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.
6 * This file is part of Lustre, http://www.sf.net/projects/lustre/
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 #define DEBUG_SUBSYSTEM S_MDC
26 #include <linux/module.h>
27 #include <linux/miscdevice.h>
28 #include <linux/lustre_mds.h>
29 #include <linux/lustre_lite.h>
30 #include <linux/lustre_dlm.h>
32 #define REQUEST_MINOR 244
34 extern int mds_queue_req(struct ptlrpc_request *);
37 int mdc_getstatus(struct obd_conn *conn, struct ll_fid *rootfid,
38 __u64 *last_committed, __u64 *last_rcvd,
39 __u32 *last_xid, struct ptlrpc_request **request)
41 struct ptlrpc_request *req;
42 struct mds_body *body;
43 struct mdc_obd *mdc = mdc_conn2mdc(conn);
44 int rc, size = sizeof(*body);
47 req = ptlrpc_prep_req(mdc->mdc_client, mdc->mdc_conn,
48 MDS_GETSTATUS, 1, &size, NULL);
50 GOTO(out, rc = -ENOMEM);
52 body = lustre_msg_buf(req->rq_reqmsg, 0);
53 req->rq_level = LUSTRE_CONN_CON;
54 req->rq_replen = lustre_msg_size(1, &size);
56 mds_pack_req_body(req);
57 rc = ptlrpc_queue_wait(req);
58 rc = ptlrpc_check_status(req, rc);
61 body = lustre_msg_buf(req->rq_repmsg, 0);
62 mds_unpack_body(body);
63 memcpy(rootfid, &body->fid1, sizeof(*rootfid));
64 *last_committed = req->rq_repmsg->last_committed;
65 *last_rcvd = req->rq_repmsg->last_rcvd;
66 *last_xid = body->last_xid;
68 CDEBUG(D_NET, "root ino=%ld, last_committed=%Lu, last_rcvd=%Lu,"
70 (unsigned long)rootfid->id,
71 (unsigned long long)*last_committed,
72 (unsigned long long)*last_rcvd,
83 int mdc_getattr(struct obd_conn *conn,
84 ino_t ino, int type, unsigned long valid, size_t ea_size,
85 struct ptlrpc_request **request)
87 struct mdc_obd *mdc = mdc_conn2mdc(conn);
88 struct ptlrpc_request *req;
89 struct mds_body *body;
90 int rc, size[2] = {sizeof(*body), 0}, bufcount = 1;
93 req = ptlrpc_prep_req(mdc->mdc_client, mdc->mdc_conn,
94 MDS_GETATTR, 1, size, NULL);
96 GOTO(out, rc = -ENOMEM);
98 body = lustre_msg_buf(req->rq_reqmsg, 0);
99 ll_ino2fid(&body->fid1, ino, 0, type);
104 size[1] = sizeof(struct obdo);
105 } else if (valid & OBD_MD_LINKNAME) {
109 req->rq_replen = lustre_msg_size(bufcount, size);
110 req->rq_level = LUSTRE_CONN_FULL;
112 rc = ptlrpc_queue_wait(req);
113 rc = ptlrpc_check_status(req, rc);
116 body = lustre_msg_buf(req->rq_repmsg, 0);
117 mds_unpack_body(body);
118 CDEBUG(D_NET, "mode: %o\n", body->mode);
127 static int mdc_lock_callback(struct ldlm_lock *lock, struct ldlm_lock *new,
128 void *data, int data_len)
131 struct inode *inode = data;
135 /* Completion AST. Do nothing. */
139 if (data_len != sizeof(*inode)) {
140 CERROR("data_len should be %d, but is %d\n", sizeof(*inode),
145 /* FIXME: do something better than throwing away everything */
148 if (S_ISDIR(inode->i_mode)) {
149 CDEBUG(D_INODE, "invalidating inode %ld\n", inode->i_ino);
