lustre/lfsck/lfsck_striped_dir.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) 2014, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_striped_dir.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 /*
  32  * About the verification for striped directory. Some rules and assumptions:
  33  *
  34  * 1) lmv_magic: The magic may be wrong. But it is almost impossible (1/2^32
  35  *    probability) that a master LMV EA claims as a slave LMV EA by wrong,
  36  *    so we can ignore such race case and the reverse case.
  37  *
  38  * 2) lmv_master_mdt_index: The master index can be self-verified by compared
  39  *    with the MDT index directly. The slave stripe index can be verified by
  40  *    compared with the file name. Although both the name entry and the LMV EA
  41  *    can be wrong, it is almost impossible that they hit the same bad data
  42  *    So if they match each other, then trust them. Similarly, for the shard,
  43  *    it stores index in both slave LMV EA and in linkEA, if the two copies
  44  *    match, then trust them.
  45  *
  46  * 3) lmv_hash_type: The valid hash type should be LMV_HASH_TYPE_ALL_CHARS or
  47  *    LMV_HASH_TYPE_FNV_1A_64. If the LFSCK instance on some slave finds that
  48  *    the name hash against the hash function does not match the MDT, then it
  49  *    will change the master LMV EA hash type as LMV_HASH_TYPE_UNKNOWN. With
  50  *    such hash type, the whole striped directory still can be accessed via
  51  *    lookup/readdir, and also support unlink, but cannot add new name entry.
  52  *
  53  * 3.1) If the master hash type is one of the valid values, then trust the
  54  *      master LMV EA. Because:
  55  *
  56  * 3.1.1) The master hash type is visible to the client and used by the client.
  57  *
  58  * 3.1.2) For a given name, different hash types may map the name entry to the
  59  *        same MDT. So simply checking one name entry or some name entries may
  60  *        cannot verify whether the hash type is correct or not.
  61  *
  62  * 3.1.3) Different shards can claim different hash types, it is not easy to
  63  *        distinguish which ones are correct. Even though the master is wrong,
  64  *        as the LFSCK processing, some LFSCK instance on other MDT may finds
  65  *        unmatched name hash, then it will change the master hash type to
  66  *        LMV_HASH_TYPE_UNKNOWN as described above. The worst case is euqal
  67  *        to the case without the LFSCK.
  68  *
  69  * 3.2) If the master hash type is invalid, nor LMV_HASH_TYPE_UNKNOWN, then
  70  *      trust the first shard with valid hash type (ALL_CHARS or FNV_1A_64).
  71  *      If the shard is also worng, means there are double failures, then as
  72  *      the LFSCK processing, other LFSCK instances on the other MDTs may
  73  *      find unmatched name hash, and then, the master hash type will be
  74  *      changed to LMV_HASH_TYPE_UNKNOWN as described in the 3).
  75  *
  76  * 3.3) If the master hash type is LMV_HASH_TYPE_UNKNOWN, then it is possible
  77  *      that some other LFSCK instance on other MDT found bad name hash, then
  78  *      changed the master hash type to LMV_HASH_TYPE_UNKNOWN as described in
  79  *      the 3). But it also maybe because of data corruption in master LMV EA.
  80  *      To make such two cases to be distinguishable, when the LFSCK changes
  81  *      the master hash type to LMV_HASH_TYPE_UNKNOWN, it will mark in the
  82  *      master LMV EA (new lmv flags LMV_HASH_FLAG_BAD_TYPE). Then subsequent
  83  *      LFSCK checking can distinguish them: for former case, turst the master
  84  *      LMV EA with nothing to be done; otherwise, trust the first shard with
  85  *      valid hash type (ALL_CHARS or FNV_1A_64) as the 3.2) does.
  86  *
  87  * 4) lmv_stripe_count: For a shard of a striped directory, if its index has
  88  *    been verified as the 2), then the stripe count must be larger than its
  89  *    index. For the master object, by scanning each shard's index, the LFSCK
  90  *    can know the highest index, and the stripe count must be larger than the
  91  *    known highest index. If the stipe count in the LMV EA matches above two
  92  *    rules, then it is may be trustable. If both the master claimed stripe
  93  *    count and the slave claimed stripe count match each own rule, but they
  94  *    are not the same, then trust the master. Because the stripe count in
  95  *    the master LMV EA is visible to client and used to distribute the name
  96  *    entry to some shard, but the slave LMV EA is only used for verification
  97  *    and invisible to client.
  98  *
  99  * 5) If the master LMV EA is lost, then there are two possible cases:
 100  *
 101  * 5.1) The slave claims slave LMV EA by wrong, means that the parent was not
 102  *      a striped directory, but its sub-directory has a wrong slave LMV EA.
 103  *      It is very very race case, similar as the 1), can be ignored.
 104  *
 105  * 5.2) The parent directory is a striped directory, but the master LMV EA
 106  *      is lost or crashed. Then the LFSCK needs to re-generate the master
 107  *      LMV EA: the lmv_master_mdt_index is from the MDT device index; the
 108  *      lmv_hash_type is from the first valid shard; the lmv_stripe_count
 109  *      will be calculated via scanning all the shards.
 110  *
 111  * 5.2.1) Before re-generating the master LMV EA, the LFSCK needs to check
 112  *        whether someone has created some file(s) under the master object
 113  *        after the master LMV EA disappear. If yes, the LFSCK will cannot
 114  *        re-generate the master LMV EA, otherwise, such new created files
 115  *        will be invisible to client. Under such case, the LFSCK will mark
 116  *        the master object as read only (without master LMV EA). Then all
 117  *        things under the master MDT-object, including those new created
 118  *        files and the shards themselves, will be visibile to client. And
 119  *        then the administrator can handle the bad striped directory with
 120  *        more human knowledge.
 121  *
 122  * 5.2.2) If someone created some special sub-directory under the master
 123  *        MDT-object with the same naming rule as shard name $FID:$index,
 124  *        as to the LFSCK cannot detect it before re-generating the master
 125  *        LMV EA, then such sub-directory itself will be invisible after
 126  *        the LFSCK re-generating the master LMV EA. The sub-items under
 127  *        such sub-directory are still visible to client. As the LFSCK
 128  *        processing, if such sub-directory cause some conflict with other
 129  *        normal shard, such as the index conflict, then the LFSCK will
 130  *        remove the master LMV EA and change the master MDT-object to
 131  *        read-only mode as the 5.2.1). But if there is no conflict, the
 132  *        LFSCK will regard such sub-directory as a striped shard that
 133  *        lost its slave LMV EA, and will re-generate slave LMV EA for it.
 134  *
 135  * 5.2.3) Anytime, if the LFSCK found some shards name/index conflict,
 136  *        and cannot make the distinguish which one is right, then it
 137  *        will remove the master LMV EA and change the MDT-object to
 138  *        read-only mode as the 5.2.2).
 139  */
 140
 141 #define DEBUG_SUBSYSTEM S_LFSCK
 142
 143 #include <lustre/lustre_idl.h>
 144 #include <lu_object.h>
 145 #include <dt_object.h>
 146 #include <md_object.h>
 147 #include <lustre_fid.h>
 148 #include <lustre_lib.h>
 149 #include <lustre_net.h>
 150 #include <lustre_lmv.h>
 151 #include <lustre/lustre_user.h>
 152
 153 #include "lfsck_internal.h"
 154
 155 void lfsck_lmv_put(const struct lu_env *env, struct lfsck_lmv *llmv)
 156 {
 157         if (llmv != NULL && atomic_dec_and_test(&llmv->ll_ref)) {
 158                 if (llmv->ll_inline) {
 159                         struct lfsck_lmv_unit   *llu;
 160                         struct lfsck_instance   *lfsck;
 161
 162                         llu = list_entry(llmv, struct lfsck_lmv_unit, llu_lmv);
 163                         lfsck = llu->llu_lfsck;
 164
 165                         spin_lock(&lfsck->li_lock);
 166                         list_del(&llu->llu_link);
 167                         spin_unlock(&lfsck->li_lock);
 168
 169                         lfsck_object_put(env, llu->llu_obj);
 170
 171                         LASSERT(llmv->ll_lslr != NULL);
 172
 173                         OBD_FREE_LARGE(llmv->ll_lslr,
 174                                        sizeof(*llmv->ll_lslr) *
 175                                        llmv->ll_stripes_allocated);
 176                         OBD_FREE_PTR(llu);
 177                 } else {
 178                         if (llmv->ll_lslr != NULL)
 179                                 OBD_FREE_LARGE(llmv->ll_lslr,
 180                                         sizeof(*llmv->ll_lslr) *
 181                                         llmv->ll_stripes_allocated);
 182
 183                         OBD_FREE_PTR(llmv);
 184                 }
 185         }
 186 }
 187
 188 /**
 189  * Mark the specified directory as read-only by set LUSTRE_IMMUTABLE_FL.
 190  *
 191  * The caller has taken the ldlm lock on the @obj already.
 192  *
 193  * \param[in] env       pointer to the thread context
 194  * \param[in] com       pointer to the lfsck component
 195  * \param[in] obj       pointer to the object to be handled
 196  * \param[in] del_lmv   true if need to drop the LMV EA
 197  *
 198  * \retval              positive number if nothing to be done
 199  * \retval              zero for succeed
 200  * \retval              negative error number on failure
 201  */
 202 static int lfsck_disable_master_lmv(const struct lu_env *env,
 203                                     struct lfsck_component *com,
 204                                     struct dt_object *obj, bool del_lmv)
 205 {
 206         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 207         struct lu_attr                  *la     = &info->lti_la;
 208         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 209         struct dt_device                *dev    = lfsck_obj2dt_dev(obj);
 210         struct thandle                  *th     = NULL;
 211         int                              rc     = 0;
 212         ENTRY;
 213
 214         th = dt_trans_create(env, dev);
 215         if (IS_ERR(th))
 216                 GOTO(log, rc = PTR_ERR(th));
 217
 218         if (del_lmv) {
 219                 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LMV, th);
 220                 if (rc != 0)
 221                         GOTO(stop, rc);
 222         }
 223
 224         la->la_valid = LA_FLAGS;
 225         rc = dt_declare_attr_set(env, obj, la, th);
 226         if (rc != 0)
 227                 GOTO(stop, rc);
 228
 229         rc = dt_trans_start_local(env, dev, th);
 230         if (rc != 0)
 231                 GOTO(stop, rc);
 232
 233         dt_write_lock(env, obj, 0);
 234         if (unlikely(lfsck_is_dead_obj(obj)))
 235                 GOTO(unlock, rc = 1);
 236
 237         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
 238                 GOTO(unlock, rc = 0);
 239
 240         if (del_lmv) {
 241                 rc = dt_xattr_del(env, obj, XATTR_NAME_LMV, th, BYPASS_CAPA);
 242                 if (rc != 0)
 243                         GOTO(unlock, rc);
 244         }
 245
 246         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
 247         if (rc == 0 && !(la->la_flags & LUSTRE_IMMUTABLE_FL)) {
 248                 la->la_valid = LA_FLAGS;
 249                 la->la_flags |= LUSTRE_IMMUTABLE_FL;
 250                 rc = dt_attr_set(env, obj, la, th, BYPASS_CAPA);
 251         }
 252
 253         GOTO(unlock, rc);
 254
 255 unlock:
 256         dt_write_unlock(env, obj);
 257
 258 stop:
 259         dt_trans_stop(env, dev, th);
 260
 261 log:
 262         CDEBUG(D_LFSCK, "%s: namespace LFSCK set the master MDT-object of "
 263                "the striped directory "DFID" as read-only: rc = %d\n",
 264                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
 265
 266         if (rc <= 0) {
 267                 struct lfsck_namespace *ns = com->lc_file_ram;
 268
 269                 ns->ln_flags |= LF_INCONSISTENT;
 270                 if (rc == 0)
 271                         ns->ln_striped_dirs_disabled++;
 272         }
 273
 274         return rc;
 275 }
 276
 277 static inline bool lfsck_is_valid_slave_lmv(struct lmv_mds_md_v1 *lmv)
 278 {
 279         return lmv->lmv_stripe_count >= 1 &&
 280                lmv->lmv_stripe_count <= LFSCK_LMV_MAX_STRIPES &&
 281                lmv->lmv_stripe_count > lmv->lmv_master_mdt_index &&
 282                lmv_is_known_hash_type(lmv->lmv_hash_type);
 283 }
 284
 285 /**
 286  * Remove the striped directory's master LMV EA and mark it as read-only.
 287  *
 288  * Take ldlm lock on the striped directory before calling the
 289  * lfsck_disable_master_lmv().
