bool                     wakeup = false;
2813
2814                 l_wait_event(mthread->t_ctl_waitq,
2815                              bk->lb_async_windows == 0 ||
2816                              llmd->llmd_prefetched < bk->lb_async_windows ||
2817                              !thread_is_running(mthread) ||
2818                              thread_is_stopped(athread),
2819                              &lwi);
2820
2821                 if (unlikely(!thread_is_running(mthread)) ||
2822                              thread_is_stopped(athread))
2823                         GOTO(out, rc = 0);
2824
2825                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2826                 ostid_to_fid(fid, oi, index);
2827                 tgt = lfsck_tgt_get(ltds, index);
2828                 if (unlikely(tgt == NULL)) {
2829                         lo->ll_flags |= LF_INCOMPLETE;
2830                         goto next;
2831                 }
2832
2833                 cobj = lfsck_object_find_by_dev(env, tgt->ltd_tgt, fid);
2834                 if (IS_ERR(cobj)) {
2835                         rc = PTR_ERR(cobj);
2836                         goto next;
2837                 }
2838
2839                 rc = dt_declare_attr_get(env, cobj, BYPASS_CAPA);
2840                 if (rc != 0)
2841                         goto next;
2842
2843                 rc = dt_declare_xattr_get(env, cobj, buf, XATTR_NAME_FID,
2844                                           BYPASS_CAPA);
2845                 if (rc != 0)
2846                         goto next;
2847
2848                 if (llo == NULL) {
2849                         llo = lfsck_layout_object_init(env, parent, gen);
2850                         if (IS_ERR(llo)) {
2851                                 rc = PTR_ERR(llo);
2852                                 goto next;
2853                         }
2854                 }
2855
2856                 llr = lfsck_layout_req_init(llo, cobj, index, i);
2857                 if (IS_ERR(llr)) {
2858                         rc = PTR_ERR(llr);
2859                         goto next;
2860                 }
2861
2862                 cobj = NULL;
2863                 spin_lock(&llmd->llmd_lock);
2864                 if (llmd->llmd_assistant_status < 0) {
2865                         spin_unlock(&llmd->llmd_lock);
2866                         lfsck_layout_req_fini(env, llr);
2867                         lfsck_tgt_put(tgt);
2868                         RETURN(llmd->llmd_assistant_status);
2869                 }
2870
2871                 list_add_tail(&llr->llr_list, &llmd->llmd_req_list);
2872                 if (llmd->llmd_prefetched == 0)
2873                         wakeup = true;
2874
2875                 llmd->llmd_prefetched++;
2876                 spin_unlock(&llmd->llmd_lock);
2877                 if (wakeup)
2878                         wake_up_all(&athread->t_ctl_waitq);
2879
2880 next:
2881                 down_write(&com->lc_sem);
2882                 com->lc_new_checked++;
2883                 if (rc < 0)
2884                         lo->ll_objs_failed_phase1++;
2885                 up_write(&com->lc_sem);
2886
2887                 if (cobj != NULL && !IS_ERR(cobj))
2888                         lu_object_put(env, &cobj->do_lu);
2889
2890                 if (likely(tgt != NULL))
2891                         lfsck_tgt_put(tgt);
2892
2893                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
2894                         GOTO(out, rc);
2895         }
2896
2897         GOTO(out, rc = 0);
2898
2899 out:
2900         if (llo != NULL && !IS_ERR(llo))
2901                 lfsck_layout_object_put(env, llo);
2902
2903         return rc;
2904 }
2905
2906 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
2907  * the OST-object's attribute and generate an structure lfsck_layout_req on the
2908  * list ::llmd_req_list.
2909  *
2910  * For each request on above list, the lfsck_layout_assistant thread compares
2911  * the OST side attribute with local attribute, if inconsistent, then repair it.
2912  *
2913  * All above processing is async mode with pipeline. */
2914 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
2915                                         struct lfsck_component *com,
2916                                         struct dt_object *obj)
2917 {
2918         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2919         struct ost_id                   *oi     = &info->lti_oi;
2920         struct lfsck_layout             *lo     = com->lc_file_ram;
2921         struct lfsck_layout_master_data *llmd   = com->lc_data;
2922         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2923         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
2924         struct thandle                  *handle = NULL;
2925         struct lu_buf                   *buf    = &info->lti_big_buf;
2926         struct lov_mds_md_v1            *lmm    = NULL;
2927         struct dt_device                *dev    = lfsck->li_bottom;
2928         struct lustre_handle             lh     = { 0 };
2929         ssize_t                          buflen = buf->lb_len;
2930         int                              rc     = 0;
2931         bool                             locked = false;
2932         bool                             stripe = false;
2933         ENTRY;
2934
2935         if (!S_ISREG(lfsck_object_type(obj)))
2936                 GOTO(out, rc = 0);
2937
2938         if (llmd->llmd_assistant_status < 0)
2939                 GOTO(out, rc = -ESRCH);
2940
2941         fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
2942         lmm_oi_cpu_to_le(oi, oi);
2943         dt_read_lock(env, obj, 0);
2944         locked = true;
2945
2946 again:
2947         rc = lfsck_layout_get_lovea(env, obj, buf, &buflen);
2948         if (rc <= 0)
2949                 GOTO(out, rc);
2950
2951         buf->lb_len = rc;
2952         lmm = buf->lb_buf;
2953         rc = lfsck_layout_verify_header(lmm);
2954         if (rc != 0)
2955                 GOTO(out, rc);
2956
2957         if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0)
2958                 GOTO(out, stripe = true);
2959
2960         /* Inconsistent lmm_oi, should be repaired. */
2961         CDEBUG(D_LFSCK, "Repair bad lmm_oi for "DFID"\n",
