lustre/cmm/cmm_device.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   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.
  11  *
  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).
  17  *
  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
  21  *
  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
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/cmm/cmm_device.c
  37  *
  38  * Lustre Cluster Metadata Manager (cmm)
  39  *
  40  * Author: Mike Pershin <tappro@clusterfs.com>
  41  */
  42
  43 #ifndef EXPORT_SYMTAB
  44 # define EXPORT_SYMTAB
  45 #endif
  46 #define DEBUG_SUBSYSTEM S_MDS
  47
  48 #include <linux/module.h>
  49
  50 #include <obd.h>
  51 #include <obd_class.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_ver.h>
  54 #include "cmm_internal.h"
  55 #include "mdc_internal.h"
  56 #ifdef HAVE_QUOTA_SUPPORT
  57 # include <lustre_quota.h>
  58 #endif
  59
  60 struct obd_ops cmm_obd_device_ops = {
  61         .o_owner           = THIS_MODULE
  62 };
  63
  64 static const struct lu_device_operations cmm_lu_ops;
  65
  66 static inline int lu_device_is_cmm(struct lu_device *d)
  67 {
  68         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &cmm_lu_ops);
  69 }
  70
  71 int cmm_root_get(const struct lu_env *env, struct md_device *md,
  72                  struct lu_fid *fid)
  73 {
  74         struct cmm_device *cmm_dev = md2cmm_dev(md);
  75         /* valid only on master MDS */
  76         if (cmm_dev->cmm_local_num == 0)
  77                 return cmm_child_ops(cmm_dev)->mdo_root_get(env,
  78                                      cmm_dev->cmm_child, fid);
  79         else
  80                 return -EINVAL;
  81 }
  82
  83 static int cmm_statfs(const struct lu_env *env, struct md_device *md,
  84                       struct kstatfs *sfs)
  85 {
  86         struct cmm_device *cmm_dev = md2cmm_dev(md);
  87         int rc;
  88
  89         ENTRY;
  90         rc = cmm_child_ops(cmm_dev)->mdo_statfs(env,
  91                                                 cmm_dev->cmm_child, sfs);
  92         RETURN (rc);
  93 }
  94
  95 static int cmm_maxsize_get(const struct lu_env *env, struct md_device *md,
  96                            int *md_size, int *cookie_size)
  97 {
  98         struct cmm_device *cmm_dev = md2cmm_dev(md);
  99         int rc;
 100         ENTRY;
 101         rc = cmm_child_ops(cmm_dev)->mdo_maxsize_get(env, cmm_dev->cmm_child,
 102                                                      md_size, cookie_size);
 103         RETURN(rc);
 104 }
 105
 106 static int cmm_init_capa_ctxt(const struct lu_env *env, struct md_device *md,
 107                               int mode , unsigned long timeout, __u32 alg,
 108                               struct lustre_capa_key *keys)
 109 {
 110         struct cmm_device *cmm_dev = md2cmm_dev(md);
 111         int rc;
 112         ENTRY;
 113         LASSERT(cmm_child_ops(cmm_dev)->mdo_init_capa_ctxt);
 114         rc = cmm_child_ops(cmm_dev)->mdo_init_capa_ctxt(env, cmm_dev->cmm_child,
 115                                                         mode, timeout, alg,
 116                                                         keys);
 117         RETURN(rc);
 118 }
 119
 120 static int cmm_update_capa_key(const struct lu_env *env,
 121                                struct md_device *md,
 122                                struct lustre_capa_key *key)
 123 {
 124         struct cmm_device *cmm_dev = md2cmm_dev(md);
 125         int rc;
 126         ENTRY;
 127         rc = cmm_child_ops(cmm_dev)->mdo_update_capa_key(env,
 128                                                          cmm_dev->cmm_child,
 129                                                          key);
 130         RETURN(rc);
 131 }
 132
 133 static int cmm_llog_ctxt_get(const struct lu_env *env, struct md_device *m,
 134                              int idx, void **h)
 135 {
 136         struct cmm_device *cmm_dev = md2cmm_dev(m);
 137         int rc;
 138         ENTRY;
 139
