X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_oi.c;h=cd1db06d302de1a16f35b3290e3ffc1b686b7745;hp=0918c75be6b09ca69ab9074ccdbeb2935a1fe6c7;hb=2272a301f5e365c0ca02b5c939934c1051d910d3;hpb=4980567857699c7f902ebda336ea98fdc4b83100

diff --git a/lustre/osd-ldiskfs/osd_oi.c b/lustre/osd-ldiskfs/osd_oi.c
index 0918c75..cd1db06 100644
--- a/lustre/osd-ldiskfs/osd_oi.c
+++ b/lustre/osd-ldiskfs/osd_oi.c
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@@ -17,11 +15,7 @@
  *
  * 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
  */
@@ -29,7 +23,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2012, Whamcloud, Inc.
+ * Copyright (c) 2012, 2016, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -42,19 +36,7 @@
  * 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.
- *
- */
-
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
-#define DEBUG_SUBSYSTEM S_MDS
+#define DEBUG_SUBSYSTEM S_OSD
 
 #include <linux/module.h>
 
@@ -69,23 +51,19 @@
 
 /* fid_cpu_to_be() */
 #include <lustre_fid.h>
+#include <dt_object.h>
 
 #include "osd_oi.h"
 /* osd_lookup(), struct osd_thread_info */
 #include "osd_internal.h"
-#include "osd_igif.h"
-#include "dt_object.h"
-
-#define OSD_OI_FID_NR         (1UL << OSD_OI_FID_OID_BITS)
-#define OSD_OI_FID_NR_MAX     (1UL << OSD_OI_FID_OID_BITS_MAX)
+#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.");
+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.");
 
 /** to serialize concurrent OI index initialization */
-static cfs_mutex_t oi_init_lock;
+static struct mutex oi_init_lock;
 
 static struct dt_index_features oi_feat = {
         .dif_flags       = DT_IND_UPDATE,
@@ -97,81 +75,81 @@ static struct dt_index_features oi_feat = {
 #define OSD_OI_NAME_BASE        "oi.16"
 
 static void osd_oi_table_put(struct osd_thread_info *info,
-                             struct osd_oi **oi_table, unsigned oi_count)
+			     struct osd_oi **oi_table, unsigned oi_count)
 {
-        struct iam_container *bag;
-        int                   i;
-
-        for (i = 0; i < oi_count; i++) {
-                LASSERT(oi_table[i] != NULL);
-                LASSERT(oi_table[i]->oi_inode != NULL);
-
-                bag = &(oi_table[i]->oi_dir.od_container);
-                if (bag->ic_object == oi_table[i]->oi_inode)
-                        iam_container_fini(bag);
-                iput(oi_table[i]->oi_inode);
-                oi_table[i]->oi_inode = NULL;
-                OBD_FREE_PTR(oi_table[i]);
-        }
+	struct iam_container *bag;
+	int		      i;
+
+	for (i = 0; i < oi_count; i++) {
+		if (oi_table[i] == NULL)
+			continue;
+
+		LASSERT(oi_table[i]->oi_inode != NULL);
+
+		bag = &(oi_table[i]->oi_dir.od_container);
+		if (bag->ic_object == oi_table[i]->oi_inode)
+			iam_container_fini(bag);
+		iput(oi_table[i]->oi_inode);
+		oi_table[i]->oi_inode = NULL;
+		OBD_FREE_PTR(oi_table[i]);
+		oi_table[i] = NULL;
+	}
 }
 