2962                PFID(lfsck_dto2fid(obj)));
2963
2964         if (bk->lb_param & LPF_DRYRUN) {
2965                 down_write(&com->lc_sem);
2966                 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
2967                 up_write(&com->lc_sem);
2968
2969                 GOTO(out, stripe = true);
2970         }
2971
2972         if (!lustre_handle_is_used(&lh)) {
2973                 dt_read_unlock(env, obj);
2974                 locked = false;
2975                 buf->lb_len = buflen;
2976                 rc = lfsck_layout_lock(env, com, obj, &lh,
2977                                        MDS_INODELOCK_LAYOUT |
2978                                        MDS_INODELOCK_XATTR);
2979                 if (rc != 0)
2980                         GOTO(out, rc);
2981
2982                 handle = dt_trans_create(env, dev);
2983                 if (IS_ERR(handle))
2984                         GOTO(out, rc = PTR_ERR(handle));
2985
2986                 rc = dt_declare_xattr_set(env, obj, buf, XATTR_NAME_LOV,
2987                                           LU_XATTR_REPLACE, handle);
2988                 if (rc != 0)
2989                         GOTO(out, rc);
2990
2991                 rc = dt_trans_start_local(env, dev, handle);
2992                 if (rc != 0)
2993                         GOTO(out, rc);
2994
2995                 dt_write_lock(env, obj, 0);
2996                 locked = true;
2997
2998                 goto again;
2999         }
3000
3001         lmm->lmm_oi = *oi;
3002         rc = dt_xattr_set(env, obj, buf, XATTR_NAME_LOV,
3003                           LU_XATTR_REPLACE, handle, BYPASS_CAPA);
3004         if (rc != 0)
3005                 GOTO(out, rc);
3006
3007         down_write(&com->lc_sem);
3008         lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
3009         up_write(&com->lc_sem);
3010
3011         GOTO(out, stripe = true);
3012
3013 out:
3014         if (locked) {
3015                 if (lustre_handle_is_used(&lh))
3016                         dt_write_unlock(env, obj);
3017                 else
3018                         dt_read_unlock(env, obj);
3019         }
3020
3021         if (handle != NULL && !IS_ERR(handle))
3022                 dt_trans_stop(env, dev, handle);
3023
3024         lfsck_layout_unlock(&lh);
3025         if (stripe) {
3026                 rc = lfsck_layout_scan_stripes(env, com, obj, lmm);
3027         } else {
3028                 down_write(&com->lc_sem);
3029                 com->lc_new_checked++;
3030                 if (rc < 0)
3031                         lo->ll_objs_failed_phase1++;
3032                 up_write(&com->lc_sem);
3033         }
3034         buf->lb_len = buflen;
3035
3036         return rc;
3037 }
3038
3039 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
3040                                        struct lfsck_component *com,
3041                                        struct dt_object *obj)
3042 {
3043         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3044         struct lfsck_layout             *lo     = com->lc_file_ram;
3045         const struct lu_fid             *fid    = lfsck_dto2fid(obj);
3046         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
3047         struct lfsck_layout_seq         *lls;
3048         __u64                            seq;
3049         __u64                            oid;
3050         int                              rc;
3051         ENTRY;
3052
3053         /* XXX: Update OST-objects bitmap for orphan detection. */
3054
3055         LASSERT(llsd != NULL);
3056
3057         down_write(&com->lc_sem);
3058         if (fid_is_idif(fid))
3059                 seq = 0;
3060         else if (!fid_is_norm(fid) ||
3061                  !fid_is_for_ostobj(env, lfsck->li_next, obj, fid))
3062                 GOTO(unlock, rc = 0);
3063         else
3064                 seq = fid_seq(fid);
3065         com->lc_new_checked++;
3066
3067         lls = lfsck_layout_seq_lookup(llsd, seq);
3068         if (lls == NULL) {
3069                 OBD_ALLOC_PTR(lls);
3070                 if (unlikely(lls == NULL))
3071                         GOTO(unlock, rc = -ENOMEM);
3072
3073                 INIT_LIST_HEAD(&lls->lls_list);
3074                 lls->lls_seq = seq;
3075                 rc = lfsck_layout_lastid_load(env, com, lls);
3076                 if (rc != 0) {
3077                         lo->ll_objs_failed_phase1++;
3078                         OBD_FREE_PTR(lls);
3079                         GOTO(unlock, rc);
3080                 }
3081
3082                 lfsck_layout_seq_insert(llsd, lls);
3083         }
3084
3085         if (unlikely(fid_is_last_id(fid)))
3086                 GOTO(unlock, rc = 0);
3087
3088         oid = fid_oid(fid);
3089         if (oid > lls->lls_lastid_known)
3090                 lls->lls_lastid_known = oid;
3091
3092         if (oid > lls->lls_lastid) {
3093                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
3094                         /* OFD may create new objects during LFSCK scanning. */
3095                         rc = lfsck_layout_lastid_reload(env, com, lls);
3096                         if (unlikely(rc != 0))
3097                                 CWARN("%s: failed to reload LAST_ID for "LPX64
3098                                       ": rc = %d\n",
3099                                       lfsck_lfsck2name(com->lc_lfsck),
3100                                       lls->lls_seq, rc);
3101                         if (oid <= lls->lls_lastid)
3102                                 GOTO(unlock, rc = 0);
3103
3104                         LASSERT(lfsck->li_out_notify != NULL);
3105
3106                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
3107                                              LE_LASTID_REBUILDING);
3108                         lo->ll_flags |= LF_CRASHED_LASTID;
3109                 }
3110
3111                 lls->lls_lastid = oid;
3112                 lls->lls_dirty = 1;
3113         }
3114
3115         GOTO(unlock, rc = 0);
3116
3117 unlock:
3118         up_write(&com->lc_sem);
3119
3120         return rc;
3121 }
3122
3123 static int lfsck_layout_exec_dir(const struct lu_env *env,
3124                                  struct lfsck_component *com,
3125                                  struct dt_object *obj,
