Whamcloud - gitweb
LU-5777 quota: reserve enough credits for setattr
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_quota.c
index c39f829..38cf125 100644 (file)
  * GPL HEADER END
  */
 /*
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2012, 2013, Intel Corporation.
  * Use is subject to license terms.
  *
  * Author: Johann Lombardi <johann@whamcloud.com>
  * Author: Niu    Yawei    <niu@whamcloud.com>
  */
 
-#include <lquota.h>
+#include <lustre_quota.h>
 #include "osd_internal.h"
 
 /**
@@ -65,6 +65,10 @@ int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
                        const struct lu_fid *fid, struct osd_inode_id *id)
 {
        struct super_block *sb = osd_sb(osd);
+        unsigned long qf_inums[2] = {
+               le32_to_cpu(LDISKFS_SB(sb)->s_es->s_usr_quota_inum),
+               le32_to_cpu(LDISKFS_SB(sb)->s_es->s_grp_quota_inum)
+       };
 
        ENTRY;
        LASSERT(fid_is_acct(fid));
@@ -74,7 +78,7 @@ int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
                RETURN(-ENOENT);
 
        id->oii_gen = OSD_OII_NOGEN;
-       id->oii_ino = LDISKFS_SB(sb)->s_qf_inums[fid2type(fid)];
+       id->oii_ino = qf_inums[fid2type(fid)];
        if (!ldiskfs_valid_inum(sb, id->oii_ino))
                RETURN(-ENOENT);
        RETURN(0);
@@ -100,27 +104,44 @@ static int osd_acct_index_lookup(const struct lu_env *env,
                                 struct lustre_capa *capa)
 {
        struct osd_thread_info  *info = osd_oti_get(env);
+#ifdef HAVE_DQUOT_FS_DISK_QUOTA
+       struct fs_disk_quota    *dqblk = &info->oti_fdq;
+#else
        struct if_dqblk         *dqblk = &info->oti_dqblk;
+#endif
        struct super_block      *sb = osd_sb(osd_obj2dev(osd_dt_obj(dtobj)));
-       struct acct_rec         *rec = (struct acct_rec *)dtrec;
+       struct lquota_acct_rec  *rec = (struct lquota_acct_rec *)dtrec;
        __u64                    id = *((__u64 *)dtkey);
        int                      rc;
+#ifdef HAVE_DQUOT_KQID
+       struct kqid             qid;
+#endif
 
        ENTRY;
 
        memset((void *)dqblk, 0, sizeof(struct obd_dqblk));
+#ifdef HAVE_DQUOT_KQID
+       qid = make_kqid(&init_user_ns, obj2type(dtobj), id);
+       rc = sb->s_qcop->get_dqblk(sb, qid, dqblk);
+#else
        rc = sb->s_qcop->get_dqblk(sb, obj2type(dtobj), (qid_t) id, dqblk);
+#endif
        if (rc)
                RETURN(rc);
+#ifdef HAVE_DQUOT_FS_DISK_QUOTA
+       rec->bspace = dqblk->d_bcount;
+       rec->ispace = dqblk->d_icount;
+#else
        rec->bspace = dqblk->dqb_curspace;
        rec->ispace = dqblk->dqb_curinodes;
+#endif
        RETURN(+1);
 }
 
 #define QUOTA_IT_READ_ERROR(it, rc)                                    \
        CERROR("%s: Error while trying to read quota information, "    \
               "failed with %d\n",                                     \
-              it->oiq_obj->oo_dt.do_lu.lo_dev->ld_obd->obd_name, rc); \
+              osd_dev(it->oiq_obj->oo_dt.do_lu.lo_dev)->od_svname, rc); \
 
 /**
  * Initialize osd Iterator for given osd index object.