150 invalidate_inode_pages(inode);
153 rc = ldlm_cli_cancel(lock->l_client, lock);
155 CERROR("ldlm_cli_cancel: %d\n", rc);
161 int mdc_enqueue(struct obd_conn *conn, int lock_type, struct lookup_intent *it,
162 int lock_mode, struct inode *dir, struct dentry *de,
163 struct lustre_handle *lockh, __u64 id, char *tgt, int tgtlen,
164 void *data, int datalen)
166 struct ptlrpc_request *req;
167 struct obd_device *obddev = conn->oc_dev;
168 struct mdc_obd *mdc = mdc_conn2mdc(conn);
169 __u64 res_id[RES_NAME_SIZE] = {dir->i_ino};
170 int size[5] = {sizeof(struct ldlm_request), sizeof(struct ldlm_intent)};
172 struct ldlm_reply *dlm_rep;
173 struct ldlm_intent *lit;
178 it->it_mode = (it->it_mode | S_IFDIR) & ~current->fs->umask;
181 it->it_op = IT_GETATTR;
183 case (IT_CREAT|IT_OPEN):
185 it->it_mode = (it->it_mode | S_IFREG) & ~current->fs->umask;
188 it->it_mode = (it->it_mode | S_IFLNK) & ~current->fs->umask;
192 if (it->it_op & (IT_MKDIR | IT_CREAT | IT_SYMLINK | IT_MKNOD)) {
193 size[2] = sizeof(struct mds_rec_create);
194 size[3] = de->d_name.len + 1;
195 size[4] = tgtlen + 1;
196 req = ptlrpc_prep_req(mdc->mdc_ldlm_client, mdc->mdc_conn,
197 LDLM_ENQUEUE, 5, size, NULL);
201 /* pack the intent */
202 lit = lustre_msg_buf(req->rq_reqmsg, 1);
203 lit->opc = NTOH__u64((__u64)it->it_op);
205 /* pack the intended request */
206 mds_create_pack(req, 2, dir, it->it_mode, id, current->fsuid,
207 current->fsgid, CURRENT_TIME, de->d_name.name,
208 de->d_name.len, tgt, tgtlen);
210 size[0] = sizeof(struct ldlm_reply);
211 size[1] = sizeof(struct mds_body);
212 size[2] = sizeof(struct obdo);
213 req->rq_replen = lustre_msg_size(3, size);
214 } else if ( it->it_op == IT_RENAME2 ) {
215 struct dentry *old_de = it->it_data;
217 size[2] = sizeof(struct mds_rec_rename);
218 size[3] = old_de->d_name.len + 1;
219 size[4] = de->d_name.len + 1;
220 req = ptlrpc_prep_req(mdc->mdc_ldlm_client, mdc->mdc_conn,
221 LDLM_ENQUEUE, 5, size, NULL);
225 /* pack the intent */
226 lit = lustre_msg_buf(req->rq_reqmsg, 1);
227 lit->opc = NTOH__u64((__u64)it->it_op);
229 /* pack the intended request */
230 mds_rename_pack(req, 2, old_de->d_inode, dir,
231 old_de->d_parent->d_name.name,
232 old_de->d_parent->d_name.len,
233 de->d_name.name, de->d_name.len);
235 size[0] = sizeof(struct ldlm_reply);
236 size[1] = sizeof(struct mds_body);
237 req->rq_replen = lustre_msg_size(2, size);
238 } else if ( it->it_op == IT_GETATTR || it->it_op == IT_RENAME ||
239 it->it_op == IT_OPEN ) {
240 size[2] = sizeof(struct mds_body);
241 size[3] = de->d_name.len + 1;
243 req = ptlrpc_prep_req(mdc->mdc_ldlm_client, mdc->mdc_conn,
244 LDLM_ENQUEUE, 4, size, NULL);
248 /* pack the intent */
249 lit = lustre_msg_buf(req->rq_reqmsg, 1);
250 lit->opc = NTOH__u64((__u64)it->it_op);
252 /* pack the intended request */
253 mds_getattr_pack(req, 2, dir, de->d_name.name, de->d_name.len);
