Whamcloud - gitweb
LU-9934 build: address issues raised by gcc7
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_oi.c
index bb3efa4..5dceea7 100644 (file)
  *
  * You should have received a copy of the GNU General Public License
  * version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
  * GPL HEADER END
  */
@@ -27,7 +23,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2012, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
  * Author: Nikita Danilov <nikita@clusterfs.com>
  */
 
-/*
- * oi uses two mechanisms to implement fid->cookie mapping:
- *
- *     - persistent index, where cookie is a record and fid is a key, and
- *
- *     - algorithmic mapping for "igif" fids.
- *
- */
-
-#define DEBUG_SUBSYSTEM S_MDS
+#define DEBUG_SUBSYSTEM S_OSD
 
 #include <linux/module.h>
 
-/* LUSTRE_VERSION_CODE */
-#include <lustre_ver.h>
 /*
  * struct OBD_{ALLOC,FREE}*()
  * OBD_FAIL_CHECK
 /* fid_cpu_to_be() */
 #include <lustre_fid.h>
 #include <dt_object.h>
+#include <lustre_scrub.h>
 
 #include "osd_oi.h"
 /* osd_lookup(), struct osd_thread_info */
 #include "osd_internal.h"
-#include "osd_scrub.h"
-
-static unsigned int osd_oi_count = OSD_OI_FID_NR;
-CFS_MODULE_PARM(osd_oi_count, "i", int, 0444,
-                "Number of Object Index containers to be created, "
-                "it's only valid for new filesystem.");
 
-/** to serialize concurrent OI index initialization */
-static struct mutex oi_init_lock;
+unsigned int osd_oi_count = OSD_OI_FID_NR;
+module_param(osd_oi_count, int, 0444);
+MODULE_PARM_DESC(osd_oi_count, "Number of Object Index containers to be created, it's only valid for new filesystem.");
 
 static struct dt_index_features oi_feat = {
         .dif_flags       = DT_IND_UPDATE,
@@ -126,8 +107,8 @@ static int osd_oi_index_create_one(struct osd_thread_info *info,
        int                              rc;
 
        dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
-       bh = osd_ldiskfs_find_entry(dir, dentry, &de, NULL);
-       if (bh) {
+       bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, NULL);
+       if (!IS_ERR(bh)) {
                osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
                brelse(bh);
                inode = osd_iget(info, osd, id);
@@ -138,7 +119,10 @@ static int osd_oi_index_create_one(struct osd_thread_info *info,
                return PTR_ERR(inode);
        }
 
-       jh = ldiskfs_journal_start_sb(sb, 100);
+       if (osd->od_dt_dev.dd_rdonly)
+               RETURN(-EROFS);
+
+       jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, 100);
        if (IS_ERR(jh))
                return PTR_ERR(jh);
 
@@ -148,6 +132,9 @@ static int osd_oi_index_create_one(struct osd_thread_info *info,
                return PTR_ERR(inode);
        }
 
+       ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
+       unlock_new_inode(inode);
+
        if (feat->dif_flags & DT_IND_VARKEY)
                rc = iam_lvar_create(inode, feat->dif_keysize_max,
                                     feat->dif_ptrsize, feat->dif_recsize_max,
@@ -157,7 +144,7 @@ static int osd_oi_index_create_one(struct osd_thread_info *info,
                                     feat->dif_ptrsize, feat->dif_recsize_max,
                                     jh);
        dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
-       rc = osd_ldiskfs_add_entry(jh, dentry, inode, NULL);
+       rc = osd_ldiskfs_add_entry(info, osd, jh, dentry, inode, NULL);
        ldiskfs_journal_stop(jh);
        iput(inode);
        return rc;
@@ -237,7 +224,17 @@ static int osd_oi_open(struct osd_thread_info *info, struct osd_device *osd,
         if (IS_ERR(inode))
                 RETURN(PTR_ERR(inode));
 
