1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
5 * Lustre Metadata Server (mds) request handler
7 * Copyright (c) 2001, 2002 Cluster File Systems, Inc.
8 * Author: Peter Braam <braam@clusterfs.com>
9 * Author: Andreas Dilger <adilger@clusterfs.com>
10 * Author: Phil Schwan <phil@clusterfs.com>
11 * Author: Mike Shaver <shaver@clusterfs.com>
13 * This file is part of Lustre, http://www.lustre.org.
15 * Lustre is free software; you can redistribute it and/or
16 * modify it under the terms of version 2 of the GNU General Public
17 * License as published by the Free Software Foundation.
19 * Lustre is distributed in the hope that it will be useful,
20 * but WITHOUT ANY WARRANTY; without even the implied warranty of
21 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
22 * GNU General Public License for more details.
24 * You should have received a copy of the GNU General Public License
25 * along with Lustre; if not, write to the Free Software
26 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
30 #define DEBUG_SUBSYSTEM S_MDS
32 #include <linux/module.h>
33 #include <linux/lustre_mds.h>
34 #include <linux/lustre_dlm.h>
35 #include <linux/init.h>
36 #include <linux/obd_class.h>
37 #include <linux/random.h>
38 #include <linux/locks.h>
39 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
40 #include <linux/buffer_head.h>
42 #include <linux/obd_lov.h>
43 #include <linux/lprocfs_status.h>
45 static kmem_cache_t *mds_file_cache;
47 extern int mds_get_lovtgts(struct mds_obd *obd, int tgt_count,
48 obd_uuid_t *uuidarray);
49 extern int mds_get_lovdesc(struct mds_obd *obd, struct lov_desc *desc);
50 extern void mds_start_transno(struct mds_obd *mds);
51 extern int mds_finish_transno(struct mds_obd *mds, void *handle,
52 struct ptlrpc_request *req, int rc);
53 static int mds_cleanup(struct obd_device * obddev);
55 extern struct lprocfs_vars status_var_nm_1[];
56 extern struct lprocfs_vars status_class_var[];
58 inline struct mds_obd *mds_req2mds(struct ptlrpc_request *req)
60 return &req->rq_export->exp_obd->u.mds;
63 static int mds_bulk_timeout(void *data)
65 struct ptlrpc_bulk_desc *desc = data;
68 recovd_conn_fail(desc->bd_connection);
72 /* Assumes caller has already pushed into the kernel filesystem context */
73 static int mds_sendpage(struct ptlrpc_request *req, struct file *file,
77 struct mds_obd *mds = mds_req2mds(req);
78 struct ptlrpc_bulk_desc *desc;
79 struct ptlrpc_bulk_page *bulk;
80 struct l_wait_info lwi;
84 desc = ptlrpc_prep_bulk(req->rq_connection);
86 GOTO(out, rc = -ENOMEM);
88 bulk = ptlrpc_prep_bulk_page(desc);
90 GOTO(cleanup_bulk, rc = -ENOMEM);
92 OBD_ALLOC(buf, PAGE_SIZE);
94 GOTO(cleanup_bulk, rc = -ENOMEM);
96 rc = mds_fs_readpage(mds, file, buf, PAGE_SIZE, (loff_t *)&offset);
99 GOTO(cleanup_buf, rc = -EIO);
101 bulk->bp_xid = req->rq_xid;
103 bulk->bp_buflen = PAGE_SIZE;
104 desc->bd_portal = MDS_BULK_PORTAL;
106 rc = ptlrpc_send_bulk(desc);
108 GOTO(cleanup_buf, rc);
110 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) {
111 CERROR("obd_fail_loc=%x, fail operation rc=%d\n",
112 OBD_FAIL_MDS_SENDPAGE, rc);
113 ptlrpc_abort_bulk(desc);
114 GOTO(cleanup_buf, rc);
117 lwi = LWI_TIMEOUT(obd_timeout * HZ, mds_bulk_timeout, desc);
118 rc = l_wait_event(desc->bd_waitq, desc->bd_flags & PTL_BULK_FL_SENT,
121 if (rc != -ETIMEDOUT)
123 GOTO(cleanup_buf, rc);
128 OBD_FREE(buf, PAGE_SIZE);
130 ptlrpc_free_bulk(desc);
136 * Look up a named entry in a directory, and get an LDLM lock on it.
137 * 'dir' is a inode for which an LDLM lock has already been taken.
139 * If we do not need an exclusive or write lock on this entry (e.g.
140 * a read lock for attribute lookup only) then we do not hold the
141 * directory semaphore on return. It is up to the caller to know what
142 * type of lock it is getting, and clean up appropriately.
144 struct dentry *mds_name2locked_dentry(struct obd_device *obd,
145 struct dentry *dir, struct vfsmount **mnt,
146 char *name, int namelen, int lock_mode,
147 struct lustre_handle *lockh,
150 struct dentry *dchild;
152 __u64 res_id[3] = {0};
155 down(&dir->d_inode->i_sem);
156 dchild = lookup_one_len(name, dir, namelen);
157 if (IS_ERR(dchild)) {
158 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
159 up(&dir->d_inode->i_sem);
163 if (dir_lock_mode != LCK_EX && dir_lock_mode != LCK_PW) {
164 up(&dir->d_inode->i_sem);
165 ldlm_lock_decref(lockh, dir_lock_mode);
168 if (lock_mode == 0 || !dchild->d_inode)
171 res_id[0] = dchild->d_inode->i_ino;
172 res_id[1] = dchild->d_inode->i_generation;
173 rc = ldlm_match_or_enqueue(NULL, NULL, obd->obd_namespace, NULL,
174 res_id, LDLM_PLAIN, NULL, 0, lock_mode,
175 &flags, ldlm_completion_ast,
176 mds_blocking_ast, NULL, 0, lockh);
177 if (rc != ELDLM_OK) {
179 up(&dir->d_inode->i_sem);
180 RETURN(ERR_PTR(-ENOLCK)); /* XXX translate ldlm code */
186 struct dentry *mds_fid2locked_dentry(struct obd_device *obd, struct ll_fid *fid,
187 struct vfsmount **mnt, int lock_mode,
188 struct lustre_handle *lockh)
190 struct mds_obd *mds = &obd->u.mds;
191 struct dentry *de = mds_fid2dentry(mds, fid, mnt), *retval = de;
193 __u64 res_id[3] = {0};
199 res_id[0] = de->d_inode->i_ino;
200 res_id[1] = de->d_inode->i_generation;
201 rc = ldlm_match_or_enqueue(NULL, NULL, obd->obd_namespace, NULL,
202 res_id, LDLM_PLAIN, NULL, 0, lock_mode,
203 &flags, ldlm_completion_ast,
204 mds_blocking_ast, NULL, 0, lockh);
205 if (rc != ELDLM_OK) {
207 retval = ERR_PTR(-ENOLCK); /* XXX translate ldlm code */
213 #ifndef DCACHE_DISCONNECTED
214 #define DCACHE_DISCONNECTED DCACHE_NFSD_DISCONNECTED
217 /* Look up an entry by inode number. */
218 struct dentry *mds_fid2dentry(struct mds_obd *mds, struct ll_fid *fid,
219 struct vfsmount **mnt)
221 /* stolen from NFS */
222 struct super_block *sb = mds->mds_sb;
223 unsigned long ino = fid->id;
224 __u32 generation = fid->generation;
226 struct list_head *lp;
227 struct dentry *result;
230 RETURN(ERR_PTR(-ESTALE));
232 inode = iget(sb, ino);
234 RETURN(ERR_PTR(-ENOMEM));
236 CDEBUG(D_DENTRY, "--> mds_fid2dentry: sb %p\n", inode->i_sb);
238 if (is_bad_inode(inode) ||
239 (generation && inode->i_generation != generation)) {
240 /* we didn't find the right inode.. */
241 CERROR("bad inode %lu, link: %d ct: %d or version %u/%u\n",
242 inode->i_ino, inode->i_nlink,
243 atomic_read(&inode->i_count), inode->i_generation,
246 RETURN(ERR_PTR(-ENOENT));
249 /* now to find a dentry. If possible, get a well-connected one */
251 *mnt = mds->mds_vfsmnt;
252 spin_lock(&dcache_lock);
253 list_for_each(lp, &inode->i_dentry) {
254 result = list_entry(lp, struct dentry, d_alias);
255 if (!(result->d_flags & DCACHE_DISCONNECTED)) {
257 result->d_vfs_flags |= DCACHE_REFERENCED;
258 spin_unlock(&dcache_lock);
265 spin_unlock(&dcache_lock);
266 result = d_alloc_root(inode);
267 if (result == NULL) {
269 return ERR_PTR(-ENOMEM);
273 result->d_flags |= DCACHE_DISCONNECTED;
277 /* Establish a connection to the MDS.
