lustre/lmv/lmv_obd.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of Lustre, http://www.lustre.org.
   7  *
   8  *   Lustre is free software; you can redistribute it and/or
   9  *   modify it under the terms of version 2 of the GNU General Public
  10  *   License as published by the Free Software Foundation.
  11  *
  12  *   Lustre is distributed in the hope that it will be useful,
  13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *   GNU General Public License for more details.
  16  *
  17  *   You should have received a copy of the GNU General Public License
  18  *   along with Lustre; if not, write to the Free Software
  19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  */
  21
  22 #ifndef EXPORT_SYMTAB
  23 # define EXPORT_SYMTAB
  24 #endif
  25 #define DEBUG_SUBSYSTEM S_LMV
  26 #ifdef __KERNEL__
  27 #include <linux/slab.h>
  28 #include <linux/module.h>
  29 #include <linux/init.h>
  30 #include <linux/slab.h>
  31 #include <linux/pagemap.h>
  32 #include <asm/div64.h>
  33 #include <linux/seq_file.h>
  34 #include <linux/namei.h>
  35 #else
  36 #include <liblustre.h>
  37 #endif
  38 #include <linux/ext2_fs.h>
  39
  40 #include <linux/obd_support.h>
  41 #include <linux/lustre_lib.h>
  42 #include <linux/lustre_net.h>
  43 #include <linux/lustre_idl.h>
  44 #include <linux/lustre_dlm.h>
  45 #include <linux/lustre_mds.h>
  46 #include <linux/obd_class.h>
  47 #include <linux/obd_ost.h>
  48 #include <linux/lprocfs_status.h>
  49 #include <linux/lustre_fsfilt.h>
  50 #include <linux/obd_lmv.h>
  51 #include <linux/lustre_lite.h>
  52 #include <linux/lustre_audit.h>
  53 #include "lmv_internal.h"
  54
  55 /* not defined for liblustre building */
  56 #if !defined(ATOMIC_INIT)
  57 #define ATOMIC_INIT(val) { (val) }
  58 #endif
  59
  60 /* object cache. */
  61 kmem_cache_t *obj_cache;
  62 atomic_t obj_cache_count = ATOMIC_INIT(0);
  63
  64 static void lmv_activate_target(struct lmv_obd *lmv,
  65                                 struct lmv_tgt_desc *tgt,
  66                                 int activate)
  67 {
  68         if (tgt->active == activate)
  69                 return;
  70
  71         tgt->active = activate;
  72         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  73 }
  74
  75 /* Error codes:
  76  *
  77  *  -EINVAL  : UUID can't be found in the LMV's target list
  78  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  79  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  80  */
  81 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
  82                               int activate)
  83 {
  84         struct lmv_tgt_desc *tgt;
  85         struct obd_device *obd;
  86         int i, rc = 0;
  87         ENTRY;
  88
  89         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  90                lmv, uuid->uuid, activate);
  91
  92         spin_lock(&lmv->lmv_lock);
  93         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
  94                 if (tgt->ltd_exp == NULL)
  95                         continue;
  96
  97                 CDEBUG(D_INFO, "lmv idx %d is %s conn "LPX64"\n",
  98                        i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
  99
 100                 if (obd_uuid_equals(uuid, &tgt->uuid))
 101                         break;
 102         }
 103
 104         if (i == lmv->desc.ld_tgt_count)
 105                 GOTO(out_lmv_lock, rc = -EINVAL);
 106
 107         obd = class_exp2obd(tgt->ltd_exp);
 108         if (obd == NULL)
 109                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 110
 111         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 112                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 113                obd->obd_type->typ_name, i);
 114         LASSERT(strcmp(obd->obd_type->typ_name, OBD_MDC_DEVICENAME) == 0);
 115
 116         if (tgt->active == activate) {
 117                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 118                        activate ? "" : "in");
 119                 GOTO(out_lmv_lock, rc);
 120         }
 121
 122         CDEBUG(D_INFO, "Marking OBD %p %sactive\n",
 123                obd, activate ? "" : "in");
 124
 125         lmv_activate_target(lmv, tgt, activate);
 126
 127         EXIT;
 128
 129  out_lmv_lock:
 130         spin_unlock(&lmv->lmv_lock);
 131         return rc;
 132 }
 133
 134 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 135                       int active, void *data)
 136 {
 137         struct obd_uuid *uuid;
 138         int rc;
 139         ENTRY;
 140
 141         if (strcmp(watched->obd_type->typ_name, OBD_MDC_DEVICENAME)) {
 142                 CERROR("unexpected notification of %s %s!\n",
 143                        watched->obd_type->typ_name,
 144                        watched->obd_name);
 145                 RETURN(-EINVAL);
 146         }
 147         uuid = &watched->u.cli.cl_import->imp_target_uuid;
 148
 149         /* Set MDC as active before notifying the observer, so the observer can
 150          * use the MDC normally.
 151          */
 152         rc = lmv_set_mdc_active(&obd->u.lmv, uuid, active);
 153         if (rc) {
 154                 CERROR("%sactivation of %s failed: %d\n",
 155                        active ? "" : "de", uuid->uuid, rc);
 156                 RETURN(rc);
 157         }
 158
 159         if (obd->obd_observer)
 160                 /* Pass the notification up the chain. */
 161                 rc = obd_notify(obd->obd_observer, watched, active, data);
 162
 163         RETURN(rc);
 164 }
 165
 166 static int lmv_attach(struct obd_device *dev, obd_count len, void *data)
 167 {
 168         struct lprocfs_static_vars lvars;
 169         int rc;
 170         ENTRY;
 171
 172         lprocfs_init_vars(lmv, &lvars);
 173         rc = lprocfs_obd_attach(dev, lvars.obd_vars);
 174 #ifdef __KERNEL__
 175         if (rc == 0) {
 176                 struct proc_dir_entry *entry;
 177
 178                 entry = create_proc_entry("target_obd_status", 0444,
 179                                            dev->obd_proc_entry);
 180                 if (entry == NULL)
 181                         RETURN(-ENOMEM);
 182                 entry->proc_fops = &lmv_proc_target_fops;
 183                 entry->data = dev;
 184        }
 185 #endif
 186         RETURN (rc);
 187 }
 188
 189 static int lmv_detach(struct obd_device *dev)
 190 {
 191         return lprocfs_obd_detach(dev);
 192 }
 193
 194 /* this is fake connect function. Its purpose is to initialize lmv and say
 195  * caller that everything is okay. Real connection will be performed later. */
 196 static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd,
 197                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 198                        unsigned long flags)
 199 {
 200 #ifdef __KERNEL__
 201         struct proc_dir_entry *lmv_proc_dir;
 202 #endif
 203         struct lmv_obd *lmv = &obd->u.lmv;
 204         struct obd_export *exp;
 205         int rc = 0;
 206         ENTRY;
 207
 208         rc = class_connect(conn, obd, cluuid);
 209         if (rc) {
 210                 CERROR("class_connection() returned %d\n", rc);
 211                 RETURN(rc);
 212         }
 213
 214         exp = class_conn2export(conn);
 215
 216         /* we don't want to actually do the underlying connections more than
 217          * once, so keep track. */
 218         lmv->refcount++;
 219         if (lmv->refcount > 1) {
 220                 class_export_put(exp);
 221                 RETURN(0);
 222         }
 223
 224         lmv->exp = exp;
 225         lmv->connected = 0;
 226         lmv->cluuid = *cluuid;
 227         lmv->connect_flags = flags;
 228         if (data)
 229                 memcpy(&lmv->conn_data, data, sizeof(*data));
 230
 231 #ifdef __KERNEL__
 232         lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
 233                                         NULL, NULL);
 234         if (IS_ERR(lmv_proc_dir)) {
 235                 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 236                        obd->obd_type->typ_name, obd->obd_name);
 237                 lmv_proc_dir = NULL;
 238         }
 239 #endif
 240
 241         /*
 242          * all real clients should perform actual connection rightaway, because
 243          * it is possible, that LMV will not have opportunity to connect
 244          * targets, as MDC stuff will bit called directly, for instance while
 245          * reading ../mdc/../kbytesfree procfs file, etc.
 246          */
 247         if (flags & OBD_OPT_REAL_CLIENT)
 248                 rc = lmv_check_connect(obd);
 249
 250 #ifdef __KERNEL__
 251         if (rc) {
 252                 if (lmv_proc_dir)
 253                         lprocfs_remove(lmv_proc_dir);
 254         }
 255 #endif
 256
 257         RETURN(rc);
 258 }
 259
 260 static void lmv_set_timeouts(struct obd_device *obd)
 261 {
 262         struct lmv_tgt_desc *tgts;
 263         struct lmv_obd *lmv;
 264         int i;
 265
 266         lmv = &obd->u.lmv;
 267         if (lmv->server_timeout == 0)
 268                 return;
 269
 270         if (lmv->connected == 0)
 271                 return;
 272
 273         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
 274                 if (tgts->ltd_exp == NULL)
 275                         continue;
 276
 277                 obd_set_info(tgts->ltd_exp, strlen("inter_mds"),
 278                              "inter_mds", 0, NULL);
 279         }
 280 }
 281
 282 static int lmv_init_ea_size(struct obd_export *exp, int easize,
 283                             int cookiesize)
 284 {
 285         struct obd_device *obd = exp->exp_obd;
 286         struct lmv_obd *lmv = &obd->u.lmv;
 287         int i, rc = 0, change = 0;
 288         ENTRY;
 289
 290         if (lmv->max_easize < easize) {
 291                 lmv->max_easize = easize;
 292                 change = 1;
 293         }
 294         if (lmv->max_cookiesize < cookiesize) {
 295                 lmv->max_cookiesize = cookiesize;
 296                 change = 1;
 297         }
 298         if (change == 0)
 299                 RETURN(0);
 300
 301         if (lmv->connected == 0)
 302                 RETURN(0);
 303
 304         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 305                 if (lmv->tgts[i].ltd_exp == NULL) {
 306                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 307                         continue;
 308                 }
 309
 310                 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
 311                 if (rc) {
 312                         CERROR("obd_init_ea_size() failed on MDT target %d, "
 313                                "error %d.\n", i, rc);
 314                         break;
 315                 }
 316         }
 317         RETURN(rc);
 318 }
 319
 320 #define MAX_STRING_SIZE 128
 321
 322 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 323 {
 324         struct lmv_obd *lmv = &obd->u.lmv;
 325         struct obd_uuid *cluuid = &lmv->cluuid;
 326         struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
 327         struct lustre_handle conn = {0, };
 328         struct obd_device *mdc_obd;
 329         struct obd_export *mdc_exp;
 330         int rc;
 331 #ifdef __KERNEL__
 332         struct proc_dir_entry *lmv_proc_dir;
 333 #endif
 334         ENTRY;
 335
 336         /* for MDS: don't connect to yourself */
 337         if (obd_uuid_equals(&tgt->uuid, cluuid)) {
 338                 CDEBUG(D_CONFIG, "don't connect back to %s\n", cluuid->uuid);
 339                 /* XXX - the old code didn't increment active tgt count.
