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         /* all real clients should perform actual connection right away, because
 242          * it is possible, that LMV will not have opportunity to connect
 243          * targets, as MDC stuff will be called directly, for instance while
 244          * reading ../mdc/../kbytesfree procfs file, etc.
 245          */
 246         if (flags & OBD_OPT_REAL_CLIENT)
 247                 rc = lmv_check_connect(obd);
 248
 249 #ifdef __KERNEL__
 250         if (rc) {
 251                 if (lmv_proc_dir)
 252                         lprocfs_remove(lmv_proc_dir);
 253         }
 254 #endif
 255
 256         RETURN(rc);
 257 }
 258
 259 static void lmv_set_timeouts(struct obd_device *obd)
 260 {
 261         struct lmv_tgt_desc *tgts;
 262         struct lmv_obd *lmv;
 263         int i;
 264
 265         lmv = &obd->u.lmv;
 266         if (lmv->server_timeout == 0)
 267                 return;
 268
 269         if (lmv->connected == 0)
 270                 return;
 271
 272         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
 273                 if (tgts->ltd_exp == NULL)
 274                         continue;
 275
 276                 obd_set_info(tgts->ltd_exp, strlen("inter_mds"),
 277                              "inter_mds", 0, NULL);
 278         }
 279 }
 280
 281 static int lmv_init_ea_size(struct obd_export *exp, int easize,
 282                             int cookiesize)
 283 {
 284         struct obd_device *obd = exp->exp_obd;
 285         struct lmv_obd *lmv = &obd->u.lmv;
 286         int i, rc = 0, change = 0;
 287         ENTRY;
 288
 289         if (lmv->max_easize < easize) {
 290                 lmv->max_easize = easize;
 291                 change = 1;
 292         }
 293         if (lmv->max_cookiesize < cookiesize) {
 294                 lmv->max_cookiesize = cookiesize;
 295                 change = 1;
 296         }
 297         if (change == 0)
 298                 RETURN(0);
 299
 300         if (lmv->connected == 0)
 301                 RETURN(0);
 302
 303         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 304                 if (lmv->tgts[i].ltd_exp == NULL) {
 305                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 306                         continue;
 307                 }
 308
 309                 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
 310                 if (rc) {
 311                         CERROR("obd_init_ea_size() failed on MDT target %d, "
 312                                "error %d.\n", i, rc);
 313                         break;
 314                 }
 315         }
 316         RETURN(rc);
 317 }
 318
 319 #define MAX_STRING_SIZE 128
 320
 321 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 322 {
 323         struct lmv_obd *lmv = &obd->u.lmv;
 324         struct obd_uuid *cluuid = &lmv->cluuid;
 325         struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
 326         struct lustre_handle conn = {0, };
 327         struct obd_device *mdc_obd;
 328         struct obd_export *mdc_exp;
 329         int rc;
 330 #ifdef __KERNEL__
 331         struct proc_dir_entry *lmv_proc_dir;
 332 #endif
 333         ENTRY;
 334
 335         /* for MDS: don't connect to yourself */
 336         if (obd_uuid_equals(&tgt->uuid, cluuid)) {
 337                 CDEBUG(D_CONFIG, "don't connect back to %s\n", cluuid->uuid);
 338                 /* XXX - the old code didn't increment active tgt count.
 339                  *       should we ? */
 340                 RETURN(0);
 341         }
 342
 343         mdc_obd = class_find_client_obd(&tgt->uuid, OBD_MDC_DEVICENAME,
 344                                         &obd->obd_uuid);
 345         if (!mdc_obd) {
 346                 CERROR("target %s not attached\n", tgt->uuid.uuid);
 347                 RETURN(-EINVAL);
 348         }
 349
 350         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 351                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 352                 tgt->uuid.uuid, obd->obd_uuid.uuid,
 353                 cluuid->uuid);
 354
 355         if (!mdc_obd->obd_set_up) {
 356                 CERROR("target %s not set up\n", tgt->uuid.uuid);
 357                 RETURN(-EINVAL);
 358         }
 359
 360         rc = obd_connect(&conn, mdc_obd, &lmv_mdc_uuid, &lmv->conn_data,
 361                          lmv->connect_flags);
 362         if (rc) {
 363                 CERROR("target %s connect error %d\n", tgt->uuid.uuid, rc);
 364                 RETURN(rc);
 365         }
 366
 367         mdc_exp = class_conn2export(&conn);
 368
 369         rc = obd_register_observer(mdc_obd, obd);
 370         if (rc) {
 371                 obd_disconnect(mdc_exp, 0);
 372                 CERROR("target %s register_observer error %d\n",
 373                        tgt->uuid.uuid, rc);
 374                 RETURN(rc);
 375         }
 376
 377         if (obd->obd_observer) {
 378                 /* tell the mds_lmv about the new target */
 379                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd, 1,
 380                                 (void *)(tgt - lmv->tgts));
 381                 if (rc) {
 382                         obd_disconnect(mdc_exp, 0);
 383                         RETURN(rc);
 384                 }
 385         }
 386
 387         tgt->ltd_exp = mdc_exp;
 388         tgt->active = 1;
 389         lmv->desc.ld_active_tgt_count++;
 390
 391         obd_init_ea_size(tgt->ltd_exp, lmv->max_easize,
 392                          lmv->max_cookiesize);
 393         CDEBUG(D_CONFIG, "connected to %s(%s) successfully (%d)\n",
 394                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 395                 atomic_read(&obd->obd_refcount));
 396
 397 #ifdef __KERNEL__
 398         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 399         if (lmv_proc_dir) {
 400                 struct proc_dir_entry *mdc_symlink;
 401                 char name[MAX_STRING_SIZE + 1];
 402
 403                 LASSERT(mdc_obd->obd_type != NULL);
 404                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 405                 name[MAX_STRING_SIZE] = '\0';
 406                 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
 407                          mdc_obd->obd_type->typ_name,
 408                          mdc_obd->obd_name);
 409                 mdc_symlink = proc_symlink(mdc_obd->obd_name,
 410                                            lmv_proc_dir, name);
 411                 if (mdc_symlink == NULL) {
 412                         CERROR("could not register LMV target "
 413                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 414                                obd->obd_type->typ_name, obd->obd_name,
 415                                mdc_obd->obd_name);
 416                         lprocfs_remove(lmv_proc_dir);
 417                         lmv_proc_dir = NULL;
 418                 }
 419         }
 420 #endif
 421         RETURN(0);
 422 }
 423
 424 int lmv_add_mdc(struct obd_device *obd, struct obd_uuid *tgt_uuid)
 425 {
 426         struct lmv_obd *lmv = &obd->u.lmv;
 427         struct lmv_tgt_desc *tgt;
 428         int rc = 0;
 429         ENTRY;
 430
 431         CDEBUG(D_CONFIG, "tgt_uuid: %s.\n", tgt_uuid->uuid);
 432
 433         lmv_init_lock(lmv);
 434
 435         if (lmv->desc.ld_active_tgt_count >= LMV_MAX_TGT_COUNT) {
 436                 lmv_init_unlock(lmv);
 437                 CERROR("can't add %s, LMV module compiled for %d MDCs. "
 438                        "That many MDCs already configured.\n",
 439                        tgt_uuid->uuid, LMV_MAX_TGT_COUNT);
 440                 RETURN(-EINVAL);
 441         }
 442         if (lmv->desc.ld_tgt_count == 0) {
 443                 struct obd_device *mdc_obd;
 444
 445                 mdc_obd = class_find_client_obd(tgt_uuid, OBD_MDC_DEVICENAME,
 446                                                 &obd->obd_uuid);
 447                 if (!mdc_obd) {
 448                         lmv_init_unlock(lmv);
 449                         CERROR("Target %s not attached\n", tgt_uuid->uuid);
 450                         RETURN(-EINVAL);
 451                 }
 452
 453                 rc = obd_llog_init(obd, &obd->obd_llogs, mdc_obd, 0, NULL);
 454                 if (rc) {
 455                         lmv_init_unlock(lmv);
 456                         CERROR("lmv failed to setup llogging subsystems\n");
 457                 }
 458         }
 459         spin_lock(&lmv->lmv_lock);
 460         tgt = lmv->tgts + lmv->desc.ld_tgt_count++;
 461         tgt->uuid = *tgt_uuid;
 462         spin_unlock(&lmv->lmv_lock);
 463
 464         if (lmv->connected) {
 465                 rc = lmv_connect_mdc(obd, tgt);
 466                 if (rc) {
 467                         spin_lock(&lmv->lmv_lock);
 468                         lmv->desc.ld_tgt_count--;
 469                         memset(tgt, 0, sizeof(*tgt));