255 /* get ready for the reply */
256 size[0] = sizeof(struct ldlm_reply);
257 size[1] = sizeof(struct mds_body);
258 size[2] = sizeof(struct obdo);
259 req->rq_replen = lustre_msg_size(3, size);
260 } else if ( it->it_op == IT_SETATTR) {
261 size[2] = sizeof(struct mds_rec_setattr);
262 size[3] = de->d_name.len + 1;
263 req = ptlrpc_prep_req(mdc->mdc_ldlm_client, mdc->mdc_conn,
264 LDLM_ENQUEUE, 5, size, NULL);
268 lit = lustre_msg_buf(req->rq_reqmsg, 1);
269 lit->opc = NTOH__u64((__u64)it->it_op);
274 mds_setattr_pack(req, 2, dir, it->it_iattr,
275 de->d_name.name, de->d_name.len);
276 size[0] = sizeof(struct ldlm_reply);
277 size[1] = sizeof(struct mds_body);
278 req->rq_replen = lustre_msg_size(2, size);
279 } else if ( it->it_op == IT_READDIR ) {
280 req = ptlrpc_prep_req(mdc->mdc_ldlm_client, mdc->mdc_conn,
281 LDLM_ENQUEUE, 1, size, NULL);
285 /* get ready for the reply */
286 size[0] = sizeof(struct ldlm_reply);
287 req->rq_replen = lustre_msg_size(1, size);
291 #warning FIXME: the data here needs to be different if a lock was granted for a different inode
292 rc = ldlm_cli_enqueue(mdc->mdc_ldlm_client, mdc->mdc_conn, req,
293 obddev->obd_namespace, NULL, res_id, lock_type,
294 NULL, 0, lock_mode, &flags,
295 (void *)mdc_lock_callback, data, datalen, lockh);
298 CERROR("ldlm_cli_enqueue: %d\n", rc);
302 dlm_rep = lustre_msg_buf(req->rq_repmsg, 0);
303 it->it_disposition = (int) dlm_rep->lock_policy_res1;
304 it->it_status = (int) dlm_rep->lock_policy_res2;
305 it->it_lock_mode = lock_mode;
311 int mdc_open(struct obd_conn *conn, ino_t ino, int type, int flags,
313 __u64 cookie, __u64 *fh, struct ptlrpc_request **request)
315 struct mdc_obd *mdc = mdc_conn2mdc(conn);
316 struct mds_body *body;
317 int rc, size[2] = {sizeof(*body)}, bufcount = 1;
318 struct ptlrpc_request *req;
322 size[1] = sizeof(*obdo);
325 req = ptlrpc_prep_req(mdc->mdc_client, mdc->mdc_conn,
326 MDS_OPEN, bufcount, size, NULL);
328 GOTO(out, rc = -ENOMEM);
330 req->rq_flags |= PTL_RPC_FL_REPLAY;
331 req->rq_level = LUSTRE_CONN_FULL;
332 body = lustre_msg_buf(req->rq_reqmsg, 0);
334 ll_ino2fid(&body->fid1, ino, 0, type);
335 body->flags = HTON__u32(flags);
336 body->extra = cookie;
339 memcpy(lustre_msg_buf(req->rq_reqmsg, 1), obdo, sizeof(*obdo));
341 req->rq_replen = lustre_msg_size(1, size);
343 rc = ptlrpc_queue_wait(req);
344 rc = ptlrpc_check_status(req, rc);
347 body = lustre_msg_buf(req->rq_repmsg, 0);
348 mds_unpack_body(body);
358 int mdc_close(struct obd_conn *conn,
359 ino_t ino, int type, __u64 fh, struct ptlrpc_request **request)
361 struct mdc_obd *mdc = mdc_conn2mdc(conn);
362 struct mds_body *body;
363 int rc, size = sizeof(*body);
364 struct ptlrpc_request *req;
366 req = ptlrpc_prep_req(mdc->mdc_client, mdc->mdc_conn,
367 MDS_CLOSE, 1, &size, NULL);
369 GOTO(out, rc = -ENOMEM);
