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 [sun.com URL with a
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 2008 Sun Microsystems, Inc. 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>
49 #include <linux/random.h>
51 #include <linux/jbd.h>
52 #include <linux/smp_lock.h>
53 #include <linux/buffer_head.h>
54 #include <linux/workqueue.h>
55 #include <linux/mount.h>
57 #include <linux/lustre_acl.h>
58 #include <obd_class.h>
59 #include <lustre_dlm.h>
61 #include <lustre_fsfilt.h>
62 #include <lprocfs_status.h>
63 #include <lustre_commit_confd.h>
64 #include <lustre_quota.h>
65 #include <lustre_disk.h>
66 #include <lustre_param.h>
68 #include "mds_internal.h"
70 __u32 mds_max_ost_index=0xFFFF;
71 CFS_MODULE_PARM(mds_max_ost_index, "i", int, 0444,
74 /* Look up an entry by inode number. */
75 /* this function ONLY returns valid dget'd dentries with an initialized inode
77 struct dentry *mds_fid2dentry(struct mds_obd *mds, struct ll_fid *fid,
78 struct vfsmount **mnt)
81 unsigned long ino = fid->id;
82 __u32 generation = fid->generation;
84 struct dentry *result;
87 RETURN(ERR_PTR(-ESTALE));
89 snprintf(fid_name, sizeof(fid_name), "0x%lx", ino);
91 CDEBUG(D_DENTRY, "--> mds_fid2dentry: ino/gen %lu/%u, sb %p\n",
92 ino, generation, mds->mds_obt.obt_sb);
94 /* under ext3 this is neither supposed to return bad inodes
96 result = ll_lookup_one_len(fid_name, mds->mds_fid_de, strlen(fid_name));
100 inode = result->d_inode;
102 RETURN(ERR_PTR(-ENOENT));
104 if (inode->i_generation == 0 || inode->i_nlink == 0) {
105 LCONSOLE_WARN("Found inode with zero generation or link -- this"
106 " may indicate disk corruption (inode: %lu/%u, "
107 "link %lu, count %d)\n", inode->i_ino,
108 inode->i_generation,(unsigned long)inode->i_nlink,
109 atomic_read(&inode->i_count));
111 RETURN(ERR_PTR(-ENOENT));
114 if (generation && inode->i_generation != generation) {
115 /* we didn't find the right inode.. */
116 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
117 "count: %d, generation %u/%u\n", inode->i_ino,
118 (unsigned long)inode->i_nlink,
119 atomic_read(&inode->i_count), inode->i_generation,
122 RETURN(ERR_PTR(-ENOENT));
126 *mnt = mds->mds_vfsmnt;
133 static int mds_lov_presetup (struct mds_obd *mds, struct lustre_cfg *lcfg)
138 if (lcfg->lcfg_bufcount >= 4 && LUSTRE_CFG_BUFLEN(lcfg, 3) > 0) {
141 ll_generate_random_uuid(uuid);
142 class_uuid_unparse(uuid, &mds->mds_lov_uuid);
144 OBD_ALLOC(mds->mds_profile, LUSTRE_CFG_BUFLEN(lcfg, 3));
145 if (mds->mds_profile == NULL)
148 strncpy(mds->mds_profile, lustre_cfg_string(lcfg, 3),
149 LUSTRE_CFG_BUFLEN(lcfg, 3));
154 static int mds_lov_clean(struct obd_device *obd)
156 struct mds_obd *mds = &obd->u.mds;
157 struct obd_device *osc = mds->mds_osc_obd;
160 if (mds->mds_profile) {
161 class_del_profile(mds->mds_profile);
162 OBD_FREE(mds->mds_profile, strlen(mds->mds_profile) + 1);
163 mds->mds_profile = NULL;
166 /* There better be a lov */
170 RETURN(PTR_ERR(osc));
172 obd_register_observer(osc, NULL);
174 /* Give lov our same shutdown flags */
175 osc->obd_force = obd->obd_force;
176 osc->obd_fail = obd->obd_fail;
178 /* Cleanup the lov */
179 obd_disconnect(mds->mds_osc_exp);
180 class_manual_cleanup(osc);
185 static int mds_postsetup(struct obd_device *obd)
187 struct mds_obd *mds = &obd->u.mds;
191 rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd, 0, NULL,
196 rc = llog_setup(obd, &obd->obd_olg, LLOG_LOVEA_ORIG_CTXT, obd, 0, NULL,
201 if (mds->mds_profile) {
202 struct lustre_profile *lprof;
203 /* The profile defines which osc and mdc to connect to, for a
204 client. We reuse that here to figure out the name of the
205 lov to use (and ignore lprof->lp_md).
