Whamcloud - gitweb
- b=15936. Unified last_rcvd format for MDS and OSS
[fs/lustre-release.git] / lustre / mdt / mdt_open.c
index 0507d97..0e4dfb6 100644 (file)
@@ -49,8 +49,8 @@ struct mdt_file_data *mdt_mfd_new(void)
 
         OBD_ALLOC_PTR(mfd);
         if (mfd != NULL) {
-                INIT_LIST_HEAD(&mfd->mfd_handle.h_link);
-                INIT_LIST_HEAD(&mfd->mfd_list);
+                CFS_INIT_LIST_HEAD(&mfd->mfd_handle.h_link);
+                CFS_INIT_LIST_HEAD(&mfd->mfd_list);
                 class_handle_hash(&mfd->mfd_handle, mdt_mfd_get);
         }
         RETURN(mfd);
@@ -98,8 +98,7 @@ static int mdt_create_data(struct mdt_thread_info *info,
         int rc;
         ENTRY;
 
-        if ((spec->sp_cr_flags & MDS_OPEN_DELAY_CREATE) ||
-            !(spec->sp_cr_flags & FMODE_WRITE))
+        if (!md_should_create(spec->sp_cr_flags))
                 RETURN(0);
 
         ma->ma_need = MA_INODE | MA_LOV;
@@ -126,7 +125,7 @@ void mdt_sizeonmds_enable(struct mdt_thread_info *info,
 {
        spin_lock(&info->mti_mdt->mdt_ioepoch_lock);
        if (info->mti_epoch->ioepoch == mo->mot_ioepoch) {
-                LASSERT(!mdt_epoch_opened(mo)); 
+                LASSERT(!mdt_epoch_opened(mo));
                 mo->mot_ioepoch = 0;
                 mo->mot_flags = 0;
        }
@@ -142,7 +141,7 @@ int mdt_epoch_open(struct mdt_thread_info *info, struct mdt_object *o)
         int rc = 0;
         ENTRY;
 
-        if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) || 
+        if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
             !S_ISREG(lu_object_attr(&o->mot_obj.mo_lu)))
                 RETURN(0);
 
@@ -378,7 +377,7 @@ static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p,
         int                      rc = 0, isdir, isreg;
         ENTRY;
 
-        repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
+        repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 
         isreg = S_ISREG(la->la_mode);
         isdir = S_ISDIR(la->la_mode);
@@ -388,7 +387,7 @@ static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p,
                  * above attr get, these size might be zero, so reset it, to
                  * retrieve the MD after create obj.
                  */
-                ma->ma_lmm_size = req_capsule_get_size(&info->mti_pill,
+                ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
                                                        &RMF_MDT_MD,
                                                        RCL_SERVER);
                 /* in replay case, p == NULL */
@@ -452,7 +451,7 @@ static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p,
                         /* Check wheather old cookie already exist in
                          * the list, becasue when do recovery, client
                          * might be disconnected from server, and
-                         * restart replay, so there maybe some orphan 
+                         * restart replay, so there maybe some orphan
                          * mfd here, we should remove them */
                         LASSERT(info->mti_rr.rr_handle != NULL);
                         old_mfd = mdt_handle2mfd(info, info->mti_rr.rr_handle);
@@ -501,7 +500,7 @@ static int mdt_finish_open(struct mdt_thread_info *info,
 
         LASSERT(ma->ma_valid & MA_INODE);
 
-        repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
+        repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 
         isreg = S_ISREG(la->la_mode);
         isdir = S_ISDIR(la->la_mode);
@@ -509,7 +508,7 @@ static int mdt_finish_open(struct mdt_thread_info *info,
         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 
         if (med->med_rmtclient) {
-                void *buf = req_capsule_server_get(&info->mti_pill, &RMF_ACL);
+                void *buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 
                 rc = mdt_pack_remote_perm(info, o, buf);
                 if (rc) {
@@ -526,8 +525,8 @@ static int mdt_finish_open(struct mdt_thread_info *info,
                 struct md_object *next = mdt_object_child(o);
                 struct lu_buf *buf = &info->mti_buf;
 
-                buf->lb_buf = req_capsule_server_get(&info->mti_pill, &RMF_ACL);
-                buf->lb_len = req_capsule_get_size(&info->mti_pill, &RMF_ACL,
+                buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
+                buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
                                                    RCL_SERVER);
                 if (buf->lb_len > 0) {
                         rc = mo_xattr_get(env, next, buf,
@@ -554,7 +553,7 @@ static int mdt_finish_open(struct mdt_thread_info *info,
         if (mdt->mdt_opts.mo_mds_capa) {
                 struct lustre_capa *capa;
 
