[fs/lustre-release.git] / lustre / mdd / mdd_object.c

/*
 * GPL HEADER START
 *
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 only,
 * as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License version 2 for more details (a copy is included
 * in the LICENSE file that accompanied this code).
 *
 * You should have received a copy of the GNU General Public License
 * version 2 along with this program; If not, see
 * http://www.gnu.org/licenses/gpl-2.0.html
 *
 * GPL HEADER END
 */
/*
 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
 * Use is subject to license terms.
 *
 * Copyright (c) 2011, 2017, Intel Corporation.
 */
/*
 * This file is part of Lustre, http://www.lustre.org/
 * Lustre is a trademark of Sun Microsystems, Inc.
 *
 * lustre/mdd/mdd_object.c
 *
 * Lustre Metadata Server (mdd) routines
 *
 * Author: Wang Di <wangdi@clusterfs.com>
 */

#define DEBUG_SUBSYSTEM S_MDS

#include <linux/module.h>
#include <obd.h>
#include <obd_class.h>
#include <obd_support.h>
#include <lprocfs_status.h>
/* fid_be_cpu(), fid_cpu_to_be(). */
#include <lustre_fid.h>
#include <lustre_idmap.h>
#include <uapi/linux/lustre/lustre_param.h>
#include <lustre_mds.h>

#include "mdd_internal.h"

static const struct lu_object_operations mdd_lu_obj_ops;

static int mdd_xattr_get(const struct lu_env *env,
                         struct md_object *obj, struct lu_buf *buf,
                         const char *name);

int mdd_la_get(const struct lu_env *env, struct mdd_object *obj,
               struct lu_attr *la)
{
        int rc;

        if (mdd_object_exists(obj) == 0)
                return -ENOENT;

        rc = mdo_attr_get(env, obj, la);
        if (unlikely(rc != 0)) {
                if (rc == -ENOENT)
                        obj->mod_flags |= DEAD_OBJ;
                return rc;
        }

        if (la->la_valid & LA_FLAGS &&
            la->la_flags & LUSTRE_ORPHAN_FL)
                obj->mod_flags |= ORPHAN_OBJ | DEAD_OBJ;

        return 0;
}

struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
{
        struct mdd_thread_info *info;

        lu_env_refill((struct lu_env *)env);
        info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
        LASSERT(info != NULL);
        return info;
}

struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
{
        struct lu_buf *buf;

        buf = &mdd_env_info(env)->mti_buf[0];
        buf->lb_buf = area;
        buf->lb_len = len;
        return buf;
}

const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
                                       const void *area, ssize_t len)
{
        struct lu_buf *buf;

        buf = &mdd_env_info(env)->mti_buf[0];
        buf->lb_buf = (void *)area;
        buf->lb_len = len;
        return buf;
}

struct lu_object *mdd_object_alloc(const struct lu_env *env,
                                   const struct lu_object_header *hdr,
                                   struct lu_device *d)
{
        struct mdd_object *mdd_obj;

        OBD_SLAB_ALLOC_PTR_GFP(mdd_obj, mdd_object_kmem, GFP_NOFS);
        if (mdd_obj != NULL) {
                struct lu_object *o;

                o = mdd2lu_obj(mdd_obj);
                lu_object_init(o, NULL, d);
                mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
                mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
                mdd_obj->mod_count = 0;
                o->lo_ops = &mdd_lu_obj_ops;
                return o;
        } else {
                return NULL;
        }
}

static int mdd_object_init(const struct lu_env *env, struct lu_object *o,
                           const struct lu_object_conf *unused)
{
        struct mdd_device *d = lu2mdd_dev(o->lo_dev);
        struct mdd_object *mdd_obj = lu2mdd_obj(o);
        struct lu_object  *below;
        struct lu_device  *under;
        ENTRY;

        mdd_obj->mod_cltime = ktime_set(0, 0);
        under = &d->mdd_child->dd_lu_dev;
        below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
        if (IS_ERR(below))
                RETURN(PTR_ERR(below));

        lu_object_add(o, below);

        RETURN(0);
}

static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
{
        int rc = 0;

        if (lu_object_exists(o)) {
                struct mdd_object *mdd_obj = lu2mdd_obj(o);
                struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);

                rc = mdd_la_get(env, mdd_obj, attr);
        }

        return rc;
}

static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
{
        struct mdd_object *mdd = lu2mdd_obj(o);

        lu_object_fini(o);
        OBD_SLAB_FREE_PTR(mdd, mdd_object_kmem);
}

static int mdd_object_print(const struct lu_env *env, void *cookie,
                            lu_printer_t p, const struct lu_object *o)
{
        struct mdd_object *mdd = lu2mdd_obj((struct lu_object *)o);

        return (*p)(env, cookie,
                    LUSTRE_MDD_NAME"-object@%p(open_count=%d, valid=%x, cltime=%lldns, flags=%lx)",
                    mdd, mdd->mod_count, mdd->mod_valid,
                    ktime_to_ns(mdd->mod_cltime), mdd->mod_flags);
}

static const struct lu_object_operations mdd_lu_obj_ops = {
        .loo_object_init    = mdd_object_init,
        .loo_object_start   = mdd_object_start,
        .loo_object_free    = mdd_object_free,
        .loo_object_print   = mdd_object_print,
};

struct mdd_object *mdd_object_find(const struct lu_env *env,
                                   struct mdd_device *d,
                                   const struct lu_fid *f)
{
        return md2mdd_obj(md_object_find_slice(env, &d->mdd_md_dev, f));
}

/*
 * No permission check is needed.
 */
int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
                 struct md_attr *ma)
{
        struct mdd_object *mdd_obj = md2mdd_obj(obj);
        int               rc;

        ENTRY;

        rc = mdd_la_get(env, mdd_obj, &ma->ma_attr);
        if ((ma->ma_need & MA_INODE) != 0 && mdd_is_dead_obj(mdd_obj))
                ma->ma_attr.la_nlink = 0;

        RETURN(rc);
}

/*
 * No permission check is needed.
 */
static int mdd_xattr_get(const struct lu_env *env,
                         struct md_object *obj, struct lu_buf *buf,
                         const char *name)
{
        struct mdd_object *mdd_obj = md2mdd_obj(obj);
        int rc;

        ENTRY;

        if (mdd_object_exists(mdd_obj) == 0) {
                CERROR("%s: object "DFID" not found: rc = -2\n",
                       mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
                return -ENOENT;
        }

        /* If the object has been destroyed, then do not get LMVEA, because
         * it needs to load stripes from the iteration of the master object,
         * and it will cause problem if master object has been destroyed, see
         * LU-6427 */
        if (unlikely((mdd_obj->mod_flags & DEAD_OBJ) &&
                     !(mdd_obj->mod_flags & ORPHAN_OBJ) &&
                      strcmp(name, XATTR_NAME_LMV) == 0))
                RETURN(-ENOENT);

        /* If the object has been delete from the namespace, then
         * get linkEA should return -ENOENT as well */
        if (unlikely((mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ)) &&
                      strcmp(name, XATTR_NAME_LINK) == 0))
                RETURN(-ENOENT);

        mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
        rc = mdo_xattr_get(env, mdd_obj, buf, name);
        mdd_read_unlock(env, mdd_obj);

        RETURN(rc);
}

/*
 * Permission check is done when open,
 * no need check again.
 */
int mdd_readlink(const struct lu_env *env, struct md_object *obj,
                 struct lu_buf *buf)
{
        struct mdd_object *mdd_obj = md2mdd_obj(obj);
        struct dt_object  *next;
        loff_t             pos = 0;
        int                rc;
        ENTRY;

        if (mdd_object_exists(mdd_obj) == 0) {
                CERROR("%s: object "DFID" not found: rc = -2\n",
                       mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
                return -ENOENT;
        }

        next = mdd_object_child(mdd_obj);
        LASSERT(next != NULL);
        LASSERT(next->do_body_ops != NULL);
        LASSERT(next->do_body_ops->dbo_read != NULL);
        mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
        rc = dt_read(env, next, buf, &pos);
        mdd_read_unlock(env, mdd_obj);
        RETURN(rc);
}

/*
 * No permission check is needed.
 */
static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
                          struct lu_buf *buf)
{
        struct mdd_object *mdd_obj = md2mdd_obj(obj);
        int rc;

        ENTRY;

        mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
        rc = mdo_xattr_list(env, mdd_obj, buf);
        mdd_read_unlock(env, mdd_obj);

        /* If the buffer is NULL then we are only here to get the
         * length of the xattr name list. */
        if (rc < 0 || buf->lb_buf == NULL)
                RETURN(rc);

        /*
         * Filter out XATTR_NAME_LINK if this is an orphan object.  See
         * mdd_xattr_get().
         */
        if (unlikely(mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ))) {
                char   *end = (char *)buf->lb_buf + rc;
                char   *p = buf->lb_buf;

                while (p < end) {
                        char   *next = p + strlen(p) + 1;

                        if (strcmp(p, XATTR_NAME_LINK) == 0) {
                                if (end - next > 0)
                                        memmove(p, next, end - next);
                                rc -= next - p;
                                CDEBUG(D_INFO, "Filtered out "XATTR_NAME_LINK
                                       " of orphan "DFID"\n",
                                       PFID(mdd_object_fid(mdd_obj)));
                                break;
                        }

                        p = next;
                }
        }

        RETURN(rc);
}

int mdd_invalidate(const struct lu_env *env, struct md_object *obj)
{
        return mdo_invalidate(env, md2mdd_obj(obj));
}

int mdd_declare_create_object_internal(const struct lu_env *env,
                                       struct mdd_object *p,
                                       struct mdd_object *c,
                                       struct lu_attr *attr,
                                       struct thandle *handle,
                                       const struct md_op_spec *spec,
                                       struct dt_allocation_hint *hint)
{
        struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
        const struct dt_index_features *feat = spec->sp_feat;
        int rc;
        ENTRY;

        if (feat != &dt_directory_features && feat != NULL) {
                dof->dof_type = DFT_INDEX;
                dof->u.dof_idx.di_feat = feat;
        } else {
                dof->dof_type = dt_mode_to_dft(attr->la_mode);
                if (dof->dof_type == DFT_REGULAR) {
                        dof->u.dof_reg.striped =
                                md_should_create(spec->sp_cr_flags);
                        if (spec->sp_cr_flags & MDS_OPEN_HAS_EA)
                                dof->u.dof_reg.striped = 0;
                        /* is this replay? */
                        if (spec->no_create)
                                dof->u.dof_reg.striped = 0;
                }
        }

        rc = mdo_declare_create_object(env, c, attr, hint, dof, handle);

        RETURN(rc);
}

int mdd_create_object_internal(const struct lu_env *env, struct mdd_object *p,
                               struct mdd_object *c, struct lu_attr *attr,
                               struct thandle *handle,
                               const struct md_op_spec *spec,
                               struct dt_allocation_hint *hint)
{
        struct dt_object_format *dof = &mdd_env_info(env)->mti_dof;
        int rc;
        ENTRY;

        LASSERT(!mdd_object_exists(c));

        rc = mdo_create_object(env, c, attr, hint, dof, handle);

        RETURN(rc);
}

int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj,
                          const struct lu_attr *attr, struct thandle *handle,
                          int needacl)
{
        int rc;
        ENTRY;

        rc = mdo_attr_set(env, obj, attr, handle);
#ifdef CONFIG_FS_POSIX_ACL
        if (!rc && (attr->la_valid & LA_MODE) && needacl)
                rc = mdd_acl_chmod(env, obj, attr->la_mode, handle);
#endif
        RETURN(rc);
}

int mdd_update_time(const struct lu_env *env, struct mdd_object *obj,
                    const struct lu_attr *oattr, struct lu_attr *attr,
                    struct thandle *handle)
{
        int rc = 0;
        ENTRY;

        LASSERT(attr->la_valid & LA_CTIME);
        LASSERT(oattr != NULL);

        /* Make sure the ctime is increased only, however, it's not strictly
         * reliable at here because there is not guarantee to hold lock on
         * object, so we just bypass some unnecessary cmtime setting first
         * and OSD has to check it again. */
        if (attr->la_ctime < oattr->la_ctime)
                attr->la_valid &= ~(LA_MTIME | LA_CTIME);
        else if (attr->la_valid == LA_CTIME &&
                 attr->la_ctime == oattr->la_ctime)
                attr->la_valid &= ~LA_CTIME;

        if (attr->la_valid != 0)
                rc = mdd_attr_set_internal(env, obj, attr, handle, 0);
        RETURN(rc);
}

