lustre/lfsck/lfsck_lib.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) 2012, 2013, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_lib.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 #define DEBUG_SUBSYSTEM S_LFSCK
  32
  33 #include <libcfs/list.h>
  34 #include <lu_object.h>
  35 #include <dt_object.h>
  36 #include <md_object.h>
  37 #include <lustre_fld.h>
  38 #include <lustre_lib.h>
  39 #include <lustre_net.h>
  40 #include <lustre_lfsck.h>
  41 #include <lustre/lustre_lfsck_user.h>
  42
  43 #include "lfsck_internal.h"
  44
  45 /* define lfsck thread key */
  46 LU_KEY_INIT(lfsck, struct lfsck_thread_info);
  47
  48 static void lfsck_key_fini(const struct lu_context *ctx,
  49                            struct lu_context_key *key, void *data)
  50 {
  51         struct lfsck_thread_info *info = data;
  52
  53         lu_buf_free(&info->lti_linkea_buf);
  54         lu_buf_free(&info->lti_big_buf);
  55         OBD_FREE_PTR(info);
  56 }
  57
  58 LU_CONTEXT_KEY_DEFINE(lfsck, LCT_MD_THREAD | LCT_DT_THREAD);
  59 LU_KEY_INIT_GENERIC(lfsck);
  60
  61 static CFS_LIST_HEAD(lfsck_instance_list);
  62 static struct list_head lfsck_ost_orphan_list;
  63 static struct list_head lfsck_mdt_orphan_list;
  64 static DEFINE_SPINLOCK(lfsck_instance_lock);
  65
  66 static const char *lfsck_status_names[] = {
  67         [LS_INIT]               = "init",
  68         [LS_SCANNING_PHASE1]    = "scanning-phase1",
  69         [LS_SCANNING_PHASE2]    = "scanning-phase2",
  70         [LS_COMPLETED]          = "completed",
  71         [LS_FAILED]             = "failed",
  72         [LS_STOPPED]            = "stopped",
  73         [LS_PAUSED]             = "paused",
  74         [LS_CRASHED]            = "crashed",
  75         [LS_PARTIAL]            = "partial",
  76         [LS_CO_FAILED]          = "co-failed",
  77         [LS_CO_STOPPED]         = "co-stopped",
  78         [LS_CO_PAUSED]          = "co-paused"
  79 };
  80
  81 const char *lfsck_flags_names[] = {
  82         "scanned-once",
  83         "inconsistent",
  84         "upgrade",
  85         "incomplete",
  86         "crashed_lastid",
  87         NULL
  88 };
  89
  90 const char *lfsck_param_names[] = {
  91         NULL,
  92         "failout",
  93         "dryrun",
  94         "all_targets",
  95         "broadcast",
  96         "orphan",
  97         "create_ostobj",
  98         NULL
  99 };
 100
 101 const char *lfsck_status2names(enum lfsck_status status)
 102 {
 103         if (unlikely(status < 0 || status >= LS_MAX))
 104                 return "unknown";
 105
 106         return lfsck_status_names[status];
 107 }
 108
 109 static int lfsck_tgt_descs_init(struct lfsck_tgt_descs *ltds)
 110 {
 111         spin_lock_init(&ltds->ltd_lock);
 112         init_rwsem(&ltds->ltd_rw_sem);
 113         INIT_LIST_HEAD(&ltds->ltd_orphan);
 114         ltds->ltd_tgts_bitmap = CFS_ALLOCATE_BITMAP(BITS_PER_LONG);
 115         if (ltds->ltd_tgts_bitmap == NULL)
 116                 return -ENOMEM;
 117
 118         return 0;
 119 }
 120
 121 static void lfsck_tgt_descs_fini(struct lfsck_tgt_descs *ltds)
 122 {
 123         struct lfsck_tgt_desc   *ltd;
 124         struct lfsck_tgt_desc   *next;
 125         int                      idx;
 126
 127         down_write(&ltds->ltd_rw_sem);
 128
 129         list_for_each_entry_safe(ltd, next, &ltds->ltd_orphan,
 130                                  ltd_orphan_list) {
 131                 list_del_init(&ltd->ltd_orphan_list);
 132                 lfsck_tgt_put(ltd);
 133         }
 134
 135         if (unlikely(ltds->ltd_tgts_bitmap == NULL)) {
 136                 up_write(&ltds->ltd_rw_sem);
 137
 138                 return;
 139         }
 140
 141         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
 142                 ltd = LTD_TGT(ltds, idx);
 143                 if (likely(ltd != NULL)) {
 144                         LASSERT(list_empty(&ltd->ltd_layout_list));
 145                         LASSERT(list_empty(&ltd->ltd_layout_phase_list));
 146
 147                         ltds->ltd_tgtnr--;
 148                         cfs_bitmap_clear(ltds->ltd_tgts_bitmap, idx);
 149                         LTD_TGT(ltds, idx) = NULL;
 150                         lfsck_tgt_put(ltd);
 151                 }
 152         }
 153
 154         LASSERTF(ltds->ltd_tgtnr == 0, "tgt count unmatched: %d\n",
 155                  ltds->ltd_tgtnr);
 156
 157         for (idx = 0; idx < TGT_PTRS; idx++) {
 158                 if (ltds->ltd_tgts_idx[idx] != NULL) {
 159                         OBD_FREE_PTR(ltds->ltd_tgts_idx[idx]);
 160                         ltds->ltd_tgts_idx[idx] = NULL;
 161                 }
 162         }
 163
 164         CFS_FREE_BITMAP(ltds->ltd_tgts_bitmap);
 165         ltds->ltd_tgts_bitmap = NULL;
 166         up_write(&ltds->ltd_rw_sem);
 167 }
 168
 169 static int __lfsck_add_target(const struct lu_env *env,
 170                               struct lfsck_instance *lfsck,
 171                               struct lfsck_tgt_desc *ltd,
 172                               bool for_ost, bool locked)
 173 {
 174         struct lfsck_tgt_descs *ltds;
 175         __u32                   index = ltd->ltd_index;
 176         int                     rc    = 0;
 177         ENTRY;
 178
 179         if (for_ost)
 180                 ltds = &lfsck->li_ost_descs;
 181         else
 182                 ltds = &lfsck->li_mdt_descs;
 183
 184         if (!locked)
 185                 down_write(&ltds->ltd_rw_sem);
 186
 187         LASSERT(ltds->ltd_tgts_bitmap != NULL);
 188
 189         if (index >= ltds->ltd_tgts_bitmap->size) {
 190                 __u32 newsize = max((__u32)ltds->ltd_tgts_bitmap->size,
 191                                     (__u32)BITS_PER_LONG);
 192                 cfs_bitmap_t *old_bitmap = ltds->ltd_tgts_bitmap;
 193                 cfs_bitmap_t *new_bitmap;
 194
 195                 while (newsize < index + 1)
 196                         newsize <<= 1;
 197
 198                 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
 199                 if (new_bitmap == NULL)
 200                         GOTO(unlock, rc = -ENOMEM);
 201
 202                 if (ltds->ltd_tgtnr > 0)
 203                         cfs_bitmap_copy(new_bitmap, old_bitmap);
 204                 ltds->ltd_tgts_bitmap = new_bitmap;
 205                 CFS_FREE_BITMAP(old_bitmap);
 206         }
 207
 208         if (cfs_bitmap_check(ltds->ltd_tgts_bitmap, index)) {
 209                 CERROR("%s: the device %s (%u) is registered already\n",
 210                        lfsck_lfsck2name(lfsck),
 211                        ltd->ltd_tgt->dd_lu_dev.ld_obd->obd_name, index);
 212                 GOTO(unlock, rc = -EEXIST);
 213         }
 214
 215         if (ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 216                 OBD_ALLOC_PTR(ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK]);
 217                 if (ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK] == NULL)
 218                         GOTO(unlock, rc = -ENOMEM);
 219         }
 220
 221         LTD_TGT(ltds, index) = ltd;
 222         cfs_bitmap_set(ltds->ltd_tgts_bitmap, index);
 223         ltds->ltd_tgtnr++;
 224
 225         GOTO(unlock, rc = 0);
 226
 227 unlock:
 228         if (!locked)
 229                 up_write(&ltds->ltd_rw_sem);
 230
 231         return rc;
 232 }
 233
 234 static int lfsck_add_target_from_orphan(const struct lu_env *env,
 235                                         struct lfsck_instance *lfsck)
 236 {
 237         struct lfsck_tgt_descs  *ltds    = &lfsck->li_ost_descs;
 238         struct lfsck_tgt_desc   *ltd;
 239         struct lfsck_tgt_desc   *next;
 240         struct list_head        *head    = &lfsck_ost_orphan_list;
 241         int                      rc;
 242         bool                     for_ost = true;
 243
 244 again:
 245         spin_lock(&lfsck_instance_lock);
 246         list_for_each_entry_safe(ltd, next, head, ltd_orphan_list) {
 247                 if (ltd->ltd_key == lfsck->li_bottom) {
 248                         list_del_init(&ltd->ltd_orphan_list);
 249                         list_add_tail(&ltd->ltd_orphan_list,
 250                                       &ltds->ltd_orphan);
 251                 }
 252         }
 253         spin_unlock(&lfsck_instance_lock);
 254
 255         down_write(&ltds->ltd_rw_sem);
 256         while (!list_empty(&ltds->ltd_orphan)) {
 257                 ltd = list_entry(ltds->ltd_orphan.next,
 258                                  struct lfsck_tgt_desc,
 259                                  ltd_orphan_list);
 260                 list_del_init(&ltd->ltd_orphan_list);
 261                 rc = __lfsck_add_target(env, lfsck, ltd, for_ost, true);
 262                 /* Do not hold the semaphore for too long time. */
 263                 up_write(&ltds->ltd_rw_sem);
 264                 if (rc != 0)
 265                         return rc;
 266
 267                 down_write(&ltds->ltd_rw_sem);
 268         }
 269         up_write(&ltds->ltd_rw_sem);
 270
 271         if (for_ost) {
 272                 ltds = &lfsck->li_mdt_descs;
 273                 head = &lfsck_mdt_orphan_list;
 274                 for_ost = false;
 275                 goto again;
 276         }
 277
 278         return 0;
 279 }
 280
 281 static inline struct lfsck_component *
 282 __lfsck_component_find(struct lfsck_instance *lfsck, __u16 type, cfs_list_t *list)
 283 {
 284         struct lfsck_component *com;
 285
 286         cfs_list_for_each_entry(com, list, lc_link) {
 287                 if (com->lc_type == type)
 288                         return com;
 289         }
 290         return NULL;
 291 }
 292
 293 struct lfsck_component *
 294 lfsck_component_find(struct lfsck_instance *lfsck, __u16 type)
 295 {
 296         struct lfsck_component *com;
 297
 298         spin_lock(&lfsck->li_lock);
 299         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan);
 300         if (com != NULL)
 301                 goto unlock;
 302
 303         com = __lfsck_component_find(lfsck, type,
 304                                      &lfsck->li_list_double_scan);
 305         if (com != NULL)
 306                 goto unlock;
 307
 308         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_idle);
 309
 310 unlock:
 311         if (com != NULL)
 312                 lfsck_component_get(com);
 313         spin_unlock(&lfsck->li_lock);
 314         return com;
 315 }
 316
 317 void lfsck_component_cleanup(const struct lu_env *env,
 318                              struct lfsck_component *com)
 319 {
 320         if (!cfs_list_empty(&com->lc_link))
 321                 cfs_list_del_init(&com->lc_link);
 322         if (!cfs_list_empty(&com->lc_link_dir))
 323                 cfs_list_del_init(&com->lc_link_dir);
 324
 325         lfsck_component_put(env, com);
 326 }
 327
 328 int lfsck_fid_alloc(const struct lu_env *env, struct lfsck_instance *lfsck,
 329                     struct lu_fid *fid, bool locked)
 330 {
 331         struct lfsck_bookmark   *bk = &lfsck->li_bookmark_ram;
 332         int                      rc = 0;
 333         ENTRY;
 334
 335         if (!locked)
 336                 mutex_lock(&lfsck->li_mutex);
 337
 338         rc = seq_client_alloc_fid(env, lfsck->li_seq, fid);
 339         if (rc >= 0) {
 340                 bk->lb_last_fid = *fid;
 341                 /* We do not care about whether the subsequent sub-operations
 342                  * failed or not. The worst case is that one FID is lost that
 343                  * is not a big issue for the LFSCK since it is relative rare
 344                  * for LFSCK create. */
 345                 rc = lfsck_bookmark_store(env, lfsck);
 346         }
 347
 348         if (!locked)
 349                 mutex_unlock(&lfsck->li_mutex);
 350
 351         RETURN(rc);
 352 }
 353
 354 static const char dot[] = ".";
 355 static const char dotdot[] = "..";
 356
 357 static int lfsck_create_lpf_local(const struct lu_env *env,
 358                                   struct lfsck_instance *lfsck,
 359                                   struct dt_object *parent,
 360                                   struct dt_object *child,
 361                                   struct lu_attr *la,
 362                                   struct dt_object_format *dof,
 363                                   const char *name)
 364 {
 365         struct dt_device        *dev    = lfsck->li_bottom;
 366         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
 367         struct dt_object        *bk_obj = lfsck->li_bookmark_obj;
 368         const struct lu_fid     *cfid   = lu_object_fid(&child->do_lu);
 369         struct thandle          *th     = NULL;
 370         loff_t                   pos    = 0;