279 * This will set up an export structure for the client to hold state data
280 * about that client, like open files, the last operation number it did
281 * on the server, etc.
283 static int mds_connect(struct lustre_handle *conn, struct obd_device *obd,
284 obd_uuid_t cluuid, struct recovd_obd *recovd,
285 ptlrpc_recovery_cb_t recover)
287 struct obd_export *exp;
288 struct mds_export_data *med;
289 struct mds_client_data *mcd;
294 if (!conn || !obd || !cluuid)
299 spin_lock(&obd->obd_dev_lock);
300 list_for_each(p, &obd->obd_exports) {
301 exp = list_entry(p, struct obd_export, exp_obd_chain);
302 mcd = exp->exp_mds_data.med_mcd;
304 CERROR("FYI: NULL mcd - simultaneous connects\n");
307 if (!memcmp(cluuid, mcd->mcd_uuid, sizeof mcd->mcd_uuid)) {
308 /* XXX make handle-found-export a subroutine */
309 LASSERT(exp->exp_obd == obd);
311 spin_unlock(&obd->obd_dev_lock);
312 if (exp->exp_connection) {
313 struct lustre_handle *hdl;
314 hdl = &exp->exp_ldlm_data.led_import.imp_handle;
315 /* Might be a re-connect after a partition. */
316 if (!memcmp(conn, hdl, sizeof *conn)) {
317 CERROR("%s reconnecting\n", cluuid);
318 conn->addr = (__u64) (unsigned long)exp;
319 conn->cookie = exp->exp_cookie;
322 CERROR("%s reconnecting from %s, "
323 "handle mismatch (ours %Lx/%Lx, "
324 "theirs %Lx/%Lx)\n", cluuid,
325 exp->exp_connection->
326 c_remote_uuid, hdl->addr,
327 hdl->cookie, conn->addr,
329 /* XXX disconnect them here? */
330 memset(conn, 0, sizeof *conn);
336 conn->addr = (__u64) (unsigned long)exp;
337 conn->cookie = exp->exp_cookie;
338 CDEBUG(D_INFO, "existing export for UUID '%s' at %p\n",
340 CDEBUG(D_IOCTL,"connect: addr %Lx cookie %Lx\n",
341 (long long)conn->addr, (long long)conn->cookie);
345 spin_unlock(&obd->obd_dev_lock);
347 if (obd->u.mds.mds_recoverable_clients != 0) {
348 CERROR("denying connection for new client %s: in recovery\n",
354 /* XXX There is a small race between checking the list and adding a
355 * new connection for the same UUID, but the real threat (list
356 * corruption when multiple different clients connect) is solved.
358 * There is a second race between adding the export to the list,
359 * and filling in the client data below. Hence skipping the case
360 * of NULL mcd above. We should already be controlling multiple
361 * connects at the client, and we can't hold the spinlock over
362 * memory allocations without risk of deadlocking.
364 rc = class_connect(conn, obd, cluuid);
367 exp = class_conn2export(conn);
369 med = &exp->exp_mds_data;
371 OBD_ALLOC(mcd, sizeof(*mcd));
373 CERROR("mds: out of memory for client data\n");
374 GOTO(out_export, rc = -ENOMEM);
377 memcpy(mcd->mcd_uuid, cluuid, sizeof(mcd->mcd_uuid));
380 INIT_LIST_HEAD(&med->med_open_head);
381 spin_lock_init(&med->med_open_lock);
383 rc = mds_client_add(&obd->u.mds, med, -1);
390 OBD_FREE(mcd, sizeof(*mcd));
392 class_disconnect(conn);
399 /* Call with med->med_open_lock held, please. */
400 inline int mds_close_mfd(struct mds_file_data *mfd, struct mds_export_data *med)
402 struct file *file = mfd->mfd_file;
403 LASSERT(file->private_data == mfd);
405 list_del(&mfd->mfd_list);
406 mfd->mfd_servercookie = DEAD_HANDLE_MAGIC;
407 kmem_cache_free(mds_file_cache, mfd);
409 return filp_close(file, 0);
412 static int mds_disconnect(struct lustre_handle *conn)
414 struct obd_export *export = class_conn2export(conn);
415 struct list_head *tmp, *n;
416 struct mds_export_data *med = &export->exp_mds_data;
421 * Close any open files.
423 spin_lock(&med->med_open_lock);
424 list_for_each_safe(tmp, n, &med->med_open_head) {
425 struct mds_file_data *mfd =
426 list_entry(tmp, struct mds_file_data, mfd_list);
427 rc = mds_close_mfd(mfd, med);
429 /* XXX better diagnostics, with file path and stuff */
430 CDEBUG(D_INODE, "Error %d closing mfd %p\n", rc, mfd);
433 spin_unlock(&med->med_open_lock);
435 ldlm_cancel_locks_for_export(export);
436 mds_client_free(export);
438 rc = class_disconnect(conn);
446 * XXX This is NOT guaranteed to flush all transactions to disk (even though
447 * it is equivalent to calling sync()) because it only _starts_ the flush
448 * and does not wait for completion. It's better than nothing though.
449 * What we really want is a mild form of fsync_dev_lockfs(), but it is
450 * non-standard, or enabling do_sync_supers in ext3, just for this call.
