lustre/obdclass/obd_config.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 (c) 2003, 2010, Oracle and/or its affiliates. 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/obdclass/obd_config.c
  37  *
  38  * Config API
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_CLASS
  42 #ifdef __KERNEL__
  43 #include <obd_class.h>
  44 #include <linux/string.h>
  45 #else
  46 #include <liblustre.h>
  47 #include <obd_class.h>
  48 #include <obd.h>
  49 #endif
  50 #include <lustre_log.h>
  51 #include <lprocfs_status.h>
  52 #include <libcfs/list.h>
  53 #include <lustre_param.h>
  54
  55 static cfs_hash_ops_t uuid_hash_ops;
  56 static cfs_hash_ops_t nid_hash_ops;
  57 static cfs_hash_ops_t nid_stat_hash_ops;
  58
  59 /*********** string parsing utils *********/
  60
  61 /* returns 0 if we find this key in the buffer, else 1 */
  62 int class_find_param(char *buf, char *key, char **valp)
  63 {
  64         char *ptr;
  65
  66         if (!buf)
  67                 return 1;
  68
  69         if ((ptr = strstr(buf, key)) == NULL)
  70                 return 1;
  71
  72         if (valp)
  73                 *valp = ptr + strlen(key);
  74
  75         return 0;
  76 }
  77
  78 /**
  79  * Finds a parameter in \a params and copies it to \a copy.
  80  *
  81  * Leading spaces are skipped. Next space or end of string is the
  82  * parameter terminator with the exception that spaces inside single or double
  83  * quotes get included into a parameter. The parameter is copied into \a copy
  84  * which has to be allocated big enough by a caller, quotes are stripped in
  85  * the copy and the copy is terminated by 0.
  86  *
  87  * On return \a params is set to next parameter or to NULL if last
  88  * parameter is returned.
  89  *
  90  * \retval 0 if parameter is returned in \a copy
  91  * \retval 1 otherwise
  92  * \retval -EINVAL if unbalanced quota is found
  93  */
  94 int class_get_next_param(char **params, char *copy)
  95 {
  96         char *q1, *q2, *str;
  97         int len;
  98
  99         str = *params;
 100         while (*str == ' ')
 101                 str++;
 102
 103         if (*str == '\0') {
 104                 *params = NULL;
 105                 return 1;
 106         }
 107
 108         while (1) {
 109                 q1 = strpbrk(str, " '\"");
 110                 if (q1 == NULL) {
 111                         len = strlen(str);
 112                         memcpy(copy, str, len);
 113                         copy[len] = '\0';
 114                         *params = NULL;
 115                         return 0;
 116                 }
 117                 len = q1 - str;
 118                 if (*q1 == ' ') {
 119                         memcpy(copy, str, len);
 120                         copy[len] = '\0';
 121                         *params = str + len;
 122                         return 0;
 123                 }
 124
 125                 memcpy(copy, str, len);
 126                 copy += len;
 127
 128                 /* search for the matching closing quote */
 129                 str = q1 + 1;
 130                 q2 = strchr(str, *q1);
 131                 if (q2 == NULL) {
 132                         CERROR("Unbalanced quota in parameters: \"%s\"\n",
 133                                *params);
 134                         return -EINVAL;
 135                 }
 136                 len = q2 - str;
 137                 memcpy(copy, str, len);
 138                 copy += len;
 139                 str = q2 + 1;
 140         }
 141         return 1;
 142 }
 143
 144 /* returns 0 if this is the first key in the buffer, else 1.
 145    valp points to first char after key. */
 146 int class_match_param(char *buf, char *key, char **valp)
 147 {
 148         if (!buf)
 149                 return 1;
 150
 151         if (memcmp(buf, key, strlen(key)) != 0)
 152                 return 1;
 153
 154         if (valp)
 155                 *valp = buf + strlen(key);
 156
 157         return 0;
 158 }
 159
 160 static int parse_nid(char *buf, void *value)
 161 {
 162         lnet_nid_t *nid = (lnet_nid_t *)value;
 163
 164         *nid = libcfs_str2nid(buf);
 165         if (*nid != LNET_NID_ANY)
 166                 return 0;
 167
 168         LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", buf);
 169         return -EINVAL;
 170 }
 171
 172 static int parse_net(char *buf, void *value)
 173 {
 174         __u32 *net = (__u32 *)value;
 175
 176         *net = libcfs_str2net(buf);
 177         CDEBUG(D_INFO, "Net %s\n", libcfs_net2str(*net));
 178         return 0;
 179 }
 180
 181 enum {
 182         CLASS_PARSE_NID = 1,
 183         CLASS_PARSE_NET,
 184 };
 185
 186 /* 0 is good nid,
 187    1 not found
 188    < 0 error
 189    endh is set to next separator */
 190 static int class_parse_value(char *buf, int opc, void *value, char **endh)
 191 {
 192         char *endp;
 193         char  tmp;
 194         int   rc = 0;
 195
 196         if (!buf)
 197                 return 1;
 198         while (*buf == ',' || *buf == ':')
 199                 buf++;
 200         if (*buf == ' ' || *buf == '/' || *buf == '\0')
 201                 return 1;
 202
 203         /* nid separators or end of nids */
 204         endp = strpbrk(buf, ",: /");
 205         if (endp == NULL)
 206                 endp = buf + strlen(buf);
 207
 208         tmp = *endp;
 209         *endp = '\0';
 210         switch (opc) {
 211         default:
 212                 LBUG();
 213         case CLASS_PARSE_NID:
 214                 rc = parse_nid(buf, value);
 215                 break;
 216         case CLASS_PARSE_NET:
 217                 rc = parse_net(buf, value);
 218                 break;
 219         }
 220         *endp = tmp;
 221         if (rc != 0)
 222                 return rc;
 223         if (endh)
 224                 *endh = endp;
 225         return 0;
 226 }
 227
 228 int class_parse_nid(char *buf, lnet_nid_t *nid, char **endh)
 229 {
 230         return class_parse_value(buf, CLASS_PARSE_NID, (void *)nid, endh);
 231 }
 232
 233 int class_parse_net(char *buf, __u32 *net, char **endh)
 234 {
 235         return class_parse_value(buf, CLASS_PARSE_NET, (void *)net, endh);
 236 }
 237
 238 /* 1 param contains key and match
 239  * 0 param contains key and not match
 240  * -1 param does not contain key
 241  */
 242 int class_match_nid(char *buf, char *key, lnet_nid_t nid)
 243 {
 244         lnet_nid_t tmp;
 245         int   rc = -1;
 246
 247         while (class_find_param(buf, key, &buf) == 0) {
 248                 /* please restrict to the nids pertaining to
 249                  * the specified nids */
 250                 while (class_parse_nid(buf, &tmp, &buf) == 0) {
 251                         if (tmp == nid)
 252                                 return 1;
 253                 }
 254                 rc = 0;
 255         }
 256         return rc;
 257 }
 258
 259 int class_match_net(char *buf, char *key, __u32 net)
 260 {
 261         __u32 tmp;
 262         int   rc = -1;
 263
 264         while (class_find_param(buf, key, &buf) == 0) {
 265                 /* please restrict to the nids pertaining to
 266                  * the specified networks */
 267                 while (class_parse_net(buf, &tmp, &buf) == 0) {
 268                         if (tmp == net)
 269                                 return 1;
 270                 }
 271                 rc = 0;
 272         }
 273         return rc;
 274 }
 275
 276 EXPORT_SYMBOL(class_find_param);
 277 EXPORT_SYMBOL(class_get_next_param);
 278 EXPORT_SYMBOL(class_match_param);
 279 EXPORT_SYMBOL(class_parse_nid);
 280 EXPORT_SYMBOL(class_parse_net);
 281 EXPORT_SYMBOL(class_match_nid);
 282 EXPORT_SYMBOL(class_match_net);
 283
 284 /********************** class fns **********************/
 285
 286 /**
 287  * Create a new obd device and set the type, name and uuid.  If successful,
 288  * the new device can be accessed by either name or uuid.
