lustre/lfsck/lfsck_namespace.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2013, 2014, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_namespace.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 #define DEBUG_SUBSYSTEM S_LFSCK
  32
  33 #include <lustre/lustre_idl.h>
  34 #include <lu_object.h>
  35 #include <dt_object.h>
  36 #include <md_object.h>
  37 #include <lustre_fid.h>
  38 #include <lustre_lib.h>
  39 #include <lustre_net.h>
  40 #include <lustre/lustre_user.h>
  41
  42 #include "lfsck_internal.h"
  43
  44 #define LFSCK_NAMESPACE_MAGIC   0xA0629D03
  45
  46 enum lfsck_nameentry_check {
  47         LFSCK_NAMEENTRY_DEAD            = 1, /* The object has been unlinked. */
  48         LFSCK_NAMEENTRY_REMOVED         = 2, /* The entry has been removed. */
  49         LFSCK_NAMEENTRY_RECREATED       = 3, /* The entry has been recreated. */
  50 };
  51
  52 static const char lfsck_namespace_name[] = "lfsck_namespace";
  53
  54 static struct lfsck_namespace_req *
  55 lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck,
  56                                    struct lu_dirent *ent, __u16 type)
  57 {
  58         struct lfsck_namespace_req *lnr;
  59         int                         size;
  60
  61         size = sizeof(*lnr) + (ent->lde_namelen & ~3) + 4;
  62         OBD_ALLOC(lnr, size);
  63         if (lnr == NULL)
  64                 return ERR_PTR(-ENOMEM);
  65
  66         INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
  67         lnr->lnr_obj = lfsck_object_get(lfsck->li_obj_dir);
  68         lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
  69         lnr->lnr_fid = ent->lde_fid;
  70         lnr->lnr_oit_cookie = lfsck->li_pos_current.lp_oit_cookie;
  71         lnr->lnr_dir_cookie = ent->lde_hash;
  72         lnr->lnr_attr = ent->lde_attrs;
  73         lnr->lnr_size = size;
  74         lnr->lnr_type = type;
  75         lnr->lnr_namelen = ent->lde_namelen;
  76         memcpy(lnr->lnr_name, ent->lde_name, ent->lde_namelen);
  77
  78         return lnr;
  79 }
  80
  81 static void lfsck_namespace_assistant_req_fini(const struct lu_env *env,
  82                                                struct lfsck_assistant_req *lar)
  83 {
  84         struct lfsck_namespace_req *lnr =
  85                         container_of0(lar, struct lfsck_namespace_req, lnr_lar);
  86
  87         if (lnr->lnr_lmv != NULL)
  88                 lfsck_lmv_put(env, lnr->lnr_lmv);
  89
  90         lu_object_put(env, &lnr->lnr_obj->do_lu);
  91         OBD_FREE(lnr, lnr->lnr_size);
  92 }
  93
  94 static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
  95                                       struct lfsck_namespace *src)
  96 {
  97         dst->ln_magic = le32_to_cpu(src->ln_magic);
  98         dst->ln_status = le32_to_cpu(src->ln_status);
  99         dst->ln_flags = le32_to_cpu(src->ln_flags);
 100         dst->ln_success_count = le32_to_cpu(src->ln_success_count);
 101         dst->ln_run_time_phase1 = le32_to_cpu(src->ln_run_time_phase1);
 102         dst->ln_run_time_phase2 = le32_to_cpu(src->ln_run_time_phase2);
 103         dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
 104         dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
 105         dst->ln_time_last_checkpoint =
 106                                 le64_to_cpu(src->ln_time_last_checkpoint);
 107         lfsck_position_le_to_cpu(&dst->ln_pos_latest_start,
 108                                  &src->ln_pos_latest_start);
 109         lfsck_position_le_to_cpu(&dst->ln_pos_last_checkpoint,
 110                                  &src->ln_pos_last_checkpoint);
 111         lfsck_position_le_to_cpu(&dst->ln_pos_first_inconsistent,
 112                                  &src->ln_pos_first_inconsistent);
 113         dst->ln_items_checked = le64_to_cpu(src->ln_items_checked);
 114         dst->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
 115         dst->ln_items_failed = le64_to_cpu(src->ln_items_failed);
 116         dst->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
 117         dst->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
 118         dst->ln_objs_repaired_phase2 =
 119                                 le64_to_cpu(src->ln_objs_repaired_phase2);
 120         dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
 121         dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
 122         fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
 123                       &src->ln_fid_latest_scanned_phase2);
 124         dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
 125         dst->ln_linkea_repaired = le64_to_cpu(src->ln_linkea_repaired);
 126         dst->ln_mul_linked_checked = le64_to_cpu(src->ln_mul_linked_checked);
 127         dst->ln_mul_linked_repaired = le64_to_cpu(src->ln_mul_linked_repaired);
 128         dst->ln_unknown_inconsistency =
 129                                 le64_to_cpu(src->ln_unknown_inconsistency);
 130         dst->ln_unmatched_pairs_repaired =
 131                                 le64_to_cpu(src->ln_unmatched_pairs_repaired);
 132         dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
 133         dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
 134         dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
 135         dst->ln_lost_dirent_repaired =
 136                                 le64_to_cpu(src->ln_lost_dirent_repaired);
 137         dst->ln_striped_dirs_scanned =
 138                                 le64_to_cpu(src->ln_striped_dirs_scanned);
 139         dst->ln_striped_dirs_repaired =
 140                                 le64_to_cpu(src->ln_striped_dirs_repaired);
 141         dst->ln_striped_dirs_failed =
 142                                 le64_to_cpu(src->ln_striped_dirs_failed);
 143         dst->ln_striped_dirs_disabled =
 144                                 le64_to_cpu(src->ln_striped_dirs_disabled);
 145         dst->ln_striped_dirs_skipped =
 146                                 le64_to_cpu(src->ln_striped_dirs_skipped);
 147         dst->ln_striped_shards_scanned =
 148                                 le64_to_cpu(src->ln_striped_shards_scanned);
 149         dst->ln_striped_shards_repaired =
 150                                 le64_to_cpu(src->ln_striped_shards_repaired);
 151         dst->ln_striped_shards_failed =
 152                                 le64_to_cpu(src->ln_striped_shards_failed);
 153         dst->ln_striped_shards_skipped =
 154                                 le64_to_cpu(src->ln_striped_shards_skipped);
 155         dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
 156         dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
 157         dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
 158         dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
 159         dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
 160         dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
 161 }
 162
 163 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
 164                                       struct lfsck_namespace *src)
 165 {
 166         dst->ln_magic = cpu_to_le32(src->ln_magic);
 167         dst->ln_status = cpu_to_le32(src->ln_status);
 168         dst->ln_flags = cpu_to_le32(src->ln_flags);
 169         dst->ln_success_count = cpu_to_le32(src->ln_success_count);
 170         dst->ln_run_time_phase1 = cpu_to_le32(src->ln_run_time_phase1);
 171         dst->ln_run_time_phase2 = cpu_to_le32(src->ln_run_time_phase2);
 172         dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
 173         dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
 174         dst->ln_time_last_checkpoint =
 175                                 cpu_to_le64(src->ln_time_last_checkpoint);
 176         lfsck_position_cpu_to_le(&dst->ln_pos_latest_start,
 177                                  &src->ln_pos_latest_start);
 178         lfsck_position_cpu_to_le(&dst->ln_pos_last_checkpoint,
 179                                  &src->ln_pos_last_checkpoint);
 180         lfsck_position_cpu_to_le(&dst->ln_pos_first_inconsistent,
 181                                  &src->ln_pos_first_inconsistent);
 182         dst->ln_items_checked = cpu_to_le64(src->ln_items_checked);
 183         dst->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
 184         dst->ln_items_failed = cpu_to_le64(src->ln_items_failed);
 185         dst->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
 186         dst->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
 187         dst->ln_objs_repaired_phase2 =
 188                                 cpu_to_le64(src->ln_objs_repaired_phase2);
 189         dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
 190         dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
 191         fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
 192                       &src->ln_fid_latest_scanned_phase2);
 193         dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
 194         dst->ln_linkea_repaired = cpu_to_le64(src->ln_linkea_repaired);
 195         dst->ln_mul_linked_checked = cpu_to_le64(src->ln_mul_linked_checked);
 196         dst->ln_mul_linked_repaired = cpu_to_le64(src->ln_mul_linked_repaired);
 197         dst->ln_unknown_inconsistency =
 198                                 cpu_to_le64(src->ln_unknown_inconsistency);
 199         dst->ln_unmatched_pairs_repaired =
 200                                 cpu_to_le64(src->ln_unmatched_pairs_repaired);
 201         dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
 202         dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
 203         dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
 204         dst->ln_lost_dirent_repaired =
 205                                 cpu_to_le64(src->ln_lost_dirent_repaired);
 206         dst->ln_striped_dirs_scanned =
 207                                 cpu_to_le64(src->ln_striped_dirs_scanned);
 208         dst->ln_striped_dirs_repaired =
 209                                 cpu_to_le64(src->ln_striped_dirs_repaired);
 210         dst->ln_striped_dirs_failed =
 211                                 cpu_to_le64(src->ln_striped_dirs_failed);
 212         dst->ln_striped_dirs_disabled =
 213                                 cpu_to_le64(src->ln_striped_dirs_disabled);
 214         dst->ln_striped_dirs_skipped =
 215                                 cpu_to_le64(src->ln_striped_dirs_skipped);
 216         dst->ln_striped_shards_scanned =
 217                                 cpu_to_le64(src->ln_striped_shards_scanned);
 218         dst->ln_striped_shards_repaired =
 219                                 cpu_to_le64(src->ln_striped_shards_repaired);
 220         dst->ln_striped_shards_failed =
 221                                 cpu_to_le64(src->ln_striped_shards_failed);
 222         dst->ln_striped_shards_skipped =
 223                                 cpu_to_le64(src->ln_striped_shards_skipped);
 224         dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
 225         dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
 226         dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
 227         dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
 228         dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
 229         dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
 230 }
 231
 232 static void lfsck_namespace_record_failure(const struct lu_env *env,
 233                                            struct lfsck_instance *lfsck,
 234                                            struct lfsck_namespace *ns)
 235 {
 236         struct lfsck_position pos;
 237
 238         ns->ln_items_failed++;
 239         lfsck_pos_fill(env, lfsck, &pos, false);
 240         if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
 241             lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
 242                 ns->ln_pos_first_inconsistent = pos;
 243
 244                 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
 245                        "inconsistency at the pos ["LPU64", "DFID", "LPX64"]\n",
 246                        lfsck_lfsck2name(lfsck),
 247                        ns->ln_pos_first_inconsistent.lp_oit_cookie,
 248                        PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
 249                        ns->ln_pos_first_inconsistent.lp_dir_cookie);
 250         }
 251 }
 252
 253 /**
 254  * Load the MDT bitmap from the lfsck_namespace trace file.
 255  *
 256  * \param[in] env       pointer to the thread context
 257  * \param[in] com       pointer to the lfsck component
 258  *
 259  * \retval              0 for success
 260  * \retval              negative error number on failure or data corruption
 261  */
 262 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
 263                                        struct lfsck_component *com)
 264 {
 265         struct dt_object                *obj    = com->lc_obj;
 266         struct lfsck_assistant_data     *lad    = com->lc_data;
 267         struct lfsck_namespace          *ns     = com->lc_file_ram;
 268         cfs_bitmap_t                    *bitmap = lad->lad_bitmap;
 269         ssize_t                          size;
 270         __u32                            nbits;
 271         int                              rc;
 272         ENTRY;
 273
 274         if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
 275             ns->ln_bitmap_size)
 276                 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
 277         else
 278                 nbits = ns->ln_bitmap_size;
 279
 280         if (unlikely(nbits < BITS_PER_LONG))
 281                 nbits = BITS_PER_LONG;
 282
 283         if (nbits > bitmap->size) {
 284                 __u32 new_bits = bitmap->size;
 285                 cfs_bitmap_t *new_bitmap;
 286
 287                 while (new_bits < nbits)
 288                         new_bits <<= 1;
 289
 290                 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
 291                 if (new_bitmap == NULL)
 292                         RETURN(-ENOMEM);
 293
 294                 lad->lad_bitmap = new_bitmap;
 295                 CFS_FREE_BITMAP(bitmap);
 296                 bitmap = new_bitmap;
 297         }
 298
 299         if (ns->ln_bitmap_size == 0) {
 300                 lad->lad_incomplete = 0;
 301                 CFS_RESET_BITMAP(bitmap);
 302
 303                 RETURN(0);
 304         }
 305
 306         size = (ns->ln_bitmap_size + 7) >> 3;
 307         rc = dt_xattr_get(env, obj,
 308                           lfsck_buf_get(env, bitmap->data, size),
 309                           XATTR_NAME_LFSCK_BITMAP, BYPASS_CAPA);
 310         if (rc != size)
 311                 RETURN(rc >= 0 ? -EINVAL : rc);
 312
 313         if (cfs_bitmap_check_empty(bitmap))
 314                 lad->lad_incomplete = 0;
 315         else
 316                 lad->lad_incomplete = 1;
 317
 318         RETURN(0);
 319 }
 320
 321 /**
 322  * Load namespace LFSCK statistics information from the trace file.
 323  *
 324  * For old release (Lustre-2.6 or older), the statistics information was
 325  * stored as XATTR_NAME_LFSCK_NAMESPACE_OLD EA. But in Lustre-2.7, we need
 326  * more statistics information. To avoid confusing old MDT when downgrade,
 327  * Lustre-2.7 stores the namespace LFSCK statistics information as new
 328  * XATTR_NAME_LFSCK_NAMESPACE EA.
 329  *
 330  * \param[in] env       pointer to the thread context
 331  * \param[in] com       pointer to the lfsck component
 332  *
 333  * \retval              0 for success
 334  * \retval              negative error number on failure
 335  */
 336 static int lfsck_namespace_load(const struct lu_env *env,
 337                                 struct lfsck_component *com)
 338 {
 339         int len = com->lc_file_size;
 340         int rc;
 341
 342         rc = dt_xattr_get(env, com->lc_obj,
 343                           lfsck_buf_get(env, com->lc_file_disk, len),
 344                           XATTR_NAME_LFSCK_NAMESPACE, BYPASS_CAPA);
 345         if (rc == len) {
 346                 struct lfsck_namespace *ns = com->lc_file_ram;
 347
 348                 lfsck_namespace_le_to_cpu(ns,
 349                                 (struct lfsck_namespace *)com->lc_file_disk);
 350                 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
 351                         CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
 352                                "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
 353                                ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
 354                         rc = -ESTALE;
 355                 } else {
 356                         rc = 0;
 357                 }
 358         } else if (rc != -ENODATA) {
 359                 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
 360                        "expected = %d: rc = %d\n",
 361                        lfsck_lfsck2name(com->lc_lfsck), len, rc);
 362                 if (rc >= 0)
 363                         rc = -ESTALE;
 364         } else {
 365                 /* Check whether it is old trace file or not.
 366                  * If yes, it should be reset via returning -ESTALE. */
 367                 rc = dt_xattr_get(env, com->lc_obj,
 368                                   lfsck_buf_get(env, com->lc_file_disk, len),
 369                                   XATTR_NAME_LFSCK_NAMESPACE_OLD, BYPASS_CAPA);
 370                 if (rc >= 0)
 371                         rc = -ESTALE;
 372         }
 373
 374         return rc;
 375 }
 376
 377 static int lfsck_namespace_store(const struct lu_env *env,
 378                                  struct lfsck_component *com, bool init)
 379 {
 380         struct dt_object                *obj    = com->lc_obj;
 381         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 382         struct lfsck_namespace          *ns     = com->lc_file_ram;
 383         struct lfsck_assistant_data     *lad    = com->lc_data;
 384         cfs_bitmap_t                    *bitmap = NULL;
 385         struct thandle                  *handle;
 386         __u32                            nbits  = 0;
 387         int                              len    = com->lc_file_size;
 388         int                              rc;
 389 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
 390         struct lu_buf            tbuf   = { &len, sizeof(len) };
 391 #endif
 392         ENTRY;
 393
 394         if (lad != NULL) {
 395                 bitmap = lad->lad_bitmap;
 396                 nbits = bitmap->size;
 397
 398                 LASSERT(nbits > 0);
 399                 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
 400         }
 401
 402         ns->ln_bitmap_size = nbits;
 403         lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
 404                                   ns);
 405         handle = dt_trans_create(env, lfsck->li_bottom);
 406         if (IS_ERR(handle))
 407                 GOTO(log, rc = PTR_ERR(handle));
 408
 409         rc = dt_declare_xattr_set(env, obj,
 410                                   lfsck_buf_get(env, com->lc_file_disk, len),
 411                                   XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
 412         if (rc != 0)
 413                 GOTO(out, rc);
 414
 415         if (bitmap != NULL) {
 416                 rc = dt_declare_xattr_set(env, obj,
 417                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
 418                                 XATTR_NAME_LFSCK_BITMAP, 0, handle);
 419                 if (rc != 0)
 420                         GOTO(out, rc);
 421         }
 422
 423 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
 424         /* To be compatible with old Lustre-2.x MDT (x <= 6), generate dummy
 425          * XATTR_NAME_LFSCK_NAMESPACE_OLD EA, then when downgrade to Lustre-2.x,
 426          * the old LFSCK will find "invalid" XATTR_NAME_LFSCK_NAMESPACE_OLD EA,
 427          * then reset the namespace LFSCK trace file. */
 428         if (init) {
 429                 rc = dt_declare_xattr_set(env, obj, &tbuf,
 430                                           XATTR_NAME_LFSCK_NAMESPACE_OLD,
 431                                           LU_XATTR_CREATE, handle);
 432                 if (rc != 0)
 433                         GOTO(out, rc);
 434         }
 435 #endif
 436
 437         rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
 438         if (rc != 0)
 439                 GOTO(out, rc);
 440
 441         rc = dt_xattr_set(env, obj,
 442                           lfsck_buf_get(env, com->lc_file_disk, len),
 443                           XATTR_NAME_LFSCK_NAMESPACE, 0, handle, BYPASS_CAPA);
 444         if (rc == 0 && bitmap != NULL)
 445                 rc = dt_xattr_set(env, obj,
 446                                   lfsck_buf_get(env, bitmap->data, nbits >> 3),
 447                                   XATTR_NAME_LFSCK_BITMAP, 0, handle,
 448                                   BYPASS_CAPA);
 449
 450 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
 451         if (rc == 0 && init)
 452                 rc = dt_xattr_set(env, obj, &tbuf,
 453                                   XATTR_NAME_LFSCK_NAMESPACE_OLD,
 454                                   LU_XATTR_CREATE, handle, BYPASS_CAPA);
 455 #endif
 456
 457         GOTO(out, rc);
 458
 459 out:
 460         dt_trans_stop(env, lfsck->li_bottom, handle);
 461
 462 log:
 463         if (rc != 0)
 464                 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
 465                        lfsck_lfsck2name(lfsck), rc);
 466         return rc;
 467 }
 468
 469 static int lfsck_namespace_init(const struct lu_env *env,
 470                                 struct lfsck_component *com)
 471 {
 472         struct lfsck_namespace *ns = com->lc_file_ram;
 473         int rc;
 474
 475         memset(ns, 0, sizeof(*ns));
 476         ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
 477         ns->ln_status = LS_INIT;
 478         down_write(&com->lc_sem);
 479         rc = lfsck_namespace_store(env, com, true);
 480         up_write(&com->lc_sem);
 481         return rc;
 482 }
 483
 484 /**
 485  * Update the namespace LFSCK trace file for the given @fid
 486  *
 487  * \param[in] env       pointer to the thread context
 488  * \param[in] com       pointer to the lfsck component
 489  * \param[in] fid       the fid which flags to be updated in the lfsck
 490  *                      trace file
 491  * \param[in] add       true if add new flags, otherwise remove flags
 492  *
 493  * \retval              0 for succeed or nothing to be done
 494  * \retval              negative error number on failure
 495  */
 496 int lfsck_namespace_trace_update(const struct lu_env *env,
 497                                  struct lfsck_component *com,
 498                                  const struct lu_fid *fid,
 499                                  const __u8 flags, bool add)
 500 {
 501         struct lfsck_instance   *lfsck  = com->lc_lfsck;
 502         struct dt_object        *obj    = com->lc_obj;
 503         struct lu_fid           *key    = &lfsck_env_info(env)->lti_fid3;
 504         struct dt_device        *dev    = lfsck->li_bottom;
 505         struct thandle          *th     = NULL;
 506         int                      rc     = 0;
 507         __u8                     old    = 0;
 508         __u8                     new    = 0;
 509         ENTRY;
 510
 511         LASSERT(flags != 0);
 512
 513         down_write(&com->lc_sem);
 514         fid_cpu_to_be(key, fid);
 515         rc = dt_lookup(env, obj, (struct dt_rec *)&old,
 516                        (const struct dt_key *)key, BYPASS_CAPA);
 517         if (rc == -ENOENT) {
 518                 if (!add)
 519                         GOTO(unlock, rc = 0);
 520
 521                 old = 0;
 522                 new = flags;
 523         } else if (rc == 0) {
 524                 if (add) {
 525                         if ((old & flags) == flags)
 526                                 GOTO(unlock, rc = 0);
 527
 528                         new = old | flags;
 529                 } else {
 530                         if ((old & flags) == 0)
 531                                 GOTO(unlock, rc = 0);
 532
 533                         new = old & ~flags;
 534                 }
 535         } else {
 536                 GOTO(log, rc);
 537         }
 538
 539         th = dt_trans_create(env, dev);
 540         if (IS_ERR(th))
 541                 GOTO(log, rc = PTR_ERR(th));
 542
 543         if (old != 0) {
 544                 rc = dt_declare_delete(env, obj,
 545                                        (const struct dt_key *)key, th);
 546                 if (rc != 0)
 547                         GOTO(log, rc);
 548         }
 549
 550         if (new != 0) {
 551                 rc = dt_declare_insert(env, obj,
 552                                        (const struct dt_rec *)&new,
 553                                        (const struct dt_key *)key, th);
 554                 if (rc != 0)
 555                         GOTO(log, rc);
 556         }
 557
 558         rc = dt_trans_start_local(env, dev, th);
 559         if (rc != 0)
 560                 GOTO(log, rc);
 561
 562         if (old != 0) {
 563                 rc = dt_delete(env, obj, (const struct dt_key *)key,
 564                                th, BYPASS_CAPA);
 565                 if (rc != 0)
 566                         GOTO(log, rc);
 567         }
 568
 569         if (new != 0) {
 570                 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
 571                                (const struct dt_key *)key, th, BYPASS_CAPA, 1);
 572                 if (rc != 0)
 573                         GOTO(log, rc);
 574         }
 575
 576         GOTO(log, rc);
 577
 578 log:
 579         if (th != NULL && !IS_ERR(th))
 580                 dt_trans_stop(env, dev, th);
 581
 582         CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
 583                "trace file, flags %x, old %x, new %x: rc = %d\n",
 584                lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
 585                (__u32)flags, (__u32)old, (__u32)new, rc);
 586
 587 unlock:
 588         up_write(&com->lc_sem);
 589
 590         return rc;
 591 }
 592
 593 int lfsck_namespace_check_exist(const struct lu_env *env,
 594                                 struct dt_object *dir,
 595                                 struct dt_object *obj, const char *name)
 596 {
 597         struct lu_fid    *fid = &lfsck_env_info(env)->lti_fid;
 598         int               rc;
 599         ENTRY;
 600
 601         if (unlikely(lfsck_is_dead_obj(obj)))
 602                 RETURN(LFSCK_NAMEENTRY_DEAD);
 603
 604         rc = dt_lookup(env, dir, (struct dt_rec *)fid,
 605                        (const struct dt_key *)name, BYPASS_CAPA);
 606         if (rc == -ENOENT)
 607                 RETURN(LFSCK_NAMEENTRY_REMOVED);
 608
 609         if (rc < 0)
 610                 RETURN(rc);
 611
 612         if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
 613                 RETURN(LFSCK_NAMEENTRY_RECREATED);
 614
 615         RETURN(0);
 616 }
 617
 618 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
 619                                             struct dt_object *obj,
 620                                             struct thandle *handle)
 621 {
 622         int rc;
 623
 624         /* For destroying all invalid linkEA entries. */
 625         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
 626         if (rc != 0)
 627                 return rc;
 628
 629         /* For insert new linkEA entry. */
 630         rc = dt_declare_xattr_set(env, obj,
 631                         lfsck_buf_get_const(env, NULL, DEFAULT_LINKEA_SIZE),
 632                         XATTR_NAME_LINK, 0, handle);
 633         return rc;
 634 }
 635
 636 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
 637                        struct linkea_data *ldata)
 638 {
 639         int rc;
 640
 641         if (ldata->ld_buf->lb_buf == NULL)
 642                 return -ENOMEM;
 643
 644         if (!dt_object_exists(obj))
 645                 return -ENOENT;
 646
 647         rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK, BYPASS_CAPA);
 648         if (rc == -ERANGE) {
 649                 /* Buf was too small, figure out what we need. */
 650                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK,
 651                                   BYPASS_CAPA);
 652                 if (rc <= 0)
 653                         return rc;
 654
 655                 lu_buf_realloc(ldata->ld_buf, rc);
 656                 if (ldata->ld_buf->lb_buf == NULL)
 657                         return -ENOMEM;
 658
 659                 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK,
 660                                   BYPASS_CAPA);
 661         }
 662
 663         if (rc > 0)
 664                 rc = linkea_init(ldata);
 665
 666         return rc;
 667 }
 668
 669 /**
 670  * Remove linkEA for the given object.
 671  *
 672  * The caller should take the ldlm lock before the calling.
 673  *
 674  * \param[in] env       pointer to the thread context
 675  * \param[in] com       pointer to the lfsck component
 676  * \param[in] obj       pointer to the dt_object to be handled
 677  *
 678  * \retval              0 for repaired cases
 679  * \retval              negative error number on failure
 680  */
 681 static int lfsck_namespace_links_remove(const struct lu_env *env,
 682                                         struct lfsck_component *com,
 683                                         struct dt_object *obj)
 684 {
 685         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 686         struct dt_device                *dev    = lfsck->li_bottom;
 687         struct thandle                  *th     = NULL;
 688         int                              rc     = 0;
 689         ENTRY;
 690
 691         LASSERT(dt_object_remote(obj) == 0);
 692
 693         th = dt_trans_create(env, dev);
 694         if (IS_ERR(th))
 695                 GOTO(log, rc = PTR_ERR(th));
 696
 697         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
 698         if (rc != 0)
 699                 GOTO(stop, rc);
 700
 701         rc = dt_trans_start_local(env, dev, th);
 702         if (rc != 0)
 703                 GOTO(stop, rc);
 704
 705         dt_write_lock(env, obj, 0);
 706         if (unlikely(lfsck_is_dead_obj(obj)))
 707                 GOTO(unlock, rc = -ENOENT);
 708
 709         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
 710                 GOTO(unlock, rc = 0);
 711
 712         rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th, BYPASS_CAPA);
 713
 714         GOTO(unlock, rc);
 715
 716 unlock:
 717         dt_write_unlock(env, obj);
 718
 719 stop:
 720         dt_trans_stop(env, dev, th);
 721
 722 log:
 723         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
 724                "for the object "DFID": rc = %d\n",
 725                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
 726
 727         if (rc == 0) {
 728                 struct lfsck_namespace *ns = com->lc_file_ram;
 729
 730                 ns->ln_flags |= LF_INCONSISTENT;
 731         }
 732
 733         return rc;
 734 }
 735
 736 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
 737                              struct linkea_data *ldata, struct thandle *handle)
 738 {
 739         const struct lu_buf *buf = lfsck_buf_get_const(env,
 740                                                        ldata->ld_buf->lb_buf,
 741                                                        ldata->ld_leh->leh_len);
 742
 743         return dt_xattr_set(env, obj, buf, XATTR_NAME_LINK, 0, handle,
 744                             BYPASS_CAPA);
 745 }
 746
 747 static void lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
 748                                                 struct lu_name *cname,
 749                                                 struct lu_fid *pfid,
 750                                                 char *buf)
 751 {
 752         linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
 753         /* To guarantee the 'name' is terminated with '0'. */
 754         memcpy(buf, cname->ln_name, cname->ln_namelen);
 755         buf[cname->ln_namelen] = 0;
 756         cname->ln_name = buf;
 757 }
 758
 759 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
 760                                                struct lu_name *cname,
 761                                                struct lu_fid *pfid,
 762                                                bool remove)
 763 {
 764         struct link_ea_entry    *oldlee;
 765         int                      oldlen;
 766         int                      repeated = 0;
 767
 768         oldlee = ldata->ld_lee;
 769         oldlen = ldata->ld_reclen;
 770         linkea_next_entry(ldata);
 771         while (ldata->ld_lee != NULL) {
 772                 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
 773                                    ldata->ld_lee->lee_reclen[1];
 774                 if (unlikely(ldata->ld_reclen == oldlen &&
 775                              memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
 776                         repeated++;
 777                         if (!remove)
 778                                 break;
 779
 780                         linkea_del_buf(ldata, cname);
 781                 } else {
 782                         linkea_next_entry(ldata);
 783                 }
 784         }
 785         ldata->ld_lee = oldlee;
 786         ldata->ld_reclen = oldlen;
 787
 788         return repeated;
 789 }
 790
 791 /**
 792  * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
 793  *
 794  * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
 795  * with the given type to generate the name, the detailed rules for name
 796  * have been described as following.
 797  *
 798  * The function also generates the linkEA corresponding to the name entry
 799  * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
 800  *
 801  * \param[in] env       pointer to the thread context
 802  * \param[in] com       pointer to the lfsck component
 803  * \param[in] orphan    pointer to the orphan MDT-object
 804  * \param[in] infix     additional information for the orphan name, such as
 805  *                      the FID for original
 806  * \param[in] type      the type for describing why the orphan MDT-object is
 807  *                      created. The rules are as following:
 808  *
 809  *  type "D":           The MDT-object is a directory, it may knows its parent
 810  *                      but because there is no valid linkEA, the LFSCK cannot
 811  *                      know where to put it back to the namespace.
 812  *  type "O":           The MDT-object has no linkEA, and there is no name
 813  *                      entry that references the MDT-object.
