Whamcloud - gitweb
LU-14121 nodemap: do not force fsuid/fsgid squashing
[fs/lustre-release.git] / lustre / mdt / mdt_lib.c
index 744b5e2..be2531a 100644 (file)
@@ -23,7 +23,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2016, Intel Corporation.
+ * Copyright (c) 2011, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -45,9 +45,8 @@
 #define DEBUG_SUBSYSTEM S_MDS
 
 #include <linux/user_namespace.h>
-#ifdef HAVE_UIDGID_HEADER
-# include <linux/uidgid.h>
-#endif
+#include <linux/uidgid.h>
+
 #include "mdt_internal.h"
 #include <uapi/linux/lnet/nidstr.h>
 #include <lustre_nodemap.h>
@@ -84,15 +83,15 @@ void mdt_exit_ucred(struct mdt_thread_info *info)
        }
 }
 
-static int match_nosquash_list(struct rw_semaphore *sem,
+static int match_nosquash_list(struct spinlock *rsi_lock,
                               struct list_head *nidlist,
                               lnet_nid_t peernid)
 {
        int rc;
        ENTRY;
-       down_read(sem);
+       spin_lock(rsi_lock);
        rc = cfs_match_nid(peernid, nidlist);
-       up_read(sem);
+       spin_unlock(rsi_lock);
        RETURN(rc);
 }
 
@@ -107,7 +106,7 @@ static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
        if (!squash->rsi_uid || ucred->uc_fsuid)
                RETURN(0);
 
-       if (match_nosquash_list(&squash->rsi_sem,
+       if (match_nosquash_list(&squash->rsi_lock,
                                &squash->rsi_nosquash_nids,
                                peernid)) {
                CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
@@ -141,8 +140,33 @@ static void ucred_set_jobid(struct mdt_thread_info *info, struct lu_ucred *uc)
                uc->uc_jobid[0] = '\0';
 }
 
+static void ucred_set_nid(struct mdt_thread_info *info, struct lu_ucred *uc)
+{
+       if (info && info->mti_exp && info->mti_exp->exp_connection)
+               uc->uc_nid = info->mti_exp->exp_connection->c_peer.nid;
+       else
+               uc->uc_nid = LNET_NID_ANY;
+}
+
+static void ucred_set_audit_enabled(struct mdt_thread_info *info,
+                                   struct lu_ucred *uc)
+{
+       struct lu_nodemap *nodemap = NULL;
+       bool audit = true;
+
+       if (info && info->mti_exp) {
+               nodemap = nodemap_get_from_exp(info->mti_exp);
+               if (nodemap && !IS_ERR(nodemap)) {
+                       audit = nodemap->nmf_enable_audit;
+                       nodemap_putref(nodemap);
+               }
+       }
+
+       uc->uc_enable_audit = audit;
+}
+
 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
-                         void *buf, bool drop_fs_cap)
+                         void *buf)
 {
        struct ptlrpc_request *req = mdt_info_req(info);
        struct mdt_device *mdt = info->mti_mdt;
@@ -153,6 +177,7 @@ static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
        __u32 perm = 0;
        int setuid;
        int setgid;
+       bool is_nm_gid_squashed = false;
        int rc = 0;
 
        ENTRY;
@@ -189,12 +214,13 @@ static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
                        RETURN(-EACCES);
                }
 
-               ucred->uc_fsuid = nodemap->nm_squash_uid;
-               ucred->uc_fsgid = nodemap->nm_squash_gid;
-               ucred->uc_cap = 0;
                ucred->uc_suppgids[0] = -1;
                ucred->uc_suppgids[1] = -1;
        }
+
+       if (nodemap && ucred->uc_o_gid == nodemap->nm_squash_gid)
+               is_nm_gid_squashed = true;
+
        nodemap_putref(nodemap);
 
        if (type == BODY_INIT) {
@@ -263,7 +289,8 @@ static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
        }
 
        if (perm & CFS_SETGRP_PERM) {
-               if (pud->pud_ngroups) {
+               /* only set groups if GID is not squashed */
+               if (pud->pud_ngroups && !is_nm_gid_squashed) {
                        /* setgroups for local client */
                        ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
                        if (!ucred->uc_ginfo) {
@@ -276,6 +303,8 @@ static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
                                                pud->pud_groups);
                        lustre_groups_sort(ucred->uc_ginfo);
                } else {
+                       ucred->uc_suppgids[0] = -1;
+                       ucred->uc_suppgids[1] = -1;
                        ucred->uc_ginfo = NULL;
                }
        } else {
@@ -286,19 +315,22 @@ static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 
        ucred->uc_uid = pud->pud_uid;
        ucred->uc_gid = pud->pud_gid;
+
+       if (nodemap && ucred->uc_o_uid == nodemap->nm_squash_uid) {
+               ucred->uc_cap = 0;
+       } else {
+               ucred->uc_cap = pud->pud_cap;
+       }
        ucred->uc_fsuid = pud->pud_fsuid;
        ucred->uc_fsgid = pud->pud_fsgid;
 
        /* process root_squash here. */
        mdt_root_squash(info, peernid);
 
-       /* remove fs privilege for non-root user. */
-       if (ucred->uc_fsuid && drop_fs_cap)
-               ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
-       else
-               ucred->uc_cap = pud->pud_cap;
        ucred->uc_valid = UCRED_NEW;
        ucred_set_jobid(info, ucred);
+       ucred_set_nid(info, ucred);
+       ucred_set_audit_enabled(info, ucred);
 