-       ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
+       if (!osd->od_dt_dev.dd_rdonly) {
+               /* 'What the @fid is' is not imporatant, because these objects
+                * have no OI mappings, and only are visible inside the OSD.*/
+               lu_igif_build(&info->oti_fid, inode->i_ino,
+                             inode->i_generation);
+               rc = osd_ea_fid_set(info, inode, &info->oti_fid,
+                                   LMAC_NOT_IN_OI, 0);
+               if (rc)
+                       GOTO(out_inode, rc);
+       }
+
         OBD_ALLOC_PTR(oi);
         if (oi == NULL)
                 GOTO(out_inode, rc = -ENOMEM);
@@ -287,10 +284,10 @@ static int
 osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
                  struct osd_oi **oi_table, unsigned oi_count, bool create)
 {
-       struct scrub_file *sf = &osd->od_scrub.os_file;
-       int                count = 0;
-       int                rc = 0;
-       int                i;
+       struct scrub_file *sf = &osd->od_scrub.os_scrub.os_file;
+       int count = 0;
+       int rc = 0;
+       int i;
        ENTRY;
 
        /* NB: oi_count != 0 means that we have already created/known all OIs
@@ -298,14 +295,14 @@ osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
        LASSERT(oi_count <= OSD_OI_FID_NR_MAX);
 
        for (i = 0; i < (oi_count != 0 ? oi_count : OSD_OI_FID_NR_MAX); i++) {
-               char name[12];
+               char name[sizeof(OSD_OI_NAME_BASE) + 3 * sizeof(i) + 1];
 
                if (oi_table[i] != NULL) {
                        count++;
                        continue;
                }
 
-               sprintf(name, "%s.%d", OSD_OI_NAME_BASE, i);
+               snprintf(name, sizeof(name), "%s.%d", OSD_OI_NAME_BASE, i);
                rc = osd_oi_open(info, osd, name, &oi_table[i], create);
                if (rc == 0) {
                        count++;
@@ -321,12 +318,11 @@ osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
                        continue;
                }
 
-               CERROR("%.16s: can't open %s: rc = %d\n",
-                      LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name, name, rc);
+               CERROR("%s: can't open %s: rc = %d\n",
+                      osd_dev2name(osd), name, rc);
                if (oi_count > 0)
-                       CERROR("%.16s: expect to open total %d OI files.\n",
-                              LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name,
-                              oi_count);
+                       CERROR("%s: expect to open total %d OI files.\n",
+                              osd_dev2name(osd), oi_count);
                break;
        }
 
@@ -338,68 +334,158 @@ osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
        RETURN(count);
 }
 
-int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd)
+static int osd_remove_oi_one(struct dentry *parent, const char *name,
+                            int namelen)
+{
+       struct dentry *child;
+       int rc;
+
+       child = ll_lookup_one_len(name, parent, namelen);
+       if (IS_ERR(child)) {
+               rc = PTR_ERR(child);
+       } else {
+               rc = ll_vfs_unlink(parent->d_inode, child);
+               dput(child);
+       }
+
+       return rc == -ENOENT ? 0 : rc;
+}
+
+static int osd_remove_ois(struct osd_thread_info *info, struct osd_device *osd)
+{
+       char name[16];
+       int namelen;
+       int rc;
+       int i;
+
+       if (osd->od_dt_dev.dd_rdonly)
+               RETURN(-EROFS);
+
+       for (i = 0; i < OSD_OI_FID_NR_MAX; i++) {
+               namelen = snprintf(name, sizeof(name), "%s.%d",
+                                  OSD_OI_NAME_BASE, i);
+               rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
+               if (rc != 0) {
+                       CERROR("%s: fail to remove the stale OI file %s: "
+                              "rc = %d\n", osd_dev2name(osd), name, rc);
+                       return rc;
+               }
+       }
+
+       namelen = snprintf(name, sizeof(name), "%s", OSD_OI_NAME_BASE);
+       rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
+       if (rc != 0)
+               CERROR("%s: fail to remove the stale OI file %s: rc = %d\n",
+                      osd_dev2name(osd), name, rc);
+
+       return rc;
+}
+
+int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd,
+               bool restored)
 {
-       struct osd_scrub  *scrub = &osd->od_scrub;
+       struct lustre_scrub *scrub = &osd->od_scrub.os_scrub;
        struct scrub_file *sf = &scrub->os_file;
-       struct osd_oi    **oi;
-       int                rc;
+       struct osd_oi **oi;
+       int count;
+       int rc;
        ENTRY;
 
