Whamcloud - gitweb
LU-7268 scrub: NOT assign LMA for EA inode
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_scrub.c
index bc7bf68..ff639e2 100644 (file)
@@ -20,7 +20,7 @@
  * GPL HEADER END
  */
 /*
- * Copyright (c) 2012, 2013, Intel Corporation.
+ * Copyright (c) 2012, 2015, Intel Corporation.
  */
 /*
  * lustre/osd-ldiskfs/osd_scrub.c
@@ -37,6 +37,7 @@
 
 #define DEBUG_SUBSYSTEM S_LFSCK
 
+#include <linux/kthread.h>
 #include <lustre/lustre_idl.h>
 #include <lustre_disk.h>
 #include <dt_object.h>
@@ -180,7 +181,8 @@ static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
        if (rc < 0)
                CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
                       DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
-                      PFID(fid), id->oii_ino, id->oii_gen, rc);
+                      PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
+                      rc);
 
        RETURN(rc);
 }
@@ -271,11 +273,14 @@ void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 {
        struct scrub_file *sf = &scrub->os_file;
 
-       CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, flags = "LPX64"\n",
-              osd_scrub2name(scrub), flags);
+       CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, old flags = "
+              LPX64", add flags = "LPX64"\n",
+              osd_scrub2name(scrub), sf->sf_flags, flags);
+
        memcpy(sf->sf_uuid, uuid, 16);
        sf->sf_status = SS_INIT;
        sf->sf_flags |= flags;
+       sf->sf_flags &= ~SF_AUTO;
        sf->sf_run_time = 0;
        sf->sf_time_latest_start = 0;
        sf->sf_time_last_checkpoint = 0;
@@ -367,6 +372,7 @@ osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
 {
        struct filter_fid_old   *ff      = &info->oti_ff;
        struct dentry           *dentry  = &info->oti_obj_dentry;
+       struct lu_fid           *tfid    = &info->oti_fid;
        handle_t                *jh;
        int                      size    = 0;
        int                      rc;
@@ -377,6 +383,15 @@ osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
        if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
                RETURN(0);
 
+       if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
+               struct ost_id *oi = &info->oti_ostid;
+
+               fid_to_ostid(fid, oi);
+               ostid_to_fid(tfid, oi, 0);
+       } else {
+               *tfid = *fid;
+       }
+
        /* We want the LMA to fit into the 256-byte OST inode, so operate
         * as following:
         * 1) read old XATTR_NAME_FID and save the parent FID;
@@ -392,7 +407,7 @@ osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
        if (IS_ERR(jh)) {
                rc = PTR_ERR(jh);
                CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
-                      DFID": rc = %d\n", osd_name(dev), PFID(fid), rc);
+                      DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
                RETURN(rc);
        }
 
@@ -413,7 +428,7 @@ osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
        }
 
        /* 3) make new LMA and add it */
-       rc = osd_ea_fid_set(info, inode, fid, LMAC_FID_ON_OST, 0);
+       rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
        if (rc == 0 && reset)
                size = sizeof(struct filter_fid);
        else if (rc != 0 && removed)
@@ -436,7 +451,7 @@ stop:
        ldiskfs_journal_stop(jh);
        if (rc < 0)
                CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
-                      osd_name(dev), PFID(fid), rc);
+                      osd_name(dev), PFID(tfid), rc);
        return rc;
 }
 
@@ -482,6 +497,10 @@ osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
                        GOTO(out, rc);
                }
 
+               /* The inode has been reused as EA inode, ignore it. */
+               if (unlikely(osd_is_ea_inode(inode)))
+                       GOTO(out, rc = 0);
+
                sf->sf_flags |= SF_UPGRADE;
                sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
                dev->od_check_ff = 1;
@@ -496,14 +515,19 @@ osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
            (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
                GOTO(out, rc = 0);
 