 static int osd_oi_index_create_one(struct osd_thread_info *info,
-                                   struct osd_device *osd, const char *name,
-                                   struct dt_index_features *feat)
+				   struct osd_device *osd, const char *name,
+				   struct dt_index_features *feat)
 {
-        const struct lu_env             *env = info->oti_env;
-        struct osd_inode_id             *id  = &info->oti_id;
-        struct buffer_head              *bh;
-        struct inode                    *inode;
-        struct ldiskfs_dir_entry_2      *de;
-        struct dentry                   *dentry;
-        struct inode                    *dir;
-        handle_t                        *jh;
-        int                              rc;
-
-        dentry = osd_child_dentry_by_inode(env, osd_sb(osd)->s_root->d_inode,
-                                           name, strlen(name));
-        dir = osd_sb(osd)->s_root->d_inode;
-        bh = osd_ldiskfs_find_entry(dir, dentry, &de, NULL);
-        if (bh) {
-                brelse(bh);
-
-                id->oii_ino = le32_to_cpu(de->inode);
-                id->oii_gen = OSD_OII_NOGEN;
-
-                inode = osd_iget(info, osd, id);
-                if (!IS_ERR(inode)) {
-                        iput(inode);
-                        RETURN(-EEXIST);
-                }
-                RETURN(PTR_ERR(inode));
-        }
-
-        jh = ldiskfs_journal_start_sb(osd_sb(osd), 100);
-        LASSERT(!IS_ERR(jh));
-
-        inode = ldiskfs_create_inode(jh, osd_sb(osd)->s_root->d_inode,
-                                     (S_IFREG | S_IRUGO | S_IWUSR));
-        LASSERT(!IS_ERR(inode));
-
-        if (feat->dif_flags & DT_IND_VARKEY)
-                rc = iam_lvar_create(inode, feat->dif_keysize_max,
-                                     feat->dif_ptrsize, feat->dif_recsize_max,
-                                     jh);
-        else
-                rc = iam_lfix_create(inode, feat->dif_keysize_max,
-                                     feat->dif_ptrsize, feat->dif_recsize_max,
-                                     jh);
-
-        dentry = osd_child_dentry_by_inode(env, osd_sb(osd)->s_root->d_inode,
-                                           name, strlen(name));
-        rc = osd_ldiskfs_add_entry(jh, dentry, inode, NULL);
-        LASSERT(rc == 0);
-
-        ldiskfs_journal_stop(jh);
-        iput(inode);
-
-        return rc;
+	const struct lu_env		*env = info->oti_env;
+	struct osd_inode_id		*id  = &info->oti_id;
+	struct buffer_head		*bh;
+	struct inode			*inode;
+	struct ldiskfs_dir_entry_2	*de;
+	struct dentry			*dentry;
+	struct super_block		*sb  = osd_sb(osd);
+	struct inode			*dir = sb->s_root->d_inode;
+	handle_t			*jh;
+	int				 rc;
+
+	dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
+	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);
+		if (!IS_ERR(inode)) {
+			iput(inode);
+			inode = ERR_PTR(-EEXIST);
+		}
+		return PTR_ERR(inode);
+	}
+
+	jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, 100);
+	if (IS_ERR(jh))
+		return PTR_ERR(jh);
+
+	inode = ldiskfs_create_inode(jh, dir, (S_IFREG | S_IRUGO | S_IWUSR));
+	if (IS_ERR(inode)) {
+		ldiskfs_journal_stop(jh);
+		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,
+				     jh);
+	else
+		rc = iam_lfix_create(inode, feat->dif_keysize_max,
+				     feat->dif_ptrsize, feat->dif_recsize_max,
+				     jh);
+	dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
+	rc = osd_ldiskfs_add_entry(info, osd, jh, dentry, inode, NULL);
+	ldiskfs_journal_stop(jh);
+	iput(inode);
+	return rc;
 }
 
 static struct inode *osd_oi_index_open(struct osd_thread_info *info,
@@ -204,7 +182,7 @@ static struct inode *osd_oi_index_open(struct osd_thread_info *info,
 
         rc = osd_oi_index_create_one(info, osd, name, f);
         if (rc)
-                RETURN(ERR_PTR(rc));
+		return ERR_PTR(rc);
 
         dentry = ll_lookup_one_len(name, osd_sb(osd)->s_root, strlen(name));
         if (IS_ERR(dentry))
@@ -248,6 +226,13 @@ static int osd_oi_open(struct osd_thread_info *info, struct osd_device *osd,
         if (IS_ERR(inode))
                 RETURN(PTR_ERR(inode));
 
