1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2001, 2002 Cluster File Systems, Inc.
5 * Author: Peter J. Braam <braam@clusterfs.com>
6 * Author: Phil Schwan <phil@clusterfs.com>
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 * Storage Target Handling functions
24 * Lustre Object Server Module (OST)
26 * This server is single threaded at present (but can easily be multi
27 * threaded). For testing and management it is treated as an
28 * obd_device, although it does not export a full OBD method table
29 * (the requests are coming in over the wire, so object target
30 * modules do not have a full method table.)
34 #define DEBUG_SUBSYSTEM S_OST
36 #include <linux/module.h>
37 #include <linux/obd_ost.h>
38 #include <linux/lustre_net.h>
40 static int ost_destroy(struct ost_obd *ost, struct ptlrpc_request *req)
43 struct ost_body *body;
44 int rc, size = sizeof(*body);
47 body = lustre_msg_buf(req->rq_reqmsg, 0);
48 conn.oc_id = body->connid;
49 conn.oc_dev = ost->ost_tgt;
51 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
55 req->rq_status = obd_destroy(&conn, &body->oa);
59 static int ost_getattr(struct ost_obd *ost, struct ptlrpc_request *req)
62 struct ost_body *body, *repbody;
63 int rc, size = sizeof(*body);
66 body = lustre_msg_buf(req->rq_reqmsg, 0);
67 conn.oc_id = body->connid;
68 conn.oc_dev = ost->ost_tgt;
70 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
74 repbody = lustre_msg_buf(req->rq_repmsg, 0);
75 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
76 req->rq_status = obd_getattr(&conn, &repbody->oa);
80 static int ost_open(struct ost_obd *ost, struct ptlrpc_request *req)
83 struct ost_body *body, *repbody;
84 int rc, size = sizeof(*body);
87 body = lustre_msg_buf(req->rq_reqmsg, 0);
88 conn.oc_id = body->connid;
89 conn.oc_dev = ost->ost_tgt;
91 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
95 repbody = lustre_msg_buf(req->rq_repmsg, 0);
96 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
97 req->rq_status = obd_open(&conn, &repbody->oa);
101 static int ost_close(struct ost_obd *ost, struct ptlrpc_request *req)
103 struct obd_conn conn;
104 struct ost_body *body, *repbody;
105 int rc, size = sizeof(*body);
108 body = lustre_msg_buf(req->rq_reqmsg, 0);
109 conn.oc_id = body->connid;
110 conn.oc_dev = ost->ost_tgt;
112 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
116 repbody = lustre_msg_buf(req->rq_repmsg, 0);
117 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
118 req->rq_status = obd_close(&conn, &repbody->oa);
122 static int ost_create(struct ost_obd *ost, struct ptlrpc_request *req)
124 struct obd_conn conn;
125 struct ost_body *body, *repbody;
126 int rc, size = sizeof(*body);
129 body = lustre_msg_buf(req->rq_reqmsg, 0);
130 conn.oc_id = body->connid;
131 conn.oc_dev = ost->ost_tgt;
133 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
137 repbody = lustre_msg_buf(req->rq_repmsg, 0);
138 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
139 req->rq_status = obd_create(&conn, &repbody->oa);
143 static int ost_punch(struct ost_obd *ost, struct ptlrpc_request *req)
145 struct obd_conn conn;
146 struct ost_body *body, *repbody;
147 int rc, size = sizeof(*body);
150 body = lustre_msg_buf(req->rq_reqmsg, 0);
151 conn.oc_id = body->connid;
152 conn.oc_dev = ost->ost_tgt;
154 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
158 repbody = lustre_msg_buf(req->rq_repmsg, 0);
159 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
160 req->rq_status = obd_punch(&conn, &repbody->oa,
161 repbody->oa.o_size, repbody->oa.o_blocks);
165 static int ost_setattr(struct ost_obd *ost, struct ptlrpc_request *req)