3126                                  struct lu_dirent *ent)
3127 {
3128         return 0;
3129 }
3130
3131 static int lfsck_layout_master_post(const struct lu_env *env,
3132                                     struct lfsck_component *com,
3133                                     int result, bool init)
3134 {
3135         struct lfsck_instance           *lfsck   = com->lc_lfsck;
3136         struct lfsck_layout             *lo      = com->lc_file_ram;
3137         struct lfsck_layout_master_data *llmd    = com->lc_data;
3138         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
3139         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
3140         struct l_wait_info               lwi     = { 0 };
3141         int                              rc;
3142         ENTRY;
3143
3144
3145         llmd->llmd_post_result = result;
3146         llmd->llmd_to_post = 1;
3147         if (llmd->llmd_post_result <= 0)
3148                 llmd->llmd_exit = 1;
3149
3150         wake_up_all(&athread->t_ctl_waitq);
3151         l_wait_event(mthread->t_ctl_waitq,
3152                      (result > 0 && list_empty(&llmd->llmd_req_list)) ||
3153                      thread_is_stopped(athread),
3154                      &lwi);
3155
3156         if (llmd->llmd_assistant_status < 0)
3157                 result = llmd->llmd_assistant_status;
3158
3159         down_write(&com->lc_sem);
3160         spin_lock(&lfsck->li_lock);
3161         /* When LFSCK failed, there may be some prefetched objects those are
3162          * not been processed yet, we do not know the exactly position, then
3163          * just restart from last check-point next time. */
3164         if (!init && !llmd->llmd_exit)
3165                 lo->ll_pos_last_checkpoint =
3166                                         lfsck->li_pos_current.lp_oit_cookie;
3167
3168         if (result > 0) {
3169                 lo->ll_status = LS_SCANNING_PHASE2;
3170                 lo->ll_flags |= LF_SCANNED_ONCE;
3171                 lo->ll_flags &= ~LF_UPGRADE;
3172                 list_del_init(&com->lc_link);
3173                 list_add_tail(&com->lc_link, &lfsck->li_list_double_scan);
3174         } else if (result == 0) {
3175                 lo->ll_status = lfsck->li_status;
3176                 if (lo->ll_status == 0)
3177                         lo->ll_status = LS_STOPPED;
3178                 if (lo->ll_status != LS_PAUSED) {
3179                         list_del_init(&com->lc_link);
3180                         list_add_tail(&com->lc_link, &lfsck->li_list_idle);
3181                 }
3182         } else {
3183                 lo->ll_status = LS_FAILED;
3184                 list_del_init(&com->lc_link);
3185                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
3186         }
3187         spin_unlock(&lfsck->li_lock);
3188
3189         if (!init) {
3190                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3191                                 HALF_SEC - lfsck->li_time_last_checkpoint);
3192                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
3193                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
3194                 com->lc_new_checked = 0;
3195         }
3196
3197         rc = lfsck_layout_store(env, com);
3198         up_write(&com->lc_sem);
3199
3200         RETURN(rc);
3201 }
3202
3203 static int lfsck_layout_slave_post(const struct lu_env *env,
3204                                    struct lfsck_component *com,
3205                                    int result, bool init)
3206 {
3207         struct lfsck_instance   *lfsck = com->lc_lfsck;
3208         struct lfsck_layout     *lo    = com->lc_file_ram;
3209         int                      rc;
3210         bool                     done  = false;
3211
3212         rc = lfsck_layout_lastid_store(env, com);
3213         if (rc != 0)
3214                 result = rc;
3215
3216         LASSERT(lfsck->li_out_notify != NULL);
3217
3218         down_write(&com->lc_sem);
3219
3220         spin_lock(&lfsck->li_lock);
3221         if (!init)
3222                 lo->ll_pos_last_checkpoint =
3223                                         lfsck->li_pos_current.lp_oit_cookie;
3224         if (result > 0) {
3225                 lo->ll_status = LS_SCANNING_PHASE2;
3226                 lo->ll_flags |= LF_SCANNED_ONCE;
3227                 if (lo->ll_flags & LF_CRASHED_LASTID) {
3228                         done = true;
3229                         lo->ll_flags &= ~LF_CRASHED_LASTID;
3230                 }
3231                 lo->ll_flags &= ~LF_UPGRADE;
3232                 list_del_init(&com->lc_link);
3233                 list_add_tail(&com->lc_link, &lfsck->li_list_double_scan);
3234         } else if (result == 0) {
3235                 lo->ll_status = lfsck->li_status;
3236                 if (lo->ll_status == 0)
3237                         lo->ll_status = LS_STOPPED;
3238                 if (lo->ll_status != LS_PAUSED) {
3239                         list_del_init(&com->lc_link);
3240                         list_add_tail(&com->lc_link, &lfsck->li_list_idle);
3241                 }
3242         } else {
3243                 lo->ll_status = LS_FAILED;
3244                 list_del_init(&com->lc_link);
3245                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
3246         }
3247         spin_unlock(&lfsck->li_lock);
3248
3249         if (done)
3250                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
3251                                      LE_LASTID_REBUILT);
3252
3253         if (!init) {
3254                 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3255                                 HALF_SEC - lfsck->li_time_last_checkpoint);
3256                 lo->ll_time_last_checkpoint = cfs_time_current_sec();
3257                 lo->ll_objs_checked_phase1 += com->lc_new_checked;
3258                 com->lc_new_checked = 0;
3259         }
3260
3261         rc = lfsck_layout_store(env, com);
3262
3263         up_write(&com->lc_sem);
3264
3265         lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
3266
3267         return rc;
3268 }
3269
3270 static int lfsck_layout_dump(const struct lu_env *env,
3271                              struct lfsck_component *com, char *buf, int len)
3272 {
