Whamcloud - gitweb
LU-2211 quota: cap how long a thread can wait for quota
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_handler.c
index 426717b..819d963 100644 (file)
@@ -72,8 +72,7 @@
 
 /* llo_* api support */
 #include <md_object.h>
-/* dt_acct_features */
-#include <lquota.h>
+#include <lustre_quota.h>
 
 #ifdef HAVE_LDISKFS_PDO
 int ldiskfs_pdo = 1;
@@ -151,11 +150,7 @@ static struct lu_object *osd_object_alloc(const struct lu_env *env,
 
                 l = &mo->oo_dt.do_lu;
                 dt_object_init(&mo->oo_dt, NULL, d);
-                if (osd_dev(d)->od_iop_mode)
-                        mo->oo_dt.do_ops = &osd_obj_ea_ops;
-                else
-                        mo->oo_dt.do_ops = &osd_obj_ops;
-
+               mo->oo_dt.do_ops = &osd_obj_ea_ops;
                 l->lo_ops = &osd_lu_obj_ops;
                 cfs_init_rwsem(&mo->oo_sem);
                 cfs_init_rwsem(&mo->oo_ext_idx_sem);
@@ -276,12 +271,10 @@ osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
                return inode;
 
        rc = osd_get_lma(inode, &info->oti_obj_dentry, lma);
+       if (rc == -ENODATA)
+               return inode;
+
        if (rc != 0) {
-               if (rc == -ENODATA) {
-                       CDEBUG(D_LFSCK, "inconsistent obj: NULL, %lu, "DFID"\n",
-                              inode->i_ino, PFID(fid));
-                       rc = -EREMCHG;
-               }
                iput(inode);
                return ERR_PTR(rc);
        }
@@ -290,8 +283,9 @@ osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
                CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
                       PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
                iput(inode);
-               return ERR_PTR(EREMCHG);
+               return ERR_PTR(-EREMCHG);
        }
+
        return inode;
 }
 
@@ -321,26 +315,27 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
        info = osd_oti_get(env);
        LASSERT(info);
        oic = &info->oti_cache;
-       id  = &oic->oic_lid;
 
        if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
                RETURN(-ENOENT);
 
-       if (fid_is_norm(fid)) {
-               /* Search order: 1. per-thread cache. */
-               if (lu_fid_eq(fid, &oic->oic_fid)) {
-                       goto iget;
-               } else if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
-                       /* Search order: 2. OI scrub pending list. */
-                       result = osd_oii_lookup(dev, fid, id);
-                       if (result == 0)
-                               goto iget;
-               }
+       /* Search order: 1. per-thread cache. */
+       if (lu_fid_eq(fid, &oic->oic_fid)) {
+               id = &oic->oic_lid;
+               goto iget;
+       }
 
-               if (sf->sf_flags & SF_INCONSISTENT)
-                       verify = 1;
+       id = &info->oti_id;
+       if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
+               /* Search order: 2. OI scrub pending list. */
+               result = osd_oii_lookup(dev, fid, id);
+               if (result == 0)
+                       goto iget;
        }
 
+       if (sf->sf_flags & SF_INCONSISTENT)
+               verify = 1;
+
        /*
         * Objects are created as locking anchors or place holders for objects
         * yet to be created. No need to osd_oi_lookup() at here because FID
@@ -380,7 +375,7 @@ iget:
 trigger:
                        if (thread_is_running(&scrub->os_thread)) {
                                result = -EINPROGRESS;
-                       } else if (!scrub->os_no_scrub) {
+                       } else if (!dev->od_noscrub) {
                                result = osd_scrub_start(dev);
                                LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
                                               "for "DFID", rc = %d [1]\n",
@@ -398,10 +393,9 @@ trigger:
 
         obj->oo_inode = inode;
         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
-        if (dev->od_iop_mode) {
-                obj->oo_compat_dot_created = 1;
-                obj->oo_compat_dotdot_created = 1;
-        }
+
+       obj->oo_compat_dot_created = 1;
+       obj->oo_compat_dotdot_created = 1;
 