452 static void mds_fsync_super(struct super_block *sb)
456 if (sb->s_dirt && sb->s_op && sb->s_op->write_super)
457 sb->s_op->write_super(sb);
462 static int mds_getstatus(struct ptlrpc_request *req)
464 struct mds_obd *mds = mds_req2mds(req);
465 struct mds_body *body;
466 int rc, size = sizeof(*body);
469 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
470 if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK)) {
471 CERROR("mds: out of memory for message: size=%d\n", size);
472 req->rq_status = -ENOMEM;
476 /* Flush any outstanding transactions to disk so the client will
477 * get the latest last_committed value and can drop their local
478 * requests if they have any. This would be fsync_super() if it
481 mds_fsync_super(mds->mds_sb);
483 body = lustre_msg_buf(req->rq_repmsg, 0);
484 memcpy(&body->fid1, &mds->mds_rootfid, sizeof(body->fid1));
486 /* the last_committed and last_xid fields are filled in for all
487 * replies already - no need to do so here also.
492 static int mds_getlovinfo(struct ptlrpc_request *req)
494 struct mds_obd *mds = mds_req2mds(req);
495 struct mds_status_req *streq;
496 struct lov_desc *desc;
498 int rc, size[2] = {sizeof(*desc)};
501 streq = lustre_msg_buf(req->rq_reqmsg, 0);
502 streq->flags = NTOH__u32(streq->flags);
503 streq->repbuf = NTOH__u32(streq->repbuf);
504 size[1] = streq->repbuf;
506 rc = lustre_pack_msg(2, size, NULL, &req->rq_replen, &req->rq_repmsg);
508 CERROR("mds: out of memory for message: size=%d\n", size[1]);
509 req->rq_status = -ENOMEM;
513 desc = lustre_msg_buf(req->rq_repmsg, 0);
514 rc = mds_get_lovdesc(mds, desc);
520 tgt_count = le32_to_cpu(desc->ld_tgt_count);
521 if (tgt_count * sizeof(obd_uuid_t) > streq->repbuf) {
522 CERROR("too many targets, enlarge client buffers\n");
523 req->rq_status = -ENOSPC;
527 /* XXX the MDS should not really know about this */
528 mds->mds_max_mdsize = lov_mds_md_size(tgt_count);
529 rc = mds_get_lovtgts(mds, tgt_count,
530 lustre_msg_buf(req->rq_repmsg, 1));
532 CERROR("get_lovtgts error %d\n", rc);
539 int mds_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
540 void *data, __u32 data_len, int flag)
545 if (flag == LDLM_CB_CANCELING) {
546 /* Don't need to do anything here. */
550 /* XXX layering violation! -phil */
551 l_lock(&lock->l_resource->lr_namespace->ns_lock);
552 lock->l_flags |= LDLM_FL_CBPENDING;
553 do_ast = (!lock->l_readers && !lock->l_writers);
554 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
557 struct lustre_handle lockh;
560 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
561 ldlm_lock2handle(lock, &lockh);
562 rc = ldlm_cli_cancel(&lockh);
564 CERROR("ldlm_cli_cancel: %d\n", rc);
566 LDLM_DEBUG(lock, "Lock still has references, will be"
571 int mds_pack_md(struct mds_obd *mds, struct ptlrpc_request *req,
572 int offset, struct mds_body *body, struct inode *inode)
574 struct lov_mds_md *lmm;
575 int lmm_size = req->rq_repmsg->buflens[offset];
579 CDEBUG(D_INFO, "no space reserved for inode %u MD\n", inode->i_ino);
583 lmm = lustre_msg_buf(req->rq_repmsg, offset);
585 /* I don't really like this, but it is a sanity check on the client
586 * MD request. However, if the client doesn't know how much space
587 * to reserve for the MD, this shouldn't be fatal either...
589 if (lmm_size > mds->mds_max_mdsize) {
590 CERROR("Reading MD for inode %u of %d bytes > max %d\n",
591 inode->i_ino, lmm_size, mds->mds_max_mdsize);
595 /* We don't need to store the reply size, because this buffer is
596 * discarded right after unpacking, and the LOV can figure out the
597 * size itself from the ost count.
599 if ((rc = mds_fs_get_md(mds, inode, lmm, lmm_size)) < 0) {
600 CDEBUG(D_INFO, "No md for ino %u: rc = %d\n", inode->i_ino, rc);
602 body->valid |= OBD_MD_FLEASIZE;
609 static int mds_getattr_internal(struct mds_obd *mds, struct dentry *dentry,
610 struct ptlrpc_request *req,
611 struct mds_body *reqbody, int reply_off)
613 struct mds_body *body;
614 struct inode *inode = dentry->d_inode;
621 body = lustre_msg_buf(req->rq_repmsg, reply_off);
623 mds_pack_inode2fid(&body->fid1, inode);
624 mds_pack_inode2body(body, inode);
626 if (S_ISREG(inode->i_mode)) {
627 rc = mds_pack_md(mds, req, reply_off + 1, body, inode);
628 } else if (S_ISLNK(inode->i_mode) && reqbody->valid & OBD_MD_LINKNAME) {
629 char *symname = lustre_msg_buf(req->rq_repmsg, reply_off + 1);
630 int len = req->rq_repmsg->buflens[reply_off + 1];
632 rc = inode->i_op->readlink(dentry, symname, len);
634 CERROR("readlink failed: %d\n", rc);
636 CDEBUG(D_INODE, "read symlink dest %s\n", symname);
637 body->valid |= OBD_MD_LINKNAME;
643 static int mds_getattr_name(int offset, struct ptlrpc_request *req)
645 struct mds_obd *mds = mds_req2mds(req);
646 struct obd_device *obd = req->rq_export->exp_obd;
647 struct obd_run_ctxt saved;
648 struct mds_body *body;
649 struct dentry *de = NULL, *dchild = NULL;
651 struct lustre_handle lockh;
653 int namelen, flags = 0, lock_mode, rc = 0;
655 __u64 res_id[3] = {0, 0, 0};
658 LASSERT(!strcmp(req->rq_export->exp_obd->obd_type->typ_name, "mds"));
660 if (req->rq_reqmsg->bufcount <= offset + 1) {
662 GOTO(out_pre_de, rc = -EINVAL);
665 body = lustre_msg_buf(req->rq_reqmsg, offset);
666 name = lustre_msg_buf(req->rq_reqmsg, offset + 1);
667 namelen = req->rq_reqmsg->buflens[offset + 1];
668 /* requests were at offset 2, replies go back at 1 */
672 uc.ouc_fsuid = body->fsuid;
673 uc.ouc_fsgid = body->fsgid;
674 uc.ouc_cap = body->capability;
675 push_ctxt(&saved, &mds->mds_ctxt, &uc);
676 de = mds_fid2dentry(mds, &body->fid1, NULL);
678 GOTO(out_pre_de, rc = -ENOENT);
682 CDEBUG(D_INODE, "parent ino %ld, name %*s\n", dir->i_ino,namelen,name);
685 res_id[0] = dir->i_ino;
686 res_id[1] = dir->i_generation;