 290  *
 291  * \param[in] env       pointer to the thread context
 292  * \param[in] com       pointer to the lfsck component
 293  * \param[in] lnr       pointer to the namespace request that contains the
 294  *                      striped directory to be handled and other information
 295  *
 296  * \retval              positive number if nothing to be done
 297  * \retval              zero for succeed
 298  * \retval              negative error number on failure
 299  */
 300 static int lfsck_remove_lmv(const struct lu_env *env,
 301                             struct lfsck_component *com,
 302                             struct lfsck_namespace_req *lnr)
 303 {
 304         struct dt_object        *obj    = lnr->lnr_obj;
 305         struct lustre_handle     lh     = { 0 };
 306         int                      rc;
 307
 308         lnr->lnr_lmv->ll_ignore = 1;
 309         rc = lfsck_ibits_lock(env, com->lc_lfsck, obj, &lh,
 310                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
 311                               LCK_EX);
 312         if (rc == 0) {
 313                 rc = lfsck_disable_master_lmv(env, com, obj, true);
 314                 lfsck_ibits_unlock(&lh, LCK_EX);
 315         }
 316
 317         return rc;
 318 }
 319
 320 /**
 321  * Remove the name entry from the striped directory's master MDT-object.
 322  *
 323  * \param[in] env       pointer to the thread context
 324  * \param[in] com       pointer to the lfsck component
 325  * \param[in] dir       pointer to the striped directory
 326  * \param[in] fid       the shard's FID which name entry will be removed
 327  * \param[in] index     the shard's index which name entry will be removed
 328  *
 329  * \retval              positive number for repaired successfully
 330  * \retval              0 if nothing to be repaired
 331  * \retval              negative error number on failure
 332  */
 333 static int lfsck_remove_dirent(const struct lu_env *env,
 334                                struct lfsck_component *com,
 335                                struct dt_object *dir,
 336                                const struct lu_fid *fid, __u32 index)
 337 {
 338         struct lfsck_thread_info        *info = lfsck_env_info(env);
 339         struct dt_object                *obj;
 340         int                              rc;
 341
 342         snprintf(info->lti_tmpbuf2, sizeof(info->lti_tmpbuf2), DFID":%u",
 343                  PFID(fid), index);
 344         obj = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, fid);
 345         if (IS_ERR(obj))
 346                 return PTR_ERR(obj);
 347
 348         rc = lfsck_namespace_repair_dirent(env, com, dir, obj,
 349                                         info->lti_tmpbuf2, info->lti_tmpbuf2,
 350                                         S_IFDIR, false, false);
 351         lfsck_object_put(env, obj);
 352         if (rc > 0) {
 353                 struct lfsck_namespace *ns = com->lc_file_ram;
 354
 355                 ns->ln_dirent_repaired++;
 356         }
 357
 358         return rc;
 359 }
 360
 361 /**
 362  * Remove old shard's name entry and refill the @lslr slot with new shard.
 363  *
 364  * Some old shard held the specified @lslr slot, but it is an invalid shard.
 365  * This function will remove the bad shard's name entry, and refill the @lslr
 366  * slot with the new shard.
 367  *
 368  * \param[in] env       pointer to the thread context
 369  * \param[in] com       pointer to the lfsck component
 370  * \param[in] lslr      pointer to lfsck_disable_master_lmv slot which content
 371  *                      will be replaced by the given information
 372  * \param[in] lnr       contain the shard's FID to be used to fill the
 373  *                      @lslr slot, it also records the known max filled index
 374  *                      and the known max stripe count
 375  * \param[in] lmv       contain the slave LMV EA to be used to fill the
 376  *                      @lslr slot
 377  * \param[in] index     the old shard's index in the striped directory
 378  * \param[in] flags     the new shard's flags in the @lslr slot
 379  *
 380  * \retval              zero for succeed
 381  * \retval              negative error number on failure
 382  */
 383 static int lfsck_replace_lmv(const struct lu_env *env,
 384                              struct lfsck_component *com,
 385                              struct lfsck_slave_lmv_rec *lslr,
 386                              struct lfsck_namespace_req *lnr,
 387                              struct lmv_mds_md_v1 *lmv,
 388                              __u32 index, __u32 flags)
 389 {
 390         int rc;
 391
 392         rc = lfsck_remove_dirent(env, com, lnr->lnr_obj,
 393                                  &lslr->lslr_fid, index);
 394         if (rc < 0)
 395                 return rc;
 396
 397         lslr->lslr_fid = lnr->lnr_fid;
 398         lslr->lslr_flags = flags;
 399         if (lmv != NULL) {
 400                 struct lfsck_lmv *llmv = lnr->lnr_lmv;
 401
 402                 lslr->lslr_stripe_count = lmv->lmv_stripe_count;
 403                 lslr->lslr_index = lmv->lmv_master_mdt_index;
 404                 lslr->lslr_hash_type = lmv->lmv_hash_type;
 405
 406                 if (flags == LSLF_NONE &&
 407                     llmv->ll_hash_type == LMV_HASH_TYPE_UNKNOWN &&
 408                     lmv_is_known_hash_type(lmv->lmv_hash_type))
 409                         llmv->ll_hash_type = lmv->lmv_hash_type;
 410
 411                 if (flags == LSLF_NONE &&
 412                     lslr->lslr_stripe_count <= LFSCK_LMV_MAX_STRIPES &&
 413                     llmv->ll_max_stripe_count < lslr->lslr_stripe_count)
 414                         llmv->ll_max_stripe_count = lslr->lslr_stripe_count;
 415         } else {
 416                 lslr->lslr_stripe_count = 0;
 417                 lslr->lslr_index = 0;
 418                 lslr->lslr_hash_type = 0;
 419         }
 420
 421         return 0;
 422 }
 423
 424 /**
 425  * Record the slave LMV EA in the lfsck_lmv::ll_lslr.
 426  *
 427  * If the lfsck_lmv::ll_lslr slot corresponding to the given @shard_idx is free,
 428  * then fill the slot with the given @lnr/@lmv/@flags directly (maybe need to
 429  * extend the lfsck_lmv::ll_lslr buffer).
 430  *
 431  * If the lfsck_lmv::ll_lslr slot corresponding to the given @shard_idx is taken
 432  * by other shard, then the LFSCK will try to resolve the conflict by checking
 433  * the two conflict shards' flags, and try other possible slot (if one of them
 434  * claims another possible @shard_idx).
 435  *
 436  * 1) If one of the two conflict shards can be recorded in another slot, then
 437  *    it is OK, go ahead. Otherwise,
 438  *
 439  * 2) If one of them is dangling name entry, then remove (one of) the dangling
 440  *    name entry (and replace related @lslr slot if needed). Otherwise,
 441  *
 442  * 3) If one of them has no slave LMV EA, then check whether the master LMV
 443  *    EA has ever been lost and re-generated (LMV_HASH_FLAG_LOST_LMV in the
 444  *    master LMV EA).
 445  *
 446  * 3.1) If yes, then it is possible that such object is not a real shard of
 447  *      the striped directory, instead, it was created by someone after the
 448  *      master LMV EA lost with the name that matches the shard naming rule.
 449  *      Then the LFSCK will remove the master LMV EA and mark the striped
 450  *      directory as read-only to allow those non-shard files to be visible
 451  *      to client.
 452  *
 453  * 3.2) If no, then remove (one of) the object what has no slave LMV EA.
 454  *
 455  * 4) If all above efforts cannot work, then the LFSCK cannot know how to
 456  *    recover the striped directory. To make the administrator can see the
 457  *    conflicts, the LFSCK will remove the master LMV EA and mark the striped
 458  *    directory as read-only.
 459  *
 460  * This function may be called recursively, to prevent overflow, we define
 461  * LFSCK_REC_LMV_MAX_DEPTH to restrict the recursive call depth.
 462  *
 463  * \param[in] env       pointer to the thread context
 464  * \param[in] com       pointer to the lfsck component
 465  * \param[in] lnr       contain the shard's FID to fill the @lslr slot,
 466  *                      it also records the known max filled index and
 467  *                      the known max stripe count
 468  * \param[in] lmv       pointer to the slave LMV EA to be recorded
 469  * \param[in] shard_idx the shard's index used for locating the @lslr slot,
 470  *                      it can be the index stored in the shard's name,
 471  *                      it also can be the index stored in the slave LMV EA
 472  *                      (for recursive case)
 473  * \param[in] flags     the shard's flags to be recorded in the @lslr slot
 474  *                      to indicate the shard status, such as whether has
 475  *                      slave LMV EA, whether dangling name entry, whether
 476  *                      the name entry and slave LMV EA unmatched, and ect
 477  * \param[in] flags2    when be called recursively, the @flags2 tells the
 478  *                      former conflict shard's flags in the @lslr slot.
 479  * \param[in,out] depth To prevent to be called recurisively too deep,
 480  *                      we define the max depth can be called recursively
 481  *                      (LFSCK_REC_LMV_MAX_DEPTH)
 482  *
 483  * \retval              zero for succeed
 484  * \retval              "-ERANGE" for invalid @shard_idx
 485  * \retval              "-EEXIST" for the required lslr slot has been
 486  *                      occupied by other shard
 487  * \retval              other negative error number on failure
 488  */
 489 static int lfsck_record_lmv(const struct lu_env *env,
 490                             struct lfsck_component *com,
 491                             struct lfsck_namespace_req *lnr,
 492                             struct lmv_mds_md_v1 *lmv, __u32 shard_idx,
 493                             __u32 flags, __u32 flags2, __u32 *depth)
 494 {
 495         struct lfsck_instance      *lfsck = com->lc_lfsck;
 496         struct lfsck_lmv           *llmv  = lnr->lnr_lmv;
 497         struct dt_object           *dir   = lnr->lnr_obj;
 498         const struct lu_fid        *fid   = &lnr->lnr_fid;
 499         struct lfsck_slave_lmv_rec *lslr;
 500         struct lfsck_rec_lmv_save  *lrls;
 501         int                         index = shard_idx;
 502         int                         rc    = 0;
 503         ENTRY;
 504
 505         CDEBUG(D_LFSCK, "%s: record slave LMV EA for the striped directory "
 506                DFID": shard = "DFID", index = %u, flags = %u, flags2 = %u, "
 507                "depth = %d\n", lfsck_lfsck2name(lfsck),
 508                PFID(lfsck_dto2fid(dir)), PFID(fid),
 509                index, flags, flags2, *depth);
 510
 511         if (index < 0 || index >= LFSCK_LMV_MAX_STRIPES)
 512                 RETURN(-ERANGE);
 513
 514         if (index >= llmv->ll_stripes_allocated) {
 515                 struct lfsck_slave_lmv_rec *new_lslr;
 516                 int new_stripes = index + 1;
 517                 size_t old_size = sizeof(*lslr) * llmv->ll_stripes_allocated;
 518
 519                 OBD_ALLOC_LARGE(new_lslr, sizeof(*new_lslr) * new_stripes);
 520                 if (new_lslr == NULL) {
 521                         llmv->ll_failed = 1;
 522
 523                         RETURN(-ENOMEM);
 524                 }
 525
 526                 memcpy(new_lslr, llmv->ll_lslr, old_size);
 527                 OBD_FREE_LARGE(llmv->ll_lslr, old_size);
 528                 llmv->ll_stripes_allocated = new_stripes;
 529                 llmv->ll_lslr = new_lslr;
 530         }
 531
 532         lslr = llmv->ll_lslr + index;
 533         if (unlikely(lu_fid_eq(&lslr->lslr_fid, fid)))
 534                 RETURN(0);
 535
 536         if (fid_is_zero(&lslr->lslr_fid)) {
 537                 lslr->lslr_fid = *fid;
 538                 if (lmv != NULL) {
 539                         lslr->lslr_stripe_count = lmv->lmv_stripe_count;
 540                         lslr->lslr_index = lmv->lmv_master_mdt_index;
 541                         lslr->lslr_hash_type = lmv->lmv_hash_type;
 542
 543                         if (flags == LSLF_NONE &&
 544                             llmv->ll_hash_type == LMV_HASH_TYPE_UNKNOWN &&
 545                             lmv_is_known_hash_type(lmv->lmv_hash_type))
 546                                 llmv->ll_hash_type = lmv->lmv_hash_type;
 547
 548                         if (flags == LSLF_NONE &&
 549                             lslr->lslr_stripe_count <= LFSCK_LMV_MAX_STRIPES &&
 550                             llmv->ll_max_stripe_count < lslr->lslr_stripe_count)
 551                                 llmv->ll_max_stripe_count =
 552                                                         lslr->lslr_stripe_count;
 553                 }
 554
 555                 lslr->lslr_flags = flags;
 556                 llmv->ll_stripes_filled++;
 557
 558                 if (llmv->ll_max_filled_off < index)
 559                         llmv->ll_max_filled_off = index;
 560
 561                 RETURN(0);
 562         }
 563
 564         (*depth)++;
 565         if (flags != LSLF_BAD_INDEX2)
 566                 LASSERTF(*depth == 1, "depth = %d\n", *depth);
 567
 568         /* Handle conflict cases. */
 569         switch (lslr->lslr_flags) {
 570         case LSLF_NONE:
 571         case LSLF_BAD_INDEX2:
 572                 /* The existing one is a normal valid object. */
 573                 switch (flags) {
 574                 case LSLF_NONE:
 575                         /* The two 'valid' name entries claims the same
 576                          * index, the LFSCK cannot distinguish which one
 577                          * is correct. Then remove the master LMV EA to
 578                          * make all shards to be visible to client, and
 579                          * mark the master MDT-object as read-only. The
 580                          * administrator can handle the conflict with
 581                          * more human knowledge. */
 582                         rc = lfsck_remove_lmv(env, com, lnr);
 583                         break;
 584                 case LSLF_BAD_INDEX2:
 585                         GOTO(out, rc = -EEXIST);
 586                 case LSLF_NO_LMVEA:
 587
 588 no_lmvea:
 589                         if (llmv->ll_lmv.lmv_hash_type &
 590                             LMV_HASH_FLAG_LOST_LMV) {
 591                                 /* If the master LMV EA was re-generated
 592                                  * by the former LFSCK reparation, and
 593                                  * before such reparation, someone has
 594                                  * created the conflict object, but the
 595                                  * LFSCK did not detect such conflict,
 596                                  * then we have to remove the master
 597                                  * LMV EA and mark the master MDT-object
 598                                  * as read-only. The administrator can
 599                                  * handle the conflict with more human
 600                                  * knowledge. */
 601                                 rc = lfsck_remove_lmv(env, com, lnr);
 602                         } else {
 603                                 /* Otherwise, remove the current name entry,
 604                                  * and add its FID in the LFSCK tracing file
 605                                  * for further processing. */
 606                                 rc = lfsck_namespace_trace_update(env, com, fid,
 607                                                 LNTF_CHECK_PARENT, true);
 608                                 if (rc == 0)
 609                                         rc = lfsck_remove_dirent(env, com, dir,
 610                                                                  fid, index);
 611                         }
 612
 613                         break;
 614                 case LSLF_DANGLING:
 615                         /* Remove the current dangling name entry. */
 616                         rc = lfsck_remove_dirent(env, com, dir, fid, index);
 617                         break;
 618                 case LSLF_BAD_INDEX1:
 619                         index = lmv->lmv_master_mdt_index;
 620                         lmv->lmv_master_mdt_index = shard_idx;
 621                         /* The name entry claims an index that is conflict
 622                          * with a valid existing name entry, then try the
 623                          * index in the lmv recursively. */
 624                         rc = lfsck_record_lmv(env, com, lnr, lmv, index,
 625                                 LSLF_BAD_INDEX2, lslr->lslr_flags, depth);
 626                         lmv->lmv_master_mdt_index = index;
 627                         if (rc == -ERANGE || rc == -EEXIST)
 628                                 /* The index in the lmv is invalid or
 629                                  * also conflict with other. Then we do
 630                                  * not know how to resolve the conflict.