         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
                GOTO(out, result = 0);
@@ -592,12 +586,12 @@ static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 /*
  * Concurrency: doesn't access mutable data.
  */
-static int osd_param_is_sane(const struct osd_device *dev,
-                             const struct thandle *th)
+static int osd_param_is_not_sane(const struct osd_device *dev,
+                                const struct thandle *th)
 {
-        struct osd_thandle *oh;
-        oh = container_of0(th, struct osd_thandle, ot_super);
-        return oh->ot_credits <= osd_journal(dev)->j_max_transaction_buffers;
+       struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
+
+       return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
 }
 
 /*
@@ -693,30 +687,48 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d,
         if (rc != 0)
                 GOTO(out, rc);
 
-        if (!osd_param_is_sane(dev, th)) {
+       if (unlikely(osd_param_is_not_sane(dev, th))) {
+               static unsigned long last_printed;
+               static int last_credits;
+
                CWARN("%.16s: too many transaction credits (%d > %d)\n",
                      LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
                      oh->ot_credits,
                      osd_journal(dev)->j_max_transaction_buffers);
-                /* XXX Limit the credits to 'max_transaction_buffers', and
-                 *     let the underlying filesystem to catch the error if
-                 *     we really need so many credits.
-                 *
-                 *     This should be removed when we can calculate the
-                 *     credits precisely. */
-                oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
 #ifdef OSD_TRACK_DECLARES
-                CERROR("  attr_set: %d, punch: %d, xattr_set: %d,\n",
-                       oh->ot_declare_attr_set, oh->ot_declare_punch,
-                       oh->ot_declare_xattr_set);
-                CERROR("  create: %d, ref_add: %d, ref_del: %d, write: %d\n",
-                       oh->ot_declare_create, oh->ot_declare_ref_add,
-                       oh->ot_declare_ref_del, oh->ot_declare_write);
-                CERROR("  insert: %d, delete: %d, destroy: %d\n",
-                       oh->ot_declare_insert, oh->ot_declare_delete,
-                       oh->ot_declare_destroy);
+               CWARN("  create: %u/%u, delete: %u/%u, destroy: %u/%u\n",
+                     oh->ot_declare_create, oh->ot_declare_create_cred,
+                     oh->ot_declare_delete, oh->ot_declare_delete_cred,
+                     oh->ot_declare_destroy, oh->ot_declare_destroy_cred);
+               CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
+                     oh->ot_declare_attr_set, oh->ot_declare_attr_set_cred,
+                     oh->ot_declare_xattr_set, oh->ot_declare_xattr_set_cred);
+               CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
+                     oh->ot_declare_write, oh->ot_declare_write_cred,
+                     oh->ot_declare_punch, oh->ot_declare_punch_cred,
+                     oh->ot_declare_quota, oh->ot_declare_quota_cred);
+               CWARN("  insert: %u/%u, delete: %u/%u\n",
+                     oh->ot_declare_insert, oh->ot_declare_insert_cred,
+                     oh->ot_declare_delete, oh->ot_declare_destroy_cred);
+               CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
+                     oh->ot_declare_ref_add, oh->ot_declare_ref_add_cred,
+                     oh->ot_declare_ref_del, oh->ot_declare_ref_del_cred);
+
+               if (last_credits != oh->ot_credits &&
+                   time_after(jiffies, last_printed + 60 * HZ)) {
+                       libcfs_debug_dumpstack(NULL);
+                       last_credits = oh->ot_credits;
+                       last_printed = jiffies;
+               }
 #endif
-        }
+               /* XXX Limit the credits to 'max_transaction_buffers', and
+                *     let the underlying filesystem to catch the error if
+                *     we really need so many credits.
+                *
+                *     This should be removed when we can calculate the
+                *     credits precisely. */
+               oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
+       }
 