        EXIT;
 
@@ -428,8 +460,7 @@ out:
 }
 
 static int old_init_ucred_common(struct mdt_thread_info *info,
-                                struct lu_nodemap *nodemap,
-                                bool drop_fs_cap)
+                                struct lu_nodemap *nodemap)
 {
        struct lu_ucred         *uc = mdt_ucred(info);
        struct mdt_device       *mdt = info->mti_mdt;
@@ -440,8 +471,6 @@ static int old_init_ucred_common(struct mdt_thread_info *info,
                if (nodemap->nmf_deny_unknown)
                        RETURN(-EACCES);
 
-               uc->uc_fsuid = nodemap->nm_squash_uid;
-               uc->uc_fsgid = nodemap->nm_squash_gid;
                uc->uc_cap = 0;
                uc->uc_suppgids[0] = -1;
                uc->uc_suppgids[1] = -1;
@@ -466,11 +495,10 @@ static int old_init_ucred_common(struct mdt_thread_info *info,
        /* process root_squash here. */
        mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 
-       /* remove fs privilege for non-root user. */
-       if (uc->uc_fsuid && drop_fs_cap)
-               uc->uc_cap &= ~CFS_CAP_FS_MASK;
        uc->uc_valid = UCRED_OLD;
        ucred_set_jobid(info, uc);
+       ucred_set_nid(info, uc);
+       ucred_set_audit_enabled(info, uc);
 
        EXIT;
 
@@ -478,7 +506,7 @@ static int old_init_ucred_common(struct mdt_thread_info *info,
 }
 
 static int old_init_ucred(struct mdt_thread_info *info,
-                         struct mdt_body *body, bool drop_fs_cap)
+                         struct mdt_body *body)
 {
        struct lu_ucred *uc = mdt_ucred(info);
        struct lu_nodemap *nodemap;
@@ -509,7 +537,7 @@ static int old_init_ucred(struct mdt_thread_info *info,
        uc->uc_ginfo = NULL;
        uc->uc_cap = body->mbo_capability;
 
-       rc = old_init_ucred_common(info, nodemap, drop_fs_cap);
+       rc = old_init_ucred_common(info, nodemap);
        nodemap_putref(nodemap);
 
        RETURN(rc);
@@ -538,15 +566,14 @@ static int old_init_ucred_reint(struct mdt_thread_info *info)
        uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
        uc->uc_ginfo = NULL;
 
-       rc = old_init_ucred_common(info, nodemap, true); /* drop_fs_cap=true */
+       rc = old_init_ucred_common(info, nodemap);
        nodemap_putref(nodemap);
 
        RETURN(rc);
 }
 
 static inline int __mdt_init_ucred(struct mdt_thread_info *info,
-                                  struct mdt_body *body,
-                                  bool drop_fs_cap)
+                                  struct mdt_body *body)
 {
        struct ptlrpc_request   *req = mdt_info_req(info);
        struct lu_ucred         *uc  = mdt_ucred(info);
@@ -558,25 +585,14 @@ static inline int __mdt_init_ucred(struct mdt_thread_info *info,
        mdt_exit_ucred(info);
 
        if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
-               return old_init_ucred(info, body, drop_fs_cap);
+               return old_init_ucred(info, body);
        else
-               return new_init_ucred(info, BODY_INIT, body, drop_fs_cap);
+               return new_init_ucred(info, BODY_INIT, body);
 }
 
 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 {
-       return __mdt_init_ucred(info, body, true);
-}
-
-/* LU-6528 when "no_subtree_check" is set for NFS export, nfsd_set_fh_dentry()
- * doesn't set correct fsuid explicitely, but raise capability to allow
- * exportfs_decode_fh() to reconnect disconnected dentry into dcache. So for
- * lookup (i.e. intent_getattr), we should keep FS capability, otherwise it
- * will fail permission check. */
-int mdt_init_ucred_intent_getattr(struct mdt_thread_info *info,
-                                 struct mdt_body *body)
-{
-       return __mdt_init_ucred(info, body, false);
+       return __mdt_init_ucred(info, body);
 }
 
 int mdt_init_ucred_reint(struct mdt_thread_info *info)
@@ -599,7 +615,7 @@ int mdt_init_ucred_reint(struct mdt_thread_info *info)
        if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
                return old_init_ucred_reint(info);
        else
-               return new_init_ucred(info, REC_INIT, NULL, true);
+               return new_init_ucred(info, REC_INIT, NULL);
 }
 
 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
@@ -644,11 +660,25 @@ void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 {
        const struct lmv_mds_md_v1 *lmm1;
+       const struct lmv_foreign_md *lfm;
        int                        i;
 
        if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
                return;
 
+       /* foreign LMV case */
+       lfm = &lmv->lmv_foreign_md;
+       if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN) {
+               CDEBUG_LIMIT(level,
+                            "foreign magic 0x%08X, length %u, type %u, flags %u, value '%.*s'\n",
+                            le32_to_cpu(lfm->lfm_magic),
+                            le32_to_cpu(lfm->lfm_length),
+                            le32_to_cpu(lfm->lfm_type),
+                            le32_to_cpu(lfm->lfm_flags),
+                            le32_to_cpu(lfm->lfm_length), lfm->lfm_value);
+               return;
+       }
+
        lmm1 = &lmv->lmv_md_v1;
        CDEBUG(level,
               "magic 0x%08X, master %#X stripe_count %#x hash_type %#x\n",
@@ -739,20 +769,24 @@ int mdt_fix_reply(struct mdt_thread_info *info)
                req_capsule_shrink(pill, &RMF_LOGCOOKIES, acl_size, RCL_SERVER);
        }
 