@@ -149,7 +170,7 @@ static struct dt_it *osd_it_acct_init(const struct lu_env *env,
        memset(it, 0, sizeof(*it));
        lu_object_get(lo);
        it->oiq_obj = obj;
-       CFS_INIT_LIST_HEAD(&it->oiq_list);
+       INIT_LIST_HEAD(&it->oiq_list);
 
        /* LUSTRE_DQTREEOFF is the initial offset where the tree can be found */
        it->oiq_blk[0] = LUSTRE_DQTREEOFF;
@@ -173,8 +194,8 @@ static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di)
 
        lu_object_put(env, &it->oiq_obj->oo_dt.do_lu);
 
-       cfs_list_for_each_entry_safe(leaf, tmp, &it->oiq_list, oql_link) {
-               cfs_list_del_init(&leaf->oql_link);
+       list_for_each_entry_safe(leaf, tmp, &it->oiq_list, oql_link) {
+               list_del_init(&leaf->oql_link);
                OBD_FREE_PTR(leaf);
        }
        EXIT;
@@ -241,9 +262,9 @@ static int osd_it_add_processed(struct osd_it_quota *it, int depth)
        OBD_ALLOC_PTR(leaf);
        if (leaf == NULL)
                RETURN(-ENOMEM);
-       CFS_INIT_LIST_HEAD(&leaf->oql_link);
+       INIT_LIST_HEAD(&leaf->oql_link);
        leaf->oql_blk = it->oiq_blk[depth];
-       cfs_list_add_tail(&leaf->oql_link, &it->oiq_list);
+       list_add_tail(&leaf->oql_link, &it->oiq_list);
        RETURN(0);
 }
 
@@ -415,3 +436,708 @@ const struct dt_index_operations osd_acct_index_ops = {
        }
 };
 
+static inline void osd_quota_swab(char *ptr, size_t size)
+{
+       int offset;
+
+       LASSERT((size & (sizeof(__u64) - 1)) == 0);
+
+       for (offset = 0; offset < size; offset += sizeof(__u64))
+            __swab64s((__u64 *)(ptr + offset));
+}
+
+const struct dt_rec *osd_quota_pack(struct osd_object *obj,
+                                   const struct dt_rec *rec,
+                                   union lquota_rec *quota_rec)
+{
+#ifdef __BIG_ENDIAN
+       struct iam_descr        *descr;
+
+       LASSERT(obj->oo_dir != NULL);
+       descr = obj->oo_dir->od_container.ic_descr;
+
+       memcpy(quota_rec, rec, descr->id_rec_size);
+
+       osd_quota_swab((char *)quota_rec, descr->id_rec_size);
+       return (const struct dt_rec *)quota_rec;
+#else
+       return rec;
+#endif
+}
+
+void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec)
+{
+#ifdef __BIG_ENDIAN
+       struct iam_descr *descr;
+
+       LASSERT(obj->oo_dir != NULL);
+       descr = obj->oo_dir->od_container.ic_descr;
+
+       osd_quota_swab((char *)rec, descr->id_rec_size);
+#else
+       return;
+#endif
+}
+
+static inline int osd_qid_type(struct osd_thandle *oh, int i)
+{
+       return (oh->ot_id_type & (1 << i)) ? GRPQUOTA : USRQUOTA;
+}
+
+static inline void osd_qid_set_type(struct osd_thandle *oh, int i, int type)
+{
+       oh->ot_id_type |= ((type == GRPQUOTA) ? (1 << i) : 0);
+}
+
+/**
+ * Reserve journal credits for quota files update first, then call
+ * ->op_begin() to perform quota enforcement.