 289  */
 290 int class_attach(struct lustre_cfg *lcfg)
 291 {
 292         struct obd_device *obd = NULL;
 293         char *typename, *name, *uuid;
 294         int rc, len;
 295         ENTRY;
 296
 297         if (!LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 298                 CERROR("No type passed!\n");
 299                 RETURN(-EINVAL);
 300         }
 301         typename = lustre_cfg_string(lcfg, 1);
 302
 303         if (!LUSTRE_CFG_BUFLEN(lcfg, 0)) {
 304                 CERROR("No name passed!\n");
 305                 RETURN(-EINVAL);
 306         }
 307         name = lustre_cfg_string(lcfg, 0);
 308
 309         if (!LUSTRE_CFG_BUFLEN(lcfg, 2)) {
 310                 CERROR("No UUID passed!\n");
 311                 RETURN(-EINVAL);
 312         }
 313         uuid = lustre_cfg_string(lcfg, 2);
 314
 315         CDEBUG(D_IOCTL, "attach type %s name: %s uuid: %s\n",
 316                MKSTR(typename), MKSTR(name), MKSTR(uuid));
 317
 318         obd = class_newdev(typename, name);
 319         if (IS_ERR(obd)) {
 320                 /* Already exists or out of obds */
 321                 rc = PTR_ERR(obd);
 322                 obd = NULL;
 323                 CERROR("Cannot create device %s of type %s : %d\n",
 324                        name, typename, rc);
 325                 GOTO(out, rc);
 326         }
 327         LASSERTF(obd != NULL, "Cannot get obd device %s of type %s\n",
 328                  name, typename);
 329         LASSERTF(obd->obd_magic == OBD_DEVICE_MAGIC,
 330                  "obd %p obd_magic %08X != %08X\n",
 331                  obd, obd->obd_magic, OBD_DEVICE_MAGIC);
 332         LASSERTF(strncmp(obd->obd_name, name, strlen(name)) == 0,
 333                  "%p obd_name %s != %s\n", obd, obd->obd_name, name);
 334
 335         cfs_rwlock_init(&obd->obd_pool_lock);
 336         obd->obd_pool_limit = 0;
 337         obd->obd_pool_slv = 0;
 338
 339         CFS_INIT_LIST_HEAD(&obd->obd_exports);
 340         CFS_INIT_LIST_HEAD(&obd->obd_unlinked_exports);
 341         CFS_INIT_LIST_HEAD(&obd->obd_delayed_exports);
 342         CFS_INIT_LIST_HEAD(&obd->obd_exports_timed);
 343         CFS_INIT_LIST_HEAD(&obd->obd_nid_stats);
 344         cfs_spin_lock_init(&obd->obd_nid_lock);
 345         cfs_spin_lock_init(&obd->obd_dev_lock);
 346         cfs_sema_init(&obd->obd_dev_sem, 1);
 347         cfs_spin_lock_init(&obd->obd_osfs_lock);
 348         /* obd->obd_osfs_age must be set to a value in the distant
 349          * past to guarantee a fresh statfs is fetched on mount. */
 350         obd->obd_osfs_age = cfs_time_shift_64(-1000);
 351
 352         /* XXX belongs in setup not attach  */
 353         cfs_init_rwsem(&obd->obd_observer_link_sem);
 354         /* recovery data */
 355         cfs_init_timer(&obd->obd_recovery_timer);
 356         cfs_spin_lock_init(&obd->obd_recovery_task_lock);
 357         cfs_waitq_init(&obd->obd_next_transno_waitq);
 358         cfs_waitq_init(&obd->obd_evict_inprogress_waitq);
 359         CFS_INIT_LIST_HEAD(&obd->obd_req_replay_queue);
 360         CFS_INIT_LIST_HEAD(&obd->obd_lock_replay_queue);
 361         CFS_INIT_LIST_HEAD(&obd->obd_final_req_queue);
 362         CFS_INIT_LIST_HEAD(&obd->obd_evict_list);
 363
 364         llog_group_init(&obd->obd_olg, FID_SEQ_LLOG);
 365
 366         len = strlen(uuid);
 367         if (len >= sizeof(obd->obd_uuid)) {
 368                 CERROR("uuid must be < %d bytes long\n",
 369                        (int)sizeof(obd->obd_uuid));
 370                 GOTO(out, rc = -EINVAL);
 371         }
 372         memcpy(obd->obd_uuid.uuid, uuid, len);
 373
 374         /* do the attach */
 375         if (OBP(obd, attach)) {
 376                 rc = OBP(obd,attach)(obd, sizeof *lcfg, lcfg);
 377                 if (rc)
 378                         GOTO(out, rc = -EINVAL);
 379         }
 380
 381         /* Detach drops this */
 382         cfs_spin_lock(&obd->obd_dev_lock);
 383         cfs_atomic_set(&obd->obd_refcount, 1);
 384         cfs_spin_unlock(&obd->obd_dev_lock);
 385         lu_ref_init(&obd->obd_reference);
 386         lu_ref_add(&obd->obd_reference, "attach", obd);
 387
 388         obd->obd_attached = 1;
 389         CDEBUG(D_IOCTL, "OBD: dev %d attached type %s with refcount %d\n",
 390                obd->obd_minor, typename, cfs_atomic_read(&obd->obd_refcount));
 391         RETURN(0);
 392  out:
 393         if (obd != NULL) {
 394                 class_release_dev(obd);
 395         }
 396         return rc;
 397 }
 398
 399 /** Create hashes, self-export, and call type-specific setup.
 400  * Setup is effectively the "start this obd" call.
 401  */
 402 int class_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 403 {
 404         int err = 0;
 405         struct obd_export *exp;
 406         ENTRY;
 407
 408         LASSERT(obd != NULL);
 409         LASSERTF(obd == class_num2obd(obd->obd_minor),
 410                  "obd %p != obd_devs[%d] %p\n",
 411                  obd, obd->obd_minor, class_num2obd(obd->obd_minor));
 412         LASSERTF(obd->obd_magic == OBD_DEVICE_MAGIC,
 413                  "obd %p obd_magic %08x != %08x\n",
 414                  obd, obd->obd_magic, OBD_DEVICE_MAGIC);
 415
 416         /* have we attached a type to this device? */
 417         if (!obd->obd_attached) {
 418                 CERROR("Device %d not attached\n", obd->obd_minor);
 419                 RETURN(-ENODEV);
 420         }
 421
 422         if (obd->obd_set_up) {
 423                 CERROR("Device %d already setup (type %s)\n",
 424                        obd->obd_minor, obd->obd_type->typ_name);
 425                 RETURN(-EEXIST);
 426         }
 427
 428         /* is someone else setting us up right now? (attach inits spinlock) */
 429         cfs_spin_lock(&obd->obd_dev_lock);
 430         if (obd->obd_starting) {
 431                 cfs_spin_unlock(&obd->obd_dev_lock);
 432                 CERROR("Device %d setup in progress (type %s)\n",
 433                        obd->obd_minor, obd->obd_type->typ_name);
 434                 RETURN(-EEXIST);
 435         }
 436         /* just leave this on forever.  I can't use obd_set_up here because
 437            other fns check that status, and we're not actually set up yet. */
 438         obd->obd_starting = 1;
 439         obd->obd_uuid_hash = NULL;
 440         obd->obd_nid_hash = NULL;
 441         obd->obd_nid_stats_hash = NULL;
 442         cfs_spin_unlock(&obd->obd_dev_lock);
 443
 444         /* create an uuid-export lustre hash */
 445         obd->obd_uuid_hash = cfs_hash_create("UUID_HASH",
 446                                              HASH_UUID_CUR_BITS,
 447                                              HASH_UUID_MAX_BITS,
 448                                              HASH_UUID_BKT_BITS, 0,
 449                                              CFS_HASH_MIN_THETA,
 450                                              CFS_HASH_MAX_THETA,
 451                                              &uuid_hash_ops, CFS_HASH_DEFAULT);
 452         if (!obd->obd_uuid_hash)
 453                 GOTO(err_hash, err = -ENOMEM);
 454
 455         /* create a nid-export lustre hash */
 456         obd->obd_nid_hash = cfs_hash_create("NID_HASH",
 457                                             HASH_NID_CUR_BITS,
 458                                             HASH_NID_MAX_BITS,
 459                                             HASH_NID_BKT_BITS, 0,
 460                                             CFS_HASH_MIN_THETA,
 461                                             CFS_HASH_MAX_THETA,
 462                                             &nid_hash_ops, CFS_HASH_DEFAULT);
 463         if (!obd->obd_nid_hash)
 464                 GOTO(err_hash, err = -ENOMEM);
 465
 466         /* create a nid-stats lustre hash */
 467         obd->obd_nid_stats_hash = cfs_hash_create("NID_STATS",
 468                                                   HASH_NID_STATS_CUR_BITS,
 469                                                   HASH_NID_STATS_MAX_BITS,
 470                                                   HASH_NID_STATS_BKT_BITS, 0,
 471                                                   CFS_HASH_MIN_THETA,
 472                                                   CFS_HASH_MAX_THETA,
 473                                                   &nid_stat_hash_ops, CFS_HASH_DEFAULT);
 474         if (!obd->obd_nid_stats_hash)
 475                 GOTO(err_hash, err = -ENOMEM);
 476
 477         exp = class_new_export(obd, &obd->obd_uuid);
 478         if (IS_ERR(exp))
 479                 GOTO(err_hash, err = PTR_ERR(exp));
 480
 481         obd->obd_self_export = exp;
 482         cfs_list_del_init(&exp->exp_obd_chain_timed);
 483         class_export_put(exp);
 484
 485         err = obd_setup(obd, lcfg);
 486         if (err)
 487                 GOTO(err_exp, err);
 488
 489         obd->obd_set_up = 1;
 490
 491         cfs_spin_lock(&obd->obd_dev_lock);
 492         /* cleanup drops this */
 493         class_incref(obd, "setup", obd);
 494         cfs_spin_unlock(&obd->obd_dev_lock);
 495
 496         CDEBUG(D_IOCTL, "finished setup of obd %s (uuid %s)\n",
 497                obd->obd_name, obd->obd_uuid.uuid);
 498
 499         RETURN(0);
 500 err_exp:
 501         if (obd->obd_self_export) {
 502                 class_unlink_export(obd->obd_self_export);
 503                 obd->obd_self_export = NULL;
 504         }
 505 err_hash:
 506         if (obd->obd_uuid_hash) {
 507                 cfs_hash_putref(obd->obd_uuid_hash);
 508                 obd->obd_uuid_hash = NULL;
 509         }
 510         if (obd->obd_nid_hash) {
 511                 cfs_hash_putref(obd->obd_nid_hash);
 512                 obd->obd_nid_hash = NULL;
 513         }
 514         if (obd->obd_nid_stats_hash) {
 515                 cfs_hash_putref(obd->obd_nid_stats_hash);
 516                 obd->obd_nid_stats_hash = NULL;
 517         }
 518         obd->obd_starting = 0;
 519         CERROR("setup %s failed (%d)\n", obd->obd_name, err);
 520         return err;
 521 }
 522
 523 /** We have finished using this obd and are ready to destroy it.
