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 "lmv_internal.h"
  53
  54 /* not defined for liblustre building */
  55 #if !defined(ATOMIC_INIT)
  56 #define ATOMIC_INIT(val) { (val) }
  57 #endif
  58
  59 /* object cache. */
  60 kmem_cache_t *obj_cache;
  61 atomic_t obj_cache_count = ATOMIC_INIT(0);
  62
  63 static void lmv_activate_target(struct lmv_obd *lmv,
  64                                 struct lmv_tgt_desc *tgt,
  65                                 int activate)
  66 {
  67         if (tgt->active == activate)
  68                 return;
  69
  70         tgt->active = activate;
  71         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  72 }
  73
  74 /* Error codes:
  75  *
  76  *  -EINVAL  : UUID can't be found in the LMV's target list
  77  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  78  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  79  */
  80 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
  81                               int activate)
  82 {
  83         struct lmv_tgt_desc *tgt;
  84         struct obd_device *obd;
  85         int i, rc = 0;
  86         ENTRY;
  87
  88         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  89                lmv, uuid->uuid, activate);
  90
  91         spin_lock(&lmv->lmv_lock);
  92         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
  93                 if (tgt->ltd_exp == NULL)
  94                         continue;
  95
  96                 CDEBUG(D_INFO, "lmv idx %d is %s conn "LPX64"\n",
  97                        i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
  98
  99                 if (obd_uuid_equals(uuid, &tgt->uuid))
 100                         break;
 101         }
 102
 103         if (i == lmv->desc.ld_tgt_count)
 104                 GOTO(out_lmv_lock, rc = -EINVAL);
 105
 106         obd = class_exp2obd(tgt->ltd_exp);
 107         if (obd == NULL)
 108                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 109
 110         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 111                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 112                obd->obd_type->typ_name, i);
 113         LASSERT(strcmp(obd->obd_type->typ_name, OBD_MDC_DEVICENAME) == 0);
 114
 115         if (tgt->active == activate) {
 116                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 117                        activate ? "" : "in");
 118                 GOTO(out_lmv_lock, rc);
 119         }
 120
 121         CDEBUG(D_INFO, "Marking OBD %p %sactive\n",
 122                obd, activate ? "" : "in");
 123
 124         lmv_activate_target(lmv, tgt, activate);
 125
 126         EXIT;
 127
 128  out_lmv_lock:
 129         spin_unlock(&lmv->lmv_lock);
 130         return rc;
 131 }
 132
 133 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 134                       int active, void *data)
 135 {
 136         struct obd_uuid *uuid;
 137         int rc;
 138         ENTRY;
 139
 140         if (strcmp(watched->obd_type->typ_name, OBD_MDC_DEVICENAME)) {
 141                 CERROR("unexpected notification of %s %s!\n",
 142                        watched->obd_type->typ_name,
 143                        watched->obd_name);
 144                 RETURN(-EINVAL);
 145         }
 146         uuid = &watched->u.cli.cl_import->imp_target_uuid;
 147
 148         /* Set MDC as active before notifying the observer, so the observer can
 149          * use the MDC normally.
 150          */
 151         rc = lmv_set_mdc_active(&obd->u.lmv, uuid, active);
 152         if (rc) {
 153                 CERROR("%sactivation of %s failed: %d\n",
 154                        active ? "" : "de", uuid->uuid, rc);
 155                 RETURN(rc);
 156         }
 157
 158         if (obd->obd_observer)
 159                 /* Pass the notification up the chain. */
 160                 rc = obd_notify(obd->obd_observer, watched, active, data);
 161
 162         RETURN(rc);
 163 }
 164
 165 static int lmv_attach(struct obd_device *dev, obd_count len, void *data)
 166 {
 167         struct lprocfs_static_vars lvars;
 168         int rc;
 169         ENTRY;
 170
 171         lprocfs_init_vars(lmv, &lvars);
 172         rc = lprocfs_obd_attach(dev, lvars.obd_vars);
 173 #ifdef __KERNEL__
 174         if (rc == 0) {
 175                 struct proc_dir_entry *entry;
 176
 177                 entry = create_proc_entry("target_obd_status", 0444,
 178                                            dev->obd_proc_entry);
 179                 if (entry == NULL)
 180                         RETURN(-ENOMEM);
 181                 entry->proc_fops = &lmv_proc_target_fops;
 182                 entry->data = dev;
 183        }
 184 #endif
 185         RETURN (rc);
 186 }
 187
 188 static int lmv_detach(struct obd_device *dev)
 189 {
 190         return lprocfs_obd_detach(dev);
 191 }
 192
 193 /* this is fake connect function. Its purpose is to initialize lmv and say
 194  * caller that everything is okay. Real connection will be performed later. */
 195 static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd,
 196                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 197                        unsigned long flags)
 198 {
 199 #ifdef __KERNEL__
 200         struct proc_dir_entry *lmv_proc_dir;
 201 #endif
 202         struct lmv_obd *lmv = &obd->u.lmv;
 203         struct obd_export *exp;
 204         int rc = 0;
 205         ENTRY;
 206
 207         rc = class_connect(conn, obd, cluuid);
 208         if (rc) {
 209                 CERROR("class_connection() returned %d\n", rc);
 210                 RETURN(rc);
 211         }
 212
 213         exp = class_conn2export(conn);
 214
 215         /* we don't want to actually do the underlying connections more than
 216          * once, so keep track. */
 217         lmv->refcount++;
 218         if (lmv->refcount > 1) {
 219                 class_export_put(exp);
 220                 RETURN(0);
 221         }
 222
 223         lmv->exp = exp;
 224         lmv->connected = 0;
 225         lmv->cluuid = *cluuid;
 226         lmv->connect_flags = flags;
 227         sema_init(&lmv->init_sem, 1);
 228         if (data)
 229                 memcpy(&lmv->conn_data, data, sizeof(*data));
 230
 231 #ifdef __KERNEL__
 232         lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
 233                                         NULL, NULL);
 234         if (IS_ERR(lmv_proc_dir)) {
 235                 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 236                        obd->obd_type->typ_name, obd->obd_name);
 237                 lmv_proc_dir = NULL;
 238         }
 239 #endif
 240
 241         /*
 242          * all real clients shouls perform actual connection rightaway, because
 243          * it is possible, that LMV will not have opportunity to connect
 244          * targets, as MDC stuff will bit called directly, for instance while
 245          * reading ../mdc/../kbytesfree procfs file, etc.