 140         rc = cmm_child_ops(cmm_dev)->mdo_llog_ctxt_get(env, cmm_dev->cmm_child,
 141                                                        idx, h);
 142         RETURN(rc);
 143 }
 144
 145 #ifdef HAVE_QUOTA_SUPPORT
 146 static int cmm_quota_notify(const struct lu_env *env, struct md_device *m)
 147 {
 148         struct cmm_device *cmm_dev = md2cmm_dev(m);
 149         int rc;
 150         ENTRY;
 151
 152         rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_notify(env,
 153                                                           cmm_dev->cmm_child);
 154         RETURN(rc);
 155 }
 156
 157 static int cmm_quota_setup(const struct lu_env *env, struct md_device *m,
 158                            void *data)
 159 {
 160         struct cmm_device *cmm_dev = md2cmm_dev(m);
 161         int rc;
 162         ENTRY;
 163
 164         rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_setup(env,
 165                                                          cmm_dev->cmm_child,
 166                                                          data);
 167         RETURN(rc);
 168 }
 169
 170 static int cmm_quota_cleanup(const struct lu_env *env, struct md_device *m)
 171 {
 172         struct cmm_device *cmm_dev = md2cmm_dev(m);
 173         int rc;
 174         ENTRY;
 175
 176         rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_cleanup(env,
 177                                                            cmm_dev->cmm_child);
 178         RETURN(rc);
 179 }
 180
 181 static int cmm_quota_recovery(const struct lu_env *env, struct md_device *m)
 182 {
 183         struct cmm_device *cmm_dev = md2cmm_dev(m);
 184         int rc;
 185         ENTRY;
 186
 187         rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_recovery(env,
 188                                                             cmm_dev->cmm_child);
 189         RETURN(rc);
 190 }
 191
 192 static int cmm_quota_check(const struct lu_env *env, struct md_device *m,
 193                            struct obd_export *exp, __u32 type)
 194 {
 195         struct cmm_device *cmm_dev = md2cmm_dev(m);
 196         int rc;
 197         ENTRY;
 198
 199         /* disable quota for CMD case temporary. */
 200         if (cmm_dev->cmm_tgt_count)
 201                 RETURN(-EOPNOTSUPP);
 202
 203         rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_check(env,
 204                                                          cmm_dev->cmm_child,
 205                                                          exp, type);
 206         RETURN(rc);
 207 }
 208
 209 static int cmm_quota_on(const struct lu_env *env, struct md_device *m,
 210                         __u32 type)
 211 {
 212         struct cmm_device *cmm_dev = md2cmm_dev(m);
 213         int rc;
 214         ENTRY;
 215
 216         /* disable quota for CMD case temporary. */
 217         if (cmm_dev->cmm_tgt_count)
 218                 RETURN(-EOPNOTSUPP);
 219
 220         rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_on(env,
 221                                                       cmm_dev->cmm_child,
 222                                                       type);
 223         RETURN(rc);
 224 }
 225
 226 static int cmm_quota_off(const struct lu_env *env, struct md_device *m,
 227                          __u32 type)
 228 {
 229         struct cmm_device *cmm_dev = md2cmm_dev(m);
 230         int rc;
 231         ENTRY;
 232
 233         rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_off(env,
 234                                                        cmm_dev->cmm_child,
 235                                                        type);
 236         RETURN(rc);
 237 }
 238
 239 static int cmm_quota_setinfo(const struct lu_env *env, struct md_device *m,
 240                              __u32 type, __u32 id, struct obd_dqinfo *dqinfo)
 241 {
 242         struct cmm_device *cmm_dev = md2cmm_dev(m);
 243         int rc;
 244         ENTRY;
 245
 246         /* disable quota for CMD case temporary. */
 247         if (cmm_dev->cmm_tgt_count)
 248                 RETURN(-EOPNOTSUPP);
 249