 524  * There can be no more references to this obd.
 525  */
 526 int class_detach(struct obd_device *obd, struct lustre_cfg *lcfg)
 527 {
 528         ENTRY;
 529
 530         if (obd->obd_set_up) {
 531                 CERROR("OBD device %d still set up\n", obd->obd_minor);
 532                 RETURN(-EBUSY);
 533         }
 534
 535         cfs_spin_lock(&obd->obd_dev_lock);
 536         if (!obd->obd_attached) {
 537                 cfs_spin_unlock(&obd->obd_dev_lock);
 538                 CERROR("OBD device %d not attached\n", obd->obd_minor);
 539                 RETURN(-ENODEV);
 540         }
 541         obd->obd_attached = 0;
 542         cfs_spin_unlock(&obd->obd_dev_lock);
 543
 544         CDEBUG(D_IOCTL, "detach on obd %s (uuid %s)\n",
 545                obd->obd_name, obd->obd_uuid.uuid);
 546
 547         class_decref(obd, "attach", obd);
 548         RETURN(0);
 549 }
 550
 551 /** Start shutting down the obd.  There may be in-progess ops when
 552  * this is called.  We tell them to start shutting down with a call
 553  * to class_disconnect_exports().
 554  */
 555 int class_cleanup(struct obd_device *obd, struct lustre_cfg *lcfg)
 556 {
 557         int err = 0;
 558         char *flag;
 559         ENTRY;
 560
 561         OBD_RACE(OBD_FAIL_LDLM_RECOV_CLIENTS);
 562
 563         if (!obd->obd_set_up) {
 564                 CERROR("Device %d not setup\n", obd->obd_minor);
 565                 RETURN(-ENODEV);
 566         }
 567
 568         cfs_spin_lock(&obd->obd_dev_lock);
 569         if (obd->obd_stopping) {
 570                 cfs_spin_unlock(&obd->obd_dev_lock);
 571                 CERROR("OBD %d already stopping\n", obd->obd_minor);
 572                 RETURN(-ENODEV);
 573         }
 574         /* Leave this on forever */
 575         obd->obd_stopping = 1;
 576         cfs_spin_unlock(&obd->obd_dev_lock);
 577
 578         if (lcfg->lcfg_bufcount >= 2 && LUSTRE_CFG_BUFLEN(lcfg, 1) > 0) {
 579                 for (flag = lustre_cfg_string(lcfg, 1); *flag != 0; flag++)
 580                         switch (*flag) {
 581                         case 'F':
 582                                 obd->obd_force = 1;
 583                                 break;
 584                         case 'A':
 585                                 LCONSOLE_WARN("Failing over %s\n",
 586                                               obd->obd_name);
 587                                 obd->obd_fail = 1;
 588                                 obd->obd_no_transno = 1;
 589                                 obd->obd_no_recov = 1;
 590                                 if (OBP(obd, iocontrol)) {
 591                                         obd_iocontrol(OBD_IOC_SYNC,
 592                                                       obd->obd_self_export,
 593                                                       0, NULL, NULL);
 594                                 }
 595                                 break;
 596                         default:
 597                                 CERROR("Unrecognised flag '%c'\n", *flag);
 598                         }
 599         }
 600
 601         LASSERT(obd->obd_self_export);
 602
 603         /* The three references that should be remaining are the
 604          * obd_self_export and the attach and setup references. */
 605         if (cfs_atomic_read(&obd->obd_refcount) > 3) {
 606                 /* refcounf - 3 might be the number of real exports
 607                    (excluding self export). But class_incref is called
 608                    by other things as well, so don't count on it. */
 609                 CDEBUG(D_IOCTL, "%s: forcing exports to disconnect: %d\n",
 610                        obd->obd_name, cfs_atomic_read(&obd->obd_refcount) - 3);
 611                 dump_exports(obd, 0);
 612                 class_disconnect_exports(obd);
 613         }
 614
 615         /* destroy an uuid-export hash body */
 616         if (obd->obd_uuid_hash) {
 617                 cfs_hash_putref(obd->obd_uuid_hash);
 618                 obd->obd_uuid_hash = NULL;
 619         }
 620
 621         /* destroy a nid-export hash body */
 622         if (obd->obd_nid_hash) {
 623                 cfs_hash_putref(obd->obd_nid_hash);
 624                 obd->obd_nid_hash = NULL;
 625         }
 626
 627         /* destroy a nid-stats hash body */
 628         if (obd->obd_nid_stats_hash) {
 629                 cfs_hash_putref(obd->obd_nid_stats_hash);
 630                 obd->obd_nid_stats_hash = NULL;
 631         }
 632
 633         /* Precleanup, we must make sure all exports get destroyed. */
 634         err = obd_precleanup(obd, OBD_CLEANUP_EXPORTS);
 635         if (err)
 636                 CERROR("Precleanup %s returned %d\n",
 637                        obd->obd_name, err);
 638         class_decref(obd, "setup", obd);
 639         obd->obd_set_up = 0;
 640         RETURN(0);
 641 }
 642
 643 struct obd_device *class_incref(struct obd_device *obd,
 644                                 const char *scope, const void *source)
 645 {
 646         lu_ref_add_atomic(&obd->obd_reference, scope, source);
 647         cfs_atomic_inc(&obd->obd_refcount);
 648         CDEBUG(D_INFO, "incref %s (%p) now %d\n", obd->obd_name, obd,
 649                cfs_atomic_read(&obd->obd_refcount));
 650
 651         return obd;
 652 }
 653
 654 void class_decref(struct obd_device *obd, const char *scope, const void *source)
 655 {
 656         int err;
 657         int refs;
 658
 659         cfs_spin_lock(&obd->obd_dev_lock);
 660         cfs_atomic_dec(&obd->obd_refcount);
 661         refs = cfs_atomic_read(&obd->obd_refcount);
 662         cfs_spin_unlock(&obd->obd_dev_lock);
 663         lu_ref_del(&obd->obd_reference, scope, source);
 664
 665         CDEBUG(D_INFO, "Decref %s (%p) now %d\n", obd->obd_name, obd, refs);
 666
 667         if ((refs == 1) && obd->obd_stopping) {
 668                 /* All exports have been destroyed; there should
 669                    be no more in-progress ops by this point.*/
 670
 671                 cfs_spin_lock(&obd->obd_self_export->exp_lock);
 672                 obd->obd_self_export->exp_flags |= exp_flags_from_obd(obd);
 673                 cfs_spin_unlock(&obd->obd_self_export->exp_lock);
 674
 675                 /* note that we'll recurse into class_decref again */
 676                 class_unlink_export(obd->obd_self_export);
 677                 return;
 678         }
 679
 680         if (refs == 0) {
 681                 CDEBUG(D_CONFIG, "finishing cleanup of obd %s (%s)\n",
 682                        obd->obd_name, obd->obd_uuid.uuid);
 683                 LASSERT(!obd->obd_attached);
 684                 if (obd->obd_stopping) {
 685                         /* If we're not stopping, we were never set up */
 686                         err = obd_cleanup(obd);
 687                         if (err)
 688                                 CERROR("Cleanup %s returned %d\n",
 689                                        obd->obd_name, err);
 690                 }
 691                 if (OBP(obd, detach)) {
 692                         err = OBP(obd, detach)(obd);
 693                         if (err)
 694                                 CERROR("Detach returned %d\n", err);
 695                 }
 696                 class_release_dev(obd);
 697         }
 698 }
 699
 700 /** Add a failover nid location.