 470                         spin_unlock(&lmv->lmv_lock);
 471                 } else {
 472                         int easize = sizeof(struct mea) +
 473                                      lmv->desc.ld_tgt_count *
 474                                      sizeof(struct lustre_id);
 475                         lmv_init_ea_size(obd->obd_self_export, easize, 0);
 476                 }
 477         }
 478
 479         lmv_init_unlock(lmv);
 480         RETURN(rc);
 481 }
 482
 483 /* performs a check if passed obd is connected. If no - connect it. */
 484 int lmv_check_connect(struct obd_device *obd)
 485 {
 486         struct lmv_obd *lmv = &obd->u.lmv;
 487         struct lmv_tgt_desc *tgt;
 488         int i, rc, easize;
 489         ENTRY;
 490
 491         if (lmv->connected)
 492                 RETURN(0);
 493
 494         lmv_init_lock(lmv);
 495         if (lmv->connected) {
 496                 lmv_init_unlock(lmv);
 497                 RETURN(0);
 498         }
 499
 500         if (lmv->desc.ld_tgt_count == 0) {
 501                 CERROR("%s: no targets configured.\n", obd->obd_name);
 502                 RETURN(-EINVAL);
 503         }
 504
 505         CDEBUG(D_CONFIG, "time to connect %s to %s\n",
 506                lmv->cluuid.uuid, obd->obd_name);
 507
 508         LASSERT(lmv->tgts != NULL);
 509
 510         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
 511                 rc = lmv_connect_mdc(obd, tgt);
 512                 if (rc)
 513                         GOTO(out_disc, rc);
 514         }
 515
 516         lmv_set_timeouts(obd);
 517         class_export_put(lmv->exp);
 518         lmv->connected = 1;
 519         easize = lmv->desc.ld_tgt_count * sizeof(struct lustre_id) +
 520                  sizeof(struct mea);
 521         lmv_init_ea_size(obd->obd_self_export, easize, 0);
 522         lmv_init_unlock(lmv);
 523         RETURN(0);
 524
 525  out_disc:
 526         while (i-- > 0) {
 527                 int rc2;
 528                 --tgt;
 529                 tgt->active = 0;
 530                 if (tgt->ltd_exp) {
 531                         --lmv->desc.ld_active_tgt_count;
 532                         rc2 = obd_disconnect(tgt->ltd_exp, 0);
 533                         if (rc2) {
 534                                 CERROR("error: LMV target %s disconnect on "
 535                                        "MDC idx %d: error %d\n",
 536                                        tgt->uuid.uuid, i, rc2);
 537                         }
 538                 }
 539         }
 540         class_disconnect(lmv->exp, 0);
 541         lmv_init_unlock(lmv);
 542         RETURN(rc);
 543 }
 544
 545 static int lmv_disconnect(struct obd_export *exp, unsigned long flags)
 546 {
 547         struct obd_device *obd = class_exp2obd(exp);
 548         struct lmv_obd *lmv = &obd->u.lmv;
 549
 550 #ifdef __KERNEL__
 551         struct proc_dir_entry *lmv_proc_dir;
 552 #endif
 553         int rc, i;
 554         ENTRY;
 555
 556         if (!lmv->tgts)
 557                 goto out_local;
 558
 559         /* Only disconnect the underlying layers on the final disconnect. */
 560         lmv->refcount--;
 561         if (lmv->refcount != 0)
 562                 goto out_local;
 563
 564 #ifdef __KERNEL__
 565         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 566 #endif
 567
 568         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 569                 struct obd_device *mdc_obd;
 570
 571                 if (lmv->tgts[i].ltd_exp == NULL)
 572                         continue;
 573
 574                 mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
 575
 576                 if (mdc_obd)
 577                         mdc_obd->obd_no_recov = obd->obd_no_recov;
 578
 579 #ifdef __KERNEL__
 580                 if (lmv_proc_dir) {
 581                         struct proc_dir_entry *mdc_symlink;
 582
 583                         mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
 584                         if (mdc_symlink) {
 585                                 lprocfs_remove(mdc_symlink);
 586                         } else {
 587                                 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 588                                        obd->obd_type->typ_name, obd->obd_name,
 589                                        mdc_obd->obd_name);
 590                         }
 591                 }
 592 #endif
 593                 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
 594                         lmv->tgts[i].ltd_exp->exp_obd->obd_name,
 595                         lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
 596
 597                 obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL);
 598                 rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags);
 599                 if (rc) {
 600                         if (lmv->tgts[i].active) {
 601                                 CERROR("Target %s disconnect error %d\n",
 602                                        lmv->tgts[i].uuid.uuid, rc);
 603                         }
 604                         rc = 0;
 605                 }
 606
 607                 lmv_activate_target(lmv, &lmv->tgts[i], 0);
 608                 lmv->tgts[i].ltd_exp = NULL;
 609         }
 610
 611 #ifdef __KERNEL__
 612         if (lmv_proc_dir) {
 613                 lprocfs_remove(lmv_proc_dir);
 614         } else {
 615                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 616                        obd->obd_type->typ_name, obd->obd_name);
 617         }
 618 #endif
 619
 620 out_local:
 621         /* this is the case when no real connection is established by
 622          * lmv_check_connect(). */
 623         if (!lmv->connected)
 624                 class_export_put(exp);
 625         rc = class_disconnect(exp, 0);
 626         if (lmv->refcount == 0)
 627                 lmv->connected = 0;
 628         RETURN(rc);
 629 }
 630
 631 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 632                          int len, void *karg, void *uarg)
 633 {
 634         struct obd_device *obddev = class_exp2obd(exp);
 635         struct lmv_obd *lmv = &obddev->u.lmv;
 636         int i, rc = 0, set = 0;
 637         ENTRY;
 638
 639         if (lmv->desc.ld_tgt_count == 0)
 640                 RETURN(-ENOTTY);
 641
 642         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 643                 int err;
 644
 645                 if (lmv->tgts[i].ltd_exp == NULL)
 646                         continue;
 647
 648                 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
 649                 if (err) {
 650                         if (lmv->tgts[i].active) {
 651                                 CERROR("error: iocontrol MDC %s on MDT"
 652                                        "idx %d: err = %d\n",
 653                                        lmv->tgts[i].uuid.uuid, i, err);
 654                                 if (!rc)
 655                                         rc = err;
 656                         }
 657                 } else
 658                         set = 1;
 659         }
 660         if (!set && !rc)
 661                 rc = -EIO;
 662
 663         RETURN(rc);
 664 }
 665
 666 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
 667 {
 668         struct lmv_obd *lmv = &obd->u.lmv;
 669         struct lustre_cfg *lcfg = buf;
 670         struct lmv_desc *desc;
 671         int rc = 0;
 672         ENTRY;
 673
 674         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 675                 CERROR("LMV setup requires a descriptor\n");
 676                 RETURN(-EINVAL);
 677         }
 678
 679         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
 680         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 681                 CERROR("descriptor size wrong: %d > %d\n",
 682                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 683                 RETURN(-EINVAL);
 684         }
 685
 686         lmv->tgts_size = LMV_MAX_TGT_COUNT * sizeof(struct lmv_tgt_desc);
 687
 688         OBD_ALLOC(lmv->tgts, lmv->tgts_size);
 689         if (lmv->tgts == NULL) {
 690                 CERROR("Out of memory\n");
 691                 RETURN(-ENOMEM);
 692         }
 693
 694         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
 695         lmv->desc.ld_tgt_count = 0;
 696         lmv->desc.ld_active_tgt_count = 0;
 697         lmv->max_cookiesize = 0;
 698         lmv->max_easize = 0;
 699
 700         spin_lock_init(&lmv->lmv_lock);
 701         sema_init(&lmv->init_sem, 1);
 702
 703         rc = lmv_setup_mgr(obd);
 704         if (rc) {
 705                 CERROR("Can't setup LMV object manager, "
 706                        "error %d.\n", rc);
 707                 OBD_FREE(lmv->tgts, lmv->tgts_size);
 708                 RETURN(rc);
 709         }
 710
 711         RETURN(rc);
 712 }
 713