-       if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA))
+       if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) {
+               ops = DTO_INDEX_INSERT;
+
                goto iget;
+       }
 
        rc = osd_oi_lookup(info, dev, fid, lid2,
                (val == SCRUB_NEXT_OSTOBJ ||
                 val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
        if (rc != 0) {
-               if (rc != -ENOENT && rc != -ESTALE)
+               if (rc == -ENOENT)
+                       ops = DTO_INDEX_INSERT;
+               else if (rc != -ESTALE)
                        GOTO(out, rc);
 
 iget:
@@ -516,12 +540,15 @@ iget:
                                        rc = 0;
                                GOTO(out, rc);
                        }
+
+                       /* The inode has been reused as EA inode, ignore it. */
+                       if (unlikely(osd_is_ea_inode(inode)))
+                               GOTO(out, rc = 0);
                }
 
                if (!scrub->os_partial_scan)
                        scrub->os_full_speed = 1;
 
-               ops = DTO_INDEX_INSERT;
                idx = osd_oi_fid2idx(dev, fid);
                switch (val) {
                case SCRUB_NEXT_NOLMA:
@@ -579,9 +606,6 @@ iget:
                        sf->sf_items_updated_prior++;
                else
                        sf->sf_items_updated++;
-
-               /* The target has been changed, need to be re-loaded. */
-               lu_object_purge(info->oti_env, osd2lu_dev(dev), fid);
        }
 
        GOTO(out, rc);
@@ -599,7 +623,8 @@ out:
        /* There may be conflict unlink during the OI scrub,
         * if happend, then remove the new added OI mapping. */
        if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
-           unlikely(inode->i_nlink == 0))
+           unlikely(ldiskfs_test_inode_state(inode,
+                                             LDISKFS_STATE_LUSTRE_DESTROY)))
                osd_scrub_refresh_mapping(info, dev, fid, lid,
                                DTO_INDEX_DELETE, false,
                                (val == SCRUB_NEXT_OSTOBJ ||
@@ -611,13 +636,11 @@ out:
                iput(inode);
 
        if (oii != NULL) {
-               LASSERT(!list_empty(&oii->oii_list));
+               LASSERT(list_empty(&oii->oii_list));
 
-               spin_lock(&scrub->os_lock);
-               list_del_init(&oii->oii_list);
-               spin_unlock(&scrub->os_lock);
                OBD_FREE_PTR(oii);
        }
+
        RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 }
 
@@ -633,6 +656,9 @@ static int osd_scrub_prep(struct osd_device *dev)
        bool                  drop_dryrun = false;
        ENTRY;
 
+       CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
+              osd_scrub2name(scrub), flags);
+
        down_write(&scrub->os_rwsem);
        if (flags & SS_SET_FAILOUT)
                sf->sf_param |= SP_FAILOUT;
@@ -686,6 +712,7 @@ static int osd_scrub_prep(struct osd_device *dev)
        sf->sf_status = SS_SCANNING;
        sf->sf_time_latest_start = cfs_time_current_sec();
        sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
+       sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
        rc = osd_scrub_file_store(scrub);
        if (rc == 0) {
                spin_lock(&scrub->os_lock);
@@ -721,11 +748,15 @@ static int osd_scrub_checkpoint(struct osd_scrub *scrub)
        return rc;
 }
 
-static void osd_scrub_post(struct osd_scrub *scrub, int result)
+static int osd_scrub_post(struct osd_scrub *scrub, int result)
 {
        struct scrub_file *sf = &scrub->os_file;
+       int rc;
        ENTRY;
 
+       CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
+              osd_scrub2name(scrub), result);
+
        down_write(&scrub->os_rwsem);
        spin_lock(&scrub->os_lock);
        thread_set_flags(&scrub->os_thread, SVC_STOPPING);
@@ -760,10 +791,10 @@ static void osd_scrub_post(struct osd_scrub *scrub, int result)
        }
        sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
                                            scrub->os_time_last_checkpoint);
-       result = osd_scrub_file_store(scrub);
+       rc = osd_scrub_file_store(scrub);
        up_write(&scrub->os_rwsem);
 