 340                  *       should we ? */
 341                 RETURN(0);
 342         }
 343
 344         mdc_obd = class_find_client_obd(&tgt->uuid, OBD_MDC_DEVICENAME,
 345                                         &obd->obd_uuid);
 346         if (!mdc_obd) {
 347                 CERROR("target %s not attached\n", tgt->uuid.uuid);
 348                 RETURN(-EINVAL);
 349         }
 350
 351         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 352                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 353                 tgt->uuid.uuid, obd->obd_uuid.uuid,
 354                 cluuid->uuid);
 355
 356         if (!mdc_obd->obd_set_up) {
 357                 CERROR("target %s not set up\n", tgt->uuid.uuid);
 358                 RETURN(-EINVAL);
 359         }
 360
 361         rc = obd_connect(&conn, mdc_obd, &lmv_mdc_uuid, &lmv->conn_data,
 362                          lmv->connect_flags);
 363         if (rc) {
 364                 CERROR("target %s connect error %d\n", tgt->uuid.uuid, rc);
 365                 RETURN(rc);
 366         }
 367
 368         mdc_exp = class_conn2export(&conn);
 369
 370         rc = obd_register_observer(mdc_obd, obd);
 371         if (rc) {
 372                 obd_disconnect(mdc_exp, 0);
 373                 CERROR("target %s register_observer error %d\n",
 374                        tgt->uuid.uuid, rc);
 375                 RETURN(rc);
 376         }
 377
 378         if (obd->obd_observer) {
 379                 /* tell the mds_lmv about the new target */
 380                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd, 1,
 381                                 (void *)(tgt - lmv->tgts));
 382                 if (rc) {
 383                         obd_disconnect(mdc_exp, 0);
 384                         RETURN(rc);
 385                 }
 386         }
 387
 388         tgt->ltd_exp = mdc_exp;
 389         tgt->active = 1;
 390         lmv->desc.ld_active_tgt_count++;
 391
 392         obd_init_ea_size(tgt->ltd_exp, lmv->max_easize,
 393                          lmv->max_cookiesize);
 394         CDEBUG(D_CONFIG, "connected to %s(%s) successfully (%d)\n",
 395                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 396                 atomic_read(&obd->obd_refcount));
 397
 398 #ifdef __KERNEL__
 399         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 400         if (lmv_proc_dir) {
 401                 struct proc_dir_entry *mdc_symlink;
 402                 char name[MAX_STRING_SIZE + 1];
 403
 404                 LASSERT(mdc_obd->obd_type != NULL);
 405                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 406                 name[MAX_STRING_SIZE] = '\0';
 407                 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
 408                          mdc_obd->obd_type->typ_name,
 409                          mdc_obd->obd_name);
 410                 mdc_symlink = proc_symlink(mdc_obd->obd_name,
 411                                            lmv_proc_dir, name);
 412                 if (mdc_symlink == NULL) {
 413                         CERROR("could not register LMV target "
 414                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 415                                obd->obd_type->typ_name, obd->obd_name,
 416                                mdc_obd->obd_name);
 417                         lprocfs_remove(lmv_proc_dir);
 418                         lmv_proc_dir = NULL;
 419                 }
 420         }
 421 #endif
 422         RETURN(0);
 423 }
 424
 425 int lmv_add_mdc(struct obd_device *obd, struct obd_uuid *tgt_uuid)
 426 {
 427         struct lmv_obd *lmv = &obd->u.lmv;
 428         struct lmv_tgt_desc *tgt;
 429         int rc = 0;
 430         ENTRY;
 431
 432         CDEBUG(D_CONFIG, "tgt_uuid: %s.\n", tgt_uuid->uuid);
 433
 434         lmv_init_lock(lmv);
 435
 436         if (lmv->desc.ld_active_tgt_count >= LMV_MAX_TGT_COUNT) {
 437                 lmv_init_unlock(lmv);
 438                 CERROR("can't add %s, LMV module compiled for %d MDCs. "
 439                        "That many MDCs already configured.\n",
 440                        tgt_uuid->uuid, LMV_MAX_TGT_COUNT);
 441                 RETURN(-EINVAL);
 442         }
 443         if (lmv->desc.ld_tgt_count == 0) {
 444                 struct obd_device *mdc_obd;
 445
 446                 mdc_obd = class_find_client_obd(tgt_uuid, OBD_MDC_DEVICENAME,
 447                                                 &obd->obd_uuid);
 448                 if (!mdc_obd) {
 449                         lmv_init_unlock(lmv);
 450                         CERROR("Target %s not attached\n", tgt_uuid->uuid);
 451                         RETURN(-EINVAL);
 452                 }
 453
 454                 rc = obd_llog_init(obd, &obd->obd_llogs, mdc_obd, 0, NULL);
 455                 if (rc) {
 456                         lmv_init_unlock(lmv);
 457                         CERROR("lmv failed to setup llogging subsystems\n");
 458                 }
 459         }
 460         spin_lock(&lmv->lmv_lock);
 461         tgt = lmv->tgts + lmv->desc.ld_tgt_count++;
 462         tgt->uuid = *tgt_uuid;
 463         spin_unlock(&lmv->lmv_lock);
 464
 465         if (lmv->connected) {
 466                 rc = lmv_connect_mdc(obd, tgt);
 467                 if (rc) {
 468                         spin_lock(&lmv->lmv_lock);
 469                         lmv->desc.ld_tgt_count--;
 470                         memset(tgt, 0, sizeof(*tgt));
 471                         spin_unlock(&lmv->lmv_lock);
 472                 } else {
 473                         int easize = sizeof(struct mea) +
 474                                      lmv->desc.ld_tgt_count *
 475                                      sizeof(struct lustre_id);
 476                         lmv_init_ea_size(obd->obd_self_export, easize, 0);
 477                 }
 478         }
 479
 480         lmv_init_unlock(lmv);
 481         RETURN(rc);
 482 }
 483
 484 /* performs a check if passed obd is connected. If no - connect it. */
 485 int lmv_check_connect(struct obd_device *obd)
 486 {
 487         struct lmv_obd *lmv = &obd->u.lmv;
 488         struct lmv_tgt_desc *tgt;
 489         int i, rc, easize;
 490         ENTRY;
 491
 492         if (lmv->connected)
 493                 RETURN(0);
 494
 495         lmv_init_lock(lmv);
 496         if (lmv->connected) {
 497                 lmv_init_unlock(lmv);
 498                 RETURN(0);
 499         }
 500
 501         if (lmv->desc.ld_tgt_count == 0) {
 502                 CERROR("%s: no targets configured.\n", obd->obd_name);
 503                 RETURN(-EINVAL);
 504         }
 505
 506         CDEBUG(D_CONFIG, "time to connect %s to %s\n",
 507                lmv->cluuid.uuid, obd->obd_name);
 508
 509         LASSERT(lmv->tgts != NULL);
 510
 511         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
 512                 rc = lmv_connect_mdc(obd, tgt);
 513                 if (rc)
 514                         GOTO(out_disc, rc);
 515         }
 516
 517         lmv_set_timeouts(obd);
 518         class_export_put(lmv->exp);
 519         lmv->connected = 1;
 520         easize = lmv->desc.ld_tgt_count * sizeof(struct lustre_id) +
 521                  sizeof(struct mea);
 522         lmv_init_ea_size(obd->obd_self_export, easize, 0);
 523         lmv_init_unlock(lmv);
 524         RETURN(0);
 525
 526  out_disc:
 527         while (i-- > 0) {
 528                 int rc2;
 529                 --tgt;
 530                 tgt->active = 0;
 531                 if (tgt->ltd_exp) {
 532                         --lmv->desc.ld_active_tgt_count;
 533                         rc2 = obd_disconnect(tgt->ltd_exp, 0);
 534                         if (rc2) {
 535                                 CERROR("error: LMV target %s disconnect on "
 536                                        "MDC idx %d: error %d\n",
 537                                        tgt->uuid.uuid, i, rc2);
 538                         }
 539                 }
 540         }
 541         class_disconnect(lmv->exp, 0);
 542         lmv_init_unlock(lmv);
 543         RETURN(rc);
 544 }
 545
 546 static int lmv_disconnect(struct obd_export *exp, unsigned long flags)
 547 {
 548         struct obd_device *obd = class_exp2obd(exp);
 549         struct lmv_obd *lmv = &obd->u.lmv;
 550
 551 #ifdef __KERNEL__
 552         struct proc_dir_entry *lmv_proc_dir;
 553 #endif
 554         int rc, i;
 555         ENTRY;
 556
 557         if (!lmv->tgts)
 558                 goto out_local;
 559
 560         /* Only disconnect the underlying layers on the final disconnect. */
 561         lmv->refcount--;
 562         if (lmv->refcount != 0)
 563                 goto out_local;
 564
 565 #ifdef __KERNEL__
 566         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 567 #endif
 568
 569         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 570                 struct obd_device *mdc_obd;
 571
 572                 if (lmv->tgts[i].ltd_exp == NULL)
 573                         continue;
 574
 575                 mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
 576
 577                 if (mdc_obd)
 578                         mdc_obd->obd_no_recov = obd->obd_no_recov;
 579
 580 #ifdef __KERNEL__
 581                 if (lmv_proc_dir) {
 582                         struct proc_dir_entry *mdc_symlink;
 583
 584                         mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
 585                         if (mdc_symlink) {
 586                                 lprocfs_remove(mdc_symlink);
 587                         } else {
 588                                 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 589                                        obd->obd_type->typ_name, obd->obd_name,
 590                                        mdc_obd->obd_name);
 591                         }
 592                 }
 593 #endif
 594                 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
 595                         lmv->tgts[i].ltd_exp->exp_obd->obd_name,
 596                         lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
 597
 598                 obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL);
 599                 rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags);
 600                 if (rc) {
 601                         if (lmv->tgts[i].active) {
 602                                 CERROR("Target %s disconnect error %d\n",
 603                                        lmv->tgts[i].uuid.uuid, rc);
 604                         }
 605                         rc = 0;
 606                 }
 607
 608                 lmv_activate_target(lmv, &lmv->tgts[i], 0);
 609                 lmv->tgts[i].ltd_exp = NULL;
 610         }
 611
 612 #ifdef __KERNEL__
 613         if (lmv_proc_dir) {
 614                 lprocfs_remove(lmv_proc_dir);
 615         } else {
 616                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 617                        obd->obd_type->typ_name, obd->obd_name);
 618         }
 619 #endif
 620
 621 out_local:
 622         /* this is the case when no real connection is established by
 623          * lmv_check_connect(). */
 624         if (!lmv->connected)
 625                 class_export_put(exp);
 626         rc = class_disconnect(exp, 0);
 627         if (lmv->refcount == 0)
 628                 lmv->connected = 0;
 629         RETURN(rc);
 630 }
 631
 632 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 633                          int len, void *karg, void *uarg)
 634 {
 635         struct obd_device *obddev = class_exp2obd(exp);
 636         struct lmv_obd *lmv = &obddev->u.lmv;
 637         int i, rc = 0, set = 0;
 638         ENTRY;
 639
 640         if (lmv->desc.ld_tgt_count == 0)
 641                 RETURN(-ENOTTY);
 642
 643         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 644                 int err;
 645
 646                 if (lmv->tgts[i].ltd_exp == NULL)
 647                         continue;
 648
 649                 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
 650                 if (err) {
 651                         if (lmv->tgts[i].active) {
 652                                 CERROR("error: iocontrol MDC %s on MDT"
 653                                        "idx %d: err = %d\n",
 654                                        lmv->tgts[i].uuid.uuid, i, err);
 655                                 if (!rc)