         /*
          * XXX temporary stuff. Some abstraction layer should
@@ -840,8 +852,24 @@ static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 
         osd_index_fini(obj);
         if (inode != NULL) {
+               struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
+               qid_t                    uid = inode->i_uid;
+               qid_t                    gid = inode->i_gid;
+
                 iput(inode);
                 obj->oo_inode = NULL;
+
+               if (qsd != NULL) {
+                       struct osd_thread_info  *info = osd_oti_get(env);
+                       struct lquota_id_info   *qi = &info->oti_qi;
+
+                       /* Release granted quota to master if necessary */
+                       qi->lqi_id.qid_uid = uid;
+                       qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
+
+                       qi->lqi_id.qid_uid = gid;
+                       qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
+               }
         }
 }
 
@@ -904,6 +932,8 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d,
                if (likely(result == 0)) { /* N.B. statfs can't really fail */
                        osd->od_osfs_age = cfs_time_current_64();
                        statfs_pack(&osd->od_statfs, ksfs);
+                       if (sb->s_flags & MS_RDONLY)
+                               sfs->os_state = OS_STATE_READONLY;
                }
        }
 
@@ -917,6 +947,17 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d,
         return result;
 }
 
+/**
+ * Estimate space needed for file creations. We assume the largest filename
+ * which is 2^64 - 1, hence a filename of 20 chars.
+ * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
+ */
+#ifdef __LDISKFS_DIR_REC_LEN
+#define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
+#else
+#define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
+#endif
+
 /*
  * Concurrency: doesn't access mutable data.
  */
@@ -934,6 +975,15 @@ static void osd_conf_get(const struct lu_env *env,
         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
        param->ddp_block_shift  = sb->s_blocksize_bits;
        param->ddp_mount_type     = LDD_MT_LDISKFS;
+       param->ddp_maxbytes       = sb->s_maxbytes;
+       /* Overhead estimate should be fairly accurate, so we really take a tiny
+        * error margin which also avoids fragmenting the filesystem too much */
+       param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
+       /* inode are statically allocated, so per-inode space consumption
+        * is the space consumed by the directory entry */
+       param->ddp_inodespace     = PER_OBJ_USAGE;
+       /* per-fragment overhead to be used by the client code */
+       param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
         param->ddp_mntopts      = 0;
         if (test_opt(sb, XATTR_USER))
                 param->ddp_mntopts |= MNTOPT_USERXATTR;
@@ -1027,25 +1077,6 @@ static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
 }
 
 /**
- * Concurrency: serialization provided by callers.
- */
-static void osd_init_quota_ctxt(const struct lu_env *env, struct dt_device *d,
-                               struct dt_quota_ctxt *ctxt, void *data)
-{
-        struct obd_device *obd = (void *)ctxt;
-        struct vfsmount *mnt = (struct vfsmount *)data;
-        ENTRY;
-
-        obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
-        OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
-        obd->obd_lvfs_ctxt.pwdmnt = mnt;
-        obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
-        obd->obd_lvfs_ctxt.fs = get_ds();
-
-        EXIT;
-}
-
-/**
  * Note: we do not count into QUOTA here.
  * If we mount with --data_journal we may need more.
  */
@@ -1113,7 +1144,6 @@ static const struct dt_device_operations osd_dt_ops = {
         .dt_ro             = osd_ro,
         .dt_commit_async   = osd_commit_async,
         .dt_init_capa_ctxt = osd_init_capa_ctxt,
-        .dt_init_quota_ctxt= osd_init_quota_ctxt,
 };
 
 static void osd_object_read_lock(const struct lu_env *env,
@@ -1362,8 +1392,8 @@ static int osd_declare_attr_set(const struct lu_env *env,
        oh = container_of0(handle, struct osd_thandle, ot_super);
        LASSERT(oh->ot_handle == NULL);
 