3273         struct lfsck_instance   *lfsck = com->lc_lfsck;
3274         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
3275         struct lfsck_layout     *lo    = com->lc_file_ram;
3276         int                      save  = len;
3277         int                      ret   = -ENOSPC;
3278         int                      rc;
3279
3280         down_read(&com->lc_sem);
3281         rc = snprintf(buf, len,
3282                       "name: lfsck_layout\n"
3283                       "magic: %#x\n"
3284                       "version: %d\n"
3285                       "status: %s\n",
3286                       lo->ll_magic,
3287                       bk->lb_version,
3288                       lfsck_status2names(lo->ll_status));
3289         if (rc <= 0)
3290                 goto out;
3291
3292         buf += rc;
3293         len -= rc;
3294         rc = lfsck_bits_dump(&buf, &len, lo->ll_flags, lfsck_flags_names,
3295                              "flags");
3296         if (rc < 0)
3297                 goto out;
3298
3299         rc = lfsck_bits_dump(&buf, &len, bk->lb_param, lfsck_param_names,
3300                              "param");
3301         if (rc < 0)
3302                 goto out;
3303
3304         rc = lfsck_time_dump(&buf, &len, lo->ll_time_last_complete,
3305                              "time_since_last_completed");
3306         if (rc < 0)
3307                 goto out;
3308
3309         rc = lfsck_time_dump(&buf, &len, lo->ll_time_latest_start,
3310                              "time_since_latest_start");
3311         if (rc < 0)
3312                 goto out;
3313
3314         rc = lfsck_time_dump(&buf, &len, lo->ll_time_last_checkpoint,
3315                              "time_since_last_checkpoint");
3316         if (rc < 0)
3317                 goto out;
3318
3319         rc = snprintf(buf, len,
3320                       "latest_start_position: "LPU64"\n"
3321                       "last_checkpoint_position: "LPU64"\n"
3322                       "first_failure_position: "LPU64"\n",
3323                       lo->ll_pos_latest_start,
3324                       lo->ll_pos_last_checkpoint,
3325                       lo->ll_pos_first_inconsistent);
3326         if (rc <= 0)
3327                 goto out;
3328
3329         buf += rc;
3330         len -= rc;
3331
3332         rc = snprintf(buf, len,
3333                       "success_count: %u\n"
3334                       "repaired_dangling: "LPU64"\n"
3335                       "repaired_unmatched_pair: "LPU64"\n"
3336                       "repaired_multiple_referenced: "LPU64"\n"
3337                       "repaired_orphan: "LPU64"\n"
3338                       "repaired_inconsistent_owner: "LPU64"\n"
3339                       "repaired_others: "LPU64"\n"
3340                       "skipped: "LPU64"\n"
3341                       "failed_phase1: "LPU64"\n"
3342                       "failed_phase2: "LPU64"\n",
3343                       lo->ll_success_count,
3344                       lo->ll_objs_repaired[LLIT_DANGLING - 1],
3345                       lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
3346                       lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
3347                       lo->ll_objs_repaired[LLIT_ORPHAN - 1],
3348                       lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
3349                       lo->ll_objs_repaired[LLIT_OTHERS - 1],
3350                       lo->ll_objs_skipped,
3351                       lo->ll_objs_failed_phase1,
3352                       lo->ll_objs_failed_phase2);
3353         if (rc <= 0)
3354                 goto out;
3355
3356         buf += rc;
3357         len -= rc;
3358
3359         if (lo->ll_status == LS_SCANNING_PHASE1) {
3360                 __u64 pos;
3361                 const struct dt_it_ops *iops;
3362                 cfs_duration_t duration = cfs_time_current() -
3363                                           lfsck->li_time_last_checkpoint;
3364                 __u64 checked = lo->ll_objs_checked_phase1 + com->lc_new_checked;
3365                 __u64 speed = checked;
3366                 __u64 new_checked = com->lc_new_checked * HZ;
3367                 __u32 rtime = lo->ll_run_time_phase1 +
3368                               cfs_duration_sec(duration + HALF_SEC);
3369
3370                 if (duration != 0)
3371                         do_div(new_checked, duration);
3372                 if (rtime != 0)
3373                         do_div(speed, rtime);
3374                 rc = snprintf(buf, len,
3375                               "checked_phase1: "LPU64"\n"
3376                               "checked_phase2: "LPU64"\n"
3377                               "run_time_phase1: %u seconds\n"
3378                               "run_time_phase2: %u seconds\n"
3379                               "average_speed_phase1: "LPU64" items/sec\n"
3380                               "average_speed_phase2: N/A\n"
3381                               "real-time_speed_phase1: "LPU64" items/sec\n"
3382                               "real-time_speed_phase2: N/A\n",
3383                               checked,
3384                               lo->ll_objs_checked_phase2,
3385                               rtime,
3386                               lo->ll_run_time_phase2,
3387                               speed,
3388                               new_checked);
3389                 if (rc <= 0)
3390                         goto out;
3391
3392                 buf += rc;
3393                 len -= rc;
3394
3395                 LASSERT(lfsck->li_di_oit != NULL);
3396
3397                 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
3398
3399                 /* The low layer otable-based iteration position may NOT
3400                  * exactly match the layout-based directory traversal
3401                  * cookie. Generally, it is not a serious issue. But the
3402                  * caller should NOT make assumption on that. */
3403                 pos = iops->store(env, lfsck->li_di_oit);
3404                 if (!lfsck->li_current_oit_processed)
3405                         pos--;
3406                 rc = snprintf(buf, len, "current_position: "LPU64"\n", pos);
3407                 if (rc <= 0)
3408                         goto out;
3409
