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 /* move the reply to the client */
99 clnt_req->rq_replen = req->rq_replen;
100 clnt_req->rq_repbuf = req->rq_repbuf;
101 req->rq_repbuf = NULL;
104 /* free the request buffer */
105 kfree(req->rq_reqbuf);
106 req->rq_reqbuf = NULL;
108 /* wake up the client */
109 wake_up_interruptible(&clnt_req->rq_wait_for_rep);
114 int mds_error(struct ptlrpc_request *req)
116 struct ptlrep_hdr *hdr;
120 hdr = kmalloc(sizeof(*hdr), GFP_KERNEL);
126 memset(hdr, 0, sizeof(*hdr));
128 hdr->seqno = req->rq_reqhdr->seqno;
129 hdr->status = req->rq_status;
130 hdr->type = MDS_TYPE_ERR;
132 req->rq_repbuf = (char *)hdr;
133 req->rq_replen = sizeof(*hdr);
136 return mds_reply(req);
139 struct dentry *mds_fid2dentry(struct mds_obd *mds, struct ll_fid *fid, struct vfsmount **mnt)
141 /* stolen from NFS */
142 struct super_block *sb = mds->mds_sb;
143 unsigned long ino = fid->id;
144 //__u32 generation = fid->generation;
145 __u32 generation = 0;
147 struct list_head *lp;
148 struct dentry *result;
151 *mnt = mntget(mds->mds_vfsmnt);
155 return ERR_PTR(-ESTALE);
157 inode = iget(sb, ino);
159 return ERR_PTR(-ENOMEM);
161 printk("--> mds_fid2dentry: sb %p\n", inode->i_sb);
163 if (is_bad_inode(inode)
164 || (generation && inode->i_generation != generation)
166 /* we didn't find the right inode.. */
168 "bad inode %lu, link: %d ct: %d or version %u/%u\n",
170 inode->i_nlink, atomic_read(&inode->i_count),
174 return ERR_PTR(-ESTALE);
177 /* now to find a dentry.
178 * If possible, get a well-connected one
180 spin_lock(&dcache_lock);
181 for (lp = inode->i_dentry.next; lp != &inode->i_dentry ; lp=lp->next) {
182 result = list_entry(lp,struct dentry, d_alias);
183 if (! (result->d_flags & DCACHE_NFSD_DISCONNECTED)) {
185 result->d_vfs_flags |= DCACHE_REFERENCED;
186 spin_unlock(&dcache_lock);
191 spin_unlock(&dcache_lock);
192 result = d_alloc_root(inode);
193 if (result == NULL) {
195 return ERR_PTR(-ENOMEM);
197 result->d_flags |= DCACHE_NFSD_DISCONNECTED;
201 static inline void mds_get_objid(struct inode *inode, __u64 *id)
203 memcpy(id, &inode->u.ext2_i.i_data, sizeof(*id));
206 int mds_getattr(struct ptlrpc_request *req)
208 struct dentry *de = mds_fid2dentry(req->rq_obd, &req->rq_req.mds->fid1,
214 rc = mds_pack_rep(NULL, 0, NULL, 0, &req->rq_rephdr, &req->rq_rep.mds,
215 &req->rq_replen, &req->rq_repbuf);
218 printk("mds: out of memory\n");
219 req->rq_status = -ENOMEM;
223 req->rq_rephdr->seqno = req->rq_reqhdr->seqno;
224 rep = req->rq_rep.mds;
228 req->rq_rephdr->status = -ENOENT;
233 rep->ino = inode->i_ino;
234 rep->atime = inode->i_atime;
235 rep->ctime = inode->i_ctime;
236 rep->mtime = inode->i_mtime;
237 rep->uid = inode->i_uid;
238 rep->gid = inode->i_gid;
239 rep->size = inode->i_size;
240 rep->mode = inode->i_mode;
241 rep->nlink = inode->i_nlink;
243 mds_get_objid(inode, &rep->objid);
248 int mds_readpage(struct ptlrpc_request *req)
250 struct vfsmount *mnt;
251 struct dentry *de = mds_fid2dentry(req->rq_obd, &req->rq_req.mds->fid1,
254 struct niobuf *niobuf;
258 printk("mds_readpage: ino %ld\n", de->d_inode->i_ino);
259 rc = mds_pack_rep(NULL, 0, NULL, 0, &req->rq_rephdr, &req->rq_rep.mds,
