1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Lustre Light Super operations
6 * This code is issued under the GNU General Public License.
7 * See the file COPYING in this distribution
9 * Copryright (C) 1996 Peter J. Braam <braam@stelias.com>
10 * Copryright (C) 1999 Stelias Computing Inc. <braam@stelias.com>
11 * Copryright (C) 1999 Seagate Technology Inc.
12 * Copryright (C) 2001 Mountain View Data, Inc.
13 * Copryright (C) 2002 Cluster File Systems, Inc.
17 #include <linux/config.h>
18 #include <linux/module.h>
19 #include <linux/kmod.h>
21 #define DEBUG_SUBSYSTEM S_LLITE
23 #include <linux/lustre_lite.h>
24 #include <linux/lustre_ha.h>
26 int ll_reconnect(struct ptlrpc_connection *conn)
28 struct ptlrpc_request *request;
29 struct list_head *tmp;
32 /* XXX c_lock semantics! */
33 conn->c_level = LUSTRE_CONN_CON;
35 /* XXX this code MUST be shared with class_obd_connect! */
36 list_for_each(tmp, &conn->c_imports) {
37 struct obd_import *imp = list_entry(tmp, struct obd_import,
39 struct obd_device *obd = imp->imp_obd;
40 struct client_obd *cli = &obd->u.cli;
41 int rq_opc = (obd->obd_type->typ_ops->o_brw)
42 ? OST_CONNECT : MDS_CONNECT;
43 int size[] = { sizeof(cli->cl_target_uuid),
44 sizeof(obd->obd_uuid) };
45 char *tmp[] = {cli->cl_target_uuid, obd->obd_uuid };
46 struct lustre_handle old_hdl;
48 LASSERT(imp->imp_connection == conn);
49 request = ptlrpc_prep_req(imp, rq_opc, 2, size, tmp);
50 request->rq_level = LUSTRE_CONN_NEW;
51 request->rq_replen = lustre_msg_size(0, NULL);
52 /* XXX are (addr, cookie) right? */
53 request->rq_reqmsg->addr = imp->imp_handle.addr;
54 request->rq_reqmsg->cookie = imp->imp_handle.cookie;
55 rc = ptlrpc_queue_wait(request);
56 rc = ptlrpc_check_status(request, rc);
58 CERROR("cannot connect to %s@%s: rc = %d\n",
59 cli->cl_target_uuid, conn->c_remote_uuid, rc);
60 ptlrpc_free_req(request);
61 GOTO(out_disc, rc = -ENOTCONN);
64 old_hdl = imp->imp_handle;
65 imp->imp_handle.addr = request->rq_repmsg->addr;
66 imp->imp_handle.cookie = request->rq_repmsg->cookie;
67 CDEBUG(D_HA, "reconnected to %s@%s (%Lx/%Lx, was %Lx/%Lx)!\n",
68 cli->cl_target_uuid, conn->c_remote_uuid,
69 imp->imp_handle.addr, imp->imp_handle.cookie,
70 old_hdl.addr, old_hdl.cookie);
71 ptlrpc_free_req(request);
73 conn->c_level = LUSTRE_CONN_RECOVD;
79 static int ll_recover_upcall(struct ptlrpc_connection *conn)
86 conn->c_level = LUSTRE_CONN_RECOVD;
88 argv[0] = obd_recovery_upcall;
89 argv[1] = conn->c_remote_uuid;
93 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
96 rc = call_usermodehelper(argv[0], argv, envp);
99 * Tragically, this will never be run, because call_umh doesn't
100 * report errors like -ENOENT to its caller.
102 CERROR("Error invoking recovery upcall (%s): %d\n",
103 obd_recovery_upcall, rc);
104 CERROR("Check /proc/sys/lustre/recovery_upcall?\n");
106 CDEBUG(D_HA, "Invoked upcall %s for connection %s\n",
112 static int ll_recover_reconnect(struct ptlrpc_connection *conn)
115 struct list_head *tmp, *pos;
116 struct ptlrpc_request *req;
120 rc = ll_reconnect(conn);
124 /* 2. walk the request list */
125 spin_lock(&conn->c_lock);
127 CDEBUG(D_HA, "connection %p to %s has last_xid "LPD64"\n",
128 conn, conn->c_remote_uuid, conn->c_last_xid);
130 list_for_each_safe(tmp, pos, &conn->c_sending_head) {
131 req = list_entry(tmp, struct ptlrpc_request, rq_list);
133 /* replay what needs to be replayed */
134 if (req->rq_flags & PTL_RPC_FL_REPLAY) {
135 CDEBUG(D_HA, "FL_REPLAY: xid "LPD64" op %d @ %d\n",
136 req->rq_xid, req->rq_reqmsg->opc,
137 req->rq_import->imp_client->cli_request_portal);
138 rc = ptlrpc_replay_req(req);
140 #error We should not hold a spinlock over such a lengthy operation.