 714 static int lmv_cleanup(struct obd_device *obd, int flags)
 715 {
 716         struct lmv_obd *lmv = &obd->u.lmv;
 717         ENTRY;
 718
 719         lmv_cleanup_mgr(obd);
 720         OBD_FREE(lmv->tgts, lmv->tgts_size);
 721
 722         RETURN(0);
 723 }
 724 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
 725 {
 726         struct lustre_cfg *lcfg = buf;
 727         struct obd_uuid tgt_uuid;
 728         int rc;
 729         ENTRY;
 730
 731         switch(lcfg->lcfg_command) {
 732         case LCFG_LMV_ADD_MDC:
 733                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(tgt_uuid.uuid))
 734                         GOTO(out, rc = -EINVAL);
 735
 736                 obd_str2uuid(&tgt_uuid, lustre_cfg_string(lcfg, 1));
 737                 rc = lmv_add_mdc(obd, &tgt_uuid);
 738                 GOTO(out, rc);
 739         default: {
 740                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 741                 GOTO(out, rc = -EINVAL);
 742         }
 743         }
 744 out:
 745         RETURN(rc);
 746 }
 747
 748 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 749                       unsigned long max_age)
 750 {
 751         struct lmv_obd *lmv = &obd->u.lmv;
 752         struct obd_statfs *temp;
 753         int rc = 0, i;
 754         ENTRY;
 755
 756         rc = lmv_check_connect(obd);
 757         if (rc)
 758                 RETURN(rc);
 759
 760         OBD_ALLOC(temp, sizeof(*temp));
 761         if (temp == NULL)
 762                 RETURN(-ENOMEM);
 763
 764         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 765                 if (lmv->tgts[i].ltd_exp == NULL)
 766                         continue;
 767
 768                 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp, max_age);
 769                 if (rc) {
 770                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
 771                                lmv->tgts[i].ltd_exp->exp_obd->obd_name,
 772                                rc);
 773                         GOTO(out_free_temp, rc);
 774                 }
 775                 if (i == 0) {
 776                         memcpy(osfs, temp, sizeof(*temp));
 777                 } else {
 778                         osfs->os_bavail += temp->os_bavail;
 779                         osfs->os_blocks += temp->os_blocks;
 780                         osfs->os_ffree += temp->os_ffree;
 781                         osfs->os_files += temp->os_files;
 782                 }
 783         }
 784
 785         EXIT;
 786 out_free_temp:
 787         OBD_FREE(temp, sizeof(*temp));
 788         return rc;
 789 }
 790
 791 static int lmv_getstatus(struct obd_export *exp, struct lustre_id *id)
 792 {
 793         struct obd_device *obd = exp->exp_obd;
 794         struct lmv_obd *lmv = &obd->u.lmv;
 795         int rc;
 796         ENTRY;
 797
 798         rc = lmv_check_connect(obd);
 799         if (rc)
 800                 RETURN(rc);
 801
 802         rc = md_getstatus(lmv->tgts[0].ltd_exp, id);
 803         id_group(id) = 0;
 804
 805         RETURN(rc);
 806 }
 807
 808 static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
 809                        __u64 valid, const char *xattr_name,
 810                        const void *xattr_data, unsigned int xattr_datalen,
 811                        unsigned int ea_size, struct obd_capa *ocapa,
 812                        struct ptlrpc_request **request)
 813 {
 814         struct obd_device *obd = exp->exp_obd;
 815         struct lmv_obd *lmv = &obd->u.lmv;
 816         int rc, i = id_group(id);
 817         struct lmv_obj *obj;
 818         ENTRY;
 819
 820         rc = lmv_check_connect(obd);
 821         if (rc)
 822                 RETURN(rc);
 823
 824         LASSERT(i < lmv->desc.ld_tgt_count);
 825
 826
 827         rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid,
 828                         xattr_name, xattr_data, xattr_datalen,
 829                         ea_size, ocapa, request);
 830         if (rc)
 831                 RETURN(rc);
 832
 833         obj = lmv_grab_obj(obd, id);
 834
 835         CDEBUG(D_OTHER, "GETATTR for "DLID4" %s\n",
 836                OLID4(id), obj ? "(splitted)" : "");
 837
 838         /*
 839          * if object is splitted, then we loop over all the slaves and gather
 840          * size attribute. In ideal world we would have to gather also mds field
 841          * from all slaves, as object is spread over the cluster and this is
 842          * definitely interesting information and it is not good to loss it,
 843          * but...
 844          */
 845         if (obj) {
 846                 struct mds_body *body;
 847
 848                 if (*request == NULL) {
 849                         lmv_put_obj(obj);
 850                         RETURN(rc);
 851                 }
 852
 853                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
 854                                       sizeof(*body));
 855                 LASSERT(body != NULL);
 856
 857                 lmv_lock_obj(obj);
 858
 859                 for (i = 0; i < obj->objcount; i++) {
 860
 861                         if (lmv->tgts[i].ltd_exp == NULL) {
 862                                 CWARN("%s: NULL export for %d\n",
 863                                       obd->obd_name, i);
 864                                 continue;
 865                         }
 866
 867                         /* skip master obj. */
 868                         if (id_equal_fid(&obj->id, &obj->objs[i].id))
 869                                 continue;
 870
 871                         body->size += obj->objs[i].size;
 872                 }
 873
 874                 lmv_unlock_obj(obj);
 875                 lmv_put_obj(obj);
 876         }
 877
 878         RETURN(rc);
 879 }
 880
 881 static int lmv_access_check(struct obd_export *exp,
 882                             struct lustre_id *id,
 883                             struct ptlrpc_request **request)
 884 {
 885         struct obd_device *obd = exp->exp_obd;
 886         struct lmv_obd *lmv = &obd->u.lmv;
 887         int rc, i = id_group(id);
 888         ENTRY;
 889
 890         rc = lmv_check_connect(obd);
 891         if (rc)
 892                 RETURN(rc);
 893
 894         LASSERT(i < lmv->desc.ld_tgt_count);
 895         rc = md_access_check(lmv->tgts[i].ltd_exp, id, request);
 896         RETURN(rc);
 897 }
 898
 899 static int lmv_change_cbdata(struct obd_export *exp,
 900                              struct lustre_id *id,
 901                              ldlm_iterator_t it,
 902                              void *data)
 903 {
 904         struct obd_device *obd = exp->exp_obd;
 905         struct lmv_obd *lmv = &obd->u.lmv;
 906         int i, rc;
 907         ENTRY;
 908
 909         rc = lmv_check_connect(obd);
 910         if (rc)
 911                 RETURN(rc);
 912
 913         CDEBUG(D_OTHER, "CBDATA for "DLID4"\n", OLID4(id));
 914         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
 915
 916         /* with CMD every object can have two locks in different
 917          * namespaces: lookup lock in space of mds storing direntry
 918          * and update/open lock in space of mds storing inode */
 919         for (i = 0; i < lmv->desc.ld_tgt_count; i++)
 920                 md_change_cbdata(lmv->tgts[i].ltd_exp, id, it, data);
 921
 922         RETURN(0);
 923 }
 924
 925 static int lmv_change_cbdata_name(struct obd_export *exp,
 926                                   struct lustre_id *pid,
 927                                   char *name, int len,
 928                                   struct lustre_id *cid,
 929                                   ldlm_iterator_t it,
 930                                   void *data)
 931 {
 932         struct obd_device *obd = exp->exp_obd;
 933         struct lmv_obd *lmv = &obd->u.lmv;
 934         struct lustre_id rcid = *cid;
 935         struct lmv_obj *obj;
 936         int rc = 0, mds;
 937         ENTRY;
 938
 939         rc = lmv_check_connect(obd);
 940         if (rc)
 941                 RETURN(rc);
 942
 943         LASSERT(id_group(pid) < lmv->desc.ld_tgt_count);
 944         LASSERT(id_group(cid) < lmv->desc.ld_tgt_count);
 945
 946         CDEBUG(D_OTHER, "CBDATA for "DLID4":%*s -> "DLID4"\n",
 947                OLID4(pid), len, name, OLID4(cid));
 948
 949         /* this is default mds for directory name belongs to. */
 950         mds = id_group(pid);
 951         obj = lmv_grab_obj(obd, pid);
 952         if (obj) {
 953                 /* directory is splitted. look for right mds for this name. */
 954                 mds = raw_name2idx(obj->hashtype, obj->objcount, name, len);
 955                 rcid = obj->objs[mds].id;
 956                 mds = id_group(&rcid);
 957                 lmv_put_obj(obj);
 958         }
 959         rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, &rcid, it, data);
 960         RETURN(rc);
 961 }
