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 /* Look up an entry by inode number. */
72 /* this function ONLY returns valid dget'd dentries with an initialized inode
74 struct dentry *mds_fid2dentry(struct mds_obd *mds, struct ll_fid *fid,
75 struct vfsmount **mnt)
78 unsigned long ino = fid->id;
79 __u32 generation = fid->generation;
81 struct dentry *result;
84 RETURN(ERR_PTR(-ESTALE));
86 snprintf(fid_name, sizeof(fid_name), "0x%lx", ino);
88 /* under ext3 this is neither supposed to return bad inodes
90 result = ll_lookup_one_len(fid_name, mds->mds_fid_de, strlen(fid_name));
94 inode = result->d_inode;
96 RETURN(ERR_PTR(-ENOENT));
98 if (inode->i_generation == 0 || inode->i_nlink == 0) {
99 LCONSOLE_WARN("Found inode with zero generation or link -- this"
100 " may indicate disk corruption (inode: %lu/%u, "
101 "link %lu, count %d)\n", inode->i_ino,
102 inode->i_generation,(unsigned long)inode->i_nlink,
103 atomic_read(&inode->i_count));
105 RETURN(ERR_PTR(-ENOENT));
108 if (generation && inode->i_generation != generation) {
109 /* we didn't find the right inode.. */
110 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
111 "count: %d, generation %u/%u\n", inode->i_ino,
112 (unsigned long)inode->i_nlink,
113 atomic_read(&inode->i_count), inode->i_generation,
116 RETURN(ERR_PTR(-ENOENT));
120 *mnt = mds->mds_obt.obt_vfsmnt;
127 static int mds_lov_presetup (struct mds_obd *mds, struct lustre_cfg *lcfg)
132 if (lcfg->lcfg_bufcount >= 4 && LUSTRE_CFG_BUFLEN(lcfg, 3) > 0) {
135 ll_generate_random_uuid(uuid);
136 class_uuid_unparse(uuid, &mds->mds_lov_uuid);
138 OBD_ALLOC(mds->mds_profile, LUSTRE_CFG_BUFLEN(lcfg, 3));
139 if (mds->mds_profile == NULL)
142 strncpy(mds->mds_profile, lustre_cfg_string(lcfg, 3),
143 LUSTRE_CFG_BUFLEN(lcfg, 3));
148 static int mds_lov_clean(struct obd_device *obd)
150 struct mds_obd *mds = &obd->u.mds;
151 struct obd_device *osc = mds->mds_lov_obd;
154 if (mds->mds_profile) {
155 class_del_profile(mds->mds_profile);
156 OBD_FREE(mds->mds_profile, strlen(mds->mds_profile) + 1);
157 mds->mds_profile = NULL;
160 /* There better be a lov */
164 RETURN(PTR_ERR(osc));
166 obd_register_observer(osc, NULL);
168 /* Give lov our same shutdown flags */
169 osc->obd_force = obd->obd_force;
170 osc->obd_fail = obd->obd_fail;
172 /* Cleanup the lov */
173 obd_disconnect(mds->mds_lov_exp);
174 class_manual_cleanup(osc);
179 static int mds_postsetup(struct obd_device *obd)
181 struct mds_obd *mds = &obd->u.mds;
182 struct llog_ctxt *ctxt;
186 rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd, 0, NULL,
191 rc = llog_setup(obd, &obd->obd_olg, LLOG_LOVEA_ORIG_CTXT, obd, 0, NULL,
196 mds_changelog_llog_init(obd, obd);
198 if (mds->mds_profile) {
199 struct lustre_profile *lprof;
200 /* The profile defines which osc and mdc to connect to, for a
201 client. We reuse that here to figure out the name of the
202 lov to use (and ignore lprof->lp_md).