+ *
+ * \param  env     - the environment passed by the caller
+ * \param  oh      - osd transaction handle
+ * \param  qi      - quota id & space required for this operation
+ * \param  obj     - osd object, could be NULL when it's under create
+ * \param  enforce - whether to perform quota enforcement
+ * \param  flags   - if the operation is write, return no user quota, no
+ *                   group quota, or sync commit flags to the caller
+ *
+ * \retval 0       - success
+ * \retval -ve     - failure
+ */
+int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
+                   struct lquota_id_info *qi, struct osd_object *obj,
+                   bool enforce, int *flags)
+{
+       struct osd_thread_info  *info = osd_oti_get(env);
+       struct osd_device       *dev = info->oti_dev;
+       struct qsd_instance     *qsd = dev->od_quota_slave;
+       struct inode            *inode = NULL;
+       int                      i, rc = 0, crd;
+       bool                     found = false;
+       ENTRY;
+
+       LASSERT(oh != NULL);
+       LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%d\n",
+                oh->ot_id_cnt);
+
+       for (i = 0; i < oh->ot_id_cnt; i++) {
+               if (oh->ot_id_array[i] == qi->lqi_id.qid_uid &&
+                   osd_qid_type(oh, i) == qi->lqi_type) {
+                       found = true;
+                       break;
+               }
+       }
+
+       if (!found) {
+               /* we need to account for credits for this new ID */
+               if (i >= OSD_MAX_UGID_CNT) {
+                       CERROR("Too many(%d) trans qids!\n", i + 1);
+                       RETURN(-EOVERFLOW);
+               }
+
+               if (obj != NULL)
+                       inode = obj->oo_inode;
+
+               /* root ID entry should be always present in the quota file */
+               if (qi->lqi_id.qid_uid == 0) {
+                       crd = 1;
+               } else {
+                       /* used space for this ID could be dropped to zero,
+                        * reserve extra credits for removing ID entry from
+                        * the quota file */
+                       if (qi->lqi_space < 0)
+                               crd = LDISKFS_QUOTA_DEL_BLOCKS(osd_sb(dev));
+                       /* reserve credits for adding ID entry to the quota
+                        * file if the i_dquot isn't initialized yet. */
+                       else if (inode == NULL ||
+                                inode->i_dquot[qi->lqi_type] == NULL)
+                               crd = LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(dev));
+                       else
+                               crd = 1;
+               }
+
+               osd_trans_declare_op(env, oh, OSD_OT_QUOTA, crd);
+
+               oh->ot_id_array[i] = qi->lqi_id.qid_uid;
+               osd_qid_set_type(oh, i, qi->lqi_type);
+               oh->ot_id_cnt++;
+       }
+
+       if (unlikely(qsd == NULL))
+               /* quota slave instance hasn't been allocated yet */
+               RETURN(0);
+
+       /* check quota */
+       if (enforce)
+               rc = qsd_op_begin(env, qsd, oh->ot_quota_trans, qi, flags);
+       RETURN(rc);
+}
+
+/**
+ * Wrapper for osd_declare_qid()
+ *
+ * \param  env    - the environment passed by the caller
+ * \param  uid    - user id of the inode
+ * \param  gid    - group id of the inode
+ * \param  space  - how many blocks/inodes will be consumed/released
+ * \param  oh     - osd transaction handle
+ * \param  obj    - osd object, could be NULL when it's under create
+ * \param  is_blk - block quota or inode quota?
+ * \param  flags  - if the operation is write, return no user quota, no
+ *                  group quota, or sync commit flags to the caller
+ * \param force   - set to 1 when changes are performed by root user and thus
+ *                  can't failed with EDQUOT
+ *
+ * \retval 0      - success
+ * \retval -ve    - failure
+ */
+int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
+                         long long space, struct osd_thandle *oh,
+                         struct osd_object *obj, bool is_blk, int *flags,
+                         bool force)
+{
+       struct osd_thread_info  *info = osd_oti_get(env);
+       struct lquota_id_info   *qi = &info->oti_qi;
+       int                      rcu, rcg; /* user & group rc */
+       ENTRY;
+
+       /* let's start with user quota */
+       qi->lqi_id.qid_uid = uid;
+       qi->lqi_type       = USRQUOTA;
+       qi->lqi_space      = space;
+       qi->lqi_is_blk     = is_blk;
+       rcu = osd_declare_qid(env, oh, qi, obj, true, flags);
+
+       if (force && (rcu == -EDQUOT || rcu == -EINPROGRESS))
+               /* ignore EDQUOT & EINPROGRESS when changes are done by root */
+               rcu = 0;
+
+       /* For non-fatal error, we want to continue to get the noquota flags
+        * for group id. This is only for commit write, which has @flags passed
+        * in. See osd_declare_write_commit().