-                capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA1);
+                capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
                 LASSERT(capa);
                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
                 capa->lc_uid = 0;
@@ -567,7 +566,7 @@ static int mdt_finish_open(struct mdt_thread_info *info,
             S_ISREG(lu_object_attr(&o->mot_obj.mo_lu))) {
                 struct lustre_capa *capa;
 
-                capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA2);
+                capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
                 LASSERT(capa);
                 capa->lc_opc = CAPA_OPC_OSS_DEFAULT | capa_open_opc(flags);
                 capa->lc_uid = 0;
@@ -605,8 +604,8 @@ static int mdt_finish_open(struct mdt_thread_info *info,
         } else if (flags & MDS_OPEN_DIRECTORY)
                 RETURN(-ENOTDIR);
 
-        if (MDT_FAIL_CHECK(OBD_FAIL_MDS_OPEN_CREATE)) {
-                obd_fail_loc = OBD_FAIL_LDLM_REPLY | OBD_FAIL_ONCE;
+        if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_OPEN_CREATE,
+                                 OBD_FAIL_LDLM_REPLY | OBD_FAIL_ONCE)) {
                 RETURN(-EAGAIN);
         }
 
@@ -641,18 +640,18 @@ static int mdt_finish_open(struct mdt_thread_info *info,
         RETURN(rc);
 }
 
-extern void mdt_req_from_mcd(struct ptlrpc_request *req,
-                             struct mdt_client_data *mcd);
+extern void mdt_req_from_lcd(struct ptlrpc_request *req,
+                             struct lsd_client_data *lcd);
 
 void mdt_reconstruct_open(struct mdt_thread_info *info,
                           struct mdt_lock_handle *lhc)
 {
         const struct lu_env *env = info->mti_env;
         struct mdt_device       *mdt  = info->mti_mdt;
-        struct req_capsule      *pill = &info->mti_pill;
+        struct req_capsule      *pill = info->mti_pill;
         struct ptlrpc_request   *req  = mdt_info_req(info);
         struct mdt_export_data  *med  = &req->rq_export->exp_mdt_data;
-        struct mdt_client_data  *mcd  = med->med_mcd;
+        struct lsd_client_data  *lcd  = med->med_lcd;
         struct md_attr          *ma   = &info->mti_attr;
         struct mdt_reint_record *rr   = &info->mti_rr;
         __u32                   flags = info->mti_spec.sp_cr_flags;
@@ -673,30 +672,47 @@ void mdt_reconstruct_open(struct mdt_thread_info *info,
         ma->ma_need = MA_INODE | MA_LOV;
         ma->ma_valid = 0;
 
-        mdt_req_from_mcd(req, med->med_mcd);
-        mdt_set_disposition(info, ldlm_rep, mcd->mcd_last_data);
+        mdt_req_from_lcd(req, med->med_lcd);
+        mdt_set_disposition(info, ldlm_rep, lcd->lcd_last_data);
 
         CERROR("This is reconstruct open: disp="LPX64", result=%d\n",
                 ldlm_rep->lock_policy_res1, req->rq_status);
 
         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE) &&
-            req->rq_status != 0) {
+            req->rq_status != 0)
                 /* We did not create successfully, return error to client. */
-                mdt_shrink_reply(info);
                 GOTO(out, rc = req->rq_status);
-        }
 