-       if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
-           !(body->mbo_valid & OBD_MD_FLMDSCAPA))
-               req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
+       /* Shrink optional SECCTX buffer if it is not used */
+       if (req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_SERVER) &&
+           req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_SERVER) != 0 &&
+           !(body->mbo_valid & OBD_MD_SECCTX))
+               req_capsule_shrink(pill, &RMF_FILE_SECCTX, 0, RCL_SERVER);
 
-       if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
-           !(body->mbo_valid & OBD_MD_FLOSSCAPA))
-               req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
+       /* Shrink optional ENCCTX buffer if it is not used */
+       if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_SERVER) &&
+           req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_SERVER) != 0 &&
+           !(body->mbo_valid & OBD_MD_ENCCTX))
+               req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0, RCL_SERVER);
 
-        /*
-         * Some more field should be shrinked if needed.
-         * This should be done by those who added fields to reply message.
-         */
+       /*
+        * Some more field should be shrinked if needed.
+        * This should be done by those who added fields to reply message.
+        */
 
-        /* Grow MD buffer if needed finally */
+       /* Grow MD buffer if needed finally */
        if (info->mti_big_lmm_used) {
                 void *lmm;
 
@@ -830,7 +864,6 @@ int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
                .hai_cookie = 0,
                .hai_gid = 0,
        };
-       __u64 compound_id;
        int archive_id;
 
         ENTRY;
@@ -900,7 +933,6 @@ int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
         * file is unlinked, file is archived, so create remove request
         * for copytool!
         * If CDT is not running, requests will be logged for later. */
-       compound_id = atomic_inc_return(&cdt->cdt_compound_id);
        if (ma->ma_hsm.mh_arch_id != 0)
                archive_id = ma->ma_hsm.mh_arch_id;
        else
@@ -908,8 +940,8 @@ int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 
        hai.hai_fid = *mdt_object_fid(mo);
 
-       rc = mdt_agent_record_add(info->mti_env, info->mti_mdt,
-                                 compound_id, archive_id, 0, &hai);
+       rc = mdt_agent_record_add(info->mti_env, info->mti_mdt, archive_id, 0,
+                                 &hai);
        if (rc)
                CERROR("%s: unable to add HSM remove request for "DFID
                       ": rc=%d\n", mdt_obd_name(info->mti_mdt),
@@ -948,10 +980,14 @@ static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
                out |= LA_KILL_SGID;
        if (in & MDS_ATTR_PROJID)
                out |= LA_PROJID;
+       if (in & MDS_ATTR_LSIZE)
+               out |= LA_LSIZE;
+       if (in & MDS_ATTR_LBLOCKS)
+               out |= LA_LBLOCKS;
 
        if (in & MDS_ATTR_FROM_OPEN)
                rr->rr_flags |= MRF_OPEN_TRUNC;
-       if (in & MDS_OPEN_OWNEROVERRIDE)
+       if (in & MDS_ATTR_OVERRIDE)
                ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
        if (in & MDS_ATTR_FORCE)
                ma->ma_attr_flags |= MDS_PERM_BYPASS;
@@ -961,9 +997,10 @@ static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
                MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
                MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
                MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
-               MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
+               MDS_ATTR_FROM_OPEN | MDS_ATTR_LSIZE | MDS_ATTR_LBLOCKS |
+               MDS_ATTR_OVERRIDE);
        if (in != 0)
-               CERROR("Unknown attr bits: %#llx\n", in);
+               CDEBUG(D_INFO, "Unknown attr bits: %#llx\n", in);
        return out;
 }
 
@@ -1016,6 +1053,9 @@ static int mdt_file_secctx_unpack(struct req_capsule *pill,
        if (name_size == 0)
                return 0;
 
+       if (name_size > XATTR_NAME_MAX + 1)
+               return -EPROTO;
+
        name = req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
        if (strnlen(name, name_size) != name_size - 1)
                return -EPROTO;
@@ -1031,6 +1071,28 @@ static int mdt_file_secctx_unpack(struct req_capsule *pill,
        return 0;
 }
 
+static int mdt_file_encctx_unpack(struct req_capsule *pill,
+                                 void **encctx, size_t *encctx_size)
+{
+       *encctx = NULL;
+       *encctx_size = 0;
+
+       if (!exp_connect_encrypt(pill->rc_req->rq_export))
+               return 0;
+
+       if (!req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_CLIENT) ||
+           !req_capsule_field_present(pill, &RMF_FILE_ENCCTX, RCL_CLIENT))
+               return -EPROTO;
+
+       *encctx_size = req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_CLIENT);
+       if (*encctx_size == 0)
+               return 0;
+
+       *encctx = req_capsule_client_get(pill, &RMF_FILE_ENCCTX);
+
+       return 0;
+}
+
 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 {
        struct lu_ucred *uc = mdt_ucred(info);
@@ -1043,7 +1105,7 @@ static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 
        ENTRY;
 