 962
 963 static int lmv_valid_attrs(struct obd_export *exp, struct lustre_id *id)
 964 {
 965         struct obd_device *obd = exp->exp_obd;
 966         struct lmv_obd *lmv = &obd->u.lmv;
 967         int rc = 0;
 968         ENTRY;
 969
 970         rc = lmv_check_connect(obd);
 971         if (rc)
 972                 RETURN(rc);
 973
 974         CDEBUG(D_OTHER, "validate "DLID4"\n", OLID4(id));
 975         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
 976         rc = md_valid_attrs(lmv->tgts[id_group(id)].ltd_exp, id);
 977         RETURN(rc);
 978 }
 979
 980 static int lmv_close(struct obd_export *exp, struct mdc_op_data *op_data,
 981                      struct obd_client_handle *och,
 982                      struct ptlrpc_request **request)
 983 {
 984         struct obd_device *obd = exp->exp_obd;
 985         struct lmv_obd *lmv = &obd->u.lmv;
 986         int rc, i = id_group(&op_data->id1);
 987         ENTRY;
 988
 989         rc = lmv_check_connect(obd);
 990         if (rc)
 991                 RETURN(rc);
 992
 993         LASSERT(i < lmv->desc.ld_tgt_count);
 994         CDEBUG(D_OTHER, "CLOSE "DLID4"\n", OLID4(&op_data->id1));
 995         rc = md_close(lmv->tgts[i].ltd_exp, op_data, och, request);
 996         RETURN(rc);
 997 }
 998
 999 int lmv_get_mea_and_update_object(struct obd_export *exp,
1000                                   struct lustre_id *id)
1001 {
1002         struct obd_device *obd = exp->exp_obd;
1003         struct lmv_obd *lmv = &obd->u.lmv;
1004         struct ptlrpc_request *req = NULL;
1005         struct lmv_obj *obj;
1006         struct lustre_md md;
1007         int mealen, rc;
1008         __u64 valid;
1009         ENTRY;
1010
1011         md.mea = NULL;
1012         mealen = MEA_SIZE_LMV(lmv);
1013
1014         valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA;
1015
1016         /* time to update mea of parent id */
1017         rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp,
1018                         id, valid, NULL, NULL, 0, mealen, NULL, &req);
1019         if (rc) {
1020                 CERROR("md_getattr() failed, error %d\n", rc);
1021                 GOTO(cleanup, rc);
1022         }
1023
1024         rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
1025         if (rc) {
1026                 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
1027                 GOTO(cleanup, rc);
1028         }
1029
1030         if (md.mea == NULL)
1031                 GOTO(cleanup, rc = -ENODATA);
1032
1033         obj = lmv_create_obj(exp, id, md.mea);
1034         if (IS_ERR(obj))
1035                 rc = PTR_ERR(obj);
1036         else
1037                 lmv_put_obj(obj);
1038
1039         obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
1040
1041         EXIT;
1042 cleanup:
1043         if (req)
1044                 ptlrpc_req_finished(req);
1045         return rc;
1046 }
1047
1048 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
1049                const void *data, int datalen, int mode, __u32 uid,
1050                __u32 gid, __u64 rdev, struct ptlrpc_request **request)
1051 {
1052         struct obd_device *obd = exp->exp_obd;
1053         struct lmv_obd *lmv = &obd->u.lmv;
1054         struct mds_body *body;
1055         struct lmv_obj *obj;
1056         int rc, mds, loop = 0;
1057         ENTRY;
1058
1059         rc = lmv_check_connect(obd);
1060         if (rc)
1061                 RETURN(rc);
1062
1063         if (!lmv->desc.ld_active_tgt_count)
1064                 RETURN(-EIO);
1065 repeat:
1066         LASSERT(++loop <= 2);
1067         obj = lmv_grab_obj(obd, &op_data->id1);
1068         if (obj) {
1069                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1070                                    op_data->name, op_data->namelen);
1071                 op_data->id1 = obj->objs[mds].id;
1072                 lmv_put_obj(obj);
1073         }
1074
1075         CDEBUG(D_OTHER, "CREATE '%*s' on "DLID4"\n", op_data->namelen,
1076                op_data->name, OLID4(&op_data->id1));
1077
1078         rc = md_create(lmv->tgts[id_group(&op_data->id1)].ltd_exp,
1079                        op_data, data, datalen, mode, uid, gid, rdev,
1080                        request);
1081         if (rc == 0) {
1082                 if (*request == NULL)
1083                         RETURN(rc);
1084
1085                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
1086                                       sizeof(*body));
1087                 if (body == NULL)
1088                         RETURN(-ENOMEM);
1089
1090                 CDEBUG(D_OTHER, "created. "DLID4"\n", OLID4(&op_data->id1));
1091         } else if (rc == -ERESTART) {
1092                 /*
1093                  * directory got splitted. time to update local object and
1094                  * repeat the request with proper MDS.
1095                  */
1096                 rc = lmv_get_mea_and_update_object(exp, &op_data->id1);
1097                 if (rc == 0) {
1098                         ptlrpc_req_finished(*request);
1099                         goto repeat;
1100                 }
1101         }
1102         RETURN(rc);
1103 }
1104
1105 static int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
1106 {
1107         struct obd_device *obd = exp->exp_obd;
1108         struct lmv_obd *lmv = &obd->u.lmv;
1109         int rc;
1110         ENTRY;
1111
1112         rc = lmv_check_connect(obd);
1113         if (rc)
1114                 RETURN(rc);
1115
1116         /* FIXME: choose right MDC here */
1117         CWARN("this method isn't implemented yet\n");
1118         rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
1119         RETURN(rc);
1120 }
1121
1122 static int
1123 lmv_enqueue_slaves(struct obd_export *exp, int locktype,
1124                    struct lookup_intent *it, int lockmode,
1125                    struct mdc_op_data *data, struct lustre_handle *lockh,
1126                    void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1127                    ldlm_blocking_callback cb_blocking, void *cb_data)
1128 {
1129         struct obd_device *obd = exp->exp_obd;
1130         struct lmv_obd *lmv = &obd->u.lmv;
1131         struct mea *mea = data->mea1;
1132         struct mdc_op_data *data2;
1133         int i, rc, mds;
1134         ENTRY;
1135
1136         OBD_ALLOC(data2, sizeof(*data2));
1137         if (data2 == NULL)
1138                 RETURN(-ENOMEM);
1139
1140         LASSERT(mea != NULL);
1141         for (i = 0; i < mea->mea_count; i++) {
1142                 memset(data2, 0, sizeof(*data2));
1143                 data2->id1 = mea->mea_ids[i];
1144                 mds = id_group(&data2->id1);
1145
1146                 if (lmv->tgts[mds].ltd_exp == NULL)
1147                         continue;
1148
1149                 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it,
1150                                 lockmode, data2, lockh + i, lmm, lmmsize,
1151                                 cb_compl, cb_blocking, cb_data);
1152
1153                 CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n",
1154                        OLID4(&mea->mea_ids[i]), rc, LUSTRE_IT(it)->it_status);
1155                 if (rc)
1156                         GOTO(cleanup, rc);
1157                 if (LUSTRE_IT(it)->it_data) {
1158                         struct ptlrpc_request *req;
1159                         req = (struct ptlrpc_request *) LUSTRE_IT(it)->it_data;
1160                         ptlrpc_req_finished(req);
1161                 }
1162
1163                 if (LUSTRE_IT(it)->it_status)
1164                         GOTO(cleanup, rc = LUSTRE_IT(it)->it_status);
1165         }
1166
1167         OBD_FREE(data2, sizeof(*data2));
1168         RETURN(0);
1169 cleanup:
1170         OBD_FREE(data2, sizeof(*data2));
1171
1172         /* drop all taken locks */
1173         while (--i >= 0) {
1174                 if (lockh[i].cookie)
1175                         ldlm_lock_decref(lockh + i, lockmode);
1176                 lockh[i].cookie = 0;
1177         }
1178         return rc;
1179 }
1180
1181 static int
1182 lmv_enqueue_remote(struct obd_export *exp, int lock_type,
1183                    struct lookup_intent *it, int lock_mode,
1184                    struct mdc_op_data *data, struct lustre_handle *lockh,
1185                    void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1186                    ldlm_blocking_callback cb_blocking, void *cb_data)
1187 {
1188         struct ptlrpc_request *req = LUSTRE_IT(it)->it_data;
1189         struct obd_device *obd = exp->exp_obd;
1190         struct lmv_obd *lmv = &obd->u.lmv;
1191         struct lustre_handle plock;
1192         struct mdc_op_data rdata;
1193         struct mds_body *body = NULL;
1194         int rc = 0, pmode;
1195         ENTRY;
1196
1197         body = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body));
1198         LASSERT(body != NULL);