203 The profile was set in the config log with
204 LCFG_MOUNTOPT profilenm oscnm mdcnm */
205 lprof = class_get_profile(mds->mds_profile);
207 CERROR("No profile found: %s\n", mds->mds_profile);
208 GOTO(err_cleanup, rc = -ENOENT);
210 rc = mds_lov_connect(obd, lprof->lp_dt);
212 GOTO(err_cleanup, rc);
219 ctxt = llog_get_context(obd, LLOG_LOVEA_ORIG_CTXT);
223 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
229 int mds_postrecov(struct obd_device *obd)
237 LASSERT(!obd->obd_recovering);
238 /* clean PENDING dir */
240 if (strncmp(obd->obd_name, MDD_OBD_NAME, strlen(MDD_OBD_NAME)))
241 rc = mds_cleanup_pending(obd);
245 /* FIXME Does target_finish_recovery really need this to block? */
246 /* Notify the LOV, which will in turn call mds_notify for each tgt */
247 /* This means that we have to hack obd_notify to think we're obd_set_up
248 during mds_lov_connect. */
249 obd_notify(obd->u.mds.mds_lov_obd, NULL,
250 obd->obd_async_recov ? OBD_NOTIFY_SYNC_NONBLOCK :
251 OBD_NOTIFY_SYNC, NULL);
256 /* We need to be able to stop an mds_lov_synchronize */
257 static int mds_lov_early_clean(struct obd_device *obd)
259 struct mds_obd *mds = &obd->u.mds;
260 struct obd_device *osc = mds->mds_lov_obd;
262 if (!osc || (!obd->obd_force && !obd->obd_fail))
265 CDEBUG(D_HA, "abort inflight\n");
266 return (obd_precleanup(osc, OBD_CLEANUP_EARLY));
269 static int mds_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
271 struct mds_obd *mds = &obd->u.mds;
272 struct llog_ctxt *ctxt;
277 case OBD_CLEANUP_EARLY:
279 case OBD_CLEANUP_EXPORTS:
280 mds_lov_early_clean(obd);
281 cfs_down_write(&mds->mds_notify_lock);
282 mds_lov_disconnect(obd);
284 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
287 ctxt = llog_get_context(obd, LLOG_LOVEA_ORIG_CTXT);
290 rc = obd_llog_finish(obd, 0);
291 mds->mds_lov_exp = NULL;
292 cfs_up_write(&mds->mds_notify_lock);
298 static struct dentry *mds_lvfs_fid2dentry(__u64 id, __u32 gen, __u64 gr,
301 struct obd_device *obd = data;
304 fid.generation = gen;
305 return mds_fid2dentry(&obd->u.mds, &fid, NULL);
309 struct lvfs_callback_ops mds_lvfs_ops = {
310 l_fid2dentry: mds_lvfs_fid2dentry,
313 static void mds_init_ctxt(struct obd_device *obd, struct vfsmount *mnt)
315 struct mds_obd *mds = &obd->u.mds;
317 mds->mds_obt.obt_vfsmnt = mnt;
318 /* why not mnt->mnt_sb instead of mnt->mnt_root->d_inode->i_sb? */
319 obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
320 obd->u.obt.obt_magic = OBT_MAGIC;
321 fsfilt_setup(obd, obd->u.obt.obt_sb);
323 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
324 obd->obd_lvfs_ctxt.pwdmnt = mnt;
325 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
326 obd->obd_lvfs_ctxt.fs = get_ds();
327 obd->obd_lvfs_ctxt.cb_ops = mds_lvfs_ops;
331 /*mds still need lov setup here*/
332 static int mds_cmd_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
334 struct mds_obd *mds = &obd->u.mds;
335 struct lvfs_run_ctxt saved;
337 struct vfsmount *mnt;
338 struct lustre_sb_info *lsi;
339 struct lustre_mount_info *lmi;
340 struct dentry *dentry;
344 CDEBUG(D_INFO, "obd %s setup \n", obd->obd_name);
345 if (strncmp(obd->obd_name, MDD_OBD_NAME, strlen(MDD_OBD_NAME)))
348 if (lcfg->lcfg_bufcount < 5) {
349 CERROR("invalid arg for setup %s\n", MDD_OBD_NAME);
352 dev = lustre_cfg_string(lcfg, 4);
353 lmi = server_get_mount(dev);
354 LASSERT(lmi != NULL);
356 lsi = s2lsi(lmi->lmi_sb);
358 /* FIXME: MDD LOV initialize objects.
