4 * Lustre Metadata Server (mds) request handler
6 * Copyright (C) 2001, 2002 Cluster File Systems, Inc.
8 * This code is issued under the GNU General Public License.
9 * See the file COPYING in this distribution
11 * by Peter Braam <braam@clusterfs.com>
13 * This server is single threaded at present (but can easily be multi threaded).
20 #include <linux/version.h>
21 #include <linux/module.h>
23 #include <linux/stat.h>
24 #include <linux/locks.h>
25 #include <linux/ext2_fs.h>
26 #include <linux/quotaops.h>
27 #include <asm/unistd.h>
28 #include <asm/uaccess.h>
29 #include <linux/obd_support.h>
30 #include <linux/obd.h>
31 #include <linux/lustre_lib.h>
32 #include <linux/lustre_idl.h>
33 #include <linux/lustre_mds.h>
34 #include <linux/lustre_net.h>
35 #include <linux/obd_class.h>
38 static struct mds_obd *MDS;
40 // XXX make this networked!
41 static int mds_queue_req(struct ptlrpc_request *req)
43 struct ptlrpc_request *srv_req;
50 srv_req = kmalloc(sizeof(*srv_req), GFP_KERNEL);
56 printk("---> MDS at %d %p, incoming req %p, srv_req %p\n",
57 __LINE__, MDS, req, srv_req);
59 memset(srv_req, 0, sizeof(*req));
61 /* move the request buffer */
62 srv_req->rq_reqbuf = req->rq_reqbuf;
63 srv_req->rq_reqlen = req->rq_reqlen;
64 srv_req->rq_obd = MDS;
66 /* remember where it came from */
67 srv_req->rq_reply_handle = req;
69 list_add(&srv_req->rq_list, &MDS->mds_reqs);
70 wake_up(&MDS->mds_waitq);
74 /* XXX do this over the net */
75 int mds_sendpage(struct ptlrpc_request *req, struct file *file,
76 __u64 offset, struct niobuf *dst)
79 mm_segment_t oldfs = get_fs();
80 /* dst->addr is a user address, but in a different task! */
82 rc = generic_file_read(file, (char *)(long)dst->addr,
91 /* XXX replace with networking code */
92 int mds_reply(struct ptlrpc_request *req)
94 struct ptlrpc_request *clnt_req = req->rq_reply_handle;
98 if (req->rq_obd->mds_service != NULL) {
99 /* This is a request that came from the network via portals. */
101 /* FIXME: we need to increment the count of handled events */
102 ptl_send_buf(req, &req->rq_peer, MDS_REPLY_PORTAL, 0);
104 /* This is a local request that came from another thread. */
106 /* move the reply to the client */
107 clnt_req->rq_replen = req->rq_replen;
108 clnt_req->rq_repbuf = req->rq_repbuf;
109 req->rq_repbuf = NULL;
112 /* free the request buffer */
113 kfree(req->rq_reqbuf);
114 req->rq_reqbuf = NULL;
116 /* wake up the client */
117 wake_up_interruptible(&clnt_req->rq_wait_for_rep);
124 int mds_error(struct ptlrpc_request *req)
126 struct ptlrep_hdr *hdr;
130 hdr = kmalloc(sizeof(*hdr), GFP_KERNEL);
136 memset(hdr, 0, sizeof(*hdr));
138 hdr->seqno = req->rq_reqhdr->seqno;
139 hdr->status = req->rq_status;
140 hdr->type = MDS_TYPE_ERR;
142 req->rq_repbuf = (char *)hdr;
143 req->rq_replen = sizeof(*hdr);
146 return mds_reply(req);
149 struct dentry *mds_fid2dentry(struct mds_obd *mds, struct ll_fid *fid, struct vfsmount **mnt)
151 /* stolen from NFS */
152 struct super_block *sb = mds->mds_sb;
153 unsigned long ino = fid->id;
154 //__u32 generation = fid->generation;
155 __u32 generation = 0;
157 struct list_head *lp;
158 struct dentry *result;
161 *mnt = mntget(mds->mds_vfsmnt);
165 return ERR_PTR(-ESTALE);
167 inode = iget(sb, ino);
169 return ERR_PTR(-ENOMEM);
171 printk("--> mds_fid2dentry: sb %p\n", inode->i_sb);
173 if (is_bad_inode(inode)
174 || (generation && inode->i_generation != generation)
176 /* we didn't find the right inode.. */
178 "bad inode %lu, link: %d ct: %d or version %u/%u\n",
180 inode->i_nlink, atomic_read(&inode->i_count),
184 return ERR_PTR(-ESTALE);
187 /* now to find a dentry.