 814  *
 815  *  type "S":           The orphan MDT-object is a shard of a striped directory
 816  *
 817  * \see lfsck_layout_recreate_parent() for more types.
 818  *
 819  * The orphan name will be like:
 820  * ${FID}-${infix}-${type}-${conflict_version}
 821  *
 822  * \param[out] count    if some others inserted some linkEA entries by race,
 823  *                      then return the linkEA entries count.
 824  *
 825  * \retval              positive number for repaired cases
 826  * \retval              0 if needs to repair nothing
 827  * \retval              negative error number on failure
 828  */
 829 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
 830                                          struct lfsck_component *com,
 831                                          struct dt_object *orphan,
 832                                          const char *infix, const char *type,
 833                                          int *count)
 834 {
 835         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 836         struct lu_name                  *cname  = &info->lti_name;
 837         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
 838         struct lu_fid                   *tfid   = &info->lti_fid5;
 839         struct lu_attr                  *la     = &info->lti_la3;
 840         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
 841         const struct lu_fid             *pfid;
 842         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 843         struct dt_device                *dev    = lfsck->li_bottom;
 844         struct dt_object                *parent;
 845         struct thandle                  *th     = NULL;
 846         struct lustre_handle             plh    = { 0 };
 847         struct lustre_handle             clh    = { 0 };
 848         struct linkea_data               ldata  = { 0 };
 849         struct lu_buf                    linkea_buf;
 850         int                              namelen;
 851         int                              idx    = 0;
 852         int                              rc     = 0;
 853         bool                             exist  = false;
 854         ENTRY;
 855
 856         cname->ln_name = NULL;
 857         /* Create .lustre/lost+found/MDTxxxx when needed. */
 858         if (unlikely(lfsck->li_lpf_obj == NULL)) {
 859                 rc = lfsck_create_lpf(env, lfsck);
 860                 if (rc != 0)
 861                         GOTO(log, rc);
 862         }
 863
 864         parent = lfsck->li_lpf_obj;
 865         pfid = lfsck_dto2fid(parent);
 866
 867         /* Hold update lock on the parent to prevent others to access. */
 868         rc = lfsck_ibits_lock(env, lfsck, parent, &plh,
 869                               MDS_INODELOCK_UPDATE, LCK_EX);
 870         if (rc != 0)
 871                 GOTO(log, rc);
 872
 873         do {
 874                 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
 875                                    PFID(cfid), infix, type, idx++);
 876                 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
 877                                (const struct dt_key *)info->lti_key,
 878                                BYPASS_CAPA);
 879                 if (rc != 0 && rc != -ENOENT)
 880                         GOTO(log, rc);
 881
 882                 if (unlikely(rc == 0 && lu_fid_eq(cfid, tfid)))
 883                         exist = true;
 884         } while (rc == 0 && !exist);
 885
 886         cname->ln_name = info->lti_key;
 887         cname->ln_namelen = namelen;
 888         rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
 889         if (rc != 0)
 890                 GOTO(log, rc);
 891
 892         rc = linkea_add_buf(&ldata, cname, pfid);
 893         if (rc != 0)
 894                 GOTO(log, rc);
 895
 896         rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
 897                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP,
 898                               LCK_EX);
 899         if (rc != 0)
 900                 GOTO(log, rc);
 901
 902         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
 903                        ldata.ld_leh->leh_len);
 904         th = dt_trans_create(env, dev);
 905         if (IS_ERR(th))
 906                 GOTO(log, rc = PTR_ERR(th));
 907
 908         if (S_ISDIR(lfsck_object_type(orphan))) {
 909                 rc = dt_declare_delete(env, orphan,
 910                                        (const struct dt_key *)dotdot, th);
 911                 if (rc != 0)
 912                         GOTO(stop, rc);
 913
 914                 rec->rec_type = S_IFDIR;
 915                 rec->rec_fid = pfid;
 916                 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
 917                                        (const struct dt_key *)dotdot, th);
 918                 if (rc != 0)
 919                         GOTO(stop, rc);
 920         }
 921
 922         rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
 923                                   XATTR_NAME_LINK, 0, th);
 924         if (rc != 0)
 925                 GOTO(stop, rc);
 926
 927         if (!exist) {
 928                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
 929                 rec->rec_fid = cfid;
 930                 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
 931                                        (const struct dt_key *)cname->ln_name,
 932                                        th);
 933                 if (rc != 0)
 934                         GOTO(stop, rc);
 935
 936                 if (S_ISDIR(rec->rec_type)) {
 937                         rc = dt_declare_ref_add(env, parent, th);
 938                         if (rc != 0)
 939                                 GOTO(stop, rc);
 940                 }
 941         }
 942
 943         memset(la, 0, sizeof(*la));
 944         la->la_ctime = cfs_time_current_sec();
 945         la->la_valid = LA_CTIME;
 946         rc = dt_declare_attr_set(env, orphan, la, th);
 947         if (rc != 0)
 948                 GOTO(stop, rc);
 949
 950         rc = dt_trans_start_local(env, dev, th);
 951         if (rc != 0)
 952                 GOTO(stop, rc);
 953
 954         dt_write_lock(env, orphan, 0);
 955         rc = lfsck_links_read(env, orphan, &ldata);
 956         if (likely((rc == -ENODATA) || (rc == -EINVAL) ||
 957                    (rc == 0 && ldata.ld_leh->leh_reccount == 0))) {
 958                 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
 959                         GOTO(unlock, rc = 1);
 960
 961                 if (S_ISDIR(lfsck_object_type(orphan))) {
 962                         rc = dt_delete(env, orphan,
 963                                        (const struct dt_key *)dotdot, th,
 964                                        BYPASS_CAPA);
 965                         if (rc != 0)
 966                                 GOTO(unlock, rc);
 967
 968                         rec->rec_type = S_IFDIR;
 969                         rec->rec_fid = pfid;
 970                         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
 971                                        (const struct dt_key *)dotdot, th,
 972                                        BYPASS_CAPA, 1);
 973                         if (rc != 0)
 974                                 GOTO(unlock, rc);
 975                 }
 976
 977                 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
 978                                   th, BYPASS_CAPA);
 979         } else {
 980                 if (rc == 0 && count != NULL)
 981                         *count = ldata.ld_leh->leh_reccount;
 982
 983                 GOTO(unlock, rc);
 984         }
 985         dt_write_unlock(env, orphan);
 986
 987         if (rc == 0 && !exist) {
 988                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
 989                 rec->rec_fid = cfid;
 990                 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
 991                                (const struct dt_key *)cname->ln_name,
 992                                th, BYPASS_CAPA, 1);
 993                 if (rc == 0 && S_ISDIR(rec->rec_type)) {
 994                         dt_write_lock(env, parent, 0);
 995                         rc = dt_ref_add(env, parent, th);
 996                         dt_write_unlock(env, parent);
 997                 }
 998         }
 999
1000         if (rc == 0)
1001                 rc = dt_attr_set(env, orphan, la, th, BYPASS_CAPA);
1002
1003         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1004
1005 unlock:
1006         dt_write_unlock(env, orphan);
1007
1008 stop:
1009         dt_trans_stop(env, dev, th);
1010
1011 log:
1012         lfsck_ibits_unlock(&clh, LCK_EX);
1013         lfsck_ibits_unlock(&plh, LCK_EX);
1014         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1015                "object "DFID", name = %s: rc = %d\n",
1016                lfsck_lfsck2name(lfsck), PFID(cfid),
1017                cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1018
1019         if (rc != 0) {
1020                 struct lfsck_namespace *ns = com->lc_file_ram;
1021
1022                 ns->ln_flags |= LF_INCONSISTENT;
1023         }
1024
1025         return rc;
1026 }
1027
1028 /**
1029  * Add the specified name entry back to namespace.
1030  *
1031  * If there is a linkEA entry that back references a name entry under
1032  * some parent directory, but such parent directory does not have the
1033  * claimed name entry. On the other hand, the linkEA entries count is
1034  * not larger than the MDT-object's hard link count. Under such case,
1035  * it is quite possible that the name entry is lost. Then the LFSCK
1036  * should add the name entry back to the namespace.
1037  *
1038  * \param[in] env       pointer to the thread context
1039  * \param[in] com       pointer to the lfsck component
1040  * \param[in] parent    pointer to the directory under which the name entry
1041  *                      will be inserted into
1042  * \param[in] child     pointer to the object referenced by the name entry
1043  *                      that to be inserted into the parent
1044  * \param[in] name      the name for the child in the parent directory
1045  *
1046  * \retval              positive number for repaired cases
1047  * \retval              0 if nothing to be repaired
1048  * \retval              negative error number on failure
1049  */
1050 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1051                                          struct lfsck_component *com,
1052                                          struct dt_object *parent,
1053                                          struct dt_object *child,
1054                                          const char *name)
1055 {
1056         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1057         struct lu_attr                  *la     = &info->lti_la;
1058         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1059         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1060         struct dt_device                *dev    = lfsck->li_next;
1061         struct thandle                  *th     = NULL;
1062         struct lustre_handle             lh     = { 0 };
1063         int                              rc     = 0;
1064         ENTRY;
1065
1066         if (unlikely(!dt_try_as_dir(env, parent)))
1067                 GOTO(log, rc = -ENOTDIR);
1068
1069         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1070                 GOTO(log, rc = 1);
1071
1072         /* Hold update lock on the parent to prevent others to access. */
1073         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
1074                               MDS_INODELOCK_UPDATE, LCK_EX);
1075         if (rc != 0)
1076                 GOTO(log, rc);
1077
1078         th = dt_trans_create(env, dev);
1079         if (IS_ERR(th))
1080                 GOTO(unlock, rc = PTR_ERR(th));
1081
1082         rec->rec_type = lfsck_object_type(child) & S_IFMT;
1083         rec->rec_fid = lfsck_dto2fid(child);
1084         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1085                                (const struct dt_key *)name, th);
1086         if (rc != 0)
1087                 GOTO(stop, rc);
1088
1089         if (S_ISDIR(rec->rec_type)) {
1090                 rc = dt_declare_ref_add(env, parent, th);
1091                 if (rc != 0)
1092                         GOTO(stop, rc);
1093         }
1094
1095         memset(la, 0, sizeof(*la));
1096         la->la_ctime = cfs_time_current_sec();
1097         la->la_valid = LA_CTIME;
1098         rc = dt_declare_attr_set(env, parent, la, th);
1099         if (rc != 0)
1100                 GOTO(stop, rc);
1101
1102         rc = dt_declare_attr_set(env, child, la, th);
1103         if (rc != 0)
1104                 GOTO(stop, rc);
1105
1106         rc = dt_trans_start_local(env, dev, th);
1107         if (rc != 0)
1108                 GOTO(stop, rc);
1109
1110         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1111                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1112         if (rc != 0)
1113                 GOTO(stop, rc);
1114
1115         if (S_ISDIR(rec->rec_type)) {
1116                 dt_write_lock(env, parent, 0);
1117                 rc = dt_ref_add(env, parent, th);
1118                 dt_write_unlock(env, parent);
1119                 if (rc != 0)
1120                         GOTO(stop, rc);
1121         }
1122
1123         la->la_ctime = cfs_time_current_sec();
1124         rc = dt_attr_set(env, parent, la, th, BYPASS_CAPA);
1125         if (rc != 0)
1126                 GOTO(stop, rc);
1127
1128         rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
1129
1130         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1131
1132 stop:
1133         dt_trans_stop(env, dev, th);
1134
1135 unlock:
1136         lfsck_ibits_unlock(&lh, LCK_EX);
1137
1138 log:
1139         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1140                "the name %s and type %o to the parent "DFID": rc = %d\n",
1141                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), name,
1142                lfsck_object_type(child) & S_IFMT,
1143                PFID(lfsck_dto2fid(parent)), rc);
1144
1145         if (rc != 0) {
1146                 struct lfsck_namespace *ns = com->lc_file_ram;
1147
1148                 ns->ln_flags |= LF_INCONSISTENT;
1149                 if (rc > 0)
1150                         ns->ln_lost_dirent_repaired++;
1151         }
1152
1153         return rc;
1154 }
1155
1156 /**
1157  * Create the specified orphan MDT-object on remote MDT.
1158  *
1159  * The LFSCK instance on this MDT will send LFSCK RPC to remote MDT to
1160  * ask the remote LFSCK instance to create the specified orphan object
1161  * under .lustre/lost+found/MDTxxxx/ directory with the name:
1162  * ${FID}-P-${conflict_version}.
1163  *
1164  * \param[in] env       pointer to the thread context
1165  * \param[in] com       pointer to the lfsck component
1166  * \param[in] orphan    pointer to the orphan MDT-object
1167  * \param[in] type      the orphan's type to be created
1168  *
1169  *  type "P":           The orphan object to be created was a parent directory
1170  *                      of some MDT-object which linkEA shows that the @orphan
1171  *                      object is missing.
1172  *
1173  * \see lfsck_layout_recreate_parent() for more types.
1174  *
1175  * \param[in] lmv       pointer to master LMV EA that will be set to the orphan
1176  *
1177  * \retval              positive number for repaired cases
1178  * \retval              0 if needs to repair nothing
1179  * \retval              negative error number on failure
1180  */
1181 static int lfsck_namespace_create_orphan_remote(const struct lu_env *env,
1182                                                 struct lfsck_component *com,
1183                                                 struct dt_object *orphan,
1184                                                 __u32 type,
1185                                                 struct lmv_mds_md_v1 *lmv)
1186 {
1187         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1188         struct lfsck_request            *lr     = &info->lti_lr;
1189         struct lu_seq_range             *range  = &info->lti_range;
1190         const struct lu_fid             *fid    = lfsck_dto2fid(orphan);
1191         struct lfsck_namespace          *ns     = com->lc_file_ram;
1192         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1193         struct seq_server_site          *ss     =
1194                         lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
1195         struct lfsck_tgt_desc           *ltd    = NULL;
1196         struct ptlrpc_request           *req    = NULL;
1197         int                              rc;
1198         ENTRY;
1199
1200         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1201                 GOTO(out, rc = 1);
1202
1203         fld_range_set_mdt(range);
1204         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
1205         if (rc != 0)
1206                 GOTO(out, rc);
1207
1208         ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, range->lsr_index);
1209         if (ltd == NULL) {
1210                 ns->ln_flags |= LF_INCOMPLETE;
1211
1212                 GOTO(out, rc = -ENODEV);
1213         }
1214
1215         req = ptlrpc_request_alloc(class_exp2cliimp(ltd->ltd_exp),
1216                                    &RQF_LFSCK_NOTIFY);
1217         if (req == NULL)
1218                 GOTO(out, rc = -ENOMEM);
1219
1220         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
1221         if (rc != 0) {
1222                 ptlrpc_request_free(req);
1223
1224                 GOTO(out, rc);
1225         }
1226
1227         lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
1228         memset(lr, 0, sizeof(*lr));
1229         lr->lr_event = LE_CREATE_ORPHAN;
1230         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1231         lr->lr_active = LFSCK_TYPE_NAMESPACE;
1232         lr->lr_fid = *fid;
1233         lr->lr_type = type;
1234         if (lmv != NULL) {
1235                 lr->lr_hash_type = lmv->lmv_hash_type;
1236                 lr->lr_stripe_count = lmv->lmv_stripe_count;
1237                 lr->lr_layout_version = lmv->lmv_layout_version;
1238                 memcpy(lr->lr_pool_name, lmv->lmv_pool_name,
1239                        sizeof(lr->lr_pool_name));
1240         }
1241
1242         ptlrpc_request_set_replen(req);
1243         rc = ptlrpc_queue_wait(req);
1244         ptlrpc_req_finished(req);
1245
1246         if (rc == 0) {
1247                 orphan->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
1248                 rc = 1;
1249         } else if (rc == -EEXIST) {
1250                 orphan->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
1251                 rc = 0;
1252         }
1253
1254         GOTO(out, rc);
1255
1256 out:
1257         CDEBUG(D_LFSCK, "%s: namespace LFSCK create object "
1258                DFID" on the MDT %x remotely: rc = %d\n",
1259                lfsck_lfsck2name(lfsck), PFID(fid),
1260                ltd != NULL ? ltd->ltd_index : -1, rc);
1261
1262         if (ltd != NULL)
1263                 lfsck_tgt_put(ltd);
1264
1265         return rc;
1266 }
1267
1268 /**
1269  * Create the specified orphan MDT-object locally.
1270  *
1271  * For the case that the parent MDT-object stored in some MDT-object's
1272  * linkEA entry is lost, the LFSCK will re-create the parent object as
1273  * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1274  * with the name ${FID}-P-${conflict_version}.
1275  *
1276  * \param[in] env       pointer to the thread context
1277  * \param[in] com       pointer to the lfsck component
1278  * \param[in] orphan    pointer to the orphan MDT-object to be created
1279  * \param[in] type      the orphan's type to be created
1280  *
1281  *  type "P":           The orphan object to be created was a parent directory
1282  *                      of some MDT-object which linkEA shows that the @orphan
1283  *                      object is missing.
1284  *
1285  * \see lfsck_layout_recreate_parent() for more types.
1286  *
1287  * \param[in] lmv       pointer to master LMV EA that will be set to the orphan
1288  *
1289  * \retval              positive number for repaired cases
1290  * \retval              negative error number on failure
1291  */
1292 static int lfsck_namespace_create_orphan_local(const struct lu_env *env,
1293                                                struct lfsck_component *com,
1294                                                struct dt_object *orphan,
1295                                                __u32 type,
1296                                                struct lmv_mds_md_v1 *lmv)
1297 {
1298         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1299         struct lu_attr                  *la     = &info->lti_la;
1300         struct dt_allocation_hint       *hint   = &info->lti_hint;
1301         struct dt_object_format         *dof    = &info->lti_dof;
1302         struct lu_name                  *cname  = &info->lti_name2;
1303         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1304         struct lu_fid                   *tfid   = &info->lti_fid;
1305         struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
1306         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
1307         const struct lu_fid             *pfid;
1308         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1309         struct dt_device                *dev    = lfsck->li_bottom;
1310         struct dt_object                *parent = NULL;
1311         struct dt_object                *child  = NULL;
1312         struct thandle                  *th     = NULL;
1313         struct lustre_handle             lh     = { 0 };
1314         struct linkea_data               ldata  = { 0 };
1315         struct lu_buf                    linkea_buf;
1316         struct lu_buf                    lmv_buf;
1317         char                             name[32];
1318         int                              namelen;
1319         int                              idx    = 0;
1320         int                              rc     = 0;
1321         ENTRY;
1322
1323         LASSERT(!dt_object_exists(orphan));
1324         LASSERT(!dt_object_remote(orphan));
1325
1326         /* @orphan maybe not attached to lfsck->li_bottom */
1327         child = lfsck_object_find_by_dev(env, dev, cfid);
1328         if (IS_ERR(child))
1329                 GOTO(log, rc = PTR_ERR(child));
1330
1331         cname->ln_name = NULL;
1332         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1333                 GOTO(log, rc = 1);
1334
1335         /* Create .lustre/lost+found/MDTxxxx when needed. */
1336         if (unlikely(lfsck->li_lpf_obj == NULL)) {
1337                 rc = lfsck_create_lpf(env, lfsck);
1338                 if (rc != 0)
1339                         GOTO(log, rc);
1340         }
1341
1342         parent = lfsck->li_lpf_obj;
1343         pfid = lfsck_dto2fid(parent);
1344
1345         /* Hold update lock on the parent to prevent others to access. */
1346         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
1347                               MDS_INODELOCK_UPDATE, LCK_EX);
1348         if (rc != 0)
1349                 GOTO(log, rc);
1350
1351         do {
1352                 namelen = snprintf(name, 31, DFID"-P-%d",
1353                                    PFID(cfid), idx++);
1354                 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
1355                                (const struct dt_key *)name, BYPASS_CAPA);
1356                 if (rc != 0 && rc != -ENOENT)
1357                         GOTO(unlock1, rc);
1358         } while (rc == 0);
1359
1360         cname->ln_name = name;
1361         cname->ln_namelen = namelen;
1362
1363         memset(la, 0, sizeof(*la));
1364         la->la_mode = type | (S_ISDIR(type) ? 0700 : 0600);
1365         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1366                        LA_ATIME | LA_MTIME | LA_CTIME;
1367
1368         child->do_ops->do_ah_init(env, hint, parent, child,
1369                                   la->la_mode & S_IFMT);
1370
1371         memset(dof, 0, sizeof(*dof));
1372         dof->dof_type = dt_mode_to_dft(type);
1373
1374         rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
1375         if (rc != 0)
1376                 GOTO(unlock1, rc);
1377
1378         rc = linkea_add_buf(&ldata, cname, pfid);
1379         if (rc != 0)
1380                 GOTO(unlock1, rc);
1381
1382         th = dt_trans_create(env, dev);
1383         if (IS_ERR(th))
1384                 GOTO(unlock1, rc = PTR_ERR(th));
1385
1386         rc = dt_declare_create(env, child, la, hint, dof, th);
1387         if (rc == 0 && S_ISDIR(type))
1388                 rc = dt_declare_ref_add(env, child, th);
1389
1390         if (rc != 0)
1391                 GOTO(stop, rc);
1392
1393         if (lmv != NULL) {
1394                 lmv->lmv_magic = LMV_MAGIC;
1395                 lmv->lmv_master_mdt_index = lfsck_dev_idx(dev);
1396                 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1397                 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1398                 rc = dt_declare_xattr_set(env, child, &lmv_buf,
1399                                           XATTR_NAME_LMV, 0, th);
1400                 if (rc != 0)
1401                         GOTO(stop, rc);
1402         }
1403
1404         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1405                        ldata.ld_leh->leh_len);
1406         rc = dt_declare_xattr_set(env, child, &linkea_buf,
1407                                   XATTR_NAME_LINK, 0, th);
1408         if (rc != 0)
1409                 GOTO(stop, rc);
1410
1411         rec->rec_type = type;
1412         rec->rec_fid = cfid;
1413         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1414                                (const struct dt_key *)name, th);
1415         if (rc == 0 && S_ISDIR(type))
1416                 rc = dt_declare_ref_add(env, parent, th);
1417
1418         if (rc != 0)
1419                 GOTO(stop, rc);
1420
1421         rc = dt_trans_start_local(env, dev, th);
1422         if (rc != 0)
1423                 GOTO(stop, rc);
1424
1425         dt_write_lock(env, child, 0);
1426         rc = dt_create(env, child, la, hint, dof, th);
1427         if (rc != 0)
1428                 GOTO(unlock2, rc);
1429
1430         if (S_ISDIR(type)) {
1431                 if (unlikely(!dt_try_as_dir(env, child)))
1432                         GOTO(unlock2, rc = -ENOTDIR);
1433
1434                 rec->rec_type = S_IFDIR;
1435                 rec->rec_fid = cfid;
1436                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
1437                                (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
1438                 if (rc != 0)
1439                         GOTO(unlock2, rc);
1440
1441                 rec->rec_fid = pfid;
1442                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
1443                                (const struct dt_key *)dotdot, th,
1444                                BYPASS_CAPA, 1);
1445                 if (rc != 0)
1446                         GOTO(unlock2, rc);
1447
1448                 rc = dt_ref_add(env, child, th);
1449                 if (rc != 0)
1450                         GOTO(unlock2, rc);
1451         }
1452
1453         if (lmv != NULL) {
1454                 rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV, 0,
1455                                   th, BYPASS_CAPA);
1456                 if (rc != 0)
1457                         GOTO(unlock2, rc);
1458         }
1459
1460         rc = dt_xattr_set(env, child, &linkea_buf,
1461                           XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1462         dt_write_unlock(env, child);
1463         if (rc != 0)
1464                 GOTO(stop, rc);
1465
1466         rec->rec_type = type;
1467         rec->rec_fid = cfid;
1468         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1469                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1470         if (rc == 0 && S_ISDIR(type)) {
1471                 dt_write_lock(env, parent, 0);
1472                 rc = dt_ref_add(env, parent, th);
1473                 dt_write_unlock(env, parent);
1474         }
1475
1476         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1477
1478 unlock2:
1479         dt_write_unlock(env, child);
1480
1481 stop:
1482         dt_trans_stop(env, dev, th);
1483
1484 unlock1:
1485         lfsck_ibits_unlock(&lh, LCK_EX);
1486
1487 log:
1488         CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan locally for "
1489                "the object "DFID", name = %s, type %o: rc = %d\n",
1490                lfsck_lfsck2name(lfsck), PFID(cfid),
1491                cname->ln_name != NULL ? cname->ln_name : "<NULL>", type, rc);
1492
1493         if (child != NULL && !IS_ERR(child))
1494                 lfsck_object_put(env, child);
1495
1496         return rc;
1497 }
1498
1499 /**
1500  * Create the specified orphan MDT-object.
1501  *
1502  * For the case that the parent MDT-object stored in some MDT-object's
1503  * linkEA entry is lost, the LFSCK will re-create the parent object as
1504  * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1505  * with the name: ${FID}-P-${conflict_version}.
1506  *
1507  * \param[in] env       pointer to the thread context
1508  * \param[in] com       pointer to the lfsck component
1509  * \param[in] orphan    pointer to the orphan MDT-object
1510  *
1511  *  type "P":           The orphan object to be created was a parent directory
1512  *                      of some MDT-object which linkEA shows that the @orphan
1513  *                      object is missing.
1514  *
1515  * \see lfsck_layout_recreate_parent() for more types.
1516  *
1517  * \param[in] lmv       pointer to master LMV EA that will be set to the orphan
1518  *
1519  * \retval              positive number for repaired cases
1520  * \retval              0 if needs to repair nothing
1521  * \retval              negative error number on failure
1522  */
1523 static int lfsck_namespace_create_orphan(const struct lu_env *env,
1524                                          struct lfsck_component *com,
1525                                          struct dt_object *orphan,
1526                                          struct lmv_mds_md_v1 *lmv)
1527 {
1528         struct lfsck_namespace *ns = com->lc_file_ram;
1529         int                     rc;
1530
1531         if (dt_object_remote(orphan))
1532                 rc = lfsck_namespace_create_orphan_remote(env, com, orphan,
1533                                                           S_IFDIR, lmv);
1534         else
1535                 rc = lfsck_namespace_create_orphan_local(env, com, orphan,
1536                                                          S_IFDIR, lmv);
1537
1538         if (rc != 0)
1539                 ns->ln_flags |= LF_INCONSISTENT;
1540
1541         return rc;
1542 }
1543
1544 /**
1545  * Remove the specified entry from the linkEA.
1546  *
1547  * Locate the linkEA entry with the given @cname and @pfid, then
1548  * remove this entry or the other entries those are repeated with
1549  * this entry.
1550  *
1551  * \param[in] env       pointer to the thread context
1552  * \param[in] com       pointer to the lfsck component
1553  * \param[in] obj       pointer to the dt_object to be handled
1554  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1555  * \param[in] cname     the name for the child in the parent directory
1556  * \param[in] pfid      the parent directory's FID for the linkEA
1557  * \param[in] next      if true, then remove the first found linkEA
1558  *                      entry, and move the ldata->ld_lee to next entry
1559  *
1560  * \retval              positive number for repaired cases
1561  * \retval              0 if nothing to be repaired
1562  * \retval              negative error number on failure
1563  */
1564 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1565                                          struct lfsck_component *com,
1566                                          struct dt_object *obj,
1567                                          struct linkea_data *ldata,
1568                                          struct lu_name *cname,
1569                                          struct lu_fid *pfid,
1570                                          bool next)
1571 {
1572         struct lfsck_instance           *lfsck     = com->lc_lfsck;
1573         struct dt_device                *dev       = lfsck->li_bottom;
1574         struct lfsck_bookmark           *bk        = &lfsck->li_bookmark_ram;
1575         struct thandle                  *th        = NULL;
1576         struct lustre_handle             lh        = { 0 };
1577         struct linkea_data               ldata_new = { 0 };
1578         struct lu_buf                    linkea_buf;
1579         int                              rc        = 0;
1580         ENTRY;
1581
1582         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1583                               MDS_INODELOCK_UPDATE |
1584                               MDS_INODELOCK_XATTR, LCK_EX);
1585         if (rc != 0)
1586                 GOTO(log, rc);
1587
1588         if (next)
1589                 linkea_del_buf(ldata, cname);
1590         else
1591                 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1592                                                     true);
1593         lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1594                        ldata->ld_leh->leh_len);
1595
1596 again:
1597         th = dt_trans_create(env, dev);
1598         if (IS_ERR(th))
1599                 GOTO(unlock1, rc = PTR_ERR(th));
1600
1601         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1602                                   XATTR_NAME_LINK, 0, th);
1603         if (rc != 0)
1604                 GOTO(stop, rc);
1605
1606         rc = dt_trans_start_local(env, dev, th);
1607         if (rc != 0)
1608                 GOTO(stop, rc);
1609
1610         dt_write_lock(env, obj, 0);
1611         if (unlikely(lfsck_is_dead_obj(obj)))
1612                 GOTO(unlock2, rc = -ENOENT);
1613
1614         rc = lfsck_links_read2(env, obj, &ldata_new);
1615         if (rc != 0)
1616                 GOTO(unlock2, rc);
1617
1618         /* The specified linkEA entry has been removed by race. */
1619         rc = linkea_links_find(&ldata_new, cname, pfid);
1620         if (rc != 0)
1621                 GOTO(unlock2, rc = 0);
1622
1623         if (bk->lb_param & LPF_DRYRUN)
1624                 GOTO(unlock2, rc = 1);
1625
1626         if (next)
1627                 linkea_del_buf(&ldata_new, cname);
1628         else
1629                 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1630                                                     true);
1631
1632         if (linkea_buf.lb_len < ldata_new.ld_leh->leh_len) {
1633                 dt_write_unlock(env, obj);
1634                 dt_trans_stop(env, dev, th);
1635                 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1636                                ldata_new.ld_leh->leh_len);
1637                 goto again;
1638         }
1639
1640         lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1641                        ldata_new.ld_leh->leh_len);
1642         rc = dt_xattr_set(env, obj, &linkea_buf,
1643                           XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1644
1645         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1646
1647 unlock2:
1648         dt_write_unlock(env, obj);
1649
1650 stop:
1651         dt_trans_stop(env, dev, th);
1652
1653 unlock1:
1654         lfsck_ibits_unlock(&lh, LCK_EX);
1655
1656 log:
1657         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1658                "for the object: "DFID", parent "DFID", name %.*s\n",
1659                lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1660                PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1661                cname->ln_name);
1662
1663         if (rc != 0) {
1664                 struct lfsck_namespace *ns = com->lc_file_ram;
1665
1666                 ns->ln_flags |= LF_INCONSISTENT;
1667         }
1668
1669         return rc;
1670 }
1671
1672 /**
1673  * Conditionally remove the specified entry from the linkEA.