/*
 * This gives the same functionality as the code between
 * sys_chmod and inode_setattr
 * chown_common and inode_setattr
 * utimes and inode_setattr
 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
 */
static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
                        const struct lu_attr *oattr, struct lu_attr *la,
                        const unsigned long flags)
{
        struct lu_ucred  *uc;
        int               rc = 0;
        ENTRY;

        if (!la->la_valid)
                RETURN(0);

        /* Do not permit change file type */
        if (la->la_valid & LA_TYPE)
                RETURN(-EPERM);

        /* They should not be processed by setattr */
        if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
                RETURN(-EPERM);

        LASSERT(oattr != NULL);

        uc = lu_ucred_check(env);
        if (uc == NULL)
                RETURN(0);

        if (la->la_valid == LA_CTIME) {
                if (!(flags & MDS_PERM_BYPASS))
                        /* This is only for set ctime when rename's source is
                         * on remote MDS. */
                        rc = mdd_may_delete(env, NULL, NULL, obj, oattr, NULL,
                                            1, 0);
                if (rc == 0 && la->la_ctime <= oattr->la_ctime)
                        la->la_valid &= ~LA_CTIME;
                RETURN(rc);
        }

        if (la->la_valid == LA_ATIME) {
                /* This is an atime-only attribute update for close RPCs. */
                if (la->la_atime < (oattr->la_atime +
                                mdd_obj2mdd_dev(obj)->mdd_atime_diff))
                        la->la_valid &= ~LA_ATIME;
                RETURN(0);
        }

        /* Check if flags change. */
        if (la->la_valid & LA_FLAGS) {
                unsigned int oldflags = oattr->la_flags &
                                (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);
                unsigned int newflags = la->la_flags &
                                (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL);

                if ((uc->uc_fsuid != oattr->la_uid) &&
                    !md_capable(uc, CFS_CAP_FOWNER))
                        RETURN(-EPERM);

                /* The IMMUTABLE and APPEND_ONLY flags can
                 * only be changed by the relevant capability. */
                if ((oldflags ^ newflags) &&
                    !md_capable(uc, CFS_CAP_LINUX_IMMUTABLE))
                        RETURN(-EPERM);

                if (!S_ISDIR(oattr->la_mode))
                        la->la_flags &= ~(LUSTRE_DIRSYNC_FL | LUSTRE_TOPDIR_FL);
        }

        if (oattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL) &&
            (la->la_valid & ~LA_FLAGS) &&
            !(flags & MDS_PERM_BYPASS))
                RETURN(-EPERM);

        /* Check for setting the obj time. */
        if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
            !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
                if ((uc->uc_fsuid != oattr->la_uid) &&
                    !md_capable(uc, CFS_CAP_FOWNER)) {
                        rc = mdd_permission_internal(env, obj, oattr,
                                                     MAY_WRITE);
                        if (rc)
                                RETURN(rc);
                }
        }

        if (la->la_valid & LA_KILL_SUID) {
                la->la_valid &= ~LA_KILL_SUID;
                if ((oattr->la_mode & S_ISUID) &&
                    !(la->la_valid & LA_MODE)) {
                        la->la_mode = oattr->la_mode;
                        la->la_valid |= LA_MODE;
                }
                la->la_mode &= ~S_ISUID;
        }

        if (la->la_valid & LA_KILL_SGID) {
                la->la_valid &= ~LA_KILL_SGID;
                if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
                        (S_ISGID | S_IXGRP)) &&
                    !(la->la_valid & LA_MODE)) {
                        la->la_mode = oattr->la_mode;
                        la->la_valid |= LA_MODE;
                }
                la->la_mode &= ~S_ISGID;
        }

        /* Make sure a caller can chmod. */
        if (la->la_valid & LA_MODE) {
                if (!(flags & MDS_PERM_BYPASS) &&
                    (uc->uc_fsuid != oattr->la_uid) &&
                    !md_capable(uc, CFS_CAP_FOWNER))
                        RETURN(-EPERM);

                if (la->la_mode == (umode_t) -1)
                        la->la_mode = oattr->la_mode;
                else
                        la->la_mode = (la->la_mode & S_IALLUGO) |
                                        (oattr->la_mode & ~S_IALLUGO);

                /* Also check the setgid bit! */
                if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ?
                                       la->la_gid : oattr->la_gid) &&
                    !md_capable(uc, CFS_CAP_FSETID))
                        la->la_mode &= ~S_ISGID;
        } else {
               la->la_mode = oattr->la_mode;
        }

        /* Make sure a caller can chown. */
        if (la->la_valid & LA_UID) {
                if (la->la_uid == (uid_t) -1)
                        la->la_uid = oattr->la_uid;
                if (((uc->uc_fsuid != oattr->la_uid) ||
                     (la->la_uid != oattr->la_uid)) &&
                    !md_capable(uc, CFS_CAP_CHOWN))
                        RETURN(-EPERM);

                /* If the user or group of a non-directory has been
                 * changed by a non-root user, remove the setuid bit.
                 * 19981026 David C Niemi <niemi@tux.org>
                 *
                 * Changed this to apply to all users, including root,
                 * to avoid some races. This is the behavior we had in
                 * 2.0. The check for non-root was definitely wrong
                 * for 2.2 anyway, as it should have been using
                 * CAP_FSETID rather than fsuid -- 19990830 SD. */
                if (((oattr->la_mode & S_ISUID) == S_ISUID) &&
                !S_ISDIR(oattr->la_mode)) {
                        la->la_mode &= ~S_ISUID;
                        la->la_valid |= LA_MODE;
                }
        }

        /* Make sure caller can chgrp. */
        if (la->la_valid & LA_GID) {
                if (la->la_gid == (gid_t) -1)
                        la->la_gid = oattr->la_gid;
                if (((uc->uc_fsuid != oattr->la_uid) ||
                     ((la->la_gid != oattr->la_gid) &&
                      !lustre_in_group_p(uc, la->la_gid))) &&
                    !md_capable(uc, CFS_CAP_CHOWN))
                        RETURN(-EPERM);

                /* Likewise, if the user or group of a non-directory
                 * has been changed by a non-root user, remove the
                 * setgid bit UNLESS there is no group execute bit
                 * (this would be a file marked for mandatory
                 * locking).  19981026 David C Niemi <niemi@tux.org>
                 *
                 * Removed the fsuid check (see the comment above) --
                 * 19990830 SD. */
                if (((oattr->la_mode & (S_ISGID | S_IXGRP)) ==
                    (S_ISGID | S_IXGRP)) && !S_ISDIR(oattr->la_mode)) {
                        la->la_mode &= ~S_ISGID;
                        la->la_valid |= LA_MODE;
                }
        }

        if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
                if (!((flags & MDS_OWNEROVERRIDE) &&
                      (uc->uc_fsuid == oattr->la_uid)) &&
                    !(flags & MDS_PERM_BYPASS)) {
                        rc = mdd_permission_internal(env, obj, oattr,
                                                     MAY_WRITE);
                        if (rc != 0)
                                RETURN(rc);
                }
        }

        if (la->la_valid & LA_CTIME) {
                /* The pure setattr, it has the priority over what is
                 * already set, do not drop it if ctime is equal. */
                if (la->la_ctime < oattr->la_ctime)
                        la->la_valid &= ~(LA_ATIME | LA_MTIME | LA_CTIME);
        }

        RETURN(0);
}

static int mdd_changelog_data_store_by_fid(const struct lu_env *env,
                                    struct mdd_device *mdd,
                                    enum changelog_rec_type type, int flags,
                                    const struct lu_fid *fid,
                                    struct thandle *handle)
{
        const struct lu_ucred *uc = lu_ucred(env);
        struct llog_changelog_rec *rec;
        struct lu_buf *buf;
        int reclen;
        int xflags = CLFE_INVALID;
        int rc;

        flags = (flags & CLF_FLAGMASK) | CLF_VERSION | CLF_EXTRA_FLAGS;

        if (uc) {
                if (uc->uc_jobid[0] != '\0')
                        flags |= CLF_JOBID;
                xflags |= CLFE_UIDGID;
                xflags |= CLFE_NID;
        }
        if (type == CL_OPEN)
                xflags |= CLFE_OPEN;

        reclen = llog_data_len(LLOG_CHANGELOG_HDR_SZ +
                               changelog_rec_offset(flags & CLF_SUPPORTED,
                                                    xflags & CLFE_SUPPORTED));
        buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen);
        if (buf->lb_buf == NULL)
                RETURN(-ENOMEM);
        rec = buf->lb_buf;

        rec->cr_hdr.lrh_len = reclen;
        rec->cr.cr_flags = flags;
        rec->cr.cr_type = (__u32)type;
        rec->cr.cr_tfid = *fid;
        rec->cr.cr_namelen = 0;

        if (flags & CLF_JOBID)
                mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);

        if (flags & CLF_EXTRA_FLAGS) {
                mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
                if (xflags & CLFE_UIDGID)
                        mdd_changelog_rec_extra_uidgid(&rec->cr,
                                                       uc->uc_uid, uc->uc_gid);
                if (xflags & CLFE_NID)
                        mdd_changelog_rec_extra_nid(&rec->cr, uc->uc_nid);
                if (xflags & CLFE_OPEN)
                        mdd_changelog_rec_extra_omode(&rec->cr, flags);
        }

        rc = mdd_changelog_store(env, mdd, rec, handle);
        RETURN(rc);
}


/** Store a data change changelog record
 * If this fails, we must fail the whole transaction; we don't
 * want the change to commit without the log entry.
 * \param mdd_obj - mdd_object of change
 * \param handle - transaction handle
 */
int mdd_changelog_data_store(const struct lu_env *env, struct mdd_device *mdd,
                             enum changelog_rec_type type, int flags,
                             struct mdd_object *mdd_obj, struct thandle *handle)
{
        int                              rc;

        LASSERT(mdd_obj != NULL);
        LASSERT(handle != NULL);

        /* Not recording */
        if (!(mdd->mdd_cl.mc_flags & CLM_ON))
                RETURN(0);
        if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
                RETURN(0);

        if (mdd_is_volatile_obj(mdd_obj))
                RETURN(0);

        if ((type >= CL_MTIME) && (type <= CL_ATIME) &&
            ktime_before(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) {
                /* Don't need multiple updates in this log */
                /* Don't check under lock - no big deal if we get an extra
                   entry */
                RETURN(0);
        }

        rc = mdd_changelog_data_store_by_fid(env, mdd, type, flags,
                                             mdo2fid(mdd_obj), handle);
        if (rc == 0)
                mdd_obj->mod_cltime = ktime_get();

        RETURN(rc);
}

static int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type,
                  int flags, struct md_device *m, const struct lu_fid *fid)
{
        struct thandle *handle;
        struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev);
        int rc;
        ENTRY;

        /* Not recording */
        if (!(mdd->mdd_cl.mc_flags & CLM_ON))
                RETURN(0);
        if (!(mdd->mdd_cl.mc_mask & (1 << type)))
                RETURN(0);

        LASSERT(fid != NULL);

        handle = mdd_trans_create(env, mdd);
        if (IS_ERR(handle))
                RETURN(PTR_ERR(handle));

        rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
        if (rc)
                GOTO(stop, rc);

        rc = mdd_trans_start(env, mdd, handle);
        if (rc)
                GOTO(stop, rc);

        rc = mdd_changelog_data_store_by_fid(env, mdd, type, flags,
                                                     fid, handle);

stop:
        rc = mdd_trans_stop(env, mdd, rc, handle);

        RETURN(rc);
}

/**
 * Save LMA extended attributes with data from \a ma.
 *
 * HSM and Size-On-MDS data will be extracted from \ma if they are valid, if
 * not, LMA EA will be first read from disk, modified and write back.
 *
 */
/* Precedence for choosing record type when multiple
 * attributes change: setattr > mtime > ctime > atime
 * (ctime changes when mtime does, plus chmod/chown.
 * atime and ctime are independent.) */
static int mdd_attr_set_changelog(const struct lu_env *env,
                                  struct md_object *obj, struct thandle *handle,
                                  __u64 valid)
{
        struct mdd_device *mdd = mdo2mdd(obj);
        int bits, type = 0;

        bits =  (valid & LA_SIZE)  ? 1 << CL_TRUNC : 0;
        bits |= (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0;
        bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0;
        bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0;
        bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0;
        bits = bits & mdd->mdd_cl.mc_mask;
        /* This is an implementation limit rather than a protocol limit */
        CLASSERT(CL_LAST <= sizeof(int) * 8);
        if (bits == 0)
                return 0;

        /* The record type is the lowest non-masked set bit */
        type = __ffs(bits);

