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>
31 #include <linux/init.h>
32 #include <linux/lprocfs_status.h>
34 #define REQUEST_MINOR 244
36 extern int mds_queue_req(struct ptlrpc_request *);
37 extern struct lprocfs_vars status_var_nm_1[];
38 extern struct lprocfs_vars status_class_var[];
40 /* should become mdc_getinfo() */
41 int mdc_getstatus(struct lustre_handle *conn, struct ll_fid *rootfid)
43 struct ptlrpc_request *req;
44 struct mds_body *body;
45 int rc, size = sizeof(*body);
48 req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_GETSTATUS, 1, &size,
51 GOTO(out, rc = -ENOMEM);
53 body = lustre_msg_buf(req->rq_reqmsg, 0);
54 req->rq_level = LUSTRE_CONN_CON;
55 req->rq_replen = lustre_msg_size(1, &size);
57 mds_pack_req_body(req);
58 rc = ptlrpc_queue_wait(req);
59 rc = ptlrpc_check_status(req, rc);
62 body = lustre_msg_buf(req->rq_repmsg, 0);
63 mds_unpack_body(body);
64 memcpy(rootfid, &body->fid1, sizeof(*rootfid));
66 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
67 ", last_xid="LPU64"\n",
68 rootfid->id, req->rq_repmsg->last_committed,
69 req->rq_repmsg->last_xid);
74 ptlrpc_req_finished(req);
78 int mdc_getlovinfo(struct obd_device *obd, struct lustre_handle *mdc_connh,
79 struct ptlrpc_request **request)
81 struct ptlrpc_request *req;
82 struct mds_status_req *streq;
83 int rc, size[2] = {sizeof(*streq)};
86 req = ptlrpc_prep_req(class_conn2cliimp(mdc_connh), MDS_GETLOVINFO, 1,
89 GOTO(out, rc = -ENOMEM);
92 streq = lustre_msg_buf(req->rq_reqmsg, 0);
93 streq->flags = HTON__u32(MDS_STATUS_LOV);
94 streq->repbuf = HTON__u32(8192);
96 /* prepare for reply */
97 req->rq_level = LUSTRE_CONN_CON;
100 req->rq_replen = lustre_msg_size(2, size);
102 rc = ptlrpc_queue_wait(req);
103 rc = ptlrpc_check_status(req, rc);
110 int mdc_getattr(struct lustre_handle *conn,
111 obd_id ino, int type, unsigned long valid, size_t ea_size,
112 struct ptlrpc_request **request)
114 struct ptlrpc_request *req;
115 struct mds_body *body;
116 int rc, size[2] = {sizeof(*body), 0}, bufcount = 1;
119 req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_GETATTR, 1, size,
122 GOTO(out, rc = -ENOMEM);
124 body = lustre_msg_buf(req->rq_reqmsg, 0);
125 ll_ino2fid(&body->fid1, ino, 0, type);
129 size[bufcount] = ea_size;
131 body->size = ea_size;
132 CDEBUG(D_INODE, "reserving %d bytes for MD/symlink in packet\n",
135 req->rq_replen = lustre_msg_size(bufcount, size);
136 mds_pack_req_body(req);
138 rc = ptlrpc_queue_wait(req);
139 rc = ptlrpc_check_status(req, rc);
142 body = lustre_msg_buf(req->rq_repmsg, 0);
143 mds_unpack_body(body);
144 CDEBUG(D_NET, "mode: %o\n", body->mode);
153 void d_delete_aliases(struct inode *inode)
155 struct dentry *dentry = NULL;
156 struct list_head *tmp;
157 struct ll_sb_info *sbi = ll_i2sbi(inode);
160 spin_lock(&dcache_lock);
161 list_for_each(tmp, &inode->i_dentry) {
162 dentry = list_entry(tmp, struct dentry, d_alias);
164 list_del_init(&dentry->d_hash);
165 list_add(&dentry->d_hash, &sbi->ll_orphan_dentry_list);
168 spin_unlock(&dcache_lock);
172 static int mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
173 void *data, __u32 data_len, int flag)
176 struct lustre_handle lockh;
180 case LDLM_CB_BLOCKING:
181 ldlm_lock2handle(lock, &lockh);
182 rc = ldlm_cli_cancel(&lockh);
184 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
188 case LDLM_CB_CANCELING: {
189 /* Invalidate all dentries associated with this inode */
190 struct inode *inode = data;
192 #warning "FIXME: what tells us that 'inode' is valid at all?"
