1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mds/handler.c
38 * Author: Peter Braam <braam@clusterfs.com>
39 * Author: Andreas Dilger <adilger@clusterfs.com>
40 * Author: Phil Schwan <phil@clusterfs.com>
41 * Author: Mike Shaver <shaver@clusterfs.com>
44 #define DEBUG_SUBSYSTEM S_MDS
46 #include <lustre_mds.h>
47 #include <linux/module.h>
48 #include <linux/init.h>
50 #include <linux/jbd.h>
51 #include <linux/smp_lock.h>
52 #include <linux/buffer_head.h>
53 #include <linux/workqueue.h>
54 #include <linux/mount.h>
56 #include <lustre_acl.h>
57 #include <obd_class.h>
58 #include <lustre_dlm.h>
60 #include <lustre_fsfilt.h>
61 #include <lprocfs_status.h>
62 #include <lustre_disk.h>
63 #include <lustre_param.h>
65 #include "mds_internal.h"
67 __u32 mds_max_ost_index=0xFFFF;
68 CFS_MODULE_PARM(mds_max_ost_index, "i", int, 0444,
71 static int mds_lov_presetup (struct mds_obd *mds, struct lustre_cfg *lcfg)
76 if (lcfg->lcfg_bufcount >= 4 && LUSTRE_CFG_BUFLEN(lcfg, 3) > 0) {
79 ll_generate_random_uuid(uuid);
80 class_uuid_unparse(uuid, &mds->mds_lov_uuid);
82 OBD_ALLOC(mds->mds_profile, LUSTRE_CFG_BUFLEN(lcfg, 3));
83 if (mds->mds_profile == NULL)
86 strncpy(mds->mds_profile, lustre_cfg_string(lcfg, 3),
87 LUSTRE_CFG_BUFLEN(lcfg, 3));
92 static int mds_lov_clean(struct obd_device *obd)
94 struct mds_obd *mds = &obd->u.mds;
95 struct obd_device *osc = mds->mds_lov_obd;
98 if (mds->mds_profile) {
99 class_del_profile(mds->mds_profile);
100 OBD_FREE(mds->mds_profile, strlen(mds->mds_profile) + 1);
101 mds->mds_profile = NULL;
104 /* There better be a lov */
108 RETURN(PTR_ERR(osc));
110 obd_register_observer(osc, NULL);
112 /* Give lov our same shutdown flags */
113 osc->obd_force = obd->obd_force;
114 osc->obd_fail = obd->obd_fail;
116 /* Cleanup the lov */
117 obd_disconnect(mds->mds_lov_exp);
118 class_manual_cleanup(osc);
123 static int mds_postsetup(struct obd_device *obd)
125 struct mds_obd *mds = &obd->u.mds;
126 struct llog_ctxt *ctxt;
130 rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd, 0, NULL,
135 rc = llog_setup(obd, &obd->obd_olg, LLOG_LOVEA_ORIG_CTXT, obd, 0, NULL,
140 mds_changelog_llog_init(obd, obd);
142 if (mds->mds_profile) {
143 struct lustre_profile *lprof;
144 /* The profile defines which osc and mdc to connect to, for a
145 client. We reuse that here to figure out the name of the
146 lov to use (and ignore lprof->lp_md).