 701  * Client obd types contact server obd types using this nid list.
 702  */
 703 int class_add_conn(struct obd_device *obd, struct lustre_cfg *lcfg)
 704 {
 705         struct obd_import *imp;
 706         struct obd_uuid uuid;
 707         int rc;
 708         ENTRY;
 709
 710         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1 ||
 711             LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(struct obd_uuid)) {
 712                 CERROR("invalid conn_uuid\n");
 713                 RETURN(-EINVAL);
 714         }
 715         if (strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) &&
 716             strcmp(obd->obd_type->typ_name, LUSTRE_OSC_NAME) &&
 717             strcmp(obd->obd_type->typ_name, LUSTRE_MGC_NAME)) {
 718                 CERROR("can't add connection on non-client dev\n");
 719                 RETURN(-EINVAL);
 720         }
 721
 722         imp = obd->u.cli.cl_import;
 723         if (!imp) {
 724                 CERROR("try to add conn on immature client dev\n");
 725                 RETURN(-EINVAL);
 726         }
 727
 728         obd_str2uuid(&uuid, lustre_cfg_string(lcfg, 1));
 729         rc = obd_add_conn(imp, &uuid, lcfg->lcfg_num);
 730
 731         RETURN(rc);
 732 }
 733
 734 /** Remove a failover nid location.
 735  */
 736 int class_del_conn(struct obd_device *obd, struct lustre_cfg *lcfg)
 737 {
 738         struct obd_import *imp;
 739         struct obd_uuid uuid;
 740         int rc;
 741         ENTRY;
 742
 743         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1 ||
 744             LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(struct obd_uuid)) {
 745                 CERROR("invalid conn_uuid\n");
 746                 RETURN(-EINVAL);
 747         }
 748         if (strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) &&
 749             strcmp(obd->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 750                 CERROR("can't del connection on non-client dev\n");
 751                 RETURN(-EINVAL);
 752         }
 753
 754         imp = obd->u.cli.cl_import;
 755         if (!imp) {
 756                 CERROR("try to del conn on immature client dev\n");
 757                 RETURN(-EINVAL);
 758         }
 759
 760         obd_str2uuid(&uuid, lustre_cfg_string(lcfg, 1));
 761         rc = obd_del_conn(imp, &uuid);
 762
 763         RETURN(rc);
 764 }
 765
 766 CFS_LIST_HEAD(lustre_profile_list);
 767
 768 struct lustre_profile *class_get_profile(const char * prof)
 769 {
 770         struct lustre_profile *lprof;
 771
 772         ENTRY;
 773         cfs_list_for_each_entry(lprof, &lustre_profile_list, lp_list) {
 774                 if (!strcmp(lprof->lp_profile, prof)) {
 775                         RETURN(lprof);
 776                 }
 777         }
 778         RETURN(NULL);
 779 }
 780
 781 /** Create a named "profile".
 782  * This defines the mdc and osc names to use for a client.
 783  * This also is used to define the lov to be used by a mdt.
 784  */
 785 int class_add_profile(int proflen, char *prof, int osclen, char *osc,
 786                       int mdclen, char *mdc)
 787 {
 788         struct lustre_profile *lprof;
 789         int err = 0;
 790         ENTRY;
 791
 792         CDEBUG(D_CONFIG, "Add profile %s\n", prof);
 793
 794         OBD_ALLOC(lprof, sizeof(*lprof));
 795         if (lprof == NULL)
 796                 RETURN(-ENOMEM);
 797         CFS_INIT_LIST_HEAD(&lprof->lp_list);
 798
 799         LASSERT(proflen == (strlen(prof) + 1));
 800         OBD_ALLOC(lprof->lp_profile, proflen);
 801         if (lprof->lp_profile == NULL)
 802                 GOTO(out, err = -ENOMEM);
 803         memcpy(lprof->lp_profile, prof, proflen);
 804
 805         LASSERT(osclen == (strlen(osc) + 1));
 806         OBD_ALLOC(lprof->lp_dt, osclen);
 807         if (lprof->lp_dt == NULL)
 808                 GOTO(out, err = -ENOMEM);
 809         memcpy(lprof->lp_dt, osc, osclen);
 810
 811         if (mdclen > 0) {
 812                 LASSERT(mdclen == (strlen(mdc) + 1));
 813                 OBD_ALLOC(lprof->lp_md, mdclen);
 814                 if (lprof->lp_md == NULL)
 815                         GOTO(out, err = -ENOMEM);
 816                 memcpy(lprof->lp_md, mdc, mdclen);
 817         }
 818
 819         cfs_list_add(&lprof->lp_list, &lustre_profile_list);
 820         RETURN(err);
 821
 822 out:
 823         if (lprof->lp_md)
 824                 OBD_FREE(lprof->lp_md, mdclen);
 825         if (lprof->lp_dt)
 826                 OBD_FREE(lprof->lp_dt, osclen);
 827         if (lprof->lp_profile)
 828                 OBD_FREE(lprof->lp_profile, proflen);
 829         OBD_FREE(lprof, sizeof(*lprof));
 830         RETURN(err);
 831 }
 832
 833 void class_del_profile(const char *prof)
 834 {
 835         struct lustre_profile *lprof;
 836         ENTRY;
 837
 838         CDEBUG(D_CONFIG, "Del profile %s\n", prof);
 839
 840         lprof = class_get_profile(prof);
 841         if (lprof) {
 842                 cfs_list_del(&lprof->lp_list);
 843                 OBD_FREE(lprof->lp_profile, strlen(lprof->lp_profile) + 1);
 844                 OBD_FREE(lprof->lp_dt, strlen(lprof->lp_dt) + 1);
 845                 if (lprof->lp_md)
 846                         OBD_FREE(lprof->lp_md, strlen(lprof->lp_md) + 1);
 847                 OBD_FREE(lprof, sizeof *lprof);
 848         }
 849         EXIT;
 850 }
 851
 852 /* COMPAT_146 */
 853 void class_del_profiles(void)
 854 {
 855         struct lustre_profile *lprof, *n;
 856         ENTRY;
 857
 858         cfs_list_for_each_entry_safe(lprof, n, &lustre_profile_list, lp_list) {
 859                 cfs_list_del(&lprof->lp_list);
 860                 OBD_FREE(lprof->lp_profile, strlen(lprof->lp_profile) + 1);
 861                 OBD_FREE(lprof->lp_dt, strlen(lprof->lp_dt) + 1);
 862                 if (lprof->lp_md)
 863                         OBD_FREE(lprof->lp_md, strlen(lprof->lp_md) + 1);
 864                 OBD_FREE(lprof, sizeof *lprof);
 865         }
 866         EXIT;
 867 }
 868
 869 static int class_set_global(char *ptr, int val) {
 870         ENTRY;
 871
 872         if (class_match_param(ptr, PARAM_AT_MIN, NULL) == 0)
 873             at_min = val;
 874         else if (class_match_param(ptr, PARAM_AT_MAX, NULL) == 0)
 875                 at_max = val;
 876         else if (class_match_param(ptr, PARAM_AT_EXTRA, NULL) == 0)
 877                 at_extra = val;
 878         else if (class_match_param(ptr, PARAM_AT_EARLY_MARGIN, NULL) == 0)
 879                 at_early_margin = val;
 880         else if (class_match_param(ptr, PARAM_AT_HISTORY, NULL) == 0)
 881                 at_history = val;
 882         else
 883                 RETURN(-EINVAL);
 884
 885         CDEBUG(D_IOCTL, "global %s = %d\n", ptr, val);
 886
 887         RETURN(0);
 888 }
 889
 890
 891 /* We can't call ll_process_config directly because it lives in a module that
 892    must be loaded after this one. */
 893 static int (*client_process_config)(struct lustre_cfg *lcfg) = NULL;
 894
 895 void lustre_register_client_process_config(int (*cpc)(struct lustre_cfg *lcfg))
 896 {
 897         client_process_config = cpc;
 898 }
 899 EXPORT_SYMBOL(lustre_register_client_process_config);
 900
 901 /** Process configuration commands given in lustre_cfg form.