+        * When force is set to true, we also want to proceed with the gid */
+       if (rcu && (rcu != -EDQUOT || flags == NULL))
+               RETURN(rcu);
+
+       /* and now group quota */
+       qi->lqi_id.qid_gid = gid;
+       qi->lqi_type       = GRPQUOTA;
+       rcg = osd_declare_qid(env, oh, qi, obj, true, flags);
+
+       if (force && (rcg == -EDQUOT || rcg == -EINPROGRESS))
+               /* as before, ignore EDQUOT & EINPROGRESS for root */
+               rcg = 0;
+
+       RETURN(rcu ? rcu : rcg);
+}
+
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0)
+/* Following code is used to migrate old admin quota files (in Linux quota
+ * file v2 format) into the new quota global indexes (in IAM format). */
+
+/* copied from osd_it_acct_get(), only changed the 'type' to -1 */
+static int osd_it_admin_get(const struct lu_env *env, struct dt_it *di,
+                           const struct dt_key *key)
+{
+       struct osd_it_quota     *it = (struct osd_it_quota *)di;
+       int                      type = -1;
+       qid_t                    dqid = *(qid_t *)key;
+       loff_t                   offset;
+       int                      rc;
+       ENTRY;
+
+       offset = find_tree_dqentry(env, it->oiq_obj, type, dqid,
+                                  LUSTRE_DQTREEOFF, 0, it);
+       if (offset > 0) { /* Found */
+               RETURN(+1);
+       } else if (offset < 0) { /* Error */
+               QUOTA_IT_READ_ERROR(it, (int)offset);
+               RETURN((int)offset);
+       }
+
+       /* The @key is not found, move to the first valid entry */
+       rc = walk_tree_dqentry(env, it->oiq_obj, type, it->oiq_blk[0], 0,
+                              0, it);
+       if (rc > 0)
+               /* no valid entry found */
+               rc = -ENOENT;
+       RETURN(rc);
+}
+
+static int osd_it_admin_load(const struct lu_env *env,
+                            const struct dt_it *di, __u64 hash)
+{
+       int rc;
+       ENTRY;
+
+       rc = osd_it_admin_get(env, (struct dt_it *)di,
+                             (const struct dt_key *)&hash);
+       RETURN(rc);
+}
+
+static int osd_it_admin_rec(const struct lu_env *env,
+                           const struct dt_it *di,
+                           struct dt_rec *dtrec, __u32 attr)
+{
+       struct osd_it_quota     *it = (struct osd_it_quota *)di;
+       struct lu_buf            buf;
+       loff_t                   pos;
+       int                      rc;
+       struct lustre_disk_dqblk_v2 *dqblk =
+               (struct lustre_disk_dqblk_v2 *)dtrec;
+       ENTRY;
+
+       buf.lb_buf = dqblk;
+       buf.lb_len = sizeof(*dqblk);
+
+       pos = it->oiq_offset;
+       rc = dt_record_read(env, &it->oiq_obj->oo_dt, &buf, &pos);
+       RETURN(rc);
+}
+
+/* copied from osd_it_acct_next(), only changed the 'type' to -1 */
+static int osd_it_admin_next(const struct lu_env *env, struct dt_it *di)
+{
+       struct osd_it_quota     *it = (struct osd_it_quota *)di;
+       int                      type = -1;
+       int                      depth, rc;
+       uint                     index;
+       ENTRY;
+
+       /* Let's first check if there are any remaining valid entry in the
+        * current leaf block. Start with the next entry after the current one.