        /* FYI we only store the first CLF_FLAGMASK bits of la_valid */
        return mdd_changelog_data_store(env, mdd, type, (int)valid,
                                        md2mdd_obj(obj), handle);
}

static int mdd_declare_attr_set(const struct lu_env *env,
                                struct mdd_device *mdd,
                                struct mdd_object *obj,
                                const struct lu_attr *attr,
                                struct thandle *handle)
{
        int rc;

        rc = mdo_declare_attr_set(env, obj, attr, handle);
        if (rc)
                return rc;

#ifdef CONFIG_FS_POSIX_ACL
        if (attr->la_valid & LA_MODE) {
                mdd_read_lock(env, obj, MOR_TGT_CHILD);
                rc = mdo_xattr_get(env, obj, &LU_BUF_NULL,
                                   XATTR_NAME_ACL_ACCESS);
                mdd_read_unlock(env, obj);
                if (rc == -EOPNOTSUPP || rc == -ENODATA)
                        rc = 0;
                else if (rc < 0)
                        return rc;

                if (rc != 0) {
                        struct lu_buf *buf = mdd_buf_get(env, NULL, rc);
                        rc = mdo_declare_xattr_set(env, obj, buf,
                                                   XATTR_NAME_ACL_ACCESS, 0,
                                                   handle);
                        if (rc)
                                return rc;
                }
        }
#endif

        rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
        return rc;
}

/*
 * LU-3671
 *
 * permission changes may require sync operation, to mitigate performance
 * impact, only do this for dir and when permission is reduced.
 *
 * For regular files, version is updated with permission change (see VBR), async
 * permission won't cause any issue, while missing permission change on
 * directory may affect accessibility of other objects after recovery.
 */
static inline bool permission_needs_sync(const struct lu_attr *old,
                                         const struct lu_attr *new)
{
        if (!S_ISDIR(old->la_mode))
                return false;

        if (new->la_valid & (LA_UID | LA_GID))
                return true;

        if (new->la_valid & LA_MODE &&
            new->la_mode & (S_ISUID | S_ISGID | S_ISVTX))
                return true;

        if ((new->la_valid & LA_MODE) &&
            ((new->la_mode & old->la_mode) & S_IRWXUGO) !=
             (old->la_mode & S_IRWXUGO))
                return true;

        return false;
}

/* set attr and LOV EA at once, return updated attr */
int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
                 const struct md_attr *ma)
{
        struct mdd_object *mdd_obj = md2mdd_obj(obj);
        struct mdd_device *mdd = mdo2mdd(obj);
        struct thandle *handle = NULL;
        struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
        struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
        const struct lu_attr *la = &ma->ma_attr;
        struct lu_ucred  *uc;
        int rc;
        ENTRY;

        /* we do not use ->attr_set() for LOV/HSM EA any more */
        LASSERT((ma->ma_valid & MA_LOV) == 0);
        LASSERT((ma->ma_valid & MA_HSM) == 0);

        rc = mdd_la_get(env, mdd_obj, attr);
        if (rc)
                RETURN(rc);

        *la_copy = ma->ma_attr;
        rc = mdd_fix_attr(env, mdd_obj, attr, la_copy, ma->ma_attr_flags);
        if (rc)
                RETURN(rc);

        /* no need to setattr anymore */
        if (la_copy->la_valid == 0) {
                CDEBUG(D_INODE, "%s: no valid attribute on "DFID", previous"
                       "valid is %#llx\n", mdd2obd_dev(mdd)->obd_name,
                       PFID(mdo2fid(mdd_obj)), la->la_valid);

                RETURN(0);
        }

        /* If an unprivileged user changes group of some file,
         * the setattr operation will be processed synchronously to
         * honor the quota limit of the corresponding group. see LU-5152 */
        uc = lu_ucred_check(env);
        if (S_ISREG(attr->la_mode) && la->la_valid & LA_GID &&
            la->la_gid != attr->la_gid && uc != NULL && uc->uc_fsuid != 0) {
                la_copy->la_valid |= LA_FLAGS;
                la_copy->la_flags |= LUSTRE_SET_SYNC_FL;

                /* Flush the possible existing sync requests to OSTs to
                 * keep the order of sync for the current setattr operation
                 * will be sent directly to OSTs. see LU-5152 */
                rc = dt_sync(env, mdd->mdd_child);
                if (rc)
                        GOTO(out, rc);
        }

        handle = mdd_trans_create(env, mdd);
        if (IS_ERR(handle)) {
                rc = PTR_ERR(handle);
                handle = NULL;

                GOTO(out, rc);
        }

        rc = mdd_declare_attr_set(env, mdd, mdd_obj, la_copy, handle);
        if (rc)
                GOTO(out, rc);

        rc = mdd_trans_start(env, mdd, handle);
        if (rc)
                GOTO(out, rc);

        if (mdd->mdd_sync_permission && permission_needs_sync(attr, la))
                handle->th_sync = 1;

        if (la->la_valid & (LA_MTIME | LA_CTIME))
                CDEBUG(D_INODE, "setting mtime %llu, ctime %llu\n",
                       la->la_mtime, la->la_ctime);

        mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
        if (la_copy->la_valid) {
                rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);

                if (rc == -EDQUOT && la_copy->la_flags & LUSTRE_SET_SYNC_FL) {
                        /* rollback to the original gid */
                        la_copy->la_flags &= ~LUSTRE_SET_SYNC_FL;
                        la_copy->la_gid = attr->la_gid;
                        mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1);
                }
        }
        mdd_write_unlock(env, mdd_obj);

out:
        if (rc == 0)
                rc = mdd_attr_set_changelog(env, obj, handle,
                                            la_copy->la_valid);

        if (handle != NULL)
                rc = mdd_trans_stop(env, mdd, rc, handle);

        return rc;
}

static int mdd_xattr_sanity_check(const struct lu_env *env,
                                  struct mdd_object *obj,
                                  const struct lu_attr *attr,
                                  const char *name)
{
        struct lu_ucred *uc     = lu_ucred_assert(env);
        ENTRY;

        if (attr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
                RETURN(-EPERM);

        if (strncmp(XATTR_USER_PREFIX, name,
                    sizeof(XATTR_USER_PREFIX) - 1) == 0) {
                /* For sticky directories, only the owner and privileged user
                 * can write attributes. */
                if (S_ISDIR(attr->la_mode) && (attr->la_mode & S_ISVTX) &&
                    (uc->uc_fsuid != attr->la_uid) &&
                    !md_capable(uc, CFS_CAP_FOWNER))
                        RETURN(-EPERM);
        } else {
                if ((uc->uc_fsuid != attr->la_uid) &&
                    !md_capable(uc, CFS_CAP_FOWNER))
                        RETURN(-EPERM);
        }

        RETURN(0);
}

/**
 * Check if a string begins with a given prefix.
 *
 * \param str     String to check
 * \param prefix  Substring to check at the beginning of \a str
 * \return true/false whether the condition is verified.
 */
static inline bool has_prefix(const char *str, const char *prefix)
{
        return strncmp(prefix, str, strlen(prefix)) == 0;
}

/**
 * Indicate the kind of changelog to store (if any) for a xattr set/del.
 *
 * \param[in]  xattr_name  Full extended attribute name.
 *
 * \return The type of changelog to use, or -1 if no changelog is to be emitted.
 */
static enum changelog_rec_type
mdd_xattr_changelog_type(const struct lu_env *env, struct mdd_device *mdd,
                         const char *xattr_name)
{
        /* Layout changes systematically recorded */
        if (strcmp(XATTR_NAME_LOV, xattr_name) == 0 ||
            strncmp(XATTR_LUSTRE_LOV, xattr_name,
                    strlen(XATTR_LUSTRE_LOV)) == 0)
                return CL_LAYOUT;

        /* HSM information changes systematically recorded */
        if (strcmp(XATTR_NAME_HSM, xattr_name) == 0)
                return CL_HSM;

        if (has_prefix(xattr_name, XATTR_USER_PREFIX) ||
            has_prefix(xattr_name, XATTR_NAME_POSIX_ACL_ACCESS) ||
            has_prefix(xattr_name, XATTR_NAME_POSIX_ACL_DEFAULT))
                return CL_XATTR;

        return -1;
}

static int mdd_declare_xattr_set(const struct lu_env *env,
                                 struct mdd_device *mdd,
                                 struct mdd_object *obj,
                                 const struct lu_buf *buf,
                                 const char *name,
                                 int fl, struct thandle *handle)
{
        int     rc;

        rc = mdo_declare_xattr_set(env, obj, buf, name, fl, handle);
        if (rc)
                return rc;

        if (mdd_xattr_changelog_type(env, mdd, name) < 0)
                return 0; /* no changelog to store */

        return mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
}

/*
 * Compare current and future data of HSM EA and add a changelog if needed.
 *
 * Caller should have write-locked \param obj.
 *
 * \param buf - Future HSM EA content.
 * \retval 0 if no changelog is needed or changelog was added properly.
 * \retval -ve errno if there was a problem
 */
static int mdd_hsm_update_locked(const struct lu_env *env,
                                 struct md_object *obj,
                                 const struct lu_buf *buf,
                                 struct thandle *handle, int *cl_flags)
{
        struct mdd_thread_info *info = mdd_env_info(env);
        struct mdd_object      *mdd_obj = md2mdd_obj(obj);
        struct lu_buf          *current_buf;
        struct md_hsm          *current_mh;
        struct md_hsm          *new_mh;
        int                     rc;
        ENTRY;

        OBD_ALLOC_PTR(current_mh);
        if (current_mh == NULL)
                RETURN(-ENOMEM);

        /* Read HSM attrs from disk */
        current_buf = lu_buf_check_and_alloc(&info->mti_xattr_buf,
                                mdo2mdd(obj)->mdd_dt_conf.ddp_max_ea_size);
        rc = mdo_xattr_get(env, mdd_obj, current_buf, XATTR_NAME_HSM);
        rc = lustre_buf2hsm(current_buf->lb_buf, rc, current_mh);
        if (rc < 0 && rc != -ENODATA)
                GOTO(free, rc);
        else if (rc == -ENODATA)
                current_mh->mh_flags = 0;

        /* Map future HSM xattr */
        OBD_ALLOC_PTR(new_mh);
        if (new_mh == NULL)
                GOTO(free, rc = -ENOMEM);
        lustre_buf2hsm(buf->lb_buf, buf->lb_len, new_mh);

        rc = 0;

        /* Flags differ, set flags for the changelog that will be added */
        if (current_mh->mh_flags != new_mh->mh_flags) {
                hsm_set_cl_event(cl_flags, HE_STATE);
                if (new_mh->mh_flags & HS_DIRTY)
                        hsm_set_cl_flags(cl_flags, CLF_HSM_DIRTY);
        }

        OBD_FREE_PTR(new_mh);
        EXIT;
free:
        OBD_FREE_PTR(current_mh);
        return rc;
}

static int mdd_object_pfid_replace(const struct lu_env *env,
                                   struct mdd_object *o)
{
        struct mdd_device *mdd = mdo2mdd(&o->mod_obj);
        struct thandle *handle;
        int rc;

        handle = mdd_trans_create(env, mdd);
        if (IS_ERR(handle))
                RETURN(PTR_ERR(handle));

        handle->th_complex = 1;

        /* it doesn't need to track the PFID update via llog, because LFSCK
         * will repair it even it goes wrong */
        rc = mdd_declare_xattr_set(env, mdd, o, NULL, XATTR_NAME_FID,
                                   0, handle);
        if (rc)
                GOTO(out, rc);

        rc = mdd_trans_start(env, mdd, handle);
        if (rc != 0)
                GOTO(out, rc);

        rc = mdo_xattr_set(env, o, NULL, XATTR_NAME_FID, 0, handle);
        if (rc)
                GOTO(out, rc);

out:
        mdd_trans_stop(env, mdd, rc, handle);
        return rc;
}


static int mdd_declare_xattr_del(const struct lu_env *env,
                                 struct mdd_device *mdd,
                                 struct mdd_object *obj,
                                 const char *name,
                                 struct thandle *handle);

static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
                         const char *name);

static int mdd_xattr_merge(const struct lu_env *env, struct md_object *md_obj,
                           struct md_object *md_vic)
{
        struct mdd_device *mdd = mdo2mdd(md_obj);
        struct mdd_object *obj = md2mdd_obj(md_obj);
        struct mdd_object *vic = md2mdd_obj(md_vic);
        struct lu_buf *buf = &mdd_env_info(env)->mti_buf[0];
        struct lu_buf *buf_vic = &mdd_env_info(env)->mti_buf[1];
        struct lov_mds_md *lmm;
        struct thandle *handle;
        int rc;
        ENTRY;

        rc = lu_fid_cmp(mdo2fid(obj), mdo2fid(vic));
        if (rc == 0) /* same fid */
                RETURN(-EPERM);

        handle = mdd_trans_create(env, mdd);
        if (IS_ERR(handle))
                RETURN(PTR_ERR(handle));

        if (rc > 0) {
                mdd_write_lock(env, obj, MOR_TGT_CHILD);
                mdd_write_lock(env, vic, MOR_TGT_CHILD);
        } else {
                mdd_write_lock(env, vic, MOR_TGT_CHILD);
                mdd_write_lock(env, obj, MOR_TGT_CHILD);
        }