 631                                  * We will handle it as handle the case
 632                                  * of 'LSLF_NONE' vs 'LSLF_NONE'. */
 633                                 rc = lfsck_remove_lmv(env, com, lnr);
 634
 635                         break;
 636                 default:
 637                         break;
 638                 }
 639
 640                 break;
 641         case LSLF_NO_LMVEA:
 642                 /* The existing one has no slave LMV EA. */
 643                 switch (flags) {
 644                 case LSLF_NONE:
 645
 646 none:
 647                         if (llmv->ll_lmv.lmv_hash_type &
 648                             LMV_HASH_FLAG_LOST_LMV) {
 649                                 /* If the master LMV EA was re-generated
 650                                  * by the former LFSCK reparation, and
 651                                  * before such reparation, someone has
 652                                  * created the conflict object, but the
 653                                  * LFSCK did not detect such conflict,
 654                                  * then we have to remove the master
 655                                  * LMV EA and mark the master MDT-object
 656                                  * as read-only. The administrator can
 657                                  * handle the conflict with more human
 658                                  * knowledge. */
 659                                 rc = lfsck_remove_lmv(env, com, lnr);
 660                         } else {
 661                                 lrls = &lfsck->li_rec_lmv_save[*depth - 1];
 662                                 lrls->lrls_fid = lslr->lslr_fid;
 663                                 /* Otherwise, remove the existing name entry,
 664                                  * and add its FID in the LFSCK tracing file
 665                                  * for further processing. Refill the slot
 666                                  * with current slave LMV EA. */
 667                                 rc = lfsck_namespace_trace_update(env,
 668                                                 com, &lrls->lrls_fid,
 669                                                 LNTF_CHECK_PARENT, true);
 670                                 if (rc == 0)
 671                                         rc = lfsck_replace_lmv(env, com, lslr,
 672                                                         lnr, lmv, index, flags);
 673                         }
 674
 675                         break;
 676                 case LSLF_BAD_INDEX2:
 677                         if (flags2 >= lslr->lslr_flags)
 678                                 GOTO(out, rc = -EEXIST);
 679
 680                         goto none;
 681                 case LSLF_NO_LMVEA:
 682                         goto no_lmvea;
 683                 case LSLF_DANGLING:
 684                         /* Remove the current dangling name entry. */
 685                         rc = lfsck_remove_dirent(env, com, dir, fid, index);
 686                         break;
 687                 case LSLF_BAD_INDEX1:
 688                         index = lmv->lmv_master_mdt_index;
 689                         lmv->lmv_master_mdt_index = shard_idx;
 690                         /* The name entry claims an index that is conflict
 691                          * with a valid existing name entry, then try the
 692                          * index in the lmv recursively. */
 693                         rc = lfsck_record_lmv(env, com, lnr, lmv, index,
 694                                 LSLF_BAD_INDEX2, lslr->lslr_flags, depth);
 695                         lmv->lmv_master_mdt_index = index;
 696                         if (rc == -ERANGE || rc == -EEXIST) {
 697                                 index = shard_idx;
 698                                 goto no_lmvea;
 699                         }
 700
 701                         break;
 702                 default:
 703                         break;
 704                 }
 705
 706                 break;
 707         case LSLF_DANGLING:
 708                 /* The existing one is a dangling name entry. */
 709                 switch (flags) {
 710                 case LSLF_NONE:
 711                 case LSLF_BAD_INDEX2:
 712                 case LSLF_NO_LMVEA:
 713                         /* Remove the existing dangling name entry.
 714                          * Refill the lslr slot with the given LMV. */
 715                         rc = lfsck_replace_lmv(env, com, lslr, lnr,
 716                                                lmv, index, flags);
 717                         break;
 718                 case LSLF_DANGLING:
 719                         /* Two dangling name entries conflict,
 720                          * remove the current one. */
 721                         rc = lfsck_remove_dirent(env, com, dir, fid, index);
 722                         break;
 723                 case LSLF_BAD_INDEX1:
 724                         index = lmv->lmv_master_mdt_index;
 725                         lmv->lmv_master_mdt_index = shard_idx;
 726                         /* The name entry claims an index that is conflict
 727                          * with a valid existing name entry, then try the
 728                          * index in the lmv recursively. */
 729                         rc = lfsck_record_lmv(env, com, lnr, lmv, index,
 730                                 LSLF_BAD_INDEX2, lslr->lslr_flags, depth);
 731                         lmv->lmv_master_mdt_index = index;
 732                         if (rc == -ERANGE || rc == -EEXIST)
 733                                 /* If the index in the lmv is invalid or
 734                                  * also conflict with other, then remove
 735                                  * the existing dangling name entry.
 736                                  * Refill the lslr slot with the given LMV. */
 737                                 rc = lfsck_replace_lmv(env, com, lslr, lnr,
 738                                                        lmv, shard_idx, flags);
 739
 740                         break;
 741                 default:
 742                         break;
 743                 }
 744
 745                 break;
 746         case LSLF_BAD_INDEX1: {
 747                 if (*depth >= LFSCK_REC_LMV_MAX_DEPTH)
 748                         goto conflict;
 749
 750                 lrls = &lfsck->li_rec_lmv_save[*depth - 1];
 751                 lrls->lrls_fid = lnr->lnr_fid;
 752                 lrls->lrls_lmv = *lmv;
 753
 754                 lnr->lnr_fid = lslr->lslr_fid;
 755                 lmv->lmv_master_mdt_index = index;
 756                 lmv->lmv_stripe_count = lslr->lslr_stripe_count;
 757                 lmv->lmv_hash_type = lslr->lslr_hash_type;
 758                 index = lslr->lslr_index;
 759
 760                 /* The existing one has another possible slot,
 761                  * try it recursively. */
 762                 rc = lfsck_record_lmv(env, com, lnr, lmv, index,
 763                                       LSLF_BAD_INDEX2, flags, depth);
 764                 *lmv = lrls->lrls_lmv;
 765                 lnr->lnr_fid = lrls->lrls_fid;
 766                 index = shard_idx;
 767                 if (rc != 0) {
 768                         if (rc == -ERANGE || rc == -EEXIST)
 769                                 goto conflict;
 770
 771                         break;
 772                 }
 773
 774                 lslr->lslr_fid = *fid;
 775                 lslr->lslr_flags = flags;
 776                 if (lmv != NULL) {
 777                         lslr->lslr_stripe_count = lmv->lmv_stripe_count;
 778                         lslr->lslr_index = lmv->lmv_master_mdt_index;
 779                         lslr->lslr_hash_type = lmv->lmv_hash_type;
 780
 781                         if (flags == LSLF_NONE &&
 782                             llmv->ll_hash_type == LMV_HASH_TYPE_UNKNOWN &&
 783                             lmv_is_known_hash_type(lmv->lmv_hash_type))
 784                                 llmv->ll_hash_type = lmv->lmv_hash_type;
 785
 786                         if (flags == LSLF_NONE &&
 787                             lslr->lslr_stripe_count <= LFSCK_LMV_MAX_STRIPES &&
 788                             llmv->ll_max_stripe_count < lslr->lslr_stripe_count)
 789                                 llmv->ll_max_stripe_count =
 790                                                         lslr->lslr_stripe_count;
 791                 } else {
 792                         lslr->lslr_stripe_count = 0;
 793                         lslr->lslr_index = 0;
 794                         lslr->lslr_hash_type = 0;
 795                 }
 796
 797                 break;
 798
 799 conflict:
 800                 switch (flags) {
 801                 case LSLF_NONE:
 802                         /* The two 'valid' name entries claims the same
 803                          * index, the LFSCK cannot distinguish which one
 804                          * is correct. Then remove the master LMV EA to
 805                          * make all shards to be visible to client, and
 806                          * mark the master MDT-object as read-only. The
 807                          * administrator can handle the conflict with
 808                          * more human knowledge. */
 809                         rc = lfsck_remove_lmv(env, com, lnr);
 810                         break;
 811                 case LSLF_BAD_INDEX2:
 812                         GOTO(out, rc = -EEXIST);
 813                 case LSLF_NO_LMVEA:
 814                         goto no_lmvea;
 815                 case LSLF_DANGLING:
 816                         /* Remove the current dangling name entry. */
 817                         rc = lfsck_remove_dirent(env, com, dir, fid, index);
 818                         break;
 819                 case LSLF_BAD_INDEX1:
 820                         index = lmv->lmv_master_mdt_index;
 821                         lmv->lmv_master_mdt_index = shard_idx;
 822                         /* The name entry claims an index that is conflict
 823                          * with a valid existing name entry, then try the
 824                          * index in the lmv recursively. */
 825                         rc = lfsck_record_lmv(env, com, lnr, lmv, index,
 826                                 LSLF_BAD_INDEX2, lslr->lslr_flags, depth);
 827                         lmv->lmv_master_mdt_index = index;
 828                         if (rc == -ERANGE || rc == -EEXIST)
 829                                 /* The index in the lmv is invalid or
 830                                  * also conflict with other. Then we do
 831                                  * not know how to resolve the conflict.