 656                                         rc = err;
 657                         }
 658                 } else
 659                         set = 1;
 660         }
 661         if (!set && !rc)
 662                 rc = -EIO;
 663
 664         RETURN(rc);
 665 }
 666
 667 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
 668 {
 669         struct lmv_obd *lmv = &obd->u.lmv;
 670         struct lustre_cfg *lcfg = buf;
 671         struct lmv_desc *desc;
 672         int rc = 0;
 673         ENTRY;
 674
 675         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 676                 CERROR("LMV setup requires a descriptor\n");
 677                 RETURN(-EINVAL);
 678         }
 679
 680         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
 681         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 682                 CERROR("descriptor size wrong: %d > %d\n",
 683                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 684                 RETURN(-EINVAL);
 685         }
 686
 687         lmv->tgts_size = LMV_MAX_TGT_COUNT * sizeof(struct lmv_tgt_desc);
 688
 689         OBD_ALLOC(lmv->tgts, lmv->tgts_size);
 690         if (lmv->tgts == NULL) {
 691                 CERROR("Out of memory\n");
 692                 RETURN(-ENOMEM);
 693         }
 694
 695         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
 696         lmv->desc.ld_tgt_count = 0;
 697         lmv->desc.ld_active_tgt_count = 0;
 698         lmv->max_cookiesize = 0;
 699         lmv->max_easize = 0;
 700
 701         spin_lock_init(&lmv->lmv_lock);
 702         sema_init(&lmv->init_sem, 1);
 703
 704         rc = lmv_setup_mgr(obd);
 705         if (rc) {
 706                 CERROR("Can't setup LMV object manager, "
 707                        "error %d.\n", rc);
 708                 OBD_FREE(lmv->tgts, lmv->tgts_size);
 709                 RETURN(rc);
 710         }
 711
 712         RETURN(rc);
 713 }
 714
 715 static int lmv_cleanup(struct obd_device *obd, int flags)
 716 {
 717         struct lmv_obd *lmv = &obd->u.lmv;
 718         ENTRY;
 719
 720         lmv_cleanup_mgr(obd);
 721         OBD_FREE(lmv->tgts, lmv->tgts_size);
 722
 723         RETURN(0);
 724 }
 725 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
 726 {
 727         struct lustre_cfg *lcfg = buf;
 728         struct obd_uuid tgt_uuid;
 729         int rc;
 730         ENTRY;
 731
 732         switch(lcfg->lcfg_command) {
 733         case LCFG_LMV_ADD_MDC:
 734                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(tgt_uuid.uuid))
 735                         GOTO(out, rc = -EINVAL);
 736
 737                 obd_str2uuid(&tgt_uuid, lustre_cfg_string(lcfg, 1));
 738                 rc = lmv_add_mdc(obd, &tgt_uuid);
 739                 GOTO(out, rc);
 740         default: {
 741                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 742                 GOTO(out, rc = -EINVAL);
 743         }
 744         }
 745 out:
 746         RETURN(rc);
 747 }
 748
 749 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 750                       unsigned long max_age)
 751 {
 752         struct lmv_obd *lmv = &obd->u.lmv;
 753         struct obd_statfs *temp;
 754         int rc = 0, i;
 755         ENTRY;
 756
 757         rc = lmv_check_connect(obd);
 758         if (rc)
 759                 RETURN(rc);
 760
 761         OBD_ALLOC(temp, sizeof(*temp));
 762         if (temp == NULL)
 763                 RETURN(-ENOMEM);
 764
 765         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 766                 if (lmv->tgts[i].ltd_exp == NULL)
 767                         continue;
 768
 769                 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp, max_age);
 770                 if (rc) {
 771                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
 772                                lmv->tgts[i].ltd_exp->exp_obd->obd_name,
 773                                rc);
 774                         GOTO(out_free_temp, rc);
 775                 }
 776                 if (i == 0) {
 777                         memcpy(osfs, temp, sizeof(*temp));
 778                 } else {
 779                         osfs->os_bavail += temp->os_bavail;
 780                         osfs->os_blocks += temp->os_blocks;
 781                         osfs->os_ffree += temp->os_ffree;
 782                         osfs->os_files += temp->os_files;
 783                 }
 784         }
 785
 786         EXIT;
 787 out_free_temp:
 788         OBD_FREE(temp, sizeof(*temp));
 789         return rc;
 790 }
 791
 792 static int lmv_getstatus(struct obd_export *exp, struct lustre_id *id)
 793 {
 794         struct obd_device *obd = exp->exp_obd;
 795         struct lmv_obd *lmv = &obd->u.lmv;
 796         int rc;
 797         ENTRY;
 798
 799         rc = lmv_check_connect(obd);
 800         if (rc)
 801                 RETURN(rc);
 802
 803         rc = md_getstatus(lmv->tgts[0].ltd_exp, id);
 804         id_group(id) = 0;
 805
 806         RETURN(rc);
 807 }
 808
 809 static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
 810                        __u64 valid, const char *xattr_name,
 811                        const void *xattr_data, unsigned int xattr_datalen,
 812                        unsigned int ea_size, struct obd_capa *ocapa,
 813                        struct ptlrpc_request **request)
 814 {
 815         struct obd_device *obd = exp->exp_obd;
 816         struct lmv_obd *lmv = &obd->u.lmv;
 817         int rc, i = id_group(id);
 818         struct lmv_obj *obj;
 819         ENTRY;
 820
 821         rc = lmv_check_connect(obd);
 822         if (rc)
 823                 RETURN(rc);
 824
 825         LASSERT(i < lmv->desc.ld_tgt_count);
 826
 827
 828         rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid,
 829                         xattr_name, xattr_data, xattr_datalen,
 830                         ea_size, ocapa, request);
 831         if (rc)
 832                 RETURN(rc);
 833
 834         obj = lmv_grab_obj(obd, id);
 835
 836         CDEBUG(D_OTHER, "GETATTR for "DLID4" %s\n",
 837                OLID4(id), obj ? "(splitted)" : "");
 838
 839         /*
 840          * if object is splitted, then we loop over all the slaves and gather
 841          * size attribute. In ideal world we would have to gather also mds field
 842          * from all slaves, as object is spread over the cluster and this is
 843          * definitely interesting information and it is not good to loss it,
 844          * but...
 845          */
 846         if (obj) {
 847                 struct mds_body *body;
 848
 849                 if (*request == NULL) {
 850                         lmv_put_obj(obj);
 851                         RETURN(rc);
 852                 }
 853
 854                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
 855                                       sizeof(*body));
 856                 LASSERT(body != NULL);
 857
 858                 lmv_lock_obj(obj);
 859
 860                 for (i = 0; i < obj->objcount; i++) {
 861
 862                         if (lmv->tgts[i].ltd_exp == NULL) {
 863                                 CWARN("%s: NULL export for %d\n",
 864                                       obd->obd_name, i);
 865                                 continue;
 866                         }
 867
 868                         /* skip master obj. */
 869                         if (id_equal_fid(&obj->id, &obj->objs[i].id))
 870                                 continue;
 871
 872                         body->size += obj->objs[i].size;
 873                 }
 874
 875                 lmv_unlock_obj(obj);
 876                 lmv_put_obj(obj);
 877         }
 878
 879         RETURN(rc);
 880 }
 881
 882 static int lmv_access_check(struct obd_export *exp,
 883                             struct lustre_id *id,
 884                             struct ptlrpc_request **request)
 885 {
 886         struct obd_device *obd = exp->exp_obd;
 887         struct lmv_obd *lmv = &obd->u.lmv;
 888         int rc, i = id_group(id);
 889         ENTRY;
 890
 891         rc = lmv_check_connect(obd);
 892         if (rc)
 893                 RETURN(rc);
 894
 895         LASSERT(i < lmv->desc.ld_tgt_count);
 896         rc = md_access_check(lmv->tgts[i].ltd_exp, id, request);
 897         RETURN(rc);
 898 }
 899
 900 static int lmv_change_cbdata(struct obd_export *exp,
 901                              struct lustre_id *id,
 902                              ldlm_iterator_t it,
 903                              void *data)
 904 {
 905         struct obd_device *obd = exp->exp_obd;
 906         struct lmv_obd *lmv = &obd->u.lmv;
 907         int i, rc;
 908         ENTRY;
 909
 910         rc = lmv_check_connect(obd);
 911         if (rc)
 912                 RETURN(rc);
 913
 914         CDEBUG(D_OTHER, "CBDATA for "DLID4"\n", OLID4(id));
 915         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
 916
 917         /* with CMD every object can have two locks in different
 918          * namespaces: lookup lock in space of mds storing direntry
 919          * and update/open lock in space of mds storing inode */
 920         for (i = 0; i < lmv->desc.ld_tgt_count; i++)
 921                 md_change_cbdata(lmv->tgts[i].ltd_exp, id, it, data);
 922
 923         RETURN(0);
 924 }
 925
 926 static int lmv_change_cbdata_name(struct obd_export *exp,
 927                                   struct lustre_id *pid,
 928                                   char *name, int len,
 929                                   struct lustre_id *cid,
 930                                   ldlm_iterator_t it,
 931                                   void *data)
 932 {
 933         struct obd_device *obd = exp->exp_obd;
 934         struct lmv_obd *lmv = &obd->u.lmv;
 935         struct lustre_id rcid = *cid;
 936         struct lmv_obj *obj;
 937         int rc = 0, mds;
 938         ENTRY;
 939
 940         rc = lmv_check_connect(obd);
 941         if (rc)
 942                 RETURN(rc);
 943
 944         LASSERT(id_group(pid) < lmv->desc.ld_tgt_count);
 945         LASSERT(id_group(cid) < lmv->desc.ld_tgt_count);
 946
 947         CDEBUG(D_OTHER, "CBDATA for "DLID4":%*s -> "DLID4"\n",
 948                OLID4(pid), len, name, OLID4(cid));
 949
 950         /* this is default mds for directory name belongs to. */
 951         mds = id_group(pid);
 952         obj = lmv_grab_obj(obd, pid);
 953         if (obj) {
 954                 /* directory is splitted. look for right mds for this name. */
 955                 mds = raw_name2idx(obj->hashtype, obj->objcount, name, len);
 956                 rcid = obj->objs[mds].id;
 957                 mds = id_group(&rcid);
 958                 lmv_put_obj(obj);
 959         }
 960         rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, &rcid, it, data);
 961         RETURN(rc);
 962 }
 963
 964 static int lmv_valid_attrs(struct obd_export *exp, struct lustre_id *id)
 965 {
 966         struct obd_device *obd = exp->exp_obd;
 967         struct lmv_obd *lmv = &obd->u.lmv;
 968         int rc = 0;
 969         ENTRY;
 970
 971         rc = lmv_check_connect(obd);
 972         if (rc)
 973                 RETURN(rc);
 974
 975         CDEBUG(D_OTHER, "validate "DLID4"\n", OLID4(id));
 976         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
 977         rc = md_valid_attrs(lmv->tgts[id_group(id)].ltd_exp, id);
 978         RETURN(rc);
 979 }
 980
 981 static int lmv_close(struct obd_export *exp, struct mdc_op_data *op_data,
 982                      struct obd_client_handle *och,
 983                      struct ptlrpc_request **request)
 984 {
 985         struct obd_device *obd = exp->exp_obd;
 986         struct lmv_obd *lmv = &obd->u.lmv;
 987         int rc, i = id_group(&op_data->id1);
 988         ENTRY;
 989
 990         rc = lmv_check_connect(obd);
 991         if (rc)
 992                 RETURN(rc);
 993
 994         LASSERT(i < lmv->desc.ld_tgt_count);
 995         CDEBUG(D_OTHER, "CLOSE "DLID4"\n", OLID4(&op_data->id1));
 996         rc = md_close(lmv->tgts[i].ltd_exp, op_data, och, request);
 997         RETURN(rc);
 998 }
 999
1000 int lmv_get_mea_and_update_object(struct obd_export *exp,
1001                                   struct lustre_id *id)
1002 {
1003         struct obd_device *obd = exp->exp_obd;
1004         struct lmv_obd *lmv = &obd->u.lmv;
1005         struct ptlrpc_request *req = NULL;