 250         rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_setinfo(env,
 251                                                            cmm_dev->cmm_child,
 252                                                            type, id, dqinfo);
 253         RETURN(rc);
 254 }
 255
 256 static int cmm_quota_getinfo(const struct lu_env *env,
 257                              const struct md_device *m,
 258                              __u32 type, __u32 id, struct obd_dqinfo *dqinfo)
 259 {
 260         struct cmm_device *cmm_dev = md2cmm_dev((struct md_device *)m);
 261         int rc;
 262         ENTRY;
 263
 264         /* disable quota for CMD case temporary. */
 265         if (cmm_dev->cmm_tgt_count)
 266                 RETURN(-EOPNOTSUPP);
 267
 268         rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_getinfo(env,
 269                                                            cmm_dev->cmm_child,
 270                                                            type, id, dqinfo);
 271         RETURN(rc);
 272 }
 273
 274 static int cmm_quota_setquota(const struct lu_env *env, struct md_device *m,
 275                               __u32 type, __u32 id, struct obd_dqblk *dqblk)
 276 {
 277         struct cmm_device *cmm_dev = md2cmm_dev(m);
 278         int rc;
 279         ENTRY;
 280
 281         /* disable quota for CMD case temporary. */
 282         if (cmm_dev->cmm_tgt_count)
 283                 RETURN(-EOPNOTSUPP);
 284
 285         rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_setquota(env,
 286                                                             cmm_dev->cmm_child,
 287                                                             type, id, dqblk);
 288         RETURN(rc);
 289 }
 290
 291 static int cmm_quota_getquota(const struct lu_env *env,
 292                               const struct md_device *m,
 293                               __u32 type, __u32 id, struct obd_dqblk *dqblk)
 294 {
 295         struct cmm_device *cmm_dev = md2cmm_dev((struct md_device *)m);
 296         int rc;
 297         ENTRY;
 298
 299         /* disable quota for CMD case temporary. */
 300         if (cmm_dev->cmm_tgt_count)
 301                 RETURN(-EOPNOTSUPP);
 302
 303         rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_getquota(env,
 304                                                             cmm_dev->cmm_child,
 305                                                             type, id, dqblk);
 306         RETURN(rc);
 307 }
 308
 309 static int cmm_quota_getoinfo(const struct lu_env *env,
 310                               const struct md_device *m,
 311                               __u32 type, __u32 id, struct obd_dqinfo *dqinfo)
 312 {
 313         struct cmm_device *cmm_dev = md2cmm_dev((struct md_device *)m);
 314         int rc;
 315         ENTRY;
 316
 317         /* disable quota for CMD case temporary. */
 318         if (cmm_dev->cmm_tgt_count)
 319                 RETURN(-EOPNOTSUPP);
 320
 321         rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_getoinfo(env,
 322                                                             cmm_dev->cmm_child,
 323                                                             type, id, dqinfo);
 324         RETURN(rc);
 325 }
 326
 327 static int cmm_quota_getoquota(const struct lu_env *env,
 328                                const struct md_device *m,
 329                                __u32 type, __u32 id, struct obd_dqblk *dqblk)
 330 {
 331         struct cmm_device *cmm_dev = md2cmm_dev((struct md_device *)m);
 332         int rc;
 333         ENTRY;
 334
 335         /* disable quota for CMD case temporary. */
 336         if (cmm_dev->cmm_tgt_count)
 337                 RETURN(-EOPNOTSUPP);
 338
 339         rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_getoquota(env,
 340                                                              cmm_dev->cmm_child,
 341                                                              type, id, dqblk);
 342         RETURN(rc);
 343 }
 344