 832                                  * We will handle it as handle the case
 833                                  * of 'LSLF_NONE' vs 'LSLF_NONE'. */
 834                                 rc = lfsck_remove_lmv(env, com, lnr);
 835
 836                         break;
 837                 }
 838
 839                 break;
 840         }
 841         default:
 842                 break;
 843         }
 844
 845         if (rc < 0)
 846                 llmv->ll_failed = 1;
 847
 848         GOTO(out, rc);
 849
 850 out:
 851         (*depth)--;
 852
 853         return rc > 0 ? 0 : rc;
 854 }
 855
 856 int lfsck_read_stripe_lmv(const struct lu_env *env, struct dt_object *obj,
 857                           struct lmv_mds_md_v1 *lmv)
 858 {
 859         struct dt_object *bottom;
 860         int               rc;
 861
 862         /* Currently, we only store the LMV header on disk. It is the LOD's
 863          * duty to iterate the master MDT-object's directory to compose the
 864          * integrated LMV EA. But here, we only want to load the LMV header,
 865          * so we need to bypass LOD to avoid unnecessary iteration in LOD. */
 866         bottom = lu2dt(container_of0(obj->do_lu.lo_header->loh_layers.prev,
 867                                      struct lu_object, lo_linkage));
 868         if (unlikely(bottom == NULL))
 869                 return -ENOENT;
 870
 871         dt_read_lock(env, bottom, 0);
 872         rc = dt_xattr_get(env, bottom, lfsck_buf_get(env, lmv, sizeof(*lmv)),
 873                           XATTR_NAME_LMV, BYPASS_CAPA);
 874         dt_read_unlock(env, bottom);
 875         if (rc != sizeof(*lmv))
 876                 return rc > 0 ? -EINVAL : rc;
 877
 878         lfsck_lmv_header_le_to_cpu(lmv, lmv);
 879         if ((lmv->lmv_magic == LMV_MAGIC &&
 880              !(lmv->lmv_hash_type & LMV_HASH_FLAG_MIGRATION)) ||
 881             (lmv->lmv_magic == LMV_MAGIC_STRIPE &&
 882              !(lmv->lmv_hash_type & LMV_HASH_FLAG_DEAD)))
 883                 return 0;
 884
 885         return -ENODATA;
 886 }
 887
 888 /**
 889  * Parse the shard's index from the given shard name.
 890  *
 891  * The valid shard name/type should be:
 892  * 1) The type must be S_IFDIR
 893  * 2) The name should be $FID:$index
 894  * 3) the index should within valid range.
 895  *
 896  * \param[in] env       pointer to the thread context
 897  * \param[in] name      the shard name
 898  * \param[in] namelen   the name length
 899  * \param[in] type      the entry's type
 900  * \param[in] fid       the entry's FID
 901  *
 902  * \retval              zero or positive number for the index from the name
 903  * \retval              negative error number on failure
 904  */
 905 int lfsck_shard_name_to_index(const struct lu_env *env, const char *name,
 906                               int namelen, __u16 type, const struct lu_fid *fid)
 907 {
 908         char    *name2  = lfsck_env_info(env)->lti_tmpbuf2;
 909         int      len;
 910         int      idx    = 0;
 911
 912         if (!S_ISDIR(type))
 913                 return -ENOTDIR;
 914
 915         LASSERT(name != name2);
 916
 917         len = snprintf(name2, sizeof(lfsck_env_info(env)->lti_tmpbuf2),
 918                        DFID":", PFID(fid));
 919         if (namelen < len + 1 || memcmp(name, name2, len) != 0)
 920                 return -EINVAL;
 921
 922         do {
 923                 if (!isdigit(name[len]))
 924                         return -EINVAL;
 925
 926                 idx = idx * 10 + name[len++] - '0';
 927         } while (len < namelen);
 928
 929         if (idx >= LFSCK_LMV_MAX_STRIPES)
 930                 return -EINVAL;
 931
 932         return idx;
 933 }
 934
 935 bool lfsck_is_valid_slave_name_entry(const struct lu_env *env,
 936                                      struct lfsck_lmv *llmv,
 937                                      const char *name, int namelen)
 938 {
 939         struct lmv_mds_md_v1    *lmv;
 940         int                      idx;
 941
 942         if (llmv == NULL || !llmv->ll_lmv_slave || !llmv->ll_lmv_verified)
 943                 return true;
 944
 945         lmv = &llmv->ll_lmv;
 946         idx = lmv_name_to_stripe_index(lmv->lmv_hash_type,
 947                                        lmv->lmv_stripe_count,
 948                                        name, namelen);
 949         if (unlikely(idx != lmv->lmv_master_mdt_index))
 950                 return false;
 951
 952         return true;
 953 }
 954
 955 /**
 956  * Check whether the given name is a valid entry under the @parent.
 957  *
 958  * If the @parent is a striped directory then the @child should one
 959  * shard of the striped directory, its name should be $FID:$index.
 960  *
 961  * If the @parent is a shard of a striped directory, then the name hash
 962  * should match the MDT, otherwise it is invalid.
 963  *
 964  * \param[in] env       pointer to the thread context
 965  * \param[in] parent    the parent directory
 966  * \param[in] child     the child object to be checked
 967  * \param[in] cname     the name for the @child in the parent directory
 968  *
 969  * \retval              positive number for invalid name entry
 970  * \retval              0 if the name is valid or uncertain
 971  * \retval              negative error number on failure
 972  */
 973 int lfsck_namespace_check_name(const struct lu_env *env,
 974                                struct dt_object *parent,
 975                                struct dt_object *child,
 976                                const struct lu_name *cname)
 977 {
 978         struct lmv_mds_md_v1    *lmv = &lfsck_env_info(env)->lti_lmv;
 979         int                      idx;
 980         int                      rc;
 981
 982         rc = lfsck_read_stripe_lmv(env, parent, lmv);
 983         if (rc != 0)
 984                 RETURN(rc == -ENODATA ? 0 : rc);
 985
 986         if (lmv->lmv_magic == LMV_MAGIC_STRIPE) {
 987                 if (!lfsck_is_valid_slave_lmv(lmv))
 988                         return 0;
 989
 990                 idx = lmv_name_to_stripe_index(lmv->lmv_hash_type,
 991                                                lmv->lmv_stripe_count,
 992                                                cname->ln_name,
 993                                                cname->ln_namelen);
 994                 if (unlikely(idx != lmv->lmv_master_mdt_index))
 995                         return 1;
 996         } else if (lfsck_shard_name_to_index(env, cname->ln_name,
 997                         cname->ln_namelen, lfsck_object_type(child),
 998                         lfsck_dto2fid(child)) < 0) {
 999                 return 1;
1000         }
1001
1002         return 0;
1003 }
1004
1005 /**
1006  * Update the object's LMV EA with the given @lmv.
1007  *
1008  * \param[in] env       pointer to the thread context
1009  * \param[in] com       pointer to the lfsck component
1010  * \param[in] obj       pointer to the object which LMV EA will be updated
1011  * \param[in] lmv       pointer to buffer holding the new LMV EA
1012  * \param[in] locked    whether the caller has held ldlm lock on the @obj or not
1013  *
1014  * \retval              positive number for nothing to be done
1015  * \retval              zero if updated successfully
1016  * \retval              negative error number on failure
1017  */
1018 int lfsck_namespace_update_lmv(const struct lu_env *env,
1019                                struct lfsck_component *com,
1020                                struct dt_object *obj,
1021                                struct lmv_mds_md_v1 *lmv, bool locked)
1022 {
1023         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1024         struct lmv_mds_md_v1            *lmv4   = &info->lti_lmv4;
1025         struct lu_buf                   *buf    = &info->lti_buf;
1026         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1027         struct dt_device                *dev    = lfsck_obj2dt_dev(obj);
1028         struct thandle                  *th     = NULL;
1029         struct lustre_handle             lh     = { 0 };
1030         int                              rc     = 0;
1031         int                              rc1    = 0;
1032         ENTRY;
1033
1034         LASSERT(lmv4 != lmv);
1035
1036         lfsck_lmv_header_cpu_to_le(lmv4, lmv);
1037         lfsck_buf_init(buf, lmv4, sizeof(*lmv4));
1038
1039         if (!locked) {
1040                 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1041                                       MDS_INODELOCK_UPDATE |
1042                                       MDS_INODELOCK_XATTR, LCK_EX);
1043                 if (rc != 0)
1044                         GOTO(log, rc);
1045         }
1046
1047         th = dt_trans_create(env, dev);
1048         if (IS_ERR(th))
1049                 GOTO(log, rc = PTR_ERR(th));
1050
1051         /* For remote updating LMV EA, there will be further LFSCK action on
1052          * remote MDT after the updating, so update the LMV EA synchronously. */
1053         if (dt_object_remote(obj))
1054                 th->th_sync = 1;
1055
1056         rc = dt_declare_xattr_set(env, obj, buf, XATTR_NAME_LMV, 0, th);
1057         if (rc != 0)
1058                 GOTO(stop, rc);
1059
1060         rc = dt_trans_start_local(env, dev, th);
1061         if (rc != 0)
1062                 GOTO(stop, rc);
1063
1064         dt_write_lock(env, obj, 0);
1065         if (unlikely(lfsck_is_dead_obj(obj)))
1066                 GOTO(unlock, rc = 1);
1067
1068         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1069                 GOTO(unlock, rc = 0);
1070
1071         rc = dt_xattr_set(env, obj, buf, XATTR_NAME_LMV, 0, th, BYPASS_CAPA);
1072
1073         GOTO(unlock, rc);
1074
1075 unlock:
1076         dt_write_unlock(env, obj);
1077
1078 stop:
1079         rc1 = dt_trans_stop(env, dev, th);
1080         if (rc == 0)
1081                 rc = rc1;
1082
1083 log:
1084         lfsck_ibits_unlock(&lh, LCK_EX);
1085         CDEBUG(D_LFSCK, "%s: namespace LFSCK updated the %s LMV EA "
1086                "for the object "DFID": rc = %d\n",
1087                lfsck_lfsck2name(lfsck),
1088                lmv->lmv_magic == LMV_MAGIC ? "master" : "slave",
1089                PFID(lfsck_dto2fid(obj)), rc);
1090
1091         return rc;
1092 }
1093
1094 /**
1095  * Check whether allow to re-genereate the lost master LMV EA.
1096  *
1097  * If the master MDT-object of the striped directory lost its master LMV EA,
1098  * then before the LFSCK repaired the striped directory, some ones may have
1099  * created some objects (that are not normal shards of the striped directory)
1100  * under the master MDT-object. If such case happend, then the LFSCK cannot
1101  * re-generate the lost master LMV EA to keep those objects to be visible to
1102  * client.
1103  *
1104  * \param[in] env       pointer to the thread context
1105  * \param[in] com       pointer to the lfsck component
1106  * \param[in] obj       pointer to the master MDT-object to be checked
1107  * \param[in] cfid      the shard's FID used for verification
1108  * \param[in] cidx      the shard's index used for verification
1109  *
1110  * \retval              positive number if not allow to re-generate LMV EA
1111  * \retval              zero if allow to re-generate LMV EA
1112  * \retval              negative error number on failure
1113  */
1114 static int lfsck_allow_regenerate_master_lmv(const struct lu_env *env,
1115                                              struct lfsck_component *com,
1116                                              struct dt_object *obj,
1117                                              const struct lu_fid *cfid,
1118                                              __u32 cidx)
1119 {
1120         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1121         struct lu_fid                   *tfid   = &info->lti_fid3;
1122         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1123         struct lu_dirent                *ent    =
1124                         (struct lu_dirent *)info->lti_key;
1125         const struct dt_it_ops          *iops;
1126         struct dt_it                    *di;
1127         __u64                            cookie;
1128         __u32                            args;
1129         int                              rc;
1130         __u16                            type;
1131         ENTRY;
1132
1133         if (unlikely(!dt_try_as_dir(env, obj)))
1134                 RETURN(-ENOTDIR);
1135
1136         /* Check whether the shard and the master MDT-object matches or not. */
1137         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf), DFID":%u",
1138                  PFID(cfid), cidx);
1139         rc = dt_lookup(env, obj, (struct dt_rec *)tfid,
1140                        (const struct dt_key *)info->lti_tmpbuf, BYPASS_CAPA);
1141         if (rc != 0)
1142                 RETURN(rc);
1143
1144         if (!lu_fid_eq(tfid, cfid))
1145                 RETURN(-ENOENT);
1146
1147         args = lfsck->li_args_dir & ~(LUDA_VERIFY | LUDA_VERIFY_DRYRUN);
1148         iops = &obj->do_index_ops->dio_it;
1149         di = iops->init(env, obj, args, BYPASS_CAPA);
1150         if (IS_ERR(di))
1151                 RETURN(PTR_ERR(di));
1152
1153         rc = iops->load(env, di, 0);
1154         if (rc == 0)
1155                 rc = iops->next(env, di);
1156         else if (rc > 0)
1157                 rc = 0;
1158
1159         if (rc != 0)
1160                 GOTO(out, rc);
1161
1162         do {
1163                 rc = iops->rec(env, di, (struct dt_rec *)ent, args);
1164                 if (rc == 0)
1165                         rc = lfsck_unpack_ent(ent, &cookie, &type);
1166
1167                 if (rc != 0)
1168                         GOTO(out, rc);
1169
1170                 /* skip dot and dotdot entries */
1171                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
1172                         goto next;
1173
1174                 /* If the subdir name does not match the shard name rule, then
1175                  * it is quite possible that it is NOT a shard, but created by
1176                  * someone after the master MDT-object lost the master LMV EA.