 902  * These may come from direct calls (e.g. class_manual_cleanup)
 903  * or processing the config llog, or ioctl from lctl.
 904  */
 905 int class_process_config(struct lustre_cfg *lcfg)
 906 {
 907         struct obd_device *obd;
 908         int err;
 909
 910         LASSERT(lcfg && !IS_ERR(lcfg));
 911         CDEBUG(D_IOCTL, "processing cmd: %x\n", lcfg->lcfg_command);
 912
 913         /* Commands that don't need a device */
 914         switch(lcfg->lcfg_command) {
 915         case LCFG_ATTACH: {
 916                 err = class_attach(lcfg);
 917                 GOTO(out, err);
 918         }
 919         case LCFG_ADD_UUID: {
 920                 CDEBUG(D_IOCTL, "adding mapping from uuid %s to nid "LPX64
 921                        " (%s)\n", lustre_cfg_string(lcfg, 1),
 922                        lcfg->lcfg_nid, libcfs_nid2str(lcfg->lcfg_nid));
 923
 924                 err = class_add_uuid(lustre_cfg_string(lcfg, 1), lcfg->lcfg_nid);
 925                 GOTO(out, err);
 926         }
 927         case LCFG_DEL_UUID: {
 928                 CDEBUG(D_IOCTL, "removing mappings for uuid %s\n",
 929                        (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) == 0)
 930                        ? "<all uuids>" : lustre_cfg_string(lcfg, 1));
 931
 932                 err = class_del_uuid(lustre_cfg_string(lcfg, 1));
 933                 GOTO(out, err);
 934         }
 935         case LCFG_MOUNTOPT: {
 936                 CDEBUG(D_IOCTL, "mountopt: profile %s osc %s mdc %s\n",
 937                        lustre_cfg_string(lcfg, 1),
 938                        lustre_cfg_string(lcfg, 2),
 939                        lustre_cfg_string(lcfg, 3));
 940                 /* set these mount options somewhere, so ll_fill_super
 941                  * can find them. */
 942                 err = class_add_profile(LUSTRE_CFG_BUFLEN(lcfg, 1),
 943                                         lustre_cfg_string(lcfg, 1),
 944                                         LUSTRE_CFG_BUFLEN(lcfg, 2),
 945                                         lustre_cfg_string(lcfg, 2),
 946                                         LUSTRE_CFG_BUFLEN(lcfg, 3),
 947                                         lustre_cfg_string(lcfg, 3));
 948                 GOTO(out, err);
 949         }
 950         case LCFG_DEL_MOUNTOPT: {
 951                 CDEBUG(D_IOCTL, "mountopt: profile %s\n",
 952                        lustre_cfg_string(lcfg, 1));
 953                 class_del_profile(lustre_cfg_string(lcfg, 1));
 954                 GOTO(out, err = 0);
 955         }
 956         case LCFG_SET_TIMEOUT: {
 957                 CDEBUG(D_IOCTL, "changing lustre timeout from %d to %d\n",
 958                        obd_timeout, lcfg->lcfg_num);
 959                 obd_timeout = max(lcfg->lcfg_num, 1U);
 960                 GOTO(out, err = 0);
 961         }
 962         case LCFG_SET_LDLM_TIMEOUT: {
 963                 CDEBUG(D_IOCTL, "changing lustre ldlm_timeout from %d to %d\n",
 964                        ldlm_timeout, lcfg->lcfg_num);
 965                 ldlm_timeout = max(lcfg->lcfg_num, 1U);
 966                 if (ldlm_timeout >= obd_timeout)
 967                         ldlm_timeout = max(obd_timeout / 3, 1U);
 968
 969                 GOTO(out, err = 0);
 970         }
 971         case LCFG_SET_UPCALL: {
 972                 LCONSOLE_ERROR_MSG(0x15a, "recovery upcall is deprecated\n");
 973                 /* COMPAT_146 Don't fail on old configs */
 974                 GOTO(out, err = 0);
 975         }
 976         case LCFG_MARKER: {
 977                 struct cfg_marker *marker;
 978                 marker = lustre_cfg_buf(lcfg, 1);
 979                 CDEBUG(D_IOCTL, "marker %d (%#x) %.16s %s\n", marker->cm_step,
 980                        marker->cm_flags, marker->cm_tgtname, marker->cm_comment);
 981                 GOTO(out, err = 0);
 982         }
 983         case LCFG_PARAM: {
 984                 char *tmp;
 985                 /* llite has no obd */
 986                 if ((class_match_param(lustre_cfg_string(lcfg, 1),
 987                                        PARAM_LLITE, 0) == 0) &&
 988                     client_process_config) {
 989                         err = (*client_process_config)(lcfg);
 990                         GOTO(out, err);
 991                 } else if ((class_match_param(lustre_cfg_string(lcfg, 1),
 992                                               PARAM_SYS, &tmp) == 0)) {
 993                         /* Global param settings */
 994                         err = class_set_global(tmp, lcfg->lcfg_num);
 995                         /* Note that since LCFG_PARAM is LCFG_REQUIRED, new
 996                            unknown globals would cause config to fail */
 997                         if (err)
 998                                 CWARN("Ignoring unknown param %s\n", tmp);
 999                         GOTO(out, 0);
1000                 }
1001
1002                 /* Fall through */
1003                 break;
1004         }
1005         }
1006
1007         /* Commands that require a device */
1008         obd = class_name2obd(lustre_cfg_string(lcfg, 0));
1009         if (obd == NULL) {
1010                 if (!LUSTRE_CFG_BUFLEN(lcfg, 0))
1011                         CERROR("this lcfg command requires a device name\n");
1012                 else
1013                         CERROR("no device for: %s\n",
1014                                lustre_cfg_string(lcfg, 0));
1015
1016                 GOTO(out, err = -EINVAL);
1017         }
1018
1019         switch(lcfg->lcfg_command) {
1020         case LCFG_SETUP: {
1021                 err = class_setup(obd, lcfg);
1022                 GOTO(out, err);
1023         }
1024         case LCFG_DETACH: {
1025                 err = class_detach(obd, lcfg);
1026                 GOTO(out, err = 0);
1027         }
1028         case LCFG_CLEANUP: {
1029                 err = class_cleanup(obd, lcfg);
1030                 GOTO(out, err = 0);
1031         }
1032         case LCFG_ADD_CONN: {
1033                 err = class_add_conn(obd, lcfg);
1034                 GOTO(out, err = 0);
1035         }
1036         case LCFG_DEL_CONN: {
1037                 err = class_del_conn(obd, lcfg);
1038                 GOTO(out, err = 0);
1039         }
1040         case LCFG_POOL_NEW: {
1041                 err = obd_pool_new(obd, lustre_cfg_string(lcfg, 2));
1042                 GOTO(out, err = 0);
1043                 break;
1044         }
1045         case LCFG_POOL_ADD: {
1046                 err = obd_pool_add(obd, lustre_cfg_string(lcfg, 2),
1047                                    lustre_cfg_string(lcfg, 3));
1048                 GOTO(out, err = 0);
1049                 break;
1050         }
1051         case LCFG_POOL_REM: {
1052                 err = obd_pool_rem(obd, lustre_cfg_string(lcfg, 2),
1053                                    lustre_cfg_string(lcfg, 3));
1054                 GOTO(out, err = 0);
1055                 break;
1056         }
1057         case LCFG_POOL_DEL: {
1058                 err = obd_pool_del(obd, lustre_cfg_string(lcfg, 2));
1059                 GOTO(out, err = 0);
1060                 break;
1061         }
1062         default: {
1063                 err = obd_process_config(obd, sizeof(*lcfg), lcfg);
1064                 GOTO(out, err);
1065
1066         }
1067         }
1068 out:
1069         if ((err < 0) && !(lcfg->lcfg_command & LCFG_REQUIRED)) {
1070                 CWARN("Ignoring error %d on optional command %#x\n", err,
1071                       lcfg->lcfg_command);
1072                 err = 0;
1073         }
1074         return err;
1075 }
1076
1077 int class_process_proc_param(char *prefix, struct lprocfs_vars *lvars,
1078                              struct lustre_cfg *lcfg, void *data)
1079 {
1080 #ifdef __KERNEL__
1081         struct lprocfs_vars *var;
1082         char *key, *sval;
1083         int i, keylen, vallen;
1084         int matched = 0, j = 0;
1085         int rc = 0;
1086         int skip = 0;
1087         ENTRY;
1088
1089         if (lcfg->lcfg_command != LCFG_PARAM) {
1090                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1091                 RETURN(-EINVAL);
1092         }
1093
1094         /* e.g. tunefs.lustre --param mdt.group_upcall=foo /r/tmp/lustre-mdt
1095            or   lctl conf_param lustre-MDT0000.mdt.group_upcall=bar
1096            or   lctl conf_param lustre-OST0000.osc.max_dirty_mb=36 */
1097         for (i = 1; i < lcfg->lcfg_bufcount; i++) {
1098                 key = lustre_cfg_buf(lcfg, i);
1099                 /* Strip off prefix */
1100                 class_match_param(key, prefix, &key);
1101                 sval = strchr(key, '=');
1102                 if (!sval || (*(sval + 1) == 0)) {
1103                         CERROR("Can't parse param %s (missing '=')\n", key);
1104                         /* rc = -EINVAL;        continue parsing other params */
1105                         continue;
1106                 }
1107                 keylen = sval - key;
1108                 sval++;
1109                 vallen = strlen(sval);
1110                 matched = 0;
1111                 j = 0;
1112                 /* Search proc entries */
1113                 while (lvars[j].name) {
1114                         var = &lvars[j];
1115                         if (class_match_param(key, (char *)var->name, 0) == 0 &&
1116                             keylen == strlen(var->name)) {
1117                                 matched++;
1118                                 rc = -EROFS;
1119                                 if (var->write_fptr) {
1120                                         mm_segment_t oldfs;
1121                                         oldfs = get_fs();
1122                                         set_fs(KERNEL_DS);
1123                                         rc = (var->write_fptr)(NULL, sval,
1124                                                                vallen, data);
1125                                         set_fs(oldfs);
1126                                 }
1127                                 break;
1128                         }
1129                         j++;
1130                 }
1131                 if (!matched) {
1132                         /* If the prefix doesn't match, return error so we
1133                            can pass it down the stack */
1134                         if (strnchr(key, keylen, '.'))
