4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
31 * This file is part of Lustre, http://www.lustre.org/
32 * Lustre is a trademark of Sun Microsystems, Inc.
34 * lustre/mds/handler.c
36 * Author: Peter Braam <braam@clusterfs.com>
37 * Author: Andreas Dilger <adilger@clusterfs.com>
38 * Author: Phil Schwan <phil@clusterfs.com>
39 * Author: Mike Shaver <shaver@clusterfs.com>
42 #define DEBUG_SUBSYSTEM S_MDS
44 #include <lustre_mds.h>
45 #include <linux/module.h>
46 #include <linux/init.h>
48 #include <linux/jbd.h>
49 #include <linux/buffer_head.h>
50 #include <linux/workqueue.h>
51 #include <linux/mount.h>
53 #include <lustre_acl.h>
54 #include <obd_class.h>
55 #include <lustre_dlm.h>
57 #include <lustre_fsfilt.h>
58 #include <lprocfs_status.h>
59 #include <lustre_disk.h>
60 #include <lustre_param.h>
61 #include <lustre_log.h>
63 #include "mds_internal.h"
65 __u32 mds_max_ost_index=0xFFFF;
66 CFS_MODULE_PARM(mds_max_ost_index, "i", int, 0444,
69 static int mds_lov_presetup (struct mds_obd *mds, struct lustre_cfg *lcfg)
74 if (lcfg->lcfg_bufcount >= 4 && LUSTRE_CFG_BUFLEN(lcfg, 3) > 0) {
77 ll_generate_random_uuid(uuid);
78 class_uuid_unparse(uuid, &mds->mds_lov_uuid);
80 OBD_ALLOC(mds->mds_profile, LUSTRE_CFG_BUFLEN(lcfg, 3));
81 if (mds->mds_profile == NULL)
84 strncpy(mds->mds_profile, lustre_cfg_string(lcfg, 3),
85 LUSTRE_CFG_BUFLEN(lcfg, 3));
90 static int mds_lov_clean(struct obd_device *obd)
92 struct mds_obd *mds = &obd->u.mds;
93 struct obd_device *osc = mds->mds_lov_obd;
96 if (mds->mds_profile) {
97 class_del_profile(mds->mds_profile);
98 OBD_FREE(mds->mds_profile, strlen(mds->mds_profile) + 1);
99 mds->mds_profile = NULL;
102 /* There better be a lov */
106 RETURN(PTR_ERR(osc));
108 obd_register_observer(osc, NULL);
110 /* Give lov our same shutdown flags */
111 osc->obd_force = obd->obd_force;
112 osc->obd_fail = obd->obd_fail;
114 /* Cleanup the lov */
115 obd_disconnect(mds->mds_lov_exp);
116 class_manual_cleanup(osc);
121 static int mds_postsetup(struct obd_device *obd)
123 struct mds_obd *mds = &obd->u.mds;
124 struct llog_ctxt *ctxt;
128 rc = llog_setup(NULL, obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd,
133 if (mds->mds_profile) {
134 struct lustre_profile *lprof;
135 /* The profile defines which osc and mdc to connect to, for a
136 client. We reuse that here to figure out the name of the
137 lov to use (and ignore lprof->lp_md).
138 The profile was set in the config log with
139 LCFG_MOUNTOPT profilenm oscnm mdcnm */
140 lprof = class_get_profile(mds->mds_profile);
142 CERROR("No profile found: %s\n", mds->mds_profile);
143 GOTO(err_cleanup, rc = -ENOENT);
145 rc = mds_lov_connect(obd, lprof->lp_dt);
147 GOTO(err_cleanup, rc);
154 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
156 llog_cleanup(NULL, ctxt);
160 int mds_postrecov(struct obd_device *obd)
168 LASSERT(!obd->obd_recovering);
169 /* clean PENDING dir */
171 if (strncmp(obd->obd_name, MDD_OBD_NAME, strlen(MDD_OBD_NAME)))
172 rc = mds_cleanup_pending(obd);
176 /* FIXME Does target_finish_recovery really need this to block? */
177 /* Notify the LOV, which will in turn call mds_notify for each tgt */
178 /* This means that we have to hack obd_notify to think we're obd_set_up
179 during mds_lov_connect. */
180 obd_notify(obd->u.mds.mds_lov_obd, NULL,
181 obd->obd_async_recov ? OBD_NOTIFY_SYNC_NONBLOCK :
182 OBD_NOTIFY_SYNC, NULL);
187 /* We need to be able to stop an mds_lov_synchronize */
188 static int mds_lov_early_clean(struct obd_device *obd)
190 struct mds_obd *mds = &obd->u.mds;
191 struct obd_device *osc = mds->mds_lov_obd;
193 if (!osc || (!obd->obd_force && !obd->obd_fail))
196 CDEBUG(D_HA, "abort inflight\n");
197 return (obd_precleanup(osc, OBD_CLEANUP_EARLY));
200 static int mds_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
202 struct mds_obd *mds = &obd->u.mds;
203 struct llog_ctxt *ctxt;
208 case OBD_CLEANUP_EARLY:
210 case OBD_CLEANUP_EXPORTS:
211 mds_lov_early_clean(obd);
212 cfs_down_write(&mds->mds_notify_lock);
213 mds_lov_disconnect(obd);
215 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
217 llog_cleanup(NULL, ctxt);
218 rc = obd_llog_finish(obd, 0);
219 mds->mds_lov_exp = NULL;
220 cfs_up_write(&mds->mds_notify_lock);
226 /* Look up an entry by inode number. */
227 /* this function ONLY returns valid dget'd dentries with an initialized inode
229 static struct dentry *mds_lvfs_fid2dentry(__u64 id, __u32 gen, __u64 gr,