1674  *
1675  * Take the parent lock firstly, then check whether the specified
1676  * name entry exists or not: if yes, do nothing; otherwise, call
1677  * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1678  *
1679  * \param[in] env       pointer to the thread context
1680  * \param[in] com       pointer to the lfsck component
1681  * \param[in] parent    pointer to the parent directory
1682  * \param[in] child     pointer to the child object that holds the linkEA
1683  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1684  * \param[in] cname     the name for the child in the parent directory
1685  * \param[in] pfid      the parent directory's FID for the linkEA
1686  *
1687  * \retval              positive number for repaired cases
1688  * \retval              0 if nothing to be repaired
1689  * \retval              negative error number on failure
1690  */
1691 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1692                                               struct lfsck_component *com,
1693                                               struct dt_object *parent,
1694                                               struct dt_object *child,
1695                                               struct linkea_data *ldata,
1696                                               struct lu_name *cname,
1697                                               struct lu_fid *pfid)
1698 {
1699         struct lu_fid           *cfid   = &lfsck_env_info(env)->lti_fid3;
1700         struct lustre_handle     lh     = { 0 };
1701         int                      rc;
1702         ENTRY;
1703
1704         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
1705                               MDS_INODELOCK_UPDATE, LCK_EX);
1706         if (rc != 0)
1707                 RETURN(rc);
1708
1709         dt_read_lock(env, parent, 0);
1710         if (unlikely(lfsck_is_dead_obj(parent))) {
1711                 dt_read_unlock(env, parent);
1712                 lfsck_ibits_unlock(&lh, LCK_EX);
1713                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1714                                                    cname, pfid, true);
1715
1716                 RETURN(rc);
1717         }
1718
1719         rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1720                        (const struct dt_key *)cname->ln_name,
1721                        BYPASS_CAPA);
1722         dt_read_unlock(env, parent);
1723
1724         /* It is safe to release the ldlm lock, because when the logic come
1725          * here, we have got all the needed information above whether the
1726          * linkEA entry is valid or not. It is not important that others
1727          * may add new linkEA entry after the ldlm lock released. If other
1728          * has removed the specified linkEA entry by race, then it is OK,
1729          * because the subsequent lfsck_namespace_shrink_linkea() can handle
1730          * such case. */
1731         lfsck_ibits_unlock(&lh, LCK_EX);
1732         if (rc == -ENOENT) {
1733                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1734                                                    cname, pfid, true);
1735
1736                 RETURN(rc);
1737         }
1738
1739         if (rc != 0)
1740                 RETURN(rc);
1741
1742         /* The LFSCK just found some internal status of cross-MDTs
1743          * create operation. That is normal. */
1744         if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1745                 linkea_next_entry(ldata);
1746
1747                 RETURN(0);
1748         }
1749
1750         rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1751                                            pfid, true);
1752
1753         RETURN(rc);
1754 }
1755
1756 /**
1757  * Conditionally replace name entry in the parent.
1758  *
1759  * As required, the LFSCK may re-create the lost MDT-object for dangling
1760  * name entry, but such repairing may be wrong because of bad FID in the
1761  * name entry. As the LFSCK processing, the real MDT-object may be found,
1762  * then the LFSCK should check whether the former re-created MDT-object
1763  * has been modified or not, if not, then destroy it and update the name
1764  * entry in the parent to reference the real MDT-object.
1765  *
1766  * \param[in] env       pointer to the thread context
1767  * \param[in] com       pointer to the lfsck component
1768  * \param[in] parent    pointer to the parent directory
1769  * \param[in] child     pointer to the MDT-object that may be the real
1770  *                      MDT-object corresponding to the name entry in parent
1771  * \param[in] cfid      the current FID in the name entry
1772  * \param[in] cname     contains the name of the child in the parent directory
1773  *
1774  * \retval              positive number for repaired cases
1775  * \retval              0 if nothing to be repaired
1776  * \retval              negative error number on failure
1777  */
1778 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1779                                         struct lfsck_component *com,
1780                                         struct dt_object *parent,
1781                                         struct dt_object *child,
1782                                         const struct lu_fid *cfid,
1783                                         const struct lu_name *cname)
1784 {
1785         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1786         struct lu_fid                   *tfid   = &info->lti_fid5;
1787         struct lu_attr                  *la     = &info->lti_la;
1788         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1789         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1790         struct dt_device                *dev    = lfsck->li_next;
1791         const char                      *name   = cname->ln_name;
1792         struct dt_object                *obj    = NULL;
1793         struct lustre_handle             plh    = { 0 };
1794         struct lustre_handle             clh    = { 0 };
1795         struct linkea_data               ldata  = { 0 };
1796         struct thandle                  *th     = NULL;
1797         bool                             exist  = true;
1798         int                              rc     = 0;
1799         ENTRY;
1800
1801         rc = lfsck_ibits_lock(env, lfsck, parent, &plh,
1802                               MDS_INODELOCK_UPDATE, LCK_EX);
1803         if (rc != 0)
1804                 GOTO(log, rc);
1805
1806         if (!fid_is_sane(cfid)) {
1807                 exist = false;
1808                 goto replace;
1809         }
1810
1811         obj = lfsck_object_find(env, lfsck, cfid);
1812         if (IS_ERR(obj)) {
1813                 rc = PTR_ERR(obj);
1814                 if (rc == -ENOENT) {
1815                         exist = false;
1816                         goto replace;
1817                 }
1818
1819                 GOTO(log, rc);
1820         }
1821
1822         if (!dt_object_exists(obj)) {
1823                 exist = false;
1824                 goto replace;
1825         }
1826
1827         rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
1828                        (const struct dt_key *)name, BYPASS_CAPA);
1829         if (rc == -ENOENT) {
1830                 exist = false;
1831                 goto replace;
1832         }
1833
1834         if (rc != 0)
1835                 GOTO(log, rc);
1836
1837         /* Someone changed the name entry, cannot replace it. */
1838         if (!lu_fid_eq(cfid, tfid))
1839                 GOTO(log, rc = 0);
1840
1841         /* lock the object to be destroyed. */
1842         rc = lfsck_ibits_lock(env, lfsck, obj, &clh,
1843                               MDS_INODELOCK_UPDATE |
1844                               MDS_INODELOCK_XATTR, LCK_EX);
1845         if (rc != 0)
1846                 GOTO(log, rc);
1847
1848         if (unlikely(lfsck_is_dead_obj(obj))) {
1849                 exist = false;
1850                 goto replace;
1851         }
1852
1853         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
1854         if (rc != 0)
1855                 GOTO(log, rc);
1856
1857         /* The object has been modified by other(s), or it is not created by
1858          * LFSCK, the two cases are indistinguishable. So cannot replace it. */
1859         if (la->la_ctime != 0)
1860                 GOTO(log, rc);
1861
1862         if (S_ISREG(la->la_mode)) {
1863                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV,
1864                                   BYPASS_CAPA);
1865                 /* If someone has created related OST-object(s),
1866                  * then keep it. */
1867                 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
1868                         GOTO(log, rc = (rc > 0 ? 0 : rc));
1869         }
1870
1871 replace:
1872         dt_read_lock(env, child, 0);
1873         rc = lfsck_links_read2(env, child, &ldata);
1874         dt_read_unlock(env, child);
1875
1876         /* Someone changed the child, no need to replace. */
1877         if (rc == -ENODATA)
1878                 GOTO(log, rc = 0);
1879
1880         if (rc != 0)
1881                 GOTO(log, rc);
1882
1883         rc = linkea_links_find(&ldata, cname, lfsck_dto2fid(parent));
1884         /* Someone moved the child, no need to replace. */
1885         if (rc != 0)
1886                 GOTO(log, rc = 0);
1887
1888         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1889                 GOTO(log, rc = 1);
1890
1891         th = dt_trans_create(env, dev);
1892         if (IS_ERR(th))
1893                 GOTO(log, rc = PTR_ERR(th));
1894
1895         if (exist) {
1896                 rc = dt_declare_destroy(env, obj, th);
1897                 if (rc != 0)
1898                         GOTO(stop, rc);
1899         }
1900
1901         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
1902         if (rc != 0)
1903                 GOTO(stop, rc);
1904
1905         rec->rec_type = S_IFDIR;
1906         rec->rec_fid = lfsck_dto2fid(child);
1907         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1908                                (const struct dt_key *)name, th);
1909         if (rc != 0)
1910                 GOTO(stop, rc);
1911
1912         rc = dt_trans_start(env, dev, th);
1913         if (rc != 0)
1914                 GOTO(stop, rc);
1915
1916         if (exist) {
1917                 rc = dt_destroy(env, obj, th);
1918                 if (rc != 0)
1919                         GOTO(stop, rc);
1920         }
1921
1922         /* The old name entry maybe not exist. */
1923         dt_delete(env, parent, (const struct dt_key *)name, th,
1924                   BYPASS_CAPA);
1925
1926         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1927                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1928
1929         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1930
1931 stop:
1932         dt_trans_stop(env, dev, th);
1933
1934 log:
1935         lfsck_ibits_unlock(&clh, LCK_EX);
1936         lfsck_ibits_unlock(&plh, LCK_EX);
1937         if (obj != NULL && !IS_ERR(obj))
1938                 lfsck_object_put(env, obj);
1939
1940         CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
1941                "object "DFID" because of conflict with the object "DFID
1942                " under the parent "DFID" with name %s: rc = %d\n",
1943                lfsck_lfsck2name(lfsck), PFID(cfid),
1944                PFID(lfsck_dto2fid(child)), PFID(lfsck_dto2fid(parent)),
1945                name, rc);
1946
1947         return rc;
1948 }
1949
1950 /**
1951  * Overwrite the linkEA for the object with the given ldata.
1952  *
1953  * The caller should take the ldlm lock before the calling.
1954  *
1955  * \param[in] env       pointer to the thread context
1956  * \param[in] com       pointer to the lfsck component
1957  * \param[in] obj       pointer to the dt_object to be handled
1958  * \param[in] ldata     pointer to the new linkEA data
1959  *
1960  * \retval              positive number for repaired cases
1961  * \retval              0 if nothing to be repaired
1962  * \retval              negative error number on failure
1963  */
1964 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
1965                                    struct lfsck_component *com,
1966                                    struct dt_object *obj,
1967                                    struct linkea_data *ldata)
1968 {
1969         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1970         struct dt_device                *dev    = lfsck->li_bottom;
1971         struct thandle                  *th     = NULL;
1972         struct lu_buf                    linkea_buf;
1973         int                              rc     = 0;
1974         ENTRY;
1975
1976         th = dt_trans_create(env, dev);
1977         if (IS_ERR(th))
1978                 GOTO(log, rc = PTR_ERR(th));
1979
1980         lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1981                        ldata->ld_leh->leh_len);
1982         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1983                                   XATTR_NAME_LINK, 0, th);
1984         if (rc != 0)
1985                 GOTO(stop, rc);
1986
1987         rc = dt_trans_start_local(env, dev, th);
1988         if (rc != 0)
1989                 GOTO(stop, rc);
1990
1991         dt_write_lock(env, obj, 0);
1992         if (unlikely(lfsck_is_dead_obj(obj)))
1993                 GOTO(unlock, rc = 0);
1994
1995         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1996                 GOTO(unlock, rc = 1);
1997
1998         rc = dt_xattr_set(env, obj, &linkea_buf,
1999                           XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
2000
2001         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2002
2003 unlock:
2004         dt_write_unlock(env, obj);
2005
2006 stop:
2007         dt_trans_stop(env, dev, th);
2008
2009 log:
2010         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
2011                "object "DFID": rc = %d\n",
2012                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
2013
2014         if (rc != 0) {
2015                 struct lfsck_namespace *ns = com->lc_file_ram;
2016
2017                 ns->ln_flags |= LF_INCONSISTENT;
2018         }
2019
2020         return rc;
2021 }
2022
2023 /**
2024  * Repair invalid name entry.
2025  *
2026  * If the name entry contains invalid information, such as bad file type
2027  * or (and) corrupted object FID, then either remove the name entry or
2028  * udpate the name entry with the given (right) information.
2029  *
2030  * \param[in] env       pointer to the thread context
2031  * \param[in] com       pointer to the lfsck component
2032  * \param[in] parent    pointer to the parent directory
2033  * \param[in] child     pointer to the object referenced by the name entry
2034  * \param[in] name      the old name of the child under the parent directory
2035  * \param[in] name2     the new name of the child under the parent directory
2036  * \param[in] type      the type claimed by the name entry
2037  * \param[in] update    update the name entry if true; otherwise, remove it
2038  * \param[in] dec       decrease the parent nlink count if true
2039  *
2040  * \retval              positive number for repaired successfully
2041  * \retval              0 if nothing to be repaired
2042  * \retval              negative error number on failure
2043  */
2044 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2045                                   struct lfsck_component *com,
2046                                   struct dt_object *parent,
2047                                   struct dt_object *child,
2048                                   const char *name, const char *name2,
2049                                   __u16 type, bool update, bool dec)
2050 {
2051         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2052         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2053         const struct lu_fid             *cfid   = lfsck_dto2fid(child);
2054         struct lu_fid                   *tfid   = &info->lti_fid5;
2055         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2056         struct dt_device                *dev    = lfsck->li_next;
2057         struct thandle                  *th     = NULL;
2058         struct lustre_handle             lh     = { 0 };
2059         int                              rc     = 0;
2060         ENTRY;
2061
2062         if (unlikely(!dt_try_as_dir(env, parent)))
2063                 GOTO(log, rc = -ENOTDIR);
2064
2065         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2066                               MDS_INODELOCK_UPDATE, LCK_EX);
2067         if (rc != 0)
2068                 GOTO(log, rc);
2069
2070         th = dt_trans_create(env, dev);
2071         if (IS_ERR(th))
2072                 GOTO(unlock1, rc = PTR_ERR(th));
2073
2074         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2075         if (rc != 0)
2076                 GOTO(stop, rc);
2077
2078         if (update) {
2079                 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2080                 rec->rec_fid = cfid;
2081                 rc = dt_declare_insert(env, parent,
2082                                        (const struct dt_rec *)rec,
2083                                        (const struct dt_key *)name2, th);
2084                 if (rc != 0)
2085                         GOTO(stop, rc);
2086         }
2087
2088         if (dec) {
2089                 rc = dt_declare_ref_del(env, parent, th);
2090                 if (rc != 0)
2091                         GOTO(stop, rc);
2092         }
2093
2094         rc = dt_trans_start(env, dev, th);
2095         if (rc != 0)
2096                 GOTO(stop, rc);
2097
2098         dt_write_lock(env, parent, 0);
2099         rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
2100                        (const struct dt_key *)name, BYPASS_CAPA);
2101         /* Someone has removed the bad name entry by race. */
2102         if (rc == -ENOENT)
2103                 GOTO(unlock2, rc = 0);
2104
2105         if (rc != 0)
2106                 GOTO(unlock2, rc);
2107
2108         /* Someone has removed the bad name entry and reused it for other
2109          * object by race. */
2110         if (!lu_fid_eq(tfid, cfid))
2111                 GOTO(unlock2, rc = 0);
2112
2113         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2114                 GOTO(unlock2, rc = 1);
2115
2116         rc = dt_delete(env, parent, (const struct dt_key *)name, th,
2117                        BYPASS_CAPA);
2118         if (rc != 0)
2119                 GOTO(unlock2, rc);
2120
2121         if (update) {
2122                 rc = dt_insert(env, parent,
2123                                (const struct dt_rec *)rec,
2124                                (const struct dt_key *)name2, th,
2125                                BYPASS_CAPA, 1);
2126                 if (rc != 0)
2127                         GOTO(unlock2, rc);
2128         }
2129
2130         if (dec) {
2131                 rc = dt_ref_del(env, parent, th);
2132                 if (rc != 0)
2133                         GOTO(unlock2, rc);
2134         }
2135
2136         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2137
2138 unlock2:
2139         dt_write_unlock(env, parent);
2140
2141 stop:
2142         dt_trans_stop(env, dev, th);
2143
2144         /* We are not sure whether the child will become orphan or not.
2145          * Record it in the LFSCK trace file for further checking in
2146          * the second-stage scanning. */
2147         if (!update && !dec && rc == 0)
2148                 lfsck_namespace_trace_update(env, com, cfid,
2149                                              LNTF_CHECK_LINKEA, true);
2150
2151 unlock1:
2152         lfsck_ibits_unlock(&lh, LCK_EX);
2153
2154 log:
2155         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2156                "entry for: parent "DFID", child "DFID", name %s, type "
2157                "in name entry %o, type claimed by child %o. repair it "
2158                "by %s with new name2 %s: rc = %d\n", lfsck_lfsck2name(lfsck),
2159                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
2160                name, type, update ? lfsck_object_type(child) : 0,
2161                update ? "updating" : "removing", name2, rc);
2162
2163         if (rc != 0) {
2164                 struct lfsck_namespace *ns = com->lc_file_ram;
2165
2166                 ns->ln_flags |= LF_INCONSISTENT;
2167         }
2168
2169         return rc;
2170 }
2171
2172 /**
2173  * Update the ".." name entry for the given object.
2174  *
2175  * The object's ".." is corrupted, this function will update the ".." name
2176  * entry with the given pfid, and the linkEA with the given ldata.
2177  *
2178  * The caller should take the ldlm lock before the calling.
2179  *
2180  * \param[in] env       pointer to the thread context
2181  * \param[in] com       pointer to the lfsck component
2182  * \param[in] obj       pointer to the dt_object to be handled
2183  * \param[in] pfid      the new fid for the object's ".." name entry
2184  * \param[in] cname     the name for the @obj in the parent directory
2185  *
2186  * \retval              positive number for repaired cases
2187  * \retval              0 if nothing to be repaired
2188  * \retval              negative error number on failure
2189  */
2190 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2191                                                   struct lfsck_component *com,
2192                                                   struct dt_object *obj,
2193                                                   const struct lu_fid *pfid,
2194                                                   struct lu_name *cname)
2195 {
2196         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2197         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2198         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2199         struct dt_device                *dev    = lfsck->li_bottom;
2200         struct thandle                  *th     = NULL;
2201         struct linkea_data               ldata  = { 0 };
2202         struct lu_buf                    linkea_buf;
2203         int                              rc     = 0;
2204         ENTRY;
2205
2206         LASSERT(!dt_object_remote(obj));
2207         LASSERT(S_ISDIR(lfsck_object_type(obj)));
2208
2209         rc = linkea_data_new(&ldata, &info->lti_big_buf);
2210         if (rc != 0)
2211                 GOTO(log, rc);
2212
2213         rc = linkea_add_buf(&ldata, cname, pfid);
2214         if (rc != 0)
2215                 GOTO(log, rc);
2216
2217         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2218                        ldata.ld_leh->leh_len);
2219
2220         th = dt_trans_create(env, dev);
2221         if (IS_ERR(th))
2222                 GOTO(log, rc = PTR_ERR(th));
2223
2224         rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2225         if (rc != 0)
2226                 GOTO(stop, rc);
2227
2228         rec->rec_type = S_IFDIR;
2229         rec->rec_fid = pfid;
2230         rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2231                                (const struct dt_key *)dotdot, th);
2232         if (rc != 0)
2233                 GOTO(stop, rc);
2234
2235         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2236                                   XATTR_NAME_LINK, 0, th);
2237         if (rc != 0)
2238                 GOTO(stop, rc);
2239
2240         rc = dt_trans_start_local(env, dev, th);
2241         if (rc != 0)
2242                 GOTO(stop, rc);
2243
2244         dt_write_lock(env, obj, 0);
2245         if (unlikely(lfsck_is_dead_obj(obj)))
2246                 GOTO(unlock, rc = 0);
2247
2248         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2249                 GOTO(unlock, rc = 1);
2250
2251         /* The old ".." name entry maybe not exist. */
2252         dt_delete(env, obj, (const struct dt_key *)dotdot, th,
2253                   BYPASS_CAPA);
2254
2255         rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2256                        (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1);
2257         if (rc != 0)
2258                 GOTO(unlock, rc);
2259
2260         rc = dt_xattr_set(env, obj, &linkea_buf,
2261                           XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
2262
2263         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2264
2265 unlock:
2266         dt_write_unlock(env, obj);
2267
2268 stop:
2269         dt_trans_stop(env, dev, th);
2270
2271 log:
2272         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2273                "the object "DFID", new parent "DFID": rc = %d\n",
2274                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2275                PFID(pfid), rc);
2276
2277         if (rc != 0) {
2278                 struct lfsck_namespace *ns = com->lc_file_ram;
2279
2280                 ns->ln_flags |= LF_INCONSISTENT;
2281         }
2282
2283         return rc;
2284 }
2285
2286 /**
2287  * Handle orphan @obj during Double Scan Directory.
2288  *
2289  * Remove the @obj's current (invalid) linkEA entries, and insert
2290  * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2291  * ${FID}-${PFID}-D-${conflict_version}
2292  *
2293  * The caller should take the ldlm lock before the calling.
2294  *
2295  * \param[in] env       pointer to the thread context
2296  * \param[in] com       pointer to the lfsck component
2297  * \param[in] obj       pointer to the orphan object to be handled
2298  * \param[in] pfid      the new fid for the object's ".." name entry
2299  * \param[in,out] lh    ldlm lock handler for the given @obj
2300  * \param[out] type     to tell the caller what the inconsistency is
2301  *
2302  * \retval              positive number for repaired cases
2303  * \retval              0 if nothing to be repaired
2304  * \retval              negative error number on failure
2305  */
2306 static int
2307 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2308                            struct lfsck_component *com,
2309                            struct dt_object *obj,
2310                            const struct lu_fid *pfid,
2311                            struct lustre_handle *lh,
2312                            enum lfsck_namespace_inconsistency_type *type)
2313 {
2314         struct lfsck_thread_info *info = lfsck_env_info(env);
2315         struct lfsck_namespace   *ns   = com->lc_file_ram;
2316         int                       rc;
2317         ENTRY;
2318
2319         /* Remove the unrecognized linkEA. */
2320         rc = lfsck_namespace_links_remove(env, com, obj);
2321         lfsck_ibits_unlock(lh, LCK_EX);
2322         if (rc < 0 && rc != -ENODATA)
2323                 RETURN(rc);
2324
2325         *type = LNIT_MUL_REF;
2326
2327         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2328          * ever tried to verify some remote MDT-object that resides on this
2329          * MDT, but this MDT failed to respond such request. So means there
2330          * may be some remote name entry on other MDT that references this
2331          * object with another name, so we cannot know whether this linkEA
2332          * is valid or not. So keep it there and maybe resolved when next
2333          * LFSCK run. */
2334         if (ns->ln_flags & LF_INCOMPLETE)
2335                 RETURN(0);
2336
2337         /* The unique linkEA is invalid, even if the ".." name entry may be
2338          * valid, we still cannot know via which name entry this directory
2339          * will be referenced. Then handle it as pure orphan. */
2340         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2341                  "-"DFID, PFID(pfid));
2342         rc = lfsck_namespace_insert_orphan(env, com, obj,
2343                                            info->lti_tmpbuf, "D", NULL);
2344
2345         RETURN(rc);
2346 }
2347
2348 /**
2349  * Double Scan Directory object for single linkEA entry case.
2350  *
2351  * The given @child has unique linkEA entry. If the linkEA entry is valid,
2352  * then check whether the name is in the namespace or not, if not, add the
2353  * missing name entry back to namespace. If the linkEA entry is invalid,
2354  * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2355  * as an orphan.
2356  *
2357  * \param[in] env       pointer to the thread context
2358  * \param[in] com       pointer to the lfsck component
2359  * \param[in] child     pointer to the directory to be double scanned
2360  * \param[in] pfid      the FID corresponding to the ".." entry
2361  * \param[in] ldata     pointer to the linkEA data for the given @child
2362  * \param[in,out] lh    ldlm lock handler for the given @child
2363  * \param[out] type     to tell the caller what the inconsistency is
2364  * \param[in] retry     if found inconsistency, but the caller does not hold
2365  *                      ldlm lock on the @child, then set @retry as true
2366  *
2367  * \retval              positive number for repaired cases
2368  * \retval              0 if nothing to be repaired
2369  * \retval              negative error number on failure
2370  */
2371 static int
2372 lfsck_namespace_dsd_single(const struct lu_env *env,
2373                            struct lfsck_component *com,
2374                            struct dt_object *child,
2375                            const struct lu_fid *pfid,
2376                            struct linkea_data *ldata,
2377                            struct lustre_handle *lh,
2378                            enum lfsck_namespace_inconsistency_type *type,
2379                            bool *retry)
2380 {
2381         struct lfsck_thread_info *info          = lfsck_env_info(env);
2382         struct lu_name           *cname         = &info->lti_name;
2383         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2384         struct lu_fid            *tfid          = &info->lti_fid3;
2385         struct lfsck_namespace   *ns            = com->lc_file_ram;
2386         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2387         struct dt_object         *parent        = NULL;
2388         struct lmv_mds_md_v1     *lmv;
2389         int                       rc            = 0;
2390         ENTRY;
2391
2392         lfsck_namespace_unpack_linkea_entry(ldata, cname, tfid, info->lti_key);
2393         /* The unique linkEA entry with bad parent will be handled as orphan. */
2394         if (!fid_is_sane(tfid)) {
2395                 if (!lustre_handle_is_used(lh) && retry != NULL)
2396                         *retry = true;
2397                 else
2398                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2399                                                         pfid, lh, type);
2400
2401                 GOTO(out, rc);
2402         }
2403
2404         parent = lfsck_object_find_bottom(env, lfsck, tfid);
2405         if (IS_ERR(parent))
2406                 GOTO(out, rc = PTR_ERR(parent));
2407
2408         /* We trust the unique linkEA entry in spite of whether it matches the
2409          * ".." name entry or not. Because even if the linkEA entry is wrong
2410          * and the ".." name entry is right, we still cannot know via which
2411          * name entry the child will be referenced, since all known entries
2412          * have been verified during the first-stage scanning. */
2413         if (!dt_object_exists(parent)) {
2414                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2415                  * has ever tried to verify some remote MDT-object that resides
2416                  * on this MDT, but this MDT failed to respond such request. So
2417                  * means there may be some remote name entry on other MDT that
2418                  * references this object with another name, so we cannot know
2419                  * whether this linkEA is valid or not. So keep it there and
2420                  * maybe resolved when next LFSCK run. */
2421                 if (ns->ln_flags & LF_INCOMPLETE)
2422                         GOTO(out, rc = 0);
2423
2424                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2425                         *retry = true;
2426
2427                         GOTO(out, rc = 0);
2428                 }
2429
2430                 lfsck_ibits_unlock(lh, LCK_EX);
2431
2432 lost_parent:
2433                 lmv = &info->lti_lmv;
2434                 rc = lfsck_read_stripe_lmv(env, child, lmv);
2435                 if (rc != 0 && rc != -ENODATA)
2436                         GOTO(out, rc);
2437
2438                 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2439                         lmv = NULL;
2440                 } else if (lfsck_shard_name_to_index(env,
2441                                         cname->ln_name, cname->ln_namelen,
2442                                         S_IFDIR, cfid) < 0) {
2443                         /* It is an invalid name entry, we
2444                          * cannot trust the parent also. */
2445                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2446                                                 ldata, cname, tfid, true);
2447                         if (rc < 0)
2448                                 GOTO(out, rc);
2449
2450                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2451                                  "-"DFID, PFID(pfid));
2452                         rc = lfsck_namespace_insert_orphan(env, com, child,
2453                                                 info->lti_tmpbuf, "S", NULL);
2454
2455                         GOTO(out, rc);
2456                 }
2457
2458                 /* Create the lost parent as an orphan. */
2459                 rc = lfsck_namespace_create_orphan(env, com, parent, lmv);
2460                 if (rc >= 0) {
2461                         /* Add the missing name entry to the parent. */
2462                         rc = lfsck_namespace_insert_normal(env, com, parent,
2463                                                         child, cname->ln_name);
2464                         if (unlikely(rc == -EEXIST)) {
2465                                 /* Unfortunately, someone reused the name
2466                                  * under the parent by race. So we have
2467                                  * to remove the linkEA entry from
2468                                  * current child object. It means that the
2469                                  * LFSCK cannot recover the system
2470                                  * totally back to its original status,
2471                                  * but it is necessary to make the
2472                                  * current system to be consistent. */
2473                                 rc = lfsck_namespace_shrink_linkea(env,
2474                                                 com, child, ldata,
2475                                                 cname, tfid, true);
2476                                 if (rc >= 0) {
2477                                         snprintf(info->lti_tmpbuf,
2478                                                  sizeof(info->lti_tmpbuf),
2479                                                  "-"DFID, PFID(pfid));
2480                                         rc = lfsck_namespace_insert_orphan(env,
2481                                                 com, child, info->lti_tmpbuf,
2482                                                 "D", NULL);
2483                                 }
2484                         }
2485                 }
2486
2487                 GOTO(out, rc);
2488         }
2489
2490         /* The unique linkEA entry with bad parent will be handled as orphan. */
2491         if (unlikely(!dt_try_as_dir(env, parent))) {
2492                 if (!lustre_handle_is_used(lh) && retry != NULL)
2493                         *retry = true;
2494                 else
2495                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2496                                                         pfid, lh, type);
2497
2498                 GOTO(out, rc);
2499         }
2500
2501         rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
2502                        (const struct dt_key *)cname->ln_name, BYPASS_CAPA);
2503         if (rc == -ENOENT) {
2504                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2505                  * has ever tried to verify some remote MDT-object that resides
2506                  * on this MDT, but this MDT failed to respond such request. So
2507                  * means there may be some remote name entry on other MDT that
2508                  * references this object with another name, so we cannot know
2509                  * whether this linkEA is valid or not. So keep it there and
2510                  * maybe resolved when next LFSCK run. */
2511                 if (ns->ln_flags & LF_INCOMPLETE)
2512                         GOTO(out, rc = 0);
2513
2514                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2515                         *retry = true;
2516
2517                         GOTO(out, rc = 0);
2518                 }
2519
2520                 lfsck_ibits_unlock(lh, LCK_EX);
2521                 rc = lfsck_namespace_check_name(env, parent, child, cname);
2522                 if (rc == -ENOENT)
2523                         goto lost_parent;
2524
2525                 if (rc < 0)
2526                         GOTO(out, rc);
2527
2528                 /* It is an invalid name entry, drop it. */
2529                 if (unlikely(rc > 0)) {
2530                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2531                                                 ldata, cname, tfid, true);
2532                         if (rc >= 0) {
2533                                 snprintf(info->lti_tmpbuf,
2534                                          sizeof(info->lti_tmpbuf),
2535                                          "-"DFID, PFID(pfid));
2536                                 rc = lfsck_namespace_insert_orphan(env, com,
2537                                         child, info->lti_tmpbuf, "D", NULL);
2538                         }
2539
2540                         GOTO(out, rc);
2541                 }
2542
2543                 /* Add the missing name entry back to the namespace. */
2544                 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2545                                                    cname->ln_name);
2546                 if (unlikely(rc == -ESTALE))
2547                         /* It may happen when the remote object has been
2548                          * removed, but the local MDT is not aware of that. */
2549                         goto lost_parent;
2550
2551                 if (unlikely(rc == -EEXIST)) {
2552                         /* Unfortunately, someone reused the name under the
2553                          * parent by race. So we have to remove the linkEA
2554                          * entry from current child object. It means that the
2555                          * LFSCK cannot recover the system totally back to
2556                          * its original status, but it is necessary to make
2557                          * the current system to be consistent.