 371         int                      len    = sizeof(struct lfsck_bookmark);
 372         int                      rc     = 0;
 373         ENTRY;
 374
 375         th = dt_trans_create(env, dev);
 376         if (IS_ERR(th))
 377                 RETURN(PTR_ERR(th));
 378
 379         /* 1a. create child */
 380         rc = dt_declare_create(env, child, la, NULL, dof, th);
 381         if (rc != 0)
 382                 GOTO(stop, rc);
 383
 384         /* 2a. increase child nlink */
 385         rc = dt_declare_ref_add(env, child, th);
 386         if (rc != 0)
 387                 GOTO(stop, rc);
 388
 389         /* 3a. insert name into parent dir */
 390         rc = dt_declare_insert(env, parent, (const struct dt_rec *)cfid,
 391                                (const struct dt_key *)name, th);
 392         if (rc != 0)
 393                 GOTO(stop, rc);
 394
 395         /* 4a. increase parent nlink */
 396         rc = dt_declare_ref_add(env, parent, th);
 397         if (rc != 0)
 398                 GOTO(stop, rc);
 399
 400         /* 5a. update bookmark */
 401         rc = dt_declare_record_write(env, bk_obj,
 402                                      lfsck_buf_get(env, bk, len), 0, th);
 403         if (rc != 0)
 404                 GOTO(stop, rc);
 405
 406         rc = dt_trans_start_local(env, dev, th);
 407         if (rc != 0)
 408                 GOTO(stop, rc);
 409
 410         dt_write_lock(env, child, 0);
 411         /* 1b.1 create child */
 412         rc = dt_create(env, child, la, NULL, dof, th);
 413         if (rc != 0)
 414                 GOTO(unlock, rc);
 415
 416         if (unlikely(!dt_try_as_dir(env, child)))
 417                 GOTO(unlock, rc = -ENOTDIR);
 418
 419         /* 1b.2 insert dot into child dir */
 420         rc = dt_insert(env, child, (const struct dt_rec *)cfid,
 421                        (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
 422         if (rc != 0)
 423                 GOTO(unlock, rc);
 424
 425         /* 1b.3 insert dotdot into child dir */
 426         rc = dt_insert(env, child, (const struct dt_rec *)&LU_LPF_FID,
 427                        (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1);
 428         if (rc != 0)
 429                 GOTO(unlock, rc);
 430
 431         /* 2b. increase child nlink */
 432         rc = dt_ref_add(env, child, th);
 433         dt_write_unlock(env, child);
 434         if (rc != 0)
 435                 GOTO(stop, rc);
 436
 437         /* 3b. insert name into parent dir */
 438         rc = dt_insert(env, parent, (const struct dt_rec *)cfid,
 439                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
 440         if (rc != 0)
 441                 GOTO(stop, rc);
 442
 443         dt_write_lock(env, parent, 0);
 444         /* 4b. increase parent nlink */
 445         rc = dt_ref_add(env, parent, th);
 446         dt_write_unlock(env, parent);
 447         if (rc != 0)
 448                 GOTO(stop, rc);
 449
 450         bk->lb_lpf_fid = *cfid;
 451         lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
 452
 453         /* 5b. update bookmark */
 454         rc = dt_record_write(env, bk_obj,
 455                              lfsck_buf_get(env, bk, len), &pos, th);
 456
 457         GOTO(stop, rc);
 458
 459 unlock:
 460         dt_write_unlock(env, child);
 461
 462 stop:
 463         dt_trans_stop(env, dev, th);
 464
 465         return rc;
 466 }
 467
 468 static int lfsck_create_lpf_remote(const struct lu_env *env,
 469                                    struct lfsck_instance *lfsck,
 470                                    struct dt_object *parent,
 471                                    struct dt_object *child,
 472                                    struct lu_attr *la,
 473                                    struct dt_object_format *dof,
 474                                    const char *name)
 475 {
 476         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
 477         struct dt_object        *bk_obj = lfsck->li_bookmark_obj;
 478         const struct lu_fid     *cfid   = lu_object_fid(&child->do_lu);
 479         struct thandle          *th     = NULL;
 480         struct dt_device        *dev;
 481         loff_t                   pos    = 0;
 482         int                      len    = sizeof(struct lfsck_bookmark);
 483         int                      rc     = 0;
 484         ENTRY;
 485
 486         /* Create .lustre/lost+found/MDTxxxx. */
 487
 488         /* XXX: Currently, cross-MDT create operation needs to create the child
 489          *      object firstly, then insert name into the parent directory. For
 490          *      this case, the child object resides on current MDT (local), but
 491          *      the parent ".lustre/lost+found" may be on remote MDT. It is not
 492          *      easy to contain all the sub-modifications orderly within single
 493          *      transaction.
 494          *
 495          *      To avoid more inconsistency, we split the create operation into
 496          *      two transactions:
 497          *
 498          *      1) create the child locally.
 499          *      2) insert the name "MDTXXXX" in the parent ".lustre/lost+found"
 500          *         remotely and update the lfsck_bookmark::lb_lpf_fid locally.
 501          *
 502          *      If 1) done but 2) failed, then the worst case is that we lose
 503          *      one object locally, which is not a big issue. (can be repaird
 504          *      by LFSCK phase III) */
 505
 506         /* Transaction I: */
 507
 508         dev = lfsck->li_bottom;
 509         th = dt_trans_create(env, dev);
 510         if (IS_ERR(th))
 511                 RETURN(PTR_ERR(th));
 512
 513         /* 1a. create child locally. */
 514         rc = dt_declare_create(env, child, la, NULL, dof, th);
 515         if (rc != 0)
 516                 GOTO(stop, rc);
 517
 518         /* 2a. increase child nlink locally. */
 519         rc = dt_declare_ref_add(env, child, th);
 520         if (rc != 0)
 521                 GOTO(stop, rc);
 522
 523         rc = dt_trans_start_local(env, dev, th);
 524         if (rc != 0)
 525                 GOTO(stop, rc);
 526
 527         dt_write_lock(env, child, 0);
 528         /* 1b. create child locally. */
 529         rc = dt_create(env, child, la, NULL, dof, th);
 530         if (rc != 0)
 531                 GOTO(unlock, rc);
 532
 533         if (unlikely(!dt_try_as_dir(env, child)))
 534                 GOTO(unlock, rc = -ENOTDIR);
 535
 536         /* 2b.1 insert dot into child dir locally. */
 537         rc = dt_insert(env, child, (const struct dt_rec *)cfid,
 538                        (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
 539         if (rc != 0)
 540                 GOTO(unlock, rc);
 541
 542         /* 2b.2 insert dotdot into child dir locally. */
 543         rc = dt_insert(env, child, (const struct dt_rec *)&LU_LPF_FID,
 544                        (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1);
 545         if (rc != 0)
 546                 GOTO(unlock, rc);
 547
 548         /* 2b.3 increase child nlink locally. */
 549         rc = dt_ref_add(env, child, th);
 550         dt_write_unlock(env, child);
 551         dt_trans_stop(env, dev, th);
 552         if (rc != 0)
 553                 RETURN(rc);
 554
 555         /* Transaction II: */
 556
 557         dev = lfsck->li_next;
 558         th = dt_trans_create(env, dev);
 559         if (IS_ERR(th))
 560                 RETURN(PTR_ERR(th));
 561
 562         /* 3a. insert name into parent dir remotely. */
 563         rc = dt_declare_insert(env, parent, (const struct dt_rec *)cfid,
 564                                (const struct dt_key *)name, th);
 565         if (rc != 0)
 566                 GOTO(stop, rc);
 567
 568         /* 4a. increase parent nlink remotely. */
 569         rc = dt_declare_ref_add(env, parent, th);
 570         if (rc != 0)
 571                 GOTO(stop, rc);
 572
 573         /* 5a. decrease child nlink for dotdot locally if former remote
 574          *     update failed. */
 575         rc = dt_declare_ref_del(env, child, th);
 576         if (rc != 0)
 577                 GOTO(stop, rc);
 578
 579         /* 6a. decrease child nlink for dot locally if former remote
 580          *     update failed. */
 581         rc = dt_declare_ref_del(env, child, th);
 582         if (rc != 0)
 583                 GOTO(stop, rc);
 584
 585         /* 7a. destroy child locally if former remote update failed. */
 586         rc = dt_declare_destroy(env, child, th);
 587         if (rc != 0)
 588                 GOTO(stop, rc);
 589
 590         /* 8a. update bookmark locally. */
 591         rc = dt_declare_record_write(env, bk_obj,
 592                                      lfsck_buf_get(env, bk, len), 0, th);
 593         if (rc != 0)
 594                 GOTO(stop, rc);
 595
 596         rc = dt_trans_start(env, dev, th);
 597         if (rc != 0)
 598                 GOTO(stop, rc);
 599
 600         /* 3b. insert name into parent dir remotely. */
 601         rc = dt_insert(env, parent, (const struct dt_rec *)cfid,
 602                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
 603         if (rc == 0) {
 604                 dt_write_lock(env, parent, 0);
 605                 /* 4b. increase parent nlink remotely. */
 606                 rc = dt_ref_add(env, parent, th);
 607                 dt_write_unlock(env, parent);
 608         }
 609         if (rc != 0) {
 610                 /* 5b. decrease child nlink for dotdot locally. */
 611                 dt_ref_del(env, child, th);
 612                 /* 6b. decrease child nlink for dot locally. */
 613                 dt_ref_del(env, child, th);
 614                 /* 7b. destroy child locally. */
 615                 dt_destroy(env, child, th);
 616                 GOTO(stop, rc);
 617         }
 618
 619         bk->lb_lpf_fid = *cfid;
 620         lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
 621
 622         /* 8b. update bookmark locally. */
 623         rc = dt_record_write(env, bk_obj,
 624                              lfsck_buf_get(env, bk, len), &pos, th);
 625
 626         GOTO(stop, rc);
 627
 628 unlock:
 629         dt_write_unlock(env, child);
 630 stop:
 631         dt_trans_stop(env, dev, th);
 632
 633         return rc;
 634 }
 635
 636 /* Do NOT create .lustre/lost+found/MDTxxxx when register the lfsck instance,
 637  * because the MDT0 maybe not reaady for sequence allocation yet. We do that
 638  * only when it is required, such as orphan OST-objects repairing. */
 639 int lfsck_create_lpf(const struct lu_env *env, struct lfsck_instance *lfsck)
 640 {
 641         struct lfsck_bookmark    *bk    = &lfsck->li_bookmark_ram;
 642         struct lfsck_thread_info *info  = lfsck_env_info(env);
 643         struct lu_fid            *cfid  = &info->lti_fid2;
 644         struct lu_attr           *la    = &info->lti_la;
 645         struct dt_object_format  *dof   = &info->lti_dof;
 646         struct dt_object         *parent = NULL;
 647         struct dt_object         *child = NULL;
 648         char                      name[8];
 649         int                       node  = lfsck_dev_idx(lfsck->li_bottom);
 650         int                       rc    = 0;
 651         ENTRY;
 652
 653         LASSERT(lfsck->li_master);
 654
 655         sprintf(name, "MDT%04x", node);
 656         if (node == 0) {
 657                 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
 658                                                   &LU_LPF_FID);
 659         } else {
 660                 struct lfsck_tgt_desc *ltd;
 661
 662                 ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, 0);
 663                 if (unlikely(ltd == NULL))
 664                         RETURN(-ENXIO);
 665
 666                 parent = lfsck_object_find_by_dev(env, ltd->ltd_tgt,
 667                                                   &LU_LPF_FID);
 668                 lfsck_tgt_put(ltd);
 669         }
 670         if (IS_ERR(parent))
 671                 RETURN(PTR_ERR(parent));
 672
 673         if (unlikely(!dt_try_as_dir(env, parent)))
 674                 GOTO(out, rc = -ENOTDIR);
 675
 676         mutex_lock(&lfsck->li_mutex);
 677         if (lfsck->li_lpf_obj != NULL)
 678                 GOTO(unlock, rc = 0);
 679
 680         if (fid_is_zero(&bk->lb_lpf_fid)) {
 681                 /* There is corner case that: in former LFSCK scanning we have
 682                  * created the .lustre/lost+found/MDTxxxx but failed to update
 683                  * the lfsck_bookmark::lb_lpf_fid successfully. So need lookup
 684                  * it from MDT0 firstly. */
 685                 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
 686                                (const struct dt_key *)name, BYPASS_CAPA);