-       OSD_DECLARE_OP(oh, attr_set);
-       oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
+       OSD_DECLARE_OP(oh, attr_set,
+                      osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
 
        if (attr == NULL || obj->oo_inode == NULL)
                RETURN(rc);
@@ -1571,6 +1601,7 @@ static int osd_attr_set(const struct lu_env *env,
         OSD_EXEC_OP(handle, attr_set);
 
         inode = obj->oo_inode;
+       ll_vfs_dq_init(inode);
 
        rc = osd_quota_transfer(inode, attr);
        if (rc)
@@ -1654,7 +1685,6 @@ static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
 {
         int result;
         struct osd_thandle *oth;
-        struct osd_device *osd = osd_obj2dev(obj);
         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
 
         LASSERT(S_ISDIR(attr->la_mode));
@@ -1662,16 +1692,7 @@ static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
         oth = container_of(th, struct osd_thandle, ot_super);
         LASSERT(oth->ot_handle->h_transaction != NULL);
         result = osd_mkfile(info, obj, mode, hint, th);
-        if (result == 0 && osd->od_iop_mode == 0) {
-                LASSERT(obj->oo_inode != NULL);
-                /*
-                 * XXX uh-oh... call low-level iam function directly.
-                 */
 
-                result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
-                                         sizeof (struct osd_fid_pack),
-                                         oth->ot_handle);
-        }
         return result;
 }
 
@@ -1898,42 +1919,40 @@ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
 }
 
 static int osd_declare_object_create(const struct lu_env *env,
-                                     struct dt_object *dt,
-                                     struct lu_attr *attr,
-                                     struct dt_allocation_hint *hint,
-                                     struct dt_object_format *dof,
-                                     struct thandle *handle)
+                                    struct dt_object *dt,
+                                    struct lu_attr *attr,
+                                    struct dt_allocation_hint *hint,
+                                    struct dt_object_format *dof,
+                                    struct thandle *handle)
 {
        struct osd_thandle      *oh;
        int                      rc;
        ENTRY;
 
-        LASSERT(handle != NULL);
+       LASSERT(handle != NULL);
 
-        oh = container_of0(handle, struct osd_thandle, ot_super);
-        LASSERT(oh->ot_handle == NULL);
+       oh = container_of0(handle, struct osd_thandle, ot_super);
+       LASSERT(oh->ot_handle == NULL);
 
-        OSD_DECLARE_OP(oh, create);
-        oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
-        /* XXX: So far, only normal fid needs be inserted into the oi,
-         *      things could be changed later. Revise following code then. */
-        if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
-                OSD_DECLARE_OP(oh, insert);
-                oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
+       OSD_DECLARE_OP(oh, create, osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
+       /* XXX: So far, only normal fid needs be inserted into the oi,
+        *      things could be changed later. Revise following code then. */
+       if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
                /* Reuse idle OI block may cause additional one OI block
                 * to be changed. */
-               oh->ot_credits += 1;
-        }
-        /* If this is directory, then we expect . and .. to be inserted as
-         * well. The one directory block always needs to be created for the
-         * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
-         * block), there is no danger of needing a tree for the first block.
-         */
-        if (attr && S_ISDIR(attr->la_mode)) {
-                OSD_DECLARE_OP(oh, insert);
-                OSD_DECLARE_OP(oh, insert);
-                oh->ot_credits += osd_dto_credits_noquota[DTO_WRITE_BASE];
-        }
+               OSD_DECLARE_OP(oh, insert,
+                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
+       }
+       /* If this is directory, then we expect . and .. to be inserted as
+        * well. The one directory block always needs to be created for the
+        * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
+        * block), there is no danger of needing a tree for the first block.
+        */
+       if (attr && S_ISDIR(attr->la_mode)) {
+               OSD_DECLARE_OP(oh, insert,
+                              osd_dto_credits_noquota[DTO_WRITE_BASE]);
+               OSD_DECLARE_OP(oh, insert, 0);
+       }
 
        if (!attr)
                RETURN(0);
@@ -1983,40 +2002,36 @@ static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
  * Concurrency: must be locked
  */
 static int osd_declare_object_destroy(const struct lu_env *env,
-                                      struct dt_object *dt,
-                                      struct thandle *th)
+                                     struct dt_object *dt,
+                                     struct thandle *th)
 {
-        struct osd_object  *obj = osd_dt_obj(dt);
-        struct inode       *inode = obj->oo_inode;
-        struct osd_thandle *oh;
-       int                 rc;
-        ENTRY;
+       struct osd_object  *obj = osd_dt_obj(dt);
+       struct inode       *inode = obj->oo_inode;
+       struct osd_thandle *oh;
+       int                 rc;
+       ENTRY;
 