-       EXIT;
+       RETURN(rc < 0 ? rc : result);
 }
 
 /* iteration engine */
@@ -935,6 +966,10 @@ static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
                RETURN(rc);
        }
 
+       /* It is an EA inode, no OI mapping for it, skip it. */
+       if (osd_is_ea_inode(inode))
+               GOTO(put, rc = SCRUB_NEXT_CONTINUE);
+
        if (scrub &&
            ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
                /* Only skip it for the first OI scrub accessing. */
@@ -965,10 +1000,11 @@ static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
                struct l_wait_info lwi;
 
                lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
-               l_wait_event(thread->t_ctl_waitq,
-                            !list_empty(&scrub->os_inconsistent_items) ||
-                            !thread_is_running(thread),
-                            &lwi);
+               if (likely(lwi.lwi_timeout > 0))
+                       l_wait_event(thread->t_ctl_waitq,
+                               !list_empty(&scrub->os_inconsistent_items) ||
+                               !thread_is_running(thread),
+                               &lwi);
        }
 
        if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
@@ -985,13 +1021,21 @@ static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
                return SCRUB_NEXT_EXIT;
 
        if (!list_empty(&scrub->os_inconsistent_items)) {
-               struct osd_inconsistent_item *oii;
+               spin_lock(&scrub->os_lock);
+               if (likely(!list_empty(&scrub->os_inconsistent_items))) {
+                       struct osd_inconsistent_item *oii;
 
-               oii = list_entry(scrub->os_inconsistent_items.next,
-                                struct osd_inconsistent_item, oii_list);
-               *oic = &oii->oii_cache;
-               scrub->os_in_prior = 1;
-               return 0;
+                       oii = list_entry(scrub->os_inconsistent_items.next,
+                               struct osd_inconsistent_item, oii_list);
+                       list_del_init(&oii->oii_list);
+                       spin_unlock(&scrub->os_lock);
+
+                       *oic = &oii->oii_cache;
+                       scrub->os_in_prior = 1;
+
+                       return 0;
+               }
+               spin_unlock(&scrub->os_lock);
        }
 
        if (noslot)
@@ -1078,8 +1122,10 @@ static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
        }
 
        rc = osd_scrub_check_update(info, dev, oic, rc);
-       if (rc != 0)
+       if (rc != 0) {
+               scrub->os_in_prior = 0;
                return rc;
+       }
 
        rc = osd_scrub_checkpoint(scrub);
        if (rc != 0) {
@@ -1142,7 +1188,8 @@ static int osd_preload_exec(struct osd_thread_info *info,
 #define SCRUB_IT_ALL   1
 #define SCRUB_IT_CRASH 2
 
-static void osd_scrub_join(struct osd_device *dev, __u32 flags)
+static void osd_scrub_join(struct osd_device *dev, __u32 flags,
+                          bool inconsistent)
 {
        struct osd_scrub     *scrub  = &dev->od_scrub;
        struct ptlrpc_thread *thread = &scrub->os_thread;
@@ -1164,12 +1211,19 @@ static void osd_scrub_join(struct osd_device *dev, __u32 flags)
        else if (flags & SS_CLEAR_DRYRUN)
                sf->sf_param &= ~SP_DRYRUN;
 
-       if (flags & SS_RESET)
+       if (flags & SS_RESET) {
                osd_scrub_file_reset(scrub,
-                       LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
+                       LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
+                       inconsistent ? SF_INCONSISTENT : 0);
+               sf->sf_status = SS_SCANNING;
+       }
+
+       if (flags & SS_AUTO_FULL) {
+               sf->sf_flags |= SF_AUTO;
+               scrub->os_full_speed = 1;
+       }
 