1006         struct lmv_obj *obj;
1007         struct lustre_md md;
1008         int mealen, rc;
1009         __u64 valid;
1010         ENTRY;
1011
1012         md.mea = NULL;
1013         mealen = MEA_SIZE_LMV(lmv);
1014
1015         valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA;
1016
1017         /* time to update mea of parent id */
1018         rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp,
1019                         id, valid, NULL, NULL, 0, mealen, NULL, &req);
1020         if (rc) {
1021                 CERROR("md_getattr() failed, error %d\n", rc);
1022                 GOTO(cleanup, rc);
1023         }
1024
1025         rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
1026         if (rc) {
1027                 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
1028                 GOTO(cleanup, rc);
1029         }
1030
1031         if (md.mea == NULL)
1032                 GOTO(cleanup, rc = -ENODATA);
1033
1034         obj = lmv_create_obj(exp, id, md.mea);
1035         if (IS_ERR(obj))
1036                 rc = PTR_ERR(obj);
1037         else
1038                 lmv_put_obj(obj);
1039
1040         obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
1041
1042         EXIT;
1043 cleanup:
1044         if (req)
1045                 ptlrpc_req_finished(req);
1046         return rc;
1047 }
1048
1049 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
1050                const void *data, int datalen, int mode, __u32 uid,
1051                __u32 gid, __u64 rdev, struct ptlrpc_request **request)
1052 {
1053         struct obd_device *obd = exp->exp_obd;
1054         struct lmv_obd *lmv = &obd->u.lmv;
1055         struct mds_body *body;
1056         struct lmv_obj *obj;
1057         int rc, mds, loop = 0;
1058         ENTRY;
1059
1060         rc = lmv_check_connect(obd);
1061         if (rc)
1062                 RETURN(rc);
1063
1064         if (!lmv->desc.ld_active_tgt_count)
1065                 RETURN(-EIO);
1066 repeat:
1067         LASSERT(++loop <= 2);
1068         obj = lmv_grab_obj(obd, &op_data->id1);
1069         if (obj) {
1070                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1071                                    op_data->name, op_data->namelen);
1072                 op_data->id1 = obj->objs[mds].id;
1073                 lmv_put_obj(obj);
1074         }
1075
1076         CDEBUG(D_OTHER, "CREATE '%*s' on "DLID4"\n", op_data->namelen,
1077                op_data->name, OLID4(&op_data->id1));
1078
1079         rc = md_create(lmv->tgts[id_group(&op_data->id1)].ltd_exp,
1080                        op_data, data, datalen, mode, uid, gid, rdev,
1081                        request);
1082         if (rc == 0) {
1083                 if (*request == NULL)
1084                         RETURN(rc);
1085
1086                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
1087                                       sizeof(*body));
1088                 if (body == NULL)
1089                         RETURN(-ENOMEM);
1090
1091                 CDEBUG(D_OTHER, "created. "DLID4"\n", OLID4(&op_data->id1));
1092         } else if (rc == -ERESTART) {
1093                 /*
1094                  * directory got splitted. time to update local object and
1095                  * repeat the request with proper MDS.
1096                  */
1097                 rc = lmv_get_mea_and_update_object(exp, &op_data->id1);
1098                 if (rc == 0) {
1099                         ptlrpc_req_finished(*request);
1100                         goto repeat;
1101                 }
1102         }
1103         RETURN(rc);
1104 }
1105
1106 static int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
1107 {
1108         struct obd_device *obd = exp->exp_obd;
1109         struct lmv_obd *lmv = &obd->u.lmv;
1110         int rc;
1111         ENTRY;
1112
1113         rc = lmv_check_connect(obd);
1114         if (rc)
1115                 RETURN(rc);
1116
1117         /* FIXME: choose right MDC here */
1118         CWARN("this method isn't implemented yet\n");
1119         rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
1120         RETURN(rc);
1121 }
1122
1123 static int
1124 lmv_enqueue_slaves(struct obd_export *exp, int locktype,
1125                    struct lookup_intent *it, int lockmode,
1126                    struct mdc_op_data *data, struct lustre_handle *lockh,
1127                    void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1128                    ldlm_blocking_callback cb_blocking, void *cb_data)
1129 {
1130         struct obd_device *obd = exp->exp_obd;
1131         struct lmv_obd *lmv = &obd->u.lmv;
1132         struct mea *mea = data->mea1;
1133         struct mdc_op_data *data2;
1134         int i, rc, mds;
1135         ENTRY;
1136
1137         OBD_ALLOC(data2, sizeof(*data2));
1138         if (data2 == NULL)
1139                 RETURN(-ENOMEM);
1140
1141         LASSERT(mea != NULL);
1142         for (i = 0; i < mea->mea_count; i++) {
1143                 memset(data2, 0, sizeof(*data2));
1144                 data2->id1 = mea->mea_ids[i];
1145                 mds = id_group(&data2->id1);
1146
1147                 if (lmv->tgts[mds].ltd_exp == NULL)
1148                         continue;
1149
1150                 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it,
1151                                 lockmode, data2, lockh + i, lmm, lmmsize,
1152                                 cb_compl, cb_blocking, cb_data);
1153
1154                 CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n",
1155                        OLID4(&mea->mea_ids[i]), rc, LUSTRE_IT(it)->it_status);
1156                 if (rc)
1157                         GOTO(cleanup, rc);
1158                 if (LUSTRE_IT(it)->it_data) {
1159                         struct ptlrpc_request *req;
1160                         req = (struct ptlrpc_request *) LUSTRE_IT(it)->it_data;
1161                         ptlrpc_req_finished(req);
1162                 }
1163
1164                 if (LUSTRE_IT(it)->it_status)
1165                         GOTO(cleanup, rc = LUSTRE_IT(it)->it_status);
1166         }
1167
1168         OBD_FREE(data2, sizeof(*data2));
1169         RETURN(0);
1170 cleanup:
1171         OBD_FREE(data2, sizeof(*data2));
1172
1173         /* drop all taken locks */
1174         while (--i >= 0) {
1175                 if (lockh[i].cookie)
1176                         ldlm_lock_decref(lockh + i, lockmode);
1177                 lockh[i].cookie = 0;
1178         }
1179         return rc;
1180 }
1181
1182 static int
1183 lmv_enqueue_remote(struct obd_export *exp, int lock_type,
1184                    struct lookup_intent *it, int lock_mode,
1185                    struct mdc_op_data *data, struct lustre_handle *lockh,
1186                    void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1187                    ldlm_blocking_callback cb_blocking, void *cb_data)
1188 {
1189         struct ptlrpc_request *req = LUSTRE_IT(it)->it_data;
1190         struct obd_device *obd = exp->exp_obd;
1191         struct lmv_obd *lmv = &obd->u.lmv;
1192         struct lustre_handle plock;
1193         struct mdc_op_data rdata;
1194         struct mds_body *body = NULL;
1195         int rc = 0, pmode;
1196         ENTRY;
1197
1198         body = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body));
1199         LASSERT(body != NULL);
1200
1201         if (!(body->valid & OBD_MD_MDS))
1202                 RETURN(0);
1203
1204         CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4" -> "DLID4"\n",
1205                LL_IT2STR(it), OLID4(&data->id1), OLID4(&body->id1));
1206
1207         /* we got LOOKUP lock, but we really need attrs */
1208         pmode = LUSTRE_IT(it)->it_lock_mode;
1209         LASSERT(pmode != 0);
1210         memcpy(&plock, lockh, sizeof(plock));
1211         LUSTRE_IT(it)->it_lock_mode = 0;
1212         LUSTRE_IT(it)->it_data = NULL;
1213         LASSERT((body->valid & OBD_MD_FID) != 0);
1214
1215         memcpy(&rdata, data, sizeof(rdata));
1216         rdata.id1 = body->id1;
1217         rdata.name = NULL;
1218         rdata.namelen = 0;
1219
1220         LUSTRE_IT(it)->it_disposition &= ~DISP_ENQ_COMPLETE;
1221         ptlrpc_req_finished(req);
1222
1223         rc = md_enqueue(lmv->tgts[id_group(&rdata.id1)].ltd_exp,
1224                         lock_type, it, lock_mode, &rdata, lockh, lmm,
1225                         lmmsize, cb_compl, cb_blocking, cb_data);
1226         ldlm_lock_decref(&plock, pmode);
1227         RETURN(rc);
1228 }
1229
1230 static int
1231 lmv_enqueue(struct obd_export *exp, int lock_type,
1232             struct lookup_intent *it, int lock_mode,
1233             struct mdc_op_data *data, struct lustre_handle *lockh,
1234             void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1235             ldlm_blocking_callback cb_blocking, void *cb_data)
1236 {
1237         struct obd_device *obd = exp->exp_obd;
1238         struct lmv_obd *lmv = &obd->u.lmv;
1239         struct lmv_obj *obj;
1240         int rc, mds;
1241         ENTRY;
1242
1243         rc = lmv_check_connect(obd);
1244         if (rc)
1245                 RETURN(rc);
1246
1247         if (data->mea1 && it->it_op == IT_UNLINK) {
1248                 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
1249                                         data, lockh, lmm, lmmsize,
1250                                         cb_compl, cb_blocking, cb_data);
1251                 RETURN(rc);
1252         }
1253
1254         if (data->namelen) {
1255                 obj = lmv_grab_obj(obd, &data->id1);
1256                 if (obj) {
1257                         /* directory is splitted. look for right mds for this
1258                          * name */
1259                         mds = raw_name2idx(obj->hashtype, obj->objcount,
1260                                            (char *)data->name, data->namelen);
1261                         data->id1 = obj->objs[mds].id;
1262                         lmv_put_obj(obj);
1263                 }
1264         }
1265         CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4"\n", LL_IT2STR(it),
1266                OLID4(&data->id1));
1267
1268         rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp,
1269                         lock_type, it, lock_mode, data, lockh, lmm,
1270                         lmmsize, cb_compl, cb_blocking, cb_data);
1271         if (rc == 0 && it->it_op == IT_OPEN)
1272                 rc = lmv_enqueue_remote(exp, lock_type, it, lock_mode,
1273                                         data, lockh, lmm, lmmsize,
1274                                         cb_compl, cb_blocking, cb_data);
1275         RETURN(rc);
1276 }
1277
1278 static int
1279 lmv_getattr_lock(struct obd_export *exp, struct lustre_id *id,
1280                  char *filename, int namelen, __u64 valid,
1281                  unsigned int ea_size, struct ptlrpc_request **request)
1282 {
1283         int rc, mds = id_group(id), loop = 0;
1284         struct obd_device *obd = exp->exp_obd;
1285         struct lmv_obd *lmv = &obd->u.lmv;
1286         struct lustre_id rid = *id;
1287         struct mds_body *body;
1288         struct lmv_obj *obj;
1289         ENTRY;
1290
1291         rc = lmv_check_connect(obd);
1292         if (rc)
1293                 RETURN(rc);
1294 repeat:
1295         LASSERT(++loop <= 2);
1296         obj = lmv_grab_obj(obd, id);
1297         if (obj) {
1298                 /* directory is splitted. look for right mds for this name */
1299                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1300                                    filename, namelen - 1);
1301                 rid = obj->objs[mds].id;
1302                 lmv_put_obj(obj);
1303         }
1304
1305         CDEBUG(D_OTHER, "getattr_lock for %*s on "DLID4" -> "DLID4"\n",
1306                namelen, filename, OLID4(id), OLID4(&rid));
1307
1308         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1309                              &rid, filename, namelen,
1310                              valid == OBD_MD_FLID ? valid : valid | OBD_MD_FID,
1311                              ea_size, request);
1312         if (rc == 0) {
1313                 /*
1314                  * this could be cross-node reference. in this case all we have
1315                  * right now is lustre_id triple. we'd like to find other
1316                  * attributes.