 345 static int cmm_quota_invalidate(const struct lu_env *env, struct md_device *m,
 346                                 __u32 type)
 347 {
 348         struct cmm_device *cmm_dev = md2cmm_dev(m);
 349         int rc;
 350         ENTRY;
 351
 352         /* disable quota for CMD case temporary. */
 353         if (cmm_dev->cmm_tgt_count)
 354                 RETURN(-EOPNOTSUPP);
 355
 356         rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_invalidate(env,
 357                                                               cmm_dev->cmm_child,
 358                                                               type);
 359         RETURN(rc);
 360 }
 361
 362 static int cmm_quota_finvalidate(const struct lu_env *env, struct md_device *m,
 363                                  __u32 type)
 364 {
 365         struct cmm_device *cmm_dev = md2cmm_dev(m);
 366         int rc;
 367         ENTRY;
 368
 369         /* disable quota for CMD case temporary. */
 370         if (cmm_dev->cmm_tgt_count)
 371                 RETURN(-EOPNOTSUPP);
 372
 373         rc = cmm_child_ops(cmm_dev)->mdo_quota.mqo_finvalidate(env,
 374                                                                cmm_dev->cmm_child,
 375                                                                type);
 376         RETURN(rc);
 377 }
 378 #endif
 379
 380 int cmm_iocontrol(const struct lu_env *env, struct md_device *m,
 381                   unsigned int cmd, int len, void *data)
 382 {
 383         struct md_device *next = md2cmm_dev(m)->cmm_child;
 384         int rc;
 385
 386         ENTRY;
 387         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
 388         RETURN(rc);
 389 }
 390
 391
 392 static const struct md_device_operations cmm_md_ops = {
 393         .mdo_statfs          = cmm_statfs,
 394         .mdo_root_get        = cmm_root_get,
 395         .mdo_maxsize_get     = cmm_maxsize_get,
 396         .mdo_init_capa_ctxt  = cmm_init_capa_ctxt,
 397         .mdo_update_capa_key = cmm_update_capa_key,
 398         .mdo_llog_ctxt_get   = cmm_llog_ctxt_get,
 399         .mdo_iocontrol       = cmm_iocontrol,
 400 #ifdef HAVE_QUOTA_SUPPORT
 401         .mdo_quota           = {
 402                 .mqo_notify      = cmm_quota_notify,
 403                 .mqo_setup       = cmm_quota_setup,
 404                 .mqo_cleanup     = cmm_quota_cleanup,
 405                 .mqo_recovery    = cmm_quota_recovery,
 406                 .mqo_check       = cmm_quota_check,
 407                 .mqo_on          = cmm_quota_on,
 408                 .mqo_off         = cmm_quota_off,
 409                 .mqo_setinfo     = cmm_quota_setinfo,
 410                 .mqo_getinfo     = cmm_quota_getinfo,
 411                 .mqo_setquota    = cmm_quota_setquota,
 412                 .mqo_getquota    = cmm_quota_getquota,
 413                 .mqo_getoinfo    = cmm_quota_getoinfo,
 414                 .mqo_getoquota   = cmm_quota_getoquota,
 415                 .mqo_invalidate  = cmm_quota_invalidate,
 416                 .mqo_finvalidate = cmm_quota_finvalidate
 417         }
 418 #endif
 419 };
 420
 421 extern struct lu_device_type mdc_device_type;
 422
 423 static int cmm_post_init_mdc(const struct lu_env *env,
 424                              struct cmm_device *cmm)
 425 {
 426         int max_mdsize, max_cookiesize, rc;
 427         struct mdc_device *mc, *tmp;
 428
 429         /* get the max mdsize and cookiesize from lower layer */
 430         rc = cmm_maxsize_get(env, &cmm->cmm_md_dev, &max_mdsize,
 431                                                 &max_cookiesize);
 432         if (rc)
 433                 RETURN(rc);
 434
 435         spin_lock(&cmm->cmm_tgt_guard);
 436         list_for_each_entry_safe(mc, tmp, &cmm->cmm_targets,
 437                                  mc_linkage) {
 438                 cmm_mdc_init_ea_size(env, mc, max_mdsize, max_cookiesize);
 439         }
 440         spin_unlock(&cmm->cmm_tgt_guard);
 441         RETURN(rc);
 442 }
 443
 444 /* --- cmm_lu_operations --- */