-       if (flags & SS_AUTO_FULL ||
-           sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
+       if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
                scrub->os_full_speed = 1;
        else
                scrub->os_full_speed = 0;
@@ -1183,6 +1237,7 @@ static void osd_scrub_join(struct osd_device *dev, __u32 flags)
        scrub->os_pos_current = sf->sf_pos_latest_start;
        sf->sf_time_latest_start = cfs_time_current_sec();
        sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
+       sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
        rc = osd_scrub_file_store(scrub);
        if (rc != 0)
                CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
@@ -1211,7 +1266,7 @@ static int osd_inode_iteration(struct osd_thread_info *info,
        osd_iit_exec_policy   exec;
        __u32                *pos;
        __u32                *count;
-       struct osd_iit_param  param  = { 0 };
+       struct osd_iit_param  param  = { NULL };
        struct l_wait_info    lwi    = { 0 };
        __u32                 limit;
        int                   rc;
@@ -1235,7 +1290,7 @@ static int osd_inode_iteration(struct osd_thread_info *info,
                        RETURN(-EINVAL);
                case SCRUB_NEXT_WAIT: {
                        struct kstatfs *ksfs = &info->oti_ksfs;
-                       __u64 used;
+                       __u64 saved_flags;
 
                        if (dev->od_full_scrub_ratio == OFSR_NEVER ||
                            unlikely(sf->sf_items_updated_prior == 0))
@@ -1243,21 +1298,24 @@ static int osd_inode_iteration(struct osd_thread_info *info,
 
                        if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
                            scrub->os_full_scrub) {
-                               osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET);
+                               osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
+                                              true);
                                goto full;
                        }
 
                        rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
-                       if (rc != 0)
-                               goto wait;
-
-                       used = ksfs->f_files - ksfs->f_ffree;
-                       do_div(used, sf->sf_items_updated_prior);
-                       /* If we hit too much inconsistent OI mappings during
-                        * the partial scan, then scan the device completely. */
-                       if (used < dev->od_full_scrub_ratio) {
-                               osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET);
-                               goto full;
+                       if (rc == 0) {
+                               __u64 used = ksfs->f_files - ksfs->f_ffree;
+
+                               do_div(used, sf->sf_items_updated_prior);
+                               /* If we hit too much inconsistent OI
+                                * mappings during the partial scan,
+                                * then scan the device completely. */
+                               if (used < dev->od_full_scrub_ratio) {
+                                       osd_scrub_join(dev,
+                                               SS_AUTO_FULL | SS_RESET, true);
+                                       goto full;
+                               }
                        }
 
 wait:
@@ -1265,15 +1323,18 @@ wait:
                            cfs_fail_val > 0)
                                continue;
 
-                       sf->sf_status = SS_COMPLETED;
+                       saved_flags = sf->sf_flags;
                        sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
                                          SF_UPGRADE | SF_AUTO);
+                       sf->sf_status = SS_COMPLETED;
                        l_wait_event(thread->t_ctl_waitq,
                                     !thread_is_running(thread) ||
                                     !scrub->os_partial_scan ||
                                     scrub->os_in_join ||
                                     !list_empty(&scrub->os_inconsistent_items),
                                     &lwi);
+                       sf->sf_flags = saved_flags;
+                       sf->sf_status = SS_SCANNING;
 
                        if (unlikely(!thread_is_running(thread)))
                                RETURN(0);
@@ -1286,8 +1347,6 @@ wait:
                default:
                        LASSERTF(rc == 0, "rc = %d\n", rc);
 
-                       sf->sf_status = SS_SCANNING;
-                       sf->sf_flags |= SF_AUTO;
                        osd_scrub_exec(info, dev, &param, oic, &noslot, rc);
                        break;
                }
@@ -1295,7 +1354,6 @@ wait:
 
 full:
        if (!preload) {
-               sf->sf_status = SS_SCANNING;
                l_wait_event(thread->t_ctl_waitq,
                             !thread_is_running(thread) || !scrub->os_in_join,
                             &lwi);
@@ -1457,7 +1515,7 @@ static int osd_scrub_main(void *args)
        GOTO(post, rc);
 
 post:
-       osd_scrub_post(scrub, rc);
+       rc = osd_scrub_post(scrub, rc);
        CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
               osd_scrub2name(scrub), scrub->os_pos_current, rc);
 