1317                  */
1318                 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
1319                 LASSERT(body != NULL);
1320                 LASSERT((body->valid & OBD_MD_FID) != 0
1321                                 || body->valid == OBD_MD_FLID);
1322
1323                 if (body->valid & OBD_MD_MDS) {
1324                         struct ptlrpc_request *req = NULL;
1325
1326                         rid = body->id1;
1327                         CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid));
1328
1329                         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1330                                              &rid, NULL, 1, valid, ea_size, &req);
1331                         ptlrpc_req_finished(*request);
1332                         *request = req;
1333                 }
1334         } else if (rc == -ERESTART) {
1335                 /* directory got splitted. time to update local object and
1336                  * repeat the request with proper MDS */
1337                 rc = lmv_get_mea_and_update_object(exp, &rid);
1338                 if (rc == 0) {
1339                         ptlrpc_req_finished(*request);
1340                         goto repeat;
1341                 }
1342         }
1343         RETURN(rc);
1344 }
1345
1346 /*
1347  * llite passes id of an target inode in data->id1 and id of directory in
1348  * data->id2
1349  */
1350 static int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
1351                     struct ptlrpc_request **request)
1352 {
1353         struct obd_device *obd = exp->exp_obd;
1354         struct lmv_obd *lmv = &obd->u.lmv;
1355         struct lmv_obj *obj;
1356         int rc, mds;
1357         ENTRY;
1358
1359         rc = lmv_check_connect(obd);
1360         if (rc)
1361                 RETURN(rc);
1362
1363         if (data->namelen != 0) {
1364                 /* usual link request */
1365                 obj = lmv_grab_obj(obd, &data->id2);
1366                 if (obj) {
1367                         rc = raw_name2idx(obj->hashtype, obj->objcount,
1368                                           data->name, data->namelen);
1369                         data->id2 = obj->objs[rc].id;
1370                         lmv_put_obj(obj);
1371                 }
1372
1373                 mds = id_group(&data->id2);
1374
1375                 CDEBUG(D_OTHER,"link "DLID4":%*s to "DLID4"\n",
1376                        OLID4(&data->id2), data->namelen, data->name,
1377                        OLID4(&data->id1));
1378         } else {
1379                 mds = id_group(&data->id1);
1380
1381                 /* request from MDS to acquire i_links for inode by id1 */
1382                 CDEBUG(D_OTHER, "inc i_nlinks for "DLID4"\n",
1383                        OLID4(&data->id1));
1384         }
1385
1386         CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n",
1387                mds, OLID4(&data->id1));
1388         rc = md_link(lmv->tgts[mds].ltd_exp, data, request);
1389
1390         RETURN(rc);
1391 }
1392
1393 static int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1394                       const char *old, int oldlen, const char *new, int newlen,
1395                       struct ptlrpc_request **request)
1396 {
1397         struct obd_device *obd = exp->exp_obd;
1398         struct lmv_obd *lmv = &obd->u.lmv;
1399         struct lmv_obj *obj;
1400         int rc, mds;
1401         ENTRY;
1402
1403         CDEBUG(D_OTHER, "rename %*s in "DLID4" to %*s in "DLID4"\n",
1404                oldlen, old, OLID4(&data->id1), newlen, new,
1405                OLID4(&data->id2));
1406
1407         rc = lmv_check_connect(obd);
1408         if (rc)
1409                 RETURN(rc);
1410
1411         if (oldlen == 0) {
1412                 /*
1413                  * MDS with old dir entry is asking another MDS to create name
1414                  * there.
1415                  */
1416                 CDEBUG(D_OTHER,
1417                        "create %*s(%d/%d) in "DLID4" pointing "
1418                        "to "DLID4"\n", newlen, new, oldlen, newlen,
1419                        OLID4(&data->id2), OLID4(&data->id1));
1420
1421                 mds = id_group(&data->id2);
1422
1423                 /*
1424                  * target directory can be splitted, sowe should forward request
1425                  * to the right MDS.
1426                  */
1427                 obj = lmv_grab_obj(obd, &data->id2);
1428                 if (obj) {
1429                         mds = raw_name2idx(obj->hashtype, obj->objcount,
1430                                            (char *)new, newlen);
1431                         data->id2 = obj->objs[mds].id;
1432                         CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1433                                OLID4(&data->id2));
1434                         lmv_put_obj(obj);
1435                 }
1436                 goto request;
1437         }
1438
1439         obj = lmv_grab_obj(obd, &data->id1);
1440         if (obj) {
1441                 /*
1442                  * directory is already splitted, so we have to forward request
1443                  * to the right MDS.
1444                  */
1445                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1446                                    (char *)old, oldlen);
1447                 data->id1 = obj->objs[mds].id;
1448                 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1449                        OLID4(&data->id1));
1450                 lmv_put_obj(obj);
1451         }
1452
1453         obj = lmv_grab_obj(obd, &data->id2);
1454         if (obj) {
1455                 /*
1456                  * directory is already splitted, so we have to forward request
1457                  * to the right MDS.
1458                  */
1459                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1460                                    (char *)new, newlen);
1461
1462                 data->id2 = obj->objs[mds].id;
1463                 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1464                        OLID4(&data->id2));
1465                 lmv_put_obj(obj);
1466         }
1467
1468         mds = id_group(&data->id1);
1469
1470 request:
1471         if (id_group(&data->id1) != id_group(&data->id2)) {
1472                 CDEBUG(D_OTHER,"cross-node rename "DLID4"/%*s to "DLID4"/%*s\n",
1473                        OLID4(&data->id1), oldlen, old, OLID4(&data->id2),
1474                        newlen, new);
1475         }
1476
1477         rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1478                        new, newlen, request);
1479         RETURN(rc);
1480 }
1481
1482 static int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1483                        struct iattr *iattr, void *ea, int ealen, void *ea2,
1484                        int ea2len, void *ea3, int ea3len,
1485                        struct ptlrpc_request **request)
1486 {
1487         struct obd_device *obd = exp->exp_obd;
1488         struct lmv_obd *lmv = &obd->u.lmv;
1489         struct ptlrpc_request *req;
1490         struct mds_body *body;
1491         struct lmv_obj *obj;
1492         int rc = 0, i;
1493         ENTRY;
1494
1495         rc = lmv_check_connect(obd);
1496         if (rc)
1497                 RETURN(rc);
1498
1499         obj = lmv_grab_obj(obd, &data->id1);
1500
1501         CDEBUG(D_OTHER, "SETATTR for "DLID4", valid 0x%x%s\n",
1502                OLID4(&data->id1), iattr->ia_valid, obj ? ", splitted" : "");
1503
1504         if (obj) {
1505                 for (i = 0; i < obj->objcount; i++) {
1506                         data->id1 = obj->objs[i].id;
1507
1508                         rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1509                                         data, iattr, ea, ealen, ea2, ea2len,
1510                                         ea3, ea3len, &req);
1511
1512                         if (id_equal_fid(&obj->id, &obj->objs[i].id)) {
1513                                 /*
1514                                  * this is master object and this request should
1515                                  * be returned back to llite.