3410                 buf += rc;
3411                 len -= rc;
3412         } else {
3413                 /* XXX: LS_SCANNING_PHASE2 will be handled in the future. */
3414                 __u64 speed1 = lo->ll_objs_checked_phase1;
3415                 __u64 speed2 = lo->ll_objs_checked_phase2;
3416
3417                 if (lo->ll_run_time_phase1 != 0)
3418                         do_div(speed1, lo->ll_run_time_phase1);
3419                 if (lo->ll_run_time_phase2 != 0)
3420                         do_div(speed2, lo->ll_run_time_phase2);
3421                 rc = snprintf(buf, len,
3422                               "checked_phase1: "LPU64"\n"
3423                               "checked_phase2: "LPU64"\n"
3424                               "run_time_phase1: %u seconds\n"
3425                               "run_time_phase2: %u seconds\n"
3426                               "average_speed_phase1: "LPU64" items/sec\n"
3427                               "average_speed_phase2: "LPU64" objs/sec\n"
3428                               "real-time_speed_phase1: N/A\n"
3429                               "real-time_speed_phase2: N/A\n"
3430                               "current_position: N/A\n",
3431                               lo->ll_objs_checked_phase1,
3432                               lo->ll_objs_checked_phase2,
3433                               lo->ll_run_time_phase1,
3434                               lo->ll_run_time_phase2,
3435                               speed1,
3436                               speed2);
3437                 if (rc <= 0)
3438                         goto out;
3439
3440                 buf += rc;
3441                 len -= rc;
3442         }
3443         ret = save - len;
3444
3445 out:
3446         up_read(&com->lc_sem);
3447
3448         return ret;
3449 }
3450
3451 static int lfsck_layout_master_double_scan(const struct lu_env *env,
3452                                            struct lfsck_component *com)
3453 {
3454         struct lfsck_layout_master_data *llmd    = com->lc_data;
3455         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
3456         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
3457         struct lfsck_layout             *lo      = com->lc_file_ram;
3458         struct l_wait_info               lwi     = { 0 };
3459
3460         if (unlikely(lo->ll_status != LS_SCANNING_PHASE2))
3461                 return 0;
3462
3463         llmd->llmd_to_double_scan = 1;
3464         wake_up_all(&athread->t_ctl_waitq);
3465         l_wait_event(mthread->t_ctl_waitq,
3466                      llmd->llmd_in_double_scan ||
3467                      thread_is_stopped(athread),
3468                      &lwi);
3469         if (llmd->llmd_assistant_status < 0)
3470                 return llmd->llmd_assistant_status;
3471
3472         return 0;
3473 }
3474
3475 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
3476                                           struct lfsck_component *com)
3477 {
3478         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3479         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
3480         struct lfsck_layout             *lo     = com->lc_file_ram;
3481         struct ptlrpc_thread            *thread = &lfsck->li_thread;
3482         int                              rc;
3483         ENTRY;
3484
3485         if (unlikely(lo->ll_status != LS_SCANNING_PHASE2))
3486                 RETURN(0);
3487
3488         atomic_inc(&lfsck->li_double_scan_count);
3489
3490         com->lc_new_checked = 0;
3491         com->lc_new_scanned = 0;
3492         com->lc_time_last_checkpoint = cfs_time_current();
3493         com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
3494                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
3495
3496         while (1) {
3497                 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(30),
3498                                                      NULL, NULL);
3499
3500                 rc = lfsck_layout_slave_query_master(env, com);
3501                 if (list_empty(&llsd->llsd_master_list)) {
3502                         if (unlikely(!thread_is_running(thread)))
3503                                 rc = 0;
3504                         else
3505                                 rc = 1;
3506
3507                         GOTO(done, rc);
3508                 }
3509
3510                 if (rc < 0)
3511                         GOTO(done, rc);
3512
3513                 rc = l_wait_event(thread->t_ctl_waitq,
3514                                   !thread_is_running(thread) ||
3515                                   list_empty(&llsd->llsd_master_list),
3516                                   &lwi);
3517                 if (unlikely(!thread_is_running(thread)))
3518                         GOTO(done, rc = 0);
3519
3520                 if (rc == -ETIMEDOUT)
3521                         continue;
3522
3523                 GOTO(done, rc = (rc < 0 ? rc : 1));
3524         }
3525
3526 done:
3527         rc = lfsck_layout_double_scan_result(env, com, rc);
3528
3529         if (atomic_dec_and_test(&lfsck->li_double_scan_count))
3530                 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
3531
3532         return rc;
3533 }
3534
3535 static void lfsck_layout_master_data_release(const struct lu_env *env,
3536                                              struct lfsck_component *com)
3537 {
3538         struct lfsck_layout_master_data *llmd   = com->lc_data;
3539         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3540         struct lfsck_tgt_descs          *ltds;
3541         struct lfsck_tgt_desc           *ltd;
3542         struct lfsck_tgt_desc           *next;
3543
3544         LASSERT(llmd != NULL);
3545         LASSERT(thread_is_init(&llmd->llmd_thread) ||
3546                 thread_is_stopped(&llmd->llmd_thread));
3547         LASSERT(list_empty(&llmd->llmd_req_list));
3548
3549         com->lc_data = NULL;
3550
3551         ltds = &lfsck->li_ost_descs;
3552         spin_lock(&ltds->ltd_lock);
3553         list_for_each_entry_safe(ltd, next, &llmd->llmd_ost_phase1_list,
3554                                  ltd_layout_phase_list) {
3555                 list_del_init(&ltd->ltd_layout_phase_list);
3556         }
3557         list_for_each_entry_safe(ltd, next, &llmd->llmd_ost_phase2_list,