        /* get EA of victim file */
        memset(buf_vic, 0, sizeof(*buf_vic));
        rc = mdd_get_lov_ea(env, vic, buf_vic);
        if (rc < 0) {
                if (rc == -ENODATA)
                        rc = 0;
                GOTO(out, rc);
        }

        /* parse the layout of victim file */
        lmm = buf_vic->lb_buf;
        if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
                GOTO(out, rc = -EINVAL);

        /* save EA of target file for restore */
        memset(buf, 0, sizeof(*buf));
        rc = mdd_get_lov_ea(env, obj, buf);
        if (rc < 0)
                GOTO(out, rc);

        /* Get rid of the layout from victim object */
        rc = mdd_declare_xattr_del(env, mdd, vic, XATTR_NAME_LOV, handle);
        if (rc)
                GOTO(out, rc);

        rc = mdd_declare_xattr_set(env, mdd, obj, buf_vic, XATTR_LUSTRE_LOV,
                                   LU_XATTR_MERGE, handle);
        if (rc)
                GOTO(out, rc);

        rc = mdd_trans_start(env, mdd, handle);
        if (rc != 0)
                GOTO(out, rc);

        rc = mdo_xattr_set(env, obj, buf_vic, XATTR_LUSTRE_LOV, LU_XATTR_MERGE,
                           handle);
        if (rc)
                GOTO(out, rc);

        rc = mdo_xattr_del(env, vic, XATTR_NAME_LOV, handle);
        if (rc) /* wtf? */
                GOTO(out_restore, rc);

        (void)mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, obj, handle);
        (void)mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, vic, handle);
        EXIT;

out_restore:
        if (rc) {
                int rc2 = mdo_xattr_set(env, obj, buf, XATTR_NAME_LOV,
                                        LU_XATTR_REPLACE, handle);
                if (rc2)
                        CERROR("%s: failed to rollback of layout of: "DFID
                               ": %d, file state unknown\n",
                               mdd_obj_dev_name(obj), PFID(mdo2fid(obj)), rc2);
        }

out:
        mdd_trans_stop(env, mdd, rc, handle);
        mdd_write_unlock(env, obj);
        mdd_write_unlock(env, vic);
        lu_buf_free(buf);
        lu_buf_free(buf_vic);

        if (!rc)
                (void) mdd_object_pfid_replace(env, obj);

        return rc;
}

/**
 * Extract the mirror with specified mirror id, and store the splitted
 * mirror layout to @buf.
 *
 * \param[in] comp_v1   mirrored layout
 * \param[in] mirror_id the mirror with mirror_id to be extracted
 * \param[out] buf      store the layout excluding the extracted mirror,
 *                      caller free the buffer we allocated in this function
 * \param[out] buf_vic  store the extracted layout, caller free the buffer
 *                      we allocated in this function
 *
 * \retval      0 on success; < 0 if error happens
 */
static int mdd_split_ea(struct lov_comp_md_v1 *comp_v1, __u16 mirror_id,
                        struct lu_buf *buf, struct lu_buf *buf_vic)
{
        struct lov_comp_md_v1 *comp_rem;
        struct lov_comp_md_v1 *comp_vic;
        struct lov_comp_md_entry_v1 *entry;
        struct lov_comp_md_entry_v1 *entry_rem;
        struct lov_comp_md_entry_v1 *entry_vic;
        __u16 mirror_cnt;
        __u16 comp_cnt, count = 0;
        int lmm_size, lmm_size_vic = 0;
        int i, j, k;
        int offset, offset_rem, offset_vic;

        mirror_cnt = le16_to_cpu(comp_v1->lcm_mirror_count) + 1;
        /* comp_v1 should contains more than 1 mirror */
        if (mirror_cnt <= 1)
                return -EINVAL;
        comp_cnt = le16_to_cpu(comp_v1->lcm_entry_count);
        lmm_size = le32_to_cpu(comp_v1->lcm_size);

        for (i = 0; i < comp_cnt; i++) {
                entry = &comp_v1->lcm_entries[i];
                if (mirror_id_of(le32_to_cpu(entry->lcme_id)) == mirror_id) {
                        count++;
                        lmm_size_vic += sizeof(*entry);
                        lmm_size_vic += le32_to_cpu(entry->lcme_size);
                } else if (count > 0) {
                        /* find the specified mirror */
                        break;
                }
        }

        if (count == 0)
                return -EINVAL;

        lu_buf_alloc(buf, lmm_size - lmm_size_vic);
        if (!buf->lb_buf)
                return -ENOMEM;

        lu_buf_alloc(buf_vic, sizeof(*comp_vic) + lmm_size_vic);
        if (!buf_vic->lb_buf) {
                lu_buf_free(buf);
                return -ENOMEM;
        }

        comp_rem = (struct lov_comp_md_v1 *)buf->lb_buf;
        comp_vic = (struct lov_comp_md_v1 *)buf_vic->lb_buf;

        memcpy(comp_rem, comp_v1, sizeof(*comp_v1));
        comp_rem->lcm_mirror_count = cpu_to_le16(mirror_cnt - 2);
        comp_rem->lcm_entry_count = cpu_to_le32(comp_cnt - count);
        comp_rem->lcm_size = cpu_to_le32(lmm_size - lmm_size_vic);
        if (!comp_rem->lcm_mirror_count)
                comp_rem->lcm_flags = cpu_to_le16(LCM_FL_NOT_FLR);

        memset(comp_vic, 0, sizeof(*comp_v1));
        comp_vic->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
        comp_vic->lcm_mirror_count = 0;
        comp_vic->lcm_entry_count = cpu_to_le32(count);
        comp_vic->lcm_size = cpu_to_le32(lmm_size_vic + sizeof(*comp_vic));
        comp_vic->lcm_flags = cpu_to_le16(LCM_FL_NOT_FLR);
        comp_vic->lcm_layout_gen = 0;

        offset = sizeof(*comp_v1) + sizeof(*entry) * comp_cnt;
        offset_rem = sizeof(*comp_rem) +
                     sizeof(*entry_rem) * (comp_cnt - count);
        offset_vic = sizeof(*comp_vic) + sizeof(*entry_vic) * count;
        for (i = j = k = 0; i < comp_cnt; i++) {
                struct lov_mds_md *lmm, *lmm_dst;
                bool vic = false;

                entry = &comp_v1->lcm_entries[i];
                entry_vic = &comp_vic->lcm_entries[j];
                entry_rem = &comp_rem->lcm_entries[k];

                if (mirror_id_of(le32_to_cpu(entry->lcme_id)) == mirror_id)
                        vic = true;

                /* copy component entry */
                if (vic) {
                        memcpy(entry_vic, entry, sizeof(*entry));
                        entry_vic->lcme_flags &= cpu_to_le32(LCME_FL_INIT);
                        entry_vic->lcme_offset = cpu_to_le32(offset_vic);
                        j++;
                } else {
                        memcpy(entry_rem, entry, sizeof(*entry));
                        entry_rem->lcme_offset = cpu_to_le32(offset_rem);
                        k++;
                }

                lmm = (struct lov_mds_md *)((char *)comp_v1 + offset);
                if (vic)
                        lmm_dst = (struct lov_mds_md *)
                                        ((char *)comp_vic + offset_vic);
                else
                        lmm_dst = (struct lov_mds_md *)
                                        ((char *)comp_rem + offset_rem);

                /* copy component entry blob */
                memcpy(lmm_dst, lmm, le32_to_cpu(entry->lcme_size));

                /* blob offset advance */
                offset += le32_to_cpu(entry->lcme_size);
                if (vic)
                        offset_vic += le32_to_cpu(entry->lcme_size);
                else
                        offset_rem += le32_to_cpu(entry->lcme_size);
        }

        return 0;
}

static int mdd_xattr_split(const struct lu_env *env, struct md_object *md_obj,
                           struct md_rejig_data *mrd)
{
        struct mdd_device *mdd = mdo2mdd(md_obj);
        struct mdd_object *obj = md2mdd_obj(md_obj);
        struct mdd_object *vic = md2mdd_obj(mrd->mrd_obj);
        struct lu_buf *buf = &mdd_env_info(env)->mti_buf[0];
        struct lu_buf *buf_save = &mdd_env_info(env)->mti_buf[1];
        struct lu_buf *buf_vic = &mdd_env_info(env)->mti_buf[2];
        struct lov_comp_md_v1 *lcm;
        struct thandle *handle;
        int rc;
        ENTRY;

        rc = lu_fid_cmp(mdo2fid(obj), mdo2fid(vic));
        if (rc == 0) /* same fid */
                RETURN(-EPERM);

        handle = mdd_trans_create(env, mdd);
        if (IS_ERR(handle))
                RETURN(PTR_ERR(handle));

        if (rc > 0) {
                mdd_write_lock(env, obj, MOR_TGT_CHILD);
                mdd_write_lock(env, vic, MOR_TGT_CHILD);
        } else {
                mdd_write_lock(env, vic, MOR_TGT_CHILD);
                mdd_write_lock(env, obj, MOR_TGT_CHILD);
        }

        /* get EA of mirrored file */
        memset(buf_save, 0, sizeof(*buf));
        rc = mdd_get_lov_ea(env, obj, buf_save);
        if (rc < 0)
                GOTO(out, rc);

        lcm = buf_save->lb_buf;
        if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
                GOTO(out, rc = -EINVAL);

        /**
         * Extract the mirror with specified mirror id, and store the splitted
         * mirror layout to the victim file.
         */
        memset(buf, 0, sizeof(*buf));
        memset(buf_vic, 0, sizeof(*buf_vic));
        rc = mdd_split_ea(lcm, mrd->mrd_mirror_id, buf, buf_vic);
        if (rc < 0)
                GOTO(out, rc);

        rc = mdd_declare_xattr_set(env, mdd, obj, buf, XATTR_NAME_LOV,
                                   LU_XATTR_SPLIT, handle);
        if (rc)
                GOTO(out, rc);
        rc = mdd_declare_xattr_set(env, mdd, vic, buf_vic, XATTR_NAME_LOV,
                                   LU_XATTR_SPLIT, handle);
        if (rc)
                GOTO(out, rc);

        rc = mdd_trans_start(env, mdd, handle);
        if (rc)
                GOTO(out, rc);

        rc = mdo_xattr_set(env, obj, buf, XATTR_NAME_LOV, LU_XATTR_REPLACE,
                           handle);
        if (rc)
                GOTO(out, rc);

        rc = mdo_xattr_set(env, vic, buf_vic, XATTR_NAME_LOV, LU_XATTR_CREATE,
                           handle);
        if (rc)
                GOTO(out_restore, rc);

        rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, obj, handle);
        if (rc)
                GOTO(out, rc);

        rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, vic, handle);
        if (rc)
                GOTO(out, rc);
        EXIT;

out_restore:
        if (rc) {
                /* restore obj's layout */
                int rc2 = mdo_xattr_set(env, obj, buf_save, XATTR_NAME_LOV,
                                        LU_XATTR_REPLACE, handle);
                if (rc2)
                        CERROR("%s: failed to rollback of layout of: "DFID
                               ": %d, file state unkonwn.\n",
                               mdd_obj_dev_name(obj), PFID(mdo2fid(obj)), rc2);
        }
out:
        mdd_trans_stop(env, mdd, rc, handle);
        mdd_write_unlock(env, obj);
        mdd_write_unlock(env, vic);
        lu_buf_free(buf_save);
        lu_buf_free(buf);
        lu_buf_free(buf_vic);

        if (!rc)
                (void) mdd_object_pfid_replace(env, obj);

        return rc;
}

static int mdd_layout_merge_allowed(const struct lu_env *env,
                                    struct md_object *target,
                                    struct md_object *victim)
{
        struct mdd_object *o1 = md2mdd_obj(target);

        /* cannot extend directory's LOVEA */
        if (S_ISDIR(mdd_object_type(o1))) {
                CERROR("%s: Don't extend directory's LOVEA, just set it.\n",
                       mdd_obj_dev_name(o1));
                RETURN(-EISDIR);
        }