1516                                  */
1517                                 *request = req;
1518                         } else {
1519                                 ptlrpc_req_finished(req);
1520                         }
1521
1522                         if (rc)
1523                                 break;
1524                 }
1525                 lmv_put_obj(obj);
1526         } else {
1527                 LASSERT(id_group(&data->id1) < lmv->desc.ld_tgt_count);
1528                 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1529                                 data, iattr, ea, ealen, ea2, ea2len, ea3,
1530                                 ea3len, request);
1531                 if (rc == 0) {
1532                         body = lustre_msg_buf((*request)->rq_repmsg, 0,
1533                                               sizeof(*body));
1534                         LASSERT(body != NULL);
1535                         LASSERT((body->valid & OBD_MD_FID) != 0);
1536                         LASSERT(id_group(&body->id1) == id_group(&data->id1));
1537                 }
1538         }
1539         RETURN(rc);
1540 }
1541
1542 static int lmv_sync(struct obd_export *exp, struct lustre_id *id,
1543                     struct ptlrpc_request **request)
1544 {
1545         struct obd_device *obd = exp->exp_obd;
1546         struct lmv_obd *lmv = &obd->u.lmv;
1547         int rc;
1548         ENTRY;
1549
1550         rc = lmv_check_connect(obd);
1551         if (rc)
1552                 RETURN(rc);
1553
1554         rc = md_sync(lmv->tgts[id_group(id)].ltd_exp,
1555                      id, request);
1556         RETURN(rc);
1557 }
1558
1559 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock,
1560                             struct ldlm_lock_desc *desc,
1561                             void *data, int flag)
1562 {
1563         struct lustre_handle lockh;
1564         struct lmv_obj *obj;
1565         int rc;
1566         ENTRY;
1567
1568         switch (flag) {
1569         case LDLM_CB_BLOCKING:
1570                 ldlm_lock2handle(lock, &lockh);
1571                 rc = ldlm_cli_cancel(&lockh);
1572                 if (rc < 0) {
1573                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1574                         RETURN(rc);
1575                 }
1576                 break;
1577         case LDLM_CB_CANCELING:
1578                 /* time to drop cached attrs for dirobj */
1579                 obj = lock->l_ast_data;
1580                 if (obj) {
1581                         CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
1582                                ", master "DLID4"\n",
1583                                lock->l_resource->lr_name.name[3] == 1 ?
1584                                "LOOKUP" : "UPDATE",
1585                                lock->l_resource->lr_name.name[0],
1586                                lock->l_resource->lr_name.name[1],
1587                                OLID4(&obj->id));
1588                         lmv_put_obj(obj);
1589                 }
1590                 break;
1591         default:
1592                 LBUG();
1593         }
1594         RETURN(0);
1595 }
1596
1597 static void lmv_remove_dots(struct page *page)
1598 {
1599         unsigned limit = PAGE_CACHE_SIZE;
1600         char *kaddr = page_address(page);
1601         struct ext2_dir_entry_2 *p;
1602         unsigned offs, rec_len;
1603
1604         for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1605                 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1606                 rec_len = le16_to_cpu(p->rec_len);
1607
1608                 if ((p->name_len == 1 && p->name[0] == '.') ||
1609                     (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1610                         p->inode = 0;
1611         }
1612 }
1613
1614 static int lmv_readpage(struct obd_export *exp, struct lustre_id *id,
1615                         __u64 offset, struct page *page,
1616                         struct ptlrpc_request **request)
1617 {
1618         struct obd_device *obd = exp->exp_obd;
1619         struct lmv_obd *lmv = &obd->u.lmv;
1620         struct lustre_id rid = *id;
1621         struct lmv_obj *obj;
1622         int rc, i;
1623         ENTRY;
1624
1625 #warning "we need well-desgined readdir() implementation"
1626         rc = lmv_check_connect(obd);
1627         if (rc)
1628                 RETURN(rc);
1629
1630         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
1631         CDEBUG(D_OTHER, "READPAGE at %llu from "DLID4"\n",
1632                offset, OLID4(&rid));
1633
1634         obj = lmv_grab_obj(obd, id);
1635         if (obj) {
1636                 lmv_lock_obj(obj);
1637
1638                 /* find dirobj containing page with requested offset. */
1639                 for (i = 0; i < obj->objcount; i++) {
1640                         if (offset < obj->objs[i].size)
1641                                 break;
1642                         offset -= obj->objs[i].size;
1643                 }
1644                 rid = obj->objs[i].id;
1645
1646                 lmv_unlock_obj(obj);
1647                 lmv_put_obj(obj);
1648
1649                 CDEBUG(D_OTHER, "forward to "DLID4" with offset %lu\n",
1650                        OLID4(&rid), (unsigned long)offset);
1651         }
1652         rc = md_readpage(lmv->tgts[id_group(&rid)].ltd_exp, &rid,
1653                          offset, page, request);
1654
1655         if (rc == 0 && !id_equal_fid(&rid, id))
1656                 /* this page isn't from master object. To avoid "." and ".."
1657                  * duplication in directory, we have to remove them from all
1658                  * slave objects */
1659                 lmv_remove_dots(page);
1660
1661         RETURN(rc);
1662 }
1663
1664 static int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1665                              struct ptlrpc_request **req)
1666 {
1667         struct obd_device *obd = exp->exp_obd;
1668         struct lmv_obd *lmv = &obd->u.lmv;
1669         struct mea *mea = data->mea1;
1670         struct mdc_op_data *data2;
1671         int i, rc = 0;
1672         ENTRY;
1673
1674         OBD_ALLOC(data2, sizeof(*data2));
1675         if (data2 == NULL)
1676                 RETURN(-ENOMEM);
1677
1678         LASSERT(mea != NULL);
1679         for (i = 0; i < mea->mea_count; i++) {
1680                 memset(data2, 0, sizeof(*data2));
1681                 data2->id1 = mea->mea_ids[i];
1682                 data2->create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1683
1684                 if (lmv->tgts[id_group(&data2->id1)].ltd_exp == NULL)
1685                         continue;
1686
1687                 rc = md_unlink(lmv->tgts[id_group(&data2->id1)].ltd_exp,
1688                                data2, req);
1689
1690                 CDEBUG(D_OTHER, "unlink slave "DLID4" -> %d\n",
1691                        OLID4(&mea->mea_ids[i]), rc);
1692
1693                 if (*req) {
1694                         ptlrpc_req_finished(*req);
1695                         *req = NULL;
1696                 }
1697                 if (rc)
1698                         RETURN(rc);
1699         }
1700         OBD_FREE(data2, sizeof(*data2));
1701         RETURN(rc);
1702 }
1703
1704 static int lmv_delete_inode(struct obd_export *exp, struct lustre_id *id)
1705 {
1706         ENTRY;
1707
1708         LASSERT(exp && id);
1709         if (lmv_delete_obj(exp, id)) {
1710                 CDEBUG(D_OTHER, "lmv object "DLID4" is destroyed.\n",
1711                        OLID4(id));
1712         }
1713         RETURN(0);
1714 }
1715
1716 static int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1717                       struct ptlrpc_request **request)
1718 {
1719         struct obd_device *obd = exp->exp_obd;
1720         struct lmv_obd *lmv = &obd->u.lmv;
1721         int rc, i = 0;
1722         ENTRY;
1723
1724         rc = lmv_check_connect(obd);
1725         if (rc)
1726                 RETURN(rc);
1727
1728         if (data->namelen == 0 && data->mea1 != NULL) {
1729                 /* mds asks to remove slave objects */
1730                 rc = lmv_unlink_slaves(exp, data, request);
1731                 RETURN(rc);
1732         }
1733
1734         if (data->namelen != 0) {
1735                 struct lmv_obj *obj;
1736
1737                 obj = lmv_grab_obj(obd, &data->id1);
1738                 if (obj) {
1739                         i = raw_name2idx(obj->hashtype, obj->objcount,
1740                                          data->name, data->namelen);
1741                         data->id1 = obj->objs[i].id;
1742                         lmv_put_obj(obj);
1743                 }
1744                 CDEBUG(D_OTHER, "unlink '%*s' in "DLID4" -> %u\n",
1745                        data->namelen, data->name, OLID4(&data->id1),
1746                        i);
1747         } else {
1748                 CDEBUG(D_OTHER, "drop i_nlink on "DLID4"\n",
1749                        OLID4(&data->id1));
1750         }
1751         rc = md_unlink(lmv->tgts[id_group(&data->id1)].ltd_exp,
1752                        data, request);
1753         RETURN(rc);
1754 }
1755
1756 static struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1757                                            struct lustre_id *id)
1758 {
1759         struct obd_device *obd = exp->exp_obd;
1760         struct lmv_obd *lmv = &obd->u.lmv;
1761         int rc;
1762         ENTRY;
1763
1764         rc = lmv_check_connect(obd);
1765         if (rc)
1766                 RETURN(ERR_PTR(rc));
1767         obd = lmv->tgts[id_group(id)].ltd_exp->exp_obd;
1768         EXIT;
1769
1770         return obd;
1771 }
1772
1773 static int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1774                                  void *acl, int acl_size,
1775                                  struct lov_stripe_md **ea,
1776                                  struct obd_trans_info *oti)
1777 {
1778         struct obd_device *obd = exp->exp_obd;
1779         struct lmv_obd *lmv = &obd->u.lmv;
1780         struct lov_stripe_md obj_md;
1781         struct lov_stripe_md *obj_mdp = &obj_md;
1782         int rc = 0;
1783         ENTRY;
1784
1785         LASSERT(ea == NULL);
1786         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1787
1788         rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa,
1789                         acl, acl_size, &obj_mdp, oti);
1790
1791         RETURN(rc);
1792 }
1793
1794 /*
1795  * to be called from MDS only. @oa should have correct store cookie and o_fid
1796  * values for "master" object, as it will be used.
1797  */
1798 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
1799                    void *acl, int acl_size,
1800                    struct lov_stripe_md **ea, struct obd_trans_info *oti)
1801 {
1802         struct obd_device *obd = exp->exp_obd;
1803         struct lmv_obd *lmv = &obd->u.lmv;
1804         struct lustre_id mid;
1805         int i, c, rc = 0;
1806         struct mea *mea;
1807         ENTRY;
1808
1809         rc = lmv_check_connect(obd);
1810         if (rc)
1811                 RETURN(rc);
1812
1813         LASSERT(oa != NULL);
1814
1815         if (ea == NULL) {
1816                 rc = lmv_obd_create_single(exp, oa, acl, acl_size, NULL, oti);
1817                 if (rc)
1818                         CERROR("Can't create object, rc = %d\n", rc);
1819                 RETURN(rc);
1820         }
1821
1822         /* acl is only suppied when mds create single remote obj */
1823         LASSERT(acl == NULL && acl_size == 0);
1824
1825         if (*ea == NULL) {
1826                 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1827                 if (rc < 0) {
1828                         CERROR("obd_alloc_diskmd() failed, error %d\n",
1829                                rc);
1830                         RETURN(rc);
1831                 } else
1832                         rc = 0;
1833
1834                 if (*ea == NULL)
1835                         RETURN(-ENOMEM);
1836         }
1837
1838         /*
1839          * here we should take care about splitted dir, so store cookie and fid
1840          * for "master" object should already be allocated and passed in @oa.