 687                 if (rc != 0 && rc != -ENOENT)
 688                         GOTO(unlock, rc);
 689
 690                 if (rc == 0) {
 691                         bk->lb_lpf_fid = *cfid;
 692                         rc = lfsck_bookmark_store(env, lfsck);
 693                 } else {
 694                         rc = lfsck_fid_alloc(env, lfsck, cfid, true);
 695                 }
 696                 if (rc != 0)
 697                         GOTO(unlock, rc);
 698         } else {
 699                 *cfid = bk->lb_lpf_fid;
 700         }
 701
 702         child = lfsck_object_find_by_dev(env, lfsck->li_bottom, cfid);
 703         if (IS_ERR(child))
 704                 GOTO(unlock, rc = PTR_ERR(child));
 705
 706         if (dt_object_exists(child) != 0) {
 707                 if (unlikely(!dt_try_as_dir(env, child)))
 708                         GOTO(unlock, rc = -ENOTDIR);
 709
 710                 lfsck->li_lpf_obj = child;
 711                 GOTO(unlock, rc = 0);
 712         }
 713
 714         memset(la, 0, sizeof(*la));
 715         la->la_atime = la->la_mtime = la->la_ctime = cfs_time_current_sec();
 716         la->la_mode = S_IFDIR | S_IRWXU;
 717         la->la_valid = LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
 718                        LA_UID | LA_GID;
 719         memset(dof, 0, sizeof(*dof));
 720         dof->dof_type = dt_mode_to_dft(S_IFDIR);
 721
 722         if (node == 0)
 723                 rc = lfsck_create_lpf_local(env, lfsck, parent, child, la,
 724                                             dof, name);
 725         else
 726                 rc = lfsck_create_lpf_remote(env, lfsck, parent, child, la,
 727                                              dof, name);
 728         if (rc == 0)
 729                 lfsck->li_lpf_obj = child;
 730
 731         GOTO(unlock, rc);
 732
 733 unlock:
 734         mutex_unlock(&lfsck->li_mutex);
 735         if (rc != 0 && child != NULL && !IS_ERR(child))
 736                 lu_object_put(env, &child->do_lu);
 737 out:
 738         if (parent != NULL && !IS_ERR(parent))
 739                 lu_object_put(env, &parent->do_lu);
 740
 741         return rc;
 742 }
 743
 744 static int lfsck_fid_init(struct lfsck_instance *lfsck)
 745 {
 746         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
 747         struct seq_server_site  *ss;
 748         char                    *prefix;
 749         int                      rc     = 0;
 750         ENTRY;
 751
 752         ss = lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
 753         if (unlikely(ss == NULL))
 754                 RETURN(-ENXIO);
 755
 756         OBD_ALLOC_PTR(lfsck->li_seq);
 757         if (lfsck->li_seq == NULL)
 758                 RETURN(-ENOMEM);
 759
 760         OBD_ALLOC(prefix, MAX_OBD_NAME + 7);
 761         if (prefix == NULL)
 762                 GOTO(out, rc = -ENOMEM);
 763
 764         snprintf(prefix, MAX_OBD_NAME + 7, "lfsck-%s", lfsck_lfsck2name(lfsck));
 765         rc = seq_client_init(lfsck->li_seq, NULL, LUSTRE_SEQ_METADATA, prefix,
 766                              ss->ss_server_seq);
 767         OBD_FREE(prefix, MAX_OBD_NAME + 7);
 768         if (rc != 0)
 769                 GOTO(out, rc);
 770
 771         if (fid_is_sane(&bk->lb_last_fid))
 772                 lfsck->li_seq->lcs_fid = bk->lb_last_fid;
 773
 774         RETURN(0);
 775
 776 out:
 777         OBD_FREE_PTR(lfsck->li_seq);
 778         lfsck->li_seq = NULL;
 779
 780         return rc;
 781 }
 782
 783 static void lfsck_fid_fini(struct lfsck_instance *lfsck)
 784 {
 785         if (lfsck->li_seq != NULL) {
 786                 seq_client_fini(lfsck->li_seq);
 787                 OBD_FREE_PTR(lfsck->li_seq);
 788                 lfsck->li_seq = NULL;
 789         }
 790 }
 791
 792 void lfsck_instance_cleanup(const struct lu_env *env,
 793                             struct lfsck_instance *lfsck)
 794 {
 795         struct ptlrpc_thread    *thread = &lfsck->li_thread;
 796         struct lfsck_component  *com;
 797         ENTRY;
 798
 799         LASSERT(list_empty(&lfsck->li_link));
 800         LASSERT(thread_is_init(thread) || thread_is_stopped(thread));
 801
 802         if (lfsck->li_obj_oit != NULL) {
 803                 lu_object_put_nocache(env, &lfsck->li_obj_oit->do_lu);
 804                 lfsck->li_obj_oit = NULL;
 805         }
 806
 807         LASSERT(lfsck->li_obj_dir == NULL);
 808
 809         while (!cfs_list_empty(&lfsck->li_list_scan)) {
 810                 com = cfs_list_entry(lfsck->li_list_scan.next,
 811                                      struct lfsck_component,
 812                                      lc_link);
 813                 lfsck_component_cleanup(env, com);
 814         }
 815
 816         LASSERT(cfs_list_empty(&lfsck->li_list_dir));
 817
 818         while (!cfs_list_empty(&lfsck->li_list_double_scan)) {
 819                 com = cfs_list_entry(lfsck->li_list_double_scan.next,
 820                                      struct lfsck_component,
 821                                      lc_link);
 822                 lfsck_component_cleanup(env, com);
 823         }
 824
 825         while (!cfs_list_empty(&lfsck->li_list_idle)) {
 826                 com = cfs_list_entry(lfsck->li_list_idle.next,
 827                                      struct lfsck_component,
 828                                      lc_link);
 829                 lfsck_component_cleanup(env, com);
 830         }
 831
 832         lfsck_tgt_descs_fini(&lfsck->li_ost_descs);
 833         lfsck_tgt_descs_fini(&lfsck->li_mdt_descs);
 834
 835         if (lfsck->li_bookmark_obj != NULL) {
 836                 lu_object_put_nocache(env, &lfsck->li_bookmark_obj->do_lu);
 837                 lfsck->li_bookmark_obj = NULL;
 838         }
 839
 840         if (lfsck->li_lpf_obj != NULL) {
 841                 lu_object_put(env, &lfsck->li_lpf_obj->do_lu);
 842                 lfsck->li_lpf_obj = NULL;
 843         }
 844
 845         if (lfsck->li_los != NULL) {
 846                 local_oid_storage_fini(env, lfsck->li_los);
 847                 lfsck->li_los = NULL;
 848         }
 849
 850         lfsck_fid_fini(lfsck);
 851
 852         OBD_FREE_PTR(lfsck);
 853 }
 854
 855 static inline struct lfsck_instance *
 856 __lfsck_instance_find(struct dt_device *key, bool ref, bool unlink)
 857 {
 858         struct lfsck_instance *lfsck;
 859
 860         cfs_list_for_each_entry(lfsck, &lfsck_instance_list, li_link) {
 861                 if (lfsck->li_bottom == key) {
 862                         if (ref)
 863                                 lfsck_instance_get(lfsck);
 864                         if (unlink)
 865                                 list_del_init(&lfsck->li_link);
 866
 867                         return lfsck;
 868                 }
 869         }
 870
 871         return NULL;
 872 }
 873
 874 struct lfsck_instance *lfsck_instance_find(struct dt_device *key, bool ref,
 875                                            bool unlink)
 876 {
 877         struct lfsck_instance *lfsck;
 878
 879         spin_lock(&lfsck_instance_lock);
 880         lfsck = __lfsck_instance_find(key, ref, unlink);
 881         spin_unlock(&lfsck_instance_lock);
 882
 883         return lfsck;
 884 }
 885
 886 static inline int lfsck_instance_add(struct lfsck_instance *lfsck)
 887 {
 888         struct lfsck_instance *tmp;
 889
 890         spin_lock(&lfsck_instance_lock);
 891         cfs_list_for_each_entry(tmp, &lfsck_instance_list, li_link) {
 892                 if (lfsck->li_bottom == tmp->li_bottom) {
 893                         spin_unlock(&lfsck_instance_lock);
 894                         return -EEXIST;
 895                 }
 896         }
 897
 898         cfs_list_add_tail(&lfsck->li_link, &lfsck_instance_list);
 899         spin_unlock(&lfsck_instance_lock);
 900         return 0;
 901 }
 902
 903 int lfsck_bits_dump(struct seq_file *m, int bits, const char *names[],
 904                     const char *prefix)
 905 {
 906         int flag;
 907         int i;
 908         bool newline = (bits != 0 ? false : true);
 909
 910         seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
 911
 912         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
 913                 if (flag & bits) {
 914                         bits &= ~flag;
 915                         if (names[i] != NULL) {
 916                                 if (bits == 0)
 917                                         newline = true;
 918
 919                                 seq_printf(m, "%s%c", names[i],
 920                                            newline ? '\n' : ',');
 921                         }
 922                 }
 923         }
 924
 925         if (!newline)
 926                 seq_printf(m, "\n");
 927         return 0;
 928 }
 929
 930 int lfsck_time_dump(struct seq_file *m, __u64 time, const char *prefix)
 931 {
 932         if (time != 0)
 933                 seq_printf(m, "%s: "LPU64" seconds\n", prefix,
 934                           cfs_time_current_sec() - time);
 935         else
 936                 seq_printf(m, "%s: N/A\n", prefix);
 937         return 0;
 938 }
 939
 940 int lfsck_pos_dump(struct seq_file *m, struct lfsck_position *pos,
 941                    const char *prefix)
 942 {
 943         if (fid_is_zero(&pos->lp_dir_parent)) {
 944                 if (pos->lp_oit_cookie == 0)
 945                         seq_printf(m, "%s: N/A, N/A, N/A\n",
 946                                    prefix);
 947                 else
 948                         seq_printf(m, "%s: "LPU64", N/A, N/A\n",
 949                                    prefix, pos->lp_oit_cookie);
 950         } else {
 951                 seq_printf(m, "%s: "LPU64", "DFID", "LPU64"\n",
 952                            prefix, pos->lp_oit_cookie,
 953                            PFID(&pos->lp_dir_parent), pos->lp_dir_cookie);
 954         }
 955         return 0;
 956 }
 957
 958 void lfsck_pos_fill(const struct lu_env *env, struct lfsck_instance *lfsck,
 959                     struct lfsck_position *pos, bool init)
 960 {
 961         const struct dt_it_ops *iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
 962
 963         if (unlikely(lfsck->li_di_oit == NULL)) {
 964                 memset(pos, 0, sizeof(*pos));
 965                 return;
 966         }
 967
 968         pos->lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
 969         if (!lfsck->li_current_oit_processed && !init)
 970                 pos->lp_oit_cookie--;
 971
 972         LASSERT(pos->lp_oit_cookie > 0);
 973
 974         if (lfsck->li_di_dir != NULL) {
 975                 struct dt_object *dto = lfsck->li_obj_dir;
 976
 977                 pos->lp_dir_cookie = dto->do_index_ops->dio_it.store(env,
 978                                                         lfsck->li_di_dir);
 979
 980                 if (pos->lp_dir_cookie >= MDS_DIR_END_OFF) {
 981                         fid_zero(&pos->lp_dir_parent);
 982                         pos->lp_dir_cookie = 0;
 983                 } else {
 984                         pos->lp_dir_parent = *lfsck_dto2fid(dto);
 985                 }
 986         } else {
 987                 fid_zero(&pos->lp_dir_parent);
 988                 pos->lp_dir_cookie = 0;
 989         }
 990 }
 991
 992 bool __lfsck_set_speed(struct lfsck_instance *lfsck, __u32 limit)
 993 {
 994         bool dirty = false;
 995
 996         if (limit != LFSCK_SPEED_NO_LIMIT) {
 997                 if (limit > HZ) {
 998                         lfsck->li_sleep_rate = limit / HZ;
 999                         lfsck->li_sleep_jif = 1;
1000                 } else {
1001                         lfsck->li_sleep_rate = 1;
1002                         lfsck->li_sleep_jif = HZ / limit;
1003                 }
1004         } else {
1005                 lfsck->li_sleep_jif = 0;
1006                 lfsck->li_sleep_rate = 0;
1007         }
1008
1009         if (lfsck->li_bookmark_ram.lb_speed_limit != limit) {
1010                 lfsck->li_bookmark_ram.lb_speed_limit = limit;
1011                 dirty = true;
1012         }
1013
1014         return dirty;
1015 }
1016
1017 void lfsck_control_speed(struct lfsck_instance *lfsck)
1018 {
1019         struct ptlrpc_thread *thread = &lfsck->li_thread;
1020         struct l_wait_info    lwi;
1021
1022         if (lfsck->li_sleep_jif > 0 &&
1023             lfsck->li_new_scanned >= lfsck->li_sleep_rate) {
1024                 lwi = LWI_TIMEOUT_INTR(lfsck->li_sleep_jif, NULL,
1025                                        LWI_ON_SIGNAL_NOOP, NULL);
1026
1027                 l_wait_event(thread->t_ctl_waitq,
1028                              !thread_is_running(thread),
1029                              &lwi);
1030                 lfsck->li_new_scanned = 0;
1031         }
1032 }
1033