1199
1200         if (!(body->valid & OBD_MD_MDS))
1201                 RETURN(0);
1202
1203         CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4" -> "DLID4"\n",
1204                LL_IT2STR(it), OLID4(&data->id1), OLID4(&body->id1));
1205
1206         /* we got LOOKUP lock, but we really need attrs */
1207         pmode = LUSTRE_IT(it)->it_lock_mode;
1208         LASSERT(pmode != 0);
1209         memcpy(&plock, lockh, sizeof(plock));
1210         LUSTRE_IT(it)->it_lock_mode = 0;
1211         LUSTRE_IT(it)->it_data = NULL;
1212         LASSERT((body->valid & OBD_MD_FID) != 0);
1213
1214         memcpy(&rdata, data, sizeof(rdata));
1215         rdata.id1 = body->id1;
1216         rdata.name = NULL;
1217         rdata.namelen = 0;
1218
1219         LUSTRE_IT(it)->it_disposition &= ~DISP_ENQ_COMPLETE;
1220         ptlrpc_req_finished(req);
1221
1222         rc = md_enqueue(lmv->tgts[id_group(&rdata.id1)].ltd_exp,
1223                         lock_type, it, lock_mode, &rdata, lockh, lmm,
1224                         lmmsize, cb_compl, cb_blocking, cb_data);
1225         ldlm_lock_decref(&plock, pmode);
1226         RETURN(rc);
1227 }
1228
1229 static int
1230 lmv_enqueue(struct obd_export *exp, int lock_type,
1231             struct lookup_intent *it, int lock_mode,
1232             struct mdc_op_data *data, struct lustre_handle *lockh,
1233             void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1234             ldlm_blocking_callback cb_blocking, void *cb_data)
1235 {
1236         struct obd_device *obd = exp->exp_obd;
1237         struct lmv_obd *lmv = &obd->u.lmv;
1238         struct lmv_obj *obj;
1239         int rc, mds;
1240         ENTRY;
1241
1242         rc = lmv_check_connect(obd);
1243         if (rc)
1244                 RETURN(rc);
1245
1246         if (data->mea1 && it->it_op == IT_UNLINK) {
1247                 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
1248                                         data, lockh, lmm, lmmsize,
1249                                         cb_compl, cb_blocking, cb_data);
1250                 RETURN(rc);
1251         }
1252
1253         if (data->namelen) {
1254                 obj = lmv_grab_obj(obd, &data->id1);
1255                 if (obj) {
1256                         /* directory is splitted. look for right mds for this
1257                          * name */
1258                         mds = raw_name2idx(obj->hashtype, obj->objcount,
1259                                            (char *)data->name, data->namelen);
1260                         data->id1 = obj->objs[mds].id;
1261                         lmv_put_obj(obj);
1262                 }
1263         }
1264         CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4"\n", LL_IT2STR(it),
1265                OLID4(&data->id1));
1266
1267         rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp,
1268                         lock_type, it, lock_mode, data, lockh, lmm,
1269                         lmmsize, cb_compl, cb_blocking, cb_data);
1270         if (rc == 0 && it->it_op == IT_OPEN)
1271                 rc = lmv_enqueue_remote(exp, lock_type, it, lock_mode,
1272                                         data, lockh, lmm, lmmsize,
1273                                         cb_compl, cb_blocking, cb_data);
1274         RETURN(rc);
1275 }
1276
1277 static int
1278 lmv_getattr_lock(struct obd_export *exp, struct lustre_id *id,
1279                  char *filename, int namelen, __u64 valid,
1280                  unsigned int ea_size, struct ptlrpc_request **request)
1281 {
1282         int rc, mds = id_group(id), loop = 0;
1283         struct obd_device *obd = exp->exp_obd;
1284         struct lmv_obd *lmv = &obd->u.lmv;
1285         struct lustre_id rid = *id;
1286         struct mds_body *body;
1287         struct lmv_obj *obj;
1288         ENTRY;
1289
1290         rc = lmv_check_connect(obd);
1291         if (rc)
1292                 RETURN(rc);
1293 repeat:
1294         LASSERT(++loop <= 2);
1295         obj = lmv_grab_obj(obd, id);
1296         if (obj) {
1297                 /* directory is splitted. look for right mds for this name */
1298                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1299                                    filename, namelen - 1);
1300                 rid = obj->objs[mds].id;
1301                 lmv_put_obj(obj);
1302         }
1303
1304         CDEBUG(D_OTHER, "getattr_lock for %*s on "DLID4" -> "DLID4"\n",
1305                namelen, filename, OLID4(id), OLID4(&rid));
1306
1307         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1308                              &rid, filename, namelen,
1309                              valid == OBD_MD_FLID ? valid : valid | OBD_MD_FID,
1310                              ea_size, request);
1311         if (rc == 0) {
1312                 /*
1313                  * this could be cross-node reference. in this case all we have
1314                  * right now is lustre_id triple. we'd like to find other
1315                  * attributes.
1316                  */
1317                 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
1318                 LASSERT(body != NULL);
1319                 LASSERT((body->valid & OBD_MD_FID) != 0
1320                                 || body->valid == OBD_MD_FLID);
1321
1322                 if (body->valid & OBD_MD_MDS) {
1323                         struct ptlrpc_request *req = NULL;
1324
1325                         rid = body->id1;
1326                         CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid));
1327
1328                         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1329                                              &rid, NULL, 1, valid, ea_size, &req);
1330                         ptlrpc_req_finished(*request);
1331                         *request = req;
1332                 }
1333         } else if (rc == -ERESTART) {
1334                 /* directory got splitted. time to update local object and
1335                  * repeat the request with proper MDS */
1336                 rc = lmv_get_mea_and_update_object(exp, &rid);
1337                 if (rc == 0) {
1338                         ptlrpc_req_finished(*request);
1339                         goto repeat;
1340                 }
1341         }
1342         RETURN(rc);
1343 }
1344
1345 /*
1346  * llite passes id of an target inode in data->id1 and id of directory in
1347  * data->id2
1348  */
1349 static int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
1350                     struct ptlrpc_request **request)
1351 {
1352         struct obd_device *obd = exp->exp_obd;
1353         struct lmv_obd *lmv = &obd->u.lmv;
1354         struct lmv_obj *obj;
1355         int rc, mds;
1356         ENTRY;
1357
1358         rc = lmv_check_connect(obd);
1359         if (rc)
1360                 RETURN(rc);
1361
1362         if (data->namelen != 0) {
1363                 /* usual link request */
1364                 obj = lmv_grab_obj(obd, &data->id2);
1365                 if (obj) {
1366                         rc = raw_name2idx(obj->hashtype, obj->objcount,
1367                                           data->name, data->namelen);
1368                         data->id2 = obj->objs[rc].id;
1369                         lmv_put_obj(obj);
1370                 }
1371
1372                 mds = id_group(&data->id2);
1373
1374                 CDEBUG(D_OTHER,"link "DLID4":%*s to "DLID4"\n",
1375                        OLID4(&data->id2), data->namelen, data->name,
1376                        OLID4(&data->id1));
1377         } else {
1378                 mds = id_group(&data->id1);
1379
1380                 /* request from MDS to acquire i_links for inode by id1 */
1381                 CDEBUG(D_OTHER, "inc i_nlinks for "DLID4"\n",
1382                        OLID4(&data->id1));
1383         }
1384
1385         CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n",
1386                mds, OLID4(&data->id1));
1387         rc = md_link(lmv->tgts[mds].ltd_exp, data, request);
1388
1389         RETURN(rc);
1390 }
1391
1392 static int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1393                       const char *old, int oldlen, const char *new, int newlen,
1394                       struct ptlrpc_request **request)
1395 {
1396         struct obd_device *obd = exp->exp_obd;
1397         struct lmv_obd *lmv = &obd->u.lmv;
1398         struct lmv_obj *obj;
1399         int rc, mds;
1400         ENTRY;
1401
1402         CDEBUG(D_OTHER, "rename %*s in "DLID4" to %*s in "DLID4"\n",
1403                oldlen, old, OLID4(&data->id1), newlen, new,
1404                OLID4(&data->id2));
1405
1406         rc = lmv_check_connect(obd);
1407         if (rc)
1408                 RETURN(rc);
1409
1410         if (oldlen == 0) {
1411                 /*
1412                  * MDS with old dir entry is asking another MDS to create name
1413                  * there.