147 The profile was set in the config log with
148 LCFG_MOUNTOPT profilenm oscnm mdcnm */
149 lprof = class_get_profile(mds->mds_profile);
151 CERROR("No profile found: %s\n", mds->mds_profile);
152 GOTO(err_cleanup, rc = -ENOENT);
154 rc = mds_lov_connect(obd, lprof->lp_dt);
156 GOTO(err_cleanup, rc);
163 ctxt = llog_get_context(obd, LLOG_LOVEA_ORIG_CTXT);
167 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
173 int mds_postrecov(struct obd_device *obd)
181 LASSERT(!obd->obd_recovering);
182 /* clean PENDING dir */
184 if (strncmp(obd->obd_name, MDD_OBD_NAME, strlen(MDD_OBD_NAME)))
185 rc = mds_cleanup_pending(obd);
189 /* FIXME Does target_finish_recovery really need this to block? */
190 /* Notify the LOV, which will in turn call mds_notify for each tgt */
191 /* This means that we have to hack obd_notify to think we're obd_set_up
192 during mds_lov_connect. */
193 obd_notify(obd->u.mds.mds_lov_obd, NULL,
194 obd->obd_async_recov ? OBD_NOTIFY_SYNC_NONBLOCK :
195 OBD_NOTIFY_SYNC, NULL);
200 /* We need to be able to stop an mds_lov_synchronize */
201 static int mds_lov_early_clean(struct obd_device *obd)
203 struct mds_obd *mds = &obd->u.mds;
204 struct obd_device *osc = mds->mds_lov_obd;
206 if (!osc || (!obd->obd_force && !obd->obd_fail))
209 CDEBUG(D_HA, "abort inflight\n");
210 return (obd_precleanup(osc, OBD_CLEANUP_EARLY));
213 static int mds_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
215 struct mds_obd *mds = &obd->u.mds;
216 struct llog_ctxt *ctxt;
221 case OBD_CLEANUP_EARLY:
223 case OBD_CLEANUP_EXPORTS:
224 mds_lov_early_clean(obd);
225 cfs_down_write(&mds->mds_notify_lock);
226 mds_lov_disconnect(obd);
228 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
231 ctxt = llog_get_context(obd, LLOG_LOVEA_ORIG_CTXT);
234 rc = obd_llog_finish(obd, 0);
235 mds->mds_lov_exp = NULL;
236 cfs_up_write(&mds->mds_notify_lock);
242 /* Look up an entry by inode number. */
243 /* this function ONLY returns valid dget'd dentries with an initialized inode
245 static struct dentry *mds_lvfs_fid2dentry(__u64 id, __u32 gen, __u64 gr,
248 struct fsfilt_fid fid;
249 struct obd_device *obd = (struct obd_device *)data;
252 RETURN(ERR_PTR(-ESTALE));
257 RETURN(fsfilt_fid2dentry(obd, obd->u.mds.mds_obt.obt_vfsmnt, &fid, 0));
261 struct lvfs_callback_ops mds_lvfs_ops = {
262 l_fid2dentry: mds_lvfs_fid2dentry,
265 static void mds_init_ctxt(struct obd_device *obd, struct vfsmount *mnt)
267 struct mds_obd *mds = &obd->u.mds;
269 mds->mds_obt.obt_vfsmnt = mnt;
270 /* why not mnt->mnt_sb instead of mnt->mnt_root->d_inode->i_sb? */
271 obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
272 obd->u.obt.obt_magic = OBT_MAGIC;
273 fsfilt_setup(obd, obd->u.obt.obt_sb);
275 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
276 obd->obd_lvfs_ctxt.pwdmnt = mnt;
277 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
278 obd->obd_lvfs_ctxt.fs = get_ds();
279 obd->obd_lvfs_ctxt.cb_ops = mds_lvfs_ops;
283 /*mds still need lov setup here*/
284 static int mds_cmd_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
286 struct mds_obd *mds = &obd->u.mds;
287 struct lvfs_run_ctxt saved;
289 struct vfsmount *mnt;
290 struct lustre_sb_info *lsi;
291 struct lustre_mount_info *lmi;
292 struct dentry *dentry;
296 CDEBUG(D_INFO, "obd %s setup \n", obd->obd_name);
297 if (strncmp(obd->obd_name, MDD_OBD_NAME, strlen(MDD_OBD_NAME)))
300 if (lcfg->lcfg_bufcount < 5) {
301 CERROR("invalid arg for setup %s\n", MDD_OBD_NAME);
304 dev = lustre_cfg_string(lcfg, 4);
305 lmi = server_get_mount(dev);
306 LASSERT(lmi != NULL);
308 lsi = s2lsi(lmi->lmi_sb);
310 /* FIXME: MDD LOV initialize objects.