188 * If possible, get a well-connected one
190 spin_lock(&dcache_lock);
191 for (lp = inode->i_dentry.next; lp != &inode->i_dentry ; lp=lp->next) {
192 result = list_entry(lp,struct dentry, d_alias);
193 if (! (result->d_flags & DCACHE_NFSD_DISCONNECTED)) {
195 result->d_vfs_flags |= DCACHE_REFERENCED;
196 spin_unlock(&dcache_lock);
201 spin_unlock(&dcache_lock);
202 result = d_alloc_root(inode);
203 if (result == NULL) {
205 return ERR_PTR(-ENOMEM);
207 result->d_flags |= DCACHE_NFSD_DISCONNECTED;
211 static inline void mds_get_objid(struct inode *inode, __u64 *id)
213 memcpy(id, &inode->u.ext2_i.i_data, sizeof(*id));
216 int mds_getattr(struct ptlrpc_request *req)
218 struct dentry *de = mds_fid2dentry(req->rq_obd, &req->rq_req.mds->fid1,
224 rc = mds_pack_rep(NULL, 0, NULL, 0, &req->rq_rephdr, &req->rq_rep.mds,
225 &req->rq_replen, &req->rq_repbuf);
228 printk("mds: out of memory\n");
229 req->rq_status = -ENOMEM;
233 req->rq_rephdr->seqno = req->rq_reqhdr->seqno;
234 rep = req->rq_rep.mds;
238 req->rq_rephdr->status = -ENOENT;
243 rep->ino = inode->i_ino;
244 rep->atime = inode->i_atime;
245 rep->ctime = inode->i_ctime;
246 rep->mtime = inode->i_mtime;
247 rep->uid = inode->i_uid;
248 rep->gid = inode->i_gid;
249 rep->size = inode->i_size;
250 rep->mode = inode->i_mode;
251 rep->nlink = inode->i_nlink;
253 mds_get_objid(inode, &rep->objid);
258 int mds_readpage(struct ptlrpc_request *req)
260 struct vfsmount *mnt;
261 struct dentry *de = mds_fid2dentry(req->rq_obd, &req->rq_req.mds->fid1,
264 struct niobuf *niobuf;
268 printk("mds_readpage: ino %ld\n", de->d_inode->i_ino);
269 rc = mds_pack_rep(NULL, 0, NULL, 0, &req->rq_rephdr, &req->rq_rep.mds,
270 &req->rq_replen, &req->rq_repbuf);
273 printk("mds: out of memory\n");
274 req->rq_status = -ENOMEM;
278 req->rq_rephdr->seqno = req->rq_reqhdr->seqno;
279 rep = req->rq_rep.mds;
283 req->rq_rephdr->status = PTR_ERR(de);
287 file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
288 /* note: in case of an error, dentry_open puts dentry */
291 req->rq_rephdr->status = PTR_ERR(file);
295 niobuf = mds_req_tgt(req->rq_req.mds);
297 /* to make this asynchronous make sure that the handling function
298 doesn't send a reply when this function completes. Instead a
299 callback function would send the reply */
300 rc = mds_sendpage(req, file, req->rq_req.mds->size, niobuf);
303 req->rq_rephdr->status = rc;
308 int mds_reint(struct ptlrpc_request *req)
311 char *buf = mds_req_tgt(req->rq_req.mds);
312 int len = req->rq_req.mds->tgtlen;
313 struct mds_update_record rec;
315 rc = mds_update_unpack(buf, len, &rec);
317 printk(__FUNCTION__ ": invalid record\n");