167 struct obd_conn conn;
168 struct ost_body *body, *repbody;
169 int rc, size = sizeof(*body);
172 body = lustre_msg_buf(req->rq_reqmsg, 0);
173 conn.oc_id = body->connid;
174 conn.oc_dev = ost->ost_tgt;
176 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
180 repbody = lustre_msg_buf(req->rq_repmsg, 0);
181 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
182 req->rq_status = obd_setattr(&conn, &repbody->oa);
186 static int ost_connect(struct ost_obd *ost, struct ptlrpc_request *req)
188 struct obd_conn conn;
189 struct ost_body *body;
190 int rc, size = sizeof(*body);
193 conn.oc_dev = ost->ost_tgt;
195 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
199 req->rq_status = obd_connect(&conn);
201 CDEBUG(D_IOCTL, "rep buffer %p, id %d\n", req->rq_repmsg, conn.oc_id);
202 body = lustre_msg_buf(req->rq_repmsg, 0);
203 body->connid = conn.oc_id;
207 static int ost_disconnect(struct ost_obd *ost, struct ptlrpc_request *req)
209 struct obd_conn conn;
210 struct ost_body *body;
211 int rc, size = sizeof(*body);
214 body = lustre_msg_buf(req->rq_reqmsg, 0);
215 conn.oc_id = body->connid;
216 conn.oc_dev = ost->ost_tgt;
218 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
222 CDEBUG(D_IOCTL, "Disconnecting %d\n", conn.oc_id);
223 req->rq_status = obd_disconnect(&conn);
227 static int ost_get_info(struct ost_obd *ost, struct ptlrpc_request *req)
229 struct obd_conn conn;
230 struct ost_body *body;
231 int rc, size[2] = {sizeof(*body)};
232 char *bufs[2] = {NULL, NULL}, *ptr;
235 body = lustre_msg_buf(req->rq_reqmsg, 0);
236 conn.oc_id = body->connid;
237 conn.oc_dev = ost->ost_tgt;
239 ptr = lustre_msg_buf(req->rq_reqmsg, 1);
243 req->rq_status = obd_get_info(&conn, req->rq_reqmsg->buflens[1], ptr,
244 &(size[1]), (void **)&(bufs[1]));
246 rc = lustre_pack_msg(2, size, bufs, &req->rq_replen, &req->rq_repmsg);
248 CERROR("cannot pack reply\n");
253 static int ost_brw_read(struct ost_obd *obddev, struct ptlrpc_request *req)
255 struct ptlrpc_bulk_desc *bulk = NULL;
256 struct obd_conn conn;
257 void *tmp1, *tmp2, *end2;
258 struct niobuf *nb, *dst, *res = NULL;
259 struct obd_ioobj *ioo;
260 struct ost_body *body;
261 int rc, cmd, i, j, objcount, niocount, size = sizeof(*body);
264 body = lustre_msg_buf(req->rq_reqmsg, 0);
265 tmp1 = lustre_msg_buf(req->rq_reqmsg, 1);
266 tmp2 = lustre_msg_buf(req->rq_reqmsg, 2);
267 end2 = (char *)tmp2 + req->rq_reqmsg->buflens[2];
268 objcount = req->rq_reqmsg->buflens[1] / sizeof(*ioo);
269 niocount = req->rq_reqmsg->buflens[2] / sizeof(*nb);
272 conn.oc_id = body->connid;
273 conn.oc_dev = req->rq_obd->u.ost.ost_tgt;
275 for (i = 0; i < objcount; i++) {
276 ost_unpack_ioo(&tmp1, &ioo);
277 if (tmp2 + ioo->ioo_bufcnt > end2) {
282 for (j = 0; j < ioo->ioo_bufcnt; j++) {
283 ost_unpack_niobuf(&tmp2, &nb);
287 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
290 OBD_ALLOC(res, sizeof(*res) * niocount);
294 /* The unpackers move tmp1 and tmp2, so reset them before using */
295 tmp1 = lustre_msg_buf(req->rq_reqmsg, 1);
296 tmp2 = lustre_msg_buf(req->rq_reqmsg, 2);
297 req->rq_status = obd_preprw(cmd, &conn, objcount,
298 tmp1, niocount, tmp2, res);
303 for (i = 0; i < niocount; i++) {
304 bulk = ptlrpc_prep_bulk(req->rq_connection);
306 CERROR("cannot alloc bulk desc\n");
307 GOTO(out_res, rc = -ENOMEM);
310 dst = &(((struct niobuf *)tmp2)[i]);
311 bulk->b_xid = dst->xid;
312 bulk->b_buf = (void *)(unsigned long)res[i].addr;
313 bulk->b_buflen = PAGE_SIZE;
314 rc = ptlrpc_send_bulk(bulk, OST_BULK_PORTAL);
317 wait_event_interruptible(bulk->b_waitq,
318 ptlrpc_check_bulk_sent(bulk));