-        oh = container_of0(th, struct osd_thandle, ot_super);
-        LASSERT(oh->ot_handle == NULL);
-        LASSERT(inode);
+       oh = container_of0(th, struct osd_thandle, ot_super);
+       LASSERT(oh->ot_handle == NULL);
+       LASSERT(inode);
 
-        OSD_DECLARE_OP(oh, destroy);
-        OSD_DECLARE_OP(oh, delete);
-        oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_DELETE];
-        /* XXX: So far, only normal fid needs to be inserted into the OI,
-         *      so only normal fid needs to be removed from the OI also. */
-        if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
-               oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
-               /* Recycle idle OI leaf may cause additional three OI blocks
-                * to be changed. */
-               oh->ot_credits += 3;
-        }
+       OSD_DECLARE_OP(oh, delete, osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
+       /* XXX: So far, only normal fid needs to be inserted into the OI,
+        *      so only normal fid needs to be removed from the OI also.
+        * Recycle idle OI leaf may cause additional three OI blocks
+        * to be changed. */
+       OSD_DECLARE_OP(oh, destroy, fid_is_norm(lu_object_fid(&dt->do_lu)) ?
+                       osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3 : 0);
 
        /* one less inode */
-        rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
+       rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
                                   false, true, NULL, false);
        if (rc)
                RETURN(rc);
        /* data to be truncated */
-       rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh, true,
-                                  true, NULL, false);
-        RETURN(rc);
+       rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
+                                  true, true, NULL, false);
+       RETURN(rc);
 }
 
 static int osd_object_destroy(const struct lu_env *env,
@@ -2089,6 +2104,7 @@ static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
         if (fl & LU_XATTR_CREATE)
                 fs_flags |= XATTR_CREATE;
 
+       ll_vfs_dq_init(inode);
         dentry->d_inode = inode;
         rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
                                    buf->lb_len, fs_flags);
@@ -2220,10 +2236,9 @@ static int osd_declare_object_ref_add(const struct lu_env *env,
         oh = container_of0(handle, struct osd_thandle, ot_super);
         LASSERT(oh->ot_handle == NULL);
 
-        OSD_DECLARE_OP(oh, ref_add);
-        oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
+       OSD_DECLARE_OP(oh, ref_add, osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
 
-        return 0;
+       return 0;
 }
 
 /*
@@ -2279,10 +2294,9 @@ static int osd_declare_object_ref_del(const struct lu_env *env,
         oh = container_of0(handle, struct osd_thandle, ot_super);
         LASSERT(oh->ot_handle == NULL);
 
-        OSD_DECLARE_OP(oh, ref_del);
-        oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
+       OSD_DECLARE_OP(oh, ref_del, osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
 
-        return 0;
+       return 0;
 }
 
 /*
@@ -2374,11 +2388,9 @@ static int osd_declare_xattr_set(const struct lu_env *env,
        oh = container_of0(handle, struct osd_thandle, ot_super);
        LASSERT(oh->ot_handle == NULL);
 
-       OSD_DECLARE_OP(oh, xattr_set);
-       if (strcmp(name, XATTR_NAME_VERSION) == 0)
-               oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
-       else
-               oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
+       OSD_DECLARE_OP(oh, xattr_set, strcmp(name, XATTR_NAME_VERSION) == 0 ?
+                      osd_dto_credits_noquota[DTO_ATTR_SET_BASE] :
+                      osd_dto_credits_noquota[DTO_XATTR_SET]);
 
        return 0;
 }
@@ -2460,10 +2472,9 @@ static int osd_declare_xattr_del(const struct lu_env *env,
         oh = container_of0(handle, struct osd_thandle, ot_super);
         LASSERT(oh->ot_handle == NULL);
 