@@ -1491,6 +1549,8 @@ static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
                           loff_t offset, __u64 ino, unsigned d_type);
 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
                           loff_t offset, __u64 ino, unsigned d_type);
+static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
+                           loff_t offset, __u64 ino, unsigned d_type);
 
 static int
 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
@@ -1508,12 +1568,14 @@ enum osd_lf_flags {
        OLF_HIDE_FID            = 0x0002,
        OLF_SHOW_NAME           = 0x0004,
        OLF_NO_OI               = 0x0008,
+       OLF_IDX_IN_FID          = 0x0010,
 };
 
 struct osd_lf_map {
        char            *olm_name;
        struct lu_fid    olm_fid;
        __u16            olm_flags;
+       __u16            olm_namelen;
        scandir_t        olm_scandir;
        filldir_t        olm_filldir;
 };
@@ -1522,107 +1584,144 @@ struct osd_lf_map {
 static const struct osd_lf_map osd_lf_maps[] = {
        /* CATALOGS */
        { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
-               NULL, NULL },
+               sizeof(CATLIST) - 1, NULL, NULL },
 
        /* CONFIGS */
        { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
-               OLF_SCAN_SUBITEMS, osd_ios_general_scan,
-               osd_ios_varfid_fill },
+               OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
+               osd_ios_general_scan, osd_ios_varfid_fill },
 
        /* NIDTBL_VERSIONS */
        { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
-               osd_ios_general_scan, osd_ios_varfid_fill },
+               sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
+               osd_ios_varfid_fill },
 
        /* PENDING */
-       { "PENDING", { 0, 0, 0 }, 0, NULL, NULL },
+       { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
 
        /* ROOT */
-       { "ROOT", { FID_SEQ_ROOT, 1, 0 },
-               OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL },
+       { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
+               OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
+               osd_ios_ROOT_scan, NULL },
 
        /* changelog_catalog */
-       { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL },
+       { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
+               NULL, NULL },
 
        /* changelog_users */
-       { CHANGELOG_USERS, { 0, 0, 0 }, 0, NULL, NULL },
+       { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
+               NULL, NULL },
 
        /* fld */
        { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
-               NULL, NULL },
+               sizeof("fld") - 1, NULL, NULL },
 
        /* last_rcvd */
        { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
-               NULL, NULL },
+               sizeof(LAST_RCVD) - 1, NULL, NULL },
 
-       /* lfsck_bookmark */
-       { "lfsck_bookmark", { 0, 0, 0 }, 0, NULL, NULL },
+       /* reply_data */
+       { REPLY_DATA, { FID_SEQ_LOCAL_FILE, REPLY_DATA_OID, 0 }, OLF_SHOW_NAME,
+               sizeof(REPLY_DATA) - 1, NULL, NULL },
 
        /* lov_objid */
        { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
-               NULL, NULL },
+               sizeof(LOV_OBJID) - 1, NULL, NULL },
 
        /* lov_objseq */
        { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
-               OLF_SHOW_NAME, NULL, NULL },
+               OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
 
        /* quota_master */
-       { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
+       { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
                osd_ios_general_scan, osd_ios_varfid_fill },
 
        /* quota_slave */
-       { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
+       { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
                osd_ios_general_scan, osd_ios_varfid_fill },
 
        /* seq_ctl */
        { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
-               OLF_SHOW_NAME, NULL, NULL },
+               OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
 
        /* seq_srv */
        { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
-               OLF_SHOW_NAME, NULL, NULL },
+               OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
 