1135                             RETURN(-ENOSYS);
1136                         CERROR("%s: unknown param %s\n",
1137                                (char *)lustre_cfg_string(lcfg, 0), key);
1138                         /* rc = -EINVAL;        continue parsing other params */
1139                         skip++;
1140                 } else if (rc < 0) {
1141                         CERROR("writing proc entry %s err %d\n",
1142                                var->name, rc);
1143                         rc = 0;
1144                 } else {
1145                         CDEBUG(D_CONFIG, "%s.%.*s: set parameter %.*s=%s\n",
1146                                       lustre_cfg_string(lcfg, 0),
1147                                       (int)strlen(prefix) - 1, prefix,
1148                                       (int)(sval - key - 1), key, sval);
1149                 }
1150         }
1151
1152         if (rc > 0)
1153                 rc = 0;
1154         if (!rc && skip)
1155                 rc = skip;
1156         RETURN(rc);
1157 #else
1158         CDEBUG(D_CONFIG, "liblustre can't process params.\n");
1159         /* Don't throw config error */
1160         RETURN(0);
1161 #endif
1162 }
1163
1164 int class_config_dump_handler(struct llog_handle * handle,
1165                               struct llog_rec_hdr *rec, void *data);
1166
1167 #ifdef __KERNEL__
1168 extern int lustre_check_exclusion(struct super_block *sb, char *svname);
1169 #else
1170 #define lustre_check_exclusion(a,b)  0
1171 #endif
1172
1173 /** Parse a configuration llog, doing various manipulations on them
1174  * for various reasons, (modifications for compatibility, skip obsolete
1175  * records, change uuids, etc), then class_process_config() resulting
1176  * net records.
1177  */
1178 static int class_config_llog_handler(struct llog_handle * handle,
1179                                      struct llog_rec_hdr *rec, void *data)
1180 {
1181         struct config_llog_instance *clli = data;
1182         int cfg_len = rec->lrh_len;
1183         char *cfg_buf = (char*) (rec + 1);
1184         int rc = 0;
1185         ENTRY;
1186
1187         //class_config_dump_handler(handle, rec, data);
1188
1189         switch (rec->lrh_type) {
1190         case OBD_CFG_REC: {
1191                 struct lustre_cfg *lcfg, *lcfg_new;
1192                 struct lustre_cfg_bufs bufs;
1193                 char *inst_name = NULL;
1194                 int inst_len = 0;
1195                 int inst = 0, swab = 0;
1196
1197                 lcfg = (struct lustre_cfg *)cfg_buf;
1198                 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
1199                         lustre_swab_lustre_cfg(lcfg);
1200                         swab = 1;
1201                 }
1202
1203                 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
1204                 if (rc)
1205                         GOTO(out, rc);
1206
1207                 /* Figure out config state info */
1208                 if (lcfg->lcfg_command == LCFG_MARKER) {
1209                         struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
1210                         lustre_swab_cfg_marker(marker, swab,
1211                                                LUSTRE_CFG_BUFLEN(lcfg, 1));
1212                         CDEBUG(D_CONFIG, "Marker, inst_flg=%#x mark_flg=%#x\n",
1213                                clli->cfg_flags, marker->cm_flags);
1214                         if (marker->cm_flags & CM_START) {
1215                                 /* all previous flags off */
1216                                 clli->cfg_flags = CFG_F_MARKER;
1217                                 if (marker->cm_flags & CM_SKIP) {
1218                                         clli->cfg_flags |= CFG_F_SKIP;
1219                                         CDEBUG(D_CONFIG, "SKIP #%d\n",
1220                                                marker->cm_step);
1221                                 } else if ((marker->cm_flags & CM_EXCLUDE) ||
1222                                            (clli->cfg_sb &&
1223                                             lustre_check_exclusion(clli->cfg_sb,
1224                                                          marker->cm_tgtname))) {
1225                                         clli->cfg_flags |= CFG_F_EXCLUDE;
1226                                         CDEBUG(D_CONFIG, "EXCLUDE %d\n",
1227                                                marker->cm_step);
1228                                 }
1229                         } else if (marker->cm_flags & CM_END) {
1230                                 clli->cfg_flags = 0;
1231                         }
1232                 }
1233                 /* A config command without a start marker before it is
1234                    illegal (post 146) */
1235                 if (!(clli->cfg_flags & CFG_F_COMPAT146) &&
1236                     !(clli->cfg_flags & CFG_F_MARKER) &&
1237                     (lcfg->lcfg_command != LCFG_MARKER)) {
1238                         CWARN("Config not inside markers, ignoring! "
1239                               "(inst: %p, uuid: %s, flags: %#x)\n",
1240                               clli->cfg_instance,
1241                               clli->cfg_uuid.uuid, clli->cfg_flags);
1242                         clli->cfg_flags |= CFG_F_SKIP;
1243                 }
1244                 if (clli->cfg_flags & CFG_F_SKIP) {
1245                         CDEBUG(D_CONFIG, "skipping %#x\n",
1246                                clli->cfg_flags);
1247                         rc = 0;
1248                         /* No processing! */
1249                         break;
1250                 }
1251
1252                 /*
1253                  * For interoperability between 1.8 and 2.0,
1254                  * rename "mds" obd device type to "mdt".