        RETURN(0);
}

/**
 * The caller should guarantee to update the object ctime
 * after xattr_set if needed.
 */
static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
                         const struct lu_buf *buf, const char *name,
                         int fl)
{
        struct mdd_object       *mdd_obj = md2mdd_obj(obj);
        struct lu_attr          *attr = MDD_ENV_VAR(env, cattr);
        struct mdd_device       *mdd = mdo2mdd(obj);
        struct thandle          *handle;
        enum changelog_rec_type  cl_type;
        int                      cl_flags = 0;
        int                      rc;
        ENTRY;

        rc = mdd_la_get(env, mdd_obj, attr);
        if (rc)
                RETURN(rc);

        rc = mdd_xattr_sanity_check(env, mdd_obj, attr, name);
        if (rc)
                RETURN(rc);

        if (strcmp(name, XATTR_LUSTRE_LOV) == 0 &&
            (fl == LU_XATTR_MERGE || fl == LU_XATTR_SPLIT)) {
                struct md_rejig_data *mrd = buf->lb_buf;
                struct md_object *victim = mrd->mrd_obj;

                if (buf->lb_len != sizeof(*mrd))
                        RETURN(-EINVAL);

                rc = mdd_layout_merge_allowed(env, obj, victim);
                if (rc)
                        RETURN(rc);

                if (fl == LU_XATTR_MERGE)
                        /* merge layout of victim as a mirror of obj's. */
                        rc = mdd_xattr_merge(env, obj, victim);
                else
                        rc = mdd_xattr_split(env, obj, mrd);
                RETURN(rc);
        }

        if (strcmp(name, XATTR_NAME_ACL_ACCESS) == 0 ||
            strcmp(name, XATTR_NAME_ACL_DEFAULT) == 0) {
                struct posix_acl *acl;

                /* user may set empty ACL, which should be treated as removing
                 * ACL. */
                acl = posix_acl_from_xattr(&init_user_ns, buf->lb_buf,
                                           buf->lb_len);
                if (IS_ERR(acl))
                        RETURN(PTR_ERR(acl));
                if (acl == NULL) {
                        rc = mdd_xattr_del(env, obj, name);
                        RETURN(rc);
                }
                posix_acl_release(acl);
        }

        if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) {
                rc = mdd_acl_set(env, mdd_obj, attr, buf, fl);
                RETURN(rc);
        }

        handle = mdd_trans_create(env, mdd);
        if (IS_ERR(handle))
                RETURN(PTR_ERR(handle));

        rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, fl, handle);
        if (rc)
                GOTO(stop, rc);

        rc = mdd_trans_start(env, mdd, handle);
        if (rc)
                GOTO(stop, rc);

        mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);

        if (strcmp(XATTR_NAME_HSM, name) == 0) {
                rc = mdd_hsm_update_locked(env, obj, buf, handle, &cl_flags);
                if (rc) {
                        mdd_write_unlock(env, mdd_obj);
                        GOTO(stop, rc);
                }
        }

        rc = mdo_xattr_set(env, mdd_obj, buf, name, fl, handle);
        mdd_write_unlock(env, mdd_obj);
        if (rc)
                GOTO(stop, rc);

        cl_type = mdd_xattr_changelog_type(env, mdd, name);
        if (cl_type < 0)
                GOTO(stop, rc = 0);

        rc = mdd_changelog_data_store(env, mdd, cl_type, cl_flags, mdd_obj,
                                      handle);

        EXIT;
stop:
        return mdd_trans_stop(env, mdd, rc, handle);
}

static int mdd_declare_xattr_del(const struct lu_env *env,
                                 struct mdd_device *mdd,
                                 struct mdd_object *obj,
                                 const char *name,
                                 struct thandle *handle)
{
        int rc;

        rc = mdo_declare_xattr_del(env, obj, name, handle);
        if (rc)
                return rc;

        if (mdd_xattr_changelog_type(env, mdd, name) < 0)
                return 0; /* no changelog to store */

        return mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
}

/**
 * The caller should guarantee to update the object ctime
 * after xattr_set if needed.
 */
static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
                         const char *name)
{
        struct mdd_object *mdd_obj = md2mdd_obj(obj);
        struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
        struct mdd_device *mdd = mdo2mdd(obj);
        struct thandle *handle;
        int rc;
        ENTRY;

        rc = mdd_la_get(env, mdd_obj, attr);
        if (rc)
                RETURN(rc);

        rc = mdd_xattr_sanity_check(env, mdd_obj, attr, name);
        if (rc)
                RETURN(rc);

        handle = mdd_trans_create(env, mdd);
        if (IS_ERR(handle))
                RETURN(PTR_ERR(handle));

        rc = mdd_declare_xattr_del(env, mdd, mdd_obj, name, handle);
        if (rc)
                GOTO(stop, rc);

        rc = mdd_trans_start(env, mdd, handle);
        if (rc)
                GOTO(stop, rc);

        mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
        rc = mdo_xattr_del(env, mdd_obj, name, handle);
        mdd_write_unlock(env, mdd_obj);
        if (rc)
                GOTO(stop, rc);

        if (mdd_xattr_changelog_type(env, mdd, name) < 0)
                GOTO(stop, rc = 0);

        rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj, handle);

        EXIT;
stop:
        return mdd_trans_stop(env, mdd, rc, handle);
}

/*
 * read lov EA of an object
 * return the lov EA in an allocated lu_buf
 */
int mdd_get_lov_ea(const struct lu_env *env, struct mdd_object *obj,
                   struct lu_buf *lmm_buf)
{
        struct lu_buf   *buf = &mdd_env_info(env)->mti_big_buf;
        int              rc, bufsize;
        ENTRY;

repeat:
        rc = mdo_xattr_get(env, obj, buf, XATTR_NAME_LOV);

        if (rc == -ERANGE) {
                /* mti_big_buf is allocated but is too small
                 * we need to increase it */
                buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
                                             buf->lb_len * 2);
                if (buf->lb_buf == NULL)
                        GOTO(out, rc = -ENOMEM);
                goto repeat;
        }

        if (rc < 0)
                RETURN(rc);

        if (rc == 0)
                RETURN(-ENODATA);

        bufsize = rc;
        if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) {
                /* mti_big_buf was not allocated, so we have to
                 * allocate it based on the ea size */
                buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf,
                                             bufsize);
                if (buf->lb_buf == NULL)
                        GOTO(out, rc = -ENOMEM);
                goto repeat;
        }

        lu_buf_alloc(lmm_buf, bufsize);
        if (lmm_buf->lb_buf == NULL)
                GOTO(out, rc = -ENOMEM);

        memcpy(lmm_buf->lb_buf, buf->lb_buf, bufsize);
        rc = 0;
        EXIT;

out:
        if (rc < 0)
                lu_buf_free(lmm_buf);
        return rc;
}

static int mdd_xattr_hsm_replace(const struct lu_env *env,
                                 struct mdd_object *o, struct lu_buf *buf,
                                 struct thandle *handle)
{
        struct hsm_attrs *attrs;
        __u32 hsm_flags;
        int flags = 0;
        int rc;
        ENTRY;

        rc = mdo_xattr_set(env, o, buf, XATTR_NAME_HSM, LU_XATTR_REPLACE,
                           handle);
        if (rc != 0)
                RETURN(rc);

        attrs = buf->lb_buf;
        hsm_flags = le32_to_cpu(attrs->hsm_flags);
        if (!(hsm_flags & HS_RELEASED) || mdd_is_dead_obj(o))
                RETURN(0);

        /* Add a changelog record for release. */
        hsm_set_cl_event(&flags, HE_RELEASE);
        rc = mdd_changelog_data_store(env, mdo2mdd(&o->mod_obj), CL_HSM,
                                      flags, o, handle);
        RETURN(rc);
}

/*
 *  check if layout swapping between 2 objects is allowed
 *  the rules are:
 *  - only normal FIDs or non-system IGIFs
 *  - same type of objects
 *  - same owner/group (so quotas are still valid) unless this is from HSM
 *    release.
 */
static int mdd_layout_swap_allowed(const struct lu_env *env,
                                   struct mdd_object *o1,
                                   const struct lu_attr *attr1,
                                   struct mdd_object *o2,
                                   const struct lu_attr *attr2,
                                   __u64 flags)
{
        const struct lu_fid     *fid1, *fid2;
        ENTRY;

        fid1 = mdo2fid(o1);
        fid2 = mdo2fid(o2);

        if (!fid_is_norm(fid1) &&
            (!fid_is_igif(fid1) || IS_ERR(mdd_links_get(env, o1))))
                RETURN(-EBADF);

        if (!fid_is_norm(fid2) &&
            (!fid_is_igif(fid2) || IS_ERR(mdd_links_get(env, o2))))
                RETURN(-EBADF);

        if (mdd_object_type(o1) != mdd_object_type(o2)) {
                if (S_ISDIR(mdd_object_type(o1)))
                        RETURN(-ENOTDIR);
                if (S_ISREG(mdd_object_type(o1)))
                        RETURN(-EISDIR);
                RETURN(-EBADF);
        }

        if (flags & SWAP_LAYOUTS_MDS_HSM)
                RETURN(0);

        if ((attr1->la_uid != attr2->la_uid) ||
            (attr1->la_gid != attr2->la_gid))
                RETURN(-EPERM);

        RETURN(0);
}

/* XXX To set the proper lmm_oi & lmm_layout_gen when swap layouts, we have to
 *     look into the layout in MDD layer. */
static int mdd_lmm_oi(struct lov_mds_md *lmm, struct ost_id *oi, bool get)
{
        struct lov_comp_md_v1   *comp_v1;
        struct lov_mds_md       *v1;
        int                      i, ent_count;
        __u32                    off;

        if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1) {
                comp_v1 = (struct lov_comp_md_v1 *)lmm;
                ent_count = le16_to_cpu(comp_v1->lcm_entry_count);

                if (ent_count == 0)
                        return -EINVAL;

                if (get) {
                        off = le32_to_cpu(comp_v1->lcm_entries[0].lcme_offset);
                        v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
                        *oi = v1->lmm_oi;
                } else {
                        for (i = 0; i < le32_to_cpu(ent_count); i++) {
                                off = le32_to_cpu(comp_v1->lcm_entries[i].
                                                lcme_offset);
                                v1 = (struct lov_mds_md *)((char *)comp_v1 +
                                                off);
                                v1->lmm_oi = *oi;
                        }
                }
        } else if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1 ||
                   le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V3) {
                if (get)
                        *oi = lmm->lmm_oi;
                else
                        lmm->lmm_oi = *oi;
        } else {
                return -EINVAL;
        }
        return 0;
}

static inline int mdd_get_lmm_oi(struct lov_mds_md *lmm, struct ost_id *oi)
{
        return mdd_lmm_oi(lmm, oi, true);
}

static inline int mdd_set_lmm_oi(struct lov_mds_md *lmm, struct ost_id *oi)
{
        return mdd_lmm_oi(lmm, oi, false);
}

static int mdd_lmm_gen(struct lov_mds_md *lmm, __u32 *gen, bool get)
{
        struct lov_comp_md_v1 *comp_v1;

        if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1) {
                comp_v1 = (struct lov_comp_md_v1 *)lmm;
                if (get)
                        *gen = le32_to_cpu(comp_v1->lcm_layout_gen);
                else
                        comp_v1->lcm_layout_gen = cpu_to_le32(*gen);
        } else if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1 ||
                   le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V3) {
                __u16 tmp_gen = *gen;
                if (get)
                        *gen = le16_to_cpu(lmm->lmm_layout_gen);
                else
                        lmm->lmm_layout_gen = cpu_to_le16(tmp_gen);
        } else {
                return -EINVAL;
        }
        return 0;
}

static inline int mdd_get_lmm_gen(struct lov_mds_md *lmm, __u32 *gen)
{
        return mdd_lmm_gen(lmm, gen, true);
}

static inline int mdd_set_lmm_gen(struct lov_mds_md *lmm, __u32 *gen)
{
        return mdd_lmm_gen(lmm, gen, false);
}