1034 void lfsck_control_speed_by_self(struct lfsck_component *com)
1035 {
1036         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1037         struct ptlrpc_thread    *thread = &lfsck->li_thread;
1038         struct l_wait_info       lwi;
1039
1040         if (lfsck->li_sleep_jif > 0 &&
1041             com->lc_new_scanned >= lfsck->li_sleep_rate) {
1042                 lwi = LWI_TIMEOUT_INTR(lfsck->li_sleep_jif, NULL,
1043                                        LWI_ON_SIGNAL_NOOP, NULL);
1044
1045                 l_wait_event(thread->t_ctl_waitq,
1046                              !thread_is_running(thread),
1047                              &lwi);
1048                 com->lc_new_scanned = 0;
1049         }
1050 }
1051
1052 static int lfsck_parent_fid(const struct lu_env *env, struct dt_object *obj,
1053                             struct lu_fid *fid)
1054 {
1055         if (unlikely(!S_ISDIR(lfsck_object_type(obj)) ||
1056                      !dt_try_as_dir(env, obj)))
1057                 return -ENOTDIR;
1058
1059         return dt_lookup(env, obj, (struct dt_rec *)fid,
1060                          (const struct dt_key *)"..", BYPASS_CAPA);
1061 }
1062
1063 static int lfsck_needs_scan_dir(const struct lu_env *env,
1064                                 struct lfsck_instance *lfsck,
1065                                 struct dt_object *obj)
1066 {
1067         struct lu_fid *fid   = &lfsck_env_info(env)->lti_fid;
1068         int            depth = 0;
1069         int            rc;
1070
1071         if (!lfsck->li_master || !S_ISDIR(lfsck_object_type(obj)) ||
1072             cfs_list_empty(&lfsck->li_list_dir))
1073                RETURN(0);
1074
1075         while (1) {
1076                 /* XXX: Currently, we do not scan the "/REMOTE_PARENT_DIR",
1077                  *      which is the agent directory to manage the objects
1078                  *      which name entries reside on remote MDTs. Related
1079                  *      consistency verification will be processed in LFSCK
1080                  *      phase III. */
1081                 if (lu_fid_eq(lfsck_dto2fid(obj), &lfsck->li_global_root_fid)) {
1082                         if (depth > 0)
1083                                 lfsck_object_put(env, obj);
1084                         return 1;
1085                 }
1086
1087                 /* No need to check .lustre and its children. */
1088                 if (fid_seq_is_dot_lustre(fid_seq(lfsck_dto2fid(obj)))) {
1089                         if (depth > 0)
1090                                 lfsck_object_put(env, obj);
1091                         return 0;
1092                 }
1093
1094                 dt_read_lock(env, obj, MOR_TGT_CHILD);
1095                 if (unlikely(lfsck_is_dead_obj(obj))) {
1096                         dt_read_unlock(env, obj);
1097                         if (depth > 0)
1098                                 lfsck_object_put(env, obj);
1099                         return 0;
1100                 }
1101
1102                 rc = dt_xattr_get(env, obj,
1103                                   lfsck_buf_get(env, NULL, 0), XATTR_NAME_LINK,
1104                                   BYPASS_CAPA);
1105                 dt_read_unlock(env, obj);
1106                 if (rc >= 0) {
1107                         if (depth > 0)
1108                                 lfsck_object_put(env, obj);
1109                         return 1;
1110                 }
1111
1112                 if (rc < 0 && rc != -ENODATA) {
1113                         if (depth > 0)
1114                                 lfsck_object_put(env, obj);
1115                         return rc;
1116                 }
1117
1118                 rc = lfsck_parent_fid(env, obj, fid);
1119                 if (depth > 0)
1120                         lfsck_object_put(env, obj);
1121                 if (rc != 0)
1122                         return rc;
1123
1124                 if (unlikely(lu_fid_eq(fid, &lfsck->li_local_root_fid)))
1125                         return 0;
1126
1127                 obj = lfsck_object_find(env, lfsck, fid);
1128                 if (obj == NULL)
1129                         return 0;
1130                 else if (IS_ERR(obj))
1131                         return PTR_ERR(obj);
1132
1133                 if (!dt_object_exists(obj)) {
1134                         lfsck_object_put(env, obj);
1135                         return 0;
1136                 }
1137
1138                 if (dt_object_remote(obj)) {
1139                         /* .lustre/lost+found/MDTxxx can be remote directory. */
1140                         if (fid_seq_is_dot_lustre(fid_seq(lfsck_dto2fid(obj))))
1141                                 rc = 0;
1142                         else
1143                                 /* Other remote directory should be client
1144                                  * visible and need to be checked. */
1145                                 rc = 1;
1146                         lfsck_object_put(env, obj);
1147                         return rc;
1148                 }
1149
1150                 depth++;
1151         }
1152         return 0;
1153 }
1154
1155 struct lfsck_thread_args *lfsck_thread_args_init(struct lfsck_instance *lfsck,
1156                                                  struct lfsck_component *com,
1157                                                  struct lfsck_start_param *lsp)
1158 {
1159         struct lfsck_thread_args *lta;
1160         int                       rc;
1161
1162         OBD_ALLOC_PTR(lta);
1163         if (lta == NULL)
1164                 return ERR_PTR(-ENOMEM);
1165
1166         rc = lu_env_init(&lta->lta_env, LCT_MD_THREAD | LCT_DT_THREAD);
1167         if (rc != 0) {
1168                 OBD_FREE_PTR(lta);
1169                 return ERR_PTR(rc);
1170         }
1171
1172         lta->lta_lfsck = lfsck_instance_get(lfsck);
1173         if (com != NULL)
1174                 lta->lta_com = lfsck_component_get(com);
1175
1176         lta->lta_lsp = lsp;
1177
1178         return lta;
1179 }
1180
1181 void lfsck_thread_args_fini(struct lfsck_thread_args *lta)
1182 {
1183         if (lta->lta_com != NULL)
1184                 lfsck_component_put(&lta->lta_env, lta->lta_com);
1185         lfsck_instance_put(&lta->lta_env, lta->lta_lfsck);
1186         lu_env_fini(&lta->lta_env);
1187         OBD_FREE_PTR(lta);
1188 }
1189
1190 /* LFSCK wrap functions */
1191
1192 void lfsck_fail(const struct lu_env *env, struct lfsck_instance *lfsck,
1193                 bool new_checked)
1194 {
1195         struct lfsck_component *com;
1196
1197         cfs_list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
1198                 com->lc_ops->lfsck_fail(env, com, new_checked);
1199         }
1200 }
1201
1202 int lfsck_checkpoint(const struct lu_env *env, struct lfsck_instance *lfsck)
1203 {
1204         struct lfsck_component *com;
1205         int                     rc  = 0;
1206         int                     rc1 = 0;
1207
1208         if (likely(cfs_time_beforeq(cfs_time_current(),
1209                                     lfsck->li_time_next_checkpoint)))
1210                 return 0;
1211
1212         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_current, false);
1213         cfs_list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
1214                 rc = com->lc_ops->lfsck_checkpoint(env, com, false);
1215                 if (rc != 0)
1216                         rc1 = rc;
1217         }
1218
1219         lfsck->li_time_last_checkpoint = cfs_time_current();
1220         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
1221                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1222         return rc1 != 0 ? rc1 : rc;
1223 }
1224
1225 int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck,
1226                struct lfsck_start_param *lsp)
1227 {
1228         struct dt_object       *obj     = NULL;
1229         struct lfsck_component *com;
1230         struct lfsck_component *next;
1231         struct lfsck_position  *pos     = NULL;
1232         const struct dt_it_ops *iops    =
1233                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
1234         struct dt_it           *di;
1235         int                     rc;
1236         ENTRY;
1237
1238         LASSERT(lfsck->li_obj_dir == NULL);
1239         LASSERT(lfsck->li_di_dir == NULL);
1240
1241         lfsck->li_current_oit_processed = 0;
1242         cfs_list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
1243                 com->lc_new_checked = 0;
1244                 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1245                         com->lc_journal = 0;
1246
1247                 rc = com->lc_ops->lfsck_prep(env, com, lsp);
1248                 if (rc != 0)
1249                         GOTO(out, rc);
1250
1251                 if ((pos == NULL) ||
1252                     (!lfsck_pos_is_zero(&com->lc_pos_start) &&
1253                      lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0))
1254                         pos = &com->lc_pos_start;
1255         }
1256
1257         /* Init otable-based iterator. */
1258         if (pos == NULL) {
1259                 rc = iops->load(env, lfsck->li_di_oit, 0);
1260                 if (rc > 0) {
1261                         lfsck->li_oit_over = 1;
1262                         rc = 0;
1263                 }
1264
1265                 GOTO(out, rc);
1266         }
1267
1268         rc = iops->load(env, lfsck->li_di_oit, pos->lp_oit_cookie);
1269         if (rc < 0)
1270                 GOTO(out, rc);
1271         else if (rc > 0)
1272                 lfsck->li_oit_over = 1;
1273
1274         if (!lfsck->li_master || fid_is_zero(&pos->lp_dir_parent))
1275                 GOTO(out, rc = 0);
1276
1277         /* Find the directory for namespace-based traverse. */
1278         obj = lfsck_object_find(env, lfsck, &pos->lp_dir_parent);
1279         if (obj == NULL)
1280                 GOTO(out, rc = 0);
1281         else if (IS_ERR(obj))
1282                 RETURN(PTR_ERR(obj));
1283
1284         /* XXX: Currently, skip remote object, the consistency for
1285          *      remote object will be processed in LFSCK phase III. */
1286         if (!dt_object_exists(obj) || dt_object_remote(obj) ||
1287             unlikely(!S_ISDIR(lfsck_object_type(obj))))
1288                 GOTO(out, rc = 0);
1289
1290         if (unlikely(!dt_try_as_dir(env, obj)))
1291                 GOTO(out, rc = -ENOTDIR);
1292
1293         /* Init the namespace-based directory traverse. */
1294         iops = &obj->do_index_ops->dio_it;
1295         di = iops->init(env, obj, lfsck->li_args_dir, BYPASS_CAPA);
1296         if (IS_ERR(di))
1297                 GOTO(out, rc = PTR_ERR(di));
1298
1299         LASSERT(pos->lp_dir_cookie < MDS_DIR_END_OFF);
1300
1301         rc = iops->load(env, di, pos->lp_dir_cookie);
1302         if ((rc == 0) || (rc > 0 && pos->lp_dir_cookie > 0))
1303                 rc = iops->next(env, di);
1304         else if (rc > 0)
1305                 rc = 0;
1306
1307         if (rc != 0) {
1308                 iops->put(env, di);
1309                 iops->fini(env, di);
1310                 GOTO(out, rc);
1311         }
1312
1313         lfsck->li_obj_dir = lfsck_object_get(obj);
1314         lfsck->li_cookie_dir = iops->store(env, di);
1315         spin_lock(&lfsck->li_lock);
1316         lfsck->li_di_dir = di;
1317         spin_unlock(&lfsck->li_lock);
1318
1319         GOTO(out, rc = 0);
1320
1321 out:
1322         if (obj != NULL)
1323                 lfsck_object_put(env, obj);
1324
1325         if (rc < 0) {
1326                 cfs_list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
1327                                              lc_link)
1328                         com->lc_ops->lfsck_post(env, com, rc, true);
1329
1330                 return rc;
1331         }
1332
1333         rc = 0;
1334         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_current, true);
1335         cfs_list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
1336                 rc = com->lc_ops->lfsck_checkpoint(env, com, true);
1337                 if (rc != 0)
1338                         break;
1339         }
1340
1341         lfsck->li_time_last_checkpoint = cfs_time_current();
1342         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
1343                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1344         return rc;
1345 }
1346
1347 int lfsck_exec_oit(const struct lu_env *env, struct lfsck_instance *lfsck,
1348                    struct dt_object *obj)
1349 {
1350         struct lfsck_component *com;
1351         const struct dt_it_ops *iops;
1352         struct dt_it           *di;
1353         int                     rc;
1354         ENTRY;
1355
1356         LASSERT(lfsck->li_obj_dir == NULL);
1357
1358         cfs_list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