1177                  * But it is also possible that the subdir name entry crashed,
1178                  * under such double failure cases, the LFSCK cannot know how
1179                  * to repair the inconsistency. For data safe, the LFSCK will
1180                  * mark the master MDT-object as read-only. The administrator
1181                  * can fix the bad shard name manually, then run LFSCK again.
1182                  *
1183                  * XXX: If the subdir name matches the shard name rule, but it
1184                  *      is not a real shard of the striped directory, instead,
1185                  *      it was created by someone after the master MDT-object
1186                  *      lost the LMV EA, then re-generating the master LMV EA
1187                  *      will cause such subdir to be invisible to client, and
1188                  *      if its index occupies some lost shard index, then the
1189                  *      LFSCK will use it to replace the bad shard, and cause
1190                  *      the subdir (itself) to be invisible for ever. */
1191                 if (lfsck_shard_name_to_index(env, ent->lde_name,
1192                                 ent->lde_namelen, type, &ent->lde_fid) < 0)
1193                         GOTO(out, rc = 1);
1194
1195 next:
1196                 rc = iops->next(env, di);
1197         } while (rc == 0);
1198
1199         GOTO(out, rc = 0);
1200
1201 out:
1202         iops->put(env, di);
1203         iops->fini(env, di);
1204
1205         return rc;
1206 }
1207
1208 /**
1209  * Notify remote LFSCK instance that the object's LMV EA has been updated.
1210  *
1211  * \param[in] env       pointer to the thread context
1212  * \param[in] com       pointer to the lfsck component
1213  * \param[in] obj       pointer to the object on which the LMV EA will be set
1214  * \param[in] event     indicate either master or slave LMV EA has been updated
1215  * \param[in] flags     indicate which element(s) in the LMV EA has been updated
1216  * \param[in] index     the MDT index on which the LFSCK instance to be notified
1217  *
1218  * \retval              positive number if nothing to be done
1219  * \retval              zero for succeed
1220  * \retval              negative error number on failure
1221  */
1222 static int lfsck_namespace_notify_lmv_remote(const struct lu_env *env,
1223                                              struct lfsck_component *com,
1224                                              struct dt_object *obj,
1225                                              __u32 event, __u32 flags,
1226                                              __u32 index)
1227 {
1228         struct lfsck_request            *lr     = &lfsck_env_info(env)->lti_lr;
1229         const struct lu_fid             *fid    = lfsck_dto2fid(obj);
1230         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1231         struct lfsck_tgt_desc           *ltd    = NULL;
1232         struct ptlrpc_request           *req    = NULL;
1233         int                              rc;
1234         ENTRY;
1235
1236         ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, index);
1237         if (ltd == NULL)
1238                 GOTO(out, rc = -ENODEV);
1239
1240         req = ptlrpc_request_alloc(class_exp2cliimp(ltd->ltd_exp),
1241                                    &RQF_LFSCK_NOTIFY);
1242         if (req == NULL)
1243                 GOTO(out, rc = -ENOMEM);
1244
1245         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
1246         if (rc != 0) {
1247                 ptlrpc_request_free(req);
1248
1249                 GOTO(out, rc);
1250         }
1251
1252         lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
1253         memset(lr, 0, sizeof(*lr));
1254         lr->lr_event = event;
1255         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1256         lr->lr_active = LFSCK_TYPE_NAMESPACE;
1257         lr->lr_fid = *fid;
1258         lr->lr_flags = flags;
1259
1260         ptlrpc_request_set_replen(req);
1261         rc = ptlrpc_queue_wait(req);
1262         ptlrpc_req_finished(req);
1263
1264         GOTO(out, rc = (rc == -ENOENT ? 1 : rc));
1265
1266 out:
1267         CDEBUG(D_LFSCK, "%s: namespace LFSCK notify LMV EA updated for the "
1268                "object "DFID" on MDT %x remotely with event %u, flags %u: "
1269                "rc = %d\n", lfsck_lfsck2name(lfsck), PFID(fid), index,
1270                event, flags, rc);
1271
1272         if (ltd != NULL)
1273                 lfsck_tgt_put(ltd);
1274
1275         return rc;
1276 }
1277
1278 /**
1279  * Generate request for local LFSCK instance to rescan the striped directory.
1280  *
1281  * \param[in] env       pointer to the thread context
1282  * \param[in] com       pointer to the lfsck component
1283  * \param[in] obj       pointer to the striped directory to be rescanned
1284  *
1285  * \retval              positive number if nothing to be done
1286  * \retval              zero for succeed
1287  * \retval              negative error number on failure
1288  */
1289 int lfsck_namespace_notify_lmv_master_local(const struct lu_env *env,
1290                                             struct lfsck_component *com,
1291                                             struct dt_object *obj)
1292 {
1293         struct lfsck_instance      *lfsck = com->lc_lfsck;
1294         struct lfsck_namespace     *ns    = com->lc_file_ram;
1295         struct lmv_mds_md_v1       *lmv4  = &lfsck_env_info(env)->lti_lmv4;
1296         struct lfsck_lmv_unit      *llu;
1297         struct lfsck_lmv           *llmv;
1298         struct lfsck_slave_lmv_rec *lslr;
1299         int                         count = 0;
1300         int                         rc;
1301         ENTRY;
1302
1303         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1304                 RETURN(0);
1305
1306         rc = lfsck_read_stripe_lmv(env, obj, lmv4);
1307         if (rc != 0)
1308                 RETURN(rc);
1309
1310         OBD_ALLOC_PTR(llu);
1311         if (unlikely(llu == NULL))
1312                 RETURN(-ENOMEM);
1313
1314         if (lmv4->lmv_stripe_count < 1)
1315                 count = LFSCK_LMV_DEF_STRIPES;
1316         else if (lmv4->lmv_stripe_count > LFSCK_LMV_MAX_STRIPES)
1317                 count = LFSCK_LMV_MAX_STRIPES;
1318         else
1319                 count = lmv4->lmv_stripe_count;
1320
1321         OBD_ALLOC_LARGE(lslr, sizeof(struct lfsck_slave_lmv_rec) * count);
1322         if (lslr == NULL) {
1323                 OBD_FREE_PTR(llu);
1324
1325                 RETURN(-ENOMEM);
1326         }
1327
1328         INIT_LIST_HEAD(&llu->llu_link);
1329         llu->llu_lfsck = lfsck;
1330         llu->llu_obj = lfsck_object_get(obj);
1331         llmv = &llu->llu_lmv;
1332         llmv->ll_lmv_master = 1;
1333         llmv->ll_inline = 1;
1334         atomic_set(&llmv->ll_ref, 1);
1335         llmv->ll_stripes_allocated = count;
1336         llmv->ll_hash_type = LMV_HASH_TYPE_UNKNOWN;
1337         llmv->ll_lslr = lslr;
1338         llmv->ll_lmv = *lmv4;
1339
1340         down_write(&com->lc_sem);
1341         if (ns->ln_status != LS_SCANNING_PHASE1 &&
1342             ns->ln_status != LS_SCANNING_PHASE2) {
1343                 ns->ln_striped_dirs_skipped++;
1344                 up_write(&com->lc_sem);
1345                 lfsck_lmv_put(env, llmv);
1346         } else {
1347                 ns->ln_striped_dirs_repaired++;
1348                 spin_lock(&lfsck->li_lock);
1349                 list_add_tail(&llu->llu_link, &lfsck->li_list_lmv);
1350                 spin_unlock(&lfsck->li_lock);
1351                 up_write(&com->lc_sem);
1352         }
1353
1354         RETURN(0);
1355 }
1356
1357 /**
1358  * Set master LMV EA for the specified striped directory.
1359  *
1360  * First, if the master MDT-object of a striped directory lost its LMV EA,
1361  * then there may be some users have created some files under the master
1362  * MDT-object directly. Under such case, the LFSCK cannot re-generate LMV
1363  * EA for the master MDT-object, because we should keep the existing files
1364  * to be visible to client. Then the LFSCK will mark the striped directory
1365  * as read-only and keep it there to be handled by administrator manually.
1366  *
1367  * If nobody has created files under the master MDT-object of the striped
1368  * directory, then we will set the master LMV EA and generate a new rescan
1369  * (the striped directory) request that will be handled later by the LFSCK
1370  * instance on the MDT later.
1371  *
1372  * \param[in] env       pointer to the thread context
1373  * \param[in] com       pointer to the lfsck component
1374  * \param[in] dir       pointer to the object on which the LMV EA will be set
1375  * \param[in] lmv       pointer to the buffer holding the new LMV EA
1376  * \param[in] cfid      the shard's FID used for verification
1377  * \param[in] cidx      the shard's index used for verification
1378  * \param[in] flags     to indicate which element(s) in the LMV EA will be set
1379  *
1380  * \retval              positive number if nothing to be done
1381  * \retval              zero for succeed
1382  * \retval              negative error number on failure
1383  */
1384 static int lfsck_namespace_set_lmv_master(const struct lu_env *env,
1385                                           struct lfsck_component *com,
1386                                           struct dt_object *dir,
1387                                           struct lmv_mds_md_v1 *lmv,
1388                                           const struct lu_fid *cfid,
1389                                           __u32 cidx, __u32 flags)
1390 {
1391         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1392         struct lmv_mds_md_v1            *lmv3   = &info->lti_lmv3;
1393         struct lu_seq_range             *range  = &info->lti_range;
1394         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1395         struct seq_server_site          *ss     =
1396                         lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
1397         struct dt_object                *obj;
1398         struct lustre_handle             lh     = { 0 };
1399         int                              pidx   = -1;
1400         int                              rc     = 0;
1401         ENTRY;
1402
1403         /* Find the bottom object to bypass LOD when set LMV EA. */
1404         obj = lu2dt(container_of0(dir->do_lu.lo_header->loh_layers.prev,
1405                                   struct lu_object, lo_linkage));
1406         if (unlikely(obj == NULL))
1407                 RETURN(-ENOENT);
1408
1409         fld_range_set_mdt(range);
1410         rc = fld_server_lookup(env, ss->ss_server_fld,
1411                                fid_seq(lfsck_dto2fid(obj)), range);
1412         if (rc != 0)
1413                 GOTO(log, rc);
1414
1415         pidx = range->lsr_index;
1416         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1417                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1418                               LCK_EX);
1419         if (rc != 0)
1420                 GOTO(log, rc);
1421
1422         rc = lfsck_read_stripe_lmv(env, obj, lmv3);
1423         if (rc == -ENODATA) {
1424                 if (!(flags & LEF_SET_LMV_ALL))
1425                         GOTO(log, rc);
1426
1427                 *lmv3 = *lmv;
1428         } else if (rc == 0) {
1429                 if (flags & LEF_SET_LMV_ALL)
1430                         GOTO(log, rc = 1);
1431
1432                 if (flags & LEF_SET_LMV_HASH)
1433                         lmv3->lmv_hash_type = lmv->lmv_hash_type;
1434         } else {
1435                 GOTO(log, rc);
1436         }
1437
1438         lmv3->lmv_magic = LMV_MAGIC;
1439         lmv3->lmv_master_mdt_index = pidx;
1440
1441         if (flags & LEF_SET_LMV_ALL) {
1442                 rc = lfsck_allow_regenerate_master_lmv(env, com, obj,
1443                                                        cfid, cidx);
1444                 if (rc > 0) {
1445                         rc = lfsck_disable_master_lmv(env, com, obj, false);
1446
1447                         GOTO(log, rc = (rc == 0 ? 1 : rc));
1448                 }
1449
1450                 if (rc < 0)
1451                         GOTO(log, rc);
1452
1453                 /* To indicate that the master has ever lost LMV EA. */
1454                 lmv3->lmv_hash_type |= LMV_HASH_FLAG_LOST_LMV;
1455         }
1456
1457         rc = lfsck_namespace_update_lmv(env, com, obj, lmv3, true);
1458         if (rc == 0 && flags & LEF_SET_LMV_ALL) {
1459                 if (dt_object_remote(obj))
1460                         rc = lfsck_namespace_notify_lmv_remote(env, com, obj,
1461                                                 LE_SET_LMV_MASTER, 0, pidx);
1462                 else
1463                         rc = lfsck_namespace_notify_lmv_master_local(env, com,
1464                                                                      obj);
1465         }
1466
1467         GOTO(log, rc);
1468
1469 log:
1470         lfsck_ibits_unlock(&lh, LCK_EX);
1471         CDEBUG(D_LFSCK, "%s: namespace LFSCK set master LMV EA for the object "
1472                DFID" on the %s MDT %d, flags %x: rc = %d\n",
1473                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
1474                dt_object_remote(obj) ? "remote" : "local", pidx, flags, rc);
1475
1476         if (rc <= 0) {
1477                 struct lfsck_namespace *ns = com->lc_file_ram;
1478
1479                 ns->ln_flags |= LF_INCONSISTENT;
1480         }
1481
1482         return rc;
1483 }
1484
1485 /**
1486  * Repair the bad name hash.