-       CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
+       BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
        rec = req_capsule_client_get(pill, &RMF_REC_REINT);
        if (rec == NULL)
                RETURN(-EFAULT);
@@ -1078,13 +1140,9 @@ static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
        la->la_mtime = rec->sa_mtime;
        ma->ma_valid = MA_INODE;
 
-       if (rec->sa_bias & MDS_DATA_MODIFIED)
-               ma->ma_attr_flags |= MDS_DATA_MODIFIED;
-       else
-               ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
-
-       ma->ma_attr_flags &= ~MDS_CLOSE_INTENT;
-       ma->ma_attr_flags |= rec->sa_bias & MDS_CLOSE_INTENT;
+       ma->ma_attr_flags |= rec->sa_bias & (MDS_CLOSE_INTENT |
+                               MDS_DATA_MODIFIED | MDS_TRUNC_KEEP_LEASE |
+                               MDS_PCC_ATTACH);
        RETURN(0);
 }
 
@@ -1102,7 +1160,7 @@ static int mdt_close_handle_unpack(struct mdt_thread_info *info)
        if (ioepoch == NULL)
                RETURN(-EPROTO);
 
-       info->mti_close_handle = ioepoch->mio_handle;
+       info->mti_open_handle = ioepoch->mio_open_handle;
 
        RETURN(0);
 }
@@ -1135,6 +1193,7 @@ static int mdt_setattr_unpack(struct mdt_thread_info *info)
                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
                                                         RCL_CLIENT);
+
                if (rr->rr_eadatalen > 0) {
                        const struct lmv_user_md        *lum;
 
@@ -1201,13 +1260,13 @@ static int mdt_create_unpack(struct mdt_thread_info *info)
        struct mdt_rec_create *rec;
        struct lu_attr *attr = &info->mti_attr.ma_attr;
        struct mdt_reint_record *rr = &info->mti_rr;
-       struct req_capsule *pill = info->mti_pill;
-       struct md_op_spec *sp = &info->mti_spec;
+       struct req_capsule      *pill = info->mti_pill;
+       struct md_op_spec       *sp = &info->mti_spec;
        int rc;
 
        ENTRY;
 
-       CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
+       BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
        rec = req_capsule_client_get(pill, &RMF_REC_REINT);
        if (rec == NULL)
                RETURN(-EFAULT);
@@ -1267,6 +1326,15 @@ static int mdt_create_unpack(struct mdt_thread_info *info)
        if (rc < 0)
                RETURN(rc);
 
+       rc = mdt_file_encctx_unpack(pill, &sp->sp_cr_file_encctx,
+                                   &sp->sp_cr_file_encctx_size);
+       if (rc < 0)
+               RETURN(rc);
+
+       rc = req_check_sepol(pill);
+       if (rc)
+               RETURN(rc);
+
        rc = mdt_dlmreq_unpack(info);
        RETURN(rc);
 }
@@ -1282,7 +1350,7 @@ static int mdt_link_unpack(struct mdt_thread_info *info)
 
        ENTRY;
 
-       CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
+       BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
        rec = req_capsule_client_get(pill, &RMF_REC_REINT);
        if (rec == NULL)
                RETURN(-EFAULT);
@@ -1306,6 +1374,10 @@ static int mdt_link_unpack(struct mdt_thread_info *info)
        if (rc < 0)
                RETURN(rc);
 
+       rc = req_check_sepol(pill);
+       if (rc)
+               RETURN(rc);
+
        rc = mdt_dlmreq_unpack(info);
 
        RETURN(rc);
@@ -1315,15 +1387,14 @@ static int mdt_unlink_unpack(struct mdt_thread_info *info)
 {
        struct lu_ucred *uc  = mdt_ucred(info);
        struct mdt_rec_unlink *rec;
-       struct md_attr *ma = &info->mti_attr;
        struct lu_attr *attr = &info->mti_attr.ma_attr;
        struct mdt_reint_record *rr = &info->mti_rr;
-       struct req_capsule *pill = info->mti_pill;
+       struct req_capsule      *pill = info->mti_pill;
        int rc;
 
        ENTRY;
 
-       CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
+       BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
        rec = req_capsule_client_get(pill, &RMF_REC_REINT);
        if (rec == NULL)
                RETURN(-EFAULT);
@@ -1335,28 +1406,27 @@ static int mdt_unlink_unpack(struct mdt_thread_info *info)
        uc->uc_suppgids[0] = rec->ul_suppgid1;
        uc->uc_suppgids[1] = -1;
 
-        attr->la_uid = rec->ul_fsuid;
-        attr->la_gid = rec->ul_fsgid;
-        rr->rr_fid1 = &rec->ul_fid1;
-        rr->rr_fid2 = &rec->ul_fid2;
-        attr->la_ctime = rec->ul_time;
-        attr->la_mtime = rec->ul_time;
-        attr->la_mode  = rec->ul_mode;
-        attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
+       attr->la_uid = rec->ul_fsuid;
+       attr->la_gid = rec->ul_fsgid;
+       rr->rr_fid1 = &rec->ul_fid1;
+       rr->rr_fid2 = &rec->ul_fid2;
+       attr->la_ctime = rec->ul_time;
+       attr->la_mtime = rec->ul_time;
+       attr->la_mode  = rec->ul_mode;
+       attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
 
        rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
        if (rc < 0)
                RETURN(rc);
 