193 if (inode->i_state & I_FREEING)
196 LASSERT(inode != NULL);
197 LASSERT(data_len == sizeof(*inode));
199 if (S_ISDIR(inode->i_mode)) {
200 CDEBUG(D_INODE, "invalidating inode %ld\n",
203 ll_invalidate_inode_pages(inode);
206 if ( inode != inode->i_sb->s_root->d_inode ) {
207 /* XXX should this igrab move up 12 lines? */
208 LASSERT(igrab(inode) == inode);
209 d_delete_aliases(inode);
221 /* This should be called with both the request and the reply still packed. */
222 void mdc_store_inode_generation(struct ptlrpc_request *req, int reqoff,
225 struct mds_rec_create *rec = lustre_msg_buf(req->rq_reqmsg, reqoff);
226 struct mds_body *body = lustre_msg_buf(req->rq_repmsg, repoff);
228 memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
231 int mdc_enqueue(struct lustre_handle *conn, int lock_type,
232 struct lookup_intent *it, int lock_mode, struct inode *dir,
233 struct dentry *de, struct lustre_handle *lockh,
234 char *tgt, int tgtlen, void *data, int datalen)
236 struct ptlrpc_request *req;
237 struct obd_device *obddev = class_conn2obd(conn);
238 __u64 res_id[RES_NAME_SIZE] = {dir->i_ino, (__u64)dir->i_generation};
239 int size[6] = {sizeof(struct ldlm_request), sizeof(struct ldlm_intent)};
240 int rc, flags = LDLM_FL_HAS_INTENT;
241 int repsize[3] = {sizeof(struct ldlm_reply),
242 sizeof(struct mds_body),
243 obddev->u.cli.cl_max_mds_easize};
244 struct ldlm_reply *dlm_rep;
245 struct ldlm_intent *lit;
246 struct ldlm_request *lockreq;
249 LDLM_DEBUG_NOLOCK("mdsintent %s dir %ld", ldlm_it2str(it->it_op),
252 if (it->it_op & (IT_MKDIR | IT_CREAT | IT_SYMLINK | IT_MKNOD)) {
255 it->it_mode |= S_IFDIR;
257 case (IT_CREAT|IT_OPEN):
259 it->it_mode |= S_IFREG;
262 it->it_mode |= S_IFLNK;
265 it->it_mode &= ~current->fs->umask;
267 size[2] = sizeof(struct mds_rec_create);
268 size[3] = de->d_name.len + 1;
269 size[4] = tgtlen + 1;
270 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 5,
275 /* pack the intent */
276 lit = lustre_msg_buf(req->rq_reqmsg, 1);
277 lit->opc = NTOH__u64((__u64)it->it_op);
279 /* pack the intended request */
280 mds_create_pack(req, 2, dir, it->it_mode, 0, current->fsuid,
281 current->fsgid, CURRENT_TIME, de->d_name.name,
282 de->d_name.len, tgt, tgtlen);
283 req->rq_replen = lustre_msg_size(3, repsize);
284 } else if (it->it_op == IT_RENAME2) {
285 struct dentry *old_de = it->it_data;
287 size[2] = sizeof(struct mds_rec_rename);
288 size[3] = old_de->d_name.len + 1;
289 size[4] = de->d_name.len + 1;
290 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 5,
295 /* pack the intent */
296 lit = lustre_msg_buf(req->rq_reqmsg, 1);
297 lit->opc = NTOH__u64((__u64)it->it_op);
299 /* pack the intended request */
300 mds_rename_pack(req, 2, old_de->d_parent->d_inode, dir,
301 old_de->d_name.name, old_de->d_name.len,
302 de->d_name.name, de->d_name.len);
303 req->rq_replen = lustre_msg_size(3, repsize);
304 } else if (it->it_op == IT_LINK2) {
305 struct dentry *old_de = it->it_data;
307 size[2] = sizeof(struct mds_rec_link);