1359                 rc = com->lc_ops->lfsck_exec_oit(env, com, obj);
1360                 if (rc != 0)
1361                         RETURN(rc);
1362         }
1363
1364         rc = lfsck_needs_scan_dir(env, lfsck, obj);
1365         if (rc <= 0)
1366                 GOTO(out, rc);
1367
1368         if (unlikely(!dt_try_as_dir(env, obj)))
1369                 GOTO(out, rc = -ENOTDIR);
1370
1371         iops = &obj->do_index_ops->dio_it;
1372         di = iops->init(env, obj, lfsck->li_args_dir, BYPASS_CAPA);
1373         if (IS_ERR(di))
1374                 GOTO(out, rc = PTR_ERR(di));
1375
1376         rc = iops->load(env, di, 0);
1377         if (rc == 0)
1378                 rc = iops->next(env, di);
1379         else if (rc > 0)
1380                 rc = 0;
1381
1382         if (rc != 0) {
1383                 iops->put(env, di);
1384                 iops->fini(env, di);
1385                 GOTO(out, rc);
1386         }
1387
1388         lfsck->li_obj_dir = lfsck_object_get(obj);
1389         lfsck->li_cookie_dir = iops->store(env, di);
1390         spin_lock(&lfsck->li_lock);
1391         lfsck->li_di_dir = di;
1392         spin_unlock(&lfsck->li_lock);
1393
1394         GOTO(out, rc = 0);
1395
1396 out:
1397         if (rc < 0)
1398                 lfsck_fail(env, lfsck, false);
1399         return (rc > 0 ? 0 : rc);
1400 }
1401
1402 int lfsck_exec_dir(const struct lu_env *env, struct lfsck_instance *lfsck,
1403                    struct dt_object *obj, struct lu_dirent *ent)
1404 {
1405         struct lfsck_component *com;
1406         int                     rc;
1407
1408         cfs_list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
1409                 rc = com->lc_ops->lfsck_exec_dir(env, com, obj, ent);
1410                 if (rc != 0)
1411                         return rc;
1412         }
1413         return 0;
1414 }
1415
1416 int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck,
1417                int result)
1418 {
1419         struct lfsck_component *com;
1420         struct lfsck_component *next;
1421         int                     rc  = 0;
1422         int                     rc1 = 0;
1423
1424         lfsck_pos_fill(env, lfsck, &lfsck->li_pos_current, false);
1425         cfs_list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
1426                 rc = com->lc_ops->lfsck_post(env, com, result, false);
1427                 if (rc != 0)
1428                         rc1 = rc;
1429         }
1430
1431         lfsck->li_time_last_checkpoint = cfs_time_current();
1432         lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
1433                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1434
1435         /* Ignore some component post failure to make other can go ahead. */
1436         return result;
1437 }
1438
1439 static void lfsck_interpret(const struct lu_env *env,
1440                             struct lfsck_instance *lfsck,
1441                             struct ptlrpc_request *req, void *args, int result)
1442 {
1443         struct lfsck_async_interpret_args *laia = args;
1444         struct lfsck_component            *com;
1445
1446         LASSERT(laia->laia_com == NULL);
1447         LASSERT(laia->laia_shared);
1448
1449         spin_lock(&lfsck->li_lock);
1450         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
1451                 if (com->lc_ops->lfsck_interpret != NULL) {
1452                         laia->laia_com = com;
1453                         com->lc_ops->lfsck_interpret(env, req, laia, result);
1454                 }
1455         }
1456
1457         list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
1458                 if (com->lc_ops->lfsck_interpret != NULL) {
1459                         laia->laia_com = com;
1460                         com->lc_ops->lfsck_interpret(env, req, laia, result);
1461                 }
1462         }
1463         spin_unlock(&lfsck->li_lock);
1464 }
1465
1466 int lfsck_double_scan(const struct lu_env *env, struct lfsck_instance *lfsck)
1467 {
1468         struct lfsck_component *com;
1469         struct lfsck_component *next;
1470         struct l_wait_info      lwi = { 0 };
1471         int                     rc  = 0;
1472         int                     rc1 = 0;
1473
1474         list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
1475                 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1476                         com->lc_journal = 0;
1477
1478                 rc = com->lc_ops->lfsck_double_scan(env, com);
1479                 if (rc != 0)
1480                         rc1 = rc;
1481         }
1482
1483         l_wait_event(lfsck->li_thread.t_ctl_waitq,
1484                      atomic_read(&lfsck->li_double_scan_count) == 0,
1485                      &lwi);
1486
1487         if (lfsck->li_status != LS_PAUSED &&
1488             lfsck->li_status != LS_CO_PAUSED) {
1489                 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
1490                                          lc_link) {
1491                         spin_lock(&lfsck->li_lock);
1492                         list_del_init(&com->lc_link);
1493                         list_add_tail(&com->lc_link, &lfsck->li_list_idle);
1494                         spin_unlock(&lfsck->li_lock);
1495                 }
1496         }
1497
1498         return rc1 != 0 ? rc1 : rc;
1499 }
1500
1501 static int lfsck_stop_notify(const struct lu_env *env,
1502                              struct lfsck_instance *lfsck,
1503                              struct lfsck_tgt_descs *ltds,
1504                              struct lfsck_tgt_desc *ltd, __u16 type)
1505 {
1506         struct ptlrpc_request_set *set;
1507         struct lfsck_component    *com;
1508         int                        rc  = 0;
1509         ENTRY;
1510
1511         spin_lock(&lfsck->li_lock);
1512         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan);
1513         if (com == NULL)
1514                 com = __lfsck_component_find(lfsck, type,
1515                                              &lfsck->li_list_double_scan);
1516         if (com != NULL)
1517                 lfsck_component_get(com);
1518         spin_lock(&lfsck->li_lock);
1519
1520         if (com != NULL) {
1521                 if (com->lc_ops->lfsck_stop_notify != NULL) {
1522                         set = ptlrpc_prep_set();
1523                         if (set == NULL) {
1524                                 lfsck_component_put(env, com);
1525
1526                                 RETURN(-ENOMEM);
1527                         }
1528
1529                         rc = com->lc_ops->lfsck_stop_notify(env, com, ltds,
1530                                                             ltd, set);
1531                         if (rc == 0)
1532                                 rc = ptlrpc_set_wait(set);
1533
1534                         ptlrpc_set_destroy(set);
1535                 }
1536
1537                 lfsck_component_put(env, com);
1538         }
1539
1540         RETURN(rc);
1541 }
1542
1543 void lfsck_quit(const struct lu_env *env, struct lfsck_instance *lfsck)
1544 {
1545         struct lfsck_component *com;
1546         struct lfsck_component *next;
1547
1548         list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
1549                                  lc_link) {
1550                 if (com->lc_ops->lfsck_quit != NULL)
1551                         com->lc_ops->lfsck_quit(env, com);
1552
1553                 spin_lock(&lfsck->li_lock);
1554                 list_del_init(&com->lc_link);
1555                 list_del_init(&com->lc_link_dir);
1556                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
1557                 spin_unlock(&lfsck->li_lock);
1558         }
1559
1560         list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
1561                                  lc_link) {
1562                 if (com->lc_ops->lfsck_quit != NULL)
1563                         com->lc_ops->lfsck_quit(env, com);
1564
1565                 spin_lock(&lfsck->li_lock);
1566                 list_del_init(&com->lc_link);
1567                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
1568                 spin_unlock(&lfsck->li_lock);
1569         }
1570 }
1571
1572 static int lfsck_async_interpret(const struct lu_env *env,
1573                                  struct ptlrpc_request *req,
1574                                  void *args, int rc)
1575 {
1576         struct lfsck_async_interpret_args *laia = args;
1577         struct lfsck_instance             *lfsck;
1578
1579         lfsck = container_of0(laia->laia_ltds, struct lfsck_instance,
1580                               li_mdt_descs);
1581         lfsck_interpret(env, lfsck, req, laia, rc);
1582         lfsck_tgt_put(laia->laia_ltd);
1583         if (rc != 0 && laia->laia_result != -EALREADY)
1584                 laia->laia_result = rc;
1585
1586         return 0;
1587 }
1588
1589 int lfsck_async_request(const struct lu_env *env, struct obd_export *exp,
1590                         struct lfsck_request *lr,
1591                         struct ptlrpc_request_set *set,
1592                         ptlrpc_interpterer_t interpreter,
1593                         void *args, int request)
1594 {
1595         struct lfsck_async_interpret_args *laia;
1596         struct ptlrpc_request             *req;
1597         struct lfsck_request              *tmp;
1598         struct req_format                 *format;
1599         int                                rc;
1600
1601         switch (request) {
1602         case LFSCK_NOTIFY:
1603                 format = &RQF_LFSCK_NOTIFY;
1604                 break;
1605         case LFSCK_QUERY:
1606                 format = &RQF_LFSCK_QUERY;
1607                 break;
1608         default:
1609                 CERROR("%s: unknown async request: opc = %d\n",
1610                        exp->exp_obd->obd_name, request);
1611                 return -EINVAL;
1612         }
1613
1614         req = ptlrpc_request_alloc(class_exp2cliimp(exp), format);
1615         if (req == NULL)
1616                 return -ENOMEM;
1617
1618         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, request);
1619         if (rc != 0) {
1620                 ptlrpc_request_free(req);
1621
1622                 return rc;
1623         }
1624
1625         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
1626         *tmp = *lr;
1627         ptlrpc_request_set_replen(req);
1628
1629         laia = ptlrpc_req_async_args(req);
1630         *laia = *(struct lfsck_async_interpret_args *)args;
1631         if (laia->laia_com != NULL)
1632                 lfsck_component_get(laia->laia_com);
1633         req->rq_interpret_reply = interpreter;
1634         ptlrpc_set_add_req(set, req);
1635
1636         return 0;
1637 }
1638
1639 /* external interfaces */
1640
1641 int lfsck_get_speed(struct seq_file *m, struct dt_device *key)
1642 {
1643         struct lu_env           env;
1644         struct lfsck_instance  *lfsck;
1645         int                     rc;
1646         ENTRY;
1647
1648         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
1649         if (rc != 0)
1650                 RETURN(rc);
1651
1652         lfsck = lfsck_instance_find(key, true, false);
1653         if (likely(lfsck != NULL)) {
1654                 seq_printf(m, "%u\n", lfsck->li_bookmark_ram.lb_speed_limit);
1655                 lfsck_instance_put(&env, lfsck);
1656         } else {
1657                 rc = -ENXIO;
1658         }
1659
1660         lu_env_fini(&env);
1661
1662         RETURN(rc);
1663 }
1664 EXPORT_SYMBOL(lfsck_get_speed);
1665
1666 int lfsck_set_speed(struct dt_device *key, int val)
1667 {
1668         struct lu_env           env;
1669         struct lfsck_instance  *lfsck;
1670         int                     rc;
1671         ENTRY;
1672
1673         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
1674         if (rc != 0)
1675                 RETURN(rc);
1676
1677         lfsck = lfsck_instance_find(key, true, false);
1678         if (likely(lfsck != NULL)) {
1679                 mutex_lock(&lfsck->li_mutex);
1680                 if (__lfsck_set_speed(lfsck, val))
1681                         rc = lfsck_bookmark_store(&env, lfsck);
1682                 mutex_unlock(&lfsck->li_mutex);
1683                 lfsck_instance_put(&env, lfsck);
1684         } else {
1685                 rc = -ENXIO;
1686         }
1687
1688         lu_env_fini(&env);
1689
1690         RETURN(rc);
1691 }
1692 EXPORT_SYMBOL(lfsck_set_speed);
1693
1694 int lfsck_get_windows(struct seq_file *m, struct dt_device *key)
1695 {
1696         struct lu_env           env;
1697         struct lfsck_instance  *lfsck;
1698         int                     rc;
1699         ENTRY;
1700
1701         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