688 rc = ldlm_lock_match(obd->obd_namespace, res_id, LDLM_PLAIN,
689 NULL, 0, lock_mode, &lockh);
691 LDLM_DEBUG_NOLOCK("enqueue res "LPU64, res_id[0]);
692 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
693 res_id, LDLM_PLAIN, NULL, 0, lock_mode,
694 &flags, ldlm_completion_ast,
695 mds_blocking_ast, NULL, 0, &lockh);
696 if (rc != ELDLM_OK) {
697 CERROR("lock enqueue: err: %d\n", rc);
698 GOTO(out_create_de, rc = -EIO);
701 ldlm_lock_dump((void *)(unsigned long)lockh.addr);
704 dchild = lookup_one_len(name, de, namelen - 1);
705 if (IS_ERR(dchild)) {
706 CDEBUG(D_INODE, "child lookup error %ld\n", PTR_ERR(dchild));
708 GOTO(out_create_dchild, rc = PTR_ERR(dchild));
711 rc = mds_getattr_internal(mds, dchild, req, body, offset);
717 ldlm_lock_decref(&lockh, lock_mode);
722 pop_ctxt(&saved, &mds->mds_ctxt, &uc);
726 static int mds_getattr(int offset, struct ptlrpc_request *req)
728 struct mds_obd *mds = mds_req2mds(req);
729 struct obd_run_ctxt saved;
732 struct mds_body *body;
734 int rc = 0, size[2] = {sizeof(*body)}, bufcount = 1;
737 body = lustre_msg_buf(req->rq_reqmsg, offset);
738 uc.ouc_fsuid = body->fsuid;
739 uc.ouc_fsgid = body->fsgid;
740 uc.ouc_cap = body->capability;
741 push_ctxt(&saved, &mds->mds_ctxt, &uc);
742 de = mds_fid2dentry(mds, &body->fid1, NULL);
744 rc = req->rq_status = -ENOENT;
745 GOTO(out_pop, PTR_ERR(de));
749 if (S_ISREG(body->fid1.f_type)) {
750 int rc = mds_fs_get_md(mds, inode, NULL, 0);
751 CDEBUG(D_INODE, "got %d bytes MD data for inode %u\n",
755 CERROR("error getting inode %u MD: rc = %d\n",
758 } else if (rc > mds->mds_max_mdsize) {
760 CERROR("MD size %d larger than maximum possible %u\n",
761 rc, mds->mds_max_mdsize);
765 } else if (body->valid & OBD_MD_LINKNAME) {
766 size[bufcount] = MIN(inode->i_size + 1, body->size);
768 CDEBUG(D_INODE, "symlink size: %d, reply space: %d\n",
769 inode->i_size + 1, body->size);
772 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) {
773 CERROR("failed MDS_GETATTR_PACK test\n");
774 req->rq_status = -ENOMEM;
775 GOTO(out, rc = -ENOMEM);
778 rc = lustre_pack_msg(bufcount, size, NULL, &req->rq_replen,
781 CERROR("out of memoryK\n");
786 req->rq_status = mds_getattr_internal(mds, de, req, body, 0);
791 pop_ctxt(&saved, &mds->mds_ctxt, &uc);
795 static int mds_statfs(struct ptlrpc_request *req)
797 struct mds_obd *mds = mds_req2mds(req);
798 struct obd_statfs *osfs;
800 int rc, size = sizeof(*osfs);
803 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
804 if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
805 CERROR("mds: statfs lustre_pack_msg failed: rc = %d\n", rc);
809 rc = mds_fs_statfs(mds, &sfs);
811 CERROR("mds: statfs failed: rc %d\n", rc);
814 osfs = lustre_msg_buf(req->rq_repmsg, 0);
815 memset(osfs, 0, size);
816 statfs_pack(osfs, &sfs);
817 obd_statfs_pack(osfs, osfs);
824 static struct mds_file_data *mds_handle2mfd(struct lustre_handle *handle)
826 struct mds_file_data *mfd = NULL;
828 if (!handle || !handle->addr)
831 mfd = (struct mds_file_data *)(unsigned long)(handle->addr);
832 if (!kmem_cache_validate(mds_file_cache, mfd))
835 if (mfd->mfd_servercookie != handle->cookie)
841 static int mds_store_md(struct mds_obd *mds, struct ptlrpc_request *req,
842 int offset, struct mds_body *body, struct inode *inode)
844 struct lov_mds_md *lmm = lustre_msg_buf(req->rq_reqmsg, offset);
845 int lmm_size = req->rq_reqmsg->buflens[offset];
846 struct obd_run_ctxt saved;
852 /* I don't really like this, but it is a sanity check on the client
855 if (lmm_size > mds->mds_max_mdsize) {
856 CERROR("Saving MD for inode %u of %d bytes > max %d\n",
857 inode->i_ino, lmm_size, mds->mds_max_mdsize);
861 CDEBUG(D_INODE, "storing %d bytes MD for inode %u\n",
862 lmm_size, inode->i_ino);
863 uc.ouc_fsuid = body->fsuid;
864 uc.ouc_fsgid = body->fsgid;
865 uc.ouc_cap = body->capability;
866 push_ctxt(&saved, &mds->mds_ctxt, &uc);
867 mds_start_transno(mds);
868 handle = mds_fs_start(mds, inode, MDS_FSOP_SETATTR);
869 if (IS_ERR(handle)) {
870 rc = PTR_ERR(handle);
871 mds_finish_transno(mds, handle, req, rc);
875 rc = mds_fs_set_md(mds, inode, handle, lmm, lmm_size);
876 rc = mds_finish_transno(mds, handle, req, rc);
878 rc2 = mds_fs_commit(mds, inode, handle);
882 pop_ctxt(&saved, &mds->mds_ctxt, &uc);
887 static int mds_open(struct ptlrpc_request *req)
889 struct mds_obd *mds = mds_req2mds(req);
890 struct mds_body *body;
891 struct mds_export_data *med;
892 struct mds_file_data *mfd;
895 struct vfsmount *mnt;
897 struct list_head *tmp;
898 int rc, size = sizeof(*body);
901 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK)) {
902 CERROR("test case OBD_FAIL_MDS_OPEN_PACK\n");
903 req->rq_status = -ENOMEM;
907 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
909 CERROR("mds: pack error: rc = %d\n", rc);
914 body = lustre_msg_buf(req->rq_reqmsg, 0);
916 /* was this animal open already and the client lost the reply? */
917 /* XXX need some way to detect a reopen, to avoid locked list walks */
918 med = &req->rq_export->exp_mds_data;
919 spin_lock(&med->med_open_lock);
920 list_for_each(tmp, &med->med_open_head) {
921 mfd = list_entry(tmp, typeof(*mfd), mfd_list);
922 if (!memcmp(&mfd->mfd_clienthandle, &body->handle,
923 sizeof(mfd->mfd_clienthandle)) &&
924 body->fid1.id == mfd->mfd_file->f_dentry->d_inode->i_ino) {
925 de = mfd->mfd_file->f_dentry;
926 spin_unlock(&med->med_open_lock);
927 CERROR("Re opening "LPD64"\n", body->fid1.id);
928 GOTO(out_pack, rc = 0);
931 spin_unlock(&med->med_open_lock);
933 mfd = kmem_cache_alloc(mds_file_cache, GFP_KERNEL);
935 CERROR("mds: out of memory\n");
936 req->rq_status = -ENOMEM;
940 de = mds_fid2dentry(mds, &body->fid1, &mnt);
942 GOTO(out_free, rc = PTR_ERR(de));
944 /* check if this inode has seen a delayed object creation */