308 size[3] = de->d_name.len + 1;
309 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
314 /* pack the intent */
315 lit = lustre_msg_buf(req->rq_reqmsg, 1);
316 lit->opc = NTOH__u64((__u64)it->it_op);
318 /* pack the intended request */
319 mds_link_pack(req, 2, old_de->d_inode, dir,
320 de->d_name.name, de->d_name.len);
321 req->rq_replen = lustre_msg_size(3, repsize);
322 } else if (it->it_op == IT_UNLINK || it->it_op == IT_RMDIR) {
323 size[2] = sizeof(struct mds_rec_unlink);
324 size[3] = de->d_name.len + 1;
325 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
330 /* pack the intent */
331 lit = lustre_msg_buf(req->rq_reqmsg, 1);
332 lit->opc = NTOH__u64((__u64)it->it_op);
334 /* pack the intended request */
335 mds_unlink_pack(req, 2, dir, NULL,
336 it->it_op == IT_UNLINK ? S_IFREG : S_IFDIR,
337 de->d_name.name, de->d_name.len);
339 req->rq_replen = lustre_msg_size(3, repsize);
340 } else if (it->it_op & (IT_GETATTR | IT_RENAME | IT_LINK |
341 IT_OPEN | IT_SETATTR | IT_LOOKUP | IT_READLINK)) {
342 size[2] = sizeof(struct mds_body);
343 size[3] = de->d_name.len + 1;
345 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
350 /* pack the intent */
351 lit = lustre_msg_buf(req->rq_reqmsg, 1);
352 lit->opc = NTOH__u64((__u64)it->it_op);
354 /* pack the intended request */
355 mds_getattr_pack(req, 2, dir, de->d_name.name, de->d_name.len);
357 /* get ready for the reply */
358 req->rq_replen = lustre_msg_size(3, repsize);
359 } else if (it->it_op == IT_READDIR) {
360 req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 1,
365 /* get ready for the reply */
366 req->rq_replen = lustre_msg_size(1, repsize);
372 rc = ldlm_cli_enqueue(conn, req, obddev->obd_namespace, NULL, res_id,
373 lock_type, NULL, 0, lock_mode, &flags,
374 ldlm_completion_ast, mdc_blocking_ast, data,
377 if (it->it_op != IT_READDIR) {
378 /* XXX This should become a lustre_msg flag, but for now... */
379 __u32 *opp = lustre_msg_buf(req->rq_reqmsg, 2);
380 *opp |= REINT_REPLAYING;
384 /* This can go when we're sure that this can never happen */
387 if (rc == ELDLM_LOCK_ABORTED) {
389 memset(lockh, 0, sizeof(*lockh));
391 } else if (rc != 0) {
392 CERROR("ldlm_cli_enqueue: %d\n", rc);
395 /* The server almost certainly gave us a lock other than the one
396 * that we asked for. If we already have a matching lock, then
397 * cancel this one--we don't need two. */
398 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
399 struct lustre_handle lockh2;
402 LDLM_DEBUG(lock, "matching against this");
404 memcpy(&lockh2, lockh, sizeof(lockh2));
405 if (ldlm_lock_match(NULL, NULL, LDLM_PLAIN, NULL, 0, LCK_NL,
407 /* We already have a lock; cancel the old one */
408 ldlm_lock_decref(lockh, lock_mode);
409 ldlm_cli_cancel(lockh);
410 memcpy(lockh, &lockh2, sizeof(lockh2));
415 /* On replay, we don't want the lock granted. */
416 lockreq = lustre_msg_buf(req->rq_reqmsg, 0);
417 lockreq->lock_flags |= LDLM_FL_INTENT_ONLY;
419 dlm_rep = lustre_msg_buf(req->rq_repmsg, 0);