+       if (unlikely(sf->sf_oi_count & (sf->sf_oi_count - 1)) != 0) {
+               LCONSOLE_WARN("%s: Invalid OI count in scrub file %d\n",
+                             osd_dev2name(osd), sf->sf_oi_count);
+               sf->sf_oi_count = 0;
+       }
+
+       if (restored) {
+               rc = osd_remove_ois(info, osd);
+               if (rc)
+                       RETURN(rc);
+       }
+
        OBD_ALLOC(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
        if (oi == NULL)
                RETURN(-ENOMEM);
 
-       mutex_lock(&oi_init_lock);
        /* try to open existing multiple OIs first */
-       rc = osd_oi_table_open(info, osd, oi, sf->sf_oi_count, false);
-       if (rc < 0)
-               GOTO(out, rc);
-
-       if (rc > 0) {
-               if (rc == sf->sf_oi_count || sf->sf_oi_count == 0)
-                       GOTO(out, rc);
+       count = osd_oi_table_open(info, osd, oi, sf->sf_oi_count, false);
+       if (count < 0)
+               GOTO(out, rc = count);
+
+       if (count > 0) {
+               if (count == sf->sf_oi_count)
+                       GOTO(out, rc = count);
+
+               if (sf->sf_oi_count == 0) {
+                       if (likely((count & (count - 1)) == 0))
+                               GOTO(out, rc = count);
+
+                       LCONSOLE_WARN("%s: invalid oi count %d, remove them, "
+                                     "then set it to %d\n", osd_dev2name(osd),
+                                     count, osd_oi_count);
+                       osd_oi_table_put(info, oi, count);
+                       rc = osd_remove_ois(info, osd);
+                       if (rc)
+                               GOTO(out, rc);
+
+                       sf->sf_oi_count = osd_oi_count;
+               }
 
-               osd_scrub_file_reset(scrub,
-                                    LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
-                                    SF_RECREATED);
-               osd_oi_count = sf->sf_oi_count;
+               scrub_file_reset(scrub, LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
+                                SF_RECREATED);
+               count = sf->sf_oi_count;
                goto create;
        }
 
        /* if previous failed then try found single OI from old filesystem */
        rc = osd_oi_open(info, osd, OSD_OI_NAME_BASE, &oi[0], false);
        if (rc == 0) { /* found single OI from old filesystem */
+               count = 1;
+               ldiskfs_clear_bit(0, sf->sf_oi_bitmap);
+               if (sf->sf_success_count == 0)
+                       /* XXX: There is one corner case that if the OI_scrub
+                        *      file crashed or lost and we regard it upgrade,
+                        *      then we allow IGIF lookup to bypass OI files.
+                        *
+                        *      The risk is that osd_fid_lookup() may found
+                        *      a wrong inode with the given IGIF especially
+                        *      when the MDT has performed file-level backup
+                        *      and restored after former upgrading from 1.8
+                        *      to 2.x. Fortunately, the osd_fid_lookup()can
+                        *      verify the inode to decrease the risk. */
+                       scrub_file_reset(scrub,
+                                        LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
+                                        SF_UPGRADE);
                GOTO(out, rc = 1);
        } else if (rc != -ENOENT) {
-               CERROR("%.16s: can't open %s: rc = %d\n",
-                      LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name,
-                      OSD_OI_NAME_BASE, rc);
+               CERROR("%s: can't open %s: rc = %d\n",
+                      osd_dev2name(osd), OSD_OI_NAME_BASE, rc);
                GOTO(out, rc);
        }
 
        if (sf->sf_oi_count > 0) {
                int i;
 
+               count = sf->sf_oi_count;
                memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
-               for (i = 0; i < osd_oi_count; i++)
+               for (i = 0; i < count; i++)
                        ldiskfs_set_bit(i, sf->sf_oi_bitmap);
-               osd_scrub_file_reset(scrub,
-                                    LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
-                                    SF_RECREATED);
+               scrub_file_reset(scrub, LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
+                                SF_RECREATED);
+       } else {
+               count = sf->sf_oi_count = osd_oi_count;
        }
-       sf->sf_oi_count = osd_oi_count;
 
 create:
-       rc = osd_scrub_file_store(scrub);
+       rc = scrub_file_store(info->oti_env, scrub);
        if (rc < 0) {
-               osd_oi_table_put(info, oi, sf->sf_oi_count);
+               osd_oi_table_put(info, oi, count);
                GOTO(out, rc);
        }
 