945 if (lustre_msg_get_op_flags(req->rq_reqmsg) & MDS_OPEN_HAS_EA) {
946 rc = mds_store_md(mds, req, 1, body, de->d_inode);
955 /* dentry_open does a dput(de) and mntput(mnt) on error */
956 file = dentry_open(de, mnt, flags & ~O_DIRECT);
962 file->private_data = mfd;
963 mfd->mfd_file = file;
964 memcpy(&mfd->mfd_clienthandle, &body->handle, sizeof(body->handle));
965 get_random_bytes(&mfd->mfd_servercookie, sizeof(mfd->mfd_servercookie));
966 spin_lock(&med->med_open_lock);
967 list_add(&mfd->mfd_list, &med->med_open_head);
968 spin_unlock(&med->med_open_lock);
971 body = lustre_msg_buf(req->rq_repmsg, 0);
972 mds_pack_inode2fid(&body->fid1, de->d_inode);
973 mds_pack_inode2body(body, de->d_inode);
974 body->handle.addr = (__u64)(unsigned long)mfd;
975 body->handle.cookie = mfd->mfd_servercookie;
976 CDEBUG(D_INODE, "llite file "LPX64": addr %p, cookie "LPX64"\n",
977 mfd->mfd_clienthandle.addr, mfd, mfd->mfd_servercookie);
981 mfd->mfd_servercookie = DEAD_HANDLE_MAGIC;
982 kmem_cache_free(mds_file_cache, mfd);
987 static int mds_close(struct ptlrpc_request *req)
989 struct mds_export_data *med = &req->rq_export->exp_mds_data;
990 struct mds_body *body;
991 struct mds_file_data *mfd;
995 body = lustre_msg_buf(req->rq_reqmsg, 0);
997 mfd = mds_handle2mfd(&body->handle);
999 CERROR("no handle for file close "LPD64
1000 ": addr "LPX64", cookie "LPX64"\n",
1001 body->fid1.id, body->handle.addr, body->handle.cookie);
1005 spin_lock(&med->med_open_lock);
1006 req->rq_status = mds_close_mfd(mfd, med);
1007 spin_unlock(&med->med_open_lock);
1009 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK)) {
1010 CERROR("test case OBD_FAIL_MDS_CLOSE_PACK\n");
1011 req->rq_status = -ENOMEM;
1015 rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
1017 CERROR("mds: lustre_pack_msg: rc = %d\n", rc);
1018 req->rq_status = rc;
1024 static int mds_readpage(struct ptlrpc_request *req)
1026 struct mds_obd *mds = mds_req2mds(req);
1027 struct vfsmount *mnt;
1030 struct mds_body *body, *repbody;
1031 struct obd_run_ctxt saved;
1032 int rc, size = sizeof(*body);
1033 struct obd_ucred uc;
1036 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
1037 if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK)) {
1038 CERROR("mds: out of memory\n");
1039 GOTO(out, rc = -ENOMEM);
1042 body = lustre_msg_buf(req->rq_reqmsg, 0);
1043 uc.ouc_fsuid = body->fsuid;
1044 uc.ouc_fsgid = body->fsgid;
1045 uc.ouc_cap = body->capability;
1046 push_ctxt(&saved, &mds->mds_ctxt, &uc);
1047 de = mds_fid2dentry(mds, &body->fid1, &mnt);
1049 GOTO(out_pop, rc = PTR_ERR(de));
1051 CDEBUG(D_INODE, "ino %ld\n", de->d_inode->i_ino);
1053 file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
1054 /* note: in case of an error, dentry_open puts dentry */
1056 GOTO(out_pop, rc = PTR_ERR(file));
1058 repbody = lustre_msg_buf(req->rq_repmsg, 0);
1059 repbody->size = file->f_dentry->d_inode->i_size;
1060 repbody->valid = OBD_MD_FLSIZE;
1062 /* to make this asynchronous make sure that the handling function
1063 doesn't send a reply when this function completes. Instead a
1064 callback function would send the reply */
1065 /* note: in case of an error, dentry_open puts dentry */
1066 rc = mds_sendpage(req, file, body->size);
1068 filp_close(file, 0);
1070 pop_ctxt(&saved, &mds->mds_ctxt, &uc);
1072 req->rq_status = rc;
1076 int mds_reint(struct ptlrpc_request *req, int offset)
1079 struct mds_update_record rec;
1081 rc = mds_update_unpack(req, offset, &rec);
1082 if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK)) {
1083 CERROR("invalid record\n");
1084 req->rq_status = -EINVAL;
1087 /* rc will be used to interrupt a for loop over multiple records */
1088 rc = mds_reint_rec(&rec, offset, req);
1092 /* forward declaration */
1093 int mds_handle(struct ptlrpc_request *req);
1095 static int check_for_next_transno(struct mds_obd *mds)
1097 struct ptlrpc_request *req;
1098 req = list_entry(mds->mds_recovery_queue.next,
1099 struct ptlrpc_request, rq_list);
1100 return req->rq_reqmsg->transno == mds->mds_next_recovery_transno;
1103 static void process_recovery_queue(struct mds_obd *mds)
1105 struct ptlrpc_request *req;
1108 spin_lock(&mds->mds_processing_task_lock);
1109 req = list_entry(mds->mds_recovery_queue.next,
1110 struct ptlrpc_request, rq_list);
1112 if (req->rq_reqmsg->transno != mds->mds_next_recovery_transno) {
1113 spin_unlock(&mds->mds_processing_task_lock);
1114 wait_event(mds->mds_next_transno_waitq,
1115 check_for_next_transno(mds));
1118 list_del(&req->rq_list);
1119 spin_unlock(&mds->mds_processing_task_lock);
1121 DEBUG_REQ(D_HA, req, "");
1124 if (list_empty(&mds->mds_recovery_queue))
1129 static int queue_recovery_request(struct ptlrpc_request *req,
1130 struct mds_obd *mds)
1132 struct list_head *tmp;
1133 int inserted = 0, transno = req->rq_reqmsg->transno;
1136 DEBUG_REQ(D_HA, req, "not queueing");
1140 spin_lock(&mds->mds_processing_task_lock);
1142 if (mds->mds_processing_task == current->pid) {
1143 /* Processing the queue right now, don't re-add. */
1144 spin_unlock(&mds->mds_processing_task_lock);
1149 list_for_each(tmp, &mds->mds_recovery_queue) {
1150 struct ptlrpc_request *reqiter =
1151 list_entry(tmp, struct ptlrpc_request, rq_list);
1152 if (reqiter->rq_reqmsg->transno > transno) {
1153 list_add_tail(&req->rq_list, &reqiter->rq_list);
1160 list_add_tail(&req->rq_list, &mds->mds_recovery_queue);
1162 if (mds->mds_processing_task != 0) {
1163 /* Someone else is processing this queue, we'll leave it to
1166 spin_unlock(&mds->mds_processing_task_lock);
1167 if (transno == mds->mds_next_recovery_transno)
1168 wake_up(&mds->mds_next_transno_waitq);
1172 /* Nobody is processing, and we know there's (at least) one to process
1173 * now, so we'll do the honours.