420 it->it_disposition = (int) dlm_rep->lock_policy_res1;
421 it->it_status = (int) dlm_rep->lock_policy_res2;
422 it->it_lock_mode = lock_mode;
428 int mdc_cancel_unused(struct lustre_handle *conn, struct inode *inode,
431 __u64 res_id[RES_NAME_SIZE] = {inode->i_ino, inode->i_generation};
432 struct obd_device *obddev = class_conn2obd(conn);
434 RETURN(ldlm_cli_cancel_unused(obddev->obd_namespace, res_id, flags));
437 struct replay_open_data {
438 struct lustre_handle *fh;
441 static void mdc_replay_open(struct ptlrpc_request *req)
444 struct replay_open_data *saved;
445 struct mds_body *body = lustre_msg_buf(req->rq_repmsg, 0);
447 if (lustre_msg_get_op_flags(req->rq_reqmsg) & MDS_OPEN_HAS_EA)
452 saved = lustre_msg_buf(req->rq_reqmsg, offset);
453 mds_unpack_body(body);
454 CDEBUG(D_HA, "updating from "LPD64"/"LPD64" to "LPD64"/"LPD64"\n",
455 saved->fh->addr, saved->fh->cookie,
456 body->handle.addr, body->handle.cookie);
457 memcpy(saved->fh, &body->handle, sizeof(body->handle));
460 int mdc_open(struct lustre_handle *conn, obd_id ino, int type, int flags,
461 struct lov_mds_md *lmm, int lmm_size, struct lustre_handle *fh,
462 struct ptlrpc_request **request)
464 struct mds_body *body;
465 struct replay_open_data *replay_data;
466 int rc, size[3] = {sizeof(*body), sizeof(*replay_data)}, bufcount = 2;
467 struct ptlrpc_request *req;
470 if (lmm && lmm_size) {
472 size[2] = size[1]; /* shuffle the spare data along */
476 req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_OPEN, bufcount, size,
479 GOTO(out, rc = -ENOMEM);
481 req->rq_flags |= PTL_RPC_FL_REPLAY;
482 body = lustre_msg_buf(req->rq_reqmsg, 0);
484 ll_ino2fid(&body->fid1, ino, 0, type);
485 body->flags = HTON__u32(flags);
486 memcpy(&body->handle, fh, sizeof(body->handle));
488 if (lmm && lmm_size) {
489 CDEBUG(D_INODE, "sending %u bytes MD for ino LPU64\n",
491 lustre_msg_set_op_flags(req->rq_reqmsg, MDS_OPEN_HAS_EA);
492 memcpy(lustre_msg_buf(req->rq_reqmsg, 1), lmm, lmm_size);
493 body->flags |= HTON__u32(OBD_MD_FLEASIZE);
496 req->rq_replen = lustre_msg_size(1, size);
498 rc = ptlrpc_queue_wait(req);
499 rc = ptlrpc_check_status(req, rc);
501 body = lustre_msg_buf(req->rq_repmsg, 0);
502 mds_unpack_body(body);
503 memcpy(fh, &body->handle, sizeof(*fh));
506 /* If open is replayed, we need to fix up the fh. */
507 req->rq_replay_cb = mdc_replay_open;
508 replay_data = lustre_msg_buf(req->rq_reqmsg, lmm ? 2 : 1);
509 replay_data->fh = fh;
517 int mdc_close(struct lustre_handle *conn, obd_id ino, int type,
518 struct lustre_handle *fh, struct ptlrpc_request **request)
520 struct mds_body *body;
521 int rc, size = sizeof(*body);
522 struct ptlrpc_request *req;
524 req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_CLOSE, 1, &size,
527 GOTO(out, rc = -ENOMEM);
529 body = lustre_msg_buf(req->rq_reqmsg, 0);
530 ll_ino2fid(&body->fid1, ino, 0, type);
531 memcpy(&body->handle, fh, sizeof(body->handle));
533 req->rq_replen = lustre_msg_size(0, NULL);
535 rc = ptlrpc_queue_wait(req);