318 req->rq_status = -EINVAL;
321 /* rc will be used to interrupt a for loop over multiple records */
322 rc = mds_reint_rec(&rec, req);
326 //int mds_handle(struct mds_conn *conn, int len, char *buf)
327 int mds_handle(struct ptlrpc_request *req)
330 struct ptlreq_hdr *hdr;
334 hdr = (struct ptlreq_hdr *)req->rq_reqbuf;
336 if (NTOH__u32(hdr->type) != MDS_TYPE_REQ) {
337 printk("lustre_mds: wrong packet type sent %d\n",
338 NTOH__u32(hdr->type));
343 rc = mds_unpack_req(req->rq_reqbuf, req->rq_reqlen,
344 &req->rq_reqhdr, &req->rq_req.mds);
346 printk("lustre_mds: Invalid request\n");
351 switch (req->rq_reqhdr->opc) {
354 CDEBUG(D_INODE, "getattr\n");
355 rc = mds_getattr(req);
359 CDEBUG(D_INODE, "readpage\n");
360 rc = mds_readpage(req);
364 CDEBUG(D_INODE, "reint\n");
369 return mds_error(req);
374 printk(__FUNCTION__ ": no header\n");
378 if( req->rq_status) {
381 CDEBUG(D_INODE, "sending reply\n");
389 static void mds_timer_run(unsigned long __data)
391 struct task_struct * p = (struct task_struct *) __data;
396 int mds_main(void *arg)
398 struct mds_obd *mds = (struct mds_obd *) arg;
399 struct timer_list timer;
403 spin_lock_irq(¤t->sigmask_lock);
404 sigfillset(¤t->blocked);
405 recalc_sigpending(current);
406 spin_unlock_irq(¤t->sigmask_lock);
408 sprintf(current->comm, "lustre_mds");
410 /* Set up an interval timer which can be used to trigger a
411 wakeup after the interval expires */
413 timer.data = (unsigned long) current;
414 timer.function = mds_timer_run;
415 mds->mds_timer = &timer;
417 /* Record that the thread is running */
418 mds->mds_thread = current;
419 wake_up(&mds->mds_done_waitq);
421 printk(KERN_INFO "lustre_mds starting. Commit interval %d seconds\n",
422 mds->mds_interval / HZ);
424 /* XXX maintain a list of all managed devices: insert here */
426 /* And now, wait forever for commit wakeup events. */
430 if (mds->mds_flags & MDS_UNMOUNT)
433 wake_up(&mds->mds_done_waitq);
434 interruptible_sleep_on(&mds->mds_waitq);
436 CDEBUG(D_INODE, "lustre_mds wakes\n");
437 CDEBUG(D_INODE, "pick up req here and continue\n");
439 if (mds->mds_service != NULL) {
443 struct ptlrpc_request request;
445 rc = PtlEQGet(mds->mds_service->srv_eq, &ev);
446 if (rc != PTL_OK && rc != PTL_EQ_DROPPED)
448 /* FIXME: If we move to an event-driven model,
449 * we should put the request on the stack of
450 * mds_handle instead. */
451 memset(&request, 0, sizeof(request));
452 request.rq_reqbuf = ev.mem_desc.start +
454 request.rq_reqlen = ev.mem_desc.length;
455 request.rq_obd = MDS;
456 request.rq_xid = ev.match_bits;
458 request.rq_peer.peer_nid = ev.initiator.nid;
459 /* FIXME: this NI should be the incoming NI.