        /* No OIs exist, new filesystem, create OI objects */
-       rc = osd_oi_table_open(info, osd, oi, osd_oi_count, true);
-       LASSERT(ergo(rc >= 0, rc == osd_oi_count));
+       rc = osd_oi_table_open(info, osd, oi, count, true);
+       LASSERT(ergo(rc >= 0, rc == count));
 
        GOTO(out, rc);
 
@@ -407,26 +493,35 @@ out:
        if (rc < 0) {
                OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
        } else {
-               LASSERT((rc & (rc - 1)) == 0);
+               LASSERTF((rc & (rc - 1)) == 0, "Invalid OI count %d\n", rc);
+
                osd->od_oi_table = oi;
                osd->od_oi_count = rc;
-               rc = 0;
+               if (sf->sf_oi_count != rc) {
+                       sf->sf_oi_count = rc;
+                       rc = scrub_file_store(info->oti_env, scrub);
+                       if (rc < 0) {
+                               osd_oi_table_put(info, oi, count);
+                               OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
+                       }
+               } else {
+                       rc = 0;
+               }
        }
 
-       mutex_unlock(&oi_init_lock);
        return rc;
 }
 
 void osd_oi_fini(struct osd_thread_info *info, struct osd_device *osd)
 {
-       if (unlikely(osd->od_oi_table == NULL))
+       if (unlikely(!osd->od_oi_table))
                return;
 
-        osd_oi_table_put(info, osd->od_oi_table, osd->od_oi_count);
+       osd_oi_table_put(info, osd->od_oi_table, osd->od_oi_count);
 
-        OBD_FREE(osd->od_oi_table,
-                 sizeof(*(osd->od_oi_table)) * OSD_OI_FID_NR_MAX);
-        osd->od_oi_table = NULL;
+       OBD_FREE(osd->od_oi_table,
+                sizeof(*(osd->od_oi_table)) * OSD_OI_FID_NR_MAX);
+       osd->od_oi_table = NULL;
 }
 
 static inline int fid_is_fs_root(const struct lu_fid *fid)
@@ -470,33 +565,53 @@ static int osd_oi_iam_lookup(struct osd_thread_info *oti,
 }
 
 int fid_is_on_ost(struct osd_thread_info *info, struct osd_device *osd,
-                 const struct lu_fid *fid)
+                 const struct lu_fid *fid, enum oi_check_flags flags)
 {
-       struct lu_seq_range *range = &info->oti_seq_range;
-       int rc;
+       struct lu_seq_range     *range = &info->oti_seq_range;
+       int                     rc;
        ENTRY;
 
+       if (flags & OI_KNOWN_ON_OST)
+               RETURN(1);
+
+       if (unlikely(fid_is_local_file(fid) || fid_is_igif(fid) ||
+                    fid_is_llog(fid)) || fid_is_name_llog(fid) ||
+                    fid_is_quota(fid))
+               RETURN(0);
+
        if (fid_is_idif(fid) || fid_is_last_id(fid))
                RETURN(1);
 
-       rc = osd_fld_lookup(info->oti_env, osd, fid, range);
+       if (!(flags & OI_CHECK_FLD))
+               RETURN(0);
+
+       if (osd_seq_site(osd)->ss_server_fld == NULL)
+               RETURN(0);
+
+       rc = osd_fld_lookup(info->oti_env, osd, fid_seq(fid), range);
        if (rc != 0) {
-               CERROR("%s: Can not lookup fld for "DFID"\n",
-                      osd2lu_dev(osd)->ld_obd->obd_name, PFID(fid));
-               RETURN(rc);
+               /* During upgrade, OST FLDB might not be loaded because
+                * OST FLDB is not created until 2.6, so if some DNE
+                * filesystem upgrade from 2.5 to 2.7/2.8, they will
+                * not be able to find the sequence from local FLDB
+                * cache see fld_index_init(). */
+               if (rc == -ENOENT && osd->od_is_ost)
+                       RETURN(1);
+
+               if (rc != -ENOENT)
+                       CERROR("%s: lookup FLD "DFID": rc = %d\n",
+                              osd_name(osd), PFID(fid), rc);
+               RETURN(0);
        }
 
-       CDEBUG(D_INFO, "fid "DFID" range "DRANGE"\n", PFID(fid),
-              PRANGE(range));
-
-       if (range->lsr_flags == LU_SEQ_RANGE_OST)
+       if (fld_range_is_ost(range))
                RETURN(1);
 