-        if (rec->ul_bias & MDS_VTX_BYPASS)
-                ma->ma_attr_flags |= MDS_VTX_BYPASS;
-        else
-                ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
-
        info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
 
-        rc = mdt_dlmreq_unpack(info);
-        RETURN(rc);
+       rc = req_check_sepol(pill);
+       if (rc)
+               RETURN(rc);
+
+       rc = mdt_dlmreq_unpack(info);
+       RETURN(rc);
 }
 
 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
@@ -1369,15 +1439,15 @@ static int mdt_rename_unpack(struct mdt_thread_info *info)
 {
        struct lu_ucred *uc = mdt_ucred(info);
        struct mdt_rec_rename *rec;
-       struct md_attr *ma = &info->mti_attr;
        struct lu_attr *attr = &info->mti_attr.ma_attr;
        struct mdt_reint_record *rr = &info->mti_rr;
        struct req_capsule *pill = info->mti_pill;
+       struct md_op_spec *spec = &info->mti_spec;
        int rc;
 
        ENTRY;
 
-       CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
+       BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
        rec = req_capsule_client_get(pill, &RMF_REC_REINT);
        if (rec == NULL)
                RETURN(-EFAULT);
@@ -1389,15 +1459,15 @@ static int mdt_rename_unpack(struct mdt_thread_info *info)
        uc->uc_suppgids[0] = rec->rn_suppgid1;
        uc->uc_suppgids[1] = rec->rn_suppgid2;
 
-        attr->la_uid = rec->rn_fsuid;
-        attr->la_gid = rec->rn_fsgid;
-        rr->rr_fid1 = &rec->rn_fid1;
-        rr->rr_fid2 = &rec->rn_fid2;
-        attr->la_ctime = rec->rn_time;
-        attr->la_mtime = rec->rn_time;
-        /* rename_tgt contains the mode already */
-        attr->la_mode = rec->rn_mode;
-        attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
+       attr->la_uid = rec->rn_fsuid;
+       attr->la_gid = rec->rn_fsgid;
+       rr->rr_fid1 = &rec->rn_fid1;
+       rr->rr_fid2 = &rec->rn_fid2;
+       attr->la_ctime = rec->rn_time;
+       attr->la_mtime = rec->rn_time;
+       /* rename_tgt contains the mode already */
+       attr->la_mode = rec->rn_mode;
+       attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
 
        rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
        if (rc < 0)
@@ -1407,21 +1477,88 @@ static int mdt_rename_unpack(struct mdt_thread_info *info)
        if (rc < 0)
                RETURN(rc);
 
-        if (rec->rn_bias & MDS_VTX_BYPASS)
-                ma->ma_attr_flags |= MDS_VTX_BYPASS;
-        else
-                ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
+       spec->no_create = !!req_is_replay(mdt_info_req(info));
+
+       rc = req_check_sepol(pill);
+       if (rc)
+               RETURN(rc);
+
+       rc = mdt_dlmreq_unpack(info);
+
+       RETURN(rc);
+}
+
+static int mdt_migrate_unpack(struct mdt_thread_info *info)
+{
+       struct lu_ucred *uc = mdt_ucred(info);
+       struct mdt_rec_rename *rec;
+       struct lu_attr *attr = &info->mti_attr.ma_attr;
+       struct mdt_reint_record *rr = &info->mti_rr;
+       struct req_capsule *pill = info->mti_pill;
+       struct md_op_spec *spec = &info->mti_spec;
+       int rc;
+
+       ENTRY;
+
+       BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
+       rec = req_capsule_client_get(pill, &RMF_REC_REINT);
+       if (rec == NULL)
+               RETURN(-EFAULT);
+
+       /* This prior initialization is needed for old_init_ucred_reint() */
+       uc->uc_fsuid = rec->rn_fsuid;
+       uc->uc_fsgid = rec->rn_fsgid;
+       uc->uc_cap   = rec->rn_cap;
+       uc->uc_suppgids[0] = rec->rn_suppgid1;
+       uc->uc_suppgids[1] = rec->rn_suppgid2;
+
+       attr->la_uid = rec->rn_fsuid;
+       attr->la_gid = rec->rn_fsgid;
+       rr->rr_fid1 = &rec->rn_fid1;
+       rr->rr_fid2 = &rec->rn_fid2;
+       attr->la_ctime = rec->rn_time;
+       attr->la_mtime = rec->rn_time;
+       /* rename_tgt contains the mode already */
+       attr->la_mode = rec->rn_mode;
+       attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
+       spec->sp_cr_flags = 0;
 
-       if (rec->rn_bias & MDS_RENAME_MIGRATE) {
-               req_capsule_extend(info->mti_pill, &RQF_MDS_REINT_MIGRATE);
+       rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
+       if (rc < 0)
+               RETURN(rc);
+
+       if (rec->rn_bias & MDS_CLOSE_MIGRATE) {
                rc = mdt_close_handle_unpack(info);
-               if (rc < 0)
+               if (rc)
                        RETURN(rc);
-               info->mti_spec.sp_migrate_close = 1;
-       }
 