460 * We don't know how to find that from here. */
461 request.rq_peer.peer_ni =
462 mds->mds_service->srv_self.peer_ni;
463 rc = mds_handle(&request);
466 struct ptlrpc_request *request;
468 if (list_empty(&mds->mds_reqs)) {
469 CDEBUG(D_INODE, "woke because of timer\n");
471 request = list_entry(mds->mds_reqs.next,
472 struct ptlrpc_request,
474 list_del(&request->rq_list);
475 rc = mds_handle(request);
480 del_timer_sync(mds->mds_timer);
482 /* XXX maintain a list of all managed devices: cleanup here */
484 mds->mds_thread = NULL;
485 wake_up(&mds->mds_done_waitq);
486 printk("lustre_mds: exiting\n");
490 static void mds_stop_srv_thread(struct mds_obd *mds)
492 mds->mds_flags |= MDS_UNMOUNT;
494 while (mds->mds_thread) {
495 wake_up(&mds->mds_waitq);
496 sleep_on(&mds->mds_done_waitq);
500 static void mds_start_srv_thread(struct mds_obd *mds)
502 init_waitqueue_head(&mds->mds_waitq);
503 init_waitqueue_head(&mds->mds_done_waitq);
504 kernel_thread(mds_main, (void *)mds,
505 CLONE_VM | CLONE_FS | CLONE_FILES);
506 while (!mds->mds_thread)
507 sleep_on(&mds->mds_done_waitq);
510 /* mount the file system (secretly) */
511 static int mds_setup(struct obd_device *obddev, obd_count len,
515 struct obd_ioctl_data* data = buf;
516 struct mds_obd *mds = &obddev->u.mds;
517 struct vfsmount *mnt;
518 struct lustre_peer peer;
523 mnt = do_kern_mount(data->ioc_inlbuf2, 0, data->ioc_inlbuf1, NULL);
530 mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
531 if (!obddev->u.mds.mds_sb) {
536 mds->mds_vfsmnt = mnt;
537 obddev->u.mds.mds_fstype = strdup(data->ioc_inlbuf2);
539 mds->mds_ctxt.pwdmnt = mnt;
540 mds->mds_ctxt.pwd = mnt->mnt_root;
541 mds->mds_ctxt.fs = KERNEL_DS;
542 mds->mds_remote_nid = 0;
544 INIT_LIST_HEAD(&mds->mds_reqs);
545 mds->mds_thread = NULL;
547 mds->mds_interval = 3 * HZ;
550 spin_lock_init(&obddev->u.mds.mds_lock);
552 err = kportal_uuid_to_peer("self", &peer);
554 mds->mds_service = kmalloc(sizeof(*mds->mds_service),
556 if (mds->mds_service == NULL)
558 mds->mds_service->srv_buf_size = 64 * 1024;
559 mds->mds_service->srv_portal = MDS_REQUEST_PORTAL;
560 memcpy(&mds->mds_service->srv_self, &peer, sizeof(peer));
561 mds->mds_service->srv_wait_queue = &mds->mds_waitq;
563 rpc_register_service(mds->mds_service, "self");
566 mds_start_srv_thread(mds);
573 static int mds_cleanup(struct obd_device * obddev)
575 struct super_block *sb;
576 struct mds_obd *mds = &obddev->u.mds;
580 if ( !(obddev->obd_flags & OBD_SET_UP) ) {
585 if ( !list_empty(&obddev->obd_gen_clients) ) {
586 printk(KERN_WARNING __FUNCTION__ ": still has clients!\n");
592 mds_stop_srv_thread(mds);
599 if (!list_empty(&mds->mds_reqs)) {
600 // XXX reply with errors and clean up
601 CDEBUG(D_INODE, "Request list not empty!\n");
605 mntput(mds->mds_vfsmnt);
607 kfree(mds->mds_fstype);
616 /* use obd ops to offer management infrastructure */
617 static struct obd_ops mds_obd_ops = {
619 o_cleanup: mds_cleanup,
622 static int __init mds_init(void)
624 obd_register_type(&mds_obd_ops, LUSTRE_MDS_NAME);
628 static void __exit mds_exit(void)
630 obd_unregister_type(LUSTRE_MDS_NAME);
633 MODULE_AUTHOR("Peter J. Braam <braam@clusterfs.com>");
634 MODULE_DESCRIPTION("Lustre Metadata Server (MDS) v0.01");
635 MODULE_LICENSE("GPL");
638 // for testing (maybe this stays)
639 EXPORT_SYMBOL(mds_queue_req);
641 module_init(mds_init);
642 module_exit(mds_exit);