536 rc = ptlrpc_check_status(req, rc);
544 int mdc_readpage(struct lustre_handle *conn, obd_id ino, int type, __u64 offset,
545 char *addr, struct ptlrpc_request **request)
547 struct ptlrpc_connection *connection =
548 client_conn2cli(conn)->cl_import.imp_connection;
549 struct ptlrpc_request *req = NULL;
550 struct ptlrpc_bulk_desc *desc = NULL;
551 struct ptlrpc_bulk_page *bulk = NULL;
552 struct mds_body *body;
553 int rc, size = sizeof(*body);
556 CDEBUG(D_INODE, "inode: %ld\n", (long)ino);
558 desc = ptlrpc_prep_bulk(connection);
560 GOTO(out, rc = -ENOMEM);
562 req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_READPAGE, 1, &size,
565 GOTO(out2, rc = -ENOMEM);
567 bulk = ptlrpc_prep_bulk_page(desc);
568 bulk->bp_buflen = PAGE_SIZE;
570 bulk->bp_xid = req->rq_xid;
571 desc->bd_portal = MDS_BULK_PORTAL;
573 rc = ptlrpc_register_bulk(desc);
575 CERROR("couldn't setup bulk sink: error %d.\n", rc);
579 mds_readdir_pack(req, 0, ino, type);
581 req->rq_replen = lustre_msg_size(1, &size);
582 rc = ptlrpc_queue_wait(req);
583 rc = ptlrpc_check_status(req, rc);
585 ptlrpc_abort_bulk(desc);
588 body = lustre_msg_buf(req->rq_repmsg, 0);
589 mds_unpack_body(body);
594 ptlrpc_free_bulk(desc);
600 static int mdc_statfs(struct lustre_handle *conn, struct obd_statfs *osfs)
602 struct ptlrpc_request *req;
603 int rc, size = sizeof(*osfs);
606 req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_STATFS, 0, NULL,
611 req->rq_replen = lustre_msg_size(1, &size);
613 rc = ptlrpc_queue_wait(req);
614 rc = ptlrpc_check_status(req, rc);
619 obd_statfs_unpack(osfs, lustre_msg_buf(req->rq_repmsg, 0));
623 ptlrpc_req_finished(req);
627 int mdc_attach(struct obd_device *dev, obd_count len, void *data)
629 return lprocfs_reg_obd(dev, status_var_nm_1, dev);
632 int mdc_detach(struct obd_device *dev)
634 return lprocfs_dereg_obd(dev);
636 struct obd_ops mdc_obd_ops = {
637 o_attach: mdc_attach,
638 o_detach: mdc_detach,
639 o_setup: client_obd_setup,
640 o_cleanup: client_obd_cleanup,
641 o_connect: client_obd_connect,
642 o_disconnect: client_obd_disconnect,
643 o_statfs: mdc_statfs,
646 static int __init ptlrpc_request_init(void)
648 return class_register_type(&mdc_obd_ops, status_class_var,
652 static void __exit ptlrpc_request_exit(void)
654 class_unregister_type(LUSTRE_MDC_NAME);
657 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
658 MODULE_DESCRIPTION("Lustre Metadata Client v1.0");
659 MODULE_LICENSE("GPL");
661 EXPORT_SYMBOL(d_delete_aliases);
662 EXPORT_SYMBOL(mdc_getstatus);
663 EXPORT_SYMBOL(mdc_getlovinfo);
664 EXPORT_SYMBOL(mdc_enqueue);
665 EXPORT_SYMBOL(mdc_cancel_unused);
666 EXPORT_SYMBOL(mdc_getattr);
667 EXPORT_SYMBOL(mdc_create);
668 EXPORT_SYMBOL(mdc_unlink);
669 EXPORT_SYMBOL(mdc_rename);
670 EXPORT_SYMBOL(mdc_link);
671 EXPORT_SYMBOL(mdc_readpage);
672 EXPORT_SYMBOL(mdc_setattr);
673 EXPORT_SYMBOL(mdc_close);
674 EXPORT_SYMBOL(mdc_open);
676 EXPORT_SYMBOL(mdc_store_inode_generation);
678 module_init(ptlrpc_request_init);
679 module_exit(ptlrpc_request_exit);