/**
 * swap layouts between 2 lustre objects
 */
static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1,
                            struct md_object *obj2, __u64 flags)
{
        struct mdd_thread_info  *info = mdd_env_info(env);
        struct mdd_object       *fst_o = md2mdd_obj(obj1);
        struct mdd_object       *snd_o = md2mdd_obj(obj2);
        struct lu_attr          *fst_la = MDD_ENV_VAR(env, cattr);
        struct lu_attr          *snd_la = MDD_ENV_VAR(env, tattr);
        struct mdd_device       *mdd = mdo2mdd(obj1);
        struct lov_mds_md       *fst_lmm, *snd_lmm;
        struct lu_buf           *fst_buf = &info->mti_buf[0];
        struct lu_buf           *snd_buf = &info->mti_buf[1];
        struct lu_buf           *fst_hsm_buf = &info->mti_buf[2];
        struct lu_buf           *snd_hsm_buf = &info->mti_buf[3];
        struct ost_id           *saved_oi = NULL;
        struct thandle          *handle;
        __u32                    fst_gen, snd_gen, saved_gen;
        int                      fst_fl;
        int                      rc;
        int                      rc2;
        ENTRY;

        CLASSERT(ARRAY_SIZE(info->mti_buf) >= 4);
        memset(info->mti_buf, 0, sizeof(info->mti_buf));

        /* we have to sort the 2 obj, so locking will always
         * be in the same order, even in case of 2 concurrent swaps */
        rc = lu_fid_cmp(mdo2fid(fst_o), mdo2fid(snd_o));
        if (rc == 0) /* same fid ? */
                RETURN(-EPERM);

        if (rc < 0)
                swap(fst_o, snd_o);

        rc = mdd_la_get(env, fst_o, fst_la);
        if (rc != 0)
                RETURN(rc);

        rc = mdd_la_get(env, snd_o, snd_la);
        if (rc != 0)
                RETURN(rc);

        /* check if layout swapping is allowed */
        rc = mdd_layout_swap_allowed(env, fst_o, fst_la, snd_o, snd_la, flags);
        if (rc != 0)
                RETURN(rc);

        handle = mdd_trans_create(env, mdd);
        if (IS_ERR(handle))
                RETURN(PTR_ERR(handle));

        /* objects are already sorted */
        mdd_write_lock(env, fst_o, MOR_TGT_CHILD);
        mdd_write_lock(env, snd_o, MOR_TGT_CHILD);

        rc = mdd_get_lov_ea(env, fst_o, fst_buf);
        if (rc < 0 && rc != -ENODATA)
                GOTO(stop, rc);

        rc = mdd_get_lov_ea(env, snd_o, snd_buf);
        if (rc < 0 && rc != -ENODATA)
                GOTO(stop, rc);

        /* swapping 2 non existant layouts is a success */
        if (fst_buf->lb_buf == NULL && snd_buf->lb_buf == NULL)
                GOTO(stop, rc = 0);

        /* to help inode migration between MDT, it is better to
         * start by the no layout file (if one), so we order the swap */
        if (snd_buf->lb_buf == NULL) {
                swap(fst_o, snd_o);
                swap(fst_buf, snd_buf);
        }

        fst_gen = snd_gen = 0;
        /* lmm and generation layout initialization */
        if (fst_buf->lb_buf != NULL) {
                fst_lmm = fst_buf->lb_buf;
                mdd_get_lmm_gen(fst_lmm, &fst_gen);
                fst_fl  = LU_XATTR_REPLACE;
        } else {
                fst_lmm = NULL;
                fst_gen = 0;
                fst_fl  = LU_XATTR_CREATE;
        }

        snd_lmm = snd_buf->lb_buf;
        mdd_get_lmm_gen(snd_lmm, &snd_gen);

        saved_gen = fst_gen;
        /* increase the generation layout numbers */
        snd_gen++;
        fst_gen++;

        /*
         * XXX The layout generation is used to generate component IDs for
         *     the composite file, we have to do some special tweaks to make
         *     sure the layout generation is always adequate for that job.
         */

        /* Skip invalid generation number for composite layout */
        if ((snd_gen & LCME_ID_MASK) == 0)
                snd_gen++;
        if ((fst_gen & LCME_ID_MASK) == 0)
                fst_gen++;
        /* Make sure the generation is greater than all the component IDs */
        if (fst_gen < snd_gen)
                fst_gen = snd_gen;
        else if (fst_gen > snd_gen)
                snd_gen = fst_gen;

        /* set the file specific informations in lmm */
        if (fst_lmm != NULL) {
                saved_oi = &info->mti_oa.o_oi;
                mdd_get_lmm_oi(fst_lmm, saved_oi);
                mdd_set_lmm_gen(fst_lmm, &snd_gen);
                mdd_set_lmm_oi(fst_lmm, &snd_lmm->lmm_oi);
                mdd_set_lmm_oi(snd_lmm, saved_oi);
        } else {
                if ((snd_lmm->lmm_magic & cpu_to_le32(LOV_MAGIC_MASK)) ==
                    cpu_to_le32(LOV_MAGIC_MAGIC))
                        snd_lmm->lmm_magic |= cpu_to_le32(LOV_MAGIC_DEFINED);
                else
                        GOTO(stop, rc = -EPROTO);
        }
        mdd_set_lmm_gen(snd_lmm, &fst_gen);

        /* Prepare HSM attribute if it's required */
        if (flags & SWAP_LAYOUTS_MDS_HSM) {
                const int buflen = sizeof(struct hsm_attrs);

                lu_buf_alloc(fst_hsm_buf, buflen);
                lu_buf_alloc(snd_hsm_buf, buflen);
                if (fst_hsm_buf->lb_buf == NULL || snd_hsm_buf->lb_buf == NULL)
                        GOTO(stop, rc = -ENOMEM);

                /* Read HSM attribute */
                rc = mdo_xattr_get(env, fst_o, fst_hsm_buf, XATTR_NAME_HSM);
                if (rc < 0)
                        GOTO(stop, rc);

                rc = mdo_xattr_get(env, snd_o, snd_hsm_buf, XATTR_NAME_HSM);
                if (rc < 0)
                        GOTO(stop, rc);

                rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_hsm_buf,
                                           XATTR_NAME_HSM, LU_XATTR_REPLACE,
                                           handle);
                if (rc < 0)
                        GOTO(stop, rc);

                rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_hsm_buf,
                                           XATTR_NAME_HSM, LU_XATTR_REPLACE,
                                           handle);
                if (rc < 0)
                        GOTO(stop, rc);
        }

        /* prepare transaction */
        rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_buf, XATTR_NAME_LOV,
                                   fst_fl, handle);
        if (rc != 0)
                GOTO(stop, rc);

        if (fst_buf->lb_buf != NULL)
                rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_buf,
                                           XATTR_NAME_LOV, LU_XATTR_REPLACE,
                                           handle);
        else
                rc = mdd_declare_xattr_del(env, mdd, snd_o, XATTR_NAME_LOV,
                                           handle);
        if (rc != 0)
                GOTO(stop, rc);

        rc = mdd_trans_start(env, mdd, handle);
        if (rc != 0)
                GOTO(stop, rc);

        if (flags & SWAP_LAYOUTS_MDS_HSM) {
                rc = mdd_xattr_hsm_replace(env, fst_o, snd_hsm_buf, handle);
                if (rc < 0)
                        GOTO(stop, rc);

                rc = mdd_xattr_hsm_replace(env, snd_o, fst_hsm_buf, handle);
                if (rc < 0) {
                        rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf,
                                                    handle);
                        if (rc2 < 0)
                                CERROR("%s: restore "DFID" HSM error: %d/%d\n",
                                       mdd_obj_dev_name(fst_o),
                                       PFID(mdo2fid(fst_o)), rc, rc2);
                        GOTO(stop, rc);
                }
        }

        rc = mdo_xattr_set(env, fst_o, snd_buf, XATTR_NAME_LOV, fst_fl, handle);
        if (rc != 0)
                GOTO(stop, rc);

        if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_MDS_HSM_SWAP_LAYOUTS))) {
                rc = -EOPNOTSUPP;
        } else {
                if (fst_buf->lb_buf != NULL)
                        rc = mdo_xattr_set(env, snd_o, fst_buf, XATTR_NAME_LOV,
                                           LU_XATTR_REPLACE, handle);
                else
                        rc = mdo_xattr_del(env, snd_o, XATTR_NAME_LOV, handle);
        }
        if (rc != 0)
                GOTO(out_restore, rc);

        /* Issue one changelog record per file */
        rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle);
        if (rc)
                GOTO(stop, rc);

        rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle);
        if (rc)
                GOTO(stop, rc);
        EXIT;

out_restore:
        if (rc != 0) {
                int steps = 0;

                /* failure on second file, but first was done, so we have
                 * to roll back first. */
                if (fst_buf->lb_buf != NULL) {
                        mdd_set_lmm_oi(fst_lmm, saved_oi);
                        mdd_set_lmm_gen(fst_lmm, &saved_gen);
                        rc2 = mdo_xattr_set(env, fst_o, fst_buf, XATTR_NAME_LOV,
                                            LU_XATTR_REPLACE, handle);
                } else {
                        rc2 = mdo_xattr_del(env, fst_o, XATTR_NAME_LOV, handle);
                }
                if (rc2 < 0)
                        goto do_lbug;

                ++steps;
                rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, handle);
                if (rc2 < 0)
                        goto do_lbug;

                ++steps;
                rc2 = mdd_xattr_hsm_replace(env, snd_o, snd_hsm_buf, handle);

        do_lbug:
                if (rc2 < 0) {
                        /* very bad day */
                        CERROR("%s: unable to roll back layout swap. FIDs: "
                               DFID" and "DFID "error: %d/%d, steps: %d\n",
                               mdd_obj_dev_name(fst_o),
                               PFID(mdo2fid(snd_o)), PFID(mdo2fid(fst_o)),
                               rc, rc2, steps);
                        /* a solution to avoid journal commit is to panic,
                         * but it has strong consequences so we use LBUG to
                         * allow sysdamin to choose to panic or not
                         */
                        LBUG();
                }
        }

stop:
        rc = mdd_trans_stop(env, mdd, rc, handle);

        mdd_write_unlock(env, snd_o);
        mdd_write_unlock(env, fst_o);

        lu_buf_free(fst_buf);
        lu_buf_free(snd_buf);
        lu_buf_free(fst_hsm_buf);
        lu_buf_free(snd_hsm_buf);

        if (!rc) {
                (void) mdd_object_pfid_replace(env, fst_o);
                (void) mdd_object_pfid_replace(env, snd_o);
        }
        return rc;
}

static int mdd_declare_layout_change(const struct lu_env *env,
                                     struct mdd_device *mdd,
                                     struct mdd_object *obj,
                                     struct md_layout_change *mlc,
                                     struct thandle *handle)
{
        int rc;

        rc = mdo_declare_layout_change(env, obj, mlc, handle);
        if (rc)
                return rc;

        return mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
}

/* For PFL, this is used to instantiate necessary component objects. */
static int
mdd_layout_instantiate_component(const struct lu_env *env,
                struct mdd_object *obj, struct md_layout_change *mlc,
                struct thandle *handle)
{
        struct mdd_device *mdd = mdd_obj2mdd_dev(obj);
        int rc;
        ENTRY;

        if (mlc->mlc_opc != MD_LAYOUT_WRITE)
                RETURN(-ENOTSUPP);

        rc = mdd_declare_layout_change(env, mdd, obj, mlc, handle);
        /**
         * It's possible that another layout write intent has already
         * instantiated our objects, so a -EALREADY returned, and we need to
         * do nothing.
         */
        if (rc)
                RETURN(rc == -EALREADY ? 0 : rc);

        rc = mdd_trans_start(env, mdd, handle);
        if (rc)
                RETURN(rc);

        mdd_write_lock(env, obj, MOR_TGT_CHILD);
        rc = mdo_layout_change(env, obj, mlc, handle);
        mdd_write_unlock(env, obj);
        if (rc)
                RETURN(rc);

        rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, obj, handle);
        RETURN(rc);
}

/**
 * Change the FLR layout from RDONLY to WRITE_PENDING.
 *
 * It picks the primary mirror, and bumps the layout version, and set
 * layout version xattr to OST objects in a sync tx. In order to facilitate
 * the handling of phantom writers from evicted clients, the clients carry
 * layout version of the file with write RPC, so that the OSTs can verify
 * if the write RPCs are legitimate, meaning not from evicted clients.
 */
static int
mdd_layout_update_rdonly(const struct lu_env *env, struct mdd_object *obj,
                         struct md_layout_change *mlc, struct thandle *handle)
{
        struct mdd_device *mdd = mdd_obj2mdd_dev(obj);
        int rc;
        ENTRY;

        /* Verify acceptable operations */
        switch (mlc->mlc_opc) {
        case MD_LAYOUT_WRITE:
                break;
        case MD_LAYOUT_RESYNC:
                /* these are legal operations - this represents the case that
                 * a few mirrors were missed in the last resync.
                 * XXX: it will be supported later */
        case MD_LAYOUT_RESYNC_DONE:
        default:
                RETURN(0);
        }

        rc = mdd_declare_layout_change(env, mdd, obj, mlc, handle);
        if (rc)
                GOTO(out, rc);

        rc = mdd_declare_xattr_del(env, mdd, obj, XATTR_NAME_SOM, handle);
        if (rc)
                GOTO(out, rc);