 445 /* add new MDC to the CMM, create MDC lu_device and connect it to mdc_obd */
 446 static int cmm_add_mdc(const struct lu_env *env,
 447                        struct cmm_device *cm, struct lustre_cfg *cfg)
 448 {
 449         struct lu_device_type *ldt = &mdc_device_type;
 450         char *p, *num = lustre_cfg_string(cfg, 2);
 451         struct mdc_device *mc, *tmp;
 452         struct lu_fld_target target;
 453         struct lu_device *ld;
 454         struct lu_device *cmm_lu = cmm2lu_dev(cm);
 455         mdsno_t mdc_num;
 456         struct lu_site *site = cmm2lu_dev(cm)->ld_site;
 457         int rc;
 458 #ifdef HAVE_QUOTA_SUPPORT
 459         int first;
 460 #endif
 461         ENTRY;
 462
 463         /* find out that there is no such mdc */
 464         LASSERT(num);
 465         mdc_num = simple_strtol(num, &p, 10);
 466         if (*p) {
 467                 CERROR("Invalid index in lustre_cgf, offset 2\n");
 468                 RETURN(-EINVAL);
 469         }
 470
 471         spin_lock(&cm->cmm_tgt_guard);
 472         list_for_each_entry_safe(mc, tmp, &cm->cmm_targets,
 473                                  mc_linkage) {
 474                 if (mc->mc_num == mdc_num) {
 475                         spin_unlock(&cm->cmm_tgt_guard);
 476                         RETURN(-EEXIST);
 477                 }
 478         }
 479         spin_unlock(&cm->cmm_tgt_guard);
 480         ld = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
 481         if (IS_ERR(ld))
 482                 RETURN(PTR_ERR(ld));
 483
 484         ld->ld_site = site;
 485
 486         rc = ldt->ldt_ops->ldto_device_init(env, ld, NULL, NULL);
 487         if (rc) {
 488                 ldt->ldt_ops->ldto_device_free(env, ld);
 489                 RETURN(rc);
 490         }
 491         /* pass config to the just created MDC */
 492         rc = ld->ld_ops->ldo_process_config(env, ld, cfg);
 493         if (rc) {
 494                 ldt->ldt_ops->ldto_device_fini(env, ld);
 495                 ldt->ldt_ops->ldto_device_free(env, ld);
 496                 RETURN(rc);
 497         }
 498
 499         spin_lock(&cm->cmm_tgt_guard);
 500         list_for_each_entry_safe(mc, tmp, &cm->cmm_targets,
 501                                  mc_linkage) {
 502                 if (mc->mc_num == mdc_num) {
 503                         spin_unlock(&cm->cmm_tgt_guard);
 504                         ldt->ldt_ops->ldto_device_fini(env, ld);
 505                         ldt->ldt_ops->ldto_device_free(env, ld);
 506                         RETURN(-EEXIST);
 507                 }
 508         }
 509         mc = lu2mdc_dev(ld);
 510         list_add_tail(&mc->mc_linkage, &cm->cmm_targets);
 511         cm->cmm_tgt_count++;
 512 #ifdef HAVE_QUOTA_SUPPORT
 513         first = cm->cmm_tgt_count;
 514 #endif
 515         spin_unlock(&cm->cmm_tgt_guard);
 516
 517         lu_device_get(cmm_lu);
 518         lu_ref_add(&cmm_lu->ld_reference, "mdc-child", ld);
 519
 520         target.ft_srv = NULL;
 521         target.ft_idx = mc->mc_num;
 522         target.ft_exp = mc->mc_desc.cl_exp;
 523         fld_client_add_target(cm->cmm_fld, &target);
 524
 525         if (mc->mc_num == 0) {
 526                 /* this is mdt0 -> mc export, fld lookup need this export
 527                    to forward fld lookup request. */
 528                 LASSERT(!lu_site2md(site)->ms_server_fld->lsf_control_exp);
 529                 lu_site2md(site)->ms_server_fld->lsf_control_exp =
 530                                           mc->mc_desc.cl_exp;
 531         }
 532 #ifdef HAVE_QUOTA_SUPPORT
 533         /* XXX: Disable quota for CMD case temporary. */
 534         if (first == 1) {
 535                 CWARN("Disable quota for CMD case temporary!\n");
 536                 cmm_child_ops(cm)->mdo_quota.mqo_off(env, cm->cmm_child, UGQUOTA);
 537         }
 538 #endif
 539         /* Set max md size for the mdc. */
 540         rc = cmm_post_init_mdc(env, cm);
 541         RETURN(rc);
 542 }
 543
 544 static void cmm_device_shutdown(const struct lu_env *env,
 545                                 struct cmm_device *cm,