3558                                  ltd_layout_phase_list) {
3559                 list_del_init(&ltd->ltd_layout_phase_list);
3560         }
3561         list_for_each_entry_safe(ltd, next, &llmd->llmd_ost_list,
3562                                  ltd_layout_list) {
3563                 list_del_init(&ltd->ltd_layout_list);
3564         }
3565         list_for_each_entry_safe(ltd, next, &llmd->llmd_mdt_phase1_list,
3566                                  ltd_layout_phase_list) {
3567                 list_del_init(&ltd->ltd_layout_phase_list);
3568         }
3569         list_for_each_entry_safe(ltd, next, &llmd->llmd_mdt_phase2_list,
3570                                  ltd_layout_phase_list) {
3571                 list_del_init(&ltd->ltd_layout_phase_list);
3572         }
3573         list_for_each_entry_safe(ltd, next, &llmd->llmd_mdt_list,
3574                                  ltd_layout_list) {
3575                 list_del_init(&ltd->ltd_layout_list);
3576         }
3577         spin_unlock(&ltds->ltd_lock);
3578
3579         OBD_FREE_PTR(llmd);
3580 }
3581
3582 static void lfsck_layout_slave_data_release(const struct lu_env *env,
3583                                             struct lfsck_component *com)
3584 {
3585         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3586         struct lfsck_layout_seq          *lls;
3587         struct lfsck_layout_seq          *next;
3588         struct lfsck_layout_slave_target *llst;
3589         struct lfsck_layout_slave_target *tmp;
3590
3591         LASSERT(llsd != NULL);
3592
3593         com->lc_data = NULL;
3594
3595         list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
3596                                      lls_list) {
3597                 list_del_init(&lls->lls_list);
3598                 lfsck_object_put(env, lls->lls_lastid_obj);
3599                 OBD_FREE_PTR(lls);
3600         }
3601
3602         list_for_each_entry_safe(llst, tmp, &llsd->llsd_master_list,
3603                                  llst_list) {
3604                 list_del_init(&llst->llst_list);
3605                 OBD_FREE_PTR(llst);
3606         }
3607
3608         OBD_FREE_PTR(llsd);
3609 }
3610
3611 static void lfsck_layout_master_quit(const struct lu_env *env,
3612                                      struct lfsck_component *com)
3613 {
3614         struct lfsck_layout_master_data *llmd    = com->lc_data;
3615         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
3616         struct ptlrpc_thread            *athread = &llmd->llmd_thread;
3617         struct l_wait_info               lwi     = { 0 };
3618
3619         llmd->llmd_exit = 1;
3620         wake_up_all(&athread->t_ctl_waitq);
3621         l_wait_event(mthread->t_ctl_waitq,
3622                      thread_is_init(athread) ||
3623                      thread_is_stopped(athread),
3624                      &lwi);
3625 }
3626
3627 static int lfsck_layout_master_in_notify(const struct lu_env *env,
3628                                          struct lfsck_component *com,
3629                                          struct lfsck_request *lr)
3630 {
3631         struct lfsck_instance           *lfsck = com->lc_lfsck;
3632         struct lfsck_layout             *lo    = com->lc_file_ram;
3633         struct lfsck_layout_master_data *llmd  = com->lc_data;
3634         struct lfsck_tgt_descs          *ltds;
3635         struct lfsck_tgt_desc           *ltd;
3636         ENTRY;
3637
3638         if (lr->lr_event != LE_PHASE1_DONE &&
3639             lr->lr_event != LE_PHASE2_DONE &&
3640             lr->lr_event != LE_STOP)
3641                 RETURN(-EINVAL);
3642
3643         if (lr->lr_flags & LEF_FROM_OST)
3644                 ltds = &lfsck->li_ost_descs;
3645         else
3646                 ltds = &lfsck->li_mdt_descs;
3647         spin_lock(&ltds->ltd_lock);
3648         ltd = LTD_TGT(ltds, lr->lr_index);
3649         if (ltd == NULL) {
3650                 spin_unlock(&ltds->ltd_lock);
3651
3652                 RETURN(-ENODEV);
3653         }
3654
3655         list_del_init(&ltd->ltd_layout_phase_list);
3656         switch (lr->lr_event) {
3657         case LE_PHASE1_DONE:
3658                 if (lr->lr_status <= 0) {
3659                         ltd->ltd_layout_done = 1;
3660                         list_del_init(&ltd->ltd_layout_list);
3661                         lo->ll_flags |= LF_INCOMPLETE;
3662                         break;
3663                 }
3664
3665                 if (lr->lr_flags & LEF_FROM_OST) {
3666                         if (list_empty(&ltd->ltd_layout_list))
3667                                 list_add_tail(&ltd->ltd_layout_list,
3668                                               &llmd->llmd_ost_list);
3669                         list_add_tail(&ltd->ltd_layout_phase_list,
3670                                       &llmd->llmd_ost_phase2_list);
3671                 } else {
3672                         if (list_empty(&ltd->ltd_layout_list))
3673                                 list_add_tail(&ltd->ltd_layout_list,
3674                                               &llmd->llmd_mdt_list);
3675                         list_add_tail(&ltd->ltd_layout_phase_list,
3676                                       &llmd->llmd_mdt_phase2_list);
3677                 }
3678                 break;
3679         case LE_PHASE2_DONE:
3680                 ltd->ltd_layout_done = 1;
3681                 list_del_init(&ltd->ltd_layout_list);
3682                 break;
3683         case LE_STOP:
3684                 ltd->ltd_layout_done = 1;
3685                 list_del_init(&ltd->ltd_layout_list);
3686                 if (!(lr->lr_flags & LEF_FORCE_STOP))
3687                         lo->ll_flags |= LF_INCOMPLETE;
3688                 break;
3689         default:
3690                 break;
3691         }
3692         spin_unlock(&ltds->ltd_lock);
3693
3694         if (lr->lr_flags & LEF_FORCE_STOP) {
3695                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
3696
3697                 memset(stop, 0, sizeof(*stop));
3698                 stop->ls_status = lr->lr_status;
3699                 stop->ls_flags = lr->lr_param;
3700                 lfsck_stop(env, lfsck->li_bottom, stop);
3701         } else if (lfsck_layout_master_to_orphan(llmd)) {