        /* record a changelog for data mover to consume */
        rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
        if (rc)
                GOTO(out, rc);

        rc = mdd_trans_start(env, mdd, handle);
        if (rc)
                GOTO(out, rc);

        /* it needs a sync tx to make FLR to work properly */
        handle->th_sync = 1;

        mdd_write_lock(env, obj, MOR_TGT_CHILD);
        rc = mdo_layout_change(env, obj, mlc, handle);
        if (!rc) {
                rc = mdo_xattr_del(env, obj, XATTR_NAME_SOM, handle);
                if (rc == -ENODATA)
                        rc = 0;
        }
        mdd_write_unlock(env, obj);
        if (rc)
                GOTO(out, rc);

        rc = mdd_changelog_data_store(env, mdd, CL_FLRW, 0, obj, handle);
        if (rc)
                GOTO(out, rc);

        EXIT;

out:
        return rc;
}

/**
 * Handle mirrored file state transition when it's in WRITE_PENDING.
 *
 * Only MD_LAYOUT_RESYNC, which represents start of resync, is allowed when
 * the file is in WRITE_PENDING state. If everything goes fine, the file's
 * layout version will be increased, and the file's state will be changed to
 * SYNC_PENDING.
 */
static int
mdd_layout_update_write_pending(const struct lu_env *env,
                struct mdd_object *obj, struct md_layout_change *mlc,
                struct thandle *handle)
{
        struct mdd_device *mdd = mdd_obj2mdd_dev(obj);
        int rc;
        ENTRY;

        switch (mlc->mlc_opc) {
        case MD_LAYOUT_RESYNC:
                /* Upon receiving the resync request, it should
                 * instantiate all stale components right away to get ready
                 * for mirror copy. In order to avoid layout version change,
                 * client should avoid sending LAYOUT_WRITE request at the
                 * resync state. */
                break;
        case MD_LAYOUT_WRITE:
                /* legal race for concurrent write, the file state has been
                 * changed by another client. */
                break;
        default:
                RETURN(-EBUSY);
        }

        rc = mdd_declare_layout_change(env, mdd, obj, mlc, handle);
        if (rc)
                GOTO(out, rc);

        rc = mdd_trans_start(env, mdd, handle);
        if (rc)
                GOTO(out, rc);

        /* it needs a sync tx to make FLR to work properly */
        handle->th_sync = 1;

        mdd_write_lock(env, obj, MOR_TGT_CHILD);
        rc = mdo_layout_change(env, obj, mlc, handle);
        mdd_write_unlock(env, obj);
        if (rc)
                GOTO(out, rc);

        EXIT;

out:
        return rc;
}

/**
 * Handle the requests when a FLR file's state is in SYNC_PENDING.
 *
 * Only concurrent write and sync complete requests are possible when the
 * file is in SYNC_PENDING. For the latter request, it will pass in the
 * mirrors that have been synchronized, then the stale bit will be cleared
 * to make the file's state turn into RDONLY.
 * For concurrent write reqeust, it just needs to change the file's state
 * to WRITE_PENDING in a sync tx. It doesn't have to change the layout
 * version because the version will be increased in the transition to
 * SYNC_PENDING later so that it can deny the write request from potential
 * evicted SYNC clients. */
static int
mdd_object_update_sync_pending(const struct lu_env *env, struct mdd_object *obj,
                struct md_layout_change *mlc, struct thandle *handle)
{
        struct mdd_device *mdd = mdd_obj2mdd_dev(obj);
        struct lu_buf *som_buf = &mdd_env_info(env)->mti_buf[1];
        int fl = 0;
        int rc;
        ENTRY;

        /* operation validation */
        switch (mlc->mlc_opc) {
        case MD_LAYOUT_RESYNC_DONE:
                /* resync complete. */
        case MD_LAYOUT_WRITE:
                /* concurrent write. */
                break;
        case MD_LAYOUT_RESYNC:
                /* resync again, most likely the previous run failed.
                 * no-op if it's already in SYNC_PENDING state */
                RETURN(0);
        default:
                RETURN(-EBUSY);
        }

        if (mlc->mlc_som.lsa_valid & LSOM_FL_VALID) {
                rc = mdo_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_SOM);
                if (rc && rc != -ENODATA)
                        RETURN(rc);

                fl = rc == -ENODATA ? LU_XATTR_CREATE : LU_XATTR_REPLACE;
                som_buf->lb_buf = &mlc->mlc_som;
                som_buf->lb_len = sizeof(mlc->mlc_som);
        }

        rc = mdd_declare_layout_change(env, mdd, obj, mlc, handle);
        if (rc)
                GOTO(out, rc);

        /* record a changelog for the completion of resync */
        rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
        if (rc)
                GOTO(out, rc);

        /* RESYNC_DONE has piggybacked size and blocks */
        if (fl) {
                rc = mdd_declare_xattr_set(env, mdd, obj, som_buf,
                                           XATTR_NAME_SOM, fl, handle);
                if (rc)
                        GOTO(out, rc);
        }

        rc = mdd_trans_start(env, mdd, handle);
        if (rc)
                GOTO(out, rc);

        /* it needs a sync tx to make FLR to work properly */
        handle->th_sync = 1;

        rc = mdo_layout_change(env, obj, mlc, handle);
        if (rc)
                GOTO(out, rc);

        if (fl) {
                rc = mdo_xattr_set(env, obj, som_buf, XATTR_NAME_SOM,
                                   fl, handle);
                if (rc)
                        GOTO(out, rc);
        }

        rc = mdd_changelog_data_store(env, mdd, CL_RESYNC, 0, obj, handle);
        if (rc)
                GOTO(out, rc);
        EXIT;
out:
        return rc;
}

/**
 * Layout change callback for object.
 *
 * This is only used by FLR for now. In the future, it can be exteneded to
 * handle all layout change.
 */
static int
mdd_layout_change(const struct lu_env *env, struct md_object *o,
                  struct md_layout_change *mlc)
{
        struct mdd_object       *obj = md2mdd_obj(o);
        struct mdd_device       *mdd = mdd_obj2mdd_dev(obj);
        struct lu_buf           *buf = mdd_buf_get(env, NULL, 0);
        struct lov_comp_md_v1   *lcm;
        struct thandle          *handle;
        int flr_state;
        int rc;
        ENTRY;

        /* Verify acceptable operations */
        switch (mlc->mlc_opc) {
        case MD_LAYOUT_WRITE:
        case MD_LAYOUT_RESYNC:
        case MD_LAYOUT_RESYNC_DONE:
                break;
        default:
                RETURN(-ENOTSUPP);
        }

        handle = mdd_trans_create(env, mdd);
        if (IS_ERR(handle))
                RETURN(PTR_ERR(handle));

        rc = mdd_get_lov_ea(env, obj, buf);
        if (rc < 0) {
                if (rc == -ENODATA)
                        rc = -EINVAL;
                GOTO(out, rc);
        }

        /* analyze the layout to make sure it's a FLR file */
        lcm = buf->lb_buf;
        if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
                GOTO(out, rc = -EINVAL);

        flr_state = le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK;

        /* please refer to HLD of FLR for state transition */
        switch (flr_state) {
        case LCM_FL_NOT_FLR:
                rc = mdd_layout_instantiate_component(env, obj, mlc, handle);
                break;
        case LCM_FL_WRITE_PENDING:
                rc = mdd_layout_update_write_pending(env, obj, mlc, handle);
                break;
        case LCM_FL_RDONLY:
                rc = mdd_layout_update_rdonly(env, obj, mlc, handle);
                break;
        case LCM_FL_SYNC_PENDING:
                rc = mdd_object_update_sync_pending(env, obj, mlc, handle);
                break;
        default:
                rc = 0;
                break;
        }
        EXIT;

out:
        mdd_trans_stop(env, mdd, rc, handle);
        lu_buf_free(buf);
        return rc;
}

void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent,
                          struct mdd_object *child, const struct lu_attr *attr,
                          const struct md_op_spec *spec,
                          struct dt_allocation_hint *hint)
{
        struct dt_object *np = parent ?  mdd_object_child(parent) : NULL;
        struct dt_object *nc = mdd_object_child(child);

        memset(hint, 0, sizeof(*hint));

        /* For striped directory, give striping EA to lod_ah_init, which will
         * decide the stripe_offset and stripe count by it. */
        if (S_ISDIR(attr->la_mode) &&
            unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
                hint->dah_eadata = spec->u.sp_ea.eadata;
                hint->dah_eadata_len = spec->u.sp_ea.eadatalen;
        } else {
                hint->dah_eadata = NULL;
                hint->dah_eadata_len = 0;
        }

        CDEBUG(D_INFO, DFID" eadata %p len %d\n", PFID(mdd_object_fid(child)),
               hint->dah_eadata, hint->dah_eadata_len);
        /* @hint will be initialized by underlying device. */
        nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT);
}

/*
 * do NOT or the MAY_*'s, you'll get the weakest
 */
int accmode(const struct lu_env *env, const struct lu_attr *la, int flags)
{
        int res = 0;

        /* Sadly, NFSD reopens a file repeatedly during operation, so the
         * "acc_mode = 0" allowance for newly-created files isn't honoured.
         * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
         * owner can write to a file even if it is marked readonly to hide
         * its brokenness. (bug 5781) */
        if (flags & MDS_OPEN_OWNEROVERRIDE) {
                struct lu_ucred *uc = lu_ucred_check(env);

                if ((uc == NULL) || (la->la_uid == uc->uc_fsuid))
                        return 0;
        }

        if (flags & FMODE_READ)
                res |= MAY_READ;
        if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
                res |= MAY_WRITE;
        if (flags & MDS_FMODE_EXEC)
                res = MAY_EXEC;
        return res;
}

static int mdd_open_sanity_check(const struct lu_env *env,
                                struct mdd_object *obj,
                                const struct lu_attr *attr, int flag)
{
        int mode, rc;
        ENTRY;

        /* EEXIST check */
        if (mdd_is_dead_obj(obj))
                RETURN(-ENOENT);

        if (S_ISLNK(attr->la_mode))
                RETURN(-ELOOP);

        mode = accmode(env, attr, flag);

        if (S_ISDIR(attr->la_mode) && (mode & MAY_WRITE))
                RETURN(-EISDIR);

        if (!(flag & MDS_OPEN_CREATED)) {
                rc = mdd_permission_internal(env, obj, attr, mode);
                if (rc)
                        RETURN(rc);
        }

        if (S_ISFIFO(attr->la_mode) || S_ISSOCK(attr->la_mode) ||
            S_ISBLK(attr->la_mode) || S_ISCHR(attr->la_mode))
                flag &= ~MDS_OPEN_TRUNC;

        /* For writing append-only file must open it with append mode. */
        if (attr->la_flags & LUSTRE_APPEND_FL) {
                if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
                        RETURN(-EPERM);
                if (flag & MDS_OPEN_TRUNC)
                        RETURN(-EPERM);
        }

        RETURN(0);
}

static int mdd_open(const struct lu_env *env, struct md_object *obj,
                    int flags)
{
        struct mdd_object *mdd_obj = md2mdd_obj(obj);
        struct md_device *md_dev = lu2md_dev(mdd2lu_dev(mdo2mdd(obj)));
        struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
        int rc = 0;

        mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);

        rc = mdd_la_get(env, mdd_obj, attr);
        if (rc != 0)
                GOTO(out, rc);

        rc = mdd_open_sanity_check(env, mdd_obj, attr, flags);
        if (rc != 0)
                GOTO(out, rc);

        mdd_obj->mod_count++;

        mdd_changelog(env, CL_OPEN, flags, md_dev, mdo2fid(mdd_obj));

        EXIT;
out:
        mdd_write_unlock(env, mdd_obj);
        return rc;
}

static int mdd_declare_close(const struct lu_env *env,
                             struct mdd_object *obj,
                             struct md_attr *ma,
                             struct thandle *handle)
{
        int rc;

        rc = orph_declare_index_delete(env, obj, handle);
        if (rc)
                return rc;

        return mdo_declare_destroy(env, obj, handle);
}

/*
 * No permission check is needed.
 */
static int mdd_close(const struct lu_env *env, struct md_object *obj,
                     struct md_attr *ma, int mode)
{
        struct mdd_object *mdd_obj = md2mdd_obj(obj);
        struct mdd_device *mdd = mdo2mdd(obj);
        struct thandle *handle = NULL;
        int is_orphan = 0;
        int rc;
        bool blocked = false;
        ENTRY;

        if (ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_KEEP_ORPHAN) {
                mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
                mdd_obj->mod_count--;
                mdd_write_unlock(env, mdd_obj);

                if (mdd_obj->mod_flags & ORPHAN_OBJ && !mdd_obj->mod_count)
                        CDEBUG(D_HA, "Object "DFID" is retained in orphan "
                                "list\n", PFID(mdd_object_fid(mdd_obj)));
                RETURN(0);
        }