311 * we need only lmi here but not get mount
312 * OSD did mount already, so put mount back
314 cfs_atomic_dec(&lsi->lsi_mounts);
316 cfs_init_rwsem(&mds->mds_notify_lock);
318 obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
319 mds_init_ctxt(obd, mnt);
321 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
322 dentry = simple_mkdir(cfs_fs_pwd(current->fs), mnt, "OBJECTS", 0777, 1);
323 if (IS_ERR(dentry)) {
324 rc = PTR_ERR(dentry);
325 CERROR("cannot create OBJECTS directory: rc = %d\n", rc);
328 mds->mds_objects_dir = dentry;
330 rc = mds_lov_init_objids(obd);
332 CERROR("cannot init lov objid rc = %d\n", rc);
333 GOTO(err_objects, rc );
336 rc = mds_lov_presetup(mds, lcfg);
338 GOTO(err_objects, rc);
340 /* Don't wait for mds_postrecov trying to clear orphans */
341 obd->obd_async_recov = 1;
342 rc = mds_postsetup(obd);
343 /* Bug 11557 - allow async abort_recov start
344 FIXME can remove most of this obd_async_recov plumbing
345 obd->obd_async_recov = 0;
349 GOTO(err_objects, rc);
352 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
355 dput(mds->mds_objects_dir);
357 fsfilt_put_ops(obd->obd_fsops);
361 static int mds_cmd_cleanup(struct obd_device *obd)
363 struct mds_obd *mds = &obd->u.mds;
364 struct lvfs_run_ctxt saved;
368 mds->mds_lov_exp = NULL;
371 LCONSOLE_WARN("%s: shutting down for failover; client state "
372 "will be preserved.\n", obd->obd_name);
374 if (strncmp(obd->obd_name, MDD_OBD_NAME, strlen(MDD_OBD_NAME)))
377 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
379 mds_lov_destroy_objids(obd);
381 if (mds->mds_objects_dir != NULL) {
382 l_dput(mds->mds_objects_dir);
383 mds->mds_objects_dir = NULL;
386 ll_vfs_dq_off(obd->u.obt.obt_sb, 0);
387 shrink_dcache_sb(mds->mds_obt.obt_sb);
388 fsfilt_put_ops(obd->obd_fsops);
390 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
395 static int mds_cmd_health_check(struct obd_device *obd)
400 static struct obd_ops mds_cmd_obd_ops = {
401 .o_owner = THIS_MODULE,
402 .o_setup = mds_cmd_setup,
403 .o_cleanup = mds_cmd_cleanup,
404 .o_precleanup = mds_precleanup,
405 .o_create = mds_obd_create,
406 .o_destroy = mds_obd_destroy,
407 .o_llog_init = mds_llog_init,
408 .o_llog_finish = mds_llog_finish,
409 .o_notify = mds_notify,
410 .o_postrecov = mds_postrecov,
411 // .o_health_check = mds_cmd_health_check,
414 quota_interface_t *mds_quota_interface_ref;
415 extern quota_interface_t mds_quota_interface;
417 static int __init mds_cmd_init(void)
419 struct lprocfs_static_vars lvars;
422 cfs_request_module("%s", "lquota");
423 mds_quota_interface_ref = PORTAL_SYMBOL_GET(mds_quota_interface);
424 rc = lquota_init(mds_quota_interface_ref);
426 if (mds_quota_interface_ref)
427 PORTAL_SYMBOL_PUT(mds_quota_interface);
430 init_obd_quota_ops(mds_quota_interface_ref, &mds_cmd_obd_ops);
432 lprocfs_mds_init_vars(&lvars);
433 class_register_type(&mds_cmd_obd_ops, NULL, lvars.module_vars,
434 LUSTRE_MDS_NAME, NULL);
439 static void /*__exit*/ mds_cmd_exit(void)
441 lquota_exit(mds_quota_interface_ref);
442 if (mds_quota_interface_ref)
443 PORTAL_SYMBOL_PUT(mds_quota_interface);
445 class_unregister_type(LUSTRE_MDS_NAME);
448 EXPORT_SYMBOL(mds_quota_interface_ref);
449 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
450 MODULE_DESCRIPTION("Lustre Metadata Server (MDS)");
451 MODULE_LICENSE("GPL");
453 module_init(mds_cmd_init);
454 module_exit(mds_cmd_exit);