        RETURN(0);
 }
 
-int __osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
-                   const struct lu_fid *fid, struct osd_inode_id *id)
+static int __osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
+                          const struct lu_fid *fid, struct osd_inode_id *id)
 {
        struct lu_fid *oi_fid = &info->oti_fid2;
        int            rc;
@@ -515,134 +630,195 @@ int __osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 
 int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
                  const struct lu_fid *fid, struct osd_inode_id *id,
-                 bool check_fld)
+                 enum oi_check_flags flags)
 {
-       int                  rc = 0;
-
-       if ((!fid_is_last_id(fid) && check_fld &&
-            fid_is_on_ost(info, osd, fid)) ||
-           fid_is_llog(fid)) {
-               /* old OSD obj id */
-               /* FIXME: actually for all of the OST object */
-               rc = osd_obj_map_lookup(info, osd, fid, id);
-       } else if (fid_is_igif(fid)) {
-               osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
-       } else if (fid_is_fs_root(fid)) {
-               osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
-                          osd_sb(osd)->s_root->d_inode->i_generation);
-       } else {
+       if (unlikely(fid_is_last_id(fid)))
+               return osd_obj_spec_lookup(info, osd, fid, id);
+
+       if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
+               return osd_obj_map_lookup(info, osd, fid, id);
+
+
+       if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
+               int rc;
+               if (fid_is_fs_root(fid)) {
+                       osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
+                                  osd_sb(osd)->s_root->d_inode->i_generation);
+                       return 0;
+               }
                if (unlikely(fid_is_acct(fid)))
                        return osd_acct_obj_lookup(info, osd, fid, id);
-               else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE) ||
-                        fid_is_last_id(fid))
-                       return osd_obj_spec_lookup(info, osd, fid, id);
 
+               /* For other special FIDs, try OI first, then do spec lookup */
                rc = __osd_oi_lookup(info, osd, fid, id);
+               if (rc == -ENOENT)
+                       return osd_obj_spec_lookup(info, osd, fid, id);
+               return rc;
        }
-       return rc;
+
+       if (!osd->od_igif_inoi && fid_is_igif(fid)) {
+               osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
+               return 0;
+       }
+
+       return __osd_oi_lookup(info, osd, fid, id);
 }
 
-static int osd_oi_iam_insert(struct osd_thread_info *oti, struct osd_oi *oi,
+static int osd_oi_iam_refresh(struct osd_thread_info *oti, struct osd_oi *oi,
                             const struct dt_rec *rec, const struct dt_key *key,
-                            struct thandle *th)
+                            handle_t *th, bool insert)
 {
-        struct iam_container  *bag;
-        struct iam_rec        *iam_rec = (struct iam_rec *)oti->oti_ldp;
-        struct iam_path_descr *ipd;
-        struct osd_thandle    *oh;
-        int                    rc;
-        ENTRY;
+       struct iam_container    *bag;
+       struct iam_path_descr   *ipd;
+       int                     rc;
+       ENTRY;
 
-        LASSERT(oi);
-        LASSERT(oi->oi_inode);
+       LASSERT(oi);
+       LASSERT(oi->oi_inode);
        ll_vfs_dq_init(oi->oi_inode);
 
-        bag = &oi->oi_dir.od_container;
-        ipd = osd_idx_ipd_get(oti->oti_env, bag);
-        if (unlikely(ipd == NULL))
-                RETURN(-ENOMEM);
+       bag = &oi->oi_dir.od_container;
+       ipd = osd_idx_ipd_get(oti->oti_env, bag);
+       if (unlikely(ipd == NULL))
+               RETURN(-ENOMEM);
 
-        oh = container_of0(th, struct osd_thandle, ot_super);
-        LASSERT(oh->ot_handle != NULL);
-        LASSERT(oh->ot_handle->h_transaction != NULL);
-        if (S_ISDIR(oi->oi_inode->i_mode))
-                osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
-                             &oti->oti_fid);
-        else
-                iam_rec = (struct iam_rec *) rec;
-        rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
-                        iam_rec, ipd);
-        osd_ipd_put(oti->oti_env, bag, ipd);
-        LINVRNT(osd_invariant(obj));
-        RETURN(rc);
+       LASSERT(th != NULL);
+       LASSERT(th->h_transaction != NULL);
+       if (insert)
+               rc = iam_insert(th, bag, (const struct iam_key *)key,
+                               (const struct iam_rec *)rec, ipd);
+       else
+               rc = iam_update(th, bag, (const struct iam_key *)key,
+                               (const struct iam_rec *)rec, ipd);
+       osd_ipd_put(oti->oti_env, bag, ipd);
+       LINVRNT(osd_invariant(obj));
+       RETURN(rc);
 }
 