320 if (bulk->b_flags & PTL_RPC_FL_INTR)
323 ptlrpc_free_bulk(bulk);
327 /* The unpackers move tmp1 and tmp2, so reset them before using */
328 tmp1 = lustre_msg_buf(req->rq_reqmsg, 1);
329 tmp2 = lustre_msg_buf(req->rq_reqmsg, 2);
330 req->rq_status = obd_commitrw(cmd, &conn, objcount,
331 tmp1, niocount, res);
336 ptlrpc_free_bulk(bulk);
339 OBD_FREE(res, sizeof(*res) * niocount);
344 static int ost_commit_page(struct obd_conn *conn, struct page *page)
346 struct obd_ioobj obj;
351 memset(&buf, 0, sizeof(buf));
352 memset(&obj, 0, sizeof(obj));
357 rc = obd_commitrw(OBD_BRW_WRITE, conn, 1, &obj, 1, &buf);
361 static int ost_brw_write_cb(struct ptlrpc_bulk_desc *bulk, void *data)
367 rc = ost_commit_page(&bulk->b_conn, bulk->b_page);
369 CERROR("ost_commit_page failed: %d\n", rc);
374 static int ost_brw_write(struct ost_obd *obddev, struct ptlrpc_request *req)
376 struct obd_conn conn;
377 struct niobuf *nb, *res;
378 struct obd_ioobj *ioo;
379 struct ost_body *body;
380 int cmd, rc, i, j, objcount, niocount, size[2] = {sizeof(*body)};
381 void *tmp1, *tmp2, *end2;
384 body = lustre_msg_buf(req->rq_reqmsg, 0);
385 tmp1 = lustre_msg_buf(req->rq_reqmsg, 1);
386 tmp2 = lustre_msg_buf(req->rq_reqmsg, 2);
387 end2 = (char *)tmp2 + req->rq_reqmsg->buflens[2];
388 objcount = req->rq_reqmsg->buflens[1] / sizeof(*ioo);
389 niocount = req->rq_reqmsg->buflens[2] / sizeof(*nb);
392 conn.oc_id = body->connid;
393 conn.oc_dev = req->rq_obd->u.ost.ost_tgt;
395 for (i = 0; i < objcount; i++) {
396 ost_unpack_ioo((void *)&tmp1, &ioo);
397 if (tmp2 + ioo->ioo_bufcnt > end2) {
401 for (j = 0; j < ioo->ioo_bufcnt; j++) {
402 ost_unpack_niobuf((void *)&tmp2, &nb);
406 size[1] = niocount * sizeof(*nb);
407 rc = lustre_pack_msg(2, size, NULL, &req->rq_replen, &req->rq_repmsg);
411 res = lustre_msg_buf(req->rq_repmsg, 1);
413 /* The unpackers move tmp1 and tmp2, so reset them before using */
414 tmp1 = lustre_msg_buf(req->rq_reqmsg, 1);
415 tmp2 = lustre_msg_buf(req->rq_reqmsg, 2);
416 req->rq_status = obd_preprw(cmd, &conn, objcount,
417 tmp1, niocount, tmp2, res);
422 for (i = 0; i < niocount; i++, res++) {
423 struct ptlrpc_bulk_desc *bulk;
424 struct ptlrpc_service *srv = req->rq_obd->u.ost.ost_service;
426 bulk = ptlrpc_prep_bulk(req->rq_connection);
428 GOTO(out, rc = -ENOMEM);
430 spin_lock(&srv->srv_lock);
431 bulk->b_xid = srv->srv_xid++;
432 spin_unlock(&srv->srv_lock);
434 res->xid = HTON__u32(bulk->b_xid);
436 bulk->b_buf = (void *)(unsigned long)res->addr;
437 bulk->b_cb = ost_brw_write_cb;
438 bulk->b_page = res->page;
439 memcpy(&(bulk->b_conn), &conn, sizeof(conn));
440 bulk->b_buflen = PAGE_SIZE;
441 bulk->b_portal = OSC_BULK_PORTAL;
442 rc = ptlrpc_register_bulk(bulk);
449 /* FIXME: should we return 'rc' here? */
453 static int ost_brw(struct ost_obd *obddev, struct ptlrpc_request *req)
455 struct ost_body *body = lustre_msg_buf(req->rq_reqmsg, 0);
457 if (body->data == OBD_BRW_READ)
458 return ost_brw_read(obddev, req);
460 return ost_brw_write(obddev, req);
463 static int ost_handle(struct obd_device *obddev, struct ptlrpc_service *svc,
464 struct ptlrpc_request *req)
467 struct ost_obd *ost = &obddev->u.ost;
470 rc = lustre_unpack_msg(req->rq_reqmsg, req->rq_reqlen);
471 if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_HANDLE_UNPACK)) {
472 CERROR("lustre_mds: Invalid request\n");
476 if (req->rq_reqmsg->type != PTL_RPC_MSG_REQUEST) {
477 CERROR("lustre_mds: wrong packet type sent %d\n",
478 req->rq_reqmsg->type);
479 GOTO(out, rc = -EINVAL);
482 switch (req->rq_reqmsg->opc) {
484 CDEBUG(D_INODE, "connect\n");
485 OBD_FAIL_RETURN(OBD_FAIL_OST_CONNECT_NET, 0);