1702         if (rc != 0)
1703                 RETURN(rc);
1704
1705         lfsck = lfsck_instance_find(key, true, false);
1706         if (likely(lfsck != NULL)) {
1707                 seq_printf(m, "%u\n", lfsck->li_bookmark_ram.lb_async_windows);
1708                 lfsck_instance_put(&env, lfsck);
1709         } else {
1710                 rc = -ENXIO;
1711         }
1712
1713         lu_env_fini(&env);
1714
1715         RETURN(rc);
1716 }
1717 EXPORT_SYMBOL(lfsck_get_windows);
1718
1719 int lfsck_set_windows(struct dt_device *key, int val)
1720 {
1721         struct lu_env           env;
1722         struct lfsck_instance  *lfsck;
1723         int                     rc;
1724         ENTRY;
1725
1726         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
1727         if (rc != 0)
1728                 RETURN(rc);
1729
1730         lfsck = lfsck_instance_find(key, true, false);
1731         if (likely(lfsck != NULL)) {
1732                 if (val > LFSCK_ASYNC_WIN_MAX) {
1733                         CWARN("%s: Too large async window size, which "
1734                               "may cause memory issues. The valid range "
1735                               "is [0 - %u]. If you do not want to restrict "
1736                               "the window size for async requests pipeline, "
1737                               "just set it as 0.\n",
1738                               lfsck_lfsck2name(lfsck), LFSCK_ASYNC_WIN_MAX);
1739                         rc = -EINVAL;
1740                 } else if (lfsck->li_bookmark_ram.lb_async_windows != val) {
1741                         mutex_lock(&lfsck->li_mutex);
1742                         lfsck->li_bookmark_ram.lb_async_windows = val;
1743                         rc = lfsck_bookmark_store(&env, lfsck);
1744                         mutex_unlock(&lfsck->li_mutex);
1745                 }
1746                 lfsck_instance_put(&env, lfsck);
1747         } else {
1748                 rc = -ENXIO;
1749         }
1750
1751         lu_env_fini(&env);
1752
1753         RETURN(rc);
1754 }
1755 EXPORT_SYMBOL(lfsck_set_windows);
1756
1757 int lfsck_dump(struct seq_file *m, struct dt_device *key, enum lfsck_type type)
1758 {
1759         struct lu_env           env;
1760         struct lfsck_instance  *lfsck;
1761         struct lfsck_component *com;
1762         int                     rc;
1763         ENTRY;
1764
1765         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
1766         if (rc != 0)
1767                 RETURN(rc);
1768
1769         lfsck = lfsck_instance_find(key, true, false);
1770         if (likely(lfsck != NULL)) {
1771                 com = lfsck_component_find(lfsck, type);
1772                 if (likely(com != NULL)) {
1773                         rc = com->lc_ops->lfsck_dump(&env, com, m);
1774                         lfsck_component_put(&env, com);
1775                 } else {
1776                         rc = -ENOTSUPP;
1777                 }
1778
1779                 lfsck_instance_put(&env, lfsck);
1780         } else {
1781                 rc = -ENXIO;
1782         }
1783
1784         lu_env_fini(&env);
1785
1786         RETURN(rc);
1787 }
1788 EXPORT_SYMBOL(lfsck_dump);
1789
1790 static int lfsck_stop_all(const struct lu_env *env,
1791                           struct lfsck_instance *lfsck,
1792                           struct lfsck_stop *stop)
1793 {
1794         struct lfsck_thread_info          *info   = lfsck_env_info(env);
1795         struct lfsck_request              *lr     = &info->lti_lr;
1796         struct lfsck_async_interpret_args *laia   = &info->lti_laia;
1797         struct ptlrpc_request_set         *set;
1798         struct lfsck_tgt_descs            *ltds   = &lfsck->li_mdt_descs;
1799         struct lfsck_tgt_desc             *ltd;
1800         struct lfsck_bookmark             *bk     = &lfsck->li_bookmark_ram;
1801         __u32                              idx;
1802         int                                rc     = 0;
1803         int                                rc1    = 0;
1804         ENTRY;
1805
1806         LASSERT(stop->ls_flags & LPF_BROADCAST);
1807
1808         set = ptlrpc_prep_set();
1809         if (unlikely(set == NULL)) {
1810                 CERROR("%s: cannot allocate memory for stop LFSCK on "
1811                        "all targets\n", lfsck_lfsck2name(lfsck));
1812
1813                 RETURN(-ENOMEM);
1814         }
1815
1816         memset(lr, 0, sizeof(*lr));
1817         lr->lr_event = LE_STOP;
1818         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1819         lr->lr_status = stop->ls_status;
1820         lr->lr_version = bk->lb_version;
1821         lr->lr_active = LFSCK_TYPES_ALL;
1822         lr->lr_param = stop->ls_flags;
1823
1824         laia->laia_com = NULL;
1825         laia->laia_ltds = ltds;
1826         laia->laia_lr = lr;
1827         laia->laia_result = 0;
1828         laia->laia_shared = 1;
1829
1830         down_read(&ltds->ltd_rw_sem);
1831         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1832                 ltd = lfsck_tgt_get(ltds, idx);
1833                 LASSERT(ltd != NULL);
1834
1835                 laia->laia_ltd = ltd;
1836                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1837                                          lfsck_async_interpret, laia,
1838                                          LFSCK_NOTIFY);
1839                 if (rc != 0) {
1840                         lfsck_interpret(env, lfsck, NULL, laia, rc);
1841                         lfsck_tgt_put(ltd);
1842                         CWARN("%s: cannot notify MDT %x for LFSCK stop: "
1843                               "rc = %d\n", lfsck_lfsck2name(lfsck), idx, rc);
1844                         rc1 = rc;
1845                 }
1846         }
1847         up_read(&ltds->ltd_rw_sem);
1848
1849         rc = ptlrpc_set_wait(set);
1850         ptlrpc_set_destroy(set);
1851
1852         if (rc == 0)
1853                 rc = laia->laia_result;
1854
1855         if (rc == -EALREADY)
1856                 rc = 0;
1857
1858         if (rc != 0)
1859                 CWARN("%s: fail to stop LFSCK on some MDTs: rc = %d\n",
1860                       lfsck_lfsck2name(lfsck), rc);
1861
1862         RETURN(rc != 0 ? rc : rc1);
1863 }
1864
1865 static int lfsck_start_all(const struct lu_env *env,
1866                            struct lfsck_instance *lfsck,
1867                            struct lfsck_start *start)
1868 {
1869         struct lfsck_thread_info          *info   = lfsck_env_info(env);
1870         struct lfsck_request              *lr     = &info->lti_lr;
1871         struct lfsck_async_interpret_args *laia   = &info->lti_laia;
1872         struct ptlrpc_request_set         *set;
1873         struct lfsck_tgt_descs            *ltds   = &lfsck->li_mdt_descs;
1874         struct lfsck_tgt_desc             *ltd;
1875         struct lfsck_bookmark             *bk     = &lfsck->li_bookmark_ram;
1876         __u32                              idx;
1877         int                                rc     = 0;
1878         ENTRY;
1879
1880         LASSERT(start->ls_flags & LPF_BROADCAST);
1881
1882         set = ptlrpc_prep_set();
1883         if (unlikely(set == NULL)) {
1884                 if (bk->lb_param & LPF_FAILOUT) {
1885                         CERROR("%s: cannot allocate memory for start LFSCK on "
1886                                "all targets, failout.\n",
1887                                lfsck_lfsck2name(lfsck));
1888
1889                         RETURN(-ENOMEM);
1890                 } else {
1891                         CWARN("%s: cannot allocate memory for start LFSCK on "
1892                               "all targets, partly scan.\n",
1893                               lfsck_lfsck2name(lfsck));
1894
1895                         RETURN(0);
1896                 }
1897         }
1898
1899         memset(lr, 0, sizeof(*lr));
1900         lr->lr_event = LE_START;
1901         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1902         lr->lr_speed = bk->lb_speed_limit;
1903         lr->lr_version = bk->lb_version;
1904         lr->lr_active = start->ls_active;
1905         lr->lr_param = start->ls_flags;
1906         lr->lr_async_windows = bk->lb_async_windows;
1907         lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
1908                        LSV_ASYNC_WINDOWS;
1909
1910         laia->laia_com = NULL;
1911         laia->laia_ltds = ltds;
1912         laia->laia_lr = lr;
1913         laia->laia_result = 0;
1914         laia->laia_shared = 1;
1915
1916         down_read(&ltds->ltd_rw_sem);
1917         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1918                 ltd = lfsck_tgt_get(ltds, idx);
1919                 LASSERT(ltd != NULL);
1920
1921                 laia->laia_ltd = ltd;
1922                 ltd->ltd_layout_done = 0;
1923                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1924                                          lfsck_async_interpret, laia,
1925                                          LFSCK_NOTIFY);
1926                 if (rc != 0) {
1927                         lfsck_interpret(env, lfsck, NULL, laia, rc);
1928                         lfsck_tgt_put(ltd);
1929                         if (bk->lb_param & LPF_FAILOUT) {
1930                                 CERROR("%s: cannot notify MDT %x for LFSCK "
1931                                        "start, failout: rc = %d\n",
1932                                        lfsck_lfsck2name(lfsck), idx, rc);
1933                                 break;
1934                         } else {
1935                                 CWARN("%s: cannot notify MDT %x for LFSCK "
1936                                       "start, partly scan: rc = %d\n",
1937                                       lfsck_lfsck2name(lfsck), idx, rc);
1938                                 rc = 0;
1939                         }
1940                 }
1941         }
1942         up_read(&ltds->ltd_rw_sem);
1943
1944         if (rc != 0) {
1945                 ptlrpc_set_destroy(set);
1946
1947                 RETURN(rc);
1948         }
1949
1950         rc = ptlrpc_set_wait(set);
1951         ptlrpc_set_destroy(set);
1952
1953         if (rc == 0)
1954                 rc = laia->laia_result;
1955
1956         if (rc != 0) {
1957                 if (bk->lb_param & LPF_FAILOUT) {
1958                         struct lfsck_stop *stop = &info->lti_stop;
1959
1960                         CERROR("%s: cannot start LFSCK on some MDTs, "
1961                                "stop all: rc = %d\n",
1962                                lfsck_lfsck2name(lfsck), rc);
1963                         if (rc != -EALREADY) {
1964                                 stop->ls_status = LS_FAILED;
1965                                 stop->ls_flags = LPF_ALL_TGT | LPF_BROADCAST;
1966                                 lfsck_stop_all(env, lfsck, stop);
1967                         }
1968                 } else {
1969                         CWARN("%s: cannot start LFSCK on some MDTs, "
1970                               "partly scan: rc = %d\n",
1971                               lfsck_lfsck2name(lfsck), rc);
1972                         rc = 0;
1973                 }
1974         }
1975
1976         RETURN(rc);
1977 }
1978
1979 int lfsck_start(const struct lu_env *env, struct dt_device *key,
1980                 struct lfsck_start_param *lsp)
1981 {
1982         struct lfsck_start              *start  = lsp->lsp_start;
1983         struct lfsck_instance           *lfsck;
1984         struct lfsck_bookmark           *bk;
1985         struct ptlrpc_thread            *thread;
1986         struct lfsck_component          *com;
1987         struct l_wait_info               lwi    = { 0 };
1988         struct lfsck_thread_args        *lta;
1989         struct task_struct              *task;
1990         int                              rc     = 0;
1991         __u16                            valid  = 0;
1992         __u16                            flags  = 0;
1993         __u16                            type   = 1;
1994         ENTRY;
1995
1996         lfsck = lfsck_instance_find(key, true, false);
1997         if (unlikely(lfsck == NULL))
1998                 RETURN(-ENXIO);
1999
2000         /* System is not ready, try again later. */
2001         if (unlikely(lfsck->li_namespace == NULL))
2002                 GOTO(put, rc = -EAGAIN);
2003
2004         /* start == NULL means auto trigger paused LFSCK. */
2005         if ((start == NULL) &&
2006             (cfs_list_empty(&lfsck->li_list_scan) ||
2007              OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AUTO)))
2008                 GOTO(put, rc = 0);
2009
2010         bk = &lfsck->li_bookmark_ram;
2011         thread = &lfsck->li_thread;
2012         mutex_lock(&lfsck->li_mutex);