232 struct fsfilt_fid fid;
233 struct obd_device *obd = (struct obd_device *)data;
236 RETURN(ERR_PTR(-ESTALE));
241 RETURN(fsfilt_fid2dentry(obd, obd->u.mds.mds_obt.obt_vfsmnt, &fid, 0));
245 struct lvfs_callback_ops mds_lvfs_ops = {
246 l_fid2dentry: mds_lvfs_fid2dentry,
249 static int mds_init_ctxt(struct obd_device *obd, struct vfsmount *mnt)
251 struct mds_obd *mds = &obd->u.mds;
254 mds->mds_obt.obt_vfsmnt = mnt;
255 /* why not mnt->mnt_sb instead of mnt->mnt_root->d_inode->i_sb? */
256 obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
257 obd->u.obt.obt_magic = OBT_MAGIC;
258 rc = fsfilt_setup(obd, obd->u.obt.obt_sb);
262 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
263 obd->obd_lvfs_ctxt.pwdmnt = mnt;
264 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
265 obd->obd_lvfs_ctxt.fs = get_ds();
266 obd->obd_lvfs_ctxt.cb_ops = mds_lvfs_ops;
270 /*mds still need lov setup here*/
271 static int mds_cmd_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
273 struct mds_obd *mds = &obd->u.mds;
274 struct lvfs_run_ctxt saved;
276 struct vfsmount *mnt;
277 struct lustre_sb_info *lsi;
278 struct lustre_mount_info *lmi;
279 struct dentry *dentry;
283 CDEBUG(D_INFO, "obd %s setup \n", obd->obd_name);
284 if (strncmp(obd->obd_name, MDD_OBD_NAME, strlen(MDD_OBD_NAME)))
287 if (lcfg->lcfg_bufcount < 5) {
288 CERROR("invalid arg for setup %s\n", MDD_OBD_NAME);
291 dev = lustre_cfg_string(lcfg, 4);
292 lmi = server_get_mount(dev);
293 LASSERT(lmi != NULL);
295 lsi = s2lsi(lmi->lmi_sb);
297 /* FIXME: MDD LOV initialize objects.
298 * we need only lmi here but not get mount
299 * OSD did mount already, so put mount back
301 cfs_atomic_dec(&lsi->lsi_mounts);
302 cfs_init_rwsem(&mds->mds_notify_lock);
304 obd->obd_fsops = fsfilt_get_ops(lsi->lsi_fstype);
305 rc = mds_init_ctxt(obd, mnt);
309 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
310 dentry = simple_mkdir(cfs_fs_pwd(current->fs), mnt, "OBJECTS", 0777, 1);
311 if (IS_ERR(dentry)) {
312 rc = PTR_ERR(dentry);
313 CERROR("cannot create OBJECTS directory: rc = %d\n", rc);
316 mds->mds_objects_dir = dentry;
318 rc = mds_lov_init_objids(obd);
320 CERROR("cannot init lov objid rc = %d\n", rc);
321 GOTO(err_objects, rc );
324 rc = mds_lov_presetup(mds, lcfg);
326 GOTO(err_objects, rc);
328 /* Don't wait for mds_postrecov trying to clear orphans */
329 obd->obd_async_recov = 1;
330 rc = mds_postsetup(obd);
331 /* Bug 11557 - allow async abort_recov start
332 FIXME can remove most of this obd_async_recov plumbing
333 obd->obd_async_recov = 0;
337 GOTO(err_objects, rc);
339 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
343 dput(mds->mds_objects_dir);
345 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
347 fsfilt_put_ops(obd->obd_fsops);
351 static int mds_cmd_cleanup(struct obd_device *obd)
353 struct mds_obd *mds = &obd->u.mds;
354 struct lvfs_run_ctxt saved;
358 mds->mds_lov_exp = NULL;
361 LCONSOLE_WARN("%s: shutting down for failover; client state "
362 "will be preserved.\n", obd->obd_name);
364 if (strncmp(obd->obd_name, MDD_OBD_NAME, strlen(MDD_OBD_NAME)))
367 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
369 mds_lov_destroy_objids(obd);
371 if (mds->mds_objects_dir != NULL) {
372 l_dput(mds->mds_objects_dir);
373 mds->mds_objects_dir = NULL;
376 ll_vfs_dq_off(obd->u.obt.obt_sb, 0);
377 shrink_dcache_sb(mds->mds_obt.obt_sb);
378 fsfilt_put_ops(obd->obd_fsops);
380 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
385 static int mds_cmd_health_check(struct obd_device *obd)
390 static struct obd_ops mds_cmd_obd_ops = {
391 .o_owner = THIS_MODULE,
392 .o_setup = mds_cmd_setup,
393 .o_cleanup = mds_cmd_cleanup,
394 .o_precleanup = mds_precleanup,
395 .o_create = mds_obd_create,
396 .o_destroy = mds_obd_destroy,
397 .o_llog_init = mds_llog_init,
398 .o_llog_finish = mds_llog_finish,
399 .o_notify = mds_notify,
400 .o_postrecov = mds_postrecov,
401 // .o_health_check = mds_cmd_health_check,
404 static int __init mds_cmd_init(void)
406 struct lprocfs_static_vars lvars;
408 lprocfs_mds_init_vars(&lvars);
409 class_register_type(&mds_cmd_obd_ops, NULL, lvars.module_vars,
410 LUSTRE_MDS_NAME, NULL);
415 static void /*__exit*/ mds_cmd_exit(void)
417 class_unregister_type(LUSTRE_MDS_NAME);
420 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
421 MODULE_DESCRIPTION("Lustre Metadata Server (MDS)");
422 MODULE_LICENSE("GPL");
424 module_init(mds_cmd_init);
425 module_exit(mds_cmd_exit);