1487  *
1488  * If the name hash of some name entry under the striped directory does not
1489  * match the shard of the striped directory, then the LFSCK will repair the
1490  * inconsistency. Ideally, the LFSCK should migrate the name entry from the
1491  * current MDT to the right MDT (another one), but before the async commit
1492  * finished, the LFSCK will change the striped directory's hash type as
1493  * LMV_HASH_TYPE_UNKNOWN and mark the lmv flags as LMV_HASH_FLAG_BAD_TYPE.
1494  *
1495  * \param[in] env       pointer to the thread context
1496  * \param[in] com       pointer to the lfsck component
1497  * \param[in] shard     pointer to the shard of the striped directory that
1498  *                      contains the bad name entry
1499  * \param[in] llmv      pointer to lfsck LMV EA structure
1500  * \param[in] name      the name of the bad name hash
1501  *
1502  * \retval              positive number if nothing to be done
1503  * \retval              zero for succeed
1504  * \retval              negative error number on failure
1505  */
1506 int lfsck_namespace_repair_bad_name_hash(const struct lu_env *env,
1507                                          struct lfsck_component *com,
1508                                          struct dt_object *shard,
1509                                          struct lfsck_lmv *llmv,
1510                                          const char *name)
1511 {
1512         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1513         struct lu_fid                   *pfid   = &info->lti_fid3;
1514         struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
1515         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1516         struct dt_object                *parent = NULL;
1517         int                              rc     = 0;
1518         ENTRY;
1519
1520         rc = dt_lookup(env, shard, (struct dt_rec *)pfid,
1521                        (const struct dt_key *)dotdot, BYPASS_CAPA);
1522         if (rc != 0 || !fid_is_sane(pfid))
1523                 GOTO(log, rc);
1524
1525         parent = lfsck_object_find_bottom(env, lfsck, pfid);
1526         if (IS_ERR(parent))
1527                 GOTO(log, rc = PTR_ERR(parent));
1528
1529         *lmv2 = llmv->ll_lmv;
1530         lmv2->lmv_hash_type = LMV_HASH_TYPE_UNKNOWN | LMV_HASH_FLAG_BAD_TYPE;
1531         rc = lfsck_namespace_set_lmv_master(env, com, parent, lmv2,
1532                                             lfsck_dto2fid(shard),
1533                                             llmv->ll_lmv.lmv_master_mdt_index,
1534                                             LEF_SET_LMV_HASH);
1535
1536         GOTO(log, rc);
1537
1538 log:
1539         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name hash "
1540                "on the MDT %x, parent "DFID", name %s, shard_%x "DFID
1541                ": rc = %d\n",
1542                lfsck_lfsck2name(lfsck), lfsck_dev_idx(lfsck->li_bottom),
1543                PFID(pfid), name, llmv->ll_lmv.lmv_master_mdt_index,
1544                PFID(lfsck_dto2fid(shard)), rc);
1545
1546         if (parent != NULL && !IS_ERR(parent))
1547                 lfsck_object_put(env, parent);
1548
1549         return rc;
1550 }
1551
1552 /**
1553  * Scan the shard of a striped directory for name hash verification.
1554  *
1555  * During the first-stage scanning, if the LFSCK cannot make sure whether
1556  * the shard of a stripe directory contains valid slave LMV EA or not, then
1557  * it will skip the name hash verification for this shard temporarily, and
1558  * record the shard's FID in the LFSCK tracing file. As the LFSCK processing,
1559  * the slave LMV EA may has been verified/fixed by LFSCK instance on master.
1560  * Then in the second-stage scanning, the shard will be re-scanned, and for
1561  * every name entry under the shard, the name hash will be verified, and for
1562  * unmatched name entry, the LFSCK will try to fix it.
1563  *
1564  * \param[in] env       pointer to the thread context
1565  * \param[in] com       pointer to the lfsck component
1566  * \param[in] child     pointer to the directory object to be handled
1567  *
1568  * \retval              positive number for scanning successfully
1569  * \retval              zero for the scanning is paused
1570  * \retval              negative error number on failure
1571  */
1572 int lfsck_namespace_scan_shard(const struct lu_env *env,
1573                                struct lfsck_component *com,
1574                                struct dt_object *child)
1575 {
1576         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1577         struct lmv_mds_md_v1            *lmv    = &info->lti_lmv;
1578         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1579         struct lfsck_namespace          *ns     = com->lc_file_ram;
1580         struct ptlrpc_thread            *thread = &lfsck->li_thread;
1581         struct lu_dirent                *ent    =
1582                         (struct lu_dirent *)info->lti_key;
1583         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
1584         struct lfsck_lmv                *llmv   = NULL;
1585         const struct dt_it_ops          *iops;
1586         struct dt_it                    *di;
1587         __u64                            cookie;
1588         __u32                            args;
1589         int                              rc;
1590         __u16                            type;
1591         ENTRY;
1592
1593         rc = lfsck_read_stripe_lmv(env, child, lmv);
1594         if (rc != 0)
1595                 RETURN(rc == -ENODATA ? 1 : rc);
1596
1597         if (lmv->lmv_magic != LMV_MAGIC_STRIPE)
1598                 RETURN(1);
1599
1600         if (unlikely(!dt_try_as_dir(env, child)))
1601                 RETURN(-ENOTDIR);
1602
1603         OBD_ALLOC_PTR(llmv);
1604         if (llmv == NULL)
1605                 RETURN(-ENOMEM);
1606
1607         llmv->ll_lmv_slave = 1;
1608         llmv->ll_lmv_verified = 1;
1609         llmv->ll_lmv = *lmv;
1610         atomic_set(&llmv->ll_ref, 1);
1611
1612         args = lfsck->li_args_dir & ~(LUDA_VERIFY | LUDA_VERIFY_DRYRUN);
1613         iops = &child->do_index_ops->dio_it;
1614         di = iops->init(env, child, args, BYPASS_CAPA);
1615         if (IS_ERR(di))
1616                 GOTO(out, rc = PTR_ERR(di));
1617
1618         rc = iops->load(env, di, 0);
1619         if (rc == 0)
1620                 rc = iops->next(env, di);
1621         else if (rc > 0)
1622                 rc = 0;
1623
1624         while (rc == 0) {
1625                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
1626                     cfs_fail_val > 0) {
1627                         struct l_wait_info lwi;
1628
1629                         lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
1630                                           NULL, NULL);
1631                         l_wait_event(thread->t_ctl_waitq,
1632                                      !thread_is_running(thread),
1633                                      &lwi);
1634
1635                         if (unlikely(!thread_is_running(thread)))
1636                                 GOTO(out, rc = 0);
1637                 }
1638
1639                 rc = iops->rec(env, di, (struct dt_rec *)ent, args);
1640                 if (rc == 0)
1641                         rc = lfsck_unpack_ent(ent, &cookie, &type);
1642
1643                 if (rc != 0) {
1644                         if (bk->lb_param & LPF_FAILOUT)
1645                                 GOTO(out, rc);
1646
1647                         goto next;
1648                 }
1649
1650                 /* skip dot and dotdot entries */
1651                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
1652                         goto next;
1653
1654                 if (!lfsck_is_valid_slave_name_entry(env, llmv, ent->lde_name,
1655                                                      ent->lde_namelen)) {
1656                         ns->ln_flags |= LF_INCONSISTENT;
1657                         rc = lfsck_namespace_repair_bad_name_hash(env, com,
1658                                                 child, llmv, ent->lde_name);
1659                         if (rc >= 0)
1660                                 ns->ln_name_hash_repaired++;
1661                 }
1662
1663                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
1664                         GOTO(out, rc);
1665
1666                 /* Rate control. */
1667                 lfsck_control_speed(lfsck);
1668                 if (unlikely(!thread_is_running(thread)))
1669                         GOTO(out, rc = 0);
1670
1671                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
1672                         spin_lock(&lfsck->li_lock);
1673                         thread_set_flags(thread, SVC_STOPPING);
1674                         spin_unlock(&lfsck->li_lock);
1675
1676                         GOTO(out, rc = -EINVAL);
1677                 }
1678
1679 next:
1680                 rc = iops->next(env, di);
1681         }
1682
1683         GOTO(out, rc);
1684
1685 out:
1686         iops->put(env, di);
1687         iops->fini(env, di);
1688         lfsck_lmv_put(env, llmv);
1689
1690         return rc;
1691 }
1692
1693 /**
1694  * Verify the slave object's (of striped directory) LMV EA.
1695  *
1696  * For the slave object of a striped directory, before traversing the shard
1697  * the LFSCK will verify whether its slave LMV EA matches its parent's master
1698  * LMV EA or not.
1699  *
1700  * \param[in] env       pointer to the thread context
1701  * \param[in] com       pointer to the lfsck component
1702  * \param[in] obj       pointer to the object which LMV EA will be checked
1703  * \param[in] llmv      pointer to buffer holding the slave LMV EA
1704  *
1705  * \retval              zero for succeed
1706  * \retval              negative error number on failure
1707  */
1708 int lfsck_namespace_verify_stripe_slave(const struct lu_env *env,
1709                                         struct lfsck_component *com,
1710                                         struct dt_object *obj,
1711                                         struct lfsck_lmv *llmv)
1712 {
1713         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1714         char                            *name   = info->lti_key;
1715         char                            *name2;
1716         struct lu_fid                   *pfid   = &info->lti_fid3;
1717         struct lu_fid                   *tfid   = &info->lti_fid4;
1718         const struct lu_fid             *cfid   = lfsck_dto2fid(obj);
1719         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1720         struct lmv_mds_md_v1            *clmv   = &llmv->ll_lmv;
1721         struct lmv_mds_md_v1            *plmv   = &info->lti_lmv;
1722         struct dt_object                *parent = NULL;
1723         int                              rc     = 0;
1724         ENTRY;
1725
1726         if (!lfsck_is_valid_slave_lmv(clmv)) {
1727                 rc = lfsck_namespace_trace_update(env, com, cfid,
1728                                         LNTF_UNCERTAIN_LMV, true);
1729
1730                 GOTO(out, rc);
1731         }
1732
1733         rc = dt_lookup(env, obj, (struct dt_rec *)pfid,
1734                        (const struct dt_key *)dotdot, BYPASS_CAPA);
1735         if (rc != 0 || !fid_is_sane(pfid)) {
1736                 rc = lfsck_namespace_trace_update(env, com, cfid,
1737                                         LNTF_UNCERTAIN_LMV, true);
1738
1739                 GOTO(out, rc);
1740         }
1741
1742         parent = lfsck_object_find(env, lfsck, pfid);
1743         if (IS_ERR(parent)) {
1744                 rc = lfsck_namespace_trace_update(env, com, cfid,
1745                                         LNTF_UNCERTAIN_LMV, true);
1746
1747                 GOTO(out, rc);
1748         }
1749
1750         rc = lfsck_read_stripe_lmv(env, parent, plmv);
1751         if (rc != 0) {
1752                 int rc1;
1753
1754                 /* If the parent has no LMV EA, then it maybe because:
1755                  * 1) The parent lost the LMV EA.