2558                          *
2559                          * It also may be because of the LFSCK found some
2560                          * internal status of create operation. Under such
2561                          * case, nothing to be done. */
2562                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
2563                                         parent, child, ldata, cname, tfid);
2564                         if (rc >= 0) {
2565                                 snprintf(info->lti_tmpbuf,
2566                                          sizeof(info->lti_tmpbuf),
2567                                          "-"DFID, PFID(pfid));
2568                                 rc = lfsck_namespace_insert_orphan(env, com,
2569                                         child, info->lti_tmpbuf, "D", NULL);
2570                         }
2571                 }
2572
2573                 GOTO(out, rc);
2574         }
2575
2576         if (rc != 0)
2577                 GOTO(out, rc);
2578
2579         if (!lu_fid_eq(tfid, cfid)) {
2580                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2581                         *retry = true;
2582
2583                         GOTO(out, rc = 0);
2584                 }
2585
2586                 lfsck_ibits_unlock(lh, LCK_EX);
2587                 /* The name entry references another MDT-object that
2588                  * may be created by the LFSCK for repairing dangling
2589                  * name entry. Try to replace it. */
2590                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2591                                                   tfid, cname);
2592                 if (rc == 0)
2593                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2594                                                         pfid, lh, type);
2595
2596                 GOTO(out, rc);
2597         }
2598
2599         if (fid_is_zero(pfid))
2600                 GOTO(out, rc = 0);
2601
2602         /* The ".." name entry is wrong, update it. */
2603         if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2604                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2605                         *retry = true;
2606
2607                         GOTO(out, rc = 0);
2608                 }
2609
2610                 *type = LNIT_UNMATCHED_PAIRS;
2611                 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2612                                                 lfsck_dto2fid(parent), cname);
2613         }
2614
2615         GOTO(out, rc);
2616
2617 out:
2618         if (parent != NULL && !IS_ERR(parent))
2619                 lfsck_object_put(env, parent);
2620
2621         return rc;
2622 }
2623
2624 /**
2625  * Double Scan Directory object for multiple linkEA entries case.
2626  *
2627  * The given @child has multiple linkEA entries. There is at most one linkEA
2628  * entry will be valid, all the others will be removed. Firstly, the function
2629  * will try to find out the linkEA entry for which the name entry exists under
2630  * the given parent (@pfid). If there is no linkEA entry that matches the given
2631  * ".." name entry, then tries to find out the first linkEA entry that both the
2632  * parent and the name entry exist to rebuild a new ".." name entry.
2633  *
2634  * \param[in] env       pointer to the thread context
2635  * \param[in] com       pointer to the lfsck component
2636  * \param[in] child     pointer to the directory to be double scanned
2637  * \param[in] pfid      the FID corresponding to the ".." entry
2638  * \param[in] ldata     pointer to the linkEA data for the given @child
2639  * \param[in,out] lh    ldlm lock handler for the given @child
2640  * \param[out] type     to tell the caller what the inconsistency is
2641  * \param[in] lpf       true if the ".." entry is under lost+found/MDTxxxx/
2642  *
2643  * \retval              positive number for repaired cases
2644  * \retval              0 if nothing to be repaired
2645  * \retval              negative error number on failure
2646  */
2647 static int
2648 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2649                              struct lfsck_component *com,
2650                              struct dt_object *child,
2651                              const struct lu_fid *pfid,
2652                              struct linkea_data *ldata,
2653                              struct lustre_handle *lh,
2654                              enum lfsck_namespace_inconsistency_type *type,
2655                              bool lpf)
2656 {
2657         struct lfsck_thread_info *info          = lfsck_env_info(env);
2658         struct lu_name           *cname         = &info->lti_name;
2659         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2660         struct lu_fid            *tfid          = &info->lti_fid3;
2661         struct lu_fid            *pfid2         = &info->lti_fid4;
2662         struct lfsck_namespace   *ns            = com->lc_file_ram;
2663         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2664         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2665         struct dt_object         *parent        = NULL;
2666         struct linkea_data        ldata_new     = { 0 };
2667         int                       dirent_count  = 0;
2668         int                       linkea_count  = 0;
2669         int                       rc            = 0;
2670         bool                      once          = true;
2671         ENTRY;
2672
2673 again:
2674         while (ldata->ld_lee != NULL) {
2675                 lfsck_namespace_unpack_linkea_entry(ldata, cname, tfid,
2676                                                     info->lti_key);
2677                 /* Drop repeated linkEA entries. */
2678                 lfsck_namespace_filter_linkea_entry(ldata, cname, tfid, true);
2679                 /* Drop invalid linkEA entry. */
2680                 if (!fid_is_sane(tfid)) {
2681                         linkea_del_buf(ldata, cname);
2682                         linkea_count++;
2683                         continue;
2684                 }
2685
2686                 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2687                  * then it is possible that: the directry object has ever
2688                  * been lost, but its name entry was there. In the former
2689                  * LFSCK run, during the first-stage scanning, the LFSCK
2690                  * found the dangling name entry, but it did not recreate
2691                  * the lost object, and when moved to the second-stage
2692                  * scanning, some children objects of the lost directory
2693                  * object were found, then the LFSCK recreated such lost
2694                  * directory object as an orphan.
2695                  *
2696                  * When the LFSCK runs again, if the dangling name is still
2697                  * there, the LFSCK should move the orphan directory object
2698                  * back to the normal namespace. */
2699                 if (!lpf && !lu_fid_eq(pfid, tfid) && once) {
2700                         linkea_next_entry(ldata);
2701                         continue;
2702                 }
2703
2704                 parent = lfsck_object_find_bottom(env, lfsck, tfid);
2705                 if (IS_ERR(parent))
2706                         RETURN(PTR_ERR(parent));
2707
2708                 if (!dt_object_exists(parent)) {
2709                         lfsck_object_put(env, parent);
2710                         if (ldata->ld_leh->leh_reccount > 1) {
2711                                 /* If it is NOT the last linkEA entry, then
2712                                  * there is still other chance to make the
2713                                  * child to be visible via other parent, then
2714                                  * remove this linkEA entry. */
2715                                 linkea_del_buf(ldata, cname);
2716                                 linkea_count++;
2717                                 continue;
2718                         }
2719
2720                         break;
2721                 }
2722
2723                 /* The linkEA entry with bad parent will be removed. */
2724                 if (unlikely(!dt_try_as_dir(env, parent))) {
2725                         lfsck_object_put(env, parent);
2726                         linkea_del_buf(ldata, cname);
2727                         linkea_count++;
2728                         continue;
2729                 }
2730
2731                 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
2732                                (const struct dt_key *)cname->ln_name,
2733                                BYPASS_CAPA);
2734                 *pfid2 = *lfsck_dto2fid(parent);
2735                 if (rc == -ENOENT) {
2736                         lfsck_object_put(env, parent);
2737                         linkea_next_entry(ldata);
2738                         continue;
2739                 }
2740
2741                 if (rc != 0) {
2742                         lfsck_object_put(env, parent);
2743
2744                         RETURN(rc);
2745                 }
2746
2747                 if (lu_fid_eq(tfid, cfid)) {
2748                         lfsck_object_put(env, parent);
2749                         if (!lu_fid_eq(pfid, pfid2)) {
2750                                 *type = LNIT_UNMATCHED_PAIRS;
2751                                 rc = lfsck_namespace_repair_unmatched_pairs(env,
2752                                                 com, child, pfid2, cname);
2753
2754                                 RETURN(rc);
2755                         }
2756
2757 rebuild:
2758                         /* It is the most common case that we find the
2759                          * name entry corresponding to the linkEA entry
2760                          * that matches the ".." name entry. */
2761                         rc = linkea_data_new(&ldata_new, &info->lti_big_buf);
2762                         if (rc != 0)
2763                                 RETURN(rc);
2764
2765                         rc = linkea_add_buf(&ldata_new, cname, pfid2);
2766                         if (rc != 0)
2767                                 RETURN(rc);
2768
2769                         rc = lfsck_namespace_rebuild_linkea(env, com, child,
2770                                                             &ldata_new);
2771                         if (rc < 0)
2772                                 RETURN(rc);
2773
2774                         linkea_del_buf(ldata, cname);
2775                         linkea_count++;
2776                         linkea_first_entry(ldata);
2777                         /* There may be some invalid dangling name entries under
2778                          * other parent directories, remove all of them. */
2779                         while (ldata->ld_lee != NULL) {
2780                                 lfsck_namespace_unpack_linkea_entry(ldata,
2781                                                 cname, tfid, info->lti_key);
2782                                 if (!fid_is_sane(tfid))
2783                                         goto next;
2784
2785                                 parent = lfsck_object_find_bottom(env, lfsck,
2786                                                                   tfid);
2787                                 if (IS_ERR(parent)) {
2788                                         rc = PTR_ERR(parent);
2789                                         if (rc != -ENOENT &&
2790                                             bk->lb_param & LPF_FAILOUT)
2791                                                 RETURN(rc);
2792
2793                                         goto next;
2794                                 }
2795
2796                                 if (!dt_object_exists(parent)) {
2797                                         lfsck_object_put(env, parent);
2798                                         goto next;
2799                                 }
2800
2801                                 rc = lfsck_namespace_repair_dirent(env, com,
2802                                         parent, child, cname->ln_name,
2803                                         cname->ln_name, S_IFDIR, false, true);
2804                                 lfsck_object_put(env, parent);
2805                                 if (rc < 0) {
2806                                         if (bk->lb_param & LPF_FAILOUT)
2807                                                 RETURN(rc);
2808
2809                                         goto next;
2810                                 }
2811
2812                                 dirent_count += rc;
2813
2814 next:
2815                                 linkea_del_buf(ldata, cname);
2816                         }
2817
2818                         ns->ln_dirent_repaired += dirent_count;
2819
2820                         RETURN(rc);
2821                 }
2822
2823                 lfsck_ibits_unlock(lh, LCK_EX);
2824                 /* The name entry references another MDT-object that may be
2825                  * created by the LFSCK for repairing dangling name entry.
2826                  * Try to replace it. */
2827                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2828                                                   tfid, cname);
2829                 lfsck_object_put(env, parent);
2830                 if (rc < 0)
2831                         RETURN(rc);
2832
2833                 if (rc > 0)
2834                         goto rebuild;
2835
2836                 linkea_del_buf(ldata, cname);
2837         }
2838
2839         linkea_first_entry(ldata);
2840         if (ldata->ld_leh->leh_reccount == 1) {
2841                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2842                                                 lh, type, NULL);
2843
2844                 if (rc == 0 && fid_is_zero(pfid) && linkea_count > 0)
2845                         rc = lfsck_namespace_rebuild_linkea(env, com, child,
2846                                                             ldata);
2847
2848                 RETURN(rc);
2849         }
2850
2851         /* All linkEA entries are invalid and removed, then handle the @child
2852          * as an orphan.*/
2853         if (ldata->ld_leh->leh_reccount == 0) {
2854                 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2855                                                 type);
2856
2857                 RETURN(rc);
2858         }
2859
2860         /* If the dangling name entry for the orphan directory object has
2861          * been remvoed, then just check whether the directory object is
2862          * still under the .lustre/lost+found/MDTxxxx/ or not. */
2863         if (lpf) {
2864                 lpf = false;
2865                 goto again;
2866         }
2867
2868         /* There is no linkEA entry that matches the ".." name entry. Find
2869          * the first linkEA entry that both parent and name entry exist to
2870          * rebuild a new ".." name entry. */
2871         if (once) {
2872                 once = false;
2873                 goto again;
2874         }
2875
2876         RETURN(rc);
2877 }
2878
2879 /**
2880  * Repair the object's nlink attribute.
2881  *
2882  * If all the known name entries have been verified, then the object's hard
2883  * link attribute should match the object's linkEA entries count unless the
2884  * object's has too much hard link to be recorded in the linkEA. Such cases
2885  * should have been marked in the LFSCK trace file. Otherwise, trust the
2886  * linkEA to update the object's nlink attribute.
2887  *
2888  * \param[in] env       pointer to the thread context
2889  * \param[in] com       pointer to the lfsck component
2890  * \param[in] obj       pointer to the dt_object to be handled
2891  * \param[in,out] la    pointer to buffer to object's attribute before
2892  *                      and after the repairing
2893  *
2894  * \retval              positive number for repaired cases
2895  * \retval              0 if nothing to be repaired
2896  * \retval              negative error number on failure
2897  */
2898 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2899                                         struct lfsck_component *com,
2900                                         struct dt_object *obj,
2901                                         struct lu_attr *la)
2902 {
2903         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2904         struct lu_fid                   *tfid   = &info->lti_fid3;
2905         struct lfsck_namespace          *ns     = com->lc_file_ram;
2906         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2907         struct dt_device                *dev    = lfsck->li_bottom;
2908         const struct lu_fid             *cfid   = lfsck_dto2fid(obj);
2909         struct dt_object                *child  = NULL;
2910         struct thandle                  *th     = NULL;
2911         struct linkea_data               ldata  = { 0 };
2912         struct lustre_handle             lh     = { 0 };
2913         __u32                            old    = la->la_nlink;
2914         int                              rc     = 0;
2915         __u8                             flags;
2916         ENTRY;
2917
2918         LASSERT(!dt_object_remote(obj));
2919         LASSERT(S_ISREG(lfsck_object_type(obj)));
2920
2921         child = lfsck_object_find_by_dev(env, dev, cfid);
2922         if (IS_ERR(child))
2923                 GOTO(log, rc = PTR_ERR(child));
2924
2925         rc = lfsck_ibits_lock(env, lfsck, child, &lh,
2926                               MDS_INODELOCK_UPDATE |
2927                               MDS_INODELOCK_XATTR, LCK_EX);
2928         if (rc != 0)
2929                 GOTO(log, rc);
2930
2931         th = dt_trans_create(env, dev);
2932         if (IS_ERR(th))
2933                 GOTO(log, rc = PTR_ERR(th));
2934
2935         la->la_valid = LA_NLINK;
2936         rc = dt_declare_attr_set(env, child, la, th);
2937         if (rc != 0)
2938                 GOTO(stop, rc);
2939
2940         rc = dt_trans_start_local(env, dev, th);
2941         if (rc != 0)
2942                 GOTO(stop, rc);
2943
2944         dt_write_lock(env, child, 0);
2945         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2946          * ever tried to verify some remote MDT-object that resides on this
2947          * MDT, but this MDT failed to respond such request. So means there
2948          * may be some remote name entry on other MDT that references this
2949          * object with another name, so we cannot know whether this linkEA
2950          * is valid or not. So keep it there and maybe resolved when next
2951          * LFSCK run. */
2952         if (ns->ln_flags & LF_INCOMPLETE)
2953                 GOTO(unlock, rc = 0);
2954
2955         fid_cpu_to_be(tfid, cfid);
2956         rc = dt_lookup(env, com->lc_obj, (struct dt_rec *)&flags,
2957                        (const struct dt_key *)tfid, BYPASS_CAPA);
2958         if (rc != 0)
2959                 GOTO(unlock, rc);
2960
2961         if (flags & LNTF_SKIP_NLINK)
2962                 GOTO(unlock, rc = 0);
2963
2964         rc = dt_attr_get(env, child, la, BYPASS_CAPA);
2965         if (rc != 0)
2966                 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
2967
2968         rc = lfsck_links_read2(env, child, &ldata);
2969         if (rc != 0)
2970                 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
2971
2972         if (la->la_nlink == ldata.ld_leh->leh_reccount ||
2973             unlikely(la->la_nlink == 0))
2974                 GOTO(unlock, rc = 0);
2975
2976         la->la_nlink = ldata.ld_leh->leh_reccount;
2977         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2978                 GOTO(unlock, rc = 1);
2979
2980         rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
2981
2982         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2983
2984 unlock:
2985         dt_write_unlock(env, child);
2986
2987 stop:
2988         dt_trans_stop(env, dev, th);
2989
2990 log:
2991         lfsck_ibits_unlock(&lh, LCK_EX);
2992         if (child != NULL && !IS_ERR(child))
2993                 lfsck_object_put(env, child);
2994
2995         CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
2996                "nlink count from %u to %u: rc = %d\n",
2997                lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
2998
2999         if (rc != 0)
3000                 ns->ln_flags |= LF_INCONSISTENT;
3001
3002         return rc;
3003 }
3004
3005 /**
3006  * Double scan the directory object for namespace LFSCK.
3007  *
3008  * This function will verify the <parent, child> pairs in the namespace tree:
3009  * the parent references the child via some name entry that should be in the
3010  * child's linkEA entry, the child should back references the parent via its
3011  * ".." name entry.
3012  *
3013  * The LFSCK will scan every linkEA entry in turn until find out the first
3014  * matched pairs. If found, then all other linkEA entries will be dropped.
3015  * If all the linkEA entries cannot match the ".." name entry, then there
3016  * are serveral possible cases:
3017  *
3018  * 1) If there is only one linkEA entry, then trust it as long as the PFID
3019  *    in the linkEA entry is valid.
3020  *
3021  * 2) If there are multiple linkEA entries, then try to find the linkEA
3022  *    that matches the ".." name entry. If found, then all other entries
3023  *    are invalid; otherwise, it is quite possible that the ".." name entry
3024  *    is corrupted. Under such case, the LFSCK will rebuild the ".." name
3025  *    entry according to the first valid linkEA entry (both the parent and
3026  *    the name entry should exist).
3027  *
3028  * 3) If the directory object has no (valid) linkEA entry, then the
3029  *    directory object will be handled as pure orphan and inserted
3030  *    in the .lustre/lost+found/MDTxxxx/ with the name:
3031  *    ${self_FID}-${PFID}-D-${conflict_version}
3032  *
3033  * \param[in] env       pointer to the thread context
3034  * \param[in] com       pointer to the lfsck component
3035  * \param[in] child     pointer to the directory object to be handled
3036  * \param[in] flags     to indicate the specical checking on the @child
3037  *
3038  * \retval              positive number for repaired cases
3039  * \retval              0 if nothing to be repaired
3040  * \retval              negative error number on failure
3041  */
3042 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3043                                            struct lfsck_component *com,
3044                                            struct dt_object *child, __u8 flags)
3045 {
3046         struct lfsck_thread_info *info          = lfsck_env_info(env);
3047         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
3048         struct lu_fid            *pfid          = &info->lti_fid2;
3049         struct lfsck_namespace   *ns            = com->lc_file_ram;
3050         struct lfsck_instance    *lfsck         = com->lc_lfsck;
3051         struct lustre_handle      lh            = { 0 };
3052         struct linkea_data        ldata         = { 0 };
3053         bool                      unknown       = false;
3054         bool                      lpf           = false;
3055         bool                      retry         = false;
3056         enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3057         int                       rc            = 0;
3058         ENTRY;
3059
3060         LASSERT(!dt_object_remote(child));
3061
3062         if (flags & LNTF_UNCERTAIN_LMV) {
3063                 if (flags & LNTF_RECHECK_NAME_HASH) {
3064                         rc = lfsck_namespace_scan_shard(env, com, child);
3065                         if (rc < 0)
3066                                 RETURN(rc);
3067
3068                         ns->ln_striped_shards_scanned++;
3069                 } else {
3070                         ns->ln_striped_shards_skipped++;
3071                 }
3072         }
3073
3074         flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3075         if (flags == 0)
3076                 RETURN(0);
3077
3078         if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3079             !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3080                 CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
3081                        "the namespace LFSCK, then the LFSCK cannot guarantee"
3082                        "all the name entries have been verified in first-stage"
3083                        "scanning. So have to skip orphan related handling for"
3084                        "the directory object "DFID" with remote name entry\n",
3085                        lfsck_lfsck2name(lfsck), PFID(cfid));
3086
3087                 RETURN(0);
3088         }
3089
3090         if (unlikely(!dt_try_as_dir(env, child)))
3091                 GOTO(out, rc = -ENOTDIR);
3092
3093         /* We only take ldlm lock on the @child when required. When the
3094          * logic comes here for the first time, it is always false. */
3095         if (0) {
3096
3097 lock:
3098                 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3099                                       MDS_INODELOCK_UPDATE |
3100                                       MDS_INODELOCK_XATTR, LCK_EX);
3101                 if (rc != 0)
3102                         GOTO(out, rc);
3103         }
3104
3105         dt_read_lock(env, child, 0);
3106         if (unlikely(lfsck_is_dead_obj(child))) {
3107                 dt_read_unlock(env, child);
3108
3109                 GOTO(out, rc = 0);
3110         }
3111
3112         rc = dt_lookup(env, child, (struct dt_rec *)pfid,
3113                        (const struct dt_key *)dotdot, BYPASS_CAPA);
3114         if (rc != 0) {
3115                 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3116                         dt_read_unlock(env, child);
3117
3118                         GOTO(out, rc);
3119                 }
3120
3121                 if (!lustre_handle_is_used(&lh)) {
3122                         dt_read_unlock(env, child);
3123                         goto lock;
3124                 }
3125
3126                 fid_zero(pfid);
3127         } else if (lfsck->li_lpf_obj != NULL &&
3128                    lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3129                 lpf = true;
3130         } else if (unlikely(!fid_is_sane(pfid))) {
3131                 fid_zero(pfid);
3132         }
3133
3134         rc = lfsck_links_read(env, child, &ldata);
3135         dt_read_unlock(env, child);
3136         if (rc != 0) {
3137                 if (rc != -ENODATA && rc != -EINVAL)
3138                         GOTO(out, rc);
3139
3140                 if (!lustre_handle_is_used(&lh))
3141                         goto lock;
3142
3143                 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3144                         /* Remove the corrupted linkEA. */
3145                         rc = lfsck_namespace_links_remove(env, com, child);
3146                         if (rc == 0)
3147                                 /* Here, because of the crashed linkEA, we
3148                                  * cannot know whether there is some parent
3149                                  * that references the child directory via
3150                                  * some name entry or not. So keep it there,
3151                                  * when the LFSCK run next time, if there is
3152                                  * some parent that references this object,
3153                                  * then the LFSCK can rebuild the linkEA;
3154                                  * otherwise, this object will be handled
3155                                  * as orphan as above. */
3156                                 unknown = true;
3157                 } else {
3158                         /* 1. If we have neither ".." nor linkEA,
3159                          *    then it is an orphan.
3160                          *
3161                          * 2. If we only have the ".." name entry,
3162                          *    but no parent references this child
3163                          *    directory, then handle it as orphan. */
3164                         lfsck_ibits_unlock(&lh, LCK_EX);
3165                         type = LNIT_MUL_REF;
3166
3167                         /* If the LFSCK is marked as LF_INCOMPLETE,
3168                          * then means some MDT has ever tried to
3169                          * verify some remote MDT-object that resides
3170                          * on this MDT, but this MDT failed to respond
3171                          * such request. So means there may be some
3172                          * remote name entry on other MDT that
3173                          * references this object with another name,
3174                          * so we cannot know whether this linkEA is
3175                          * valid or not. So keep it there and maybe
3176                          * resolved when next LFSCK run. */
3177                         if (ns->ln_flags & LF_INCOMPLETE)
3178                                 GOTO(out, rc = 0);
3179
3180                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3181                                  "-"DFID, PFID(pfid));
3182                         rc = lfsck_namespace_insert_orphan(env, com, child,
3183                                                 info->lti_tmpbuf, "D", NULL);
3184                 }
3185
3186                 GOTO(out, rc);
3187         }
3188
3189         linkea_first_entry(&ldata);
3190         /* This is the most common case: the object has unique linkEA entry. */
3191         if (ldata.ld_leh->leh_reccount == 1) {
3192                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3193                                                 &lh, &type, &retry);
3194                 if (retry) {
3195                         LASSERT(!lustre_handle_is_used(&lh));
3196
3197                         retry = false;
3198                         goto lock;
3199                 }
3200
3201                 GOTO(out, rc);
3202         }
3203
3204         if (!lustre_handle_is_used(&lh))
3205                 goto lock;
3206
3207         if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3208                 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3209                                                 &type);
3210
3211                 GOTO(out, rc);
3212         }
3213
3214         /* When we come here, the cases usually like that:
3215          * 1) The directory object has a corrupted linkEA entry. During the
3216          *    first-stage scanning, the LFSCK cannot know such corruption,
3217          *    then it appends the right linkEA entry according to the found
3218          *    name entry after the bad one.
3219          *
3220          * 2) The directory object has a right linkEA entry. During the
3221          *    first-stage scanning, the LFSCK finds some bad name entry,
3222          *    but the LFSCK cannot aware that at that time, then it adds
3223          *    the bad linkEA entry for further processing. */
3224         rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3225                                           &lh, &type, lpf);
3226
3227         GOTO(out, rc);
3228
3229 out:
3230         lfsck_ibits_unlock(&lh, LCK_EX);
3231         if (rc > 0) {
3232                 switch (type) {
3233                 case LNIT_BAD_LINKEA:
3234                         ns->ln_linkea_repaired++;
3235                         break;
3236                 case LNIT_UNMATCHED_PAIRS:
3237                         ns->ln_unmatched_pairs_repaired++;
3238                         break;
3239                 case LNIT_MUL_REF:
3240                         ns->ln_mul_ref_repaired++;
3241                         break;
3242                 default:
3243                         break;
3244                 }
3245         }
3246
3247         if (unknown)
3248                 ns->ln_unknown_inconsistency++;
3249
3250         return rc;
3251 }
3252
3253 /**
3254  * Double scan the MDT-object for namespace LFSCK.
3255  *
3256  * If the MDT-object contains invalid or repeated linkEA entries, then drop
3257  * those entries from the linkEA; if the linkEA becomes empty or the object
3258  * has no linkEA, then it is an orphan and will be added into the directory
3259  * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3260  * the remote parent; if the name entry corresponding to some linkEA entry
3261  * is lost, then add the name entry back to the namespace.