1414                  */
1415                 CDEBUG(D_OTHER,
1416                        "create %*s(%d/%d) in "DLID4" pointing "
1417                        "to "DLID4"\n", newlen, new, oldlen, newlen,
1418                        OLID4(&data->id2), OLID4(&data->id1));
1419
1420                 mds = id_group(&data->id2);
1421
1422                 /*
1423                  * target directory can be splitted, sowe should forward request
1424                  * to the right MDS.
1425                  */
1426                 obj = lmv_grab_obj(obd, &data->id2);
1427                 if (obj) {
1428                         mds = raw_name2idx(obj->hashtype, obj->objcount,
1429                                            (char *)new, newlen);
1430                         data->id2 = obj->objs[mds].id;
1431                         CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1432                                OLID4(&data->id2));
1433                         lmv_put_obj(obj);
1434                 }
1435                 goto request;
1436         }
1437
1438         obj = lmv_grab_obj(obd, &data->id1);
1439         if (obj) {
1440                 /*
1441                  * directory is already splitted, so we have to forward request
1442                  * to the right MDS.
1443                  */
1444                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1445                                    (char *)old, oldlen);
1446                 data->id1 = obj->objs[mds].id;
1447                 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1448                        OLID4(&data->id1));
1449                 lmv_put_obj(obj);
1450         }
1451
1452         obj = lmv_grab_obj(obd, &data->id2);
1453         if (obj) {
1454                 /*
1455                  * directory is already splitted, so we have to forward request
1456                  * to the right MDS.
1457                  */
1458                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1459                                    (char *)new, newlen);
1460
1461                 data->id2 = obj->objs[mds].id;
1462                 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1463                        OLID4(&data->id2));
1464                 lmv_put_obj(obj);
1465         }
1466
1467         mds = id_group(&data->id1);
1468
1469 request:
1470         if (id_group(&data->id1) != id_group(&data->id2)) {
1471                 CDEBUG(D_OTHER,"cross-node rename "DLID4"/%*s to "DLID4"/%*s\n",
1472                        OLID4(&data->id1), oldlen, old, OLID4(&data->id2),
1473                        newlen, new);
1474         }
1475
1476         rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1477                        new, newlen, request);
1478         RETURN(rc);
1479 }
1480
1481 static int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1482                        struct iattr *iattr, void *ea, int ealen, void *ea2,
1483                        int ea2len, void *ea3, int ea3len,
1484                        struct ptlrpc_request **request)
1485 {
1486         struct obd_device *obd = exp->exp_obd;
1487         struct lmv_obd *lmv = &obd->u.lmv;
1488         struct ptlrpc_request *req;
1489         struct mds_body *body;
1490         struct lmv_obj *obj;
1491         int rc = 0, i;
1492         ENTRY;
1493
1494         rc = lmv_check_connect(obd);
1495         if (rc)
1496                 RETURN(rc);
1497
1498         obj = lmv_grab_obj(obd, &data->id1);
1499
1500         CDEBUG(D_OTHER, "SETATTR for "DLID4", valid 0x%x%s\n",
1501                OLID4(&data->id1), iattr->ia_valid, obj ? ", splitted" : "");
1502
1503         if (obj) {
1504                 for (i = 0; i < obj->objcount; i++) {
1505                         data->id1 = obj->objs[i].id;
1506
1507                         rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1508                                         data, iattr, ea, ealen, ea2, ea2len,
1509                                         ea3, ea3len, &req);
1510
1511                         if (id_equal_fid(&obj->id, &obj->objs[i].id)) {
1512                                 /*
1513                                  * this is master object and this request should
1514                                  * be returned back to llite.
1515                                  */
1516                                 *request = req;
1517                         } else {
1518                                 ptlrpc_req_finished(req);
1519                         }
1520
1521                         if (rc)
1522                                 break;
1523                 }
1524                 lmv_put_obj(obj);
1525         } else {
1526                 LASSERT(id_group(&data->id1) < lmv->desc.ld_tgt_count);
1527                 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1528                                 data, iattr, ea, ealen, ea2, ea2len, ea3,
1529                                 ea3len, request);
1530                 if (rc == 0) {
1531                         body = lustre_msg_buf((*request)->rq_repmsg, 0,
1532                                               sizeof(*body));
1533                         LASSERT(body != NULL);
1534                         LASSERT((body->valid & OBD_MD_FID) != 0);
1535                         LASSERT(id_group(&body->id1) == id_group(&data->id1));
1536                 }
1537         }
1538         RETURN(rc);
1539 }
1540
1541 static int lmv_sync(struct obd_export *exp, struct lustre_id *id,
1542                     struct ptlrpc_request **request)
1543 {
1544         struct obd_device *obd = exp->exp_obd;
1545         struct lmv_obd *lmv = &obd->u.lmv;
1546         int rc;
1547         ENTRY;
1548
1549         rc = lmv_check_connect(obd);
1550         if (rc)
1551                 RETURN(rc);
1552
1553         rc = md_sync(lmv->tgts[id_group(id)].ltd_exp,
1554                      id, request);
1555         RETURN(rc);
1556 }
1557
1558 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock,
1559                             struct ldlm_lock_desc *desc,
1560                             void *data, int flag)
1561 {
1562         struct lustre_handle lockh;
1563         struct lmv_obj *obj;
1564         int rc;
1565         ENTRY;
1566
1567         switch (flag) {
1568         case LDLM_CB_BLOCKING:
1569                 ldlm_lock2handle(lock, &lockh);
1570                 rc = ldlm_cli_cancel(&lockh);
1571                 if (rc < 0) {
1572                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1573                         RETURN(rc);
1574                 }
1575                 break;
1576         case LDLM_CB_CANCELING:
1577                 /* time to drop cached attrs for dirobj */
1578                 obj = lock->l_ast_data;
1579                 if (obj) {
1580                         CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
1581                                ", master "DLID4"\n",
1582                                lock->l_resource->lr_name.name[3] == 1 ?
1583                                "LOOKUP" : "UPDATE",
1584                                lock->l_resource->lr_name.name[0],
1585                                lock->l_resource->lr_name.name[1],
1586                                OLID4(&obj->id));
1587                         lmv_put_obj(obj);
1588                 }
1589                 break;
1590         default:
1591                 LBUG();
1592         }
1593         RETURN(0);
1594 }
1595
1596 static void lmv_remove_dots(struct page *page)
1597 {
1598         unsigned limit = PAGE_CACHE_SIZE;
1599         char *kaddr = page_address(page);
1600         struct ext2_dir_entry_2 *p;
1601         unsigned offs, rec_len;
1602
1603         for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1604                 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1605                 rec_len = le16_to_cpu(p->rec_len);
1606
1607                 if ((p->name_len == 1 && p->name[0] == '.') ||
1608                     (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1609                         p->inode = 0;
1610         }
1611 }
1612
1613 static int lmv_readpage(struct obd_export *exp, struct lustre_id *id,
1614                         __u64 offset, struct page *page,
1615                         struct ptlrpc_request **request)
1616 {
1617         struct obd_device *obd = exp->exp_obd;
1618         struct lmv_obd *lmv = &obd->u.lmv;
1619         struct lustre_id rid = *id;
1620         struct lmv_obj *obj;
1621         int rc, i;
1622         ENTRY;
1623
1624         rc = lmv_check_connect(obd);
1625         if (rc)
1626                 RETURN(rc);
1627
1628         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
1629         CDEBUG(D_OTHER, "READPAGE at %llu from "DLID4"\n",
1630                offset, OLID4(&rid));
1631
1632         obj = lmv_grab_obj(obd, id);
1633         if (obj) {
1634                 lmv_lock_obj(obj);
1635
1636                 /* find dirobj containing page with requested offset. */
1637                 for (i = 0; i < obj->objcount; i++) {
1638                         if (offset < obj->objs[i].size)
1639                                 break;
1640                         offset -= obj->objs[i].size;
1641                 }
1642                 rid = obj->objs[i].id;
1643
1644                 lmv_unlock_obj(obj);
1645                 lmv_put_obj(obj);
1646
1647                 CDEBUG(D_OTHER, "forward to "DLID4" with offset %lu\n",
1648                        OLID4(&rid), (unsigned long)offset);
1649         }
1650         rc = md_readpage(lmv->tgts[id_group(&rid)].ltd_exp, &rid,
1651                          offset, page, request);
1652
1653         if (rc == 0 && !id_equal_fid(&rid, id))
1654                 /* this page isn't from master object. To avoid "." and ".."