206 The profile was set in the config log with
207 LCFG_MOUNTOPT profilenm oscnm mdcnm */
208 lprof = class_get_profile(mds->mds_profile);
210 CERROR("No profile found: %s\n", mds->mds_profile);
211 GOTO(err_cleanup, rc = -ENOENT);
213 rc = mds_lov_connect(obd, lprof->lp_dt);
215 GOTO(err_cleanup, rc);
222 llog_cleanup(llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT));
223 llog_cleanup(llog_get_context(obd, LLOG_LOVEA_ORIG_CTXT));
227 int mds_postrecov(struct obd_device *obd)
235 LASSERT(!obd->obd_recovering);
236 LASSERT(!llog_ctxt_null(obd, LLOG_MDS_OST_ORIG_CTXT));
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_osc_obd, NULL,
250 obd->obd_async_recov ? OBD_NOTIFY_SYNC_NONBLOCK :
251 OBD_NOTIFY_SYNC, NULL);
254 lquota_recovery(mds_quota_interface_ref, obd);
259 /* We need to be able to stop an mds_lov_synchronize */
260 static int mds_lov_early_clean(struct obd_device *obd)
262 struct mds_obd *mds = &obd->u.mds;
263 struct obd_device *osc = mds->mds_osc_obd;
265 if (!osc || (!obd->obd_force && !obd->obd_fail))
268 CDEBUG(D_HA, "abort inflight\n");
269 return (obd_precleanup(osc, OBD_CLEANUP_EARLY));
272 static int mds_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
275 struct mds_obd *mds = &obd->u.mds;
279 case OBD_CLEANUP_EARLY:
281 case OBD_CLEANUP_EXPORTS:
282 mds_lov_early_clean(obd);
283 down_write(&mds->mds_notify_lock);
284 mds_lov_disconnect(obd);
286 llog_cleanup(llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT));
287 llog_cleanup(llog_get_context(obd, LLOG_LOVEA_ORIG_CTXT));
288 rc = obd_llog_finish(obd, 0);
289 mds->mds_osc_exp = NULL;
290 up_write(&mds->mds_notify_lock);
296 static struct dentry *mds_lvfs_fid2dentry(__u64 id, __u32 gen, __u64 gr,
299 struct obd_device *obd = data;
302 fid.generation = gen;
303 return mds_fid2dentry(&obd->u.mds, &fid, NULL);
307 struct lvfs_callback_ops mds_lvfs_ops = {
308 l_fid2dentry: mds_lvfs_fid2dentry,
311 quota_interface_t *mds_quota_interface_ref;
312 extern quota_interface_t mds_quota_interface;
314 static void mds_init_ctxt(struct obd_device *obd, struct vfsmount *mnt)
316 struct mds_obd *mds = &obd->u.mds;
318 mds->mds_vfsmnt = mnt;
319 /* why not mnt->mnt_sb instead of mnt->mnt_root->d_inode->i_sb? */
320 obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
322 fsfilt_setup(obd, obd->u.obt.obt_sb);
324 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
325 obd->obd_lvfs_ctxt.pwdmnt = mnt;
326 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
327 obd->obd_lvfs_ctxt.fs = get_ds();
328 obd->obd_lvfs_ctxt.cb_ops = mds_lvfs_ops;
332 /*mds still need lov setup here*/
333 static int mds_cmd_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
335 struct mds_obd *mds = &obd->u.mds;
336 struct lvfs_run_ctxt saved;
338 struct vfsmount *mnt;
339 struct lustre_sb_info *lsi;
340 struct lustre_mount_info *lmi;
341 struct dentry *dentry;
345 CDEBUG(D_INFO, "obd %s setup \n", obd->obd_name);
346 if (strncmp(obd->obd_name, MDD_OBD_NAME, strlen(MDD_OBD_NAME)))
349 if (lcfg->lcfg_bufcount < 5) {
350 CERROR("invalid arg for setup %s\n", MDD_OBD_NAME);
353 dev = lustre_cfg_string(lcfg, 4);
354 lmi = server_get_mount(dev);
355 LASSERT(lmi != NULL);
357 lsi = s2lsi(lmi->lmi_sb);
359 /* FIXME: MDD LOV initialize objects.