1175 mds->mds_processing_task = current->pid;
1176 spin_unlock(&mds->mds_processing_task_lock);
1178 process_recovery_queue(mds);
1182 static int filter_recovery_request(struct ptlrpc_request *req,
1183 struct mds_obd *mds, int *process)
1185 switch (req->rq_reqmsg->opc) {
1187 case MDS_DISCONNECT:
1192 case MDS_GETSTATUS: /* used in unmounting */
1195 *process = queue_recovery_request(req, mds);
1199 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
1201 RETURN(ptlrpc_error(req->rq_svc, req));
1205 static int mds_queue_final_reply(struct ptlrpc_request *req, int rc)
1207 struct mds_obd *mds = mds_req2mds(req);
1210 /* Just like ptlrpc_error, but without the sending. */
1211 lustre_pack_msg(0, NULL, NULL, &req->rq_replen,
1213 req->rq_type = PTL_RPC_MSG_ERR;
1216 list_add(&req->rq_list, &mds->mds_delayed_reply_queue);
1217 if (--mds->mds_recoverable_clients == 0) {
1218 struct list_head *tmp, *n;
1221 "all clients recovered, sending delayed replies\n");
1222 list_for_each_safe(tmp, n, &mds->mds_delayed_reply_queue) {
1223 req = list_entry(tmp, struct ptlrpc_request, rq_list);
1224 DEBUG_REQ(D_HA, req, "delayed:");
1225 ptlrpc_reply(req->rq_svc, req);
1228 CDEBUG(D_HA, "%d recoverable clients remain\n",
1229 mds->mds_recoverable_clients);
1235 static char *reint_names[] = {
1236 [REINT_SETATTR] "setattr",
1237 [REINT_CREATE] "create",
1238 [REINT_LINK] "link",
1239 [REINT_UNLINK] "unlink",
1240 [REINT_RENAME] "rename"
1243 int mds_handle(struct ptlrpc_request *req)
1247 struct mds_obd *mds = NULL; /* quell gcc overwarning */
1250 rc = lustre_unpack_msg(req->rq_reqmsg, req->rq_reqlen);
1251 if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_HANDLE_UNPACK)) {
1252 CERROR("lustre_mds: Invalid request\n");
1256 LASSERT(!strcmp(req->rq_obd->obd_type->typ_name, LUSTRE_MDT_NAME));
1258 if (req->rq_reqmsg->opc != MDS_CONNECT) {
1259 if (req->rq_export == NULL)
1260 GOTO(out, rc = -ENOTCONN);
1262 mds = mds_req2mds(req);
1263 if (mds->mds_recoverable_clients != 0) {
1264 rc = filter_recovery_request(req, mds, &should_process);
1265 if (rc || !should_process)
1270 switch (req->rq_reqmsg->opc) {
1272 DEBUG_REQ(D_INODE, req, "connect");
1273 OBD_FAIL_RETURN(OBD_FAIL_MDS_CONNECT_NET, 0);
1274 rc = target_handle_connect(req);
1275 /* Make sure that last_rcvd is correct. */
1277 /* Now that we have an export, set mds. */
1278 mds = mds_req2mds(req);
1279 mds_fsync_super(mds->mds_sb);
1283 case MDS_DISCONNECT:
1284 DEBUG_REQ(D_INODE, req, "disconnect");
1285 OBD_FAIL_RETURN(OBD_FAIL_MDS_DISCONNECT_NET, 0);
1286 rc = target_handle_disconnect(req);
1287 /* Make sure that last_rcvd is correct. */
1289 mds_fsync_super(mds->mds_sb);
1293 DEBUG_REQ(D_INODE, req, "getstatus");
1294 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETSTATUS_NET, 0);
1295 rc = mds_getstatus(req);
1298 case MDS_GETLOVINFO:
1299 DEBUG_REQ(D_INODE, req, "getlovinfo");
1300 rc = mds_getlovinfo(req);
1304 DEBUG_REQ(D_INODE, req, "getattr");
1305 OBD_FAIL_RETURN(OBD_FAIL_MDS_GETATTR_NET, 0);
1306 rc = mds_getattr(0, req);
1310 DEBUG_REQ(D_INODE, req, "statfs");
1311 OBD_FAIL_RETURN(OBD_FAIL_MDS_STATFS_NET, 0);
1312 rc = mds_statfs(req);
1316 DEBUG_REQ(D_INODE, req, "readpage\n");
1317 OBD_FAIL_RETURN(OBD_FAIL_MDS_READPAGE_NET, 0);
1318 rc = mds_readpage(req);
1320 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1325 int size = sizeof(struct mds_body);
1326 int opc = *(u32 *)lustre_msg_buf(req->rq_reqmsg, 0),
1327 realopc = opc & REINT_OPCODE_MASK;
1329 DEBUG_REQ(D_INODE, req, "reint (%s%s)",
1330 reint_names[realopc],
1331 opc & REINT_REPLAYING ? "|REPLAYING" : "");
1333 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET, 0);
1335 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
1338 req->rq_status = rc;
1341 rc = mds_reint(req, 0);
1342 OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET_REP, 0);
1347 DEBUG_REQ(D_INODE, req, "open");
1348 OBD_FAIL_RETURN(OBD_FAIL_MDS_OPEN_NET, 0);
1353 DEBUG_REQ(D_INODE, req, "close");
1354 OBD_FAIL_RETURN(OBD_FAIL_MDS_CLOSE_NET, 0);
1355 rc = mds_close(req);
1359 DEBUG_REQ(D_INODE, req, "enqueue");
1360 OBD_FAIL_RETURN(OBD_FAIL_LDLM_ENQUEUE, 0);
1361 rc = ldlm_handle_enqueue(req);
1364 DEBUG_REQ(D_INODE, req, "convert");
1365 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CONVERT, 0);
1366 rc = ldlm_handle_convert(req);
1368 case LDLM_BL_CALLBACK:
1369 case LDLM_CP_CALLBACK:
1370 DEBUG_REQ(D_INODE, req, "callback");
1371 CERROR("callbacks should not happen on MDS\n");
1373 OBD_FAIL_RETURN(OBD_FAIL_LDLM_BL_CALLBACK, 0);
1376 rc = ptlrpc_error(req->rq_svc, req);
1383 struct mds_export_data *med = &req->rq_export->exp_mds_data;
1385 req->rq_repmsg->last_xid =
1386 HTON__u64(le64_to_cpu(med->med_mcd->mcd_last_xid));
1387 req->rq_repmsg->last_committed =
1388 HTON__u64(mds->mds_last_committed);
1389 CDEBUG(D_INFO, "last_rcvd ~%Lu, last_committed %Lu, xid %d\n",
1390 (unsigned long long)mds->mds_last_rcvd,
1391 (unsigned long long)mds->mds_last_committed,
1392 cpu_to_le32(req->rq_xid));
1396 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LAST_REPLAY) {
1397 struct mds_obd *mds = mds_req2mds(req);
1398 LASSERT(mds->mds_recoverable_clients);
1399 DEBUG_REQ(D_HA, req, "LAST_REPLAY, queuing reply");
1400 return mds_queue_final_reply(req, rc);
1403 /* MDS_CONNECT / EALREADY (note: not -EALREADY!) isn't an error */
1404 if (rc && (req->rq_reqmsg->opc != MDS_CONNECT ||
1406 CERROR("mds: processing error (opcode %d): %d\n",
1407 req->rq_reqmsg->opc, rc);
1408 ptlrpc_error(req->rq_svc, req);
1410 CDEBUG(D_NET, "sending reply\n");
1411 ptlrpc_reply(req->rq_svc, req);
1416 /* Update the server data on disk. This stores the new mount_count and
1417 * also the last_rcvd value to disk. If we don't have a clean shutdown,
1418 * then the server last_rcvd value may be less than that of the clients.