-        info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
+               spec->sp_migrate_close = 1;
+       } else {
+               spec->sp_migrate_close = 0;
+       }
+       spec->sp_migrate_nsonly = 0;
+
+       /* lustre version > 2.11 migration packs lum */
+       if (req_capsule_has_field(pill, &RMF_EADATA, RCL_CLIENT)) {
+               if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
+                       rr->rr_eadatalen = req_capsule_get_size(pill,
+                                                               &RMF_EADATA,
+                                                               RCL_CLIENT);
+
+                       if (rr->rr_eadatalen > 0) {
+                               rr->rr_eadata = req_capsule_client_get(pill,
+                                                               &RMF_EADATA);
+                               spec->u.sp_ea.eadatalen = rr->rr_eadatalen;
+                               spec->u.sp_ea.eadata = rr->rr_eadata;
+                               spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
+                       }
+               } else {
+                       /* old client doesn't provide lum. */
+                       RETURN(-EOPNOTSUPP);
+               }
+       }
 
+       spec->no_create = !!req_is_replay(mdt_info_req(info));
 
        rc = mdt_dlmreq_unpack(info);
 
@@ -1458,7 +1595,8 @@ static int mdt_open_unpack(struct mdt_thread_info *info)
        int rc;
        ENTRY;
 
-       CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
+       BUILD_BUG_ON(sizeof(struct mdt_rec_create) !=
+                    sizeof(struct mdt_rec_reint));
        rec = req_capsule_client_get(pill, &RMF_REC_REINT);
        if (rec == NULL)
                RETURN(-EFAULT);
@@ -1471,18 +1609,18 @@ static int mdt_open_unpack(struct mdt_thread_info *info)
        uc->uc_suppgids[1] = rec->cr_suppgid2;
        uc->uc_umask = rec->cr_umask;
 
-        rr->rr_fid1   = &rec->cr_fid1;
-        rr->rr_fid2   = &rec->cr_fid2;
-        rr->rr_handle = &rec->cr_old_handle;
-        attr->la_mode = rec->cr_mode;
-        attr->la_rdev  = rec->cr_rdev;
-        attr->la_uid   = rec->cr_fsuid;
-        attr->la_gid   = rec->cr_fsgid;
-        attr->la_ctime = rec->cr_time;
-        attr->la_mtime = rec->cr_time;
-        attr->la_atime = rec->cr_time;
-        attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
-                         LA_CTIME | LA_MTIME | LA_ATIME;
+       rr->rr_fid1   = &rec->cr_fid1;
+       rr->rr_fid2   = &rec->cr_fid2;
+       rr->rr_open_handle = &rec->cr_open_handle_old;
+       attr->la_mode = rec->cr_mode;
+       attr->la_rdev  = rec->cr_rdev;
+       attr->la_uid   = rec->cr_fsuid;
+       attr->la_gid   = rec->cr_fsgid;
+       attr->la_ctime = rec->cr_time;
+       attr->la_mtime = rec->cr_time;
+       attr->la_atime = rec->cr_time;
+       attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
+                        LA_CTIME | LA_MTIME | LA_ATIME;
         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
         /* Do not trigger ASSERTION if client miss to set such flags. */
@@ -1496,11 +1634,13 @@ static int mdt_open_unpack(struct mdt_thread_info *info)
         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
                                                         RCL_CLIENT);
+
                 if (rr->rr_eadatalen > 0) {
                         rr->rr_eadata = req_capsule_client_get(pill,
                                                                &RMF_EADATA);
                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
                         sp->u.sp_ea.eadata = rr->rr_eadata;
+                       sp->sp_archive_id = rec->cr_archive_id;
                         sp->no_create = !!req_is_replay(req);
                        mdt_fix_lov_magic(info, rr->rr_eadata);
                 }
@@ -1518,6 +1658,17 @@ static int mdt_open_unpack(struct mdt_thread_info *info)
        rc = mdt_file_secctx_unpack(pill, &sp->sp_cr_file_secctx_name,
                                    &sp->sp_cr_file_secctx,
                                    &sp->sp_cr_file_secctx_size);
+       if (rc < 0)
+               RETURN(rc);
+
+       rc = mdt_file_encctx_unpack(pill, &sp->sp_cr_file_encctx,
+                                   &sp->sp_cr_file_encctx_size);
+       if (rc < 0)
+               RETURN(rc);
+
+       rc = req_check_sepol(pill);
+       if (rc)
+               RETURN(rc);
 
        RETURN(rc);
 }
@@ -1533,9 +1684,8 @@ static int mdt_setxattr_unpack(struct mdt_thread_info *info)
        ENTRY;
 
 
-       CLASSERT(sizeof(struct mdt_rec_setxattr) ==
-                sizeof(struct mdt_rec_reint));
-
+       BUILD_BUG_ON(sizeof(struct mdt_rec_setxattr) !=
+                    sizeof(struct mdt_rec_reint));
        rec = req_capsule_client_get(pill, &RMF_REC_REINT);
        if (rec == NULL)
                RETURN(-EFAULT);
@@ -1561,6 +1711,10 @@ static int mdt_setxattr_unpack(struct mdt_thread_info *info)
         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
                                                         RCL_CLIENT);
+
+               if (rr->rr_eadatalen > info->mti_mdt->mdt_max_ea_size)
+                       RETURN(-E2BIG);
+
                 if (rr->rr_eadatalen > 0) {
                         rr->rr_eadata = req_capsule_client_get(pill,
                                                                &RMF_EADATA);
@@ -1574,6 +1728,10 @@ static int mdt_setxattr_unpack(struct mdt_thread_info *info)
                 RETURN(-EFAULT);
         }
 