-        OSD_DECLARE_OP(oh, xattr_set);
-        oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
+       OSD_DECLARE_OP(oh, xattr_set, osd_dto_credits_noquota[DTO_XATTR_SET]);
 
-        return 0;
+       return 0;
 }
 
 /*
@@ -2489,6 +2500,7 @@ static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
 
         OSD_EXEC_OP(handle, xattr_set);
 
+       ll_vfs_dq_init(inode);
         dentry->d_inode = inode;
         rc = inode->i_op->removexattr(dentry, name);
         return rc;
@@ -2676,7 +2688,6 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
        int                      result;
        int                      skip_iam = 0;
        struct osd_object       *obj = osd_dt_obj(dt);
-       struct osd_device       *osd = osd_obj2dev(obj);
 
         LINVRNT(osd_invariant(obj));
         LASSERT(dt_object_exists(dt));
@@ -2684,7 +2695,7 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
         if (osd_object_is_root(obj)) {
                 dt->do_index_ops = &osd_index_ea_ops;
                 result = 0;
-        } else if (feat == &dt_directory_features && osd->od_iop_mode) {
+       } else if (feat == &dt_directory_features) {
                 dt->do_index_ops = &osd_index_ea_ops;
                 if (S_ISDIR(obj->oo_inode->i_mode))
                         result = 0;
@@ -2833,10 +2844,9 @@ static int osd_index_declare_iam_delete(const struct lu_env *env,
         oh = container_of0(handle, struct osd_thandle, ot_super);
         LASSERT(oh->ot_handle == NULL);
 
-        OSD_DECLARE_OP(oh, delete);
-        oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
+       OSD_DECLARE_OP(oh, delete, osd_dto_credits_noquota[DTO_INDEX_DELETE]);
 
-        return 0;
+       return 0;
 }
 
 /**
@@ -2896,23 +2906,22 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
 }
 
 static int osd_index_declare_ea_delete(const struct lu_env *env,
-                                       struct dt_object *dt,
-                                       const struct dt_key *key,
-                                       struct thandle *handle)
+                                      struct dt_object *dt,
+                                      const struct dt_key *key,
+                                      struct thandle *handle)
 {
-        struct osd_thandle *oh;
+       struct osd_thandle *oh;
        struct inode       *inode;
        int                 rc;
        ENTRY;
 
-        LASSERT(dt_object_exists(dt));
-        LASSERT(handle != NULL);
+       LASSERT(dt_object_exists(dt));
+       LASSERT(handle != NULL);
 
-        oh = container_of0(handle, struct osd_thandle, ot_super);
-        LASSERT(oh->ot_handle == NULL);
+       oh = container_of0(handle, struct osd_thandle, ot_super);
+       LASSERT(oh->ot_handle == NULL);
 
-        OSD_DECLARE_OP(oh, delete);
-        oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
+       OSD_DECLARE_OP(oh, delete, osd_dto_credits_noquota[DTO_INDEX_DELETE]);
 
        inode = osd_dt_obj(dt)->oo_inode;
        LASSERT(inode);
@@ -2974,6 +2983,7 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
                 RETURN(-EACCES);
 
+       ll_vfs_dq_init(dir);
         dentry = osd_child_dentry_get(env, obj,
                                       (char *)key, strlen((char *)key));
 
@@ -3086,10 +3096,9 @@ static int osd_index_declare_iam_insert(const struct lu_env *env,
         oh = container_of0(handle, struct osd_thandle, ot_super);
         LASSERT(oh->ot_handle == NULL);
 
-        OSD_DECLARE_OP(oh, insert);
-        oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
+       OSD_DECLARE_OP(oh, insert, osd_dto_credits_noquota[DTO_INDEX_INSERT]);
 