3262  *
3263  * \param[in] env       pointer to the thread context
3264  * \param[in] com       pointer to the lfsck component
3265  * \param[in] child     pointer to the dt_object to be handled
3266  * \param[in] flags     some hints to indicate how the @child should be handled
3267  *
3268  * \retval              positive number for repaired cases
3269  * \retval              0 if nothing to be repaired
3270  * \retval              negative error number on failure
3271  */
3272 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3273                                            struct lfsck_component *com,
3274                                            struct dt_object *child, __u8 flags)
3275 {
3276         struct lfsck_thread_info *info     = lfsck_env_info(env);
3277         struct lu_attr           *la       = &info->lti_la;
3278         struct lu_name           *cname    = &info->lti_name;
3279         struct lu_fid            *pfid     = &info->lti_fid;
3280         struct lu_fid            *cfid     = &info->lti_fid2;
3281         struct lfsck_instance    *lfsck    = com->lc_lfsck;
3282         struct lfsck_namespace   *ns       = com->lc_file_ram;
3283         struct dt_object         *parent   = NULL;
3284         struct linkea_data        ldata    = { 0 };
3285         bool                      repaired = false;
3286         int                       count    = 0;
3287         int                       rc;
3288         ENTRY;
3289
3290         dt_read_lock(env, child, 0);
3291         if (unlikely(lfsck_is_dead_obj(child))) {
3292                 dt_read_unlock(env, child);
3293
3294                 RETURN(0);
3295         }
3296
3297         if (S_ISDIR(lfsck_object_type(child))) {
3298                 dt_read_unlock(env, child);
3299                 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3300
3301                 RETURN(rc);
3302         }
3303
3304         rc = lfsck_links_read(env, child, &ldata);
3305         dt_read_unlock(env, child);
3306         if (rc != 0)
3307                 GOTO(out, rc);
3308
3309         linkea_first_entry(&ldata);
3310         while (ldata.ld_lee != NULL) {
3311                 lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3312                                                     info->lti_key);
3313                 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3314                                                          false);
3315                 /* Found repeated linkEA entries */
3316                 if (rc > 0) {
3317                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3318                                                 &ldata, cname, pfid, false);
3319                         if (rc < 0)
3320                                 GOTO(out, rc);
3321
3322                         if (rc == 0)
3323                                 continue;
3324
3325                         repaired = true;
3326
3327                         /* fall through */
3328                 }
3329
3330                 /* Invalid PFID in the linkEA entry. */
3331                 if (!fid_is_sane(pfid)) {
3332                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3333                                                 &ldata, cname, pfid, true);
3334                         if (rc < 0)
3335                                 GOTO(out, rc);
3336
3337                         if (rc > 0)
3338                                 repaired = true;
3339
3340                         continue;
3341                 }
3342
3343                 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3344                 if (IS_ERR(parent))
3345                         GOTO(out, rc = PTR_ERR(parent));
3346
3347                 if (!dt_object_exists(parent)) {
3348
3349 lost_parent:
3350                         if (ldata.ld_leh->leh_reccount > 1) {
3351                                 /* If it is NOT the last linkEA entry, then
3352                                  * there is still other chance to make the
3353                                  * child to be visible via other parent, then
3354                                  * remove this linkEA entry. */
3355                                 rc = lfsck_namespace_shrink_linkea(env, com,
3356                                         child, &ldata, cname, pfid, true);
3357                         } else {
3358                                 /* If the LFSCK is marked as LF_INCOMPLETE,
3359                                  * then means some MDT has ever tried to
3360                                  * verify some remote MDT-object that resides
3361                                  * on this MDT, but this MDT failed to respond
3362                                  * such request. So means there may be some
3363                                  * remote name entry on other MDT that
3364                                  * references this object with another name,
3365                                  * so we cannot know whether this linkEA is
3366                                  * valid or not. So keep it there and maybe
3367                                  * resolved when next LFSCK run. */
3368                                 if (ns->ln_flags & LF_INCOMPLETE) {
3369                                         lfsck_object_put(env, parent);
3370
3371                                         GOTO(out, rc = 0);
3372                                 }
3373
3374                                 /* Create the lost parent as an orphan. */
3375                                 rc = lfsck_namespace_create_orphan(env, com,
3376                                                                 parent, NULL);
3377                                 if (rc < 0) {
3378                                         lfsck_object_put(env, parent);
3379
3380                                         GOTO(out, rc);
3381                                 }
3382
3383                                 if (rc > 0)
3384                                         repaired = true;
3385
3386                                 /* Add the missing name entry to the parent. */
3387                                 rc = lfsck_namespace_insert_normal(env, com,
3388                                                 parent, child, cname->ln_name);
3389                                 if (unlikely(rc == -EEXIST))
3390                                         /* Unfortunately, someone reused the
3391                                          * name under the parent by race. So we
3392                                          * have to remove the linkEA entry from
3393                                          * current child object. It means that
3394                                          * the LFSCK cannot recover the system
3395                                          * totally back to its original status,
3396                                          * but it is necessary to make the
3397                                          * current system to be consistent. */
3398                                         rc = lfsck_namespace_shrink_linkea(env,
3399                                                         com, child, &ldata,
3400                                                         cname, pfid, true);
3401                                 else
3402                                         linkea_next_entry(&ldata);
3403                         }
3404
3405                         lfsck_object_put(env, parent);
3406                         if (rc < 0)
3407                                 GOTO(out, rc);
3408
3409                         if (rc > 0)
3410                                 repaired = true;
3411
3412                         continue;
3413                 }
3414
3415                 /* The linkEA entry with bad parent will be removed. */
3416                 if (unlikely(!dt_try_as_dir(env, parent))) {
3417                         lfsck_object_put(env, parent);
3418                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3419                                                 &ldata, cname, pfid, true);
3420                         if (rc < 0)
3421                                 GOTO(out, rc);
3422
3423                         if (rc > 0)
3424                                 repaired = true;
3425
3426                         continue;
3427                 }
3428
3429                 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3430                                (const struct dt_key *)cname->ln_name,
3431                                BYPASS_CAPA);
3432                 if (rc != 0 && rc != -ENOENT) {
3433                         lfsck_object_put(env, parent);
3434
3435                         GOTO(out, rc);
3436                 }
3437
3438                 if (rc == 0) {
3439                         if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3440                                 /* It is the most common case that we
3441                                  * find the name entry corresponding
3442                                  * to the linkEA entry. */
3443                                 lfsck_object_put(env, parent);
3444                                 linkea_next_entry(&ldata);
3445                         } else {
3446                                 /* The name entry references another
3447                                  * MDT-object that may be created by
3448                                  * the LFSCK for repairing dangling
3449                                  * name entry. Try to replace it. */
3450                                 rc = lfsck_namespace_replace_cond(env, com,
3451                                                 parent, child, cfid, cname);
3452                                 lfsck_object_put(env, parent);
3453                                 if (rc < 0)
3454                                         GOTO(out, rc);
3455
3456                                 if (rc > 0) {
3457                                         repaired = true;
3458                                         linkea_next_entry(&ldata);
3459                                 } else {
3460                                         rc = lfsck_namespace_shrink_linkea(env,
3461                                                         com, child, &ldata,
3462                                                         cname, pfid, true);
3463                                         if (rc < 0)
3464                                                 GOTO(out, rc);
3465
3466                                         if (rc > 0)
3467                                                 repaired = true;
3468                                 }
3469                         }
3470
3471                         continue;
3472                 }
3473
3474                 rc = dt_attr_get(env, child, la, BYPASS_CAPA);
3475                 if (rc != 0)
3476                         GOTO(out, rc);
3477
3478                 /* If there is no name entry in the parent dir and the object
3479                  * link count is less than the linkea entries count, then the
3480                  * linkea entry should be removed. */
3481                 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3482                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
3483                                         parent, child, &ldata, cname, pfid);
3484                         lfsck_object_put(env, parent);
3485                         if (rc < 0)
3486                                 GOTO(out, rc);
3487
3488                         if (rc > 0)
3489                                 repaired = true;
3490
3491                         continue;
3492                 }
3493
3494                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3495                  * MDT has ever tried to verify some remote MDT-object that
3496                  * resides on this MDT, but this MDT failed to respond such
3497                  * request. So means there may be some remote name entry on
3498                  * other MDT that references this object with another name,
3499                  * so we cannot know whether this linkEA is valid or not.
3500                  * So keep it there and maybe resolved when next LFSCK run. */
3501                 if (ns->ln_flags & LF_INCOMPLETE) {
3502                         lfsck_object_put(env, parent);
3503
3504                         GOTO(out, rc = 0);
3505                 }
3506
3507                 rc = lfsck_namespace_check_name(env, parent, child, cname);
3508                 if (rc == -ENOENT)
3509                         goto lost_parent;
3510
3511                 if (rc < 0) {
3512                         lfsck_object_put(env, parent);
3513
3514                         GOTO(out, rc);
3515                 }
3516
3517                 /* It is an invalid name entry, drop it. */
3518                 if (unlikely(rc > 0)) {
3519                         lfsck_object_put(env, parent);
3520                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3521                                                 &ldata, cname, pfid, true);
3522                         if (rc < 0)
3523                                 GOTO(out, rc);
3524
3525                         if (rc > 0)
3526                                 repaired = true;
3527
3528                         continue;
3529                 }
3530
3531                 /* Add the missing name entry back to the namespace. */
3532                 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3533                                                    cname->ln_name);
3534                 if (unlikely(rc == -ESTALE))
3535                         /* It may happen when the remote object has been
3536                          * removed, but the local MDT is not aware of that. */
3537                         goto lost_parent;
3538
3539                 if (unlikely(rc == -EEXIST))
3540                         /* Unfortunately, someone reused the name under the
3541                          * parent by race. So we have to remove the linkEA
3542                          * entry from current child object. It means that the
3543                          * LFSCK cannot recover the system totally back to
3544                          * its original status, but it is necessary to make
3545                          * the current system to be consistent.
3546                          *
3547                          * It also may be because of the LFSCK found some
3548                          * internal status of create operation. Under such
3549                          * case, nothing to be done. */
3550                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
3551                                         parent, child, &ldata, cname, pfid);
3552                 else
3553                         linkea_next_entry(&ldata);
3554
3555                 lfsck_object_put(env, parent);
3556                 if (rc < 0)
3557                         GOTO(out, rc);
3558
3559                 if (rc > 0)
3560                         repaired = true;
3561         }
3562
3563         GOTO(out, rc = 0);
3564
3565 out:
3566         if (rc < 0 && rc != -ENODATA)
3567                 return rc;
3568
3569         if (rc == 0) {
3570                 LASSERT(ldata.ld_leh != NULL);
3571
3572                 count = ldata.ld_leh->leh_reccount;
3573         }
3574
3575         if (count == 0) {
3576                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3577                  * MDT has ever tried to verify some remote MDT-object that
3578                  * resides on this MDT, but this MDT failed to respond such
3579                  * request. So means there may be some remote name entry on
3580                  * other MDT that references this object with another name,
3581                  * so we cannot know whether this linkEA is valid or not.
3582                  * So keep it there and maybe resolved when next LFSCK run. */
3583                 if (!(ns->ln_flags & LF_INCOMPLETE)) {
3584                         /* If the child becomes orphan, then insert it into
3585                          * the global .lustre/lost+found/MDTxxxx directory. */
3586                         rc = lfsck_namespace_insert_orphan(env, com, child,
3587                                                            "", "O", &count);
3588                         if (rc < 0)
3589                                 return rc;
3590
3591                         if (rc > 0) {
3592                                 ns->ln_mul_ref_repaired++;
3593                                 repaired = true;
3594                         }
3595                 }
3596         } else {
3597                 rc = dt_attr_get(env, child, la, BYPASS_CAPA);
3598                 if (rc != 0)
3599                         return rc;
3600
3601                 if (la->la_nlink != 0 && la->la_nlink != count) {
3602                         rc = lfsck_namespace_repair_nlink(env, com, child, la);
3603                         if (rc > 0) {
3604                                 ns->ln_objs_nlink_repaired++;
3605                                 rc = 0;
3606                         }
3607                 }
3608         }
3609
3610         if (repaired) {
3611                 if (la->la_nlink > 1)
3612                         ns->ln_mul_linked_repaired++;
3613
3614                 if (rc == 0)
3615                         rc = 1;
3616         }
3617
3618         return rc;
3619 }
3620
3621 static void lfsck_namespace_dump_statistics(struct seq_file *m,
3622                                             struct lfsck_namespace *ns,
3623                                             __u64 checked_phase1,
3624                                             __u64 checked_phase2,
3625                                             __u32 time_phase1,
3626                                             __u32 time_phase2)
3627 {
3628         seq_printf(m, "checked_phase1: "LPU64"\n"
3629                       "checked_phase2: "LPU64"\n"
3630                       "updated_phase1: "LPU64"\n"
3631                       "updated_phase2: "LPU64"\n"
3632                       "failed_phase1: "LPU64"\n"
3633                       "failed_phase2: "LPU64"\n"
3634                       "directories: "LPU64"\n"
3635                       "dirent_repaired: "LPU64"\n"
3636                       "linkea_repaired: "LPU64"\n"
3637                       "nlinks_repaired: "LPU64"\n"
3638                       "multiple_linked_checked: "LPU64"\n"
3639                       "multiple_linked_repaired: "LPU64"\n"
3640                       "unknown_inconsistency: "LPU64"\n"
3641                       "unmatched_pairs_repaired: "LPU64"\n"
3642                       "dangling_repaired: "LPU64"\n"
3643                       "multiple_referenced_repaired: "LPU64"\n"
3644                       "bad_file_type_repaired: "LPU64"\n"
3645                       "lost_dirent_repaired: "LPU64"\n"
3646                       "local_lost_found_scanned: "LPU64"\n"
3647                       "local_lost_found_moved: "LPU64"\n"
3648                       "local_lost_found_skipped: "LPU64"\n"
3649                       "local_lost_found_failed: "LPU64"\n"
3650                       "striped_dirs_scanned: "LPU64"\n"
3651                       "striped_dirs_repaired: "LPU64"\n"
3652                       "striped_dirs_failed: "LPU64"\n"
3653                       "striped_dirs_disabled: "LPU64"\n"
3654                       "striped_dirs_skipped: "LPU64"\n"
3655                       "striped_shards_scanned: "LPU64"\n"
3656                       "striped_shards_repaired: "LPU64"\n"
3657                       "striped_shards_failed: "LPU64"\n"
3658                       "striped_shards_skipped: "LPU64"\n"
3659                       "name_hash_repaired: "LPU64"\n"
3660                       "success_count: %u\n"
3661                       "run_time_phase1: %u seconds\n"
3662                       "run_time_phase2: %u seconds\n",
3663                       checked_phase1,
3664                       checked_phase2,
3665                       ns->ln_items_repaired,
3666                       ns->ln_objs_repaired_phase2,
3667                       ns->ln_items_failed,
3668                       ns->ln_objs_failed_phase2,
3669                       ns->ln_dirs_checked,
3670                       ns->ln_dirent_repaired,
3671                       ns->ln_linkea_repaired,
3672                       ns->ln_objs_nlink_repaired,
3673                       ns->ln_mul_linked_checked,
3674                       ns->ln_mul_linked_repaired,
3675                       ns->ln_unknown_inconsistency,
3676                       ns->ln_unmatched_pairs_repaired,
3677                       ns->ln_dangling_repaired,
3678                       ns->ln_mul_ref_repaired,
3679                       ns->ln_bad_type_repaired,
3680                       ns->ln_lost_dirent_repaired,
3681                       ns->ln_local_lpf_scanned,
3682                       ns->ln_local_lpf_moved,
3683                       ns->ln_local_lpf_skipped,
3684                       ns->ln_local_lpf_failed,
3685                       ns->ln_striped_dirs_scanned,
3686                       ns->ln_striped_dirs_repaired,
3687                       ns->ln_striped_dirs_failed,
3688                       ns->ln_striped_dirs_disabled,
3689                       ns->ln_striped_dirs_skipped,
3690                       ns->ln_striped_shards_scanned,
3691                       ns->ln_striped_shards_repaired,
3692                       ns->ln_striped_shards_failed,
3693                       ns->ln_striped_shards_skipped,
3694                       ns->ln_name_hash_repaired,
3695                       ns->ln_success_count,
3696                       time_phase1,
3697                       time_phase2);
3698 }
3699
3700 static void lfsck_namespace_release_lmv(const struct lu_env *env,
3701                                         struct lfsck_component *com)
3702 {
3703         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3704         struct lfsck_namespace          *ns     = com->lc_file_ram;
3705
3706         while (!list_empty(&lfsck->li_list_lmv)) {
3707                 struct lfsck_lmv_unit   *llu;
3708                 struct lfsck_lmv        *llmv;
3709
3710                 llu = list_entry(lfsck->li_list_lmv.next,
3711                                  struct lfsck_lmv_unit, llu_link);
3712                 llmv = &llu->llu_lmv;
3713
3714                 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
3715                          "still in using: %u\n",
3716                          atomic_read(&llmv->ll_ref));
3717
3718                 ns->ln_striped_dirs_skipped++;
3719                 lfsck_lmv_put(env, llmv);
3720         }
3721 }
3722
3723 /* namespace APIs */
3724
3725 static int lfsck_namespace_reset(const struct lu_env *env,
3726                                  struct lfsck_component *com, bool init)
3727 {
3728         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3729         struct lfsck_namespace          *ns     = com->lc_file_ram;
3730         struct lfsck_assistant_data     *lad    = com->lc_data;
3731         struct dt_object                *root;
3732         struct dt_object                *dto;
3733         int                              rc;
3734         ENTRY;
3735
3736         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
3737         if (IS_ERR(root))
3738                 GOTO(log, rc = PTR_ERR(root));
3739
3740         if (unlikely(!dt_try_as_dir(env, root)))
3741                 GOTO(put, rc = -ENOTDIR);
3742
3743         down_write(&com->lc_sem);
3744         if (init) {
3745                 memset(ns, 0, sizeof(*ns));
3746         } else {
3747                 __u32 count = ns->ln_success_count;
3748                 __u64 last_time = ns->ln_time_last_complete;
3749
3750                 memset(ns, 0, sizeof(*ns));
3751                 ns->ln_success_count = count;
3752                 ns->ln_time_last_complete = last_time;
3753         }
3754         ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
3755         ns->ln_status = LS_INIT;
3756
3757         rc = local_object_unlink(env, lfsck->li_bottom, root,
3758                                  lfsck_namespace_name);
3759         if (rc != 0)
3760                 GOTO(out, rc);
3761
3762         lfsck_object_put(env, com->lc_obj);
3763         com->lc_obj = NULL;
3764         dto = local_index_find_or_create(env, lfsck->li_los, root,
3765                                          lfsck_namespace_name,
3766                                          S_IFREG | S_IRUGO | S_IWUSR,
3767                                          &dt_lfsck_features);
3768         if (IS_ERR(dto))
3769                 GOTO(out, rc = PTR_ERR(dto));
3770
3771         com->lc_obj = dto;
3772         rc = dto->do_ops->do_index_try(env, dto, &dt_lfsck_features);
3773         if (rc != 0)
3774                 GOTO(out, rc);
3775
3776         lad->lad_incomplete = 0;
3777         CFS_RESET_BITMAP(lad->lad_bitmap);
3778
3779         rc = lfsck_namespace_store(env, com, true);
3780
3781         GOTO(out, rc);
3782
3783 out:
3784         up_write(&com->lc_sem);
3785
3786 put:
3787         lu_object_put(env, &root->do_lu);
3788 log:
3789         CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
3790                lfsck_lfsck2name(lfsck), rc);
3791         return rc;
3792 }
3793
3794 static void
3795 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
3796                      bool new_checked)
3797 {
3798         struct lfsck_namespace *ns = com->lc_file_ram;
3799
3800         down_write(&com->lc_sem);
3801         if (new_checked)
3802                 com->lc_new_checked++;
3803         lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
3804         up_write(&com->lc_sem);
3805 }
3806
3807 static void lfsck_namespace_close_dir(const struct lu_env *env,
3808                                       struct lfsck_component *com)
3809 {
3810         struct lfsck_namespace          *ns     = com->lc_file_ram;
3811         struct lfsck_assistant_data     *lad    = com->lc_data;
3812         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3813         struct lfsck_lmv                *llmv   = lfsck->li_lmv;
3814         struct lfsck_namespace_req      *lnr;
3815         __u32                            size   =
3816                                 sizeof(*lnr) + LFSCK_TMPBUF_LEN;
3817         bool                             wakeup = false;
3818         ENTRY;
3819
3820         if (llmv == NULL)
3821                 RETURN_EXIT;
3822
3823         OBD_ALLOC(lnr, size);
3824         if (lnr == NULL) {
3825                 ns->ln_striped_dirs_skipped++;
3826
3827                 RETURN_EXIT;
3828         }
3829
3830         /* Generate a dummy request to indicate that all shards' name entry
3831          * in this striped directory has been scanned for the first time. */
3832         INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
3833         lnr->lnr_obj = lfsck_object_get(lfsck->li_obj_dir);
3834         lnr->lnr_lmv = lfsck_lmv_get(llmv);
3835         lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
3836         lnr->lnr_oit_cookie = lfsck->li_pos_current.lp_oit_cookie;
3837         lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
3838         lnr->lnr_size = size;
3839
3840         spin_lock(&lad->lad_lock);
3841         if (lad->lad_assistant_status < 0) {
3842                 spin_unlock(&lad->lad_lock);
3843                 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
3844                 ns->ln_striped_dirs_skipped++;
3845
3846                 RETURN_EXIT;
3847         }
3848
3849         list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
3850         if (lad->lad_prefetched == 0)
3851                 wakeup = true;
3852
3853         lad->lad_prefetched++;
3854         spin_unlock(&lad->lad_lock);
3855         if (wakeup)
3856                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
3857
3858         EXIT;
3859 }
3860
3861 static int lfsck_namespace_open_dir(const struct lu_env *env,
3862                                     struct lfsck_component *com)
3863 {
3864         struct lfsck_instance   *lfsck  = com->lc_lfsck;
3865         struct lfsck_namespace  *ns     = com->lc_file_ram;
3866         struct lfsck_lmv        *llmv   = lfsck->li_lmv;
3867         int                      rc     = 0;
3868         ENTRY;
3869
3870         if (llmv == NULL)
3871                 RETURN(0);
3872
3873         if (llmv->ll_lmv_master) {
3874                 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
3875
3876                 if (lmv->lmv_master_mdt_index !=
3877                     lfsck_dev_idx(lfsck->li_bottom)) {
3878                         lmv->lmv_master_mdt_index =
3879                                 lfsck_dev_idx(lfsck->li_bottom);
3880                         ns->ln_flags |= LF_INCONSISTENT;
3881                         llmv->ll_lmv_updated = 1;
3882                 }
3883         } else {
3884                 rc = lfsck_namespace_verify_stripe_slave(env, com,
3885                                         lfsck->li_obj_dir, llmv);
3886         }
3887
3888         RETURN(rc > 0 ? 0 : rc);
3889 }
3890
3891 static int lfsck_namespace_checkpoint(const struct lu_env *env,
3892                                       struct lfsck_component *com, bool init)
3893 {
3894         struct lfsck_instance   *lfsck = com->lc_lfsck;
3895         struct lfsck_namespace  *ns    = com->lc_file_ram;
3896         int                      rc;
3897
3898         if (!init) {
3899                 rc = lfsck_checkpoint_generic(env, com);
3900                 if (rc != 0)
3901                         goto log;
3902         }
3903
3904         down_write(&com->lc_sem);
3905         if (init) {
3906                 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
3907         } else {
3908                 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
3909                 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3910                                 HALF_SEC - lfsck->li_time_last_checkpoint);
3911                 ns->ln_time_last_checkpoint = cfs_time_current_sec();
3912                 ns->ln_items_checked += com->lc_new_checked;
3913                 com->lc_new_checked = 0;
3914         }
3915
3916         rc = lfsck_namespace_store(env, com, false);
3917         up_write(&com->lc_sem);
3918
3919 log:
3920         CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos ["LPU64
3921                ", "DFID", "LPX64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
3922                lfsck->li_pos_current.lp_oit_cookie,
3923                PFID(&lfsck->li_pos_current.lp_dir_parent),
3924                lfsck->li_pos_current.lp_dir_cookie, rc);
3925
3926         return rc > 0 ? 0 : rc;
3927 }
3928
3929 static int lfsck_namespace_prep(const struct lu_env *env,
3930                                 struct lfsck_component *com,
3931                                 struct lfsck_start_param *lsp)
3932 {
3933         struct lfsck_instance   *lfsck  = com->lc_lfsck;
3934         struct lfsck_namespace  *ns     = com->lc_file_ram;
3935         struct lfsck_position   *pos    = &com->lc_pos_start;
3936         int                      rc;
3937
3938         rc = lfsck_namespace_load_bitmap(env, com);
3939         if (rc != 0 || ns->ln_status == LS_COMPLETED) {
3940                 rc = lfsck_namespace_reset(env, com, false);
3941                 if (rc == 0)
3942                         rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
3943
3944                 if (rc != 0) {
3945                         CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
3946                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
3947
3948                         return rc;
3949                 }
3950         }
3951
3952         down_write(&com->lc_sem);
3953         ns->ln_time_latest_start = cfs_time_current_sec();
3954         spin_lock(&lfsck->li_lock);
3955
3956         if (ns->ln_flags & LF_SCANNED_ONCE) {
3957                 if (!lfsck->li_drop_dryrun ||
3958                     lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
3959                         ns->ln_status = LS_SCANNING_PHASE2;
3960                         list_move_tail(&com->lc_link,
3961                                        &lfsck->li_list_double_scan);
3962                         if (!list_empty(&com->lc_link_dir))
3963                                 list_del_init(&com->lc_link_dir);
3964                         lfsck_pos_set_zero(pos);
3965                 } else {
3966                         ns->ln_status = LS_SCANNING_PHASE1;
3967                         ns->ln_run_time_phase1 = 0;
3968                         ns->ln_run_time_phase2 = 0;
3969                         ns->ln_items_checked = 0;
3970                         ns->ln_items_repaired = 0;
3971                         ns->ln_items_failed = 0;
3972                         ns->ln_dirs_checked = 0;
3973                         ns->ln_objs_checked_phase2 = 0;
3974                         ns->ln_objs_repaired_phase2 = 0;
3975                         ns->ln_objs_failed_phase2 = 0;
3976                         ns->ln_objs_nlink_repaired = 0;
3977                         ns->ln_dirent_repaired = 0;
3978                         ns->ln_linkea_repaired = 0;
3979                         ns->ln_mul_linked_checked = 0;
3980                         ns->ln_mul_linked_repaired = 0;
3981                         ns->ln_unknown_inconsistency = 0;
3982                         ns->ln_unmatched_pairs_repaired = 0;
3983                         ns->ln_dangling_repaired = 0;
3984                         ns->ln_mul_ref_repaired = 0;
3985                         ns->ln_bad_type_repaired = 0;
3986                         ns->ln_lost_dirent_repaired = 0;
3987                         ns->ln_striped_dirs_scanned = 0;
3988                         ns->ln_striped_dirs_repaired = 0;
3989                         ns->ln_striped_dirs_failed = 0;
3990                         ns->ln_striped_dirs_disabled = 0;
3991                         ns->ln_striped_dirs_skipped = 0;
3992                         ns->ln_striped_shards_scanned = 0;
3993                         ns->ln_striped_shards_repaired = 0;
3994                         ns->ln_striped_shards_failed = 0;
3995                         ns->ln_striped_shards_skipped = 0;
3996                         ns->ln_name_hash_repaired = 0;
3997                         fid_zero(&ns->ln_fid_latest_scanned_phase2);
3998                         if (list_empty(&com->lc_link_dir))
3999                                 list_add_tail(&com->lc_link_dir,
4000                                               &lfsck->li_list_dir);
4001                         *pos = ns->ln_pos_first_inconsistent;
4002                 }
4003         } else {
4004                 ns->ln_status = LS_SCANNING_PHASE1;
4005                 if (list_empty(&com->lc_link_dir))
4006                         list_add_tail(&com->lc_link_dir,
4007                                       &lfsck->li_list_dir);
4008                 if (!lfsck->li_drop_dryrun ||
4009                     lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4010                         *pos = ns->ln_pos_last_checkpoint;
4011                         pos->lp_oit_cookie++;
4012                 } else {
4013                         *pos = ns->ln_pos_first_inconsistent;
4014                 }
4015         }
4016
4017         spin_unlock(&lfsck->li_lock);
4018         up_write(&com->lc_sem);
4019
4020         rc = lfsck_start_assistant(env, com, lsp);
4021
4022         CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos ["LPU64", "
4023                DFID", "LPX64"]: rc = %d\n",
4024                lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
4025                PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
4026
4027         return rc;
4028 }
4029
4030 static int lfsck_namespace_exec_oit(const struct lu_env *env,
4031                                     struct lfsck_component *com,
4032                                     struct dt_object *obj)
4033 {
4034         struct lfsck_thread_info *info  = lfsck_env_info(env);
4035         struct lfsck_namespace   *ns    = com->lc_file_ram;
4036         struct lfsck_instance    *lfsck = com->lc_lfsck;
4037         const struct lu_fid      *fid   = lfsck_dto2fid(obj);
4038         struct lu_attr           *la    = &info->lti_la;
4039         struct lu_fid            *pfid  = &info->lti_fid2;
4040         struct lu_name           *cname = &info->lti_name;
4041         struct lu_seq_range      *range = &info->lti_range;
4042         struct dt_device         *dev   = lfsck->li_bottom;
4043         struct seq_server_site   *ss    =
4044                                 lu_site2seq(dev->dd_lu_dev.ld_site);
4045         struct linkea_data        ldata = { 0 };
4046         __u32                     idx   = lfsck_dev_idx(dev);
4047         int                       rc;
4048         ENTRY;
4049
4050         rc = lfsck_links_read(env, obj, &ldata);
4051         if (rc == -ENOENT)
4052                 GOTO(out, rc = 0);
4053
4054         /* -EINVAL means crashed linkEA, should be verified. */
4055         if (rc == -EINVAL) {
4056                 rc = lfsck_namespace_trace_update(env, com, fid,
4057                                                   LNTF_CHECK_LINKEA, true);
4058                 if (rc == 0) {
4059                         struct lustre_handle lh = { 0 };
4060
4061                         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4062                                               MDS_INODELOCK_UPDATE |
4063                                               MDS_INODELOCK_XATTR, LCK_EX);
4064                         if (rc == 0) {
4065                                 rc = lfsck_namespace_links_remove(env, com,
4066                                                                   obj);
4067                                 lfsck_ibits_unlock(&lh, LCK_EX);
4068                         }
4069                 }
4070
4071                 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4072         }
4073
4074         /* zero-linkEA object may be orphan, but it also maybe because
4075          * of upgrading. Currently, we cannot record it for double scan.