 int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
                  const struct lu_fid *fid, const struct osd_inode_id *id,
-                 struct thandle *th)
+                 handle_t *th, enum oi_check_flags flags, bool *exist)
 {
        struct lu_fid       *oi_fid = &info->oti_fid2;
-       struct osd_inode_id *oi_id = &info->oti_id2;
+       struct osd_inode_id *oi_id  = &info->oti_id2;
+       int                  rc     = 0;
 
-       if (fid_is_igif(fid) || unlikely(fid_seq(fid) == FID_SEQ_DOT_LUSTRE))
-               return 0;
+       if (unlikely(fid_is_last_id(fid)))
+               return osd_obj_spec_insert(info, osd, fid, id, th);
 
-       if ((fid_is_on_ost(info, osd, fid) && !fid_is_last_id(fid)) ||
-            fid_is_llog(fid))
+       if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
                return osd_obj_map_insert(info, osd, fid, id, th);
 
-       /* Server mount should not depends on OI files */
-       if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE) ||
-           fid_is_last_id(fid))
-               return osd_obj_spec_insert(info, osd, fid, id, th);
-
        fid_cpu_to_be(oi_fid, fid);
        osd_id_pack(oi_id, id);
-       return osd_oi_iam_insert(info, osd_fid2oi(osd, fid),
-                                (const struct dt_rec *)oi_id,
-                                (const struct dt_key *)oi_fid, th);
+       rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
+                              (const struct dt_rec *)oi_id,
+                              (const struct dt_key *)oi_fid, th, true);
+       if (rc != 0) {
+               struct inode *inode;
+               struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
+
+               if (rc != -EEXIST)
+                       return rc;
+
+               rc = osd_oi_lookup(info, osd, fid, oi_id, 0);
+               if (rc != 0)
+                       return rc;
+
+               if (unlikely(osd_id_eq(id, oi_id)))
+                       return 1;
+
+               /* Check whether the mapping for oi_id is valid or not. */
+               inode = osd_iget(info, osd, oi_id);
+               if (IS_ERR(inode)) {
+                       rc = PTR_ERR(inode);
+                       if (rc == -ENOENT || rc == -ESTALE)
+                               goto update;
+                       return rc;
+               }
+
+               /* The EA inode should NOT be in OI, old OI scrub may added
+                * such OI mapping by wrong, replace it. */
+               if (unlikely(osd_is_ea_inode(inode))) {
+                       iput(inode);
+                       goto update;
+               }
+
+               rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
+                                &info->oti_ost_attrs);
+               iput(inode);
+               if (rc == -ENODATA)
+                       goto update;
+
+               if (rc != 0)
+                       return rc;
+
+               if (!(lma->lma_compat & LMAC_NOT_IN_OI) &&
+                   lu_fid_eq(fid, &lma->lma_self_fid)) {
+                       CERROR("%s: the FID "DFID" is used by two objects: "
+                              "%u/%u %u/%u\n", osd_dev2name(osd),
+                              PFID(fid), oi_id->oii_ino, oi_id->oii_gen,
+                              id->oii_ino, id->oii_gen);
+                       return -EEXIST;
+               }
+
+update:
+               osd_id_pack(oi_id, id);
+               rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
+                                       (const struct dt_rec *)oi_id,
+                                       (const struct dt_key *)oi_fid, th, false);
+               if (rc != 0)
+                       return rc;
+
+               if (exist != NULL)
+                       *exist = true;
+       }
+
+       if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
+               rc = osd_obj_spec_insert(info, osd, fid, id, th);
+       return rc;
 }
 
 static int osd_oi_iam_delete(struct osd_thread_info *oti, struct osd_oi *oi,
-                             const struct dt_key *key, struct thandle *handle)
+                            const struct dt_key *key, handle_t *th)
 {
-        struct iam_container  *bag;
-        struct iam_path_descr *ipd;
-        struct osd_thandle    *oh;
-        int                    rc;
-        ENTRY;
+       struct iam_container    *bag;
+       struct iam_path_descr   *ipd;
+       int                      rc;
+       ENTRY;
 