1655                  * duplication in directory, we have to remove them from all
1656                  * slave objects */
1657                 lmv_remove_dots(page);
1658
1659         RETURN(rc);
1660 }
1661
1662 static int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1663                              struct ptlrpc_request **req)
1664 {
1665         struct obd_device *obd = exp->exp_obd;
1666         struct lmv_obd *lmv = &obd->u.lmv;
1667         struct mea *mea = data->mea1;
1668         struct mdc_op_data *data2;
1669         int i, rc = 0;
1670         ENTRY;
1671
1672         OBD_ALLOC(data2, sizeof(*data2));
1673         if (data2 == NULL)
1674                 RETURN(-ENOMEM);
1675
1676         LASSERT(mea != NULL);
1677         for (i = 0; i < mea->mea_count; i++) {
1678                 memset(data2, 0, sizeof(*data2));
1679                 data2->id1 = mea->mea_ids[i];
1680                 data2->create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1681
1682                 if (lmv->tgts[id_group(&data2->id1)].ltd_exp == NULL)
1683                         continue;
1684
1685                 rc = md_unlink(lmv->tgts[id_group(&data2->id1)].ltd_exp,
1686                                data2, req);
1687
1688                 CDEBUG(D_OTHER, "unlink slave "DLID4" -> %d\n",
1689                        OLID4(&mea->mea_ids[i]), rc);
1690
1691                 if (*req) {
1692                         ptlrpc_req_finished(*req);
1693                         *req = NULL;
1694                 }
1695                 if (rc)
1696                         RETURN(rc);
1697         }
1698         OBD_FREE(data2, sizeof(*data2));
1699         RETURN(rc);
1700 }
1701
1702 static int lmv_delete_inode(struct obd_export *exp, struct lustre_id *id)
1703 {
1704         ENTRY;
1705
1706         LASSERT(exp && id);
1707         if (lmv_delete_obj(exp, id)) {
1708                 CDEBUG(D_OTHER, "lmv object "DLID4" is destroyed.\n",
1709                        OLID4(id));
1710         }
1711         RETURN(0);
1712 }
1713
1714 static int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1715                       struct ptlrpc_request **request)
1716 {
1717         struct obd_device *obd = exp->exp_obd;
1718         struct lmv_obd *lmv = &obd->u.lmv;
1719         int rc, i = 0;
1720         ENTRY;
1721
1722         rc = lmv_check_connect(obd);
1723         if (rc)
1724                 RETURN(rc);
1725
1726         if (data->namelen == 0 && data->mea1 != NULL) {
1727                 /* mds asks to remove slave objects */
1728                 rc = lmv_unlink_slaves(exp, data, request);
1729                 RETURN(rc);
1730         }
1731
1732         if (data->namelen != 0) {
1733                 struct lmv_obj *obj;
1734
1735                 obj = lmv_grab_obj(obd, &data->id1);
1736                 if (obj) {
1737                         i = raw_name2idx(obj->hashtype, obj->objcount,
1738                                          data->name, data->namelen);
1739                         data->id1 = obj->objs[i].id;
1740                         lmv_put_obj(obj);
1741                 }
1742                 CDEBUG(D_OTHER, "unlink '%*s' in "DLID4" -> %u\n",
1743                        data->namelen, data->name, OLID4(&data->id1),
1744                        i);
1745         } else {
1746                 CDEBUG(D_OTHER, "drop i_nlink on "DLID4"\n",
1747                        OLID4(&data->id1));
1748         }
1749         rc = md_unlink(lmv->tgts[id_group(&data->id1)].ltd_exp,
1750                        data, request);
1751         RETURN(rc);
1752 }
1753
1754 static struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1755                                            struct lustre_id *id)
1756 {
1757         struct obd_device *obd = exp->exp_obd;
1758         struct lmv_obd *lmv = &obd->u.lmv;
1759         int rc;
1760         ENTRY;
1761
1762         rc = lmv_check_connect(obd);
1763         if (rc)
1764                 RETURN(ERR_PTR(rc));
1765         obd = lmv->tgts[id_group(id)].ltd_exp->exp_obd;
1766         EXIT;
1767
1768         return obd;
1769 }
1770
1771 static int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1772                                  void *acl, int acl_size,
1773                                  struct lov_stripe_md **ea,
1774                                  struct obd_trans_info *oti)
1775 {
1776         struct obd_device *obd = exp->exp_obd;
1777         struct lmv_obd *lmv = &obd->u.lmv;
1778         struct lov_stripe_md obj_md;
1779         struct lov_stripe_md *obj_mdp = &obj_md;
1780         int rc = 0;
1781         ENTRY;
1782
1783         LASSERT(ea == NULL);
1784         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1785
1786         rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa,
1787                         acl, acl_size, &obj_mdp, oti);
1788
1789         RETURN(rc);
1790 }
1791
1792 /*
1793  * to be called from MDS only. @oa should have correct store cookie and o_fid
1794  * values for "master" object, as it will be used.
1795  */
1796 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
1797                    void *acl, int acl_size,
1798                    struct lov_stripe_md **ea, struct obd_trans_info *oti)
1799 {
1800         struct obd_device *obd = exp->exp_obd;
1801         struct lmv_obd *lmv = &obd->u.lmv;
1802         struct lustre_id mid;
1803         int i, c, rc = 0;
1804         struct mea *mea;
1805         ENTRY;
1806
1807         rc = lmv_check_connect(obd);
1808         if (rc)
1809                 RETURN(rc);
1810
1811         LASSERT(oa != NULL);
1812
1813         if (ea == NULL) {
1814                 rc = lmv_obd_create_single(exp, oa, acl, acl_size, NULL, oti);
1815                 if (rc)
1816                         CERROR("Can't create object, rc = %d\n", rc);
1817                 RETURN(rc);
1818         }
1819
1820         /* acl is only suppied when mds create single remote obj */
1821         LASSERT(acl == NULL && acl_size == 0);
1822
1823         if (*ea == NULL) {
1824                 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1825                 if (rc < 0) {
1826                         CERROR("obd_alloc_diskmd() failed, error %d\n",
1827                                rc);
1828                         RETURN(rc);
1829                 } else
1830                         rc = 0;
1831
1832                 if (*ea == NULL)
1833                         RETURN(-ENOMEM);
1834         }
1835
1836         /*
1837          * here we should take care about splitted dir, so store cookie and fid
1838          * for "master" object should already be allocated and passed in @oa.
1839          */
1840         LASSERT(oa->o_id != 0);
1841         LASSERT(oa->o_fid != 0);
1842
1843         /* save "master" object id */
1844         obdo2id(&mid, oa);
1845
1846         mea = (struct mea *)*ea;
1847         mea->mea_master = -1;
1848         mea->mea_magic = MEA_MAGIC_ALL_CHARS;
1849
1850         if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1851                 mea->mea_count = lmv->desc.ld_tgt_count;
1852
1853         for (i = 0, c = 0; c < mea->mea_count && i < lmv->desc.ld_tgt_count; i++) {
1854                 struct lov_stripe_md obj_md;
1855                 struct lov_stripe_md *obj_mdp = &obj_md;
1856
1857                 if (lmv->tgts[i].ltd_exp == NULL) {
1858                         /* this is "master" MDS */
1859                         mea->mea_master = i;
1860                         mea->mea_ids[c] = mid;
1861                         c++;
1862                         continue;
1863                 }
1864
1865                 /*
1866                  * "master" MDS should always be part of stripped dir,
1867                  * so scan for it.
1868                  */
1869                 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1870                         continue;
1871
1872                 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE |
1873                         OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1874
1875                 rc = obd_create(lmv->tgts[c].ltd_exp, oa, NULL, 0,
1876                                 &obj_mdp, oti);
1877                 if (rc) {
1878                         CERROR("obd_create() failed on MDT target %d, "
1879                                "error %d\n", c, rc);
1880                         RETURN(rc);
1881                 }
1882
1883                 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1884                        i, oa->o_id, oa->o_generation);
1885
1886
1887                 /*
1888                  * here, when object is created (or it is master and was passed
1889                  * from caller) on desired MDS we save its fid to local mea_ids.