 546                                 struct lustre_cfg *cfg)
 547 {
 548         struct mdc_device *mc, *tmp;
 549         ENTRY;
 550
 551         /* Remove local target from FLD. */
 552         fld_client_del_target(cm->cmm_fld, cm->cmm_local_num);
 553
 554         /* Finish all mdc devices. */
 555         spin_lock(&cm->cmm_tgt_guard);
 556         list_for_each_entry_safe(mc, tmp, &cm->cmm_targets, mc_linkage) {
 557                 struct lu_device *ld_m = mdc2lu_dev(mc);
 558                 fld_client_del_target(cm->cmm_fld, mc->mc_num);
 559                 ld_m->ld_ops->ldo_process_config(env, ld_m, cfg);
 560         }
 561         spin_unlock(&cm->cmm_tgt_guard);
 562
 563         /* remove upcall device*/
 564         md_upcall_fini(&cm->cmm_md_dev);
 565
 566         EXIT;
 567 }
 568
 569 static int cmm_device_mount(const struct lu_env *env,
 570                             struct cmm_device *m, struct lustre_cfg *cfg)
 571 {
 572         const char *index = lustre_cfg_string(cfg, 2);
 573         char *p;
 574
 575         LASSERT(index != NULL);
 576
 577         m->cmm_local_num = simple_strtol(index, &p, 10);
 578         if (*p) {
 579                 CERROR("Invalid index in lustre_cgf\n");
 580                 RETURN(-EINVAL);
 581         }
 582
 583         RETURN(0);
 584 }
 585
 586 static int cmm_process_config(const struct lu_env *env,
 587                               struct lu_device *d, struct lustre_cfg *cfg)
 588 {
 589         struct cmm_device *m = lu2cmm_dev(d);
 590         struct lu_device *next = md2lu_dev(m->cmm_child);
 591         int err;
 592         ENTRY;
 593
 594         switch(cfg->lcfg_command) {
 595         case LCFG_ADD_MDC:
 596                 /* On first ADD_MDC add also local target. */
 597                 if (!(m->cmm_flags & CMM_INITIALIZED)) {
 598                         struct lu_site *ls = cmm2lu_dev(m)->ld_site;
 599                         struct lu_fld_target target;
 600
 601                         target.ft_srv = lu_site2md(ls)->ms_server_fld;
 602                         target.ft_idx = m->cmm_local_num;
 603                         target.ft_exp = NULL;
 604
 605                         fld_client_add_target(m->cmm_fld, &target);
 606                 }
 607                 err = cmm_add_mdc(env, m, cfg);
 608
 609                 /* The first ADD_MDC can be counted as setup is finished. */
 610                 if (!(m->cmm_flags & CMM_INITIALIZED))
 611                         m->cmm_flags |= CMM_INITIALIZED;
 612
 613                 break;
 614         case LCFG_SETUP:
 615         {
 616                 /* lower layers should be set up at first */
 617                 err = next->ld_ops->ldo_process_config(env, next, cfg);
 618                 if (err == 0)
 619                         err = cmm_device_mount(env, m, cfg);
 620                 break;
 621         }
 622         case LCFG_CLEANUP:
 623         {
 624                 cmm_device_shutdown(env, m, cfg);
 625         }
 626         default:
 627                 err = next->ld_ops->ldo_process_config(env, next, cfg);
 628         }
 629         RETURN(err);
 630 }
 631
 632 static int cmm_recovery_complete(const struct lu_env *env,
 633                                  struct lu_device *d)
 634 {
 635         struct cmm_device *m = lu2cmm_dev(d);
 636         struct lu_device *next = md2lu_dev(m->cmm_child);
 637         int rc;
 638         ENTRY;
 639         rc = next->ld_ops->ldo_recovery_complete(env, next);
 640         RETURN(rc);
 641 }
 642
 643 static int cmm_prepare(const struct lu_env *env,
 644                        struct lu_device *pdev,
 645                        struct lu_device *dev)
 646 {
 647         struct cmm_device *cmm = lu2cmm_dev(dev);
 648         struct lu_device *next = md2lu_dev(cmm->cmm_child);
 649         int rc;
 650
 651         ENTRY;
 652         rc = next->ld_ops->ldo_prepare(env, dev, next);
 653         RETURN(rc);
 654 }
 655
 656 static const struct lu_device_operations cmm_lu_ops = {