+       rc = req_check_sepol(pill);
+       if (rc)
+               RETURN(rc);
+
        if (mdt_dlmreq_unpack(info) < 0)
                RETURN(-EPROTO);
 
@@ -1588,7 +1746,7 @@ static int mdt_resync_unpack(struct mdt_thread_info *info)
        struct mdt_rec_resync   *rec;
        ENTRY;
 
-       CLASSERT(sizeof(*rec) == sizeof(struct mdt_rec_reint));
+       BUILD_BUG_ON(sizeof(*rec) != sizeof(struct mdt_rec_reint));
        rec = req_capsule_client_get(pill, &RMF_REC_REINT);
        if (rec == NULL)
                RETURN(-EFAULT);
@@ -1599,13 +1757,14 @@ static int mdt_resync_unpack(struct mdt_thread_info *info)
        uc->uc_cap   = rec->rs_cap;
 
        rr->rr_fid1   = &rec->rs_fid;
+       rr->rr_mirror_id = rec->rs_mirror_id;
 
        /* cookie doesn't need to be swapped but it has been swapped
         * in lustre_swab_mdt_rec_reint() as rr_mtime, so here it needs
         * restoring. */
        if (ptlrpc_req_need_swab(mdt_info_req(info)))
-               __swab64s(&rec->rs_handle.cookie);
-       rr->rr_handle = &rec->rs_handle;
+               __swab64s(&rec->rs_lease_handle.cookie);
+       rr->rr_lease_handle = &rec->rs_lease_handle;
 
        RETURN(mdt_dlmreq_unpack(info));
 }
@@ -1621,7 +1780,7 @@ static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
        [REINT_OPEN]     = mdt_open_unpack,
        [REINT_SETXATTR] = mdt_setxattr_unpack,
        [REINT_RMENTRY]  = mdt_rmentry_unpack,
-       [REINT_MIGRATE]  = mdt_rename_unpack,
+       [REINT_MIGRATE]  = mdt_migrate_unpack,
        [REINT_RESYNC]   = mdt_resync_unpack,
 };
 