359 * we need only lmi here but not get mount
360 * OSD did mount already, so put mount back
362 cfs_atomic_dec(&lsi->lsi_mounts);
364 cfs_init_rwsem(&mds->mds_notify_lock);
366 obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
367 mds_init_ctxt(obd, mnt);
369 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
370 dentry = simple_mkdir(cfs_fs_pwd(current->fs), mnt, "OBJECTS", 0777, 1);
371 if (IS_ERR(dentry)) {
372 rc = PTR_ERR(dentry);
373 CERROR("cannot create OBJECTS directory: rc = %d\n", rc);
376 mds->mds_objects_dir = dentry;
378 dentry = ll_lookup_one_len("__iopen__", cfs_fs_pwd(current->fs),
379 strlen("__iopen__"));
380 if (IS_ERR(dentry)) {
381 rc = PTR_ERR(dentry);
382 CERROR("cannot lookup __iopen__ directory: rc = %d\n", rc);
383 GOTO(err_objects, rc);
386 mds->mds_fid_de = dentry;
387 if (!dentry->d_inode || is_bad_inode(dentry->d_inode)) {
389 CERROR("__iopen__ directory has no inode? rc = %d\n", rc);
392 rc = mds_lov_init_objids(obd);
394 CERROR("cannot init lov objid rc = %d\n", rc);
398 rc = mds_lov_presetup(mds, lcfg);
400 GOTO(err_objects, rc);
402 /* Don't wait for mds_postrecov trying to clear orphans */
403 obd->obd_async_recov = 1;
404 rc = mds_postsetup(obd);
405 /* Bug 11557 - allow async abort_recov start
406 FIXME can remove most of this obd_async_recov plumbing
407 obd->obd_async_recov = 0;
411 GOTO(err_objects, rc);
414 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
417 dput(mds->mds_fid_de);
419 dput(mds->mds_objects_dir);
421 fsfilt_put_ops(obd->obd_fsops);
425 static int mds_cmd_cleanup(struct obd_device *obd)
427 struct mds_obd *mds = &obd->u.mds;
428 struct lvfs_run_ctxt saved;
432 mds->mds_lov_exp = NULL;
435 LCONSOLE_WARN("%s: shutting down for failover; client state "
436 "will be preserved.\n", obd->obd_name);
438 if (strncmp(obd->obd_name, MDD_OBD_NAME, strlen(MDD_OBD_NAME)))
441 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
443 mds_lov_destroy_objids(obd);
445 if (mds->mds_objects_dir != NULL) {
446 l_dput(mds->mds_objects_dir);
447 mds->mds_objects_dir = NULL;
450 dput(mds->mds_fid_de);
451 ll_vfs_dq_off(obd->u.obt.obt_sb, 0);
452 shrink_dcache_sb(mds->mds_obt.obt_sb);
453 fsfilt_put_ops(obd->obd_fsops);
455 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
460 static int mds_cmd_health_check(struct obd_device *obd)
465 static struct obd_ops mds_cmd_obd_ops = {
466 .o_owner = THIS_MODULE,
467 .o_setup = mds_cmd_setup,
468 .o_cleanup = mds_cmd_cleanup,
469 .o_precleanup = mds_precleanup,
470 .o_create = mds_obd_create,
471 .o_destroy = mds_obd_destroy,
472 .o_llog_init = mds_llog_init,
473 .o_llog_finish = mds_llog_finish,
474 .o_notify = mds_notify,
475 .o_postrecov = mds_postrecov,
476 // .o_health_check = mds_cmd_health_check,
479 quota_interface_t *mds_quota_interface_ref;
480 extern quota_interface_t mds_quota_interface;
482 static int __init mds_cmd_init(void)
484 struct lprocfs_static_vars lvars;
487 cfs_request_module("%s", "lquota");
488 mds_quota_interface_ref = PORTAL_SYMBOL_GET(mds_quota_interface);
489 rc = lquota_init(mds_quota_interface_ref);
491 if (mds_quota_interface_ref)
492 PORTAL_SYMBOL_PUT(mds_quota_interface);
495 init_obd_quota_ops(mds_quota_interface_ref, &mds_cmd_obd_ops);
497 lprocfs_mds_init_vars(&lvars);
498 class_register_type(&mds_cmd_obd_ops, NULL, lvars.module_vars,
499 LUSTRE_MDS_NAME, NULL);
504 static void /*__exit*/ mds_cmd_exit(void)
506 lquota_exit(mds_quota_interface_ref);
507 if (mds_quota_interface_ref)
508 PORTAL_SYMBOL_PUT(mds_quota_interface);
510 class_unregister_type(LUSTRE_MDS_NAME);
513 EXPORT_SYMBOL(mds_quota_interface_ref);
514 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
515 MODULE_DESCRIPTION("Lustre Metadata Server (MDS)");
516 MODULE_LICENSE("GPL");
518 module_init(mds_cmd_init);
519 module_exit(mds_cmd_exit);