 657         .ldo_object_alloc      = cmm_object_alloc,
 658         .ldo_process_config    = cmm_process_config,
 659         .ldo_recovery_complete = cmm_recovery_complete,
 660         .ldo_prepare           = cmm_prepare,
 661 };
 662
 663 /* --- lu_device_type operations --- */
 664 int cmm_upcall(const struct lu_env *env, struct md_device *md,
 665                enum md_upcall_event ev, void *data)
 666 {
 667         int rc;
 668         ENTRY;
 669
 670         switch (ev) {
 671                 case MD_LOV_SYNC:
 672                         rc = cmm_post_init_mdc(env, md2cmm_dev(md));
 673                         if (rc)
 674                                 CERROR("can not init md size %d\n", rc);
 675                         /* fall through */
 676                 default:
 677                         rc = md_do_upcall(env, md, ev, data);
 678         }
 679         RETURN(rc);
 680 }
 681
 682 static struct lu_device *cmm_device_free(const struct lu_env *env,
 683                                          struct lu_device *d)
 684 {
 685         struct cmm_device *m = lu2cmm_dev(d);
 686         struct lu_device  *next = md2lu_dev(m->cmm_child);
 687         ENTRY;
 688
 689         LASSERT(m->cmm_tgt_count == 0);
 690         LASSERT(list_empty(&m->cmm_targets));
 691         if (m->cmm_fld != NULL) {
 692                 OBD_FREE_PTR(m->cmm_fld);
 693                 m->cmm_fld = NULL;
 694         }
 695         md_device_fini(&m->cmm_md_dev);
 696         OBD_FREE_PTR(m);
 697         RETURN(next);
 698 }
 699
 700 static struct lu_device *cmm_device_alloc(const struct lu_env *env,
 701                                           struct lu_device_type *t,
 702                                           struct lustre_cfg *cfg)
 703 {
 704         struct lu_device  *l;
 705         struct cmm_device *m;
 706         ENTRY;
 707
 708         OBD_ALLOC_PTR(m);
 709         if (m == NULL) {
 710                 l = ERR_PTR(-ENOMEM);
 711         } else {
 712                 md_device_init(&m->cmm_md_dev, t);
 713                 m->cmm_md_dev.md_ops = &cmm_md_ops;
 714                 md_upcall_init(&m->cmm_md_dev, cmm_upcall);
 715                 l = cmm2lu_dev(m);
 716                 l->ld_ops = &cmm_lu_ops;
 717
 718                 OBD_ALLOC_PTR(m->cmm_fld);
 719                 if (!m->cmm_fld) {
 720                         cmm_device_free(env, l);
 721                         l = ERR_PTR(-ENOMEM);
 722                 }
 723         }
 724         RETURN(l);
 725 }
 726
 727 /* context key constructor/destructor: cmm_key_init, cmm_key_fini */
 728 LU_KEY_INIT_FINI(cmm, struct cmm_thread_info);
 729
 730 /* context key: cmm_thread_key */
 731 LU_CONTEXT_KEY_DEFINE(cmm, LCT_MD_THREAD);
 732
 733 struct cmm_thread_info *cmm_env_info(const struct lu_env *env)
 734 {
 735         struct cmm_thread_info *info;
 736
 737         info = lu_context_key_get(&env->le_ctx, &cmm_thread_key);
 738         LASSERT(info != NULL);
 739         return info;
 740 }
 741
 742 /* type constructor/destructor: cmm_type_init/cmm_type_fini */
 743 LU_TYPE_INIT_FINI(cmm, &cmm_thread_key);
 744
 745 /*
 746  * Kludge code : it should be moved mdc_device.c if mdc_(mds)_device
 747  * is really stacked.
 748  */
 749 static int __cmm_type_init(struct lu_device_type *t)
 750 {
 751         int rc;
 752         rc = lu_device_type_init(&mdc_device_type);
 753         if (rc == 0) {
 754                 rc = cmm_type_init(t);
 755                 if (rc)
 756                         lu_device_type_fini(&mdc_device_type);
 757         }
 758         return rc;
 759 }
 760
 761 static void __cmm_type_fini(struct lu_device_type *t)
 762 {
 763         lu_device_type_fini(&mdc_device_type);
 764         cmm_type_fini(t);
 765 }
 766
 767 static void __cmm_type_start(struct lu_device_type *t)
 768 {
 769         mdc_device_type.ldt_ops->ldto_start(&mdc_device_type);
 770         cmm_type_start(t);
 771 }
 772
 773 static void __cmm_type_stop(struct lu_device_type *t)
 774 {
 775         mdc_device_type.ldt_ops->ldto_stop(&mdc_device_type);