@@ -1640,3 +1799,225 @@ int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
         }
         RETURN(rc);
 }
+
+int mdt_pack_secctx_in_reply(struct mdt_thread_info *info,
+                            struct mdt_object *child)
+{
+       char *secctx_name;
+       struct lu_buf *buffer;
+       struct mdt_body *repbody;
+       struct req_capsule *pill = info->mti_pill;
+       int rc = 0;
+
+       if (req_capsule_has_field(pill, &RMF_FILE_SECCTX, RCL_SERVER) &&
+           req_capsule_get_size(pill, &RMF_FILE_SECCTX, RCL_SERVER) != 0) {
+               secctx_name =
+                       req_capsule_client_get(pill, &RMF_FILE_SECCTX_NAME);
+               buffer = &info->mti_buf;
+
+               /* fill reply buffer with security context now */
+               buffer->lb_len = req_capsule_get_size(pill, &RMF_FILE_SECCTX,
+                                                     RCL_SERVER);
+               buffer->lb_buf = req_capsule_server_get(info->mti_pill,
+                                                       &RMF_FILE_SECCTX);
+               rc = mo_xattr_get(info->mti_env, mdt_object_child(child),
+                                 buffer, secctx_name);
+               if (rc >= 0) {
+                       CDEBUG(D_SEC,
+                              "found security context of size %d for "DFID"\n",
+                              rc, PFID(mdt_object_fid(child)));
+
+                       repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
+                       repbody->mbo_valid |= OBD_MD_SECCTX;
+                       if (rc < buffer->lb_len)
+                               req_capsule_shrink(pill, &RMF_FILE_SECCTX, rc,
+                                                  RCL_SERVER);
+                       rc = 0;
+               } else {
+                       CDEBUG(D_SEC,
+                            "security context not found for "DFID": rc = %d\n",
+                            PFID(mdt_object_fid(child)), rc);
+                       req_capsule_shrink(pill, &RMF_FILE_SECCTX, 0,
+                                          RCL_SERVER);
+                       /* handling -ENOENT is important because it may change
+                        * object state in DNE env dropping LOHA_EXISTS flag,
+                        * it is important to return that to the caller.
+                        * Check LU-13115 for details.
+                        */
+                       if (rc != -ENOENT)
+                               rc = 0;
+               }
+       }
+       return rc;
+}
+
+/* check whether two FIDs belong to different MDT. */
+static int mdt_fids_different_target(struct mdt_thread_info *info,
+                                    const struct lu_fid *fid1,
+                                    const struct lu_fid *fid2)
+{
+       const struct lu_env *env = info->mti_env;
+       struct mdt_device *mdt = info->mti_mdt;
+       struct lu_seq_range *range = &info->mti_range;
+       struct seq_server_site *ss;
+       __u32 index1, index2;
+       int rc;
+
+       if (fid_seq(fid1) == fid_seq(fid2))
+               return 0;
+
+       ss = mdt->mdt_lu_dev.ld_site->ld_seq_site;
+
+       range->lsr_flags = LU_SEQ_RANGE_MDT;
+       rc = fld_server_lookup(env, ss->ss_server_fld, fid1->f_seq, range);
+       if (rc)
+               return rc;
+
+       index1 = range->lsr_index;
+
+       rc = fld_server_lookup(env, ss->ss_server_fld, fid2->f_seq, range);
+       if (rc)
+               return rc;
+
+       index2 = range->lsr_index;
+
+       return index1 != index2;
+}
+
+/**
+ * Check whether \a child is remote object on \a parent.
+ *
+ * \param[in]  info    thread environment
+ * \param[in]  parent  parent object, it's the same as child object in
+ *                     getattr_by_fid
+ * \param[in]  child   child object
+ *
+ * \retval 1   is remote object.
+ * \retval 0   isn't remote object.
+ * \retval < 1  error code
+ */
+int mdt_is_remote_object(struct mdt_thread_info *info,
+                        struct mdt_object *parent,
+                        struct mdt_object *child)
+{
+       struct lu_buf *buf = &info->mti_big_buf;
+       struct linkea_data ldata = { NULL };
+       struct link_ea_header *leh;
+       struct link_ea_entry *lee;
+       struct lu_name name;
+       struct lu_fid pfid;
+       int reclen;
+       int i;
+       int rc;
+
+       ENTRY;
+
+       if (fid_is_root(mdt_object_fid(child)))
+               RETURN(0);
+
+       if (likely(parent != child)) {
+               if (mdt_object_remote(parent) ^ mdt_object_remote(child))
+                       RETURN(1);
+
+               if (!mdt_object_remote(parent) && !mdt_object_remote(child))
+                       RETURN(0);
+
+               rc = mdt_fids_different_target(info, mdt_object_fid(parent),
+                                              mdt_object_fid(child));
+               RETURN(rc);
+       }
+
+       /* client < 2.13.52 getattr_by_fid parent and child are the same */
+       buf = lu_buf_check_and_alloc(buf, PATH_MAX);
+       if (!buf->lb_buf)
+               RETURN(-ENOMEM);
+
+       ldata.ld_buf = buf;
+       rc = mdt_links_read(info, child, &ldata);
+       /* can't read linkea, just assume it's remote object */
+       if (rc == -ENOENT || rc == -ENODATA)
+               RETURN(1);
+       if (rc)
+               RETURN(rc);
+
+       leh = buf->lb_buf;
+       lee = (struct link_ea_entry *)(leh + 1);
+       for (i = 0; i < leh->leh_reccount; i++) {
+               linkea_entry_unpack(lee, &reclen, &name, &pfid);
+               lee = (struct link_ea_entry *) ((char *)lee + reclen);
+               if (mdt_fids_different_target(info, &pfid,
+                                             mdt_object_fid(child)))
+                       RETURN(1);
+       }
+
+       RETURN(0);
+}
+
+int mdt_pack_encctx_in_reply(struct mdt_thread_info *info,
+                            struct mdt_object *child)
+{
+       struct lu_buf *buffer;
+       struct mdt_body *repbody;
+       struct req_capsule *pill = info->mti_pill;
+       struct obd_export *exp = mdt_info_req(info)->rq_export;
+       int rc = 0;
+
+       if (!exp_connect_encrypt(exp))
+               return rc;
+
+       if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX, RCL_SERVER) &&
+           req_capsule_get_size(pill, &RMF_FILE_ENCCTX, RCL_SERVER) != 0) {
+               struct lu_attr la = { 0 };
+               struct dt_object *dt = mdt_obj2dt(child);
+
+               if (dt && dt->do_ops && dt->do_ops->do_attr_get)
+                       dt_attr_get(info->mti_env, mdt_obj2dt(child), &la);
+
+               if (la.la_valid & LA_FLAGS && la.la_flags & LUSTRE_ENCRYPT_FL) {
+                       buffer = &info->mti_buf;
+
+                       /* fill reply buffer with encryption context now */
+                       buffer->lb_len =
+                               req_capsule_get_size(pill, &RMF_FILE_ENCCTX,
+                                                    RCL_SERVER);
+                       buffer->lb_buf =
+                               req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
+                       rc = mo_xattr_get(info->mti_env,
+                                         mdt_object_child(child),
+                                         buffer,
+                                         LL_XATTR_NAME_ENCRYPTION_CONTEXT);
+                       if (rc >= 0) {
+                               CDEBUG(D_SEC,
+                                      "found encryption ctx of size %d for "DFID"\n",
+                                      rc, PFID(mdt_object_fid(child)));
+
+                               repbody = req_capsule_server_get(pill,
+                                                                &RMF_MDT_BODY);
+                               repbody->mbo_valid |= OBD_MD_ENCCTX;
+                               if (rc < buffer->lb_len)
+                                       req_capsule_shrink(pill,
+                                                          &RMF_FILE_ENCCTX, rc,
+                                                          RCL_SERVER);
+                               rc = 0;
+                       } else {
+                               CDEBUG(D_SEC,
+                                      "encryption ctx not found for "DFID": rc = %d\n",
+                                      PFID(mdt_object_fid(child)), rc);
+                               req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0,
+                                                  RCL_SERVER);
+                               /* handling -ENOENT is important because it may
+                                * change object state in DNE env dropping
+                                * LOHA_EXISTS flag, it is important to return
+                                * that to the caller.
+                                * Check LU-13115 for details.
+                                */
+                               if (rc != -ENOENT)
+                                       rc = 0;
+                       }
+               } else {
+                       req_capsule_shrink(pill, &RMF_FILE_ENCCTX, 0,
+                                          RCL_SERVER);
+               }
+       }
+       return rc;
+}