486 rc = ost_connect(ost, req);
489 CDEBUG(D_INODE, "disconnect\n");
490 OBD_FAIL_RETURN(OBD_FAIL_OST_DISCONNECT_NET, 0);
491 rc = ost_disconnect(ost, req);
494 CDEBUG(D_INODE, "get_info\n");
495 OBD_FAIL_RETURN(OBD_FAIL_OST_GET_INFO_NET, 0);
496 rc = ost_get_info(ost, req);
499 CDEBUG(D_INODE, "create\n");
500 OBD_FAIL_RETURN(OBD_FAIL_OST_CREATE_NET, 0);
501 rc = ost_create(ost, req);
504 CDEBUG(D_INODE, "destroy\n");
505 OBD_FAIL_RETURN(OBD_FAIL_OST_DESTROY_NET, 0);
506 rc = ost_destroy(ost, req);
509 CDEBUG(D_INODE, "getattr\n");
510 OBD_FAIL_RETURN(OBD_FAIL_OST_GETATTR_NET, 0);
511 rc = ost_getattr(ost, req);
514 CDEBUG(D_INODE, "setattr\n");
515 OBD_FAIL_RETURN(OBD_FAIL_OST_SETATTR_NET, 0);
516 rc = ost_setattr(ost, req);
519 CDEBUG(D_INODE, "setattr\n");
520 OBD_FAIL_RETURN(OBD_FAIL_OST_OPEN_NET, 0);
521 rc = ost_open(ost, req);
524 CDEBUG(D_INODE, "setattr\n");
525 OBD_FAIL_RETURN(OBD_FAIL_OST_CLOSE_NET, 0);
526 rc = ost_close(ost, req);
529 CDEBUG(D_INODE, "brw\n");
530 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
531 rc = ost_brw(ost, req);
534 CDEBUG(D_INODE, "punch\n");
535 OBD_FAIL_RETURN(OBD_FAIL_OST_PUNCH_NET, 0);
536 rc = ost_punch(ost, req);
539 req->rq_status = -ENOTSUPP;
540 rc = ptlrpc_error(svc, req);
546 //req->rq_status = rc;
548 CERROR("ost: processing error %d\n", rc);
549 ptlrpc_error(svc, req);
551 CDEBUG(D_INODE, "sending reply\n");
552 ptlrpc_reply(svc, req);
558 /* mount the file system (secretly) */
559 static int ost_setup(struct obd_device *obddev, obd_count len, void *buf)
561 struct obd_ioctl_data* data = buf;
562 struct ost_obd *ost = &obddev->u.ost;
563 struct obd_device *tgt;
567 if (data->ioc_dev < 0 || data->ioc_dev > MAX_OBD_DEVICES)
571 tgt = &obd_dev[data->ioc_dev];
573 if (!(tgt->obd_flags & OBD_ATTACHED) ||
574 !(tgt->obd_flags & OBD_SET_UP)) {
575 CERROR("device not attached or not set up (%d)\n",
577 GOTO(error_dec, err = -EINVAL);
580 ost->ost_conn.oc_dev = tgt;
581 err = obd_connect(&ost->ost_conn);
583 CERROR("fail to connect to device %d\n", data->ioc_dev);
584 GOTO(error_dec, err = -EINVAL);
587 ost->ost_service = ptlrpc_init_svc(128 * 1024,
588 OST_REQUEST_PORTAL, OSC_REPLY_PORTAL,
590 if (!ost->ost_service) {
591 CERROR("failed to start service\n");
592 GOTO(error_disc, err = -EINVAL);
595 err = ptlrpc_start_thread(obddev, ost->ost_service, "lustre_ost");
597 GOTO(error_disc, err = -EINVAL);
599 err = ptlrpc_start_thread(obddev, ost->ost_service, "lustre_ost");
601 GOTO(error_disc, err = -EINVAL);
607 obd_disconnect(&ost->ost_conn);
613 static int ost_cleanup(struct obd_device * obddev)
615 struct ost_obd *ost = &obddev->u.ost;
620 if ( !list_empty(&obddev->obd_gen_clients) ) {
621 CERROR("still has clients!\n");
625 ptlrpc_stop_all_threads(ost->ost_service);
626 rpc_unregister_service(ost->ost_service);
628 if (!list_empty(&ost->ost_service->srv_reqs)) {
629 // XXX reply with errors and clean up
630 CERROR("Request list not empty!\n");
632 OBD_FREE(ost->ost_service, sizeof(*ost->ost_service));
634 err = obd_disconnect(&ost->ost_conn);
636 CERROR("lustre ost: fail to disconnect device\n");
644 /* use obd ops to offer management infrastructure */
645 static struct obd_ops ost_obd_ops = {
647 o_cleanup: ost_cleanup,
650 static int __init ost_init(void)
652 obd_register_type(&ost_obd_ops, LUSTRE_OST_NAME);
656 static void __exit ost_exit(void)
658 obd_unregister_type(LUSTRE_OST_NAME);
661 MODULE_AUTHOR("Peter J. Braam <braam@clusterfs.com>");
662 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
663 MODULE_LICENSE("GPL");
665 module_init(ost_init);
666 module_exit(ost_exit);