1841          */
1842         LASSERT(oa->o_id != 0);
1843         LASSERT(oa->o_fid != 0);
1844
1845         /* save "master" object id */
1846         obdo2id(&mid, oa);
1847
1848         mea = (struct mea *)*ea;
1849         mea->mea_master = -1;
1850         mea->mea_magic = MEA_MAGIC_ALL_CHARS;
1851
1852         if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1853                 mea->mea_count = lmv->desc.ld_tgt_count;
1854
1855         for (i = 0, c = 0; c < mea->mea_count && i < lmv->desc.ld_tgt_count; i++) {
1856                 struct lov_stripe_md obj_md;
1857                 struct lov_stripe_md *obj_mdp = &obj_md;
1858
1859                 if (lmv->tgts[i].ltd_exp == NULL) {
1860                         /* this is "master" MDS */
1861                         mea->mea_master = i;
1862                         mea->mea_ids[c] = mid;
1863                         c++;
1864                         continue;
1865                 }
1866
1867                 /*
1868                  * "master" MDS should always be part of stripped dir,
1869                  * so scan for it.
1870                  */
1871                 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1872                         continue;
1873
1874                 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE |
1875                         OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1876
1877                 rc = obd_create(lmv->tgts[c].ltd_exp, oa, NULL, 0,
1878                                 &obj_mdp, oti);
1879                 if (rc) {
1880                         CERROR("obd_create() failed on MDT target %d, "
1881                                "error %d\n", c, rc);
1882                         RETURN(rc);
1883                 }
1884
1885                 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1886                        i, oa->o_id, oa->o_generation);
1887
1888
1889                 /*
1890                  * here, when object is created (or it is master and was passed
1891                  * from caller) on desired MDS we save its fid to local mea_ids.
1892                  */
1893                 LASSERT(oa->o_fid);
1894
1895                 /*
1896                  * store cookie should be defined here for both cases (master
1897                  * object and not master), because master is already created.
1898                  */
1899                 LASSERT(oa->o_id);
1900
1901                 /* fill mea by store cookie and fid */
1902                 obdo2id(&mea->mea_ids[c], oa);
1903                 c++;
1904         }
1905         LASSERT(c == mea->mea_count);
1906
1907         CDEBUG(D_OTHER, "%d dirobjects created\n",
1908                (int)mea->mea_count);
1909
1910         RETURN(rc);
1911 }
1912
1913 static int lmv_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1914                          struct obd_device *tgt, int count,
1915                          struct llog_catid *logid)
1916 {
1917         struct llog_ctxt *ctxt;
1918         int rc;
1919         ENTRY;
1920
1921         rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1922                             &llog_client_ops);
1923         if (rc == 0) {
1924                 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1925                 ctxt->loc_imp = tgt->u.cli.cl_import;
1926         }
1927
1928         RETURN(rc);
1929 }
1930
1931 static int lmv_llog_finish(struct obd_device *obd,
1932                            struct obd_llogs *llogs, int count)
1933 {
1934         int rc;
1935         ENTRY;
1936
1937         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1938         RETURN(rc);
1939 }
1940
1941 static int lmv_precleanup(struct obd_device *obd, int flags)
1942 {
1943         int rc = 0;
1944
1945         rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1946         if (rc != 0)
1947                 CERROR("failed to cleanup llogging subsystems\n");
1948
1949         RETURN(rc);
1950 }
1951
1952 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1953                         void *key, __u32 *vallen, void *val)
1954 {
1955         struct obd_device *obd;
1956         struct lmv_obd *lmv;
1957         int rc = 0;
1958         ENTRY;
1959
1960         obd = class_exp2obd(exp);
1961         if (obd == NULL) {
1962                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1963                        exp->exp_handle.h_cookie);
1964                 RETURN(-EINVAL);
1965         }
1966
1967         lmv = &obd->u.lmv;
1968         if (keylen == strlen("mdsize") && !strcmp(key, "mdsize")) {
1969                 __u32 *mdsize = val;
1970                 *vallen = sizeof(__u32);
1971                 *mdsize = sizeof(struct lustre_id) * lmv->desc.ld_tgt_count
1972                        + sizeof(struct mea);
1973                 RETURN(0);
1974         } else if (keylen == strlen("mdsnum") && !strcmp(key, "mdsnum")) {
1975                 struct obd_uuid *cluuid = &lmv->cluuid;
1976                 struct lmv_tgt_desc *tgts;
1977                 __u32 *mdsnum = val;
1978                 int i;
1979
1980                 tgts = lmv->tgts;
1981                 for (i = 0; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1982                         if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1983                                 *vallen = sizeof(__u32);
1984                                 *mdsnum = i;
1985                                 RETURN(0);
1986                         }
1987                 }
1988                 LASSERT(0);
1989         } else if (keylen == strlen("rootid") && !strcmp(key, "rootid")) {
1990                 rc = lmv_check_connect(obd);
1991                 if (rc)
1992                         RETURN(rc);
1993
1994                 /* getting rootid from first MDS. */
1995                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
1996                                   vallen, val);
1997                 RETURN(rc);
1998         } else if (keylen >= strlen("lmvdesc") && !strcmp(key, "lmvdesc")) {
1999                 struct lmv_desc *desc_ret = val;
2000                 *desc_ret = lmv->desc;
2001                 RETURN(0);
2002         } else if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2003                 struct lmv_tgt_desc *tgts;
2004                 int i;
2005
2006                 rc = lmv_check_connect(obd);
2007                 if (rc)
2008                         RETURN(rc);
2009
2010                 LASSERT(*vallen == sizeof(__u32));
2011                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
2012                      i++, tgts++) {
2013
2014                         /* all tgts should be connected when this get called. */
2015                         if (!tgts || !tgts->ltd_exp) {
2016                                 CERROR("target not setup?\n");
2017                                 continue;
2018                         }
2019
2020                         if (!obd_get_info(tgts->ltd_exp, keylen, key,
2021                                           vallen, val))
2022                                 RETURN(0);
2023                 }
2024                 RETURN(-EINVAL);
2025         } else if (keylen >= strlen("lovdesc") && !strcmp(key, "lovdesc")) {
2026                 rc = lmv_check_connect(obd);
2027                 if (rc)
2028                         RETURN(rc);
2029
2030                 /* forwarding this request to first MDS, it should know LOV
2031                  * desc. */
2032                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
2033                                   vallen, val);
2034                 RETURN(rc);
2035         } else if (keylen >= strlen("getext") && !strcmp(key, "getext")) {
2036                 struct lmv_tgt_desc *tgts;
2037                 int i;
2038
2039                 rc = lmv_check_connect(obd);
2040                 if (rc)
2041                         RETURN(rc);
2042
2043                 LASSERT(*vallen == sizeof(struct fid_extent));
2044                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
2045                      i++, tgts++) {
2046
2047                         /* all tgts should be connected when this get called. */
2048                         if (!tgts || !tgts->ltd_exp) {
2049                                 CERROR("target not setup?\n");
2050                                 continue;
2051                         }
2052
2053                         rc = obd_get_info(tgts->ltd_exp, keylen, key,
2054                                           vallen, val);
2055                         if (rc)
2056                                 RETURN(rc);
2057                 }
2058                 RETURN(0);
2059         }
2060
2061         CDEBUG(D_IOCTL, "invalid key\n");
2062         RETURN(-EINVAL);
2063 }
2064
2065 int lmv_set_info(struct obd_export *exp, obd_count keylen,
2066                  void *key, obd_count vallen, void *val)
2067 {
2068         struct lmv_tgt_desc    *tgt;
2069         struct obd_device      *obd;
2070         struct lmv_obd         *lmv;
2071         int rc = 0;
2072         ENTRY;
2073
2074         obd = class_exp2obd(exp);
2075         if (obd == NULL) {
2076                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2077                        exp->exp_handle.h_cookie);
2078                 RETURN(-EINVAL);
2079         }
2080         lmv = &obd->u.lmv;
2081
2082         if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
2083                 lmv->server_timeout = 1;
2084                 lmv_set_timeouts(obd);
2085                 RETURN(0);
2086         }
2087
2088         /* maybe this could be default */
2089         if ((keylen == strlen("sec") && strcmp(key, "sec") == 0) ||
2090             (keylen == strlen("sec_flags") && strcmp(key, "sec_flags") == 0) ||
2091             (keylen == strlen("nllu") && strcmp(key, "nllu") == 0)) {
2092                 struct obd_export *exp;
2093                 int err, i;
2094
2095                 spin_lock(&lmv->lmv_lock);
2096                 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count;
2097                      i++, tgt++) {
2098                         exp = tgt->ltd_exp;
2099                         /* during setup time the connections to mdc might
2100                          * haven't been established.
2101                          */
2102                         if (exp == NULL) {
2103                                 struct obd_device *tgt_obd;
2104
2105                                 tgt_obd = class_find_client_obd(&tgt->uuid,
2106                                                                 OBD_MDC_DEVICENAME,
2107                                                                 &obd->obd_uuid);
2108                                 if (!tgt_obd) {
2109                                         CERROR("can't set info %s, "
2110                                                "device %s not attached?\n",
2111                                                 (char *) key, tgt->uuid.uuid);
2112                                         rc = -EINVAL;
2113                                         continue;
2114                                 }
2115                                 exp = tgt_obd->obd_self_export;
2116                         }
2117
2118                         err = obd_set_info(exp, keylen, key, vallen, val);
2119                         if (!rc)
2120                                 rc = err;
2121                 }
2122                 spin_unlock(&lmv->lmv_lock);
2123
2124                 RETURN(rc);
2125         }
2126         if (keylen == 5 && strcmp(key, "audit") == 0) {
2127                 struct audit_attr_msg * msg = val;
2128                 int mds = id_group(&msg->id);
2129                 int i;
2130                 LASSERT(mds < lmv->desc.ld_tgt_count);
2131
2132                 if (IS_AUDIT_OP(msg->attr, AUDIT_FS)) {
2133                         //FS audit, send message to all mds
2134                         for (i = 0; i < lmv->desc.ld_tgt_count;i++) {
2135                                 obd_set_info(lmv->tgts[i].ltd_exp,
2136                                                   keylen, key, vallen, val);
2137                         }
2138                 }
2139                 else if (IS_AUDIT_OP(msg->attr, AUDIT_DIR)) {
2140                         //audit for dir.