+        */
+       depth = LUSTRE_DQTREEDEPTH;
+       index = it->oiq_index[depth];
+       if (++index < LUSTRE_DQSTRINBLK) {
+               /* Search for the next valid entry from current index */
+               rc = walk_block_dqentry(env, it->oiq_obj, type,
+                                       it->oiq_blk[depth], index, it);
+               if (rc < 0) {
+                       QUOTA_IT_READ_ERROR(it, rc);
+                       RETURN(rc);
+               } else if (rc == 0) {
+                       /* Found on entry, @it is already updated to the
+                        * new position in walk_block_dqentry(). */
+                       RETURN(0);
+               } else {
+                       rc = osd_it_add_processed(it, depth);
+                       if (rc)
+                               RETURN(rc);
+               }
+       } else {
+               rc = osd_it_add_processed(it, depth);
+               if (rc)
+                       RETURN(rc);
+       }
+       rc = 1;
+
+       /* We have consumed all the entries of the current leaf block, move on
+        * to the next one. */
+       depth--;
+
+       /* We keep searching as long as walk_tree_dqentry() returns +1
+        * (= no valid entry found). */
+       for (; depth >= 0 && rc > 0; depth--) {
+               index = it->oiq_index[depth];
+               if (++index > 0xff)
+                       continue;
+               rc = walk_tree_dqentry(env, it->oiq_obj, type,
+                                      it->oiq_blk[depth], depth, index, it);
+       }
+
+       if (rc < 0)
+               QUOTA_IT_READ_ERROR(it, rc);
+       RETURN(rc);
+}
+
+const struct dt_index_operations osd_admin_index_ops = {
+       .dio_lookup     = osd_acct_index_lookup,
+       .dio_it         = {
+               .init     = osd_it_acct_init,
+               .fini     = osd_it_acct_fini,
+               .get      = osd_it_admin_get,
+               .put      = osd_it_acct_put,
+               .next     = osd_it_admin_next,
+               .key      = osd_it_acct_key,
+               .key_size = osd_it_acct_key_size,
+               .rec      = osd_it_admin_rec,
+               .store    = osd_it_acct_store,
+               .load     = osd_it_admin_load
+       }
+};
+
+static int convert_quota_file(const struct lu_env *env,
+                             struct dt_object *old, struct dt_object *new,
+                             bool isblk)
+{
+       const struct dt_it_ops  *iops = &old->do_index_ops->dio_it;
+       struct osd_object       *obj;
+       struct lu_buf            buf;
+       struct dt_it            *it;
+       struct dt_key           *key;
+       __u32                    grace;
+       struct lquota_glb_rec   *glb_rec = NULL;
+       loff_t                   pos;
+       int                      rc;
+       struct lustre_disk_dqblk_v2     *dqblk = NULL;
+       struct lustre_disk_dqinfo       *dqinfo = NULL;
+       ENTRY;
+
+       obj = osd_dt_obj(old);
+       LASSERT(obj->oo_inode);
+
+       if (i_size_read(obj->oo_inode) == 0)
+               RETURN(0);
+
+       /* allocate buffers */
+       OBD_ALLOC_PTR(dqinfo);
+       if (dqinfo == NULL)
+               RETURN(-ENOMEM);
+
+       OBD_ALLOC_PTR(glb_rec);
+       if (glb_rec == NULL)
+               GOTO(out, rc = -ENOMEM);
+
+       OBD_ALLOC_PTR(dqblk);
+       if (dqblk == NULL)
+               GOTO(out, rc = -ENOMEM);
+
+       /* convert the old igrace/bgrace */
+       buf.lb_buf = dqinfo;
+       buf.lb_len = sizeof(*dqinfo);
+       pos = LUSTRE_DQINFOOFF;
+
+       rc = dt_record_read(env, old, &buf, &pos);
+       if (rc)
+               GOTO(out, rc);
+
+       /* keep it in little endian */
+       grace = isblk ? dqinfo->dqi_bgrace : dqinfo->dqi_igrace;
+       if (grace != 0) {
+               glb_rec->qbr_time = grace;