3702                 wake_up_all(&llmd->llmd_thread.t_ctl_waitq);
3703         }
3704
3705         RETURN(0);
3706 }
3707
3708 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
3709                                         struct lfsck_component *com,
3710                                         struct lfsck_request *lr)
3711 {
3712         struct lfsck_instance            *lfsck = com->lc_lfsck;
3713         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3714         struct lfsck_layout_slave_target *llst;
3715         ENTRY;
3716
3717         if (lr->lr_event != LE_PHASE2_DONE &&
3718             lr->lr_event != LE_STOP)
3719                 RETURN(-EINVAL);
3720
3721         llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index);
3722         if (llst == NULL)
3723                 RETURN(-ENODEV);
3724
3725         lfsck_layout_llst_put(llst);
3726         if (list_empty(&llsd->llsd_master_list)) {
3727                 switch (lr->lr_event) {
3728                 case LE_PHASE2_DONE:
3729                         wake_up_all(&lfsck->li_thread.t_ctl_waitq);
3730                         break;
3731                 case LE_STOP: {
3732                         struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
3733
3734                         memset(stop, 0, sizeof(*stop));
3735                         stop->ls_status = lr->lr_status;
3736                         stop->ls_flags = lr->lr_param;
3737                         lfsck_stop(env, lfsck->li_bottom, stop);
3738                         break;
3739                 }
3740                 default:
3741                         break;
3742                 }
3743         }
3744
3745         RETURN(0);
3746 }
3747
3748 static int lfsck_layout_query(const struct lu_env *env,
3749                               struct lfsck_component *com)
3750 {
3751         struct lfsck_layout *lo = com->lc_file_ram;
3752
3753         return lo->ll_status;
3754 }
3755
3756 static int lfsck_layout_master_stop_notify(const struct lu_env *env,
3757                                            struct lfsck_component *com,
3758                                            struct lfsck_tgt_descs *ltds,
3759                                            struct lfsck_tgt_desc *ltd,
3760                                            struct ptlrpc_request_set *set)
3761 {
3762         struct lfsck_thread_info          *info  = lfsck_env_info(env);
3763         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
3764         struct lfsck_request              *lr    = &info->lti_lr;
3765         struct lfsck_instance             *lfsck = com->lc_lfsck;
3766         int                                rc;
3767
3768         LASSERT(list_empty(&ltd->ltd_layout_list));
3769         LASSERT(list_empty(&ltd->ltd_layout_phase_list));
3770
3771         memset(lr, 0, sizeof(*lr));
3772         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
3773         lr->lr_event = LE_STOP;
3774         lr->lr_active = LT_LAYOUT;
3775         if (ltds == &lfsck->li_ost_descs) {
3776                 lr->lr_flags = LEF_TO_OST;
3777         } else {
3778                 if (ltd->ltd_index == lfsck_dev_idx(lfsck->li_bottom))
3779                         return 0;
3780
3781                 lr->lr_flags = 0;
3782         }
3783         lr->lr_status = LS_CO_STOPPED;
3784
3785         laia->laia_com = com;
3786         laia->laia_ltds = ltds;
3787         laia->laia_ltd = ltd;
3788         laia->laia_lr = lr;
3789
3790         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
3791                                  lfsck_layout_master_async_interpret,
3792                                  laia, LFSCK_NOTIFY);
3793         if (rc != 0)
3794                 CERROR("%s: Fail to notify %s %x for co-stop: rc = %d\n",
3795                        lfsck_lfsck2name(lfsck),
3796                        (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
3797                        ltd->ltd_index, rc);
3798
3799         return rc;
3800 }
3801
3802 /* with lfsck::li_lock held */
3803 static int lfsck_layout_slave_join(const struct lu_env *env,
3804                                    struct lfsck_component *com,
3805                                    struct lfsck_start_param *lsp)
3806 {
3807         struct lfsck_instance            *lfsck = com->lc_lfsck;
3808         struct lfsck_layout_slave_data   *llsd  = com->lc_data;
3809         struct lfsck_layout_slave_target *llst;
3810         struct lfsck_start               *start = lsp->lsp_start;
3811         int                               rc    = 0;
3812         ENTRY;
3813
3814         if (!lsp->lsp_index_valid || start == NULL ||
3815             !(start->ls_flags & LPF_ALL_MDT))
3816                 RETURN(-EALREADY);
3817
3818         spin_unlock(&lfsck->li_lock);
3819         rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
3820         spin_lock(&lfsck->li_lock);
3821         if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
3822                 spin_unlock(&lfsck->li_lock);
3823                 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index);
3824                 if (llst != NULL)
3825                         lfsck_layout_llst_put(llst);
3826                 spin_lock(&lfsck->li_lock);
3827                 rc = -EAGAIN;
3828         }
3829
3830         RETURN(rc);
3831 }
3832
3833 static struct lfsck_operations lfsck_layout_master_ops = {
3834         .lfsck_reset            = lfsck_layout_reset,
3835         .lfsck_fail             = lfsck_layout_fail,
3836         .lfsck_checkpoint       = lfsck_layout_master_checkpoint,
3837         .lfsck_prep             = lfsck_layout_master_prep,
3838         .lfsck_exec_oit         = lfsck_layout_master_exec_oit,
3839         .lfsck_exec_dir         = lfsck_layout_exec_dir,
3840         .lfsck_post             = lfsck_layout_master_post,
3841         .lfsck_dump             = lfsck_layout_dump,
3842         .lfsck_double_scan      = lfsck_layout_master_double_scan,
3843         .lfsck_data_release     = lfsck_layout_master_data_release,
3844         .lfsck_quit             = lfsck_layout_master_quit,
3845         .lfsck_in_notify        = lfsck_layout_master_in_notify,
3846         .lfsck_query            = lfsck_layout_query,