2013         spin_lock(&lfsck->li_lock);
2014         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2015                 rc = -EALREADY;
2016                 if (unlikely(start == NULL)) {
2017                         spin_unlock(&lfsck->li_lock);
2018                         GOTO(out, rc);
2019                 }
2020
2021                 while (start->ls_active != 0) {
2022                         if (!(type & start->ls_active)) {
2023                                 type <<= 1;
2024                                 continue;
2025                         }
2026
2027                         com = __lfsck_component_find(lfsck, type,
2028                                                      &lfsck->li_list_scan);
2029                         if (com == NULL)
2030                                 com = __lfsck_component_find(lfsck, type,
2031                                                 &lfsck->li_list_double_scan);
2032                         if (com == NULL) {
2033                                 rc = -EOPNOTSUPP;
2034                                 break;
2035                         }
2036
2037                         if (com->lc_ops->lfsck_join != NULL) {
2038                                 rc = com->lc_ops->lfsck_join( env, com, lsp);
2039                                 if (rc != 0 && rc != -EALREADY)
2040                                         break;
2041                         }
2042                         start->ls_active &= ~type;
2043                         type <<= 1;
2044                 }
2045                 spin_unlock(&lfsck->li_lock);
2046                 GOTO(out, rc);
2047         }
2048         spin_unlock(&lfsck->li_lock);
2049
2050         lfsck->li_status = 0;
2051         lfsck->li_oit_over = 0;
2052         lfsck->li_start_unplug = 0;
2053         lfsck->li_drop_dryrun = 0;
2054         lfsck->li_new_scanned = 0;
2055
2056         /* For auto trigger. */
2057         if (start == NULL)
2058                 goto trigger;
2059
2060         if (start->ls_flags & LPF_BROADCAST && !lfsck->li_master) {
2061                 CERROR("%s: only allow to specify '-A | -o' via MDS\n",
2062                        lfsck_lfsck2name(lfsck));
2063
2064                 GOTO(out, rc = -EPERM);
2065         }
2066
2067         start->ls_version = bk->lb_version;
2068
2069         if (start->ls_active != 0) {
2070                 struct lfsck_component *next;
2071
2072                 if (start->ls_active == LFSCK_TYPES_ALL)
2073                         start->ls_active = LFSCK_TYPES_SUPPORTED;
2074
2075                 if (start->ls_active & ~LFSCK_TYPES_SUPPORTED) {
2076                         start->ls_active &= ~LFSCK_TYPES_SUPPORTED;
2077                         GOTO(out, rc = -ENOTSUPP);
2078                 }
2079
2080                 list_for_each_entry_safe(com, next,
2081                                          &lfsck->li_list_scan, lc_link) {
2082                         if (!(com->lc_type & start->ls_active)) {
2083                                 rc = com->lc_ops->lfsck_post(env, com, 0,
2084                                                              false);
2085                                 if (rc != 0)
2086                                         GOTO(out, rc);
2087                         }
2088                 }
2089
2090                 while (start->ls_active != 0) {
2091                         if (type & start->ls_active) {
2092                                 com = __lfsck_component_find(lfsck, type,
2093                                                         &lfsck->li_list_idle);
2094                                 if (com != NULL) {
2095                                         /* The component status will be updated
2096                                          * when its prep() is called later by
2097                                          * the LFSCK main engine. */
2098                                         list_del_init(&com->lc_link);
2099                                         list_add_tail(&com->lc_link,
2100                                                       &lfsck->li_list_scan);
2101                                 }
2102                                 start->ls_active &= ~type;
2103                         }
2104                         type <<= 1;
2105                 }
2106         }
2107
2108         if (list_empty(&lfsck->li_list_scan)) {
2109                 /* The speed limit will be used to control both the LFSCK and
2110                  * low layer scrub (if applied), need to be handled firstly. */
2111                 if (start->ls_valid & LSV_SPEED_LIMIT) {
2112                         if (__lfsck_set_speed(lfsck, start->ls_speed_limit)) {
2113                                 rc = lfsck_bookmark_store(env, lfsck);
2114                                 if (rc != 0)
2115                                         GOTO(out, rc);
2116                         }
2117                 }
2118
2119                 goto trigger;
2120         }
2121
2122         if (start->ls_flags & LPF_RESET)
2123                 flags |= DOIF_RESET;
2124
2125         rc = lfsck_set_param(env, lfsck, start, !!(flags & DOIF_RESET));
2126         if (rc != 0)
2127                 GOTO(out, rc);
2128
2129         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
2130                 start->ls_active |= com->lc_type;
2131                 if (flags & DOIF_RESET) {
2132                         rc = com->lc_ops->lfsck_reset(env, com, false);
2133                         if (rc != 0)
2134                                 GOTO(out, rc);
2135                 }
2136         }
2137
2138 trigger:
2139         lfsck->li_args_dir = LUDA_64BITHASH | LUDA_VERIFY;
2140         if (bk->lb_param & LPF_DRYRUN)
2141                 lfsck->li_args_dir |= LUDA_VERIFY_DRYRUN;
2142
2143         if (start != NULL && start->ls_valid & LSV_ERROR_HANDLE) {
2144                 valid |= DOIV_ERROR_HANDLE;
2145                 if (start->ls_flags & LPF_FAILOUT)
2146                         flags |= DOIF_FAILOUT;
2147         }
2148
2149         if (start != NULL && start->ls_valid & LSV_DRYRUN) {
2150                 valid |= DOIV_DRYRUN;
2151                 if (start->ls_flags & LPF_DRYRUN)
2152                         flags |= DOIF_DRYRUN;
2153         }
2154
2155         if (!list_empty(&lfsck->li_list_scan))
2156                 flags |= DOIF_OUTUSED;
2157
2158         lfsck->li_args_oit = (flags << DT_OTABLE_IT_FLAGS_SHIFT) | valid;
2159         thread_set_flags(thread, 0);
2160         lta = lfsck_thread_args_init(lfsck, NULL, lsp);
2161         if (IS_ERR(lta))
2162                 GOTO(out, rc = PTR_ERR(lta));
2163
2164         __lfsck_set_speed(lfsck, bk->lb_speed_limit);
2165         task = kthread_run(lfsck_master_engine, lta, "lfsck");
2166         if (IS_ERR(task)) {
2167                 rc = PTR_ERR(task);
2168                 CERROR("%s: cannot start LFSCK thread: rc = %d\n",
2169                        lfsck_lfsck2name(lfsck), rc);
2170                 lfsck_thread_args_fini(lta);
2171
2172                 GOTO(out, rc);
2173         }
2174
2175         l_wait_event(thread->t_ctl_waitq,
2176                      thread_is_running(thread) ||
2177                      thread_is_stopped(thread),
2178                      &lwi);
2179         if (start == NULL || !(start->ls_flags & LPF_BROADCAST)) {
2180                 lfsck->li_start_unplug = 1;
2181                 wake_up_all(&thread->t_ctl_waitq);
2182
2183                 GOTO(out, rc = 0);
2184         }
2185
2186         /* release lfsck::li_mutex to avoid deadlock. */
2187         mutex_unlock(&lfsck->li_mutex);
2188         rc = lfsck_start_all(env, lfsck, start);
2189         if (rc != 0) {
2190                 spin_lock(&lfsck->li_lock);
2191                 if (thread_is_stopped(thread)) {
2192                         spin_unlock(&lfsck->li_lock);
2193                 } else {
2194                         lfsck->li_status = LS_FAILED;
2195                         lfsck->li_flags = 0;
2196                         thread_set_flags(thread, SVC_STOPPING);
2197                         spin_unlock(&lfsck->li_lock);
2198
2199                         lfsck->li_start_unplug = 1;
2200                         wake_up_all(&thread->t_ctl_waitq);
2201                         l_wait_event(thread->t_ctl_waitq,
2202                                      thread_is_stopped(thread),
2203                                      &lwi);
2204                 }
2205         } else {
2206                 lfsck->li_start_unplug = 1;
2207                 wake_up_all(&thread->t_ctl_waitq);
2208         }
2209
2210         GOTO(put, rc);
2211
2212 out:
2213         mutex_unlock(&lfsck->li_mutex);
2214
2215 put:
2216         lfsck_instance_put(env, lfsck);
2217
2218         return rc < 0 ? rc : 0;
2219 }
2220 EXPORT_SYMBOL(lfsck_start);
2221
2222 int lfsck_stop(const struct lu_env *env, struct dt_device *key,
2223                struct lfsck_stop *stop)
2224 {
2225         struct lfsck_instance   *lfsck;
2226         struct ptlrpc_thread    *thread;
2227         struct l_wait_info       lwi    = { 0 };
2228         int                      rc     = 0;
2229         int                      rc1    = 0;
2230         ENTRY;
2231
2232         lfsck = lfsck_instance_find(key, true, false);
2233         if (unlikely(lfsck == NULL))
2234                 RETURN(-ENXIO);
2235
2236         thread = &lfsck->li_thread;
2237         /* release lfsck::li_mutex to avoid deadlock. */
2238         if (stop != NULL && stop->ls_flags & LPF_BROADCAST) {
2239                 if (!lfsck->li_master) {
2240                         CERROR("%s: only allow to specify '-A' via MDS\n",
2241                                lfsck_lfsck2name(lfsck));
2242
2243                         GOTO(out, rc = -EPERM);
2244                 }
2245
2246                 rc1 = lfsck_stop_all(env, lfsck, stop);
2247         }
2248
2249         mutex_lock(&lfsck->li_mutex);
2250         spin_lock(&lfsck->li_lock);
2251         /* no error if LFSCK is already stopped, or was never started */
2252         if (thread_is_init(thread) || thread_is_stopped(thread)) {
2253                 spin_unlock(&lfsck->li_lock);
2254                 GOTO(out, rc = 0);
2255         }
2256
2257         if (stop != NULL) {
2258                 lfsck->li_status = stop->ls_status;
2259                 lfsck->li_flags = stop->ls_flags;
2260         } else {
2261                 lfsck->li_status = LS_STOPPED;
2262                 lfsck->li_flags = 0;
2263         }
2264
2265         thread_set_flags(thread, SVC_STOPPING);
2266         spin_unlock(&lfsck->li_lock);
2267
2268         wake_up_all(&thread->t_ctl_waitq);
2269         l_wait_event(thread->t_ctl_waitq,
2270                      thread_is_stopped(thread),
2271                      &lwi);
2272
2273         GOTO(out, rc = 0);
2274
2275 out:
2276         mutex_unlock(&lfsck->li_mutex);
2277         lfsck_instance_put(env, lfsck);
2278
2279         return rc != 0 ? rc : rc1;
2280 }
2281 EXPORT_SYMBOL(lfsck_stop);
2282
2283 int lfsck_in_notify(const struct lu_env *env, struct dt_device *key,
2284                     struct lfsck_request *lr)
2285 {
2286         int rc = -EOPNOTSUPP;
2287         ENTRY;
2288
2289         switch (lr->lr_event) {
2290         case LE_START: {
2291                 struct lfsck_start       *start = &lfsck_env_info(env)->lti_start;
2292                 struct lfsck_start_param  lsp;
2293
2294                 memset(start, 0, sizeof(*start));
2295                 start->ls_valid = lr->lr_valid;
2296                 start->ls_speed_limit = lr->lr_speed;
2297                 start->ls_version = lr->lr_version;
2298                 start->ls_active = lr->lr_active;
2299                 start->ls_flags = lr->lr_param & ~LPF_BROADCAST;
2300                 start->ls_async_windows = lr->lr_async_windows;
2301
2302                 lsp.lsp_start = start;
2303                 lsp.lsp_index = lr->lr_index;
2304                 lsp.lsp_index_valid = 1;
2305                 rc = lfsck_start(env, key, &lsp);
2306                 break;
2307         }
2308         case LE_STOP: {
2309                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
2310
2311                 memset(stop, 0, sizeof(*stop));
2312                 stop->ls_status = lr->lr_status;
2313                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
2314                 rc = lfsck_stop(env, key, stop);
2315                 break;
2316         }
2317         case LE_PHASE1_DONE:
2318         case LE_PHASE2_DONE:
2319         case LE_FID_ACCESSED:
2320         case LE_PEER_EXIT:
2321         case LE_CONDITIONAL_DESTROY:
2322         case LE_PAIRS_VERIFY: {
2323                 struct lfsck_instance  *lfsck;
2324                 struct lfsck_component *com;
2325
2326                 lfsck = lfsck_instance_find(key, true, false);