4076          * Because it may cause the LFSCK trace file to be too large. */
4077         if (rc == -ENODATA) {
4078                 if (S_ISDIR(lfsck_object_type(obj)))
4079                         GOTO(out, rc = 0);
4080
4081                 rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
4082                 if (rc != 0)
4083                         GOTO(out, rc);
4084
4085                 /* "la_ctime" == 1 means that it has ever been removed from
4086                  * backend /lost+found directory but not been added back to
4087                  * the normal namespace yet. */
4088                 if (la->la_nlink > 1 || unlikely(la->la_ctime == 1))
4089                         rc = lfsck_namespace_trace_update(env, com, fid,
4090                                                 LNTF_CHECK_LINKEA, true);
4091
4092                 GOTO(out, rc);
4093         }
4094
4095         if (rc != 0)
4096                 GOTO(out, rc);
4097
4098         /* Record multiple-linked object. */
4099         if (ldata.ld_leh->leh_reccount > 1) {
4100                 rc = lfsck_namespace_trace_update(env, com, fid,
4101                                                   LNTF_CHECK_LINKEA, true);
4102
4103                 GOTO(out, rc);
4104         }
4105
4106         linkea_first_entry(&ldata);
4107         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
4108         if (!fid_is_sane(pfid)) {
4109                 rc = lfsck_namespace_trace_update(env, com, fid,
4110                                                   LNTF_CHECK_PARENT, true);
4111         } else {
4112                 fld_range_set_mdt(range);
4113                 rc = fld_local_lookup(env, ss->ss_server_fld,
4114                                       fid_seq(pfid), range);
4115                 if ((rc == -ENOENT) ||
4116                     (rc == 0 && range->lsr_index != idx)) {
4117                         rc = lfsck_namespace_trace_update(env, com, fid,
4118                                                 LNTF_CHECK_LINKEA, true);
4119                 } else {
4120                         if (S_ISDIR(lfsck_object_type(obj)))
4121                                 GOTO(out, rc = 0);
4122
4123                         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
4124                         if (rc != 0)
4125                                 GOTO(out, rc);
4126
4127                         /* "la_ctime" == 1 means that it has ever been
4128                          * removed from backend /lost+found directory but
4129                          * not been added back to the normal namespace yet. */
4130                         if (la->la_nlink > 1 || unlikely(la->la_ctime == 1))
4131                                 rc = lfsck_namespace_trace_update(env, com,
4132                                                 fid, LNTF_CHECK_LINKEA, true);
4133                 }
4134         }
4135
4136         GOTO(out, rc);
4137
4138 out:
4139         down_write(&com->lc_sem);
4140         com->lc_new_checked++;
4141         if (S_ISDIR(lfsck_object_type(obj)))
4142                 ns->ln_dirs_checked++;
4143         if (rc != 0)
4144                 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4145         up_write(&com->lc_sem);
4146
4147         return rc;
4148 }
4149
4150 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4151                                     struct lfsck_component *com,
4152                                     struct lu_dirent *ent, __u16 type)
4153 {
4154         struct lfsck_assistant_data     *lad     = com->lc_data;
4155         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4156         struct lfsck_namespace_req      *lnr;
4157         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
4158         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4159         struct ptlrpc_thread            *athread = &lad->lad_thread;
4160         struct l_wait_info               lwi     = { 0 };
4161         bool                             wakeup  = false;
4162
4163         l_wait_event(mthread->t_ctl_waitq,
4164                      bk->lb_async_windows == 0 ||
4165                      lad->lad_prefetched < bk->lb_async_windows ||
4166                      !thread_is_running(mthread) ||
4167                      thread_is_stopped(athread),
4168                      &lwi);
4169
4170         if (unlikely(!thread_is_running(mthread)) ||
4171                      thread_is_stopped(athread))
4172                 return 0;
4173
4174         if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4175                 return 0;
4176
4177         lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, ent, type);
4178         if (IS_ERR(lnr)) {
4179                 struct lfsck_namespace *ns = com->lc_file_ram;
4180
4181                 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4182                 return PTR_ERR(lnr);
4183         }
4184
4185         spin_lock(&lad->lad_lock);
4186         if (lad->lad_assistant_status < 0) {
4187                 spin_unlock(&lad->lad_lock);
4188                 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4189                 return lad->lad_assistant_status;
4190         }
4191
4192         list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4193         if (lad->lad_prefetched == 0)
4194                 wakeup = true;
4195
4196         lad->lad_prefetched++;
4197         spin_unlock(&lad->lad_lock);
4198         if (wakeup)
4199                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4200
4201         down_write(&com->lc_sem);
4202         com->lc_new_checked++;
4203         up_write(&com->lc_sem);
4204
4205         return 0;
4206 }
4207
4208 static int lfsck_namespace_post(const struct lu_env *env,
4209                                 struct lfsck_component *com,
4210                                 int result, bool init)
4211 {
4212         struct lfsck_instance   *lfsck = com->lc_lfsck;
4213         struct lfsck_namespace  *ns    = com->lc_file_ram;
4214         int                      rc;
4215         ENTRY;
4216
4217         lfsck_post_generic(env, com, &result);
4218
4219         down_write(&com->lc_sem);
4220         lfsck_namespace_release_lmv(env, com);
4221
4222         spin_lock(&lfsck->li_lock);
4223         if (!init)
4224                 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4225         if (result > 0) {
4226                 ns->ln_status = LS_SCANNING_PHASE2;
4227                 ns->ln_flags |= LF_SCANNED_ONCE;
4228                 ns->ln_flags &= ~LF_UPGRADE;
4229                 list_del_init(&com->lc_link_dir);
4230                 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4231         } else if (result == 0) {
4232                 if (lfsck->li_status != 0)
4233                         ns->ln_status = lfsck->li_status;
4234                 else
4235                         ns->ln_status = LS_STOPPED;
4236                 if (ns->ln_status != LS_PAUSED) {
4237                         list_del_init(&com->lc_link_dir);
4238                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4239                 }
4240         } else {
4241                 ns->ln_status = LS_FAILED;
4242                 list_del_init(&com->lc_link_dir);
4243                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4244         }
4245         spin_unlock(&lfsck->li_lock);
4246
4247         if (!init) {
4248                 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4249                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4250                 ns->ln_time_last_checkpoint = cfs_time_current_sec();
4251                 ns->ln_items_checked += com->lc_new_checked;
4252                 com->lc_new_checked = 0;
4253         }
4254
4255         rc = lfsck_namespace_store(env, com, false);
4256         up_write(&com->lc_sem);
4257
4258         CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4259                lfsck_lfsck2name(lfsck), rc);
4260
4261         RETURN(rc);
4262 }
4263
4264 static int
4265 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4266                      struct seq_file *m)
4267 {
4268         struct lfsck_instance   *lfsck = com->lc_lfsck;
4269         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
4270         struct lfsck_namespace  *ns    = com->lc_file_ram;
4271         int                      rc;
4272
4273         down_read(&com->lc_sem);
4274         seq_printf(m, "name: lfsck_namespace\n"
4275                    "magic: %#x\n"
4276                    "version: %d\n"
4277                    "status: %s\n",
4278                    ns->ln_magic,
4279                    bk->lb_version,
4280                    lfsck_status2names(ns->ln_status));
4281
4282         rc = lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4283         if (rc < 0)
4284                 goto out;
4285
4286         rc = lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4287         if (rc < 0)
4288                 goto out;
4289
4290         rc = lfsck_time_dump(m, ns->ln_time_last_complete,
4291                              "time_since_last_completed");
4292         if (rc < 0)
4293                 goto out;
4294
4295         rc = lfsck_time_dump(m, ns->ln_time_latest_start,
4296                              "time_since_latest_start");
4297         if (rc < 0)
4298                 goto out;
4299
4300         rc = lfsck_time_dump(m, ns->ln_time_last_checkpoint,
4301                              "time_since_last_checkpoint");
4302         if (rc < 0)
4303                 goto out;
4304
4305         rc = lfsck_pos_dump(m, &ns->ln_pos_latest_start,
4306                             "latest_start_position");
4307         if (rc < 0)
4308                 goto out;
4309
4310         rc = lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4311                             "last_checkpoint_position");
4312         if (rc < 0)
4313                 goto out;
4314
4315         rc = lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4316                             "first_failure_position");
4317         if (rc < 0)
4318                 goto out;
4319
4320         if (ns->ln_status == LS_SCANNING_PHASE1) {
4321                 struct lfsck_position pos;
4322                 const struct dt_it_ops *iops;
4323                 cfs_duration_t duration = cfs_time_current() -
4324                                           lfsck->li_time_last_checkpoint;
4325                 __u64 checked = ns->ln_items_checked + com->lc_new_checked;
4326                 __u64 speed = checked;
4327                 __u64 new_checked = com->lc_new_checked *
4328                                     msecs_to_jiffies(MSEC_PER_SEC);
4329                 __u32 rtime = ns->ln_run_time_phase1 +
4330                               cfs_duration_sec(duration + HALF_SEC);
4331
4332                 if (duration != 0)
4333                         do_div(new_checked, duration);
4334                 if (rtime != 0)
4335                         do_div(speed, rtime);
4336                 lfsck_namespace_dump_statistics(m, ns, checked,
4337                                                 ns->ln_objs_checked_phase2,
4338                                                 rtime, ns->ln_run_time_phase2);
4339
4340                 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4341                               "average_speed_phase2: N/A\n"
4342                               "real_time_speed_phase1: "LPU64" items/sec\n"
4343                               "real_time_speed_phase2: N/A\n",
4344                               speed,
4345                               new_checked);
4346
4347                 LASSERT(lfsck->li_di_oit != NULL);
4348
4349                 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
4350
4351                 /* The low layer otable-based iteration position may NOT
4352                  * exactly match the namespace-based directory traversal
4353                  * cookie. Generally, it is not a serious issue. But the
4354                  * caller should NOT make assumption on that. */
4355                 pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4356                 if (!lfsck->li_current_oit_processed)
4357                         pos.lp_oit_cookie--;
4358
4359                 spin_lock(&lfsck->li_lock);
4360                 if (lfsck->li_di_dir != NULL) {
4361                         pos.lp_dir_cookie = lfsck->li_cookie_dir;
4362                         if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4363                                 fid_zero(&pos.lp_dir_parent);
4364                                 pos.lp_dir_cookie = 0;
4365                         } else {
4366                                 pos.lp_dir_parent =
4367                                         *lfsck_dto2fid(lfsck->li_obj_dir);
4368                         }
4369                 } else {
4370                         fid_zero(&pos.lp_dir_parent);
4371                         pos.lp_dir_cookie = 0;
4372                 }
4373                 spin_unlock(&lfsck->li_lock);
4374                 lfsck_pos_dump(m, &pos, "current_position");
4375         } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4376                 cfs_duration_t duration = cfs_time_current() -
4377                                           lfsck->li_time_last_checkpoint;
4378                 __u64 checked = ns->ln_objs_checked_phase2 +
4379                                 com->lc_new_checked;
4380                 __u64 speed1 = ns->ln_items_checked;
4381                 __u64 speed2 = checked;
4382                 __u64 new_checked = com->lc_new_checked *
4383                                     msecs_to_jiffies(MSEC_PER_SEC);
4384                 __u32 rtime = ns->ln_run_time_phase2 +
4385                               cfs_duration_sec(duration + HALF_SEC);
4386
4387                 if (duration != 0)
4388                         do_div(new_checked, duration);
4389                 if (ns->ln_run_time_phase1 != 0)
4390                         do_div(speed1, ns->ln_run_time_phase1);
4391                 if (rtime != 0)
4392                         do_div(speed2, rtime);
4393                 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4394                                                 checked,
4395                                                 ns->ln_run_time_phase1, rtime);
4396
4397                 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4398                               "average_speed_phase2: "LPU64" objs/sec\n"
4399                               "real_time_speed_phase1: N/A\n"
4400                               "real_time_speed_phase2: "LPU64" objs/sec\n"
4401                               "current_position: "DFID"\n",
4402                               speed1,
4403                               speed2,
4404                               new_checked,
4405                               PFID(&ns->ln_fid_latest_scanned_phase2));
4406         } else {
4407                 __u64 speed1 = ns->ln_items_checked;
4408                 __u64 speed2 = ns->ln_objs_checked_phase2;
4409
4410                 if (ns->ln_run_time_phase1 != 0)
4411                         do_div(speed1, ns->ln_run_time_phase1);
4412                 if (ns->ln_run_time_phase2 != 0)
4413                         do_div(speed2, ns->ln_run_time_phase2);
4414                 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4415                                                 ns->ln_objs_checked_phase2,
4416                                                 ns->ln_run_time_phase1,
4417                                                 ns->ln_run_time_phase2);
4418
4419                 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4420                               "average_speed_phase2: "LPU64" objs/sec\n"
4421                               "real_time_speed_phase1: N/A\n"
4422                               "real_time_speed_phase2: N/A\n"
4423                               "current_position: N/A\n",
4424                               speed1,
4425                               speed2);
4426         }
4427 out:
4428         up_read(&com->lc_sem);
4429         return 0;
4430 }
4431
4432 static int lfsck_namespace_double_scan(const struct lu_env *env,
4433                                        struct lfsck_component *com)
4434 {
4435         struct lfsck_namespace          *ns     = com->lc_file_ram;
4436         struct lfsck_assistant_data     *lad    = com->lc_data;
4437         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4438         struct lfsck_tgt_desc           *ltd;
4439         struct lfsck_tgt_desc           *next;
4440         int                              rc;
4441
4442         rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4443         if (thread_is_stopped(&lad->lad_thread)) {
4444                 LASSERT(list_empty(&lad->lad_req_list));
4445                 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4446
4447                 spin_lock(&ltds->ltd_lock);
4448                 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4449                                          ltd_namespace_phase_list) {
4450                         list_del_init(&ltd->ltd_namespace_phase_list);
4451                 }
4452                 spin_unlock(&ltds->ltd_lock);
4453         }
4454
4455         return rc;
4456 }
4457
4458 static void lfsck_namespace_data_release(const struct lu_env *env,
4459                                          struct lfsck_component *com)
4460 {
4461         struct lfsck_assistant_data     *lad    = com->lc_data;
4462         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4463         struct lfsck_tgt_desc           *ltd;
4464         struct lfsck_tgt_desc           *next;
4465
4466         LASSERT(lad != NULL);
4467         LASSERT(thread_is_init(&lad->lad_thread) ||
4468                 thread_is_stopped(&lad->lad_thread));
4469         LASSERT(list_empty(&lad->lad_req_list));
4470
4471         com->lc_data = NULL;
4472         lfsck_namespace_release_lmv(env, com);
4473
4474         spin_lock(&ltds->ltd_lock);
4475         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4476                                  ltd_namespace_phase_list) {
4477                 list_del_init(&ltd->ltd_namespace_phase_list);
4478         }
4479         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4480                                  ltd_namespace_phase_list) {
4481                 list_del_init(&ltd->ltd_namespace_phase_list);
4482         }
4483         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4484                                  ltd_namespace_list) {
4485                 list_del_init(&ltd->ltd_namespace_list);
4486         }
4487         spin_unlock(&ltds->ltd_lock);
4488
4489         if (likely(lad->lad_bitmap != NULL))
4490                 CFS_FREE_BITMAP(lad->lad_bitmap);
4491
4492         OBD_FREE_PTR(lad);
4493 }
4494
4495 static void lfsck_namespace_quit(const struct lu_env *env,
4496                                  struct lfsck_component *com)
4497 {
4498         struct lfsck_assistant_data     *lad    = com->lc_data;
4499         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4500         struct lfsck_tgt_desc           *ltd;
4501         struct lfsck_tgt_desc           *next;
4502
4503         LASSERT(lad != NULL);
4504
4505         lfsck_quit_generic(env, com);
4506
4507         LASSERT(thread_is_init(&lad->lad_thread) ||
4508                 thread_is_stopped(&lad->lad_thread));
4509         LASSERT(list_empty(&lad->lad_req_list));
4510
4511         lfsck_namespace_release_lmv(env, com);
4512
4513         spin_lock(&ltds->ltd_lock);
4514         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4515                                  ltd_namespace_phase_list) {
4516                 list_del_init(&ltd->ltd_namespace_phase_list);
4517         }
4518         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4519                                  ltd_namespace_phase_list) {
4520                 list_del_init(&ltd->ltd_namespace_phase_list);
4521         }
4522         spin_unlock(&ltds->ltd_lock);
4523 }
4524
4525 static int lfsck_namespace_in_notify(const struct lu_env *env,
4526                                      struct lfsck_component *com,
4527                                      struct lfsck_request *lr,
4528                                      struct thandle *th)
4529 {
4530         struct lfsck_instance           *lfsck = com->lc_lfsck;
4531         struct lfsck_namespace          *ns    = com->lc_file_ram;
4532         struct lfsck_assistant_data     *lad   = com->lc_data;
4533         struct lfsck_tgt_descs          *ltds  = &lfsck->li_mdt_descs;
4534         struct lfsck_tgt_desc           *ltd;
4535         int                              rc;
4536         bool                             fail  = false;
4537         ENTRY;
4538
4539         switch (lr->lr_event) {
4540         case LE_CREATE_ORPHAN: {
4541                 struct dt_object *orphan = NULL;
4542                 struct lmv_mds_md_v1 *lmv;
4543
4544                 CDEBUG(D_LFSCK, "%s: namespace LFSCK handling notify from "
4545                        "MDT %x to create orphan"DFID" with type %o\n",
4546                        lfsck_lfsck2name(lfsck), lr->lr_index,
4547                        PFID(&lr->lr_fid), lr->lr_type);
4548
4549                 orphan = lfsck_object_find(env, lfsck, &lr->lr_fid);
4550                 if (IS_ERR(orphan))
4551                         GOTO(out_create, rc = PTR_ERR(orphan));
4552
4553                 if (dt_object_exists(orphan))
4554                         GOTO(out_create, rc = -EEXIST);
4555
4556                 if (lr->lr_stripe_count > 0) {
4557                         lmv = &lfsck_env_info(env)->lti_lmv;
4558                         memset(lmv, 0, sizeof(*lmv));
4559                         lmv->lmv_hash_type = lr->lr_hash_type;
4560                         lmv->lmv_stripe_count = lr->lr_stripe_count;
4561                         lmv->lmv_layout_version = lr->lr_layout_version;
4562                         memcpy(lmv->lmv_pool_name, lr->lr_pool_name,
4563                                sizeof(lmv->lmv_pool_name));
4564                 } else {
4565                         lmv = NULL;
4566                 }
4567
4568                 rc = lfsck_namespace_create_orphan_local(env, com, orphan,
4569                                                          lr->lr_type, lmv);
4570
4571                 GOTO(out_create, rc = (rc == 1) ? 0 : rc);
4572
4573 out_create:
4574                 CDEBUG(D_LFSCK, "%s: namespace LFSCK handled notify from "
4575                        "MDT %x to create orphan"DFID" with type %o: rc = %d\n",
4576                        lfsck_lfsck2name(lfsck), lr->lr_index,
4577                        PFID(&lr->lr_fid), lr->lr_type, rc);
4578
4579                 if (orphan != NULL && !IS_ERR(orphan))
4580                         lfsck_object_put(env, orphan);
4581
4582                 return rc;
4583         }
4584         case LE_SKIP_NLINK_DECLARE: {
4585                 struct dt_object        *obj   = com->lc_obj;
4586                 struct lu_fid           *key   = &lfsck_env_info(env)->lti_fid3;
4587                 __u8                     flags = 0;
4588
4589                 LASSERT(th != NULL);
4590
4591                 rc = dt_declare_delete(env, obj,
4592                                        (const struct dt_key *)key, th);
4593                 if (rc == 0)
4594                         rc = dt_declare_insert(env, obj,
4595                                                (const struct dt_rec *)&flags,
4596                                                (const struct dt_key *)key, th);
4597
4598                 RETURN(rc);
4599         }
4600         case LE_SKIP_NLINK: {
4601                 struct dt_object        *obj   = com->lc_obj;
4602                 struct lu_fid           *key   = &lfsck_env_info(env)->lti_fid3;
4603                 __u8                     flags = 0;
4604                 bool                     exist = false;
4605                 ENTRY;
4606
4607                 LASSERT(th != NULL);
4608
4609                 fid_cpu_to_be(key, &lr->lr_fid);
4610                 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
4611                                (const struct dt_key *)key, BYPASS_CAPA);
4612                 if (rc == 0) {
4613                         if (flags & LNTF_SKIP_NLINK)
4614                                 RETURN(0);
4615
4616                         exist = true;
4617                 } else if (rc != -ENOENT) {
4618                         GOTO(log, rc);
4619                 }
4620
4621                 flags |= LNTF_SKIP_NLINK;
4622                 if (exist) {
4623                         rc = dt_delete(env, obj, (const struct dt_key *)key,
4624                                        th, BYPASS_CAPA);
4625                         if (rc != 0)
4626                                 GOTO(log, rc);
4627                 }
4628
4629                 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
4630                                (const struct dt_key *)key, th, BYPASS_CAPA, 1);
4631
4632                 GOTO(log, rc);
4633
4634 log:
4635                 CDEBUG(D_LFSCK, "%s: RPC service thread mark the "DFID
4636                        " to be skipped for namespace double scan: rc = %d\n",
4637                        lfsck_lfsck2name(com->lc_lfsck), PFID(&lr->lr_fid), rc);
4638
4639                 if (rc != 0)
4640                         /* If we cannot record this object in the LFSCK tracing,
4641                          * we have to mark the LFSC as LF_INCOMPLETE, then the
4642                          * LFSCK will skip nlink attribute verification for
4643                          * all objects. */
4644                         ns->ln_flags |= LF_INCOMPLETE;
4645
4646                 return 0;
4647         }
4648         case LE_SET_LMV_MASTER: {
4649                 struct dt_object        *obj;
4650
4651                 obj = lfsck_object_find_by_dev(env, lfsck->li_bottom,
4652                                                &lr->lr_fid);
4653                 if (IS_ERR(obj))
4654                         RETURN(PTR_ERR(obj));
4655
4656                 rc = lfsck_namespace_notify_lmv_master_local(env, com, obj);
4657                 lfsck_object_put(env, obj);
4658
4659                 RETURN(rc > 0 ? 0 : rc);
4660         }
4661         case LE_SET_LMV_SLAVE: {
4662                 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
4663                         ns->ln_striped_shards_repaired++;
4664
4665                 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
4666                                                   LNTF_RECHECK_NAME_HASH, true);
4667
4668                 RETURN(rc > 0 ? 0 : rc);
4669         }
4670         case LE_PHASE1_DONE:
4671         case LE_PHASE2_DONE:
4672         case LE_PEER_EXIT:
4673                 break;
4674         default:
4675                 RETURN(-EINVAL);
4676         }
4677
4678         CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
4679                "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
4680                lr->lr_index, lr->lr_status, lr->lr_flags2);
4681
4682         spin_lock(&ltds->ltd_lock);
4683         ltd = LTD_TGT(ltds, lr->lr_index);
4684         if (ltd == NULL) {
4685                 spin_unlock(&ltds->ltd_lock);
4686
4687                 RETURN(-ENXIO);
4688         }
4689
4690         list_del_init(&ltd->ltd_namespace_phase_list);
4691         switch (lr->lr_event) {
4692         case LE_PHASE1_DONE:
4693                 if (lr->lr_status <= 0) {
4694                         ltd->ltd_namespace_done = 1;
4695                         list_del_init(&ltd->ltd_namespace_list);
4696                         CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
4697                                "phase1 for namespace LFSCK: rc = %d.\n",
4698                                lfsck_lfsck2name(lfsck),
4699                                ltd->ltd_index, lr->lr_status);
4700                         ns->ln_flags |= LF_INCOMPLETE;
4701                         fail = true;
4702                         break;
4703                 }
4704
4705                 if (lr->lr_flags2 & LF_INCOMPLETE)
4706                         ns->ln_flags |= LF_INCOMPLETE;
4707
4708                 if (list_empty(&ltd->ltd_namespace_list))
4709                         list_add_tail(&ltd->ltd_namespace_list,
4710                                       &lad->lad_mdt_list);
4711                 list_add_tail(&ltd->ltd_namespace_phase_list,
4712                               &lad->lad_mdt_phase2_list);
4713                 break;
4714         case LE_PHASE2_DONE:
4715                 ltd->ltd_namespace_done = 1;
4716                 list_del_init(&ltd->ltd_namespace_list);
4717                 break;
4718         case LE_PEER_EXIT:
4719                 fail = true;
4720                 ltd->ltd_namespace_done = 1;
4721                 list_del_init(&ltd->ltd_namespace_list);
4722                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
4723                         CDEBUG(D_LFSCK,
4724                                "%s: the peer MDT %x exit namespace LFSCK\n",
4725                                lfsck_lfsck2name(lfsck), ltd->ltd_index);
4726                         ns->ln_flags |= LF_INCOMPLETE;
4727                 }
4728                 break;
4729         default:
4730                 break;
4731         }
4732         spin_unlock(&ltds->ltd_lock);
4733
4734         if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
4735                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
4736
4737                 memset(stop, 0, sizeof(*stop));
4738                 stop->ls_status = lr->lr_status;
4739                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
4740                 lfsck_stop(env, lfsck->li_bottom, stop);
4741         } else if (lfsck_phase2_next_ready(lad)) {
4742                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4743         }
4744
4745         RETURN(0);
4746 }
4747
4748 static int lfsck_namespace_query(const struct lu_env *env,
4749                                  struct lfsck_component *com)
4750 {
4751         struct lfsck_namespace *ns = com->lc_file_ram;
4752
4753         return ns->ln_status;
4754 }
4755
4756 static struct lfsck_operations lfsck_namespace_ops = {
4757         .lfsck_reset            = lfsck_namespace_reset,
4758         .lfsck_fail             = lfsck_namespace_fail,
4759         .lfsck_close_dir        = lfsck_namespace_close_dir,
4760         .lfsck_open_dir         = lfsck_namespace_open_dir,
4761         .lfsck_checkpoint       = lfsck_namespace_checkpoint,
4762         .lfsck_prep             = lfsck_namespace_prep,
4763         .lfsck_exec_oit         = lfsck_namespace_exec_oit,
4764         .lfsck_exec_dir         = lfsck_namespace_exec_dir,
4765         .lfsck_post             = lfsck_namespace_post,
4766         .lfsck_dump             = lfsck_namespace_dump,
4767         .lfsck_double_scan      = lfsck_namespace_double_scan,
4768         .lfsck_data_release     = lfsck_namespace_data_release,
4769         .lfsck_quit             = lfsck_namespace_quit,
4770         .lfsck_in_notify        = lfsck_namespace_in_notify,
4771         .lfsck_query            = lfsck_namespace_query,
4772 };
4773
4774 /**
4775  * Repair dangling name entry.
4776  *
4777  * For the name entry with dangling reference, we need to repare the
4778  * inconsistency according to the LFSCK sponsor's requirement:
4779  *
4780  * 1) Keep the inconsistency there and report the inconsistency case,
4781  *    then give the chance to the application to find related issues,
4782  *    and the users can make the decision about how to handle it with
4783  *    more human knownledge. (by default)
4784  *
4785  * 2) Re-create the missing MDT-object with the FID information.
4786  *
4787  * \param[in] env       pointer to the thread context
4788  * \param[in] com       pointer to the lfsck component
4789  * \param[in] child     pointer to the object corresponding to the dangling
4790  *                      name entry
4791  * \param[in] lnr       pointer to the namespace request that contains the
4792  *                      name's name, parent object, parent's LMV, and ect.