360 * we need only lmi here but not get mount
361 * OSD did mount already, so put mount back
363 atomic_dec(&lsi->lsi_mounts);
365 init_rwsem(&mds->mds_notify_lock);
367 obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
368 mds_init_ctxt(obd, mnt);
370 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
371 dentry = simple_mkdir(current->fs->pwd, mnt, "OBJECTS", 0777, 1);
372 if (IS_ERR(dentry)) {
373 rc = PTR_ERR(dentry);
374 CERROR("cannot create OBJECTS directory: rc = %d\n", rc);
377 mds->mds_objects_dir = dentry;
379 dentry = lookup_one_len("__iopen__", current->fs->pwd,
380 strlen("__iopen__"));
381 if (IS_ERR(dentry)) {
382 rc = PTR_ERR(dentry);
383 CERROR("cannot lookup __iopen__ directory: rc = %d\n", rc);
384 GOTO(err_objects, rc);
387 mds->mds_fid_de = dentry;
388 if (!dentry->d_inode || is_bad_inode(dentry->d_inode)) {
390 CERROR("__iopen__ directory has no inode? rc = %d\n", rc);
393 rc = mds_lov_init_objids(obd);
395 CERROR("cannot init lov objid rc = %d\n", rc);
399 rc = mds_lov_presetup(mds, lcfg);
401 GOTO(err_objects, rc);
403 /* Don't wait for mds_postrecov trying to clear orphans */
404 obd->obd_async_recov = 1;
405 rc = mds_postsetup(obd);
406 /* Bug 11557 - allow async abort_recov start
407 FIXME can remove most of this obd_async_recov plumbing
408 obd->obd_async_recov = 0;
412 GOTO(err_objects, rc);
414 mds->mds_max_mdsize = sizeof(struct lov_mds_md);
415 mds->mds_max_cookiesize = sizeof(struct llog_cookie);
418 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
421 dput(mds->mds_fid_de);
423 dput(mds->mds_objects_dir);
425 fsfilt_put_ops(obd->obd_fsops);
429 static int mds_cmd_cleanup(struct obd_device *obd)
431 struct mds_obd *mds = &obd->u.mds;
432 struct lvfs_run_ctxt saved;
436 mds->mds_osc_exp = NULL;
439 LCONSOLE_WARN("%s: shutting down for failover; client state "
440 "will be preserved.\n", obd->obd_name);
442 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
444 mds_lov_destroy_objids(obd);
446 if (mds->mds_objects_dir != NULL) {
447 l_dput(mds->mds_objects_dir);
448 mds->mds_objects_dir = NULL;
451 shrink_dcache_parent(mds->mds_fid_de);
452 dput(mds->mds_fid_de);
453 LL_DQUOT_OFF(obd->u.obt.obt_sb);
454 fsfilt_put_ops(obd->obd_fsops);
456 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
461 static int mds_cmd_health_check(struct obd_device *obd)
466 static struct obd_ops mds_cmd_obd_ops = {
467 .o_owner = THIS_MODULE,
468 .o_setup = mds_cmd_setup,
469 .o_cleanup = mds_cmd_cleanup,
470 .o_precleanup = mds_precleanup,
471 .o_create = mds_obd_create,
472 .o_destroy = mds_obd_destroy,
473 .o_llog_init = mds_llog_init,
474 .o_llog_finish = mds_llog_finish,
475 .o_notify = mds_notify,
476 .o_postrecov = mds_postrecov,
477 // .o_health_check = mds_cmd_health_check,
480 static int __init mds_cmd_init(void)
482 struct lprocfs_static_vars lvars;
484 lprocfs_mds_init_vars(&lvars);
485 class_register_type(&mds_cmd_obd_ops, NULL, lvars.module_vars,
486 LUSTRE_MDS_NAME, NULL);
491 static void /*__exit*/ mds_cmd_exit(void)
493 class_unregister_type(LUSTRE_MDS_NAME);
496 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
497 MODULE_DESCRIPTION("Lustre Metadata Server (MDS)");
498 MODULE_LICENSE("GPL");
500 module_init(mds_cmd_init);
501 module_exit(mds_cmd_exit);