1890                  */
1891                 LASSERT(oa->o_fid);
1892
1893                 /*
1894                  * store cookie should be defined here for both cases (master
1895                  * object and not master), because master is already created.
1896                  */
1897                 LASSERT(oa->o_id);
1898
1899                 /* fill mea by store cookie and fid */
1900                 obdo2id(&mea->mea_ids[c], oa);
1901                 c++;
1902         }
1903         LASSERT(c == mea->mea_count);
1904
1905         CDEBUG(D_OTHER, "%d dirobjects created\n",
1906                (int)mea->mea_count);
1907
1908         RETURN(rc);
1909 }
1910
1911 static int lmv_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1912                          struct obd_device *tgt, int count,
1913                          struct llog_catid *logid)
1914 {
1915         struct llog_ctxt *ctxt;
1916         int rc;
1917         ENTRY;
1918
1919         rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1920                             &llog_client_ops);
1921         if (rc == 0) {
1922                 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1923                 ctxt->loc_imp = tgt->u.cli.cl_import;
1924         }
1925
1926         RETURN(rc);
1927 }
1928
1929 static int lmv_llog_finish(struct obd_device *obd,
1930                            struct obd_llogs *llogs, int count)
1931 {
1932         int rc;
1933         ENTRY;
1934
1935         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1936         RETURN(rc);
1937 }
1938
1939 static int lmv_precleanup(struct obd_device *obd, int flags)
1940 {
1941         int rc = 0;
1942
1943         rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1944         if (rc != 0)
1945                 CERROR("failed to cleanup llogging subsystems\n");
1946
1947         RETURN(rc);
1948 }
1949
1950 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1951                         void *key, __u32 *vallen, void *val)
1952 {
1953         struct obd_device *obd;
1954         struct lmv_obd *lmv;
1955         int rc = 0;
1956         ENTRY;
1957
1958         obd = class_exp2obd(exp);
1959         if (obd == NULL) {
1960                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1961                        exp->exp_handle.h_cookie);
1962                 RETURN(-EINVAL);
1963         }
1964
1965         lmv = &obd->u.lmv;
1966         if (keylen == strlen("mdsize") && !strcmp(key, "mdsize")) {
1967                 __u32 *mdsize = val;
1968                 *vallen = sizeof(__u32);
1969                 *mdsize = sizeof(struct lustre_id) * lmv->desc.ld_tgt_count
1970                        + sizeof(struct mea);
1971                 RETURN(0);
1972         } else if (keylen == strlen("mdsnum") && !strcmp(key, "mdsnum")) {
1973                 struct obd_uuid *cluuid = &lmv->cluuid;
1974                 struct lmv_tgt_desc *tgts;
1975                 __u32 *mdsnum = val;
1976                 int i;
1977
1978                 tgts = lmv->tgts;
1979                 for (i = 0; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1980                         if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1981                                 *vallen = sizeof(__u32);
1982                                 *mdsnum = i;
1983                                 RETURN(0);
1984                         }
1985                 }
1986                 LASSERT(0);
1987         } else if (keylen == strlen("rootid") && !strcmp(key, "rootid")) {
1988                 rc = lmv_check_connect(obd);
1989                 if (rc)
1990                         RETURN(rc);
1991
1992                 /* getting rootid from first MDS. */
1993                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
1994                                   vallen, val);
1995                 RETURN(rc);
1996         } else if (keylen >= strlen("lmvdesc") && !strcmp(key, "lmvdesc")) {
1997                 struct lmv_desc *desc_ret = val;
1998                 *desc_ret = lmv->desc;
1999                 RETURN(0);
2000         } else if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2001                 struct lmv_tgt_desc *tgts;
2002                 int i;
2003
2004                 rc = lmv_check_connect(obd);
2005                 if (rc)
2006                         RETURN(rc);
2007
2008                 LASSERT(*vallen == sizeof(__u32));
2009                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
2010                      i++, tgts++) {
2011
2012                         /* all tgts should be connected when this get called. */
2013                         if (!tgts || !tgts->ltd_exp) {
2014                                 CERROR("target not setup?\n");
2015                                 continue;
2016                         }
2017
2018                         if (!obd_get_info(tgts->ltd_exp, keylen, key,
2019                                           vallen, val))
2020                                 RETURN(0);
2021                 }
2022                 RETURN(-EINVAL);
2023         } else if (keylen >= strlen("lovdesc") && !strcmp(key, "lovdesc")) {
2024                 rc = lmv_check_connect(obd);
2025                 if (rc)
2026                         RETURN(rc);
2027
2028                 /* forwarding this request to first MDS, it should know LOV
2029                  * desc. */
2030                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
2031                                   vallen, val);
2032                 RETURN(rc);
2033         } else if (keylen >= strlen("getext") && !strcmp(key, "getext")) {
2034                 struct lmv_tgt_desc *tgts;
2035                 int i;
2036
2037                 rc = lmv_check_connect(obd);
2038                 if (rc)
2039                         RETURN(rc);
2040
2041                 LASSERT(*vallen == sizeof(struct fid_extent));
2042                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
2043                      i++, tgts++) {
2044
2045                         /* all tgts should be connected when this get called. */
2046                         if (!tgts || !tgts->ltd_exp) {
2047                                 CERROR("target not setup?\n");
2048                                 continue;
2049                         }
2050
2051                         rc = obd_get_info(tgts->ltd_exp, keylen, key,
2052                                           vallen, val);
2053                         if (rc)
2054                                 RETURN(rc);
2055                 }
2056                 RETURN(0);
2057         }
2058
2059         CDEBUG(D_IOCTL, "invalid key\n");
2060         RETURN(-EINVAL);
2061 }
2062
2063 int lmv_set_info(struct obd_export *exp, obd_count keylen,
2064                  void *key, obd_count vallen, void *val)
2065 {
2066         struct lmv_tgt_desc    *tgt;
2067         struct obd_device      *obd;
2068         struct lmv_obd         *lmv;
2069         int rc = 0;
2070         ENTRY;
2071
2072         obd = class_exp2obd(exp);
2073         if (obd == NULL) {
2074                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
2075                        exp->exp_handle.h_cookie);
2076                 RETURN(-EINVAL);
2077         }
2078         lmv = &obd->u.lmv;
2079
2080         if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
2081                 lmv->server_timeout = 1;
2082                 lmv_set_timeouts(obd);
2083                 RETURN(0);
2084         }
2085
2086         /* maybe this could be default */
2087         if ((keylen == strlen("sec") && strcmp(key, "sec") == 0) ||
2088             (keylen == strlen("sec_flags") && strcmp(key, "sec_flags") == 0) ||
2089             (keylen == strlen("nllu") && strcmp(key, "nllu") == 0)) {
2090                 struct obd_export *exp;
2091                 int err, i;
2092
2093                 spin_lock(&lmv->lmv_lock);
2094                 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count;
2095                      i++, tgt++) {
2096                         exp = tgt->ltd_exp;
2097                         /* during setup time the connections to mdc might
2098                          * haven't been established.
2099                          */
2100                         if (exp == NULL) {
2101                                 struct obd_device *tgt_obd;
2102
2103                                 tgt_obd = class_find_client_obd(&tgt->uuid,
2104                                                                 OBD_MDC_DEVICENAME,
2105                                                                 &obd->obd_uuid);
2106                                 if (!tgt_obd) {
2107                                         CERROR("can't set info %s, "
2108                                                "device %s not attached?\n",
2109                                                 (char *) key, tgt->uuid.uuid);
2110                                         rc = -EINVAL;
2111                                         continue;
2112                                 }
2113                                 exp = tgt_obd->obd_self_export;
2114                         }
2115
2116                         err = obd_set_info(exp, keylen, key, vallen, val);
2117                         if (!rc)
2118                                 rc = err;
2119                 }
2120                 spin_unlock(&lmv->lmv_lock);
2121
2122                 RETURN(rc);
2123         }
2124         if (keylen == 5 && strcmp(key, "audit") == 0) {
2125                 struct audit_attr_msg * msg = val;
2126                 int mds = id_group(&msg->id);
2127                 int i;
2128                 LASSERT(mds < lmv->desc.ld_tgt_count);
2129
2130                 if (IS_AUDIT_OP(msg->attr, AUDIT_FS) ||
2131                     IS_AUDIT_OP(msg->attr, AUDIT_NULL) ||
2132                     IS_AUDIT_OP(msg->attr, AUDIT_SYNC)) {
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