371 body = lustre_msg_buf(req->rq_reqmsg, 0);
372 ll_ino2fid(&body->fid1, ino, 0, type);
375 req->rq_level = LUSTRE_CONN_FULL;
376 req->rq_replen = lustre_msg_size(0, NULL);
378 rc = ptlrpc_queue_wait(req);
379 rc = ptlrpc_check_status(req, rc);
387 int mdc_readpage(struct obd_conn *conn, ino_t ino, int type, __u64 offset,
388 char *addr, struct ptlrpc_request **request)
390 struct mdc_obd *mdc = mdc_conn2mdc(conn);
391 struct ptlrpc_request *req = NULL;
392 struct ptlrpc_bulk_desc *desc = NULL;
393 struct ptlrpc_bulk_page *bulk = NULL;
394 struct mds_body *body;
395 int rc, size = sizeof(*body);
398 CDEBUG(D_INODE, "inode: %ld\n", (long)ino);
400 desc = ptlrpc_prep_bulk(mdc->mdc_conn);
402 GOTO(out, rc = -ENOMEM);
404 req = ptlrpc_prep_req(mdc->mdc_client, mdc->mdc_conn,
405 MDS_READPAGE, 1, &size, NULL);
407 GOTO(out2, rc = -ENOMEM);
409 bulk = ptlrpc_prep_bulk_page(desc);
410 bulk->b_buflen = PAGE_SIZE;
412 bulk->b_xid = req->rq_xid;
413 desc->b_portal = MDS_BULK_PORTAL;
415 rc = ptlrpc_register_bulk(desc);
417 CERROR("couldn't setup bulk sink: error %d.\n", rc);
421 body = lustre_msg_buf(req->rq_reqmsg, 0);
423 body->fid1.f_type = type;
426 req->rq_replen = lustre_msg_size(1, &size);
427 req->rq_level = LUSTRE_CONN_FULL;
428 rc = ptlrpc_queue_wait(req);
429 rc = ptlrpc_check_status(req, rc);
431 ptlrpc_abort_bulk(desc);
434 body = lustre_msg_buf(req->rq_repmsg, 0);
435 mds_unpack_body(body);
440 ptlrpc_free_bulk(desc);
447 int mdc_statfs(struct ptlrpc_client *cl, struct ptlrpc_connection *conn,
448 struct statfs *statfs,
449 struct ptlrpc_request **request)
451 struct mds_rec_setattr *rec;
452 struct ptlrpc_request *req;
453 int rc, size = sizeof(*rec);
456 req = ptlrpc_prep_req(cl, conn, MDS_STATFS, 1, &size, NULL);
460 rec = lustre_msg_buf(req->rq_reqmsg, 0);
461 mds_setattr_pack(rec, inode, iattr);
463 size = sizeof(struct mds_body);
464 req->rq_replen = lustre_msg_size(1, &size);
466 rc = mdc_reint(cl, req, LUSTRE_CONN_FULL);
468 if (rc == -ERESTARTSYS )
475 static int mdc_ioctl(long cmd, struct obd_conn *conn, int len, void *karg,
479 /* FIXME XXX : This should use the new ioc_data to pass args in */
481 struct ptlrpc_client cl;
482 struct ptlrpc_connection *conn;
483 struct ptlrpc_request *request;
487 if (_IOC_TYPE(cmd) != IOC_REQUEST_TYPE ||
488 _IOC_NR(cmd) < IOC_REQUEST_MIN_NR ||
489 _IOC_NR(cmd) > IOC_REQUEST_MAX_NR ) {
490 CDEBUG(D_IOCTL, "invalid ioctl ( type %d, nr %d, size %d )\n",
491 _IOC_TYPE(cmd), _IOC_NR(cmd), _IOC_SIZE(cmd));
495 ptlrpc_init_client(NULL, NULL,
496 MDS_REQUEST_PORTAL, MDC_REPLY_PORTAL, &cl);
497 connection = ptlrpc_uuid_to_connection("mds");
499 CERROR("cannot create client\n");
504 case IOC_REQUEST_GETATTR: {
505 CERROR("-- getting attr for ino %lu\n", arg);
506 err = mdc_getattr(&cl, connection, arg, S_IFDIR, ~0, 0,
508 CERROR("-- done err %d\n", err);