+	/* '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 != 0)
+		GOTO(out_inode, rc);
+
         OBD_ALLOC_PTR(oi);
         if (oi == NULL)
                 GOTO(out_inode, rc = -ENOMEM);
@@ -295,93 +280,216 @@ out_inode:
  */
 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 osd_oi **oi_table, unsigned oi_count, bool create)
 {
-        struct dt_device *dev = &osd->od_dt_dev;
-        int               count = 0;
-        int               rc = 0;
-        int               i;
-
-        /* NB: oi_count != 0 means that we have already created/known all OIs
-         * and have known exact number of OIs. */
-        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];
-
-                sprintf(name, "%s.%d", OSD_OI_NAME_BASE, i);
-                rc = osd_oi_open(info, osd, name, &oi_table[i], create);
-                if (rc == 0) {
-                        count++;
-                        continue;
-                }
-
-                if (rc == -ENOENT && oi_count == 0)
-                        return count;
-
-                CERROR("%s: can't open %s: rc = %d\n",
-                       dev->dd_lu_dev.ld_obd->obd_name, name, rc);
-                if (oi_count > 0) {
-                        CERROR("%s: expect to open total %d OI files.\n",
-                               dev->dd_lu_dev.ld_obd->obd_name, oi_count);
-                }
-                break;
-        }
-
-        if (rc < 0) {
-                osd_oi_table_put(info, oi_table, count);
-                return rc;
-        }
+	struct scrub_file *sf = &osd->od_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
+	 * and have known exact number of OIs. */
+	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];
+
+		if (oi_table[i] != NULL) {
+			count++;
+			continue;
+		}
+
+		sprintf(name, "%s.%d", OSD_OI_NAME_BASE, i);
+		rc = osd_oi_open(info, osd, name, &oi_table[i], create);
+		if (rc == 0) {
+			count++;
+			continue;
+		}
+
+		if (rc == -ENOENT && create == false) {
+			if (oi_count == 0)
+				return count;
+
+			rc = 0;
+			ldiskfs_set_bit(i, sf->sf_oi_bitmap);
+			continue;
+		}
+
+		CERROR("%s: can't open %s: rc = %d\n",
+		       osd_dev2name(osd), name, rc);
+		if (oi_count > 0)
+			CERROR("%s: expect to open total %d OI files.\n",
+			       osd_dev2name(osd), oi_count);
+		break;
+	}
+
+	if (rc < 0) {
+		osd_oi_table_put(info, oi_table, oi_count > 0 ? oi_count : i);
+		count = rc;
+	}
+
+	RETURN(count);
+}
 
-        return count;
+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;
 }
 
-int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd)
+static int osd_remove_ois(struct osd_thread_info *info, struct osd_device *osd)
 {
-        struct dt_device *dev = &osd->od_dt_dev;
-        struct osd_oi   **oi;
-        int               rc;
+	char name[16];
+	int namelen;
+	int rc;
+	int i;
+
+	for (i = 0; i < osd->od_scrub.os_file.sf_oi_count; 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;
+}
 
-        OBD_ALLOC(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
-        if (oi == NULL)
-                return -ENOMEM;
-
-        cfs_mutex_lock(&oi_init_lock);
-        /* try to open existing multiple OIs first */
-        rc = osd_oi_table_open(info, osd, oi, 0, false);
-        if (rc != 0)
-                goto out;
-
-        /* 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 */
-                rc = 1;
-                goto out;
-        } else if (rc != -ENOENT) {
-                CERROR("%s: can't open %s: rc = %d\n",
-                       dev->dd_lu_dev.ld_obd->obd_name, OSD_OI_NAME_BASE, rc);
-                goto out;
-        }
+int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd,
+		bool restored)
+{
+	struct osd_scrub  *scrub = &osd->od_scrub;
+	struct scrub_file *sf = &scrub->os_file;
+	struct osd_oi    **oi;
+	int		   rc;
+	ENTRY;
+
+	if (restored) {
+		rc = osd_remove_ois(info, osd);
+		if (rc != 0)
+			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);
+
+		osd_scrub_file_reset(scrub,
+				     LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
+				     SF_RECREATED);
+		osd_oi_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 */
+		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. */
+			osd_scrub_file_reset(scrub,
+					LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
+					SF_UPGRADE);
+		GOTO(out, rc = 1);
+	} else if (rc != -ENOENT) {
+		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;
+
+		memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
+		for (i = 0; i < osd_oi_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);
+	}
+	sf->sf_oi_count = osd_oi_count;
+
+create:
+	rc = osd_scrub_file_store(scrub);
+	if (rc < 0) {
+		osd_oi_table_put(info, oi, sf->sf_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));
+
+	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));
 out:
-        if (rc < 0) {
-                OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
-        } else {
-                LASSERT((rc & (rc - 1)) == 0);
-                osd->od_oi_table = oi;
-                osd->od_oi_count = rc;
-                rc = 0;
-        }
-
-        cfs_mutex_unlock(&oi_init_lock);
-        return rc;
+	if (rc < 0) {
+		OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
+	} else {
+		LASSERT((rc & (rc - 1)) == 0);
+		osd->od_oi_table = oi;
+		osd->od_oi_count = rc;
+		if (sf->sf_oi_count != rc) {
+			sf->sf_oi_count = rc;
+			rc = osd_scrub_file_store(scrub);
+			if (rc < 0) {
+				osd_oi_table_put(info, oi, sf->sf_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))
+		return;
+
         osd_oi_table_put(info, osd->od_oi_table, osd->od_oi_count);
 