4793  *
4794  * \retval              positive number if no need to repair
4795  * \retval              zero for repaired successfully
4796  * \retval              negative error number on failure
4797  */
4798 int lfsck_namespace_repair_dangling(const struct lu_env *env,
4799                                     struct lfsck_component *com,
4800                                     struct dt_object *child,
4801                                     struct lfsck_namespace_req *lnr)
4802 {
4803         struct lfsck_thread_info        *info   = lfsck_env_info(env);
4804         struct lu_attr                  *la     = &info->lti_la;
4805         struct dt_allocation_hint       *hint   = &info->lti_hint;
4806         struct dt_object_format         *dof    = &info->lti_dof;
4807         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
4808         struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
4809         struct dt_object                *parent = lnr->lnr_obj;
4810         const struct lu_name            *cname;
4811         struct linkea_data               ldata  = { 0 };
4812         struct lustre_handle             lh     = { 0 };
4813         struct lu_buf                    linkea_buf;
4814         struct lu_buf                    lmv_buf;
4815         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4816         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
4817         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
4818         struct thandle                  *th     = NULL;
4819         int                              rc     = 0;
4820         __u16                            type   = lnr->lnr_type;
4821         bool                             create;
4822         ENTRY;
4823
4824         cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
4825         if (bk->lb_param & LPF_CREATE_MDTOBJ)
4826                 create = true;
4827         else
4828                 create = false;
4829
4830         if (!create || bk->lb_param & LPF_DRYRUN)
4831                 GOTO(log, rc = 0);
4832
4833         rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
4834         if (rc != 0)
4835                 GOTO(log, rc);
4836
4837         rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(parent));
4838         if (rc != 0)
4839                 GOTO(log, rc);
4840
4841         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
4842                               MDS_INODELOCK_UPDATE, LCK_EX);
4843         if (rc != 0)
4844                 GOTO(log, rc);
4845
4846         rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
4847         if (rc != 0)
4848                 GOTO(log, rc);
4849
4850         th = dt_trans_create(env, dev);
4851         if (IS_ERR(th))
4852                 GOTO(log, rc = PTR_ERR(th));
4853
4854         /* Set the ctime as zero, then others can know it is created for
4855          * repairing dangling name entry by LFSCK. And if the LFSCK made
4856          * wrong decision and the real MDT-object has been found later,
4857          * then the LFSCK has chance to fix the incosistency properly. */
4858         memset(la, 0, sizeof(*la));
4859         la->la_mode = (type & S_IFMT) | 0600;
4860         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
4861                         LA_ATIME | LA_MTIME | LA_CTIME;
4862
4863         child->do_ops->do_ah_init(env, hint, parent, child,
4864                                   la->la_mode & S_IFMT);
4865
4866         memset(dof, 0, sizeof(*dof));
4867         dof->dof_type = dt_mode_to_dft(type);
4868         /* If the target is a regular file, then the LFSCK will only create
4869          * the MDT-object without stripes (dof->dof_reg.striped = 0). related
4870          * OST-objects will be created when write open. */
4871
4872         /* 1a. create child. */
4873         rc = dt_declare_create(env, child, la, hint, dof, th);
4874         if (rc != 0)
4875                 GOTO(stop, rc);
4876
4877         if (S_ISDIR(type)) {
4878                 if (unlikely(!dt_try_as_dir(env, child)))
4879                         GOTO(stop, rc = -ENOTDIR);
4880
4881                 /* 2a. insert dot into child dir */
4882                 rec->rec_type = S_IFDIR;
4883                 rec->rec_fid = lfsck_dto2fid(child);
4884                 rc = dt_declare_insert(env, child,
4885                                        (const struct dt_rec *)rec,
4886                                        (const struct dt_key *)dot, th);
4887                 if (rc != 0)
4888                         GOTO(stop, rc);
4889
4890                 /* 3a. insert dotdot into child dir */
4891                 rec->rec_fid = lfsck_dto2fid(parent);
4892                 rc = dt_declare_insert(env, child,
4893                                        (const struct dt_rec *)rec,
4894                                        (const struct dt_key *)dotdot, th);
4895                 if (rc != 0)
4896                         GOTO(stop, rc);
4897
4898                 /* 4a. increase child nlink */
4899                 rc = dt_declare_ref_add(env, child, th);
4900                 if (rc != 0)
4901                         GOTO(stop, rc);
4902
4903                 /* 5a. generate slave LMV EA. */
4904                 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4905                         int idx;
4906
4907                         idx = lfsck_shard_name_to_index(env,
4908                                         lnr->lnr_name, lnr->lnr_namelen,
4909                                         type, lfsck_dto2fid(child));
4910                         if (unlikely(idx < 0))
4911                                 GOTO(stop, rc = idx);
4912
4913                         *lmv2 = lnr->lnr_lmv->ll_lmv;
4914                         lmv2->lmv_magic = LMV_MAGIC_STRIPE;
4915                         lmv2->lmv_master_mdt_index = idx;
4916
4917                         lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
4918                         lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
4919                         rc = dt_declare_xattr_set(env, child, &lmv_buf,
4920                                                   XATTR_NAME_LMV, 0, th);
4921                         if (rc != 0)
4922                                 GOTO(stop, rc);
4923                 }
4924         }
4925
4926         /* 6a. insert linkEA for child */
4927         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
4928                        ldata.ld_leh->leh_len);
4929         rc = dt_declare_xattr_set(env, child, &linkea_buf,
4930                                   XATTR_NAME_LINK, 0, th);
4931         if (rc != 0)
4932                 GOTO(stop, rc);
4933
4934         rc = dt_trans_start(env, dev, th);
4935         if (rc != 0)
4936                 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
4937
4938         dt_write_lock(env, child, 0);
4939         /* 1b. create child */
4940         rc = dt_create(env, child, la, hint, dof, th);
4941         if (rc != 0)
4942                 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
4943
4944         if (S_ISDIR(type)) {
4945                 if (unlikely(!dt_try_as_dir(env, child)))
4946                         GOTO(unlock, rc = -ENOTDIR);
4947
4948                 /* 2b. insert dot into child dir */
4949                 rec->rec_type = S_IFDIR;
4950                 rec->rec_fid = lfsck_dto2fid(child);
4951                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
4952                                (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
4953                 if (rc != 0)
4954                         GOTO(unlock, rc);
4955
4956                 /* 3b. insert dotdot into child dir */
4957                 rec->rec_fid = lfsck_dto2fid(parent);
4958                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
4959                                (const struct dt_key *)dotdot, th,
4960                                BYPASS_CAPA, 1);
4961                 if (rc != 0)
4962                         GOTO(unlock, rc);
4963
4964                 /* 4b. increase child nlink */
4965                 rc = dt_ref_add(env, child, th);
4966                 if (rc != 0)
4967                         GOTO(unlock, rc);
4968
4969                 /* 5b. generate slave LMV EA. */
4970                 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4971                         rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
4972                                           0, th, BYPASS_CAPA);
4973                         if (rc != 0)
4974                                 GOTO(unlock, rc);
4975                 }
4976         }
4977
4978         /* 6b. insert linkEA for child. */
4979         rc = dt_xattr_set(env, child, &linkea_buf,
4980                           XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
4981
4982         GOTO(unlock, rc);
4983
4984 unlock:
4985         dt_write_unlock(env, child);
4986
4987 stop:
4988         dt_trans_stop(env, dev, th);
4989
4990 log:
4991         lfsck_ibits_unlock(&lh, LCK_EX);
4992         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
4993                "reference for: parent "DFID", child "DFID", type %u, "
4994                "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
4995                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
4996                type, cname->ln_name,
4997                create ? "Create the lost OST-object as required" :
4998                         "Keep the MDT-object there by default", rc);
4999
5000         if (rc <= 0) {
5001                 struct lfsck_namespace *ns = com->lc_file_ram;
5002
5003                 ns->ln_flags |= LF_INCONSISTENT;
5004         }
5005
5006         return rc;
5007 }
5008
5009 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
5010                                                 struct lfsck_component *com,
5011                                                 struct lfsck_assistant_req *lar)
5012 {
5013         struct lfsck_thread_info   *info     = lfsck_env_info(env);
5014         struct lu_attr             *la       = &info->lti_la;
5015         struct lfsck_instance      *lfsck    = com->lc_lfsck;
5016         struct lfsck_bookmark      *bk       = &lfsck->li_bookmark_ram;
5017         struct lfsck_namespace     *ns       = com->lc_file_ram;
5018         struct linkea_data          ldata    = { 0 };
5019         const struct lu_name       *cname;
5020         struct thandle             *handle   = NULL;
5021         struct lfsck_namespace_req *lnr      =
5022                         container_of0(lar, struct lfsck_namespace_req, lnr_lar);
5023         struct dt_object           *dir      = lnr->lnr_obj;
5024         struct dt_object           *obj      = NULL;
5025         const struct lu_fid        *pfid     = lfsck_dto2fid(dir);
5026         struct dt_device           *dev      = NULL;
5027         struct lustre_handle        lh       = { 0 };
5028         bool                        repaired = false;
5029         bool                        dtlocked = false;
5030         bool                        remove;
5031         bool                        newdata;
5032         bool                        log      = false;
5033         bool                        bad_hash = false;
5034         int                         idx      = 0;
5035         int                         count    = 0;
5036         int                         rc;
5037         enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
5038         ENTRY;
5039
5040         if (lnr->lnr_attr & LUDA_UPGRADE) {
5041                 ns->ln_flags |= LF_UPGRADE;
5042                 ns->ln_dirent_repaired++;
5043                 repaired = true;
5044         } else if (lnr->lnr_attr & LUDA_REPAIR) {
5045                 ns->ln_flags |= LF_INCONSISTENT;
5046                 ns->ln_dirent_repaired++;
5047                 repaired = true;
5048         }
5049
5050         if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
5051                 if (strcmp(lnr->lnr_name, dotdot) != 0)
5052                         LBUG();
5053                 else
5054                         rc = lfsck_namespace_trace_update(env, com, pfid,
5055                                                 LNTF_CHECK_PARENT, true);
5056
5057                 GOTO(out, rc);
5058         }
5059
5060         if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
5061                 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
5062
5063                 RETURN(rc);
5064         }
5065
5066         if (lnr->lnr_name[0] == '.' &&
5067             (lnr->lnr_namelen == 1 || fid_seq_is_dot(fid_seq(&lnr->lnr_fid))))
5068                 GOTO(out, rc = 0);
5069
5070         if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5071                 rc = lfsck_namespace_handle_striped_master(env, com, lnr);
5072
5073                 RETURN(rc);
5074         }
5075
5076         idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5077         if (idx < 0)
5078                 GOTO(out, rc = idx);
5079
5080         if (idx == lfsck_dev_idx(lfsck->li_bottom)) {
5081                 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5082                         GOTO(out, rc = 0);
5083
5084                 dev = lfsck->li_next;
5085         } else {
5086                 struct lfsck_tgt_desc *ltd;
5087
5088                 /* Usually, some local filesystem consistency verification
5089                  * tools can guarantee the local namespace tree consistenct.
5090                  * So the LFSCK will only verify the remote directory. */
5091                 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5092                         rc = lfsck_namespace_trace_update(env, com, pfid,
5093                                                 LNTF_CHECK_PARENT, true);
5094
5095                         GOTO(out, rc);
5096                 }
5097
5098                 ltd = LTD_TGT(&lfsck->li_mdt_descs, idx);
5099                 if (unlikely(ltd == NULL)) {
5100                         CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5101                                "did not join the namespace LFSCK\n",
5102                                lfsck_lfsck2name(lfsck), idx);
5103                         lfsck_lad_set_bitmap(env, com, idx);
5104
5105                         GOTO(out, rc = -ENODEV);
5106                 }
5107
5108                 dev = ltd->ltd_tgt;
5109         }
5110
5111         obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5112         if (IS_ERR(obj))
5113                 GOTO(out, rc = PTR_ERR(obj));
5114
5115         cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5116         if (dt_object_exists(obj) == 0) {
5117
5118 dangling:
5119                 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5120                 if (rc == 0) {
5121                         if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5122                                         lnr->lnr_name, lnr->lnr_namelen)) {
5123                                 type = LNIT_BAD_DIRENT;
5124
5125                                 GOTO(out, rc);
5126                         }
5127
5128                         type = LNIT_DANGLING;
5129                         rc = lfsck_namespace_repair_dangling(env, com,
5130                                                              obj, lnr);
5131                         if (rc == 0)
5132                                 repaired = true;
5133                 }
5134
5135                 GOTO(out, rc);
5136         }
5137
5138         if (!(bk->lb_param & LPF_DRYRUN) && repaired) {
5139
5140 again:
5141                 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5142                                       MDS_INODELOCK_UPDATE |
5143                                       MDS_INODELOCK_XATTR, LCK_EX);
5144                 if (rc != 0)
5145                         GOTO(out, rc);
5146
5147                 handle = dt_trans_create(env, dev);
5148                 if (IS_ERR(handle))
5149                         GOTO(out, rc = PTR_ERR(handle));
5150
5151                 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5152                 if (rc != 0)
5153                         GOTO(stop, rc);
5154
5155                 rc = dt_trans_start(env, dev, handle);
5156                 if (rc != 0)
5157                         GOTO(stop, rc);
5158
5159                 dt_write_lock(env, obj, 0);
5160                 dtlocked = true;
5161         }
5162
5163         rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5164         if (rc != 0)
5165                 GOTO(stop, rc);
5166
5167         rc = lfsck_links_read(env, obj, &ldata);
5168         if (unlikely(rc == -ENOENT)) {
5169                 if (handle != NULL) {
5170                         dt_write_unlock(env, obj);
5171                         dtlocked = false;
5172
5173                         dt_trans_stop(env, dev, handle);
5174                         handle = NULL;
5175
5176                         lfsck_ibits_unlock(&lh, LCK_EX);
5177                 }
5178
5179                 /* It may happen when the remote object has been removed,
5180                  * but the local MDT is not aware of that. */
5181                 goto dangling;
5182         } else if (rc == 0) {
5183                 count = ldata.ld_leh->leh_reccount;
5184                 rc = linkea_links_find(&ldata, cname, pfid);
5185                 if ((rc == 0) &&
5186                     (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5187                         if ((lfsck_object_type(obj) & S_IFMT) !=
5188                             lnr->lnr_type) {
5189                                 ns->ln_flags |= LF_INCONSISTENT;
5190                                 type = LNIT_BAD_TYPE;
5191                         }
5192
5193                         goto record;
5194                 }
5195
5196                 ns->ln_flags |= LF_INCONSISTENT;
5197
5198                 /* If the name entry hash does not match the slave striped
5199                  * directory, and the name entry does not match also, then
5200                  * it is quite possible that name entry is corrupted. */
5201                 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5202                                         lnr->lnr_name, lnr->lnr_namelen)) {
5203                         type = LNIT_BAD_DIRENT;
5204
5205                         GOTO(stop, rc = 0);
5206                 }
5207
5208                 /* If the file type stored in the name entry does not match
5209                  * the file type claimed by the object, and the object does
5210                  * not recognize the name entry, then it is quite possible
5211                  * that the name entry is corrupted. */
5212                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5213                         type = LNIT_BAD_DIRENT;
5214
5215                         GOTO(stop, rc = 0);
5216                 }
5217
5218                 /* For sub-dir object, we cannot make sure whether the sub-dir
5219                  * back references the parent via ".." name entry correctly or
5220                  * not in the LFSCK first-stage scanning. It may be that the
5221                  * (remote) sub-dir ".." name entry has no parent FID after
5222                  * file-level backup/restore and its linkEA may be wrong.
5223                  * So under such case, we should replace the linkEA according
5224                  * to current name entry. But this needs to be done during the
5225                  * LFSCK second-stage scanning. The LFSCK will record the name
5226                  * entry for further possible using. */
5227                 remove = false;
5228                 newdata = false;
5229                 goto nodata;
5230         } else if (unlikely(rc == -EINVAL)) {
5231                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5232                         type = LNIT_BAD_TYPE;
5233
5234                 count = 1;
5235                 ns->ln_flags |= LF_INCONSISTENT;
5236                 /* The magic crashed, we are not sure whether there are more
5237                  * corrupt data in the linkea, so remove all linkea entries. */
5238                 remove = true;
5239                 newdata = true;
5240                 goto nodata;
5241         } else if (rc == -ENODATA) {
5242                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5243                         type = LNIT_BAD_TYPE;
5244
5245                 count = 1;
5246                 ns->ln_flags |= LF_UPGRADE;
5247                 remove = false;
5248                 newdata = true;
5249
5250 nodata:
5251                 if (bk->lb_param & LPF_DRYRUN) {
5252                         ns->ln_linkea_repaired++;
5253                         repaired = true;
5254                         log = true;
5255                         goto record;
5256                 }
5257
5258                 if (!lustre_handle_is_used(&lh))
5259                         goto again;
5260
5261                 if (remove) {
5262                         LASSERT(newdata);
5263
5264                         rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle,
5265                                           BYPASS_CAPA);
5266                         if (rc != 0)
5267                                 GOTO(stop, rc);
5268                 }
5269
5270                 if (newdata) {
5271                         rc = linkea_data_new(&ldata,
5272                                         &lfsck_env_info(env)->lti_linkea_buf);
5273                         if (rc != 0)
5274                                 GOTO(stop, rc);
5275                 }
5276
5277                 rc = linkea_add_buf(&ldata, cname, pfid);
5278                 if (rc != 0)
5279                         GOTO(stop, rc);
5280
5281                 rc = lfsck_links_write(env, obj, &ldata, handle);
5282                 if (unlikely(rc == -ENOSPC) &&
5283                     S_ISREG(lfsck_object_type(obj)) && !dt_object_remote(obj)) {
5284                         if (handle != NULL) {
5285                                 LASSERT(dt_write_locked(env, obj));
5286
5287                                 dt_write_unlock(env, obj);
5288                                 dtlocked = false;
5289
5290                                 dt_trans_stop(env, dev, handle);
5291                                 handle = NULL;
5292
5293                                 lfsck_ibits_unlock(&lh, LCK_EX);
5294                         }
5295
5296                         rc = lfsck_namespace_trace_update(env, com,
5297                                         &lnr->lnr_fid, LNTF_SKIP_NLINK, true);
5298                         if (rc != 0)
5299                                 /* If we cannot record this object in the
5300                                  * LFSCK tracing, we have to mark the LFSCK
5301                                  * as LF_INCOMPLETE, then the LFSCK will
5302                                  * skip nlink attribute verification for
5303                                  * all objects. */
5304                                 ns->ln_flags |= LF_INCOMPLETE;
5305
5306                         GOTO(out, rc = 0);
5307                 }
5308
5309                 if (rc != 0)
5310                         GOTO(stop, rc);
5311
5312                 count = ldata.ld_leh->leh_reccount;
5313                 if (!S_ISDIR(lfsck_object_type(obj)) ||
5314                     !dt_object_remote(obj)) {
5315                         ns->ln_linkea_repaired++;
5316                         repaired = true;
5317                         log = true;
5318                 }
5319         } else {
5320                 GOTO(stop, rc);
5321         }
5322
5323 record:
5324         LASSERT(count > 0);
5325
5326         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
5327         if (rc != 0)
5328                 GOTO(stop, rc);
5329
5330         if ((count == 1 && la->la_nlink == 1) ||
5331             S_ISDIR(lfsck_object_type(obj)))
5332                 /* Usually, it is for single linked object or dir, do nothing.*/
5333                 GOTO(stop, rc);
5334
5335         /* Following modification will be in another transaction.  */
5336         if (handle != NULL) {
5337                 dt_write_unlock(env, obj);
5338                 dtlocked = false;
5339
5340                 dt_trans_stop(env, dev, handle);
5341                 handle = NULL;
5342
5343                 lfsck_ibits_unlock(&lh, LCK_EX);
5344         }
5345
5346         ns->ln_mul_linked_checked++;
5347         rc = lfsck_namespace_trace_update(env, com, &lnr->lnr_fid,
5348                                           LNTF_CHECK_LINKEA, true);
5349
5350         GOTO(out, rc);
5351
5352 stop:
5353         if (dtlocked)
5354                 dt_write_unlock(env, obj);
5355
5356         if (handle != NULL && !IS_ERR(handle))
5357                 dt_trans_stop(env, dev, handle);
5358
5359 out:
5360         lfsck_ibits_unlock(&lh, LCK_EX);
5361
5362         if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5363             !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5364                                              lnr->lnr_name, lnr->lnr_namelen) &&
5365             type != LNIT_BAD_DIRENT) {
5366                 ns->ln_flags |= LF_INCONSISTENT;
5367
5368                 log = false;
5369                 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5370                                                 lnr->lnr_lmv, lnr->lnr_name);
5371                 if (rc >= 0)
5372                         bad_hash = true;
5373         }
5374
5375         if (rc >= 0) {
5376                 switch (type) {
5377                 case LNIT_BAD_TYPE:
5378                         log = false;
5379                         rc = lfsck_namespace_repair_dirent(env, com, dir,
5380                                         obj, lnr->lnr_name, lnr->lnr_name,
5381                                         lnr->lnr_type, true, false);
5382                         if (rc > 0)
5383                                 repaired = true;
5384                         break;
5385                 case LNIT_BAD_DIRENT:
5386                         log = false;
5387                         /* XXX: This is a bad dirent, we do not know whether
5388                          *      the original name entry reference a regular
5389                          *      file or a directory, then keep the parent's
5390                          *      nlink count unchanged here. */
5391                         rc = lfsck_namespace_repair_dirent(env, com, dir,
5392                                         obj, lnr->lnr_name, lnr->lnr_name,
5393                                         lnr->lnr_type, false, false);
5394                         if (rc > 0)
5395                                 repaired = true;
5396                         break;
5397                 default:
5398                         break;
5399                 }
5400         }
5401
5402         down_write(&com->lc_sem);
5403         if (rc < 0) {
5404                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5405                        "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5406                        lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5407                        PFID(lfsck_dto2fid(lnr->lnr_obj)),
5408                        lnr->lnr_namelen, lnr->lnr_name, rc);
5409
5410                 lfsck_namespace_record_failure(env, lfsck, ns);
5411                 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5412                      rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5413                      rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5414                     dev != NULL && dev != lfsck->li_next)
5415                         lfsck_lad_set_bitmap(env, com, idx);
5416
5417                 if (!(bk->lb_param & LPF_FAILOUT))
5418                         rc = 0;
5419         } else {
5420                 if (log)
5421                         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5422                                "repaired the entry: "DFID", parent "DFID
5423                                ", name %.*s\n", lfsck_lfsck2name(lfsck),
5424                                PFID(&lnr->lnr_fid),
5425                                PFID(lfsck_dto2fid(lnr->lnr_obj)),
5426                                lnr->lnr_namelen, lnr->lnr_name);
5427
5428                 if (repaired) {
5429                         ns->ln_items_repaired++;
5430
5431                         switch (type) {
5432                         case LNIT_DANGLING:
5433                                 ns->ln_dangling_repaired++;
5434                                 break;
5435                         case LNIT_BAD_TYPE:
5436                                 ns->ln_bad_type_repaired++;
5437                                 break;
5438                         case LNIT_BAD_DIRENT:
5439                                 ns->ln_dirent_repaired++;
5440                                 break;
5441                         default:
5442                                 break;
5443                         }
5444
5445                         if (bk->lb_param & LPF_DRYRUN &&
5446                             lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5447                                 lfsck_pos_fill(env, lfsck,
5448                                                &ns->ln_pos_first_inconsistent,
5449                                                false);
5450                 }
5451
5452                 if (bad_hash) {
5453                         ns->ln_name_hash_repaired++;
5454
5455                         /* Not count repeatedly. */
5456                         if (!repaired)
5457                                 ns->ln_items_repaired++;
5458
5459                         if (bk->lb_param & LPF_DRYRUN &&
5460                             lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5461                                 lfsck_pos_fill(env, lfsck,
5462                                                &ns->ln_pos_first_inconsistent,
5463                                                false);
5464                 }
5465
5466
5467                 rc = 0;
5468         }
5469         up_write(&com->lc_sem);
5470
5471         if (obj != NULL && !IS_ERR(obj))
5472                 lfsck_object_put(env, obj);
5473
5474         return rc;
5475 }
5476
5477 /**
5478  * Handle one orphan under the backend /lost+found directory
5479  *
5480  * Insert the orphan FID into the namespace LFSCK trace file for further
5481  * processing (via the subsequent namespace LFSCK second-stage scanning).
5482  * At the same time, remove the orphan name entry from backend /lost+found
5483  * directory. There is an interval between the orphan name entry removed
5484  * from the backend /lost+found directory and the orphan FID in the LFSCK
5485  * trace file handled. In such interval, the LFSCK can be reset, then
5486  * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
5487  * To guarantee that the orphans can be found when LFSCK run next time
5488  * without e2fsck again, when remove the orphan name entry, the LFSCK
5489  * will set the orphan's ctime attribute as 1. Since normal applications
5490  * cannot change the object's ctime attribute as 1. Then when LFSCK run
5491  * next time, it can record the object (that ctime is 1) in the namespace
5492  * LFSCK trace file during the first-stage scanning.
5493  *
5494  * \param[in] env       pointer to the thread context
5495  * \param[in] com       pointer to the lfsck component
5496  * \param[in] parent    pointer to the object for the backend /lost+found
5497  * \param[in] ent       pointer to the name entry for the target under the
5498  *                      backend /lost+found
5499  *
5500  * \retval              positive for repaired
5501  * \retval              0 if needs to repair nothing
5502  * \retval              negative error number on failure
5503  */
5504 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
5505                                               struct lfsck_component *com,
5506                                               struct dt_object *parent,
5507                                               struct lu_dirent *ent)
5508 {
5509         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5510         struct lu_fid                   *key    = &info->lti_fid;
5511         struct lu_attr                  *la     = &info->lti_la;
5512         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5513         struct dt_object                *obj    = com->lc_obj;
5514         struct dt_device                *dev    = lfsck->li_bottom;
5515         struct dt_object                *child  = NULL;
5516         struct thandle                  *th     = NULL;
5517         int                              rc     = 0;
5518         __u8                             flags  = 0;
5519         bool                             exist  = false;
5520         ENTRY;
5521
5522         child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
5523         if (IS_ERR(child))
5524                 RETURN(PTR_ERR(child));
5525
5526         LASSERT(dt_object_exists(child));
5527         LASSERT(!dt_object_remote(child));
5528
5529         fid_cpu_to_be(key, &ent->lde_fid);
5530         rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
5531                        (const struct dt_key *)key, BYPASS_CAPA);
5532         if (rc == 0) {
5533                 exist = true;
5534                 flags |= LNTF_CHECK_ORPHAN;
5535         } else if (rc == -ENOENT) {
5536                 flags = LNTF_CHECK_ORPHAN;
5537         } else {
5538                 GOTO(out, rc);
5539         }
5540
5541         th = dt_trans_create(env, dev);
5542         if (IS_ERR(th))
5543                 GOTO(out, rc = PTR_ERR(th));
5544
5545         /* a1. remove name entry from backend /lost+found */
5546         rc = dt_declare_delete(env, parent,
5547                                (const struct dt_key *)ent->lde_name, th);
5548         if (rc != 0)
5549                 GOTO(stop, rc);
5550
5551         if (S_ISDIR(lfsck_object_type(child))) {
5552                 /* a2. decrease parent's nlink */
5553                 rc = dt_declare_ref_del(env, parent, th);
5554                 if (rc != 0)
5555                         GOTO(stop, rc);
5556         }
5557
5558         if (exist) {
5559                 /* a3. remove child's FID from the LFSCK trace file. */
5560                 rc = dt_declare_delete(env, obj,
5561                                        (const struct dt_key *)key, th);
5562                 if (rc != 0)
5563                         GOTO(stop, rc);
5564         } else {
5565                 /* a4. set child's ctime as 1 */
5566                 memset(la, 0, sizeof(*la));
5567                 la->la_ctime = 1;
5568                 la->la_valid = LA_CTIME;
5569                 rc = dt_declare_attr_set(env, child, la, th);
5570                 if (rc != 0)
5571                         GOTO(stop, rc);
5572         }
5573
5574         /* a5. insert child's FID into the LFSCK trace file. */
5575         rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
5576                                (const struct dt_key *)key, th);
5577         if (rc != 0)
5578                 GOTO(stop, rc);
5579
5580         rc = dt_trans_start_local(env, dev, th);
5581         if (rc != 0)
5582                 GOTO(stop, rc);
5583
5584         /* b1. remove name entry from backend /lost+found */
5585         rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th,
5586                        BYPASS_CAPA);
5587         if (rc != 0)
5588                 GOTO(stop, rc);
5589
5590         if (S_ISDIR(lfsck_object_type(child))) {
5591                 /* b2. decrease parent's nlink */
5592                 dt_write_lock(env, parent, 0);
5593                 rc = dt_ref_del(env, parent, th);
5594                 dt_write_unlock(env, parent);
5595                 if (rc != 0)
5596                         GOTO(stop, rc);
5597         }
5598
5599         if (exist) {
5600                 /* a3. remove child's FID from the LFSCK trace file. */
5601                 rc = dt_delete(env, obj, (const struct dt_key *)key, th,
5602                                BYPASS_CAPA);
5603                 if (rc != 0)
5604                         GOTO(stop, rc);
5605         } else {
5606                 /* b4. set child's ctime as 1 */
5607                 rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
5608                 if (rc != 0)
5609                         GOTO(stop, rc);
5610         }
5611
5612         /* b5. insert child's FID into the LFSCK trace file. */
5613         rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
5614                        (const struct dt_key *)key, th, BYPASS_CAPA, 1);
5615
5616         GOTO(stop, rc = (rc == 0 ? 1 : rc));
5617
5618 stop:
5619         dt_trans_stop(env, dev, th);
5620
5621 out:
5622         lu_object_put(env, &child->do_lu);
5623
5624         return rc;
5625 }
5626
5627 /**
5628  * Handle orphans under the backend /lost+found directory
5629  *
5630  * Some backend checker, such as e2fsck for ldiskfs may find some orphans
5631  * and put them under the backend /lost+found directory that is invisible
5632  * to client. The LFSCK will scan such directory, for the original client
5633  * visible orphans, add their fids into the namespace LFSCK trace file,
5634  * then the subsenquent namespace LFSCK second-stage scanning can handle
5635  * them as other objects to be double scanned: either move back to normal
5636  * namespace, or to the global visible orphan directory:
5637  * /ROOT/.lustre/lost+found/MDTxxxx/
5638  *
5639  * \param[in] env       pointer to the thread context
5640  * \param[in] com       pointer to the lfsck component
5641  */
5642 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
5643                                            struct lfsck_component *com)
5644 {
5645         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5646         struct lu_dirent                *ent    =
5647                                         (struct lu_dirent *)info->lti_key;
5648         struct lu_seq_range             *range  = &info->lti_range;
5649         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5650         struct ptlrpc_thread            *thread = &lfsck->li_thread;
5651         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
5652         struct dt_device                *dev    = lfsck->li_bottom;
5653         struct lfsck_namespace          *ns     = com->lc_file_ram;
5654         struct dt_object                *parent;
5655         const struct dt_it_ops          *iops;
5656         struct dt_it                    *di;
5657         struct seq_server_site          *ss     =
5658                                         lu_site2seq(dev->dd_lu_dev.ld_site);
5659         __u64                            cookie;
5660         int                              rc     = 0;
5661         __u16                            type;
5662         ENTRY;
5663
5664         parent = lfsck_object_find_by_dev(env, dev, &LU_BACKEND_LPF_FID);
5665         if (IS_ERR(parent)) {
5666                 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
5667                        lfsck_lfsck2name(lfsck), PTR_ERR(parent));
5668                 RETURN_EXIT;
5669         }
5670
5671         /* It is normal that the /lost+found does not exist for ZFS backend. */
5672         if (!dt_object_exists(parent))
5673                 GOTO(out, rc = 0);
5674
5675         if (unlikely(!dt_try_as_dir(env, parent)))
5676                 GOTO(out, rc = -ENOTDIR);
5677
5678         CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
5679                lfsck_lfsck2name(lfsck));
5680
5681         com->lc_new_scanned = 0;
5682         iops = &parent->do_index_ops->dio_it;
5683         di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE, BYPASS_CAPA);
5684         if (IS_ERR(di))
5685                 GOTO(out, rc = PTR_ERR(di));
5686
5687         rc = iops->load(env, di, 0);
5688         if (rc == 0)
5689                 rc = iops->next(env, di);
5690         else if (rc > 0)
5691                 rc = 0;
5692
5693         while (rc == 0) {
5694                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
5695                     cfs_fail_val > 0) {
5696                         struct l_wait_info lwi;
5697
5698                         lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
5699                                           NULL, NULL);
5700                         l_wait_event(thread->t_ctl_waitq,
5701                                      !thread_is_running(thread),
5702                                      &lwi);
5703
5704                         if (unlikely(!thread_is_running(thread)))
5705                                 break;
5706                 }
5707
5708                 rc = iops->rec(env, di, (struct dt_rec *)ent,
5709                                LUDA_64BITHASH | LUDA_TYPE);
5710                 if (rc == 0)
5711                         rc = lfsck_unpack_ent(ent, &cookie, &type);
5712
5713                 if (unlikely(rc != 0)) {
5714                         CDEBUG(D_LFSCK, "%s: fail to iterate backend "
5715                                "/lost+found: rc = %d\n",
5716                                lfsck_lfsck2name(lfsck), rc);
5717
5718                         goto skip;
5719                 }
5720
5721                 /* skip dot and dotdot entries */
5722                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5723                         goto next;
5724
5725                 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
5726                         goto skip;
5727
5728                 if (fid_is_norm(&ent->lde_fid)) {
5729                         fld_range_set_mdt(range);
5730                         rc = fld_local_lookup(env, ss->ss_server_fld,
5731                                               fid_seq(&ent->lde_fid), range);
5732                         if (rc != 0)
5733                                 goto skip;
5734                 } else if (lfsck_dev_idx(dev) != 0) {
5735                         /* If the returned FID is IGIF, then there are three
5736                          * possible cases:
5737                          *
5738                          * 1) The object is upgraded from old Lustre-1.8 with
5739                          *    IGIF assigned to such object.
5740                          * 2) The object is a backend local object and is
5741                          *    invisible to client.
5742                          * 3) The object lost its LMV EA, and since there is
5743                          *    no FID-in-dirent for the orphan in the backend
5744                          *    /lost+found directory, then the low layer will
5745                          *    return IGIF for such object.
5746                          *
5747                          * For MDTx (x != 0), it is either case 2) or case 3),
5748                          * but from the LFSCK view, they are indistinguishable.
5749                          * To be safe, the LFSCK will keep it there and report
5750                          * some message, then the adminstrator can handle that
5751                          * furtherly.