513 case IOC_REQUEST_READPAGE: {
515 OBD_ALLOC(buf, PAGE_SIZE);
520 CERROR("-- readpage 0 for ino %lu\n", arg);
521 err = mdc_readpage(&cl, connection, arg, S_IFDIR, 0, buf,
523 CERROR("-- done err %d\n", err);
524 OBD_FREE(buf, PAGE_SIZE);
529 case IOC_REQUEST_SETATTR: {
534 inode.i_generation = 0;
535 iattr.ia_mode = 040777;
537 iattr.ia_valid = ATTR_MODE | ATTR_ATIME;
539 err = mdc_setattr(&cl, connection, &inode, &iattr, &request);
540 CERROR("-- done err %d\n", err);
545 case IOC_REQUEST_CREATE: {
550 inode.i_generation = 0;
551 iattr.ia_mode = 040777;
553 iattr.ia_valid = ATTR_MODE | ATTR_ATIME;
555 err = mdc_create(&cl, connection, &inode,
556 "foofile", strlen("foofile"),
557 NULL, 0, 0100707, 47114711,
558 11, 47, 0, NULL, &request);
559 CERROR("-- done err %d\n", err);
564 case IOC_REQUEST_OPEN: {
566 copy_from_user(&ino, (__u64 *)arg, sizeof(ino));
567 CERROR("-- opening ino %llu\n", (unsigned long long)ino);
568 err = mdc_open(&cl, connection, ino, S_IFDIR, O_RDONLY, 4711,
570 copy_to_user((__u64 *)arg, &fh, sizeof(fh));
571 CERROR("-- done err %d (fh=%Lu)\n", err,
572 (unsigned long long)fh);
577 case IOC_REQUEST_CLOSE: {
578 CERROR("-- closing ino 2, filehandle %lu\n", arg);
579 err = mdc_close(&cl, connection, 2, S_IFDIR, arg, &request);
580 CERROR("-- done err %d\n", err);
586 GOTO(out, err = -EINVAL);
590 ptlrpc_free_req(request);
591 ptlrpc_put_connection(connection);
592 ptlrpc_cleanup_client(&cl);
599 static int mdc_setup(struct obd_device *obddev, obd_count len, void *buf)
601 struct obd_ioctl_data* data = buf;
602 struct mdc_obd *mdc = &obddev->u.mdc;
603 char server_uuid[37];
607 if (data->ioc_inllen1 < 1) {
608 CERROR("osc setup requires a TARGET UUID\n");
612 if (data->ioc_inllen1 > 37) {
613 CERROR("mdc UUID must be less than 38 characters\n");
617 if (data->ioc_inllen2 < 1) {
618 CERROR("mdc setup requires a SERVER UUID\n");
622 if (data->ioc_inllen2 > 37) {
623 CERROR("mdc UUID must be less than 38 characters\n");
627 memcpy(mdc->mdc_target_uuid, data->ioc_inlbuf1, data->ioc_inllen1);
628 memcpy(server_uuid, data->ioc_inlbuf2, MIN(data->ioc_inllen2,
629 sizeof(server_uuid)));
631 mdc->mdc_conn = ptlrpc_uuid_to_connection(server_uuid);
635 OBD_ALLOC(mdc->mdc_client, sizeof(*mdc->mdc_client));
636 if (mdc->mdc_client == NULL)
637 GOTO(out_conn, rc = -ENOMEM);
639 OBD_ALLOC(mdc->mdc_ldlm_client, sizeof(*mdc->mdc_ldlm_client));
640 if (mdc->mdc_ldlm_client == NULL)
641 GOTO(out_client, rc = -ENOMEM);
643 ptlrpc_init_client(NULL, NULL, MDS_REQUEST_PORTAL, MDC_REPLY_PORTAL,
645 ptlrpc_init_client(NULL, NULL, LDLM_REQUEST_PORTAL, LDLM_REPLY_PORTAL,
646 mdc->mdc_ldlm_client);
647 mdc->mdc_client->cli_name = "mdc";
648 mdc->mdc_ldlm_client->cli_name = "ldlm";
649 /* XXX get recovery hooked in here again */
650 //ptlrpc_init_client(ptlrpc_connmgr, ll_recover,...