+               rc = lquota_disk_write_glb(env, new, 0, glb_rec);
+               if (rc)
+                       GOTO(out, rc);
+               glb_rec->qbr_time = 0;
+       }
+
+       /* iterate the old admin file, insert each record into the
+        * new index file. */
+       it = iops->init(env, old, 0, BYPASS_CAPA);
+       if (IS_ERR(it))
+               GOTO(out, rc = PTR_ERR(it));
+
+       rc = iops->load(env, it, 0);
+       if (rc == -ENOENT)
+               GOTO(out_it, rc = 0);
+       else if (rc < 0)
+               GOTO(out_it, rc);
+
+       do {
+               key = iops->key(env, it);
+               if (IS_ERR(key))
+                       GOTO(out_it, rc = PTR_ERR(key));
+
+               /* skip the root user/group */
+               if (*((__u64 *)key) == 0)
+                       goto next;
+
+               rc = iops->rec(env, it, (struct dt_rec *)dqblk, 0);
+               if (rc)
+                       GOTO(out_it, rc);
+
+               /* keep the value in little endian */
+               glb_rec->qbr_hardlimit = isblk ? dqblk->dqb_bhardlimit :
+                                                dqblk->dqb_ihardlimit;
+               glb_rec->qbr_softlimit = isblk ? dqblk->dqb_bsoftlimit :
+                                                dqblk->dqb_isoftlimit;
+
+               rc = lquota_disk_write_glb(env, new, *((__u64 *)key), glb_rec);
+               if (rc)
+                       GOTO(out_it, rc);
+next:
+               rc = iops->next(env, it);
+       } while (rc == 0);
+
+       /* reach the end */
+       if (rc > 0)
+               rc = 0;
+
+out_it:
+       iops->put(env, it);
+       iops->fini(env, it);
+out:
+       if (dqblk != NULL)
+               OBD_FREE_PTR(dqblk);
+       if (glb_rec != NULL)
+               OBD_FREE_PTR(glb_rec);
+       if (dqinfo != NULL)
+               OBD_FREE_PTR(dqinfo);
+       return rc;
+}
+
+/* Nobdy else can access the global index now, it's safe to truncate and
+ * reinitialize it */
+static int truncate_quota_index(const struct lu_env *env, struct dt_object *dt,
+                               const struct dt_index_features *feat)
+{
+       struct osd_device       *osd = osd_obj2dev(osd_dt_obj(dt));
+       struct thandle          *th;
+       struct lu_attr          *attr;
+       struct osd_thandle      *oth;
+       struct inode            *inode;
+       int                      rc;
+       struct iam_container    *bag = &(osd_dt_obj(dt))->oo_dir->od_container;
+       struct lu_buf           *lb = &osd_oti_get(env)->oti_buf;
+       ENTRY;
+
+       LASSERT(bag->ic_root_bh != NULL);
+       iam_container_fini(bag);
+
+       LASSERT(fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB);
+
+       OBD_ALLOC_PTR(attr);
+       if (attr == NULL)
+               RETURN(-ENOMEM);
+
+       attr->la_size = 0;
+       attr->la_valid = LA_SIZE;
+
+       th = dt_trans_create(env, &osd->od_dt_dev);
+       if (IS_ERR(th)) {
+               OBD_FREE_PTR(attr);
+               RETURN(PTR_ERR(th));
+       }
+
+       rc = dt_declare_punch(env, dt, 0, OBD_OBJECT_EOF, th);
+       if (rc)
+               GOTO(out, rc);
+
+       rc = dt_declare_attr_set(env, dt, attr, th);
+       if (rc)
+               GOTO(out, rc);
+
+       inode = osd_dt_obj(dt)->oo_inode;
+       LASSERT(inode);
+
+       /* iam_lfix_create() writes two blocks at the beginning */
+       lb->lb_len = osd_sb(osd)->s_blocksize * 2;
+       rc = dt_declare_record_write(env, dt, lb, 0, th);
+       if (rc)
+               GOTO(out, rc);
+
+       rc = dt_trans_start_local(env, &osd->od_dt_dev, th);
+       if (rc)
+               GOTO(out, rc);
+
+       dt_write_lock(env, dt, 0);
+       rc = dt_punch(env, dt, 0, OBD_OBJECT_EOF, th, BYPASS_CAPA);
+       if (rc)
+               GOTO(out_lock, rc);