         if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE)) {
+                struct obd_export *exp = req->rq_export;
                 /*
                  * We failed after creation, but we do not know in which step
                  * we failed. So try to check the child object.
                  */
                 parent = mdt_object_find(env, mdt, rr->rr_fid1);
-                LASSERT(!IS_ERR(parent));
-
+                if (IS_ERR(parent)) {
+                        rc = PTR_ERR(parent);
+                        LCONSOLE_WARN("Parent "DFID" lookup error %d."
+                                      " Evicting client %s with export %s.\n",
+                                      PFID(mdt_object_fid(parent)), rc,
+                                      obd_uuid2str(&exp->exp_client_uuid),
+                                      obd_export_nid2str(exp));
+                        mdt_export_evict(exp);
+                        EXIT;
+                        return;
+                }
                 child = mdt_object_find(env, mdt, rr->rr_fid2);
-                LASSERT(!IS_ERR(child));
-
+                if (IS_ERR(child)) {
+                        rc = PTR_ERR(parent);
+                        LCONSOLE_WARN("Child "DFID" lookup error %d."
+                                      " Evicting client %s with export %s.\n",
+                                      PFID(mdt_object_fid(child)), rc,
+                                      obd_uuid2str(&exp->exp_client_uuid),
+                                      obd_export_nid2str(exp));
+                        mdt_export_evict(exp);
+                        EXIT;
+                        return;
+                }
                 rc = mdt_object_exists(child);
                 if (rc > 0) {
                         struct md_object *next;
@@ -720,7 +736,6 @@ void mdt_reconstruct_open(struct mdt_thread_info *info,
                 }
                 mdt_object_put(env, parent);
                 mdt_object_put(env, child);
-                mdt_shrink_reply(info);
                 GOTO(out, rc);
         } else {
 regular_open:
@@ -746,7 +761,6 @@ static int mdt_open_by_fid(struct mdt_thread_info* info,
         int                      rc;
         ENTRY;
 
-        LASSERT(info->mti_spec.u.sp_ea.no_lov_create);
         o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
         if (IS_ERR(o))
                 RETURN(rc = PTR_ERR(o));
@@ -759,13 +773,13 @@ static int mdt_open_by_fid(struct mdt_thread_info* info,
 
                 rc = mo_attr_get(env, mdt_object_child(o), ma);
                 if (rc == 0)
-                        rc = mdt_mfd_open(info, NULL, o, flags, 0);
+                        rc = mdt_finish_open(info, NULL, o, flags, 0, rep);
         } else if (rc == 0) {
                 rc = -ENOENT;
         } else  {
                 /* the child object was created on remote server */
                 struct mdt_body *repbody;
-                repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
+                repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
                 repbody->fid1 = *rr->rr_fid2;
                 repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
                 rc = 0;
@@ -847,19 +861,19 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
         int                      created = 0;
         ENTRY;
 
-        OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PAUSE_OPEN | OBD_FAIL_ONCE,
-                         (obd_timeout + 1) / 4);
+        OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_MDS_PAUSE_OPEN, OBD_FAIL_ONCE,
+                               (obd_timeout + 1) / 4);
 
-        repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
+        repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 
-        ma->ma_lmm = req_capsule_server_get(&info->mti_pill, &RMF_MDT_MD);
-        ma->ma_lmm_size = req_capsule_get_size(&info->mti_pill, &RMF_MDT_MD,
+        ma->ma_lmm = req_capsule_server_get(info->mti_pill, &RMF_MDT_MD);
+        ma->ma_lmm_size = req_capsule_get_size(info->mti_pill, &RMF_MDT_MD,
                                                RCL_SERVER);
         ma->ma_need = MA_INODE | MA_LOV;
         ma->ma_valid = 0;
 
-        LASSERT(info->mti_pill.rc_fmt == &RQF_LDLM_INTENT_OPEN);
-        ldlm_rep = req_capsule_server_get(&info->mti_pill, &RMF_DLM_REP);
+        LASSERT(info->mti_pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
+        ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
 
         /* TODO: JOIN file */
         if (create_flags & MDS_OPEN_JOIN_FILE) {
@@ -873,13 +887,17 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
                PFID(rr->rr_fid2), create_flags,
                ma->ma_attr.la_mode, lustre_msg_get_flags(req->rq_reqmsg));
 
-        if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
-                /* This is a replay request. */
+        if ((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) ||
+            (req->rq_export->exp_libclient && create_flags&MDS_OPEN_HAS_EA)) {
+                /* This is a replay request or from liblustre with ea. */
                 result = mdt_open_by_fid(info, ldlm_rep);
 
-                if (result != -ENOENT)
+                if (result != -ENOENT) {
+                        if (req->rq_export->exp_libclient &&
+                            create_flags&MDS_OPEN_HAS_EA)
+                                GOTO(out, result = 0);
                         GOTO(out, result);
-
+                }
                 /*
                  * We didn't find the correct object, so we need to re-create it
                  * via a regular replay.
@@ -892,7 +910,7 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
                        "regular open\n");
         }
 