141 #error If necessary, drop spinlock, do operation, re-get spinlock, restart loop.
142 #error If we need to avoid re-processint items, then delete them from the list
143 #error as they are replayed and re-add at the tail of this list, so the next
144 #error item to process will always be at the head of the list.
147 CERROR("recovery replay error %d for req %Ld\n",
153 /* server has seen req, we have reply: skip */
154 if ((req->rq_flags & PTL_RPC_FL_REPLIED) &&
155 req->rq_xid <= conn->c_last_xid) {
156 CDEBUG(D_HA, "REPLIED SKIP: xid "LPD64" op %d @ %d\n",
157 req->rq_xid, req->rq_reqmsg->opc,
158 req->rq_import->imp_client->cli_request_portal);
162 /* server has lost req, we have reply: resend, ign reply */
163 if ((req->rq_flags & PTL_RPC_FL_REPLIED) &&
164 req->rq_xid > conn->c_last_xid) {
165 CDEBUG(D_HA, "REPLIED RESEND: xid "LPD64" op %d @ %d\n",
166 req->rq_xid, req->rq_reqmsg->opc,
167 req->rq_import->imp_client->cli_request_portal);
168 rc = ptlrpc_replay_req(req);
170 CERROR("request resend error %d for req %Ld\n",
176 /* server has seen req, we have lost reply: -ERESTARTSYS */
177 if ( !(req->rq_flags & PTL_RPC_FL_REPLIED) &&
178 req->rq_xid <= conn->c_last_xid) {
179 CDEBUG(D_HA, "RESTARTSYS: xid "LPD64" op %d @ %d\n",
180 req->rq_xid, req->rq_reqmsg->opc,
181 req->rq_import->imp_client->cli_request_portal);
182 ptlrpc_restart_req(req);
185 /* service has not seen req, no reply: resend */
186 if ( !(req->rq_flags & PTL_RPC_FL_REPLIED) &&
187 req->rq_xid > conn->c_last_xid) {
188 CDEBUG(D_HA, "RESEND: xid "LPD64" op %d @ %d\n",
189 req->rq_xid, req->rq_reqmsg->opc,
190 req->rq_import->imp_client->cli_request_portal);
191 ptlrpc_resend_req(req);
196 conn->c_level = LUSTRE_CONN_FULL;
197 recovd_conn_fixed(conn);
199 CERROR("recovery complete on conn %p(%s), waking delayed reqs\n",
200 conn, conn->c_remote_uuid);
201 /* Finally, continue what we delayed since recovery started */
202 list_for_each_safe(tmp, pos, &conn->c_delayed_head) {
203 req = list_entry(tmp, struct ptlrpc_request, rq_list);
204 ptlrpc_continue_req(req);
209 spin_unlock(&conn->c_lock);
213 static int ll_retry_recovery(struct ptlrpc_connection *conn)
215 CDEBUG(D_HA, "Recovery has failed on conn %p\n", conn);
217 /* XXX use a timer, sideshow bob */
218 recovd_conn_fail(conn);
219 /* XXX this is disabled until I fix it so that we don't just keep
220 * XXX retrying in the case of a missing upcall.
226 int ll_recover(struct recovd_data *rd, int phase)
228 struct ptlrpc_connection *conn = class_rd2conn(rd);
234 case PTLRPC_RECOVD_PHASE_PREPARE:
235 RETURN(ll_recover_upcall(conn));
236 case PTLRPC_RECOVD_PHASE_RECOVER:
237 RETURN(ll_recover_reconnect(conn));
238 case PTLRPC_RECOVD_PHASE_FAILURE:
239 RETURN(ll_retry_recovery(conn));