2327                 if (unlikely(lfsck == NULL))
2328                         RETURN(-ENXIO);
2329
2330                 com = lfsck_component_find(lfsck, lr->lr_active);
2331                 if (likely(com != NULL)) {
2332                         rc = com->lc_ops->lfsck_in_notify(env, com, lr);
2333                         lfsck_component_put(env, com);
2334                 }
2335
2336                 lfsck_instance_put(env, lfsck);
2337                 break;
2338         }
2339         default:
2340                 break;
2341         }
2342
2343         RETURN(rc);
2344 }
2345 EXPORT_SYMBOL(lfsck_in_notify);
2346
2347 int lfsck_query(const struct lu_env *env, struct dt_device *key,
2348                 struct lfsck_request *lr)
2349 {
2350         struct lfsck_instance  *lfsck;
2351         struct lfsck_component *com;
2352         int                     rc;
2353         ENTRY;
2354
2355         lfsck = lfsck_instance_find(key, true, false);
2356         if (unlikely(lfsck == NULL))
2357                 RETURN(-ENXIO);
2358
2359         com = lfsck_component_find(lfsck, lr->lr_active);
2360         if (likely(com != NULL)) {
2361                 rc = com->lc_ops->lfsck_query(env, com);
2362                 lfsck_component_put(env, com);
2363         } else {
2364                 rc = -ENOTSUPP;
2365         }
2366
2367         lfsck_instance_put(env, lfsck);
2368
2369         RETURN(rc);
2370 }
2371 EXPORT_SYMBOL(lfsck_query);
2372
2373 int lfsck_register_namespace(const struct lu_env *env, struct dt_device *key,
2374                              struct ldlm_namespace *ns)
2375 {
2376         struct lfsck_instance  *lfsck;
2377         int                     rc      = -ENXIO;
2378
2379         lfsck = lfsck_instance_find(key, true, false);
2380         if (likely(lfsck != NULL)) {
2381                 lfsck->li_namespace = ns;
2382                 lfsck_instance_put(env, lfsck);
2383                 rc = 0;
2384         }
2385
2386         return rc;
2387 }
2388 EXPORT_SYMBOL(lfsck_register_namespace);
2389
2390 int lfsck_register(const struct lu_env *env, struct dt_device *key,
2391                    struct dt_device *next, struct obd_device *obd,
2392                    lfsck_out_notify notify, void *notify_data, bool master)
2393 {
2394         struct lfsck_instance   *lfsck;
2395         struct dt_object        *root  = NULL;
2396         struct dt_object        *obj;
2397         struct lu_fid           *fid   = &lfsck_env_info(env)->lti_fid;
2398         int                      rc;
2399         ENTRY;
2400
2401         lfsck = lfsck_instance_find(key, false, false);
2402         if (unlikely(lfsck != NULL))
2403                 RETURN(-EEXIST);
2404
2405         OBD_ALLOC_PTR(lfsck);
2406         if (lfsck == NULL)
2407                 RETURN(-ENOMEM);
2408
2409         mutex_init(&lfsck->li_mutex);
2410         spin_lock_init(&lfsck->li_lock);
2411         CFS_INIT_LIST_HEAD(&lfsck->li_link);
2412         CFS_INIT_LIST_HEAD(&lfsck->li_list_scan);
2413         CFS_INIT_LIST_HEAD(&lfsck->li_list_dir);
2414         CFS_INIT_LIST_HEAD(&lfsck->li_list_double_scan);
2415         CFS_INIT_LIST_HEAD(&lfsck->li_list_idle);
2416         atomic_set(&lfsck->li_ref, 1);
2417         atomic_set(&lfsck->li_double_scan_count, 0);
2418         init_waitqueue_head(&lfsck->li_thread.t_ctl_waitq);
2419         lfsck->li_out_notify = notify;
2420         lfsck->li_out_notify_data = notify_data;
2421         lfsck->li_next = next;
2422         lfsck->li_bottom = key;
2423         lfsck->li_obd = obd;
2424
2425         rc = lfsck_tgt_descs_init(&lfsck->li_ost_descs);
2426         if (rc != 0)
2427                 GOTO(out, rc);
2428
2429         rc = lfsck_tgt_descs_init(&lfsck->li_mdt_descs);
2430         if (rc != 0)
2431                 GOTO(out, rc);
2432
2433         fid->f_seq = FID_SEQ_LOCAL_NAME;
2434         fid->f_oid = 1;
2435         fid->f_ver = 0;
2436         rc = local_oid_storage_init(env, lfsck->li_bottom, fid, &lfsck->li_los);
2437         if (rc != 0)
2438                 GOTO(out, rc);
2439
2440         rc = dt_root_get(env, key, fid);
2441         if (rc != 0)
2442                 GOTO(out, rc);
2443
2444         root = dt_locate(env, lfsck->li_bottom, fid);
2445         if (IS_ERR(root))
2446                 GOTO(out, rc = PTR_ERR(root));
2447
2448         if (unlikely(!dt_try_as_dir(env, root)))
2449                 GOTO(out, rc = -ENOTDIR);
2450
2451         lfsck->li_local_root_fid = *fid;
2452         if (master) {
2453                 lfsck->li_master = 1;
2454                 if (lfsck_dev_idx(lfsck->li_bottom) == 0) {
2455                         rc = dt_lookup(env, root,
2456                                 (struct dt_rec *)(&lfsck->li_global_root_fid),
2457                                 (const struct dt_key *)"ROOT", BYPASS_CAPA);
2458                         if (rc != 0)
2459                                 GOTO(out, rc);
2460                 }
2461         }
2462
2463         fid->f_seq = FID_SEQ_LOCAL_FILE;
2464         fid->f_oid = OTABLE_IT_OID;
2465         fid->f_ver = 0;
2466         obj = dt_locate(env, lfsck->li_bottom, fid);
2467         if (IS_ERR(obj))
2468                 GOTO(out, rc = PTR_ERR(obj));
2469
2470         lfsck->li_obj_oit = obj;
2471         rc = obj->do_ops->do_index_try(env, obj, &dt_otable_features);
2472         if (rc != 0) {
2473                 if (rc == -ENOTSUPP)
2474                         GOTO(add, rc = 0);
2475
2476                 GOTO(out, rc);
2477         }
2478
2479         rc = lfsck_bookmark_setup(env, lfsck);
2480         if (rc != 0)
2481                 GOTO(out, rc);
2482
2483         if (master) {
2484                 rc = lfsck_fid_init(lfsck);
2485                 if (rc < 0)
2486                         GOTO(out, rc);
2487
2488                 rc = lfsck_namespace_setup(env, lfsck);
2489                 if (rc < 0)
2490                         GOTO(out, rc);
2491         }
2492
2493         rc = lfsck_layout_setup(env, lfsck);
2494         if (rc < 0)
2495                 GOTO(out, rc);
2496
2497         /* XXX: more LFSCK components initialization to be added here. */
2498
2499 add:
2500         rc = lfsck_instance_add(lfsck);
2501         if (rc == 0)
2502                 rc = lfsck_add_target_from_orphan(env, lfsck);
2503 out:
2504         if (root != NULL && !IS_ERR(root))
2505                 lu_object_put(env, &root->do_lu);
2506         if (rc != 0)
2507                 lfsck_instance_cleanup(env, lfsck);
2508         return rc;
2509 }
2510 EXPORT_SYMBOL(lfsck_register);
2511
2512 void lfsck_degister(const struct lu_env *env, struct dt_device *key)
2513 {
2514         struct lfsck_instance *lfsck;
2515
2516         lfsck = lfsck_instance_find(key, false, true);
2517         if (lfsck != NULL)
2518                 lfsck_instance_put(env, lfsck);
2519 }
2520 EXPORT_SYMBOL(lfsck_degister);
2521
2522 int lfsck_add_target(const struct lu_env *env, struct dt_device *key,
2523                      struct dt_device *tgt, struct obd_export *exp,
2524                      __u32 index, bool for_ost)
2525 {
2526         struct lfsck_instance   *lfsck;
2527         struct lfsck_tgt_desc   *ltd;
2528         int                      rc;
2529         ENTRY;
2530
2531         OBD_ALLOC_PTR(ltd);
2532         if (ltd == NULL)
2533                 RETURN(-ENOMEM);
2534
2535         ltd->ltd_tgt = tgt;
2536         ltd->ltd_key = key;
2537         ltd->ltd_exp = exp;
2538         INIT_LIST_HEAD(&ltd->ltd_orphan_list);
2539         INIT_LIST_HEAD(&ltd->ltd_layout_list);
2540         INIT_LIST_HEAD(&ltd->ltd_layout_phase_list);
2541         atomic_set(&ltd->ltd_ref, 1);
2542         ltd->ltd_index = index;
2543
2544         spin_lock(&lfsck_instance_lock);
2545         lfsck = __lfsck_instance_find(key, true, false);
2546         if (lfsck == NULL) {
2547                 if (for_ost)
2548                         list_add_tail(&ltd->ltd_orphan_list,
2549                                       &lfsck_ost_orphan_list);
2550                 else
2551                         list_add_tail(&ltd->ltd_orphan_list,
2552                                       &lfsck_mdt_orphan_list);
2553                 spin_unlock(&lfsck_instance_lock);
2554
2555                 RETURN(0);
2556         }
2557         spin_unlock(&lfsck_instance_lock);
2558
2559         rc = __lfsck_add_target(env, lfsck, ltd, for_ost, false);
2560         if (rc != 0)
2561                 lfsck_tgt_put(ltd);
2562
2563         lfsck_instance_put(env, lfsck);
2564
2565         RETURN(rc);
2566 }
2567 EXPORT_SYMBOL(lfsck_add_target);
2568
2569 void lfsck_del_target(const struct lu_env *env, struct dt_device *key,
2570                       struct dt_device *tgt, __u32 index, bool for_ost)
2571 {
2572         struct lfsck_instance   *lfsck;
2573         struct lfsck_tgt_descs  *ltds;
2574         struct lfsck_tgt_desc   *ltd    = NULL;
2575         struct list_head        *head;
2576
2577         if (for_ost)
2578                 head = &lfsck_ost_orphan_list;
2579         else
2580                 head = &lfsck_mdt_orphan_list;
2581
2582         spin_lock(&lfsck_instance_lock);
2583         list_for_each_entry(ltd, head, ltd_orphan_list) {
2584                 if (ltd->ltd_tgt == tgt) {
2585                         list_del_init(&ltd->ltd_orphan_list);
2586                         spin_unlock(&lfsck_instance_lock);
2587                         lfsck_tgt_put(ltd);
2588
2589                         return;
2590                 }
2591         }
2592
2593         lfsck = __lfsck_instance_find(key, true, false);
2594         spin_unlock(&lfsck_instance_lock);
2595         if (unlikely(lfsck == NULL))
2596                 return;
2597
2598         if (for_ost)
2599                 ltds = &lfsck->li_ost_descs;
2600         else
2601                 ltds = &lfsck->li_mdt_descs;
2602
2603         down_write(&ltds->ltd_rw_sem);
2604         LASSERT(ltds->ltd_tgts_bitmap != NULL);
2605
2606         if (unlikely(index >= ltds->ltd_tgts_bitmap->size))
2607                 goto unlock;
2608
2609         ltd = LTD_TGT(ltds, index);
2610         if (unlikely(ltd == NULL))
2611                 goto unlock;
2612
2613         LASSERT(ltds->ltd_tgtnr > 0);
2614
2615         ltds->ltd_tgtnr--;
2616         cfs_bitmap_clear(ltds->ltd_tgts_bitmap, index);
2617         LTD_TGT(ltds, index) = NULL;
2618
2619 unlock:
2620         if (ltd == NULL) {
2621                 if (for_ost)
2622                         head = &lfsck->li_ost_descs.ltd_orphan;
2623                 else
2624                         head = &lfsck->li_ost_descs.ltd_orphan;
2625
2626                 list_for_each_entry(ltd, head, ltd_orphan_list) {
2627                         if (ltd->ltd_tgt == tgt) {
2628                                 list_del_init(&ltd->ltd_orphan_list);
2629                                 break;
2630                         }
2631                 }
2632         }
2633
2634         up_write(&ltds->ltd_rw_sem);
2635         if (ltd != NULL) {
2636                 spin_lock(&ltds->ltd_lock);
2637                 ltd->ltd_dead = 1;
2638                 spin_unlock(&ltds->ltd_lock);
2639                 lfsck_stop_notify(env, lfsck, ltds, ltd, LFSCK_TYPE_LAYOUT);
2640                 lfsck_tgt_put(ltd);
2641         }
2642
2643         lfsck_instance_put(env, lfsck);
2644 }
2645 EXPORT_SYMBOL(lfsck_del_target);
2646
2647 static int __init lfsck_init(void)
2648 {
2649         int rc;
2650
2651         INIT_LIST_HEAD(&lfsck_ost_orphan_list);
2652         INIT_LIST_HEAD(&lfsck_mdt_orphan_list);
2653         lfsck_key_init_generic(&lfsck_thread_key, NULL);
2654         rc = lu_context_key_register(&lfsck_thread_key);
2655         if (rc == 0) {
2656                 tgt_register_lfsck_in_notify(lfsck_in_notify);
2657                 tgt_register_lfsck_query(lfsck_query);
2658         }
2659
2660         return rc;
2661 }
2662
2663 static void __exit lfsck_exit(void)
2664 {
2665         struct lfsck_tgt_desc *ltd;
2666         struct lfsck_tgt_desc *next;
2667
2668         LASSERT(cfs_list_empty(&lfsck_instance_list));
2669
2670         list_for_each_entry_safe(ltd, next, &lfsck_ost_orphan_list,
2671                                  ltd_orphan_list) {
2672                 list_del_init(&ltd->ltd_orphan_list);
2673                 lfsck_tgt_put(ltd);
2674         }
2675
2676         list_for_each_entry_safe(ltd, next, &lfsck_mdt_orphan_list,
2677                                  ltd_orphan_list) {
2678                 list_del_init(&ltd->ltd_orphan_list);
2679                 lfsck_tgt_put(ltd);
2680         }
2681
2682         lu_context_key_degister(&lfsck_thread_key);
2683 }
2684
2685 MODULE_AUTHOR("Intel Corporation <http://www.intel.com/>");
2686 MODULE_DESCRIPTION("LFSCK");
2687 MODULE_LICENSE("GPL");
2688
2689 cfs_module(lfsck, LUSTRE_VERSION_STRING, lfsck_init, lfsck_exit);