1756                  * 2) The child claims a wrong (slave) LMV EA. */
1757                 if (rc == -ENODATA)
1758                         rc = lfsck_namespace_set_lmv_master(env, com, parent,
1759                                         clmv, cfid, clmv->lmv_master_mdt_index,
1760                                         LEF_SET_LMV_ALL);
1761                 else
1762                         rc = 0;
1763
1764                 rc1 = lfsck_namespace_trace_update(env, com, cfid,
1765                                                    LNTF_UNCERTAIN_LMV, true);
1766
1767                 GOTO(out, rc = (rc < 0 ? rc : rc1));
1768         }
1769
1770         /* Unmatched magic or stripe count. */
1771         if (unlikely(plmv->lmv_magic != LMV_MAGIC ||
1772                      plmv->lmv_stripe_count != clmv->lmv_stripe_count)) {
1773                 rc = lfsck_namespace_trace_update(env, com, cfid,
1774                                                   LNTF_UNCERTAIN_LMV, true);
1775
1776                 GOTO(out, rc);
1777         }
1778
1779         /* If the master hash type has been set as LMV_HASH_TYPE_UNKNOWN,
1780          * then the slave hash type is not important. */
1781         if ((plmv->lmv_hash_type & LMV_HASH_TYPE_MASK) ==
1782             LMV_HASH_TYPE_UNKNOWN &&
1783             plmv->lmv_hash_type & LMV_HASH_FLAG_BAD_TYPE)
1784                 GOTO(out, rc = 0);
1785
1786         /* Unmatched hash type. */
1787         if (unlikely((plmv->lmv_hash_type & LMV_HASH_TYPE_MASK) !=
1788                      (clmv->lmv_hash_type & LMV_HASH_TYPE_MASK))) {
1789                 rc = lfsck_namespace_trace_update(env, com, cfid,
1790                                                   LNTF_UNCERTAIN_LMV, true);
1791
1792                 GOTO(out, rc);
1793         }
1794
1795         snprintf(info->lti_tmpbuf2, sizeof(info->lti_tmpbuf2), DFID":%u",
1796                  PFID(cfid), clmv->lmv_master_mdt_index);
1797         name2 = info->lti_tmpbuf2;
1798
1799         rc = lfsck_links_get_first(env, obj, name, tfid);
1800         if (rc == 0 && strcmp(name, name2) == 0 && lu_fid_eq(pfid, tfid)) {
1801                 llmv->ll_lmv_verified = 1;
1802
1803                 GOTO(out, rc);
1804         }
1805
1806         rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
1807                        (const struct dt_key *)name2, BYPASS_CAPA);
1808         if (rc != 0 || !lu_fid_eq(cfid, tfid))
1809                 rc = lfsck_namespace_trace_update(env, com, cfid,
1810                                                   LNTF_UNCERTAIN_LMV, true);
1811         else
1812                 llmv->ll_lmv_verified = 1;
1813
1814         GOTO(out, rc);
1815
1816 out:
1817         if (parent != NULL && !IS_ERR(parent))
1818                 lfsck_object_put(env, parent);
1819
1820         return rc;
1821 }
1822
1823 /**
1824  * Double scan the striped directory or the shard.
1825  *
1826  * All the shards' under the given striped directory or its shard have
1827  * been scanned, the LFSCK has got the global knownledge about the LMV
1828  * EA consistency.
1829  *
1830  * If the target is one shard of a striped directory, then only needs to
1831  * update related tracing file.
1832  *
1833  * If the target is the master MDT-object of a striped directory, then the
1834  * LFSCK will make the decision about whether the master LMV EA is invalid
1835  * or not, and repair it if inconsistenct; for every shard of the striped
1836  * directory, whether the slave LMV EA is invalid or not, and repair it if
1837  * inconsistent.
1838  *
1839  * \param[in] env       pointer to the thread context
1840  * \param[in] com       pointer to the lfsck component
1841  * \param[in] lnr       pointer to the namespace request that contains the
1842  *                      striped directory or the shard
1843  *
1844  * \retval              zero for succeed
1845  * \retval              negative error number on failure
1846  */
1847 int lfsck_namespace_striped_dir_rescan(const struct lu_env *env,
1848                                        struct lfsck_component *com,
1849                                        struct lfsck_namespace_req *lnr)
1850 {
1851         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1852         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1853         struct lfsck_namespace          *ns     = com->lc_file_ram;
1854         struct lfsck_lmv                *llmv   = lnr->lnr_lmv;
1855         struct lmv_mds_md_v1            *lmv    = &llmv->ll_lmv;
1856         struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
1857         struct dt_object                *dir    = lnr->lnr_obj;
1858         const struct lu_fid             *pfid   = lfsck_dto2fid(dir);
1859         struct lu_seq_range             *range  = &info->lti_range;
1860         struct seq_server_site          *ss     =
1861                         lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
1862         __u32                            stripe_count;
1863         __u32                            hash_type;
1864         int                              rc     = 0;
1865         int                              i;
1866         ENTRY;
1867
1868         if (llmv->ll_lmv_slave) {
1869                 if (llmv->ll_lmv_verified) {
1870                         ns->ln_striped_shards_scanned++;
1871                         lfsck_namespace_trace_update(env, com,
1872                                         lfsck_dto2fid(dir),
1873                                         LNTF_UNCERTAIN_LMV |
1874                                         LNTF_RECHECK_NAME_HASH, false);
1875                 }
1876
1877                 RETURN(0);
1878         }
1879
1880         /* Either the striped directory has been disabled or only part of
1881          * the striped directory have been scanned. The LFSCK cannot repair
1882          * something based on incompleted knowledge. So skip it. */
1883         if (llmv->ll_ignore || llmv->ll_exit_value <= 0)
1884                 RETURN(0);
1885
1886         /* There ever been some failure, as to the LFSCK cannot know whether
1887          * it has got the global knowledge about the LMV EA consistency or not,
1888          * so it cannot make reparation about the incompleted knowledge. */
1889         if (llmv->ll_failed) {
1890                 ns->ln_striped_dirs_scanned++;
1891                 ns->ln_striped_dirs_failed++;
1892
1893                 RETURN(0);
1894         }
1895
1896         if (lmv->lmv_stripe_count > LFSCK_LMV_MAX_STRIPES)
1897                 stripe_count = max(llmv->ll_max_filled_off + 1,
1898                                    llmv->ll_max_stripe_count);
1899         else
1900                 stripe_count = max(llmv->ll_max_filled_off + 1,
1901                                    lmv->lmv_stripe_count);
1902
1903         if (lmv->lmv_stripe_count != stripe_count) {
1904                 lmv->lmv_stripe_count = stripe_count;
1905                 llmv->ll_lmv_updated = 1;
1906         }
1907
1908         if (!lmv_is_known_hash_type(lmv->lmv_hash_type) &&
1909             !(lmv->lmv_hash_type & LMV_HASH_FLAG_BAD_TYPE) &&
1910             lmv_is_known_hash_type(llmv->ll_hash_type)) {
1911                 hash_type = llmv->ll_hash_type & LMV_HASH_TYPE_MASK;
1912                 lmv->lmv_hash_type = llmv->ll_hash_type;
1913                 llmv->ll_lmv_updated = 1;
1914         } else {
1915                 hash_type = lmv->lmv_hash_type & LMV_HASH_TYPE_MASK;
1916                 if (!lmv_is_known_hash_type(hash_type))
1917                         hash_type = LMV_HASH_TYPE_UNKNOWN;
1918         }
1919
1920         if (llmv->ll_lmv_updated) {
1921                 lmv->lmv_layout_version++;
1922                 rc = lfsck_namespace_update_lmv(env, com, dir, lmv, false);
1923                 if (rc != 0)
1924                         RETURN(rc);
1925
1926                 ns->ln_striped_dirs_scanned++;
1927                 ns->ln_striped_dirs_repaired++;
1928         }
1929
1930         fld_range_set_mdt(range);
1931         for (i = 0; i <= llmv->ll_max_filled_off; i++) {
1932                 struct dt_object *obj = NULL;
1933                 struct lfsck_slave_lmv_rec *lslr = llmv->ll_lslr + i;
1934                 const struct lu_fid *cfid = &lslr->lslr_fid;
1935                 const struct lu_name *cname;
1936                 struct linkea_data ldata = { 0 };
1937                 int len;
1938                 int rc1 = 0;
1939                 bool repair_linkea = false;
1940                 bool repair_lmvea = false;
1941                 bool rename = false;
1942                 bool create = false;
1943                 bool linkea_repaired = false;
1944                 bool lmvea_repaired = false;
1945                 bool rename_repaired = false;
1946                 bool create_repaired = false;
1947
1948                 /* LMV EA hole. */
1949                 if (fid_is_zero(cfid))
1950                         continue;
1951
1952                 len = snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
1953                                DFID":%u", PFID(cfid), i);
1954                 cname = lfsck_name_get_const(env, info->lti_tmpbuf, len);
1955                 memcpy(lnr->lnr_name, info->lti_tmpbuf, len);
1956
1957                 obj = lfsck_object_find_bottom(env, lfsck, cfid);
1958                 if (IS_ERR(obj)) {
1959                         rc1 = PTR_ERR(obj);
1960                         goto next;
1961                 }
1962
1963                 switch (lslr->lslr_flags) {
1964                 case LSLF_NONE:
1965                         if (llmv->ll_inline ||
1966                             lslr->lslr_stripe_count != stripe_count ||
1967                             (lslr->lslr_hash_type & LMV_HASH_TYPE_MASK) !=
1968                              hash_type)
1969                                 repair_lmvea = true;
1970                         break;
1971                 case LSLF_BAD_INDEX2:
1972                         /* The index in the slave LMV EA is right,
1973                          * the name entry should be updated. */
1974                         rename = true;
1975                         snprintf(info->lti_tmpbuf2, sizeof(info->lti_tmpbuf2),
1976                                  DFID":%u", PFID(cfid), lslr->lslr_index);
1977                         if (llmv->ll_inline ||
1978                             lslr->lslr_stripe_count != stripe_count ||
1979                             (lslr->lslr_hash_type & LMV_HASH_TYPE_MASK) !=
1980                              hash_type)
1981                                 repair_lmvea = true;
1982                         break;
1983                 case LSLF_BAD_INDEX1:
1984                         /* The index in the name entry is right,
1985                          * the slave LMV EA should be updated. */
1986                 case LSLF_NO_LMVEA:
1987                         repair_lmvea = true;
1988                         break;
1989                 case LSLF_DANGLING:
1990                         create = true;
1991                         goto repair;
1992                 default:
1993                         break;
1994                 }
1995
1996                 rc1 = lfsck_links_read(env, obj, &ldata);
1997                 if (rc1 == -ENOENT) {
1998                         create = true;
1999                         goto repair;
2000                 }
2001
2002                 if (rc1 == -EINVAL || rc1 == -ENODATA) {
2003                         repair_linkea = true;
2004                         goto repair;
2005                 }
2006
2007                 if (rc1 != 0)
2008                         goto next;
2009
2010                 if (ldata.ld_leh->leh_reccount != 1) {
2011                         repair_linkea = true;
2012                         goto repair;
2013                 }
2014
2015                 rc1 = linkea_links_find(&ldata, cname, pfid);
2016                 if (rc1 != 0)
2017                         repair_linkea = true;
2018
2019 repair:
2020                 if (create) {
2021                         rc1 = lfsck_namespace_repair_dangling(env, com,
2022                                                               obj, lnr);
2023                         if (rc1 >= 0) {
2024                                 create_repaired = true;
2025                                 if (rc == 0)
2026                                         ns->ln_dangling_repaired++;
2027                         }
2028                 }
2029
2030                 if (repair_lmvea) {
2031                         *lmv2 = *lmv;
2032                         lmv2->lmv_magic = LMV_MAGIC_STRIPE;
2033                         lmv2->lmv_stripe_count = stripe_count;
2034                         lmv2->lmv_master_mdt_index = i;
2035                         lmv2->lmv_hash_type = hash_type;
2036
2037                         rc1 = lfsck_namespace_update_lmv(env, com, obj,
2038                                                          lmv2, false);
2039                         if (rc1 < 0)
2040                                 goto next;
2041
2042                         if (dt_object_remote(obj)) {
2043                                 rc1 = fld_server_lookup(env, ss->ss_server_fld,
2044                                         fid_seq(lfsck_dto2fid(obj)), range);
2045                                 if (rc1 != 0)
2046                                         goto next;
2047
2048                                 rc1 = lfsck_namespace_notify_lmv_remote(env,
2049                                                 com, obj, LE_SET_LMV_SLAVE, 0,
2050                                                 range->lsr_index);
2051                         } else {
2052                                 ns->ln_striped_shards_repaired++;
2053                                 rc1 = lfsck_namespace_trace_update(env, com,
2054                                         cfid, LNTF_RECHECK_NAME_HASH, true);
2055                         }
2056
2057                         if (rc1 < 0)
2058                                 goto next;
2059
2060                         if (rc1 >= 0)
2061                                 lmvea_repaired = true;
2062                 } else if (llmv->ll_inline) {
2063                         if (dt_object_remote(obj)) {
2064                                 rc1 = fld_server_lookup(env, ss->ss_server_fld,
2065                                         fid_seq(lfsck_dto2fid(obj)), range);
2066                                 if (rc1 != 0)
2067                                         goto next;
2068
2069                                 /* The slave LMV EA on the remote shard is
2070                                  * correct, just notify the LFSCK instance
2071                                  * on such MDT to re-verify the name_hash. */
2072                                 rc1 = lfsck_namespace_notify_lmv_remote(env,
2073                                                 com, obj, LE_SET_LMV_SLAVE,
2074                                                 LEF_RECHECK_NAME_HASH,
2075                                                 range->lsr_index);
2076                         } else {
2077                                 rc1 = lfsck_namespace_trace_update(env, com,
2078                                         cfid, LNTF_RECHECK_NAME_HASH, true);
2079                         }
2080
2081                         if (rc1 < 0)
2082                                 goto next;
2083                 }
2084
2085                 if (rename) {
2086                         rc1 = lfsck_namespace_repair_dirent(env, com, dir, obj,
2087                                         info->lti_tmpbuf2, lnr->lnr_name,
2088                                         lnr->lnr_type, true, false);
2089                         if (rc1 >= 0) {
2090                                 rename_repaired = true;
2091                                 if (rc1 > 0) {
2092                                         ns->ln_dirent_repaired++;
2093                                         rc1 = lfsck_namespace_trace_update(env,
2094                                                 com, cfid,
2095                                                 LNTF_RECHECK_NAME_HASH, true);
2096                                 }
2097                         }
2098
2099                         if (rc1 < 0)
2100                                 goto next;
2101                 }
2102
2103                 if (repair_linkea) {
2104                         struct lustre_handle lh = { 0 };
2105
2106                         rc1 = linkea_data_new(&ldata, &info->lti_big_buf);
2107                         if (rc1 != 0)
2108                                 goto next;
2109
2110                         rc1 = linkea_add_buf(&ldata, cname, lfsck_dto2fid(dir));
2111                         if (rc1 != 0)
2112                                 goto next;
2113
2114                         rc1 = lfsck_ibits_lock(env, lfsck, obj, &lh,
2115                                                MDS_INODELOCK_UPDATE |
2116                                                MDS_INODELOCK_XATTR, LCK_EX);
2117                         lfsck_ibits_unlock(&lh, LCK_EX);
2118                         if (rc1 != 0)
2119                                 goto next;
2120
2121                         rc1 = lfsck_namespace_rebuild_linkea(env, com, obj,
2122                                                              &ldata);
2123                         if (rc1 >= 0) {
2124                                 linkea_repaired = true;
2125                                 if (rc1 > 0)
2126                                         ns->ln_linkea_repaired++;
2127                         }
2128                 }
2129
2130 next:
2131                 CDEBUG(D_LFSCK, "%s: namespace LFSCK repair the shard "
2132                       "%d "DFID" of the striped directory "DFID" with "
2133                       "dangling %s/%s, rename %s/%s, llinkea %s/%s, "
2134                       "repair_lmvea %s/%s: rc = %d\n", lfsck_lfsck2name(lfsck),
2135                       i, PFID(cfid), PFID(&lnr->lnr_fid),
2136                       create ? "yes" : "no", create_repaired ? "yes" : "no",
2137                       rename ? "yes" : "no", rename_repaired ? "yes" : "no",
2138                       repair_linkea ? "yes" : "no",
2139                       linkea_repaired ? "yes" : "no",
2140                       repair_lmvea ? "yes" : "no",
2141                       lmvea_repaired ? "yes" : "no", rc1);
2142
2143                 if (obj != NULL && !IS_ERR(obj))
2144                         lfsck_object_put(env, obj);
2145
2146                 if (rc1 < 0) {
2147                         rc = rc1;
2148                         ns->ln_striped_shards_failed++;
2149                 }
2150         }
2151
2152         RETURN(rc);
2153 }
2154
2155 /**
2156  * Verify the shard's name entry under the striped directory.