-        return 0;
+       return 0;
 }
 
 /**
@@ -3193,6 +3202,8 @@ static int __osd_ea_add_rec(struct osd_thread_info *info,
         } else {
                 child->d_fsdata = NULL;
         }
+       LASSERT(pobj->oo_inode);
+       ll_vfs_dq_init(pobj->oo_inode);
         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
 
         RETURN(rc);
@@ -3308,7 +3319,7 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
         return rc;
 }
 
-static int
+static void
 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
                      struct osd_idmap_cache *oic)
 {
@@ -3319,13 +3330,16 @@ osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
        int                  rc;
        ENTRY;
 
+       if (!fid_is_norm(fid) && !fid_is_igif(fid))
+               RETURN_EXIT;
+
 again:
        rc = osd_oi_lookup(oti, dev, fid, id);
        if (rc != 0 && rc != -ENOENT)
-               RETURN(rc);
+               RETURN_EXIT;
 
        if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
-               RETURN(0);
+               RETURN_EXIT;
 
        if (thread_is_running(&scrub->os_thread)) {
                rc = osd_oii_insert(dev, oic, rc == -ENOENT);
@@ -3336,10 +3350,10 @@ again:
                if (unlikely(rc == -EAGAIN))
                        goto again;
 
-               RETURN(rc);
+               RETURN_EXIT;
        }
 
-       if (!scrub->os_no_scrub && ++once == 1) {
+       if (!dev->od_noscrub && ++once == 1) {
                CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
                       PFID(fid));
                rc = osd_scrub_start(dev);
@@ -3351,7 +3365,7 @@ again:
                        goto again;
        }
 
-       RETURN(rc = -EREMCHG);
+       EXIT;
 }
 
 /**
@@ -3404,7 +3418,7 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
                        rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
                else
                        osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
-               if (rc != 0 || !fid_is_norm(fid)) {
+               if (rc != 0) {
                        fid_zero(&oic->oic_fid);
                        GOTO(out, rc);
                }
@@ -3414,7 +3428,7 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
                    (sf->sf_flags & SF_INCONSISTENT ||
                     ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
                                      sf->sf_oi_bitmap)))
-                       rc = osd_consistency_check(oti, dev, oic);
+                       osd_consistency_check(oti, dev, oic);
        } else {
                rc = -ENOENT;
        }
@@ -3497,24 +3511,23 @@ static inline void osd_object_put(const struct lu_env *env,
 }
 
 static int osd_index_declare_ea_insert(const struct lu_env *env,
-                                       struct dt_object *dt,
-                                       const struct dt_rec *rec,
-                                       const struct dt_key *key,
-                                       struct thandle *handle)
+                                      struct dt_object *dt,
+                                      const struct dt_rec *rec,
+                                      const struct dt_key *key,
+                                      struct thandle *handle)
 {
-        struct osd_thandle *oh;
+       struct osd_thandle *oh;
        struct inode       *inode;
        int                 rc;
        ENTRY;
 
-        LASSERT(dt_object_exists(dt));
-        LASSERT(handle != NULL);
+       LASSERT(dt_object_exists(dt));
+       LASSERT(handle != NULL);
 
-        oh = container_of0(handle, struct osd_thandle, ot_super);
-        LASSERT(oh->ot_handle == NULL);
+       oh = container_of0(handle, struct osd_thandle, ot_super);
+       LASSERT(oh->ot_handle == NULL);
 
-        OSD_DECLARE_OP(oh, insert);
-        oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
+       OSD_DECLARE_OP(oh, insert, osd_dto_credits_noquota[DTO_INDEX_INSERT]);
 
        inode = osd_dt_obj(dt)->oo_inode;
        LASSERT(inode);
@@ -3849,6 +3862,7 @@ static const struct dt_index_operations osd_index_iam_ops = {
         }
 };
 