        /* health_check */
        { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
-               OLF_SHOW_NAME, NULL, NULL },
+               OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
+
+       /* LFSCK */
+       { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
+               osd_ios_general_scan, osd_ios_varfid_fill },
+
+       /* lfsck_bookmark */
+       { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
+               NULL, NULL },
+
+       /* lfsck_layout */
+       { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
+               NULL, NULL },
 
        /* lfsck_namespace */
-       { "lfsck_namespace", { 0, 0, 0 }, 0, NULL, NULL },
+       { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
+               NULL, NULL },
 
        /* OBJECTS, upgrade from old device */
-       { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL },
+       { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
+               osd_ios_OBJECTS_scan, NULL },
 
        /* lquota_v2.user, upgrade from old device */
-       { "lquota_v2.user", { 0, 0, 0 }, 0, NULL, NULL },
+       { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
+               NULL, NULL },
 
        /* lquota_v2.group, upgrade from old device */
-       { "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL },
+       { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
+               NULL, NULL },
 
        /* LAST_GROUP, upgrade from old device */
        { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
-               OLF_SHOW_NAME, NULL, NULL },
+               OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
 
-       /* SLAVE_LOG, llog for destroy slave stripes of striped dir */
-       { "SLAVE_LOG", { FID_SEQ_LOCAL_FILE, SLAVE_LLOG_CATALOGS_OID, 0 },
-              OLF_SHOW_NAME, NULL, NULL },
+       /* committed batchid for cross-MDT operation */
+       { "BATCHID", { FID_SEQ_LOCAL_FILE, BATCHID_COMMITTED_OID, 0 },
+               OLF_SHOW_NAME, sizeof("BATCHID") - 1, NULL, NULL },
+
+       /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
+        * and f_oid = index for their log files.  See lu_update_log{_dir}_fid()
+        * for more details. */
+
+       /* update_log */
+       { "update_log", { FID_SEQ_UPDATE_LOG, 0, 0 },
+               OLF_SHOW_NAME | OLF_IDX_IN_FID, sizeof("update_log") - 1,
+               NULL, NULL },
+
+       /* update_log_dir */
+       { "update_log_dir", { FID_SEQ_UPDATE_LOG_DIR, 0, 0 },
+               OLF_SHOW_NAME | OLF_SCAN_SUBITEMS | OLF_IDX_IN_FID,
+               sizeof("update_log_dir") - 1,
+               osd_ios_general_scan, osd_ios_uld_fill },
 
        /* lost+found */
        { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
-               OLF_SCAN_SUBITEMS, osd_ios_general_scan, osd_ios_lf_fill },
+               OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
+               osd_ios_general_scan, osd_ios_lf_fill },
 
-       { NULL, { 0, 0, 0 }, 0, NULL, NULL }
+       { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
 };
 
 /* Add the new introduced files under .lustre/ in the list in the future. */
 static const struct osd_lf_map osd_dl_maps[] = {
        /* .lustre/fid */
        { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
-               NULL, NULL },
+               sizeof("fid") - 1, NULL, NULL },
 
        /* .lustre/lost+found */
        { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
-               NULL, NULL },
+               sizeof("lost+found") - 1, NULL, NULL },
 
-       { NULL, { 0, 0, 0 }, 0, NULL, NULL }
+       { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
 };
 
 struct osd_ios_item {
@@ -1648,7 +1747,19 @@ osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
        struct dentry *dentry;
 
        dentry = ll_lookup_one_len(name, parent, namelen);
-       if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
+       if (IS_ERR(dentry)) {
+               int rc = PTR_ERR(dentry);
+
+               if (rc != -ENOENT)
+                       CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
+                              namelen, name, parent->d_name.len,
+                              parent->d_name.name, parent->d_inode->i_ino,
+                              parent->d_inode->i_generation, rc);
+
+               return dentry;
+       }
+
+       if (dentry->d_inode == NULL) {
                dput(dentry);
                return ERR_PTR(-ENOENT);
        }
@@ -1706,10 +1817,16 @@ osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
 