-        if (MDT_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK))
+        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK))
                 GOTO(out, result = err_serious(-ENOMEM));
 
         mdt_set_disposition(info, ldlm_rep,
@@ -1057,7 +1075,6 @@ out_child:
 out_parent:
         mdt_object_unlock_put(info, parent, lh, result);
 out:
-        mdt_shrink_reply(info);
         if (result)
                 lustre_msg_set_transno(req->rq_repmsg, 0);
         return result;
@@ -1089,8 +1106,8 @@ int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd)
                 mdt_write_allow(info->mti_mdt, o);
         } else if (mode & FMODE_EPOCH) {
                 ret = mdt_epoch_close(info, o);
-        } 
-        
+        }
+
         /* Update atime on close only. */
         if ((mode & MDS_FMODE_EXEC || mode & FMODE_READ || mode & FMODE_WRITE)
             && (ma->ma_valid & MA_INODE) && (ma->ma_attr.la_valid & LA_ATIME)) {
@@ -1160,26 +1177,26 @@ int mdt_close(struct mdt_thread_info *info)
 
         LASSERT(info->mti_epoch);
 
-        req_capsule_set_size(&info->mti_pill, &RMF_MDT_MD, RCL_SERVER,
+        req_capsule_set_size(info->mti_pill, &RMF_MDT_MD, RCL_SERVER,
                              info->mti_mdt->mdt_max_mdsize);
-        req_capsule_set_size(&info->mti_pill, &RMF_LOGCOOKIES, RCL_SERVER,
+        req_capsule_set_size(info->mti_pill, &RMF_LOGCOOKIES, RCL_SERVER,
                              info->mti_mdt->mdt_max_cookiesize);
-        rc = req_capsule_pack(&info->mti_pill);
+        rc = req_capsule_server_pack(info->mti_pill);
         if (mdt_check_resent(info, mdt_reconstruct_generic, NULL))
                 RETURN(lustre_msg_get_status(req->rq_repmsg));
 
         /* Continue to close handle even if we can not pack reply */
         if (rc == 0) {
-                repbody = req_capsule_server_get(&info->mti_pill,
+                repbody = req_capsule_server_get(info->mti_pill,
                                                  &RMF_MDT_BODY);
-                ma->ma_lmm = req_capsule_server_get(&info->mti_pill,
+                ma->ma_lmm = req_capsule_server_get(info->mti_pill,
                                                     &RMF_MDT_MD);
-                ma->ma_lmm_size = req_capsule_get_size(&info->mti_pill,
+                ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
                                                        &RMF_MDT_MD,
                                                        RCL_SERVER);
-                ma->ma_cookie = req_capsule_server_get(&info->mti_pill,
+                ma->ma_cookie = req_capsule_server_get(info->mti_pill,
                                                        &RMF_LOGCOOKIES);
-                ma->ma_cookie_size = req_capsule_get_size(&info->mti_pill,
+                ma->ma_cookie_size = req_capsule_get_size(info->mti_pill,
                                                           &RMF_LOGCOOKIES,
                                                           RCL_SERVER);
                 ma->ma_need = MA_INODE | MA_LOV | MA_COOKIE;
@@ -1214,9 +1231,12 @@ int mdt_close(struct mdt_thread_info *info)
         if (repbody != NULL)
                 mdt_shrink_reply(info);
 
-        if (MDT_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK))
+        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK))
                 RETURN(err_serious(-ENOMEM));
 
+        if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_CLOSE_NET_REP,
+                                 OBD_FAIL_MDS_CLOSE_NET_REP))
+                info->mti_fail_id = OBD_FAIL_MDS_CLOSE_NET_REP;
         RETURN(rc ? rc : ret);
 }
 
@@ -1228,11 +1248,11 @@ int mdt_done_writing(struct mdt_thread_info *info)
         int rc;
         ENTRY;
 
-        rc = req_capsule_pack(&info->mti_pill);
+        rc = req_capsule_server_pack(info->mti_pill);
         if (rc)
                 RETURN(err_serious(rc));
 
-        repbody = req_capsule_server_get(&info->mti_pill,
+        repbody = req_capsule_server_get(info->mti_pill,
                                          &RMF_MDT_BODY);
         repbody->eadatasize = 0;
         repbody->aclsize = 0;
@@ -1254,8 +1274,8 @@ int mdt_done_writing(struct mdt_thread_info *info)
                        ": cookie = "LPX64"\n", PFID(info->mti_rr.rr_fid1),
                        info->mti_epoch->handle.cookie);
                 RETURN(-ESTALE);
-        } 
+        }
+
         LASSERT(mfd->mfd_mode == FMODE_EPOCH ||
                 mfd->mfd_mode == FMODE_EPOCHLCK);
         class_handle_unhash(&mfd->mfd_handle);