2157  *
2158  * Before all shards of the striped directory scanned, the LFSCK cannot
2159  * know whether the master LMV EA is valid or not, and also cannot know
2160  * how to repair an invalid shard exactly. For example, the stripe index
2161  * stored in the shard's name does not match the stripe index stored in
2162  * the slave LMV EA, then the LFSCK cannot know which one is correct.
2163  * If the LFSCK just assumed one is correct, and fixed the other, then
2164  * as the LFSCK processing, it may find that the former reparation is
2165  * wrong and have to roll back. Unfortunately, if some applications saw
2166  * the changes and made further modification based on such changes, then
2167  * the roll back is almost impossible.
2168  *
2169  * To avoid above trouble, the LFSCK will scan the master object of the
2170  * striped directory twice, that is NOT the same as normal two-stages
2171  * scanning, the double scanning the striped directory will happen both
2172  * during the first-stage scanning:
2173  *
2174  * 1) When the striped directory is opened for scanning, the LFSCK will
2175  *    iterate each shard in turn, and records its slave LMV EA in the
2176  *    lfsck_lmv::ll_lslr. In this step, if the 'shard' (may be fake
2177  *    shard) name does not match the shard naming rule, for example, it
2178  *    does not contains the shard's FID, or not contains index, then we
2179  *    can remove the bad name entry directly. But if the name is valid,
2180  *    but the shard has no slave LMV EA or the slave LMV EA does not
2181  *    match its name, then we just record related information in the
2182  *    lfsck_lmv::ll_lslr in RAM.
2183  *
2184  * 2) When all the known shards have been scanned, then the engine will
2185  *    generate a dummy request (via lfsck_namespace_close_dir) to tell
2186  *    the assistant thread that all the known shards have been scanned.
2187  *    Since the assistant has got the global knowledge about the index
2188  *    conflict, stripe count, hash type, and so on. Then the assistant
2189  *    thread will scan the lfsck_lmv::ll_lslr, and for every shard in
2190  *    the record, check and repair inconsistency.
2191  *
2192  * Generally, the stripe directory has only several shards, and there
2193  * will NOT be a lof of striped directory. So double scanning striped
2194  * directory will not much affect the LFSCK performance.
2195  *
2196  * \param[in] env       pointer to the thread context
2197  * \param[in] com       pointer to the lfsck component
2198  * \param[in] lnr       pointer to the namespace request that contains the
2199  *                      shard's name, parent object, parent's LMV, and ect.
2200  *
2201  * \retval              zero for succeed
2202  * \retval              negative error number on failure
2203  */
2204 int lfsck_namespace_handle_striped_master(const struct lu_env *env,
2205                                           struct lfsck_component *com,
2206                                           struct lfsck_namespace_req *lnr)
2207 {
2208         struct lfsck_thread_info   *info        = lfsck_env_info(env);
2209         struct lmv_mds_md_v1       *lmv         = &info->lti_lmv;
2210         struct lfsck_instance      *lfsck       = com->lc_lfsck;
2211         struct lfsck_namespace     *ns          = com->lc_file_ram;
2212         struct lfsck_lmv           *llmv        = lnr->lnr_lmv;
2213         struct dt_object           *dir         = lnr->lnr_obj;
2214         const struct lu_fid        *pfid        = lfsck_dto2fid(dir);
2215         struct dt_object           *obj         = NULL;
2216         struct dt_device           *dev         = NULL;
2217         int                         shard_idx   = 0;
2218         int                         stripe      = 0;
2219         int                         rc          = 0;
2220         int                         depth       = 0;
2221         bool                        repaired    = false;
2222         enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
2223         ENTRY;
2224
2225         if (unlikely(llmv->ll_ignore))
2226                 RETURN(0);
2227
2228         shard_idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
2229         if (shard_idx < 0)
2230                 GOTO(fail_lmv, rc = shard_idx);
2231
2232         if (shard_idx == lfsck_dev_idx(lfsck->li_bottom)) {
2233                 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
2234                         GOTO(out, rc = 0);
2235
2236                 dev = lfsck->li_next;
2237         } else {
2238                 struct lfsck_tgt_desc *ltd;
2239
2240                 /* Usually, some local filesystem consistency verification
2241                  * tools can guarantee the local namespace tree consistenct.
2242                  * So the LFSCK will only verify the remote directory. */
2243                 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
2244                         rc = lfsck_namespace_trace_update(env, com, pfid,
2245                                                 LNTF_CHECK_PARENT, true);
2246
2247                         GOTO(out, rc);
2248                 }
2249
2250                 ltd = LTD_TGT(&lfsck->li_mdt_descs, shard_idx);
2251                 if (unlikely(ltd == NULL)) {
2252                         CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
2253                                "did not join the namespace LFSCK\n",
2254                                lfsck_lfsck2name(lfsck), shard_idx);
2255                         lfsck_lad_set_bitmap(env, com, shard_idx);
2256
2257                         GOTO(fail_lmv, rc = -ENODEV);
2258                 }
2259
2260                 dev = ltd->ltd_tgt;
2261         }
2262
2263         obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
2264         if (IS_ERR(obj))
2265                 GOTO(fail_lmv, rc = PTR_ERR(obj));
2266
2267         if (!dt_object_exists(obj)) {
2268                 stripe = lfsck_shard_name_to_index(env, lnr->lnr_name,
2269                                 lnr->lnr_namelen, lnr->lnr_type, &lnr->lnr_fid);
2270                 if (stripe < 0) {
2271                         type = LNIT_BAD_DIRENT;
2272
2273                         GOTO(out, rc = 0);
2274                 }
2275
2276 dangling:
2277                 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
2278                 if (rc == 0)
2279                         rc = lfsck_record_lmv(env, com, lnr, NULL, stripe,
2280                                               LSLF_DANGLING, LSLF_NONE, &depth);
2281
2282                 GOTO(out, rc);
2283         }
2284
2285         stripe = lfsck_shard_name_to_index(env, lnr->lnr_name, lnr->lnr_namelen,
2286                                            lfsck_object_type(obj),
2287                                            &lnr->lnr_fid);
2288         if (stripe < 0) {
2289                 type = LNIT_BAD_DIRENT;
2290
2291                 GOTO(out, rc = 0);
2292         }
2293
2294         rc = lfsck_read_stripe_lmv(env, obj, lmv);
2295         if (unlikely(rc == -ENOENT))
2296                 /* It may happen when the remote object has been removed,
2297                  * but the local MDT does not aware of that. */
2298                 goto dangling;
2299
2300         if (rc == -ENODATA)
2301                 rc = lfsck_record_lmv(env, com, lnr, lmv, stripe,
2302                                       LSLF_NO_LMVEA, LSLF_NONE, &depth);
2303         else if (rc == 0)
2304                 rc = lfsck_record_lmv(env, com, lnr, lmv, stripe,
2305                                       lmv->lmv_master_mdt_index != stripe ?
2306                                       LSLF_BAD_INDEX1 : LSLF_NONE, LSLF_NONE,
2307                                       &depth);
2308
2309         GOTO(out, rc);
2310
2311 fail_lmv:
2312         llmv->ll_failed = 1;
2313
2314 out:
2315         if (rc >= 0 && type == LNIT_NONE && !S_ISDIR(lnr->lnr_type))
2316                 type = LNIT_BAD_TYPE;
2317
2318         switch (type) {
2319         case LNIT_BAD_TYPE:
2320                 rc = lfsck_namespace_repair_dirent(env, com, dir, obj,
2321                                                    lnr->lnr_name, lnr->lnr_name,
2322                                                    lnr->lnr_type, true, false);
2323                 if (rc > 0)
2324                         repaired = true;
2325                 break;
2326         case LNIT_BAD_DIRENT:
2327                 rc = lfsck_namespace_repair_dirent(env, com, dir, obj,
2328                                                    lnr->lnr_name, lnr->lnr_name,
2329                                                    lnr->lnr_type, false, false);
2330                 if (rc > 0)
2331                         repaired = true;
2332                 break;
2333         default:
2334                 break;
2335         }
2336
2337         if (rc < 0) {
2338                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
2339                        "the shard: "DFID", parent "DFID", name %.*s: rc = %d\n",
2340                        lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
2341                        PFID(lfsck_dto2fid(lnr->lnr_obj)),
2342                        lnr->lnr_namelen, lnr->lnr_name, rc);
2343
2344                 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
2345                      rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
2346                      rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
2347                     dev != NULL && dev != lfsck->li_next)
2348                         lfsck_lad_set_bitmap(env, com, shard_idx);
2349
2350                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT))
2351                         rc = 0;
2352         } else {
2353                 if (repaired) {
2354                         ns->ln_items_repaired++;
2355
2356                         switch (type) {
2357                         case LNIT_BAD_TYPE:
2358                                 ns->ln_bad_type_repaired++;
2359                                 break;
2360                         case LNIT_BAD_DIRENT:
2361                                 ns->ln_dirent_repaired++;
2362                                 break;
2363                         default:
2364                                 break;
2365                         }
2366                 }
2367
2368                 rc = 0;
2369         }
2370
2371         if (obj != NULL && !IS_ERR(obj))
2372                 lfsck_object_put(env, obj);
2373
2374         return rc;
2375 }