5752                          *
5753                          * For MDT0, it is more possible the case 1). The LFSCK
5754                          * will handle the orphan as an upgraded object. */
5755                         CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
5756                                "in the backend /lost+found on the MDT %04x, "
5757                                "to be safe, skip it.\n",
5758                                lfsck_lfsck2name(lfsck), ent->lde_namelen,
5759                                ent->lde_name, PFID(&ent->lde_fid),
5760                                lfsck_dev_idx(dev));
5761                         goto skip;
5762                 }
5763
5764                 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
5765
5766 skip:
5767                 down_write(&com->lc_sem);
5768                 com->lc_new_scanned++;
5769                 ns->ln_local_lpf_scanned++;
5770                 if (rc > 0)
5771                         ns->ln_local_lpf_moved++;
5772                 else if (rc == 0)
5773                         ns->ln_local_lpf_skipped++;
5774                 else
5775                         ns->ln_local_lpf_failed++;
5776                 up_write(&com->lc_sem);
5777
5778                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5779                         break;
5780
5781 next:
5782                 lfsck_control_speed_by_self(com);
5783                 if (unlikely(!thread_is_running(thread))) {
5784                         rc = 0;
5785                         break;
5786                 }
5787
5788                 rc = iops->next(env, di);
5789         }
5790
5791         iops->put(env, di);
5792         iops->fini(env, di);
5793
5794         EXIT;
5795
5796 out:
5797         CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
5798                lfsck_lfsck2name(lfsck), rc);
5799
5800         lu_object_put(env, &parent->do_lu);
5801 }
5802
5803 /**
5804  * Rescan the striped directory after the master LMV EA reset.
5805  *
5806  * Sometimes, the master LMV EA of the striped directory maybe lost, so when
5807  * the namespace LFSCK engine scan the striped directory for the first time,
5808  * it will be regarded as a normal directory. As the LFSCK processing, some
5809  * other LFSCK instance on other MDT will find the shard of this striped dir,
5810  * and find that the master MDT-object of the striped directory lost its LMV
5811  * EA, then such remote LFSCK instance will regenerate the master LMV EA and
5812  * notify the LFSCK instance on this MDT to rescan the striped directory.
5813  *
5814  * \param[in] env       pointer to the thread context
5815  * \param[in] com       pointer to the lfsck component
5816  * \param[in] llu       the lfsck_lmv_unit that contains the striped directory
5817  *                      to be rescanned.
5818  *
5819  * \retval              positive number for success
5820  * \retval              0 for LFSCK stopped/paused
5821  * \retval              negative error number on failure
5822  */
5823 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
5824                                               struct lfsck_component *com,
5825                                               struct lfsck_lmv_unit *llu)
5826 {
5827         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5828         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5829         struct lfsck_assistant_data     *lad    = com->lc_data;
5830         struct dt_object                *dir;
5831         const struct dt_it_ops          *iops;
5832         struct dt_it                    *di;
5833         struct lu_dirent                *ent    =
5834                         (struct lu_dirent *)info->lti_key;
5835         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
5836         struct ptlrpc_thread            *thread = &lfsck->li_thread;
5837         struct lfsck_namespace_req      *lnr;
5838         struct lfsck_assistant_req      *lar;
5839         int                              rc;
5840         __u16                            type;
5841         ENTRY;
5842
5843         LASSERT(list_empty(&lad->lad_req_list));
5844
5845         lfsck->li_lmv = &llu->llu_lmv;
5846         lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
5847         rc = lfsck_open_dir(env, lfsck, 0);
5848         if (rc != 0)
5849                 RETURN(rc);
5850
5851         dir = lfsck->li_obj_dir;
5852         di = lfsck->li_di_dir;
5853         iops = &dir->do_index_ops->dio_it;
5854         do {
5855                 rc = iops->rec(env, di, (struct dt_rec *)ent,
5856                                lfsck->li_args_dir);
5857                 if (rc == 0)
5858                         rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
5859                                               &type);
5860
5861                 if (rc != 0) {
5862                         if (bk->lb_param & LPF_FAILOUT)
5863                                 GOTO(out, rc);
5864
5865                         goto next;
5866                 }
5867
5868                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5869                         goto next;
5870
5871                 lnr = lfsck_namespace_assistant_req_init(lfsck, ent, type);
5872                 if (IS_ERR(lnr)) {
5873                         if (bk->lb_param & LPF_FAILOUT)
5874                                 GOTO(out, rc = PTR_ERR(lnr));
5875
5876                         goto next;
5877                 }
5878
5879                 lar = &lnr->lnr_lar;
5880                 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
5881                 lfsck_namespace_assistant_req_fini(env, lar);
5882                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
5883                         GOTO(out, rc);
5884
5885                 if (unlikely(!thread_is_running(thread)))
5886                         GOTO(out, rc = 0);
5887
5888 next:
5889                 rc = iops->next(env, di);
5890         } while (rc == 0);
5891
5892 out:
5893         lfsck_close_dir(env, lfsck, rc);
5894         if (rc <= 0)
5895                 RETURN(rc);
5896
5897         /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
5898         if (list_empty(&lad->lad_req_list))
5899                 RETURN(1);
5900
5901         spin_lock(&lad->lad_lock);
5902         lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
5903                           lar_list);
5904         list_del_init(&lar->lar_list);
5905         spin_unlock(&lad->lad_lock);
5906
5907         rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
5908         lfsck_namespace_assistant_req_fini(env, lar);
5909
5910         RETURN(rc == 0 ? 1 : rc);
5911 }
5912
5913 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
5914                                                 struct lfsck_component *com)
5915 {
5916         struct lfsck_instance   *lfsck  = com->lc_lfsck;
5917         struct ptlrpc_thread    *thread = &lfsck->li_thread;
5918         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
5919         struct lfsck_namespace  *ns     = com->lc_file_ram;
5920         struct dt_object        *obj    = com->lc_obj;
5921         const struct dt_it_ops  *iops   = &obj->do_index_ops->dio_it;
5922         struct dt_object        *target;
5923         struct dt_it            *di;
5924         struct dt_key           *key;
5925         struct lu_fid            fid;
5926         int                      rc;
5927         __u8                     flags  = 0;
5928         ENTRY;
5929
5930         while (!list_empty(&lfsck->li_list_lmv)) {
5931                 struct lfsck_lmv_unit *llu;
5932
5933                 spin_lock(&lfsck->li_lock);
5934                 llu = list_entry(lfsck->li_list_lmv.next,
5935                                  struct lfsck_lmv_unit, llu_link);
5936                 list_del_init(&llu->llu_link);
5937                 spin_unlock(&lfsck->li_lock);
5938
5939                 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
5940                 if (rc <= 0)
5941                         RETURN(rc);
5942         }
5943
5944         CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
5945                lfsck_lfsck2name(lfsck));
5946
5947         lfsck_namespace_scan_local_lpf(env, com);
5948
5949         com->lc_new_checked = 0;
5950         com->lc_new_scanned = 0;
5951         com->lc_time_last_checkpoint = cfs_time_current();
5952         com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
5953                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
5954
5955         di = iops->init(env, obj, 0, BYPASS_CAPA);
5956         if (IS_ERR(di))
5957                 RETURN(PTR_ERR(di));
5958
5959         fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
5960         rc = iops->get(env, di, (const struct dt_key *)&fid);
5961         if (rc < 0)
5962                 GOTO(fini, rc);
5963
5964         /* Skip the start one, which either has been processed or non-exist. */
5965         rc = iops->next(env, di);
5966         if (rc != 0)
5967                 GOTO(put, rc);
5968
5969         do {
5970                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
5971                     cfs_fail_val > 0) {
5972                         struct l_wait_info lwi;
5973
5974                         lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
5975                                           NULL, NULL);
5976                         l_wait_event(thread->t_ctl_waitq,
5977                                      !thread_is_running(thread),
5978                                      &lwi);
5979
5980                         if (unlikely(!thread_is_running(thread)))
5981                                 GOTO(put, rc = 0);
5982                 }
5983
5984                 key = iops->key(env, di);
5985                 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
5986                 if (!fid_is_sane(&fid)) {
5987                         rc = 0;
5988                         goto checkpoint;
5989                 }
5990
5991                 target = lfsck_object_find_by_dev(env, lfsck->li_bottom, &fid);
5992                 if (IS_ERR(target)) {
5993                         rc = PTR_ERR(target);
5994                         goto checkpoint;
5995                 }
5996
5997                 if (dt_object_exists(target)) {
5998                         rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
5999                         if (rc == 0) {
6000                                 rc = lfsck_namespace_double_scan_one(env, com,
6001                                                                 target, flags);
6002                                 if (rc == -ENOENT)
6003                                         rc = 0;
6004                         }
6005                 }
6006
6007                 lfsck_object_put(env, target);
6008
6009 checkpoint:
6010                 down_write(&com->lc_sem);
6011                 com->lc_new_checked++;
6012                 com->lc_new_scanned++;
6013                 ns->ln_fid_latest_scanned_phase2 = fid;
6014                 if (rc > 0)
6015                         ns->ln_objs_repaired_phase2++;
6016                 else if (rc < 0)
6017                         ns->ln_objs_failed_phase2++;
6018                 up_write(&com->lc_sem);
6019
6020                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6021                         GOTO(put, rc);
6022
6023                 if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
6024                                               cfs_time_current())) &&
6025                     com->lc_new_checked != 0) {
6026                         down_write(&com->lc_sem);
6027                         ns->ln_run_time_phase2 +=
6028                                 cfs_duration_sec(cfs_time_current() +
6029                                 HALF_SEC - com->lc_time_last_checkpoint);
6030                         ns->ln_time_last_checkpoint = cfs_time_current_sec();
6031                         ns->ln_objs_checked_phase2 += com->lc_new_checked;
6032                         com->lc_new_checked = 0;
6033                         rc = lfsck_namespace_store(env, com, false);
6034                         up_write(&com->lc_sem);
6035                         if (rc != 0)
6036                                 GOTO(put, rc);
6037
6038                         com->lc_time_last_checkpoint = cfs_time_current();
6039                         com->lc_time_next_checkpoint =
6040                                 com->lc_time_last_checkpoint +
6041                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
6042                 }
6043
6044                 lfsck_control_speed_by_self(com);
6045                 if (unlikely(!thread_is_running(thread)))
6046                         GOTO(put, rc = 0);
6047
6048                 rc = iops->next(env, di);
6049         } while (rc == 0);
6050
6051         GOTO(put, rc);
6052
6053 put:
6054         iops->put(env, di);
6055
6056 fini:
6057         iops->fini(env, di);
6058
6059         CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop: rc = %d\n",
6060                lfsck_lfsck2name(lfsck), rc);
6061
6062         return rc;
6063 }
6064
6065 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6066                                                struct lfsck_component *com,
6067                                                struct lfsck_position *pos)
6068 {
6069         struct lfsck_assistant_data     *lad = com->lc_data;
6070         struct lfsck_namespace_req      *lnr;
6071
6072         if (list_empty(&lad->lad_req_list))
6073                 return;
6074
6075         lnr = list_entry(lad->lad_req_list.next,
6076                          struct lfsck_namespace_req,
6077                          lnr_lar.lar_list);
6078         pos->lp_oit_cookie = lnr->lnr_oit_cookie;
6079         pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6080         pos->lp_dir_parent = *lfsck_dto2fid(lnr->lnr_obj);
6081 }
6082
6083 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6084                                               struct lfsck_component *com,
6085                                               int rc)
6086 {
6087         struct lfsck_instance   *lfsck  = com->lc_lfsck;
6088         struct lfsck_namespace  *ns     = com->lc_file_ram;
6089
6090         down_write(&com->lc_sem);
6091         ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
6092                                 HALF_SEC - lfsck->li_time_last_checkpoint);
6093         ns->ln_time_last_checkpoint = cfs_time_current_sec();
6094         ns->ln_objs_checked_phase2 += com->lc_new_checked;
6095         com->lc_new_checked = 0;
6096
6097         if (rc > 0) {
6098                 if (ns->ln_flags & LF_INCOMPLETE)
6099                         ns->ln_status = LS_PARTIAL;
6100                 else
6101                         ns->ln_status = LS_COMPLETED;
6102                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6103                         ns->ln_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
6104                 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6105                 ns->ln_success_count++;
6106         } else if (rc == 0) {
6107                 if (lfsck->li_status != 0)
6108                         ns->ln_status = lfsck->li_status;
6109                 else
6110                         ns->ln_status = LS_STOPPED;
6111         } else {
6112                 ns->ln_status = LS_FAILED;
6113         }
6114
6115         rc = lfsck_namespace_store(env, com, false);
6116         up_write(&com->lc_sem);
6117
6118         return rc;
6119 }
6120
6121 static int
6122 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6123                                                   struct ptlrpc_request *req,
6124                                                   void *args, int rc)
6125 {
6126         return 0;
6127 }
6128
6129 /**
6130  * Notify remote LFSCK instances about former failures.
6131  *
6132  * The local LFSCK instance has recorded which MDTs have ever failed to respond
6133  * some LFSCK verification requests (maybe because of network issues or the MDT
6134  * itself trouble). During the respond gap the MDT may missed some name entries
6135  * verification, then the MDT cannot know whether related MDT-objects have been
6136  * referenced by related name entries or not, then in the second-stage scanning,
6137  * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6138  * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6139  * name entry for repairing the orphan.
6140  *
6141  * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6142  * it will scan the bitmap for the ever failed MDTs, and notify them that they
6143  * have ever missed some name entries verification and should skip the handling
6144  * for orphan MDT-objects.
6145  *
6146  * \param[in] env       pointer to the thread context
6147  * \param[in] com       pointer to the lfsck component
6148  * \param[in] lr        pointer to the lfsck request
6149  */
6150 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6151                                                     struct lfsck_component *com,
6152                                                     struct lfsck_request *lr)
6153 {
6154         struct lfsck_async_interpret_args *laia  =
6155                                 &lfsck_env_info(env)->lti_laia2;
6156         struct lfsck_assistant_data       *lad   = com->lc_data;
6157         struct lfsck_namespace            *ns    = com->lc_file_ram;
6158         struct lfsck_instance             *lfsck = com->lc_lfsck;
6159         struct lfsck_tgt_descs            *ltds  = &lfsck->li_mdt_descs;
6160         struct lfsck_tgt_desc             *ltd;
6161         struct ptlrpc_request_set         *set;
6162         int                                rc    = 0;
6163         ENTRY;
6164
6165         set = ptlrpc_prep_set();
6166         if (set == NULL)
6167                 GOTO(out, rc = -ENOMEM);
6168
6169         lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6170         memset(laia, 0, sizeof(*laia));
6171         lad->lad_touch_gen++;
6172
6173         spin_lock(&ltds->ltd_lock);
6174         while (!list_empty(&lad->lad_mdt_list)) {
6175                 ltd = list_entry(lad->lad_mdt_list.next,
6176                                  struct lfsck_tgt_desc,
6177                                  ltd_namespace_list);
6178                 if (ltd->ltd_namespace_gen == lad->lad_touch_gen)
6179                         break;
6180
6181                 ltd->ltd_namespace_gen = lad->lad_touch_gen;
6182                 list_move_tail(&ltd->ltd_namespace_list,
6183                                &lad->lad_mdt_list);
6184                 if (!lad->lad_incomplete ||
6185                     !cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
6186                         ltd->ltd_namespace_failed = 0;
6187                         continue;
6188                 }
6189
6190                 ltd->ltd_namespace_failed = 1;
6191                 spin_unlock(&ltds->ltd_lock);
6192                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6193                         lfsck_namespace_assistant_sync_failures_interpret,
6194                         laia, LFSCK_NOTIFY);
6195                 if (rc != 0)
6196                         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6197                                "to sync failure with MDT %x: rc = %d\n",
6198                                lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6199
6200                 spin_lock(&ltds->ltd_lock);
6201         }
6202         spin_unlock(&ltds->ltd_lock);
6203
6204         rc = ptlrpc_set_wait(set);
6205         ptlrpc_set_destroy(set);
6206
6207         GOTO(out, rc);
6208
6209 out:
6210         if (rc != 0)
6211                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6212                        "to sync failure with MDTs, and related MDTs "
6213                        "may handle orphan improperly: rc = %d\n",
6214                        lfsck_lfsck2name(lfsck), rc);
6215
6216         EXIT;
6217 }
6218
6219 struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6220         .la_handler_p1          = lfsck_namespace_assistant_handler_p1,
6221         .la_handler_p2          = lfsck_namespace_assistant_handler_p2,
6222         .la_fill_pos            = lfsck_namespace_assistant_fill_pos,
6223         .la_double_scan_result  = lfsck_namespace_double_scan_result,
6224         .la_req_fini            = lfsck_namespace_assistant_req_fini,
6225         .la_sync_failures       = lfsck_namespace_assistant_sync_failures,
6226 };
6227
6228 /**
6229  * Verify the specified linkEA entry for the given directory object.
6230  * If the object has no such linkEA entry or it has more other linkEA
6231  * entries, then re-generate the linkEA with the given information.
6232  *
6233  * \param[in] env       pointer to the thread context
6234  * \param[in] dev       pointer to the dt_device
6235  * \param[in] obj       pointer to the dt_object to be handled
6236  * \param[in] cname     the name for the child in the parent directory
6237  * \param[in] pfid      the parent directory's FID for the linkEA
6238  *
6239  * \retval              0 for success
6240  * \retval              negative error number on failure
6241  */
6242 int lfsck_verify_linkea(const struct lu_env *env, struct dt_device *dev,
6243                         struct dt_object *obj, const struct lu_name *cname,
6244                         const struct lu_fid *pfid)
6245 {
6246         struct linkea_data       ldata  = { 0 };
6247         struct lu_buf            linkea_buf;
6248         struct thandle          *th;
6249         int                      rc;
6250         int                      fl     = LU_XATTR_CREATE;
6251         bool                     dirty  = false;
6252         ENTRY;
6253
6254         LASSERT(S_ISDIR(lfsck_object_type(obj)));
6255
6256         rc = lfsck_links_read(env, obj, &ldata);
6257         if (rc == -ENODATA) {
6258                 dirty = true;
6259         } else if (rc == 0) {
6260                 fl = LU_XATTR_REPLACE;
6261                 if (ldata.ld_leh->leh_reccount != 1) {
6262                         dirty = true;
6263                 } else {
6264                         rc = linkea_links_find(&ldata, cname, pfid);
6265                         if (rc != 0)
6266                                 dirty = true;
6267                 }
6268         }
6269
6270         if (!dirty)
6271                 RETURN(rc);
6272
6273         rc = linkea_data_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf);
6274         if (rc != 0)
6275                 RETURN(rc);
6276
6277         rc = linkea_add_buf(&ldata, cname, pfid);
6278         if (rc != 0)
6279                 RETURN(rc);
6280
6281         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6282                        ldata.ld_leh->leh_len);
6283         th = dt_trans_create(env, dev);
6284         if (IS_ERR(th))
6285                 RETURN(PTR_ERR(th));
6286
6287         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6288                                   XATTR_NAME_LINK, fl, th);
6289         if (rc != 0)
6290                 GOTO(stop, rc);
6291
6292         rc = dt_trans_start_local(env, dev, th);
6293         if (rc != 0)
6294                 GOTO(stop, rc);
6295
6296         dt_write_lock(env, obj, 0);
6297         rc = dt_xattr_set(env, obj, &linkea_buf,
6298                           XATTR_NAME_LINK, fl, th, BYPASS_CAPA);
6299         dt_write_unlock(env, obj);
6300
6301         GOTO(stop, rc);
6302
6303 stop:
6304         dt_trans_stop(env, dev, th);
6305         return rc;
6306 }
6307
6308 /**
6309  * Get the name and parent directory's FID from the first linkEA entry.
6310  *
6311  * \param[in] env       pointer to the thread context
6312  * \param[in] obj       pointer to the object which get linkEA from
6313  * \param[out] name     pointer to the buffer to hold the name
6314  *                      in the first linkEA entry
6315  * \param[out] pfid     pointer to the buffer to hold the parent
6316  *                      directory's FID in the first linkEA entry
6317  *
6318  * \retval              0 for success
6319  * \retval              negative error number on failure
6320  */
6321 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6322                           char *name, struct lu_fid *pfid)
6323 {
6324         struct lu_name           *cname = &lfsck_env_info(env)->lti_name;
6325         struct linkea_data        ldata = { 0 };
6326         int                       rc;
6327
6328         rc = lfsck_links_read(env, obj, &ldata);
6329         if (rc != 0)
6330                 return rc;
6331
6332         linkea_first_entry(&ldata);
6333         if (ldata.ld_lee == NULL)
6334                 return -ENODATA;
6335
6336         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6337         /* To guarantee the 'name' is terminated with '0'. */
6338         memcpy(name, cname->ln_name, cname->ln_namelen);
6339         name[cname->ln_namelen] = 0;
6340
6341         return 0;
6342 }
6343
6344 /**
6345  * Remove the name entry from the parent directory.
6346  *
6347  * No need to care about the object referenced by the name entry,
6348  * either the name entry is invalid or redundant, or the referenced
6349  * object has been processed has been or will be handled by others.
6350  *
6351  * \param[in] env       pointer to the thread context
6352  * \param[in] lfsck     pointer to the lfsck instance
6353  * \param[in] parent    pointer to the lost+found object
6354  * \param[in] name      the name for the name entry to be removed
6355  * \param[in] type      the type for the name entry to be removed
6356  *
6357  * \retval              0 for success
6358  * \retval              negative error number on failure
6359  */
6360 int lfsck_remove_name_entry(const struct lu_env *env,
6361                             struct lfsck_instance *lfsck,
6362                             struct dt_object *parent,
6363                             const char *name, __u32 type)
6364 {
6365         struct dt_device        *dev    = lfsck->li_next;
6366         struct thandle          *th;
6367         struct lustre_handle     lh     = { 0 };
6368         int                      rc;
6369         ENTRY;
6370
6371         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
6372                               MDS_INODELOCK_UPDATE, LCK_EX);
6373         if (rc != 0)
6374                 RETURN(rc);
6375
6376         th = dt_trans_create(env, dev);
6377         if (IS_ERR(th))
6378                 GOTO(unlock, rc = PTR_ERR(th));
6379
6380         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
6381         if (rc != 0)
6382                 GOTO(stop, rc);
6383
6384         if (S_ISDIR(type)) {
6385                 rc = dt_declare_ref_del(env, parent, th);
6386                 if (rc != 0)
6387                         GOTO(stop, rc);
6388         }
6389
6390         rc = dt_trans_start(env, dev, th);
6391         if (rc != 0)
6392                 GOTO(stop, rc);
6393
6394         rc = dt_delete(env, parent, (const struct dt_key *)name, th,
6395                        BYPASS_CAPA);
6396         if (rc != 0)
6397                 GOTO(stop, rc);
6398
6399         if (S_ISDIR(type)) {
6400                 dt_write_lock(env, parent, 0);
6401                 rc = dt_ref_del(env, parent, th);
6402                 dt_write_unlock(env, parent);
6403         }
6404
6405         GOTO(stop, rc);
6406
6407 stop:
6408         dt_trans_stop(env, dev, th);
6409
6410 unlock:
6411         lfsck_ibits_unlock(&lh, LCK_EX);
6412
6413         CDEBUG(D_LFSCK, "%s: remove name entry "DFID"/%s "
6414                "with type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6415                PFID(lfsck_dto2fid(parent)), name, type, rc);
6416
6417         return rc;
6418 }
6419
6420 /**
6421  * Update the object's name entry with the given FID.
6422  *
6423  * \param[in] env       pointer to the thread context
6424  * \param[in] lfsck     pointer to the lfsck instance
6425  * \param[in] parent    pointer to the parent directory that holds
6426  *                      the name entry
6427  * \param[in] name      the name for the entry to be updated
6428  * \param[in] pfid      the new PFID for the name entry
6429  * \param[in] type      the type for the name entry to be updated
6430  *
6431  * \retval              0 for success
6432  * \retval              negative error number on failure
6433  */
6434 int lfsck_update_name_entry(const struct lu_env *env,
6435                             struct lfsck_instance *lfsck,
6436                             struct dt_object *parent, const char *name,
6437                             const struct lu_fid *pfid, __u32 type)
6438 {
6439         struct dt_insert_rec    *rec    = &lfsck_env_info(env)->lti_dt_rec;
6440         struct dt_device        *dev    = lfsck->li_next;
6441         struct lustre_handle     lh     = { 0 };
6442         struct thandle          *th;
6443         int                      rc;
6444         bool                     exists = true;
6445         ENTRY;
6446
6447         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
6448                               MDS_INODELOCK_UPDATE, LCK_EX);
6449         if (rc != 0)
6450                 RETURN(rc);
6451
6452         th = dt_trans_create(env, dev);
6453         if (IS_ERR(th))
6454                 GOTO(unlock, rc = PTR_ERR(th));
6455
6456         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
6457         if (rc != 0)
6458                 GOTO(stop, rc);
6459
6460         rec->rec_type = type;
6461         rec->rec_fid = pfid;
6462         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
6463                                (const struct dt_key *)name, th);
6464         if (rc != 0)
6465                 GOTO(stop, rc);
6466
6467         rc = dt_declare_ref_add(env, parent, th);
6468         if (rc != 0)
6469                 GOTO(stop, rc);
6470
6471         rc = dt_trans_start(env, dev, th);
6472         if (rc != 0)
6473                 GOTO(stop, rc);
6474
6475         rc = dt_delete(env, parent, (const struct dt_key *)name, th,
6476                        BYPASS_CAPA);
6477         if (rc == -ENOENT) {
6478                 exists = false;
6479                 rc = 0;
6480         }
6481
6482         if (rc != 0)
6483                 GOTO(stop, rc);
6484
6485         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
6486                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
6487         if (rc == 0 && S_ISDIR(type) && !exists) {
6488                 dt_write_lock(env, parent, 0);
6489                 rc = dt_ref_add(env, parent, th);
6490                 dt_write_unlock(env, parent);
6491         }
6492
6493         GOTO(stop, rc);
6494
6495 stop:
6496         dt_trans_stop(env, dev, th);
6497
6498 unlock:
6499         lfsck_ibits_unlock(&lh, LCK_EX);
6500
6501         CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
6502                " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6503                PFID(lfsck_dto2fid(parent)), name, PFID(pfid), type, rc);
6504
6505         return rc;
6506 }
6507
6508 int lfsck_namespace_setup(const struct lu_env *env,
6509                           struct lfsck_instance *lfsck)
6510 {
6511         struct lfsck_component  *com;
6512         struct lfsck_namespace  *ns;
6513         struct dt_object        *root = NULL;
6514         struct dt_object        *obj;
6515         int                      rc;
6516         ENTRY;
6517
6518         LASSERT(lfsck->li_master);
6519
6520         OBD_ALLOC_PTR(com);
6521         if (com == NULL)
6522                 RETURN(-ENOMEM);
6523
6524         INIT_LIST_HEAD(&com->lc_link);
6525         INIT_LIST_HEAD(&com->lc_link_dir);
6526         init_rwsem(&com->lc_sem);
6527         atomic_set(&com->lc_ref, 1);
6528         com->lc_lfsck = lfsck;
6529         com->lc_type = LFSCK_TYPE_NAMESPACE;
6530         com->lc_ops = &lfsck_namespace_ops;
6531         com->lc_data = lfsck_assistant_data_init(
6532                         &lfsck_namespace_assistant_ops,
6533                         "lfsck_namespace");
6534         if (com->lc_data == NULL)
6535                 GOTO(out, rc = -ENOMEM);
6536
6537         com->lc_file_size = sizeof(struct lfsck_namespace);
6538         OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6539         if (com->lc_file_ram == NULL)
6540                 GOTO(out, rc = -ENOMEM);
6541
6542         OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6543         if (com->lc_file_disk == NULL)
6544                 GOTO(out, rc = -ENOMEM);
6545
6546         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6547         if (IS_ERR(root))
6548                 GOTO(out, rc = PTR_ERR(root));
6549
6550         if (unlikely(!dt_try_as_dir(env, root)))
6551                 GOTO(out, rc = -ENOTDIR);
6552
6553         obj = local_index_find_or_create(env, lfsck->li_los, root,
6554                                          lfsck_namespace_name,
6555                                          S_IFREG | S_IRUGO | S_IWUSR,
6556                                          &dt_lfsck_features);
6557         if (IS_ERR(obj))
6558                 GOTO(out, rc = PTR_ERR(obj));
6559
6560         com->lc_obj = obj;
6561         rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features);
6562         if (rc != 0)
6563                 GOTO(out, rc);
6564
6565         rc = lfsck_namespace_load(env, com);
6566         if (rc == -ENODATA)
6567                 rc = lfsck_namespace_init(env, com);
6568         else if (rc < 0)
6569                 rc = lfsck_namespace_reset(env, com, true);
6570         if (rc != 0)
6571                 GOTO(out, rc);
6572
6573         ns = com->lc_file_ram;
6574         switch (ns->ln_status) {
6575         case LS_INIT:
6576         case LS_COMPLETED:
6577         case LS_FAILED:
6578         case LS_STOPPED:
6579                 spin_lock(&lfsck->li_lock);
6580                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6581                 spin_unlock(&lfsck->li_lock);
6582                 break;
6583         default:
6584                 CERROR("%s: unknown lfsck_namespace status %d\n",
6585                        lfsck_lfsck2name(lfsck), ns->ln_status);
6586                 /* fall through */
6587         case LS_SCANNING_PHASE1:
6588         case LS_SCANNING_PHASE2:
6589                 /* No need to store the status to disk right now.
6590                  * If the system crashed before the status stored,
6591                  * it will be loaded back when next time. */
6592                 ns->ln_status = LS_CRASHED;
6593                 /* fall through */
6594         case LS_PAUSED:
6595         case LS_CRASHED:
6596                 spin_lock(&lfsck->li_lock);
6597                 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6598                 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
6599                 spin_unlock(&lfsck->li_lock);
6600                 break;
6601         }
6602
6603         GOTO(out, rc = 0);
6604
6605 out:
6606         if (root != NULL && !IS_ERR(root))
6607                 lu_object_put(env, &root->do_lu);
6608         if (rc != 0) {
6609                 lfsck_component_cleanup(env, com);
6610                 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
6611                        lfsck_lfsck2name(lfsck), rc);
6612         }
6613         return rc;
6614 }