        osd_id_gen(id, inode->i_ino, inode->i_generation);
        if (rc == -ENODATA) {
-               if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
+               if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
                        lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
-               else
+               } else {
                        tfid = *fid;
+                       if (flags & OLF_IDX_IN_FID) {
+                               LASSERT(dev->od_index >= 0);
+
+                               tfid.f_oid = dev->od_index;
+                       }
+               }
                rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
                if (rc != 0) {
                        CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
@@ -1871,7 +1988,7 @@ static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
                RETURN(0);
 
        for (map = osd_dl_maps; map->olm_name != NULL; map++) {
-               if (strlen(map->olm_name) != namelen)
+               if (map->olm_namelen != namelen)
                        continue;
 
                if (strncmp(map->olm_name, name, namelen) == 0)
@@ -1892,6 +2009,35 @@ static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
        RETURN(rc);
 }
 
+static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
+                           loff_t offset, __u64 ino, unsigned d_type)
+{
+       struct osd_ios_filldir_buf *fill_buf = buf;
+       struct dentry              *child;
+       struct lu_fid               tfid;
+       int                         rc       = 0;
+       ENTRY;
+
+       /* skip any non-DFID format name */
+       if (name[0] != '[')
+               RETURN(0);
+
+       child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
+       if (IS_ERR(child))
+               RETURN(PTR_ERR(child));
+
+       /* skip the start '[' */
+       sscanf(&name[1], SFID, RFID(&tfid));
+       if (fid_is_sane(&tfid))
+               rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
+                                     child->d_inode, &tfid, 0);
+       else
+               rc = -EIO;
+       dput(child);
+
+       RETURN(rc);
+}
+
 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
                             loff_t offset, __u64 ino, unsigned d_type)
 {
@@ -1907,7 +2053,7 @@ static int osd_ios_root_fill(void *buf, const char *name, int namelen,
                RETURN(0);
 
        for (map = osd_lf_maps; map->olm_name != NULL; map++) {
-               if (strlen(map->olm_name) != namelen)
+               if (map->olm_namelen != namelen)
                        continue;
 
                if (strncmp(map->olm_name, name, namelen) == 0)
@@ -1943,7 +2089,7 @@ osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
                                                .oifb_info = info,
                                                .oifb_dev = dev,
                                                .oifb_dentry = dentry };
-       struct file                  *filp  = &info->oti_it_ea.oie_file;
+       struct file                  *filp  = &info->oti_file;
        struct inode                 *inode = dentry->d_inode;
        const struct file_operations *fops  = inode->i_fop;
        int                           rc;
@@ -1952,7 +2098,7 @@ osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
        LASSERT(filldir != NULL);
 
        filp->f_pos = 0;
-       filp->f_dentry = dentry;
+       filp->f_path.dentry = dentry;
        filp->f_mode = FMODE_64BITHASH;
        filp->f_mapping = inode->i_mapping;
        filp->f_op = fops;
@@ -2149,7 +2295,7 @@ static int osd_initial_OI_scrub(struct osd_thread_info *info,
 
                child = osd_ios_lookup_one_len(map->olm_name,
                                               osd_sb(dev)->s_root,
-                                              strlen(map->olm_name));
+                                              map->olm_namelen);
                if (!IS_ERR(child))
                        dput(child);
                else if (PTR_ERR(child) == -ENOENT)
@@ -2198,10 +2344,11 @@ static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
 again:
        if (thread_is_running(thread)) {
                spin_unlock(&scrub->os_lock);
-               if (!scrub->os_partial_scan || flags & SS_AUTO_PARTIAL)
+               if (!(scrub->os_file.sf_flags & SF_AUTO) ||
+                    (flags & (SS_AUTO_FULL | SS_AUTO_PARTIAL)))
                        RETURN(-EALREADY);
 