1419 * This will alert us that we may need to do client recovery.
1421 * Assumes we are already in the server filesystem context.
1423 * Also assumes for mds_last_rcvd that we are not modifying it (no locking).
1425 int mds_update_server_data(struct mds_obd *mds)
1427 struct mds_server_data *msd = mds->mds_server_data;
1428 struct file *filp = mds->mds_rcvd_filp;
1432 msd->msd_last_rcvd = cpu_to_le64(mds->mds_last_rcvd);
1433 msd->msd_mount_count = cpu_to_le64(mds->mds_mount_count);
1435 CDEBUG(D_SUPER, "MDS mount_count is %Lu, last_rcvd is %Lu\n",
1436 (unsigned long long)mds->mds_mount_count,
1437 (unsigned long long)mds->mds_last_rcvd);
1438 rc = lustre_fwrite(filp, (char *)msd, sizeof(*msd), &off);
1439 if (rc != sizeof(*msd)) {
1440 CERROR("error writing MDS server data: rc = %d\n", rc);
1445 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1446 rc = fsync_dev(filp->f_dentry->d_inode->i_rdev);
1448 rc = file_fsync(filp, filp->f_dentry, 1);
1451 CERROR("error flushing MDS server data: rc = %d\n", rc);
1456 /* Do recovery actions for the MDS */
1457 static int mds_recovery_complete(struct obd_device *obddev)
1459 struct mds_obd *mds = &obddev->u.mds;
1460 struct obd_run_ctxt saved;
1463 LASSERT(mds->mds_recoverable_clients == 0);
1465 /* This happens at the end when recovery is complete */
1466 ++mds->mds_mount_count;
1467 push_ctxt(&saved, &mds->mds_ctxt, NULL);
1468 rc = mds_update_server_data(mds);
1469 pop_ctxt(&saved, &mds->mds_ctxt, NULL);
1474 /* mount the file system (secretly) */
1475 static int mds_setup(struct obd_device *obddev, obd_count len, void *buf)
1477 struct obd_ioctl_data* data = buf;
1478 struct mds_obd *mds = &obddev->u.mds;
1479 struct vfsmount *mnt;
1484 #ifdef CONFIG_DEV_RDONLY
1485 dev_clear_rdonly(2);
1487 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2)
1488 GOTO(err_dec, rc = -EINVAL);
1490 mds->mds_fstype = strdup(data->ioc_inlbuf2);
1492 mnt = do_kern_mount(mds->mds_fstype, 0, data->ioc_inlbuf1, NULL);
1495 CERROR("do_kern_mount failed: rc = %d\n", rc);
1496 GOTO(err_kfree, rc);
1499 CERROR("%s: mnt is %p\n", data->ioc_inlbuf1, mnt);
1500 mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
1502 GOTO(err_put, rc = -ENODEV);
1504 init_MUTEX(&mds->mds_transno_sem);
1505 mds->mds_max_mdsize = sizeof(struct lov_mds_md);
1506 rc = mds_fs_setup(obddev, mnt);
1508 CERROR("MDS filesystem method init failed: rc = %d\n", rc);
1512 obddev->obd_namespace =
1513 ldlm_namespace_new("mds_server", LDLM_NAMESPACE_SERVER);
1514 if (obddev->obd_namespace == NULL) {
1515 mds_cleanup(obddev);
1516 GOTO(err_fs, rc = -ENOMEM);
1519 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
1520 "mds_ldlm_client", &obddev->obd_ldlm_client);
1522 spin_lock_init(&mds->mds_processing_task_lock);
1523 mds->mds_processing_task = 0;
1524 INIT_LIST_HEAD(&mds->mds_recovery_queue);
1525 INIT_LIST_HEAD(&mds->mds_delayed_reply_queue);
1530 mds_fs_cleanup(obddev);
1533 mntput(mds->mds_vfsmnt);
1537 kfree(mds->mds_fstype);
1543 static int mds_cleanup(struct obd_device *obddev)
1545 struct super_block *sb;
1546 struct mds_obd *mds = &obddev->u.mds;
1547 struct obd_run_ctxt saved;
1554 push_ctxt(&saved, &mds->mds_ctxt, NULL);
1555 mds_update_server_data(mds);
1557 if (mds->mds_rcvd_filp) {
1558 int rc = filp_close(mds->mds_rcvd_filp, 0);
1559 mds->mds_rcvd_filp = NULL;
1562 CERROR("last_rcvd file won't close, rc=%d\n", rc);
1564 pop_ctxt(&saved, &mds->mds_ctxt, NULL);
1567 mntput(mds->mds_vfsmnt);
1569 kfree(mds->mds_fstype);
1571 ldlm_namespace_free(obddev->obd_namespace);
1574 #ifdef CONFIG_DEV_RDONLY
1575 dev_clear_rdonly(2);
1577 mds_fs_cleanup(obddev);
1583 static int ldlm_intent_policy(struct ldlm_lock *lock, void *req_cookie,
1584 ldlm_mode_t mode, int flags, void *data)
1586 struct ptlrpc_request *req = req_cookie;
1593 if (req->rq_reqmsg->bufcount > 1) {
1594 /* an intent needs to be considered */
1595 struct ldlm_intent *it = lustre_msg_buf(req->rq_reqmsg, 1);
1596 struct mds_obd *mds= &req->rq_export->exp_obd->u.mds;
1597 struct mds_body *mds_rep;
1598 struct ldlm_reply *rep;
1599 __u64 new_resid[3] = {0, 0, 0}, old_res;
1600 int rc, size[3] = {sizeof(struct ldlm_reply),
1601 sizeof(struct mds_body),
1602 mds->mds_max_mdsize};
1604 it->opc = NTOH__u64(it->opc);
1606 LDLM_DEBUG(lock, "intent policy, opc: %s",
1607 ldlm_it2str(it->opc));
1609 rc = lustre_pack_msg(3, size, NULL, &req->rq_replen,
1612 rc = req->rq_status = -ENOMEM;
1616 rep = lustre_msg_buf(req->rq_repmsg, 0);
1617 rep->lock_policy_res1 = 1;
1619 /* execute policy */
1620 switch ((long)it->opc) {
1621 case IT_CREAT|IT_OPEN:
1622 rc = mds_reint(req, 2);
1623 if (rc || (req->rq_status != 0 &&
1624 req->rq_status != -EEXIST)) {
1625 rep->lock_policy_res2 = req->rq_status;
1626 RETURN(ELDLM_LOCK_ABORTED);
1637 rc = mds_reint(req, 2);
1638 if (rc || (req->rq_status != 0 &&
1639 req->rq_status != -EISDIR &&
1640 req->rq_status != -ENOTDIR)) {
1641 rep->lock_policy_res2 = req->rq_status;
1642 RETURN(ELDLM_LOCK_ABORTED);
1653 rc = mds_getattr_name(2, req);
1654 /* FIXME: we need to sit down and decide on who should
1655 * set req->rq_status, who should return negative and
1656 * positive return values, and what they all mean. */
1657 if (rc || req->rq_status != 0) {
1658 rep->lock_policy_res2 = req->rq_status;
1659 RETURN(ELDLM_LOCK_ABORTED);
1662 case IT_READDIR|IT_OPEN:
1666 CERROR("Unhandled intent "LPD64"\n", it->opc);
1670 /* We don't bother returning a lock to the client for a file
1671 * or directory we are removing.