 776         cmm_type_stop(t);
 777 }
 778
 779 static int cmm_device_init(const struct lu_env *env, struct lu_device *d,
 780                            const char *name, struct lu_device *next)
 781 {
 782         struct cmm_device *m = lu2cmm_dev(d);
 783         struct lu_site *ls;
 784         int err = 0;
 785         ENTRY;
 786
 787         spin_lock_init(&m->cmm_tgt_guard);
 788         CFS_INIT_LIST_HEAD(&m->cmm_targets);
 789         m->cmm_tgt_count = 0;
 790         m->cmm_child = lu2md_dev(next);
 791
 792         err = fld_client_init(m->cmm_fld, name,
 793                               LUSTRE_CLI_FLD_HASH_DHT);
 794         if (err) {
 795                 CERROR("Can't init FLD, err %d\n", err);
 796                 RETURN(err);
 797         }
 798
 799         /* Assign site's fld client ref, needed for asserts in osd. */
 800         ls = cmm2lu_dev(m)->ld_site;
 801         lu_site2md(ls)->ms_client_fld = m->cmm_fld;
 802         err = cmm_procfs_init(m, name);
 803
 804         RETURN(err);
 805 }
 806
 807 static struct lu_device *cmm_device_fini(const struct lu_env *env,
 808                                          struct lu_device *ld)
 809 {
 810         struct cmm_device *cm = lu2cmm_dev(ld);
 811         struct mdc_device *mc, *tmp;
 812         struct lu_site *ls;
 813         ENTRY;
 814
 815         /* Finish all mdc devices */
 816         spin_lock(&cm->cmm_tgt_guard);
 817         list_for_each_entry_safe(mc, tmp, &cm->cmm_targets, mc_linkage) {
 818                 struct lu_device *ld_m = mdc2lu_dev(mc);
 819                 struct lu_device *ld_c = cmm2lu_dev(cm);
 820
 821                 list_del_init(&mc->mc_linkage);
 822                 lu_ref_del(&ld_c->ld_reference, "mdc-child", ld_m);
 823                 lu_device_put(ld_c);
 824                 ld_m->ld_type->ldt_ops->ldto_device_fini(env, ld_m);
 825                 ld_m->ld_type->ldt_ops->ldto_device_free(env, ld_m);
 826                 cm->cmm_tgt_count--;
 827         }
 828         spin_unlock(&cm->cmm_tgt_guard);
 829
 830         fld_client_fini(cm->cmm_fld);
 831         ls = cmm2lu_dev(cm)->ld_site;
 832         lu_site2md(ls)->ms_client_fld = NULL;
 833         cmm_procfs_fini(cm);
 834
 835         RETURN (md2lu_dev(cm->cmm_child));
 836 }
 837
 838 static struct lu_device_type_operations cmm_device_type_ops = {
 839         .ldto_init = __cmm_type_init,
 840         .ldto_fini = __cmm_type_fini,
 841
 842         .ldto_start = __cmm_type_start,
 843         .ldto_stop  = __cmm_type_stop,
 844
 845         .ldto_device_alloc = cmm_device_alloc,
 846         .ldto_device_free  = cmm_device_free,
 847
 848         .ldto_device_init = cmm_device_init,
 849         .ldto_device_fini = cmm_device_fini
 850 };
 851
 852 static struct lu_device_type cmm_device_type = {
 853         .ldt_tags     = LU_DEVICE_MD,
 854         .ldt_name     = LUSTRE_CMM_NAME,
 855         .ldt_ops      = &cmm_device_type_ops,
 856         .ldt_ctx_tags = LCT_MD_THREAD | LCT_DT_THREAD
 857 };
 858
 859 struct lprocfs_vars lprocfs_cmm_obd_vars[] = {
 860         { 0 }
 861 };
 862
 863 struct lprocfs_vars lprocfs_cmm_module_vars[] = {
 864         { 0 }
 865 };
 866
 867 static void lprocfs_cmm_init_vars(struct lprocfs_static_vars *lvars)
 868 {
 869     lvars->module_vars  = lprocfs_cmm_module_vars;
 870     lvars->obd_vars     = lprocfs_cmm_obd_vars;
 871 }
 872
 873 static int __init cmm_mod_init(void)
 874 {
 875         struct lprocfs_static_vars lvars;
 876
 877         lprocfs_cmm_init_vars(&lvars);
 878         return class_register_type(&cmm_obd_device_ops, NULL, lvars.module_vars,
 879                                    LUSTRE_CMM_NAME, &cmm_device_type);
 880 }
 881
 882 static void __exit cmm_mod_exit(void)
 883 {
 884         class_unregister_type(LUSTRE_CMM_NAME);
 885 }
 886
 887 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
 888 MODULE_DESCRIPTION("Lustre Clustered Metadata Manager ("LUSTRE_CMM_NAME")");
 889 MODULE_LICENSE("GPL");
 890
 891 cfs_module(cmm, "0.1.0", cmm_mod_init, cmm_mod_exit);