+
+       rc = dt_attr_set(env, dt, attr, th, BYPASS_CAPA);
+       if (rc)
+               GOTO(out_lock, rc);
+
+       oth = container_of(th, struct osd_thandle, ot_super);
+
+       if (feat->dif_flags & DT_IND_VARKEY)
+               rc = iam_lvar_create(osd_dt_obj(dt)->oo_inode,
+                                    feat->dif_keysize_max,
+                                    feat->dif_ptrsize,
+                                    feat->dif_recsize_max, oth->ot_handle);
+       else
+               rc = iam_lfix_create(osd_dt_obj(dt)->oo_inode,
+                                    feat->dif_keysize_max,
+                                    feat->dif_ptrsize,
+                                    feat->dif_recsize_max, oth->ot_handle);
+out_lock:
+       dt_write_unlock(env, dt);
+out:
+       dt_trans_stop(env, &osd->od_dt_dev, th);
+       OBD_FREE_PTR(attr);
+
+       if (rc == 0) {
+               rc  = iam_container_setup(bag);
+               if (rc != 0)
+                       iam_container_fini(bag);
+       }
+       RETURN(rc);
+}
+
+static int set_quota_index_version(const struct lu_env *env,
+                                  struct dt_object *dt,
+                                  dt_obj_version_t version)
+{
+       struct osd_device       *osd = osd_obj2dev(osd_dt_obj(dt));
+       struct thandle          *th;
+       int                      rc;
+       ENTRY;
+
+       th = dt_trans_create(env, &osd->od_dt_dev);
+       if (IS_ERR(th))
+               RETURN(PTR_ERR(th));
+
+       rc = dt_declare_version_set(env, dt, th);
+       if (rc)
+               GOTO(out, rc);
+
+       rc = dt_trans_start_local(env, &osd->od_dt_dev, th);
+       if (rc)
+               GOTO(out, rc);
+
+       th->th_sync = 1;
+       dt_version_set(env, dt, version, th);
+out:
+       dt_trans_stop(env, &osd->od_dt_dev, th);
+       RETURN(rc);
+}
+
+int osd_quota_migration(const struct lu_env *env, struct dt_object *dt,
+                       const struct dt_index_features *feat)
+{
+       struct osd_thread_info  *oti = osd_oti_get(env);
+       struct osd_device       *osd = osd_obj2dev(osd_dt_obj(dt));
+       struct dt_object        *root, *parent = NULL, *admin = NULL;
+       dt_obj_version_t         version;
+       char                    *fname;
+       bool                     isblk = false, converted = false;
+       int                      rc;
+       ENTRY;
+
+       /* not newly created global index */
+       version = dt_version_get(env, dt);
+       if (version != 0)
+               RETURN(0);
+
+       /* locate root */
+       rc = dt_root_get(env, &osd->od_dt_dev, &oti->oti_fid);
+       if (rc) {
+               CERROR("%s: Can't get root FID, rc:%d\n", osd->od_svname, rc);
+               RETURN(rc);
+       }
+
+       root = dt_locate(env, &osd->od_dt_dev, &oti->oti_fid);
+       if (IS_ERR(root)) {
+               CERROR("%s: Failed to locate root "DFID", rc:%ld\n",
+                      osd->od_svname, PFID(&oti->oti_fid), PTR_ERR(root));
+               RETURN(PTR_ERR(root));
+       }
+
+       /* locate /OBJECTS */
+       rc = dt_lookup_dir(env, root, OBJECTS, &oti->oti_fid);
+       if (rc == -ENOENT) {
+               GOTO(out, rc = 0);
+       } else if (rc) {
+               CERROR("%s: Failed to lookup %s, rc:%d\n",
+                      osd->od_svname, OBJECTS, rc);
+               GOTO(out, rc);
+       }
+
+       parent = dt_locate(env, &osd->od_dt_dev, &oti->oti_fid);
+       if (IS_ERR(parent)) {
+               CERROR("%s: Failed to locate %s "DFID", rc:%ld\n",
+                      osd->od_svname, OBJECTS, PFID(&oti->oti_fid),
+                      PTR_ERR(parent));
+               GOTO(out, rc = PTR_ERR(parent));
+       }
+
+       /* locate quota admin file */
+       if (feat == &dt_quota_iusr_features) {