        /* mdd_finish_unlink() will always set orphan object as DEAD_OBJ, but
         * it might fail to add the object to orphan list (w/o ORPHAN_OBJ). */
        /* check without any lock */
        is_orphan = mdd_obj->mod_count == 1 &&
                    (mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0;

again:
        if (is_orphan) {
                /* mdd_trans_create() maybe failed because of barrier_entry(),
                 * under such case, the orphan MDT-object will be left in the
                 * orphan list, and when the MDT remount next time, the unused
                 * orphans will be destroyed automatically.
                 *
                 * One exception: the former mdd_finish_unlink may failed to
                 * add the orphan MDT-object to the orphan list, then if the
                 * mdd_trans_create() failed because of barrier_entry(), the
                 * MDT-object will become real orphan that is neither in the
                 * namespace nor in the orphan list. Such bad case should be
                 * very rare and will be handled by e2fsck/lfsck. */
                handle = mdd_trans_create(env, mdo2mdd(obj));
                if (IS_ERR(handle)) {
                        rc = PTR_ERR(handle);
                        if (rc != -EINPROGRESS)
                                GOTO(stop, rc);

                        handle = NULL;
                        blocked = true;
                        goto cont;
                }

                rc = mdd_declare_close(env, mdd_obj, ma, handle);
                if (rc)
                        GOTO(stop, rc);

                rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle);
                if (rc)
                        GOTO(stop, rc);

                rc = mdd_trans_start(env, mdo2mdd(obj), handle);
                if (rc)
                        GOTO(stop, rc);
        }

cont:
        mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD);
        rc = mdd_la_get(env, mdd_obj, &ma->ma_attr);
        if (rc != 0) {
                CERROR("Failed to get lu_attr of "DFID": %d\n",
                       PFID(mdd_object_fid(mdd_obj)), rc);
                GOTO(out, rc);
        }

        /* check again with lock */
        is_orphan = (mdd_obj->mod_count == 1) &&
                    ((mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0 ||
                     ma->ma_attr.la_nlink == 0);

        if (is_orphan && !handle && !blocked) {
                mdd_write_unlock(env, mdd_obj);
                goto again;
        }

        mdd_obj->mod_count--; /*release open count */

        if (!is_orphan || blocked)
                GOTO(out, rc = 0);

        /* Orphan object */
        /* NB: Object maybe not in orphan list originally, it is rare case for
         * mdd_finish_unlink() failure, in that case, the object doesn't have
         * ORPHAN_OBJ flag */
        if ((mdd_obj->mod_flags & ORPHAN_OBJ) != 0) {
                /* remove link to object from orphan index */
                LASSERT(handle != NULL);
                rc = __mdd_orphan_del(env, mdd_obj, handle);
                if (rc != 0) {
                        CERROR("%s: unable to delete "DFID" from orphan list: "
                               "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
                               PFID(mdd_object_fid(mdd_obj)), rc);
                        /* If object was not deleted from orphan list, do not
                         * destroy OSS objects, which will be done when next
                         * recovery. */
                        GOTO(out, rc);
                }

                CDEBUG(D_HA, "Object "DFID" is deleted from orphan "
                       "list, OSS objects to be destroyed.\n",
                       PFID(mdd_object_fid(mdd_obj)));
        }

        rc = mdo_destroy(env, mdd_obj, handle);

        if (rc != 0) {
                CERROR("%s: unable to delete "DFID" from orphan list: "
                       "rc = %d\n", lu_dev_name(mdd2lu_dev(mdd)),
                       PFID(mdd_object_fid(mdd_obj)), rc);
        }
        EXIT;

out:
        mdd_write_unlock(env, mdd_obj);

        /* Record CL_CLOSE in changelog only if file was opened in write mode,
         * or if CL_OPEN was recorded.
         * Changelogs mask may change between open and close operations, but
         * this is not a big deal if we have a CL_CLOSE entry with no matching
         * CL_OPEN. Plus Changelogs mask may not change often.
         */
        if (!rc && !blocked &&
            ((mode & (FMODE_WRITE | MDS_OPEN_APPEND | MDS_OPEN_TRUNC)) ||
             (mdd->mdd_cl.mc_mask & (1 << CL_OPEN))) &&
            !(ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_RECOV_OPEN)) {
                if (handle == NULL) {
                        handle = mdd_trans_create(env, mdo2mdd(obj));
                        if (IS_ERR(handle))
                                GOTO(stop, rc = PTR_ERR(handle));

                        rc = mdd_declare_changelog_store(env, mdd, NULL, NULL,
                                                         handle);
                        if (rc)
                                GOTO(stop, rc);

                        rc = mdd_trans_start(env, mdo2mdd(obj), handle);
                        if (rc)
                                GOTO(stop, rc);
                }

                mdd_changelog_data_store(env, mdd, CL_CLOSE, mode,
                                         mdd_obj, handle);
        }

stop:
        if (handle != NULL && !IS_ERR(handle))
                rc = mdd_trans_stop(env, mdd, rc, handle);

        return rc;
}

/*
 * Permission check is done when open,
 * no need check again.
 */
static int mdd_readpage_sanity_check(const struct lu_env *env,
                                     struct mdd_object *obj)
{
        struct dt_object *next = mdd_object_child(obj);
        int rc;
        ENTRY;

        if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
                rc = 0;
        else
                rc = -ENOTDIR;

        RETURN(rc);
}

static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp,
                              size_t nob, const struct dt_it_ops *iops,
                              struct dt_it *it, __u32 attr, void *arg)
{
        struct lu_dirpage       *dp = &lp->lp_dir;
        void                    *area = dp;
        int                      result;
        __u64                    hash = 0;
        struct lu_dirent        *ent;
        struct lu_dirent        *last = NULL;
        struct lu_fid            fid;
        int                      first = 1;

        if (nob < sizeof(*dp))
                return -EINVAL;

        memset(area, 0, sizeof (*dp));
        area += sizeof (*dp);
        nob  -= sizeof (*dp);

        ent  = area;
        do {
                int    len;
                size_t recsize;

                len = iops->key_size(env, it);

                /* IAM iterator can return record with zero len. */
                if (len == 0)
                        goto next;

                hash = iops->store(env, it);
                if (unlikely(first)) {
                        first = 0;
                        dp->ldp_hash_start = cpu_to_le64(hash);
                }

                /* calculate max space required for lu_dirent */
                recsize = lu_dirent_calc_size(len, attr);

                if (nob >= recsize) {
                        result = iops->rec(env, it, (struct dt_rec *)ent, attr);
                        if (result == -ESTALE)
                                goto next;
                        if (result != 0)
                                goto out;

                        /* osd might not able to pack all attributes,
                         * so recheck rec length */
                        recsize = le16_to_cpu(ent->lde_reclen);

                        if (le32_to_cpu(ent->lde_attrs) & LUDA_FID) {
                                fid_le_to_cpu(&fid, &ent->lde_fid);
                                if (fid_is_dot_lustre(&fid))
                                        goto next;
                        }
                } else {
                        result = (last != NULL) ? 0 :-EINVAL;
                        goto out;
                }
                last = ent;
                ent = (void *)ent + recsize;
                nob -= recsize;

next:
                result = iops->next(env, it);
                if (result == -ESTALE)
                        goto next;
        } while (result == 0);

out:
        dp->ldp_hash_end = cpu_to_le64(hash);
        if (last != NULL) {
                if (last->lde_hash == dp->ldp_hash_end)
                        dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE);
                last->lde_reclen = 0; /* end mark */
        }
        if (result > 0)
                /* end of directory */
                dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
        else if (result < 0)
                CWARN("build page failed: %d!\n", result);
        return result;
}

int mdd_readpage(const struct lu_env *env, struct md_object *obj,
                 const struct lu_rdpg *rdpg)
{
        struct mdd_object *mdd_obj = md2mdd_obj(obj);
        int rc;
        ENTRY;

        if (mdd_object_exists(mdd_obj) == 0) {
                CERROR("%s: object "DFID" not found: rc = -2\n",
                       mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj)));
                return -ENOENT;
        }

        mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
        rc = mdd_readpage_sanity_check(env, mdd_obj);
        if (rc)
                GOTO(out_unlock, rc);

        if (mdd_is_dead_obj(mdd_obj)) {
                struct page *pg;
                struct lu_dirpage *dp;

                /*
                 * According to POSIX, please do not return any entry to client:
                 * even dot and dotdot should not be returned.
                 */
                CDEBUG(D_INODE, "readdir from dead object: "DFID"\n",
                       PFID(mdd_object_fid(mdd_obj)));

                if (rdpg->rp_count <= 0)
                        GOTO(out_unlock, rc = -EFAULT);
                LASSERT(rdpg->rp_pages != NULL);

                pg = rdpg->rp_pages[0];
                dp = (struct lu_dirpage *)kmap(pg);
                memset(dp, 0 , sizeof(struct lu_dirpage));
                dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
                dp->ldp_hash_end   = cpu_to_le64(MDS_DIR_END_OFF);
                dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
                kunmap(pg);
                GOTO(out_unlock, rc = LU_PAGE_SIZE);
        }

        rc = dt_index_walk(env, mdd_object_child(mdd_obj), rdpg,
                           mdd_dir_page_build, NULL);
        if (rc >= 0) {
                struct lu_dirpage       *dp;

                dp = kmap(rdpg->rp_pages[0]);
                dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash);
                if (rc == 0) {
                        /*
                         * No pages were processed, mark this for first page
                         * and send back.
                         */
                        dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF);
                        dp->ldp_flags = cpu_to_le32(LDF_EMPTY);
                        rc = min_t(unsigned int, LU_PAGE_SIZE, rdpg->rp_count);
                }
                kunmap(rdpg->rp_pages[0]);
        }

        GOTO(out_unlock, rc);
out_unlock:
        mdd_read_unlock(env, mdd_obj);
        return rc;
}

static int mdd_object_sync(const struct lu_env *env, struct md_object *obj)
{
        struct mdd_object *mdd_obj = md2mdd_obj(obj);

        if (mdd_object_exists(mdd_obj) == 0) {
                int rc = -ENOENT;

                CERROR("%s: object "DFID" not found: rc = %d\n",
                       mdd_obj_dev_name(mdd_obj),
                       PFID(mdd_object_fid(mdd_obj)), rc);
                return rc;
        }
        return dt_object_sync(env, mdd_object_child(mdd_obj),
                              0, OBD_OBJECT_EOF);
}

static int mdd_object_lock(const struct lu_env *env,
                           struct md_object *obj,
                           struct lustre_handle *lh,
                           struct ldlm_enqueue_info *einfo,
                           union ldlm_policy_data *policy)
{
        struct mdd_object *mdd_obj = md2mdd_obj(obj);
        return dt_object_lock(env, mdd_object_child(mdd_obj), lh,
                              einfo, policy);
}

static int mdd_object_unlock(const struct lu_env *env,
                             struct md_object *obj,
                             struct ldlm_enqueue_info *einfo,
                             union ldlm_policy_data *policy)
{
        struct mdd_object *mdd_obj = md2mdd_obj(obj);
        return dt_object_unlock(env, mdd_object_child(mdd_obj), einfo, policy);
}

const struct md_object_operations mdd_obj_ops = {
        .moo_permission         = mdd_permission,
        .moo_attr_get           = mdd_attr_get,
        .moo_attr_set           = mdd_attr_set,
        .moo_xattr_get          = mdd_xattr_get,
        .moo_xattr_set          = mdd_xattr_set,
        .moo_xattr_list         = mdd_xattr_list,
        .moo_invalidate         = mdd_invalidate,
        .moo_xattr_del          = mdd_xattr_del,
        .moo_swap_layouts       = mdd_swap_layouts,
        .moo_open               = mdd_open,
        .moo_close              = mdd_close,
        .moo_readpage           = mdd_readpage,
        .moo_readlink           = mdd_readlink,
        .moo_changelog          = mdd_changelog,
        .moo_object_sync        = mdd_object_sync,
        .moo_object_lock        = mdd_object_lock,
        .moo_object_unlock      = mdd_object_unlock,
        .moo_layout_change      = mdd_layout_change,
};