1673 * As for link and rename, there is no reason for the client
1674 * to get a lock on the target at this point. If they are
1675 * going to modify the file/directory later they will get a
1676 * lock at that time.
1678 if (it->opc & (IT_UNLINK | IT_RMDIR | IT_LINK | IT_LINK2 |
1679 IT_RENAME | IT_RENAME2))
1680 RETURN(ELDLM_LOCK_ABORTED);
1682 rep->lock_policy_res2 = req->rq_status;
1683 mds_rep = lustre_msg_buf(req->rq_repmsg, 1);
1685 /* If the client is about to open a file that doesn't have an MD
1686 * stripe record, it's going to need a write lock. */
1687 if (it->opc & IT_OPEN && !(mds_rep->valid & OBD_MD_FLEASIZE)) {
1688 LDLM_DEBUG(lock, "open with no EA; returning PW lock");
1689 lock->l_req_mode = LCK_PW;
1692 if (flags & LDLM_FL_INTENT_ONLY) {
1693 LDLM_DEBUG(lock, "INTENT_ONLY, aborting lock");
1694 RETURN(ELDLM_LOCK_ABORTED);
1696 /* Give the client a lock on the child object, instead of the
1697 * parent that it requested. */
1698 new_resid[0] = NTOH__u32(mds_rep->ino);
1699 new_resid[1] = NTOH__u32(mds_rep->generation);
1700 if (new_resid[0] == 0)
1702 old_res = lock->l_resource->lr_name[0];
1704 ldlm_lock_change_resource(lock, new_resid);
1705 if (lock->l_resource == NULL) {
1709 LDLM_DEBUG(lock, "intent policy, old res %ld",
1711 RETURN(ELDLM_LOCK_CHANGED);
1713 int size = sizeof(struct ldlm_reply);
1714 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
1724 int mds_attach(struct obd_device *dev, obd_count len, void *data)
1726 return lprocfs_reg_obd(dev, status_var_nm_1, dev);
1729 int mds_detach(struct obd_device *dev)
1731 return lprocfs_dereg_obd(dev);
1734 static int mdt_setup(struct obd_device *obddev, obd_count len, void *buf)
1737 // struct obd_ioctl_data* data = buf;
1738 struct mds_obd *mds = &obddev->u.mds;
1744 mds->mds_service = ptlrpc_init_svc(MDS_NEVENTS, MDS_NBUFS,
1745 MDS_BUFSIZE, MDS_MAXREQSIZE,
1746 MDS_REQUEST_PORTAL, MDC_REPLY_PORTAL,
1747 "self", mds_handle, "mds");
1748 if (!mds->mds_service) {
1749 CERROR("failed to start service\n");
1750 GOTO(err_dec, rc = -ENOMEM);
1753 for (i = 0; i < MDT_NUM_THREADS; i++) {
1755 sprintf(name, "ll_mdt_%02d", i);
1756 rc = ptlrpc_start_thread(obddev, mds->mds_service, name);
1758 CERROR("cannot start MDT thread #%d: rc %d\n", i, rc);
1759 GOTO(err_thread, rc);
1766 ptlrpc_stop_all_threads(mds->mds_service);
1767 ptlrpc_unregister_service(mds->mds_service);
1774 static int mdt_cleanup(struct obd_device *obddev)
1776 struct mds_obd *mds = &obddev->u.mds;
1779 ptlrpc_stop_all_threads(mds->mds_service);
1780 ptlrpc_unregister_service(mds->mds_service);
1786 extern int mds_iocontrol(unsigned int cmd, struct lustre_handle *conn,
1787 int len, void *karg, void *uarg);
1789 /* use obd ops to offer management infrastructure */
1790 static struct obd_ops mds_obd_ops = {
1791 o_attach: mds_attach,
1792 o_detach: mds_detach,
1793 o_connect: mds_connect,
1794 o_disconnect: mds_disconnect,
1796 o_cleanup: mds_cleanup,
1797 o_iocontrol: mds_iocontrol
1800 static struct obd_ops mdt_obd_ops = {
1802 o_cleanup: mdt_cleanup,
1806 static int __init mds_init(void)
1809 mds_file_cache = kmem_cache_create("ll_mds_file_data",
1810 sizeof(struct mds_file_data),
1812 if (mds_file_cache == NULL)
1815 class_register_type(&mds_obd_ops, status_class_var, LUSTRE_MDS_NAME);
1816 class_register_type(&mdt_obd_ops, 0, LUSTRE_MDT_NAME);
1817 ldlm_register_intent(ldlm_intent_policy);
1822 static void __exit mds_exit(void)
1826 ldlm_unregister_intent();
1827 class_unregister_type(LUSTRE_MDS_NAME);
1828 class_unregister_type(LUSTRE_MDT_NAME);
1829 if (kmem_cache_destroy(mds_file_cache))
1830 CERROR("couldn't free MDS file cache\n");
1834 MODULE_AUTHOR("Cluster File Systems <info@clusterfs.com>");
1835 MODULE_DESCRIPTION("Lustre Metadata Server (MDS) v0.01");
1836 MODULE_LICENSE("GPL");
1838 module_init(mds_init);
1839 module_exit(mds_exit);