+               fname = ADMIN_USR;
+               isblk = false;
+       } else if (feat == &dt_quota_busr_features) {
+               fname = ADMIN_USR;
+               isblk = true;
+       } else if (feat == &dt_quota_igrp_features) {
+               fname = ADMIN_GRP;
+               isblk = false;
+       } else {
+               fname = ADMIN_GRP;
+               isblk = true;
+       }
+
+       rc = dt_lookup_dir(env, parent, fname, &oti->oti_fid);
+       if (rc == -ENOENT) {
+               GOTO(out, rc = 0);
+       } else if (rc) {
+               CERROR("%s: Failed to lookup %s, rc:%d\n",
+                      osd->od_svname, fname, rc);
+               GOTO(out, rc);
+       }
+
+       admin = dt_locate(env, &osd->od_dt_dev, &oti->oti_fid);
+       if (IS_ERR(admin)) {
+               CERROR("%s: Failed to locate %s "DFID", rc:%d\n",
+                      osd->od_svname, fname, PFID(&oti->oti_fid), rc);
+               GOTO(out, rc = PTR_ERR(admin));
+       }
+
+       if (!dt_object_exists(admin)) {
+               CERROR("%s: Old admin file %s doesn't exist, but is still "
+                      " referenced in parent directory.\n",
+                      osd->od_svname, fname);
+               GOTO(out, rc = -ENOENT);
+       }
+
+       /* truncate the new quota index file in case of any leftovers
+        * from last failed migration */
+       rc = truncate_quota_index(env, dt, feat);
+       if (rc) {
+               CERROR("%s: Failed to truncate the quota index "DFID", rc:%d\n",
+                      osd->od_svname, PFID(lu_object_fid(&dt->do_lu)), rc);
+               GOTO(out, rc);
+       }
+
+       /* set up indexing operations for the admin file */
+       admin->do_index_ops = &osd_admin_index_ops;
+
+       LCONSOLE_INFO("%s: Migrate %s quota from old admin quota file(%s) to "
+                     "new IAM quota index("DFID").\n", osd->od_svname,
+                     isblk ? "block" : "inode", fname,
+                     PFID(lu_object_fid(&dt->do_lu)));
+
+       /* iterate the admin quota file, and insert each record into
+        * the new index file */
+       rc = convert_quota_file(env, admin, dt, isblk);
+       if (rc)
+               CERROR("%s: Migrate old admin quota file(%s) failed, rc:%d\n",
+                      osd->od_svname, fname, rc);
+       converted = true;
+out:
+       /* if no migration happen, we need to set the default grace time. */
+       if (!converted && rc == 0) {
+               struct lquota_glb_rec *rec = &oti->oti_quota_rec.lqr_glb_rec;
+
+               rec->qbr_hardlimit = 0;
+               rec->qbr_softlimit = 0;
+               rec->qbr_granted = 0;
+               rec->qbr_time = isblk ? MAX_DQ_TIME : MAX_IQ_TIME;
+
+               rc = lquota_disk_write_glb(env, dt, 0, rec);
+               if (rc)
+                       CERROR("%s: Failed to set default grace time for "
+                              "index("DFID"), rc:%d\n", osd->od_svname,
+                              PFID(lu_object_fid(&dt->do_lu)), rc);
+       }
+
+       /* bump index version to 1 (or 2 if migration happened), so the
+        * migration will be skipped next time. */
+       if (rc == 0) {
+               rc = set_quota_index_version(env , dt, converted ? 2 : 1);
+               if (rc)
+                       CERROR("%s: Failed to set quota index("DFID") "
+                              "version, rc:%d\n", osd->od_svname,
+                              PFID(lu_object_fid(&dt->do_lu)), rc);
+       }
+
+       if (admin && !IS_ERR(admin))
+               lu_object_put(env, &admin->do_lu);
+       if (parent && !IS_ERR(parent))
+               lu_object_put(env, &parent->do_lu);
+       lu_object_put(env, &root->do_lu);
+
+       RETURN(rc);
+}
+#endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0) */