260 &req->rq_replen, &req->rq_repbuf);
263 printk("mds: out of memory\n");
264 req->rq_status = -ENOMEM;
268 req->rq_rephdr->seqno = req->rq_reqhdr->seqno;
269 rep = req->rq_rep.mds;
273 req->rq_rephdr->status = PTR_ERR(de);
277 file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
278 /* note: in case of an error, dentry_open puts dentry */
281 req->rq_rephdr->status = PTR_ERR(file);
285 niobuf = mds_req_tgt(req->rq_req.mds);
287 /* to make this asynchronous make sure that the handling function
288 doesn't send a reply when this function completes. Instead a
289 callback function would send the reply */
290 rc = mds_sendpage(req, file, req->rq_req.mds->size, niobuf);
293 req->rq_rephdr->status = rc;
298 int mds_reint(struct ptlrpc_request *req)
301 char *buf = mds_req_tgt(req->rq_req.mds);
302 int len = req->rq_req.mds->tgtlen;
303 struct mds_update_record rec;
305 rc = mds_update_unpack(buf, len, &rec);
307 printk(__FUNCTION__ ": invalid record\n");
308 req->rq_status = -EINVAL;
311 /* rc will be used to interrupt a for loop over multiple records */
312 rc = mds_reint_rec(&rec, req);
316 //int mds_handle(struct mds_conn *conn, int len, char *buf)
317 int mds_handle(struct ptlrpc_request *req)
320 struct ptlreq_hdr *hdr;
324 hdr = (struct ptlreq_hdr *)req->rq_reqbuf;
326 if (NTOH__u32(hdr->type) != MDS_TYPE_REQ) {
327 printk("lustre_mds: wrong packet type sent %d\n",
328 NTOH__u32(hdr->type));
333 rc = mds_unpack_req(req->rq_reqbuf, req->rq_reqlen,
334 &req->rq_reqhdr, &req->rq_req.mds);
336 printk("lustre_mds: Invalid request\n");
341 switch (req->rq_reqhdr->opc) {
344 CDEBUG(D_INODE, "getattr\n");
345 rc = mds_getattr(req);
349 CDEBUG(D_INODE, "readpage\n");
350 rc = mds_readpage(req);
354 CDEBUG(D_INODE, "reint\n");
359 return mds_error(req);
364 printk(__FUNCTION__ ": no header\n");
368 if( req->rq_status) {
371 CDEBUG(D_INODE, "sending reply\n");
379 static void mds_timer_run(unsigned long __data)
381 struct task_struct * p = (struct task_struct *) __data;
386 int mds_main(void *arg)
388 struct mds_obd *mds = (struct mds_obd *) arg;
389 struct timer_list timer;
393 spin_lock_irq(¤t->sigmask_lock);
394 sigfillset(¤t->blocked);
395 recalc_sigpending(current);
396 spin_unlock_irq(¤t->sigmask_lock);
398 sprintf(current->comm, "lustre_mds");
400 /* Set up an interval timer which can be used to trigger a
401 wakeup after the interval expires */
403 timer.data = (unsigned long) current;
404 timer.function = mds_timer_run;
405 mds->mds_timer = &timer;
407 /* Record that the thread is running */
408 mds->mds_thread = current;
409 wake_up(&mds->mds_done_waitq);
411 printk(KERN_INFO "lustre_mds starting. Commit interval %d seconds\n",
412 mds->mds_interval / HZ);
414 /* XXX maintain a list of all managed devices: insert here */
416 /* And now, wait forever for commit wakeup events. */
418 struct ptlrpc_request *request;
421 if (mds->mds_flags & MDS_UNMOUNT)
425 wake_up(&mds->mds_done_waitq);
426 interruptible_sleep_on(&mds->mds_waitq);
428 CDEBUG(D_INODE, "lustre_mds wakes\n");
429 CDEBUG(D_INODE, "pick up req here and continue\n");
431 if (list_empty(&mds->mds_reqs)) {
432 CDEBUG(D_INODE, "woke because of timer\n");