         OBD_FREE(osd->od_oi_table,
@@ -402,7 +510,6 @@ static int osd_oi_iam_lookup(struct osd_thread_info *oti,
 {
         struct iam_container  *bag;
         struct iam_iterator   *it = &oti->oti_idx_it;
-        struct iam_rec        *iam_rec;
         struct iam_path_descr *ipd;
         int                    rc;
         ENTRY;
@@ -419,17 +526,8 @@ static int osd_oi_iam_lookup(struct osd_thread_info *oti,
         iam_it_init(it, bag, 0, ipd);
 
         rc = iam_it_get(it, (struct iam_key *)key);
-        if (rc >= 0) {
-                if (S_ISDIR(oi->oi_inode->i_mode))
-                        iam_rec = (struct iam_rec *)oti->oti_ldp;
-                else
-                        iam_rec = (struct iam_rec *)rec;
-
-                iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
-                if (S_ISDIR(oi->oi_inode->i_mode))
-                        osd_fid_unpack((struct lu_fid *)rec,
-                                       (struct osd_fid_pack *)iam_rec);
-        }
+	if (rc > 0)
+		iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)rec);
         iam_it_put(it);
         iam_it_fini(it);
         osd_ipd_put(oti->oti_env, bag, ipd);
@@ -439,173 +537,296 @@ static int osd_oi_iam_lookup(struct osd_thread_info *oti,
         RETURN(rc);
 }
 
-int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
-                  const struct lu_fid *fid, struct osd_inode_id *id)
+int fid_is_on_ost(struct osd_thread_info *info, struct osd_device *osd,
+		  const struct lu_fid *fid, enum oi_check_flags flags)
 {
-        struct lu_fid       *oi_fid = &info->oti_fid;
-        const struct dt_key *key;
-        int                  rc = 0;
-
-        if (fid_is_idif(fid) || fid_seq(fid) == FID_SEQ_LLOG) {
-                /* old OSD obj id */
-                rc = osd_compat_objid_lookup(info, osd, fid, id);
-        } else if (fid_is_igif(fid)) {
-                lu_igif_to_id(fid, id);
-                rc = 0;
-        } else if (fid_is_fs_root(fid)) {
-                struct inode *inode = osd_sb(osd)->s_root->d_inode;
-
-                id->oii_ino = inode->i_ino;
-                id->oii_gen = inode->i_generation;
-        } else {
-                if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
-                        rc = osd_compat_spec_lookup(info, osd, fid, id);
-                        if (rc == 0 || rc != -ERESTART)
-                                goto out;
-                }
-
-                fid_cpu_to_be(oi_fid, fid);
-                key = (struct dt_key *)oi_fid;
-
-                rc = osd_oi_iam_lookup(info, osd_fid2oi(osd, fid),
-                                       (struct dt_rec *)id, key);
-
-                if (rc > 0) {
-                        id->oii_ino = be32_to_cpu(id->oii_ino);
-                        id->oii_gen = be32_to_cpu(id->oii_gen);
-                        rc = 0;
-                } else if (rc == 0) {
-                        rc = -ENOENT;
-                }
-        }
-
-out:
-        return 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);
+
+	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) {
+		/* 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);
+	}
+
+	if (fld_range_is_ost(range))
+		RETURN(1);
+
+	RETURN(0);
 }
 
-static int osd_oi_iam_insert(struct osd_thread_info *oti, struct osd_oi *oi,
-                             const struct dt_rec *rec, const struct dt_key *key,
-                             struct thandle *th, int ignore_quota)
+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 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;
-#ifdef HAVE_QUOTA_SUPPORT
-        cfs_cap_t              save    = cfs_curproc_cap_pack();
-#endif
-        ENTRY;
-
-        LASSERT(oi);
-        LASSERT(oi->oi_inode);
+	struct lu_fid *oi_fid = &info->oti_fid2;
+	int	       rc;
+
+	fid_cpu_to_be(oi_fid, fid);
+	rc = osd_oi_iam_lookup(info, osd_fid2oi(osd, fid), (struct dt_rec *)id,
+			       (const struct dt_key *)oi_fid);
+	if (rc > 0) {
+		osd_id_unpack(id, id);
+		rc = 0;
+	} else if (rc == 0) {
+		rc = -ENOENT;
+	}
+	return rc;
+}
 