2141                         //if dir is splitted, send RPC to all mds involved
2142                         struct lmv_obj *obj;
2143                         struct lustre_id rid;
2144                         int i;
2145
2146                         obj = lmv_grab_obj(obd, &msg->id);
2147                         if (obj) {
2148                                 lmv_lock_obj(obj);
2149                                 for (i = 0; i < obj->objcount; i++) {
2150                                         rid = obj->objs[i].id;
2151                                         mds = id_group(&rid);
2152                                         obd_set_info(lmv->tgts[mds].ltd_exp,
2153                                                           keylen, key,
2154                                                           vallen, val);
2155                                 }
2156                                 lmv_unlock_obj(obj);
2157                                 lmv_put_obj(obj);
2158                         }
2159                         else {
2160                                 rc = obd_set_info(lmv->tgts[mds].ltd_exp,
2161                                                  keylen, key, vallen, val);
2162                         }
2163                 }
2164                 else {
2165                         //set audit for file
2166                         rc = obd_set_info(lmv->tgts[mds].ltd_exp,
2167                                           keylen, key, vallen, val);
2168                 }
2169                 RETURN(rc);
2170         }
2171         if (((keylen == strlen("flush_cred") &&
2172              strcmp(key, "flush_cred") == 0)) ||
2173              ((keylen == strlen("crypto_type") &&
2174              strcmp(key, "crypto_type") == 0))) {
2175                 int i;
2176
2177                 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count;
2178                      i++, tgt++) {
2179                         if (!tgt->ltd_exp)
2180                                 continue;
2181                         rc = obd_set_info(tgt->ltd_exp,
2182                                           keylen, key, vallen, val);
2183                         if (rc)
2184                                 RETURN(rc);
2185                 }
2186
2187                 RETURN(0);
2188         }
2189
2190         if (keylen == strlen("ids") && memcmp(key, "ids", keylen) == 0) {
2191                 struct lustre_id *id = (struct lustre_id *)val;
2192
2193                 rc = lmv_check_connect(obd);
2194                 if (rc)
2195                         RETURN(rc);
2196
2197                 rc = obd_set_info(lmv->tgts[id_group(id)].ltd_exp,
2198                                   keylen, key, vallen, val);
2199                 RETURN(rc);
2200         }
2201
2202         if (keylen == strlen("chkconnect") &&
2203             memcmp(key, "chkconnect", keylen) == 0) {
2204                 rc = lmv_check_connect(obd);
2205                 RETURN(rc);
2206         }
2207
2208         RETURN(-EINVAL);
2209 }
2210
2211 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2212                struct lov_stripe_md *lsm)
2213 {
2214         struct obd_device *obd = class_exp2obd(exp);
2215         struct lmv_obd *lmv = &obd->u.lmv;
2216         struct mea *meap, *lsmp;
2217         int mea_size, i;
2218         ENTRY;
2219
2220         mea_size = (sizeof(struct lustre_id) *
2221                     lmv->desc.ld_tgt_count) + sizeof(struct mea);
2222         if (!lmmp)
2223                 RETURN(mea_size);
2224
2225         if (*lmmp && !lsm) {
2226                 OBD_FREE(*lmmp, mea_size);
2227                 *lmmp = NULL;
2228                 RETURN(0);
2229         }
2230
2231         if (*lmmp == NULL) {
2232                 OBD_ALLOC(*lmmp, mea_size);
2233                 if (*lmmp == NULL)
2234                         RETURN(-ENOMEM);
2235         }
2236
2237         if (!lsm)
2238                 RETURN(mea_size);
2239
2240         lsmp = (struct mea *)lsm;
2241         meap = (struct mea *)*lmmp;
2242
2243         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2244             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2245                 RETURN(-EINVAL);
2246
2247         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2248         meap->mea_count = cpu_to_le32(lsmp->mea_count);
2249         meap->mea_master = cpu_to_le32(lsmp->mea_master);
2250
2251         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2252                 meap->mea_ids[i] = meap->mea_ids[i];
2253                 id_cpu_to_le(&meap->mea_ids[i]);
2254         }
2255
2256         RETURN(mea_size);
2257 }
2258
2259 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2260                  struct lov_mds_md *lmm, int lmm_size)
2261 {
2262         struct obd_device *obd = class_exp2obd(exp);
2263         struct mea **tmea = (struct mea **)lsmp;
2264         struct mea *mea = (struct mea *)lmm;
2265         struct lmv_obd *lmv = &obd->u.lmv;
2266         int mea_size, i, rc = 0;
2267         __u32 magic;
2268         ENTRY;
2269
2270         mea_size = sizeof(struct lustre_id) *
2271                 lmv->desc.ld_tgt_count + sizeof(struct mea);
2272
2273         if (lsmp == NULL)
2274                 return mea_size;
2275
2276         if (*lsmp != NULL && lmm == NULL) {
2277                 OBD_FREE(*tmea, mea_size);
2278                 RETURN(0);
2279         }
2280
2281         LASSERT(mea_size == lmm_size);
2282
2283         OBD_ALLOC(*tmea, mea_size);
2284         if (*tmea == NULL)
2285                 RETURN(-ENOMEM);
2286
2287         if (!lmm)
2288                 RETURN(mea_size);
2289
2290         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2291             mea->mea_magic == MEA_MAGIC_ALL_CHARS)
2292         {
2293                 magic = le32_to_cpu(mea->mea_magic);
2294         } else {
2295                 struct mea_old *old = (struct mea_old *)lmm;
2296
2297                 mea_size = sizeof(struct lustre_id) * old->mea_count +
2298                         sizeof(struct mea_old);
2299
2300                 if (old->mea_count > 256 || old->mea_master > 256 ||
2301                     lmm_size < mea_size || old->mea_master > old->mea_count) {
2302                         CWARN("bad MEA: count %u, master %u, size %u\n",
2303                               old->mea_count, old->mea_master, mea_size);
2304                         GOTO(out_free_mea, rc = -EINVAL);
2305                 }
2306                 magic = MEA_MAGIC_LAST_CHAR;
2307         }
2308
2309         (*tmea)->mea_magic = magic;
2310         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2311         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2312
2313         for (i = 0; i < (*tmea)->mea_count; i++) {
2314                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2315                 id_le_to_cpu(&(*tmea)->mea_ids[i]);
2316         }
2317         RETURN(mea_size);
2318
2319 out_free_mea:
2320         OBD_FREE(*tmea, mea_size);
2321         return rc;
2322 }
2323
2324 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
2325             struct lov_stripe_md *ea, obd_count oa_bufs,
2326             struct brw_page *pgarr, struct obd_trans_info *oti)
2327 {
2328         struct obd_device *obd = exp->exp_obd;
2329         struct lmv_obd *lmv = &obd->u.lmv;
2330         struct mea *mea = (struct mea *) ea;
2331         int err;
2332
2333         LASSERT(oa != NULL);
2334         LASSERT(ea != NULL);
2335         LASSERT(pgarr != NULL);
2336         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
2337
2338         oa->o_gr = id_gen(&mea->mea_ids[oa->o_mds]);
2339         oa->o_id = id_ino(&mea->mea_ids[oa->o_mds]);
2340         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2341
2342         err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp,
2343                       oa, NULL, oa_bufs, pgarr, oti);
2344         RETURN(err);
2345 }
2346
2347 static int lmv_cancel_unused(struct obd_export *exp,
2348                              struct lov_stripe_md *lsm,
2349                              int flags, void *opaque)
2350 {
2351         struct obd_device *obd = exp->exp_obd;
2352         struct lmv_obd *lmv = &obd->u.lmv;
2353         int rc = 0, err, i;
2354         ENTRY;
2355
2356         LASSERT(lsm == NULL);
2357
2358         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2359                 if (!lmv->tgts[i].ltd_exp || !lmv->tgts[i].active)
2360                         continue;
2361
2362                 err = obd_cancel_unused(lmv->tgts[i].ltd_exp,
2363                                         NULL, flags, opaque);
2364                 if (!rc)
2365                         rc = err;
2366         }
2367         RETURN(rc);
2368 }
2369
2370 struct obd_ops lmv_obd_ops = {
2371         .o_owner                = THIS_MODULE,
2372         .o_attach               = lmv_attach,
2373         .o_detach               = lmv_detach,
2374         .o_setup                = lmv_setup,
2375         .o_cleanup              = lmv_cleanup,
2376         .o_precleanup           = lmv_precleanup,
2377         .o_process_config       = lmv_process_config,
2378         .o_connect              = lmv_connect,
2379         .o_disconnect           = lmv_disconnect,
2380         .o_statfs               = lmv_statfs,
2381         .o_llog_init            = lmv_llog_init,
2382         .o_llog_finish          = lmv_llog_finish,
2383         .o_get_info             = lmv_get_info,
2384         .o_set_info             = lmv_set_info,
2385         .o_create               = lmv_obd_create,
2386         .o_packmd               = lmv_packmd,
2387         .o_unpackmd             = lmv_unpackmd,
2388         .o_brw                  = lmv_brw,
2389         .o_init_ea_size         = lmv_init_ea_size,
2390         .o_notify               = lmv_notify,
2391         .o_iocontrol            = lmv_iocontrol,
2392         .o_cancel_unused        = lmv_cancel_unused,
2393 };
2394
2395 struct md_ops lmv_md_ops = {
2396         .m_getstatus           = lmv_getstatus,
2397         .m_getattr             = lmv_getattr,
2398         .m_change_cbdata       = lmv_change_cbdata,
2399         .m_change_cbdata_name  = lmv_change_cbdata_name,
2400         .m_close               = lmv_close,
2401         .m_create              = lmv_create,
2402         .m_done_writing        = lmv_done_writing,
2403         .m_enqueue             = lmv_enqueue,
2404         .m_getattr_lock        = lmv_getattr_lock,
2405         .m_intent_lock         = lmv_intent_lock,
2406         .m_link                = lmv_link,
2407         .m_rename              = lmv_rename,
2408         .m_setattr             = lmv_setattr,
2409         .m_sync                = lmv_sync,
2410         .m_readpage            = lmv_readpage,
2411         .m_unlink              = lmv_unlink,
2412         .m_get_real_obd        = lmv_get_real_obd,
2413         .m_valid_attrs         = lmv_valid_attrs,
2414         .m_delete_inode        = lmv_delete_inode,
2415         .m_access_check        = lmv_access_check,
2416 };
2417
2418 int __init lmv_init(void)
2419 {
2420         struct lprocfs_static_vars lvars;
2421         int rc;
2422
2423         obj_cache = kmem_cache_create("lmv_objects",
2424                                       sizeof(struct lmv_obj),
2425                                       0, 0, NULL, NULL);
2426         if (!obj_cache) {
2427                 CERROR("error allocating lmv objects cache\n");
2428                 return -ENOMEM;
2429         }
2430
2431         lprocfs_init_vars(lmv, &lvars);
2432         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2433                                  lvars.module_vars,
2434                                  OBD_LMV_DEVICENAME);
2435         if (rc)
2436                 kmem_cache_destroy(obj_cache);
2437
2438         return rc;
2439 }
2440
2441 #ifdef __KERNEL__
2442 static void lmv_exit(void)
2443 {
2444         class_unregister_type(OBD_LMV_DEVICENAME);
2445
2446         LASSERTF(kmem_cache_destroy(obj_cache) == 0,
2447                  "can't free lmv objects cache, %d object(s)"
2448                  "still in use\n", atomic_read(&obj_cache_count));
2449 }
2450
2451 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2452 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2453 MODULE_LICENSE("GPL");
2454
2455 module_init(lmv_init);
2456 module_exit(lmv_exit);
2457 #endif