1255                  */
1256                 {
1257                         char *typename = lustre_cfg_string(lcfg, 1);
1258                         char *index = lustre_cfg_string(lcfg, 2);
1259
1260                         if ((lcfg->lcfg_command == LCFG_ATTACH && typename &&
1261                              strcmp(typename, "mds") == 0)) {
1262                                 CWARN("For 1.8 interoperability, rename obd "
1263                                        "type from mds to mdt\n");
1264                                 typename[2] = 't';
1265                         }
1266                         if ((lcfg->lcfg_command == LCFG_SETUP && index &&
1267                              strcmp(index, "type") == 0)) {
1268                                 CDEBUG(D_INFO, "For 1.8 interoperability, "
1269                                        "set this index to '0'\n");
1270                                 index[0] = '0';
1271                                 index[1] = 0;
1272                         }
1273                 }
1274
1275                 if ((clli->cfg_flags & CFG_F_EXCLUDE) &&
1276                     (lcfg->lcfg_command == LCFG_LOV_ADD_OBD))
1277                         /* Add inactive instead */
1278                         lcfg->lcfg_command = LCFG_LOV_ADD_INA;
1279
1280                 lustre_cfg_bufs_init(&bufs, lcfg);
1281
1282                 if (clli && clli->cfg_instance &&
1283                     LUSTRE_CFG_BUFLEN(lcfg, 0) > 0){
1284                         inst = 1;
1285                         inst_len = LUSTRE_CFG_BUFLEN(lcfg, 0) +
1286                                    sizeof(clli->cfg_instance) * 2 + 1;
1287                         OBD_ALLOC(inst_name, inst_len);
1288                         if (inst_name == NULL)
1289                                 GOTO(out, rc = -ENOMEM);
1290                         sprintf(inst_name, "%s-%p",
1291                                 lustre_cfg_string(lcfg, 0),
1292                                 clli->cfg_instance);
1293                         lustre_cfg_bufs_set_string(&bufs, 0, inst_name);
1294                         CDEBUG(D_CONFIG, "cmd %x, instance name: %s\n",
1295                                lcfg->lcfg_command, inst_name);
1296                 }
1297
1298                 /* we override the llog's uuid for clients, to insure they
1299                 are unique */
1300                 if (clli && clli->cfg_instance != NULL &&
1301                     lcfg->lcfg_command == LCFG_ATTACH) {
1302                         lustre_cfg_bufs_set_string(&bufs, 2,
1303                                                    clli->cfg_uuid.uuid);
1304                 }
1305                 /*
1306                  * sptlrpc config record, we expect 2 data segments:
1307                  *  [0]: fs_name/target_name,
1308                  *  [1]: rule string
1309                  * moving them to index [1] and [2], and insert MGC's
1310                  * obdname at index [0].
1311                  */
1312                 if (clli && clli->cfg_instance == NULL &&
1313                     lcfg->lcfg_command == LCFG_SPTLRPC_CONF) {
1314                         lustre_cfg_bufs_set(&bufs, 2, bufs.lcfg_buf[1],
1315                                             bufs.lcfg_buflen[1]);
1316                         lustre_cfg_bufs_set(&bufs, 1, bufs.lcfg_buf[0],
1317                                             bufs.lcfg_buflen[0]);
1318                         lustre_cfg_bufs_set_string(&bufs, 0,
1319                                                    clli->cfg_obdname);
1320                 }
1321
1322                 lcfg_new = lustre_cfg_new(lcfg->lcfg_command, &bufs);
1323
1324                 lcfg_new->lcfg_num   = lcfg->lcfg_num;
1325                 lcfg_new->lcfg_flags = lcfg->lcfg_flags;
1326
1327                 /* XXX Hack to try to remain binary compatible with
1328                  * pre-newconfig logs */
1329                 if (lcfg->lcfg_nal != 0 &&      /* pre-newconfig log? */
1330                     (lcfg->lcfg_nid >> 32) == 0) {
1331                         __u32 addr = (__u32)(lcfg->lcfg_nid & 0xffffffff);
1332
1333                         lcfg_new->lcfg_nid =
1334                                 LNET_MKNID(LNET_MKNET(lcfg->lcfg_nal, 0), addr);
1335                         CWARN("Converted pre-newconfig NAL %d NID %x to %s\n",
1336                               lcfg->lcfg_nal, addr,
1337                               libcfs_nid2str(lcfg_new->lcfg_nid));
1338                 } else {
1339                         lcfg_new->lcfg_nid = lcfg->lcfg_nid;
1340                 }
1341
1342                 lcfg_new->lcfg_nal = 0; /* illegal value for obsolete field */
1343
1344                 rc = class_process_config(lcfg_new);
1345                 lustre_cfg_free(lcfg_new);
1346
1347                 if (inst)
1348                         OBD_FREE(inst_name, inst_len);
1349                 break;
1350         }
1351         default:
1352                 CERROR("Unknown llog record type %#x encountered\n",
1353                        rec->lrh_type);
1354                 break;
1355         }
1356 out:
1357         if (rc) {
1358                 CERROR("Err %d on cfg command:\n", rc);
1359                 class_config_dump_handler(handle, rec, data);
1360         }
1361         RETURN(rc);
1362 }
1363
1364 int class_config_parse_llog(struct llog_ctxt *ctxt, char *name,
1365                             struct config_llog_instance *cfg)
1366 {
1367         struct llog_process_cat_data cd = {0, 0};
1368         struct llog_handle *llh;
1369         int rc, rc2;
1370         ENTRY;
1371
1372         CDEBUG(D_INFO, "looking up llog %s\n", name);
1373         rc = llog_create(ctxt, &llh, NULL, name);
1374         if (rc)
1375                 RETURN(rc);
1376
1377         rc = llog_init_handle(llh, LLOG_F_IS_PLAIN, NULL);
1378         if (rc)
1379                 GOTO(parse_out, rc);
1380
1381         /* continue processing from where we last stopped to end-of-log */
1382         if (cfg)
1383                 cd.lpcd_first_idx = cfg->cfg_last_idx;
1384         cd.lpcd_last_idx = 0;
1385
1386         rc = llog_process(llh, class_config_llog_handler, cfg, &cd);
1387
1388         CDEBUG(D_CONFIG, "Processed log %s gen %d-%d (rc=%d)\n", name,
1389                cd.lpcd_first_idx + 1, cd.lpcd_last_idx, rc);
1390
1391         if (cfg)
1392                 cfg->cfg_last_idx = cd.lpcd_last_idx;
1393
1394 parse_out:
1395         rc2 = llog_close(llh);
1396         if (rc == 0)
1397                 rc = rc2;
1398
1399         RETURN(rc);
1400 }
1401
1402 int class_config_dump_handler(struct llog_handle * handle,
1403                               struct llog_rec_hdr *rec, void *data)
1404 {
1405         int cfg_len = rec->lrh_len;
1406         char *cfg_buf = (char*) (rec + 1);
1407         char *outstr, *ptr, *end;
1408         int rc = 0;
1409         ENTRY;
1410
1411         OBD_ALLOC(outstr, 256);
1412         end = outstr + 256;
1413         ptr = outstr;
1414         if (!outstr) {
1415                 RETURN(-ENOMEM);
1416         }
1417         if (rec->lrh_type == OBD_CFG_REC) {
1418                 struct lustre_cfg *lcfg;
1419                 int i;
1420
1421                 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
1422                 if (rc)
1423                         GOTO(out, rc);
1424                 lcfg = (struct lustre_cfg *)cfg_buf;
1425
1426                 ptr += snprintf(ptr, end-ptr, "cmd=%05x ",
1427                                 lcfg->lcfg_command);
1428                 if (lcfg->lcfg_flags) {
1429                         ptr += snprintf(ptr, end-ptr, "flags=%#08x ",
1430                                         lcfg->lcfg_flags);
1431                 }
1432                 if (lcfg->lcfg_num) {
1433                         ptr += snprintf(ptr, end-ptr, "num=%#08x ",
1434                                         lcfg->lcfg_num);
1435                 }
1436                 if (lcfg->lcfg_nid) {
1437                         ptr += snprintf(ptr, end-ptr, "nid=%s("LPX64")\n     ",
1438                                         libcfs_nid2str(lcfg->lcfg_nid),
1439                                         lcfg->lcfg_nid);
1440                 }
1441                 if (lcfg->lcfg_command == LCFG_MARKER) {
1442                         struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
1443                         ptr += snprintf(ptr, end-ptr, "marker=%d(%#x)%s '%s'",
1444                                         marker->cm_step, marker->cm_flags,
1445                                         marker->cm_tgtname, marker->cm_comment);
1446                 } else {
1447                         for (i = 0; i <  lcfg->lcfg_bufcount; i++) {
1448                                 ptr += snprintf(ptr, end-ptr, "%d:%s  ", i,
1449                                                 lustre_cfg_string(lcfg, i));
1450                         }
1451                 }
1452                 LCONSOLE(D_WARNING, "   %s\n", outstr);
1453         } else {
1454                 LCONSOLE(D_WARNING, "unhandled lrh_type: %#x\n", rec->lrh_type);
1455                 rc = -EINVAL;
1456         }
1457 out:
1458         OBD_FREE(outstr, 256);
1459         RETURN(rc);
1460 }
1461
1462 int class_config_dump_llog(struct llog_ctxt *ctxt, char *name,
1463                            struct config_llog_instance *cfg)
1464 {
1465         struct llog_handle *llh;
1466         int rc, rc2;
1467         ENTRY;
1468
1469         LCONSOLE_INFO("Dumping config log %s\n", name);
1470
1471         rc = llog_create(ctxt, &llh, NULL, name);
1472         if (rc)
1473                 RETURN(rc);
1474
1475         rc = llog_init_handle(llh, LLOG_F_IS_PLAIN, NULL);
1476         if (rc)
1477                 GOTO(parse_out, rc);
1478
1479         rc = llog_process(llh, class_config_dump_handler, cfg, NULL);
1480 parse_out:
1481         rc2 = llog_close(llh);
1482         if (rc == 0)
1483                 rc = rc2;
1484
1485         LCONSOLE_INFO("End config log %s\n", name);
1486         RETURN(rc);
1487
1488 }
1489
1490 /** Call class_cleanup and class_detach.