3847         .lfsck_stop_notify      = lfsck_layout_master_stop_notify,
3848 };
3849
3850 static struct lfsck_operations lfsck_layout_slave_ops = {
3851         .lfsck_reset            = lfsck_layout_reset,
3852         .lfsck_fail             = lfsck_layout_fail,
3853         .lfsck_checkpoint       = lfsck_layout_slave_checkpoint,
3854         .lfsck_prep             = lfsck_layout_slave_prep,
3855         .lfsck_exec_oit         = lfsck_layout_slave_exec_oit,
3856         .lfsck_exec_dir         = lfsck_layout_exec_dir,
3857         .lfsck_post             = lfsck_layout_slave_post,
3858         .lfsck_dump             = lfsck_layout_dump,
3859         .lfsck_double_scan      = lfsck_layout_slave_double_scan,
3860         .lfsck_data_release     = lfsck_layout_slave_data_release,
3861         .lfsck_in_notify        = lfsck_layout_slave_in_notify,
3862         .lfsck_query            = lfsck_layout_query,
3863         .lfsck_join             = lfsck_layout_slave_join,
3864 };
3865
3866 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
3867 {
3868         struct lfsck_component  *com;
3869         struct lfsck_layout     *lo;
3870         struct dt_object        *root = NULL;
3871         struct dt_object        *obj;
3872         int                      rc;
3873         ENTRY;
3874
3875         OBD_ALLOC_PTR(com);
3876         if (com == NULL)
3877                 RETURN(-ENOMEM);
3878
3879         INIT_LIST_HEAD(&com->lc_link);
3880         INIT_LIST_HEAD(&com->lc_link_dir);
3881         init_rwsem(&com->lc_sem);
3882         atomic_set(&com->lc_ref, 1);
3883         com->lc_lfsck = lfsck;
3884         com->lc_type = LT_LAYOUT;
3885         if (lfsck->li_master) {
3886                 struct lfsck_layout_master_data *llmd;
3887
3888                 com->lc_ops = &lfsck_layout_master_ops;
3889                 OBD_ALLOC_PTR(llmd);
3890                 if (llmd == NULL)
3891                         GOTO(out, rc = -ENOMEM);
3892
3893                 INIT_LIST_HEAD(&llmd->llmd_req_list);
3894                 spin_lock_init(&llmd->llmd_lock);
3895                 INIT_LIST_HEAD(&llmd->llmd_ost_list);
3896                 INIT_LIST_HEAD(&llmd->llmd_ost_phase1_list);
3897                 INIT_LIST_HEAD(&llmd->llmd_ost_phase2_list);
3898                 INIT_LIST_HEAD(&llmd->llmd_mdt_list);
3899                 INIT_LIST_HEAD(&llmd->llmd_mdt_phase1_list);
3900                 INIT_LIST_HEAD(&llmd->llmd_mdt_phase2_list);
3901                 init_waitqueue_head(&llmd->llmd_thread.t_ctl_waitq);
3902                 com->lc_data = llmd;
3903         } else {
3904                 struct lfsck_layout_slave_data *llsd;
3905
3906                 com->lc_ops = &lfsck_layout_slave_ops;
3907                 OBD_ALLOC_PTR(llsd);
3908                 if (llsd == NULL)
3909                         GOTO(out, rc = -ENOMEM);
3910
3911                 INIT_LIST_HEAD(&llsd->llsd_seq_list);
3912                 INIT_LIST_HEAD(&llsd->llsd_master_list);
3913                 spin_lock_init(&llsd->llsd_lock);
3914                 com->lc_data = llsd;
3915         }
3916         com->lc_file_size = sizeof(*lo);
3917         OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
3918         if (com->lc_file_ram == NULL)
3919                 GOTO(out, rc = -ENOMEM);
3920
3921         OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
3922         if (com->lc_file_disk == NULL)
3923                 GOTO(out, rc = -ENOMEM);
3924
3925         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
3926         if (IS_ERR(root))
3927                 GOTO(out, rc = PTR_ERR(root));
3928
3929         if (unlikely(!dt_try_as_dir(env, root)))
3930                 GOTO(out, rc = -ENOTDIR);
3931
3932         obj = local_file_find_or_create(env, lfsck->li_los, root,
3933                                         lfsck_layout_name,
3934                                         S_IFREG | S_IRUGO | S_IWUSR);
3935         if (IS_ERR(obj))
3936                 GOTO(out, rc = PTR_ERR(obj));
3937
3938         com->lc_obj = obj;
3939         rc = lfsck_layout_load(env, com);
3940         if (rc > 0)
3941                 rc = lfsck_layout_reset(env, com, true);
3942         else if (rc == -ENOENT)
3943                 rc = lfsck_layout_init(env, com);
3944
3945         if (rc != 0)
3946                 GOTO(out, rc);
3947
3948         lo = com->lc_file_ram;
3949         switch (lo->ll_status) {
3950         case LS_INIT:
3951         case LS_COMPLETED:
3952         case LS_FAILED:
3953         case LS_STOPPED:
3954         case LS_PARTIAL:
3955                 spin_lock(&lfsck->li_lock);
3956                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
3957                 spin_unlock(&lfsck->li_lock);
3958                 break;
3959         default:
3960                 CERROR("%s: unknown lfsck_layout status: rc = %u\n",
3961                        lfsck_lfsck2name(lfsck), lo->ll_status);
3962                 /* fall through */
3963         case LS_SCANNING_PHASE1:
3964         case LS_SCANNING_PHASE2:
3965                 /* No need to store the status to disk right now.
3966                  * If the system crashed before the status stored,
3967                  * it will be loaded back when next time. */
3968                 lo->ll_status = LS_CRASHED;
3969                 lo->ll_flags |= LF_INCOMPLETE;
3970                 /* fall through */
3971         case LS_PAUSED:
3972         case LS_CRASHED:
3973         case LS_CO_FAILED:
3974         case LS_CO_STOPPED:
3975         case LS_CO_PAUSED:
3976                 spin_lock(&lfsck->li_lock);
3977                 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
3978                 spin_unlock(&lfsck->li_lock);
3979                 break;
3980         }
3981
3982         if (lo->ll_flags & LF_CRASHED_LASTID) {
3983                 LASSERT(lfsck->li_out_notify != NULL);
3984
3985                 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
3986                                      LE_LASTID_REBUILDING);
3987         }
3988
3989         GOTO(out, rc = 0);
3990
3991 out:
3992         if (root != NULL && !IS_ERR(root))
3993                 lu_object_put(env, &root->do_lu);
3994
3995         if (rc != 0)
3996                 lfsck_component_cleanup(env, com);
3997
3998         return rc;
3999 }