-               osd_scrub_join(dev, flags);
+               osd_scrub_join(dev, flags, false);
                spin_lock(&scrub->os_lock);
                if (!thread_is_running(thread))
                        goto again;
@@ -2328,19 +2475,19 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
                RETURN(PTR_ERR(filp));
        }
 
-       inode = filp->f_dentry->d_inode;
+       inode = filp->f_path.dentry->d_inode;
        /* 'What the @fid is' is not imporatant, because the object
         * has no OI mapping, and only is visible inside the OSD.*/
        lu_igif_build(fid, inode->i_ino, inode->i_generation);
        rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
        if (rc != 0) {
-               filp_close(filp, 0);
+               filp_close(filp, NULL);
                pop_ctxt(&saved, ctxt);
                RETURN(rc);
        }
 
        scrub->os_inode = igrab(inode);
-       filp_close(filp, 0);
+       filp_close(filp, NULL);
        pop_ctxt(&saved, ctxt);
 
        rc = osd_scrub_file_load(scrub);
@@ -2365,8 +2512,29 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
                GOTO(cleanup_inode, rc);
        } else {
                if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
+                       struct obd_uuid *old_uuid;
+                       struct obd_uuid *new_uuid;
+
+                       OBD_ALLOC_PTR(old_uuid);
+                       OBD_ALLOC_PTR(new_uuid);
+                       if (old_uuid == NULL || new_uuid == NULL) {
+                               CERROR("%.16s: UUID has been changed, but"
+                                      "failed to allocate RAM for report\n",
+                                      LDISKFS_SB(sb)->s_es->s_volume_name);
+                       } else {
+                               class_uuid_unparse(sf->sf_uuid, old_uuid);
+                               class_uuid_unparse(es->s_uuid, new_uuid);
+                               CERROR("%.16s: UUID has been changed from "
+                                      "%s to %s\n",
+                                      LDISKFS_SB(sb)->s_es->s_volume_name,
+                                      old_uuid->uuid, new_uuid->uuid);
+                       }
                        osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
                        dirty = 1;
+                       if (old_uuid != NULL)
+                               OBD_FREE_PTR(old_uuid);
+                       if (new_uuid != NULL)
+                               OBD_FREE_PTR(new_uuid);
                } else if (sf->sf_status == SS_SCANNING) {
                        sf->sf_status = SS_CRASHED;
                        dirty = 1;
@@ -2462,8 +2630,7 @@ void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
 /* object table based iteration APIs */
 
 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
-                                      struct dt_object *dt, __u32 attr,
-                                      struct lustre_capa *capa)
+                                      struct dt_object *dt, __u32 attr)
 {
        enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
        enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
@@ -2750,7 +2917,8 @@ int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
                }
 
                scrub->os_bad_oimap_time = now;
-               if (++scrub->os_bad_oimap_count > dev->od_full_scrub_speed)
+               if (++scrub->os_bad_oimap_count >
+                   dev->od_full_scrub_threshold_rate)
                        scrub->os_full_scrub = 1;
        }
 
@@ -2955,7 +3123,7 @@ int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
                              "real-time_speed: "LPU64" objects/sec\n"
                              "current_position: %u\n"
                              "lf_scanned: "LPU64"\n"
-                             "lf_reparied: "LPU64"\n"
+                             "lf_repaired: "LPU64"\n"
                              "lf_failed: "LPU64"\n",
                              rtime, speed, new_checked, scrub->os_pos_current,
                              scrub->os_lf_scanned, scrub->os_lf_repaired,
@@ -2968,7 +3136,7 @@ int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
                              "real-time_speed: N/A\n"
                              "current_position: N/A\n"
                              "lf_scanned: "LPU64"\n"
-                             "lf_reparied: "LPU64"\n"
+                             "lf_repaired: "LPU64"\n"
                              "lf_failed: "LPU64"\n",
                              sf->sf_run_time, speed, scrub->os_lf_scanned,
                              scrub->os_lf_repaired, scrub->os_lf_failed);