1491  * "Manual" only in the sense that we're faking lcfg commands.
1492  */
1493 int class_manual_cleanup(struct obd_device *obd)
1494 {
1495         char                    flags[3] = "";
1496         struct lustre_cfg      *lcfg;
1497         struct lustre_cfg_bufs  bufs;
1498         int                     rc;
1499         ENTRY;
1500
1501         if (!obd) {
1502                 CERROR("empty cleanup\n");
1503                 RETURN(-EALREADY);
1504         }
1505
1506         if (obd->obd_force)
1507                 strcat(flags, "F");
1508         if (obd->obd_fail)
1509                 strcat(flags, "A");
1510
1511         CDEBUG(D_CONFIG, "Manual cleanup of %s (flags='%s')\n",
1512                obd->obd_name, flags);
1513
1514         lustre_cfg_bufs_reset(&bufs, obd->obd_name);
1515         lustre_cfg_bufs_set_string(&bufs, 1, flags);
1516         lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
1517         if (!lcfg)
1518                 RETURN(-ENOMEM);
1519
1520         rc = class_process_config(lcfg);
1521         if (rc) {
1522                 CERROR("cleanup failed %d: %s\n", rc, obd->obd_name);
1523                 GOTO(out, rc);
1524         }
1525
1526         /* the lcfg is almost the same for both ops */
1527         lcfg->lcfg_command = LCFG_DETACH;
1528         rc = class_process_config(lcfg);
1529         if (rc)
1530                 CERROR("detach failed %d: %s\n", rc, obd->obd_name);
1531 out:
1532         lustre_cfg_free(lcfg);
1533         RETURN(rc);
1534 }
1535
1536 /*
1537  * uuid<->export lustre hash operations
1538  */
1539
1540 static unsigned
1541 uuid_hash(cfs_hash_t *hs, const void *key, unsigned mask)
1542 {
1543         return cfs_hash_djb2_hash(((struct obd_uuid *)key)->uuid,
1544                                   sizeof(((struct obd_uuid *)key)->uuid), mask);
1545 }
1546
1547 static void *
1548 uuid_key(cfs_hlist_node_t *hnode)
1549 {
1550         struct obd_export *exp;
1551
1552         exp = cfs_hlist_entry(hnode, struct obd_export, exp_uuid_hash);
1553
1554         return &exp->exp_client_uuid;
1555 }
1556
1557 /*
1558  * NOTE: It is impossible to find an export that is in failed
1559  *       state with this function
1560  */
1561 static int
1562 uuid_keycmp(const void *key, cfs_hlist_node_t *hnode)
1563 {
1564         struct obd_export *exp;
1565
1566         LASSERT(key);
1567         exp = cfs_hlist_entry(hnode, struct obd_export, exp_uuid_hash);
1568
1569         return obd_uuid_equals(key, &exp->exp_client_uuid) &&
1570                !exp->exp_failed;
1571 }
1572
1573 static void *
1574 uuid_export_object(cfs_hlist_node_t *hnode)
1575 {
1576         return cfs_hlist_entry(hnode, struct obd_export, exp_uuid_hash);
1577 }
1578
1579 static void
1580 uuid_export_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
1581 {
1582         struct obd_export *exp;
1583
1584         exp = cfs_hlist_entry(hnode, struct obd_export, exp_uuid_hash);
1585         class_export_get(exp);
1586 }
1587
1588 static void
1589 uuid_export_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
1590 {
1591         struct obd_export *exp;
1592
1593         exp = cfs_hlist_entry(hnode, struct obd_export, exp_uuid_hash);
1594         class_export_put(exp);
1595 }
1596
1597 static cfs_hash_ops_t uuid_hash_ops = {
1598         .hs_hash        = uuid_hash,
1599         .hs_key         = uuid_key,
1600         .hs_keycmp      = uuid_keycmp,
1601         .hs_object      = uuid_export_object,
1602         .hs_get         = uuid_export_get,
1603         .hs_put_locked  = uuid_export_put_locked,
1604 };
1605
1606
1607 /*
1608  * nid<->export hash operations
1609  */
1610
1611 static unsigned
1612 nid_hash(cfs_hash_t *hs, const void *key, unsigned mask)
1613 {
1614         return cfs_hash_djb2_hash(key, sizeof(lnet_nid_t), mask);
1615 }
1616
1617 static void *
1618 nid_key(cfs_hlist_node_t *hnode)
1619 {
1620         struct obd_export *exp;
1621
1622         exp = cfs_hlist_entry(hnode, struct obd_export, exp_nid_hash);
1623
1624         RETURN(&exp->exp_connection->c_peer.nid);
1625 }
1626
1627 /*
1628  * NOTE: It is impossible to find an export that is in failed
1629  *       state with this function
1630  */
1631 static int
1632 nid_kepcmp(const void *key, cfs_hlist_node_t *hnode)
1633 {
1634         struct obd_export *exp;
1635
1636         LASSERT(key);
1637         exp = cfs_hlist_entry(hnode, struct obd_export, exp_nid_hash);
1638
1639         RETURN(exp->exp_connection->c_peer.nid == *(lnet_nid_t *)key &&
1640                !exp->exp_failed);
1641 }
1642
1643 static void *
1644 nid_export_object(cfs_hlist_node_t *hnode)
1645 {
1646         return cfs_hlist_entry(hnode, struct obd_export, exp_nid_hash);
1647 }
1648
1649 static void
1650 nid_export_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
1651 {
1652         struct obd_export *exp;
1653
1654         exp = cfs_hlist_entry(hnode, struct obd_export, exp_nid_hash);
1655         class_export_get(exp);
1656 }
1657
1658 static void
1659 nid_export_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
1660 {
1661         struct obd_export *exp;
1662
1663         exp = cfs_hlist_entry(hnode, struct obd_export, exp_nid_hash);
1664         class_export_put(exp);
1665 }
1666
1667 static cfs_hash_ops_t nid_hash_ops = {
1668         .hs_hash        = nid_hash,
1669         .hs_key         = nid_key,
1670         .hs_keycmp      = nid_kepcmp,
1671         .hs_object      = nid_export_object,
1672         .hs_get         = nid_export_get,
1673         .hs_put_locked  = nid_export_put_locked,
1674 };
1675
1676
1677 /*
1678  * nid<->nidstats hash operations
1679  */
1680
1681 static void *
1682 nidstats_key(cfs_hlist_node_t *hnode)
1683 {
1684         struct nid_stat *ns;
1685
1686         ns = cfs_hlist_entry(hnode, struct nid_stat, nid_hash);
1687
1688         return &ns->nid;
1689 }
1690
1691 static int
1692 nidstats_keycmp(const void *key, cfs_hlist_node_t *hnode)
1693 {
1694         return *(lnet_nid_t *)nidstats_key(hnode) == *(lnet_nid_t *)key;
1695 }
1696
1697 static void *
1698 nidstats_object(cfs_hlist_node_t *hnode)
1699 {
1700         return cfs_hlist_entry(hnode, struct nid_stat, nid_hash);
1701 }
1702
1703 static void
1704 nidstats_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
1705 {
1706         struct nid_stat *ns;
1707
1708         ns = cfs_hlist_entry(hnode, struct nid_stat, nid_hash);
1709         nidstat_getref(ns);
1710 }
1711
1712 static void
1713 nidstats_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
1714 {
1715         struct nid_stat *ns;
1716
1717         ns = cfs_hlist_entry(hnode, struct nid_stat, nid_hash);
1718         nidstat_putref(ns);
1719 }
1720
1721 static cfs_hash_ops_t nid_stat_hash_ops = {
1722         .hs_hash        = nid_hash,
1723         .hs_key         = nidstats_key,
1724         .hs_keycmp      = nidstats_keycmp,
1725         .hs_object      = nidstats_object,
1726         .hs_get         = nidstats_get,
1727         .hs_put_locked  = nidstats_put_locked,
1728 };