-        bag = &oi->oi_dir.od_container;
-        ipd = osd_idx_ipd_get(oti->oti_env, bag);
-        if (unlikely(ipd == NULL))
-                RETURN(-ENOMEM);
+int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
+		  const struct lu_fid *fid, struct osd_inode_id *id,
+		  enum oi_check_flags flags)
+{
+	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);
+
+		/* 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;
+	}
+
+	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);
+}
 
-        oh = container_of0(th, struct osd_thandle, ot_super);
-        LASSERT(oh->ot_handle != NULL);
-        LASSERT(oh->ot_handle->h_transaction != NULL);
-#ifdef HAVE_QUOTA_SUPPORT
-        if (ignore_quota)
-                cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
-        else
-                cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
-#endif
-        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);
-#ifdef HAVE_QUOTA_SUPPORT
-        cfs_curproc_cap_unpack(save);
-#endif
-        osd_ipd_put(oti->oti_env, bag, ipd);
-        LINVRNT(osd_invariant(obj));
-        RETURN(rc);
+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,
+			     handle_t *th, bool insert)
+{
+	struct iam_container	*bag;
+	struct iam_path_descr	*ipd;
+	int			rc;
+	ENTRY;
+
+	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);
+
+	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 *id0,
-                  struct thandle *th, int ignore_quota)
+		  const struct lu_fid *fid, const struct osd_inode_id *id,
+		  handle_t *th, enum oi_check_flags flags, bool *exist)
 {
-        struct lu_fid       *oi_fid = &info->oti_fid;
-        struct osd_inode_id *id;
-        const struct dt_key *key;
-
-        if (fid_is_igif(fid))
-                return 0;
-
-        if (fid_is_idif(fid) || fid_seq(fid) == FID_SEQ_LLOG)
-                return osd_compat_objid_insert(info, osd, fid, id0, th);
-
-        /* notice we don't return immediately, but continue to get into OI */
-        if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
-                osd_compat_spec_insert(info, osd, fid, id0, th);
-
-        fid_cpu_to_be(oi_fid, fid);
-        key = (struct dt_key *)oi_fid;
-
-        id  = &info->oti_id;
-        id->oii_ino = cpu_to_be32(id0->oii_ino);
-        id->oii_gen = cpu_to_be32(id0->oii_gen);
-
-        return osd_oi_iam_insert(info, osd_fid2oi(osd, fid),
-                                 (struct dt_rec *)id, key, th, ignore_quota);
+	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_insert(info, osd, fid, id, th);
+
+	if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
+		return osd_obj_map_insert(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, 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;
-
-        LASSERT(oi);
-
-        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);
-
-        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);
+	struct iam_container	*bag;
+	struct iam_path_descr	*ipd;
+	int			 rc;
+	ENTRY;
+
+	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);
+
+	LASSERT(th != NULL);
+	LASSERT(th->h_transaction != NULL);
+
+	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)
+		  struct osd_device *osd, const struct lu_fid *fid,
+		  handle_t *th, enum oi_check_flags flags)
 {
-        struct lu_fid       *oi_fid = &info->oti_fid;
-        const struct dt_key *key;
+	struct lu_fid *oi_fid = &info->oti_fid2;
 
-        if (!fid_is_norm(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_idif(fid) || fid_seq(fid) == FID_SEQ_LLOG)
-                return osd_compat_objid_delete(info, osd, fid, th);
+	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);
-        key = (struct dt_key *)oi_fid;
+	fid_cpu_to_be(oi_fid, fid);
+	return osd_oi_iam_delete(info, osd_fid2oi(osd, fid),
+				 (const struct dt_key *)oi_fid, th);
+}
 
-        return osd_oi_iam_delete(info, osd_fid2oi(osd, fid), key, 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()
+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;
@@ -616,6 +837,6 @@ int osd_oi_mod_init()
                 osd_oi_count = size_roundup_power2(osd_oi_count);
         }
 
-        cfs_mutex_init(&oi_init_lock);
+	mutex_init(&oi_init_lock);
         return 0;
 }