+
 /**
  * Creates or initializes iterator context.
  *
@@ -4136,8 +4150,10 @@ static inline int osd_it_ea_rec(const struct lu_env *env,
 
        if (!fid_is_sane(fid)) {
                rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
-               if (rc != 0)
+               if (rc != 0) {
+                       fid_zero(&oic->oic_fid);
                        RETURN(rc);
+               }
        } else {
                osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
        }
@@ -4146,16 +4162,11 @@ static inline int osd_it_ea_rec(const struct lu_env *env,
                           it->oie_dirent->oied_name,
                           it->oie_dirent->oied_namelen,
                           it->oie_dirent->oied_type, attr);
-       if (!fid_is_norm(fid)) {
-               fid_zero(&oic->oic_fid);
-               RETURN(0);
-       }
-
        oic->oic_fid = *fid;
        if ((scrub->os_pos_current <= ino) &&
            (sf->sf_flags & SF_INCONSISTENT ||
             ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
-               rc = osd_consistency_check(oti, dev, oic);
+               osd_consistency_check(oti, dev, oic);
 
        RETURN(rc);
 }
@@ -4225,7 +4236,6 @@ static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
                 return -EACCES;
 
         rc = osd_ea_lookup_rec(env, obj, rec, key);
-
         if (rc == 0)
                 rc = +1;
         RETURN(rc);
@@ -4422,13 +4432,8 @@ static int osd_mount(const struct lu_env *env,
                GOTO(out, rc = -EINVAL);
        }
 
-       if (lmd_flags & LMD_FLG_IAM) {
-                o->od_iop_mode = 0;
-               LCONSOLE_WARN("%s: OSD: IAM mode enabled\n", name);
-        } else
-                o->od_iop_mode = 1;
        if (lmd_flags & LMD_FLG_NOSCRUB)
-               o->od_scrub.os_no_scrub = 1;
+               o->od_noscrub = 1;
 
 out:
        if (__page)
@@ -4494,6 +4499,7 @@ static int osd_device_init0(const struct lu_env *env,
 
        o->od_read_cache = 1;
        o->od_writethrough_cache = 1;
+       o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
 
        rc = osd_mount(env, o, cfg);
        if (rc)
@@ -4529,7 +4535,18 @@ static int osd_device_init0(const struct lu_env *env,
 
        LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
 
+       /* initialize quota slave instance */
+       o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
+                                    o->od_proc_entry);
+       if (IS_ERR(o->od_quota_slave)) {
+               rc = PTR_ERR(o->od_quota_slave);
+               o->od_quota_slave = NULL;
+               GOTO(out_procfs, rc);
+       }
+
        RETURN(0);
+out_procfs:
+       osd_procfs_fini(o);
 out_site:
        lu_site_fini(&o->od_site);
 out_compat:
@@ -4616,7 +4633,17 @@ static int osd_process_config(const struct lu_env *env,
 static int osd_recovery_complete(const struct lu_env *env,
                                  struct lu_device *d)
 {
-        RETURN(0);
+       struct osd_device       *osd = osd_dev(d);
+       int                      rc = 0;
+       ENTRY;
+
+       if (osd->od_quota_slave == NULL)
+               RETURN(0);
+
+       /* start qsd instance on recovery completion, this notifies the quota
+        * slave code that we are about to process new requests now */
+       rc = qsd_start(env, osd->od_quota_slave);
+       RETURN(rc);
 }
 
 /*
@@ -4687,13 +4714,9 @@ static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
                        RETURN(result);
        }
 
-       /* 2. setup quota slave instance */
-       osd->od_quota_slave = qsd_init(env, osd->od_svname, &osd->od_dt_dev,
-                                      osd->od_proc_entry);
-       if (IS_ERR(osd->od_quota_slave)) {
-               result = PTR_ERR(osd->od_quota_slave);
-               osd->od_quota_slave = NULL;
-       }
+       if (osd->od_quota_slave != NULL)
+               /* set up quota slave objects */
+               result = qsd_prepare(env, osd->od_quota_slave);
 
        RETURN(result);
 }