434 request = list_entry(mds->mds_reqs.next,
435 struct ptlrpc_request, rq_list);
436 list_del(&request->rq_list);
437 rc = mds_handle(request);
441 del_timer_sync(mds->mds_timer);
443 /* XXX maintain a list of all managed devices: cleanup here */
445 mds->mds_thread = NULL;
446 wake_up(&mds->mds_done_waitq);
447 printk("lustre_mds: exiting\n");
451 static void mds_stop_srv_thread(struct mds_obd *mds)
453 mds->mds_flags |= MDS_UNMOUNT;
455 while (mds->mds_thread) {
456 wake_up(&mds->mds_waitq);
457 sleep_on(&mds->mds_done_waitq);
461 static void mds_start_srv_thread(struct mds_obd *mds)
463 init_waitqueue_head(&mds->mds_waitq);
464 init_waitqueue_head(&mds->mds_done_waitq);
465 kernel_thread(mds_main, (void *)mds,
466 CLONE_VM | CLONE_FS | CLONE_FILES);
467 while (!mds->mds_thread)
468 sleep_on(&mds->mds_done_waitq);
471 /* mount the file system (secretly) */
472 static int mds_setup(struct obd_device *obddev, obd_count len,
476 struct obd_ioctl_data* data = buf;
477 struct mds_obd *mds = &obddev->u.mds;
480 /* If the ioctl data contains two inline buffers, this is a request to
481 * mount a local filesystem for metadata storage. If the ioctl data
482 * contains one buffer, however, it is the UUID of the remote node that
483 * we will contact for metadata. */
484 if (data->ioc_inllen2 == 0) {
487 nid = kportal_uuid_to_nid(data->ioc_inlbuf1);
489 printk("Lustre: uuid_to_nid failed; use ptlctl to "
490 "associate this uuid with a NID\n");
494 printk("Lustre MDS: remote nid is %u\n", nid);
495 mds->mds_remote_nid = nid;
497 struct vfsmount *mnt;
500 mnt = do_kern_mount(data->ioc_inlbuf2, 0,
501 data->ioc_inlbuf1, NULL);
508 mds->mds_sb = mnt->mnt_root->d_inode->i_sb;
509 if (!obddev->u.mds.mds_sb) {
514 mds->mds_vfsmnt = mnt;
515 obddev->u.mds.mds_fstype = strdup(data->ioc_inlbuf2);
517 mds->mds_ctxt.pwdmnt = mnt;
518 mds->mds_ctxt.pwd = mnt->mnt_root;
519 mds->mds_ctxt.fs = KERNEL_DS;
520 mds->mds_remote_nid = 0;
523 INIT_LIST_HEAD(&mds->mds_reqs);
524 mds->mds_thread = NULL;
526 mds->mds_interval = 3 * HZ;
529 spin_lock_init(&obddev->u.mds.mds_lock);
531 mds_start_srv_thread(mds);
538 static int mds_cleanup(struct obd_device * obddev)
540 struct super_block *sb;
541 struct mds_obd *mds = &obddev->u.mds;
545 if ( !(obddev->obd_flags & OBD_SET_UP) ) {
550 if ( !list_empty(&obddev->obd_gen_clients) ) {
551 printk(KERN_WARNING __FUNCTION__ ": still has clients!\n");
557 mds_stop_srv_thread(mds);
564 if (!list_empty(&mds->mds_reqs)) {
565 // XXX reply with errors and clean up
566 CDEBUG(D_INODE, "Request list not empty!\n");
570 mntput(mds->mds_vfsmnt);
572 kfree(mds->mds_fstype);
581 /* use obd ops to offer management infrastructure */
582 static struct obd_ops mds_obd_ops = {
584 o_cleanup: mds_cleanup,
587 static int __init mds_init(void)
589 obd_register_type(&mds_obd_ops, LUSTRE_MDS_NAME);
593 static void __exit mds_exit(void)
595 obd_unregister_type(LUSTRE_MDS_NAME);
598 MODULE_AUTHOR("Peter J. Braam <braam@clusterfs.com>");
599 MODULE_DESCRIPTION("Lustre Metadata Server (MDS) v0.01");
600 MODULE_LICENSE("GPL");
603 // for testing (maybe this stays)
604 EXPORT_SYMBOL(mds_queue_req);
606 module_init(mds_init);
607 module_exit(mds_exit);