652 ptlrpc_init_client(ptlrpc_connmgr, NULL,
653 MDS_REQUEST_PORTAL, MDC_REPLY_PORTAL,
660 OBD_FREE(mdc->mdc_client, sizeof(*mdc->mdc_client));
662 ptlrpc_put_connection(mdc->mdc_conn);
666 static int mdc_cleanup(struct obd_device * obddev)
668 struct mdc_obd *mdc = &obddev->u.mdc;
670 ptlrpc_cleanup_client(mdc->mdc_client);
671 OBD_FREE(mdc->mdc_client, sizeof(*mdc->mdc_client));
672 ptlrpc_cleanup_client(mdc->mdc_ldlm_client);
673 OBD_FREE(mdc->mdc_ldlm_client, sizeof(*mdc->mdc_ldlm_client));
674 ptlrpc_put_connection(mdc->mdc_conn);
680 static int mdc_connect(struct obd_conn *conn)
682 struct mdc_obd *mdc = mdc_conn2mdc(conn);
683 struct ptlrpc_request *request;
684 int rc, size = sizeof(mdc->mdc_target_uuid);
685 char *tmp = mdc->mdc_target_uuid;
689 conn->oc_dev->obd_namespace = ldlm_namespace_new(LDLM_NAMESPACE_CLIENT);
690 if (conn->oc_dev->obd_namespace == NULL)
693 request = ptlrpc_prep_req(mdc->mdc_client, mdc->mdc_conn,
694 MDS_CONNECT, 1, &size, &tmp);
698 request->rq_replen = lustre_msg_size(0, NULL);
700 rc = ptlrpc_queue_wait(request);
704 mdc->mdc_client->cli_target_devno = request->rq_repmsg->target_id;
705 mdc->mdc_ldlm_client->cli_target_devno =
706 mdc->mdc_client->cli_target_devno;
709 ptlrpc_free_req(request);
713 static int mdc_disconnect(struct obd_conn *conn)
715 struct mdc_obd *mdc = mdc_conn2mdc(conn);
716 struct ptlrpc_request *request;
717 struct mds_body *body;
718 int rc, size = sizeof(*body);
721 ldlm_namespace_free(conn->oc_dev->obd_namespace);
722 request = ptlrpc_prep_req(mdc->mdc_client, mdc->mdc_conn,
723 MDS_DISCONNECT, 1, &size,
728 body = lustre_msg_buf(request->rq_reqmsg, 0);
729 body->valid = conn->oc_id;
731 request->rq_replen = lustre_msg_size(0, NULL);
733 rc = ptlrpc_queue_wait(request);
736 ptlrpc_free_req(request);
740 struct obd_ops mdc_obd_ops = {
742 o_cleanup: mdc_cleanup,
743 o_connect: mdc_connect,
744 o_disconnect: mdc_disconnect,
745 o_iocontrol: mdc_ioctl
748 static int __init ptlrpc_request_init(void)
750 return obd_register_type(&mdc_obd_ops, LUSTRE_MDC_NAME);
753 static void __exit ptlrpc_request_exit(void)
755 obd_unregister_type(LUSTRE_MDC_NAME);
758 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
759 MODULE_DESCRIPTION("Lustre Metadata Client v1.0");
760 MODULE_LICENSE("GPL");
762 EXPORT_SYMBOL(mdc_getstatus);
763 EXPORT_SYMBOL(mdc_enqueue);
764 EXPORT_SYMBOL(mdc_getattr);
765 EXPORT_SYMBOL(mdc_create);
766 EXPORT_SYMBOL(mdc_unlink);
767 EXPORT_SYMBOL(mdc_rename);
768 EXPORT_SYMBOL(mdc_link);
769 EXPORT_SYMBOL(mdc_readpage);
770 EXPORT_SYMBOL(mdc_setattr);
771 EXPORT_SYMBOL(mdc_close);
772 EXPORT_SYMBOL(mdc_open);
774 module_init(ptlrpc_request_init);
775 module_exit(ptlrpc_request_exit);