-        LASSERT(oi);
+       LASSERT(oi);
        LASSERT(oi->oi_inode);
        ll_vfs_dq_init(oi->oi_inode);
 
-        bag = &oi->oi_dir.od_container;
-        ipd = osd_idx_ipd_get(oti->oti_env, bag);
-        if (unlikely(ipd == NULL))
-                RETURN(-ENOMEM);
+       bag = &oi->oi_dir.od_container;
+       ipd = osd_idx_ipd_get(oti->oti_env, bag);
+       if (unlikely(ipd == NULL))
+               RETURN(-ENOMEM);
 
-        oh = container_of0(handle, struct osd_thandle, ot_super);
-        LASSERT(oh->ot_handle != NULL);
-        LASSERT(oh->ot_handle->h_transaction != NULL);
+       LASSERT(th != NULL);
+       LASSERT(th->h_transaction != NULL);
 
-        rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
-        osd_ipd_put(oti->oti_env, bag, ipd);
-        LINVRNT(osd_invariant(obj));
-        RETURN(rc);
+       rc = iam_delete(th, bag, (const struct iam_key *)key, ipd);
+       osd_ipd_put(oti->oti_env, bag, ipd);
+       LINVRNT(osd_invariant(obj));
+       RETURN(rc);
 }
 
 int osd_oi_delete(struct osd_thread_info *info,
                  struct osd_device *osd, const struct lu_fid *fid,
-                 struct thandle *th)
+                 handle_t *th, enum oi_check_flags flags)
 {
        struct lu_fid *oi_fid = &info->oti_fid2;
 
-       if (fid_is_igif(fid) || fid_is_last_id(fid))
-               return 0;
+       /* clear idmap cache */
+       if (lu_fid_eq(fid, &info->oti_cache.oic_fid))
+               fid_zero(&info->oti_cache.oic_fid);
 
-       LASSERT(fid_seq(fid) != FID_SEQ_LOCAL_FILE);
+       if (fid_is_last_id(fid))
+               return 0;
 
-       if (fid_is_on_ost(info, osd, fid) || fid_is_llog(fid))
+       if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
                return osd_obj_map_delete(info, osd, fid, th);
 
        fid_cpu_to_be(oi_fid, fid);
@@ -650,17 +826,43 @@ int osd_oi_delete(struct osd_thread_info *info,
                                 (const struct dt_key *)oi_fid, th);
 }
 
+int osd_oi_update(struct osd_thread_info *info, struct osd_device *osd,
+                 const struct lu_fid *fid, const struct osd_inode_id *id,
+                 handle_t *th, enum oi_check_flags flags)
+{
+       struct lu_fid       *oi_fid = &info->oti_fid2;
+       struct osd_inode_id *oi_id  = &info->oti_id2;
+       int                  rc     = 0;
+
+       if (unlikely(fid_is_last_id(fid)))
+               return osd_obj_spec_update(info, osd, fid, id, th);
+
+       if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
+               return osd_obj_map_update(info, osd, fid, id, th);
+
+       fid_cpu_to_be(oi_fid, fid);
+       osd_id_pack(oi_id, id);
+       rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
+                              (const struct dt_rec *)oi_id,
+                              (const struct dt_key *)oi_fid, th, false);
+       if (rc != 0)
+               return rc;
+
+       if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
+               rc = osd_obj_spec_update(info, osd, fid, id, th);
+       return rc;
+}
+
 int osd_oi_mod_init(void)
 {
-        if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
-                osd_oi_count = OSD_OI_FID_NR;
+       if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
+               osd_oi_count = OSD_OI_FID_NR;
 
-        if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
-                LCONSOLE_WARN("Round up oi_count %d to power2 %d\n",
-                              osd_oi_count, size_roundup_power2(osd_oi_count));
-                osd_oi_count = size_roundup_power2(osd_oi_count);
-        }
+       if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
+               LCONSOLE_WARN("Round up oi_count %d to power2 %d\n",
+                             osd_oi_count, size_roundup_power2(osd_oi_count));
+               osd_oi_count = size_roundup_power2(osd_oi_count);
+       }
 
-       mutex_init(&oi_init_lock);
-        return 0;
+       return 0;
 }