 246          */
 247         if (flags & OBD_OPT_REAL_CLIENT)
 248                 rc = lmv_check_connect(obd);
 249
 250 #ifdef __KERNEL__
 251         if (rc) {
 252                 if (lmv_proc_dir)
 253                         lprocfs_remove(lmv_proc_dir);
 254         }
 255 #endif
 256
 257         RETURN(rc);
 258 }
 259
 260 static void lmv_set_timeouts(struct obd_device *obd)
 261 {
 262         struct lmv_tgt_desc *tgts;
 263         struct lmv_obd *lmv;
 264         int i;
 265
 266         lmv = &obd->u.lmv;
 267         if (lmv->server_timeout == 0)
 268                 return;
 269
 270         if (lmv->connected == 0)
 271                 return;
 272
 273         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
 274                 if (tgts->ltd_exp == NULL)
 275                         continue;
 276
 277                 obd_set_info(tgts->ltd_exp, strlen("inter_mds"),
 278                              "inter_mds", 0, NULL);
 279         }
 280 }
 281
 282 #define MAX_STRING_SIZE 128
 283
 284 /* performs a check if passed obd is connected. If no - connect it. */
 285 int lmv_check_connect(struct obd_device *obd)
 286 {
 287 #ifdef __KERNEL__
 288         struct proc_dir_entry *lmv_proc_dir;
 289 #endif
 290         struct lmv_obd *lmv = &obd->u.lmv;
 291         struct lmv_tgt_desc *tgts;
 292         struct obd_uuid *cluuid;
 293         struct obd_export *exp;
 294         int rc, rc2, i;
 295         ENTRY;
 296
 297         if (lmv->connected)
 298                 RETURN(0);
 299
 300         down(&lmv->init_sem);
 301         if (lmv->connected) {
 302                 up(&lmv->init_sem);
 303                 RETURN(0);
 304         }
 305
 306         cluuid = &lmv->cluuid;
 307         exp = lmv->exp;
 308
 309         CDEBUG(D_OTHER, "time to connect %s to %s\n",
 310                cluuid->uuid, obd->obd_name);
 311
 312         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
 313                 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
 314                 struct lustre_handle conn = {0, };
 315                 struct obd_device *tgt_obd;
 316
 317                 LASSERT(tgts != NULL);
 318
 319                 tgt_obd = class_find_client_obd(&tgts->uuid, OBD_MDC_DEVICENAME,
 320                                                 &obd->obd_uuid);
 321                 if (!tgt_obd) {
 322                         CERROR("target %s not attached\n", tgts->uuid.uuid);
 323                         GOTO(out_disc, rc = -EINVAL);
 324                 }
 325
 326                 /* for MDS: don't connect to yourself */
 327                 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
 328                         CDEBUG(D_OTHER, "don't connect back to %s\n",
 329                                cluuid->uuid);
 330                         tgts->ltd_exp = NULL;
 331                         continue;
 332                 }
 333
 334                 CDEBUG(D_OTHER, "connect to %s(%s) - %s, %s FOR %s\n",
 335                         tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
 336                         tgts->uuid.uuid, obd->obd_uuid.uuid,
 337                         cluuid->uuid);
 338
 339                 if (!tgt_obd->obd_set_up) {
 340                         CERROR("target %s not set up\n", tgts->uuid.uuid);
 341                         GOTO(out_disc, rc = -EINVAL);
 342                 }
 343
 344                 rc = obd_connect(&conn, tgt_obd, &lmv_mdc_uuid, &lmv->conn_data,
 345                                  lmv->connect_flags);
 346                 if (rc) {
 347                         CERROR("target %s connect error %d\n",
 348                                 tgts->uuid.uuid, rc);
 349                         GOTO(out_disc, rc);
 350                 }
 351                 tgts->ltd_exp = class_conn2export(&conn);
 352
 353                 obd_init_ea_size(tgts->ltd_exp, lmv->max_easize,
 354                                  lmv->max_cookiesize);
 355
 356                 rc = obd_register_observer(tgt_obd, obd);
 357                 if (rc) {
 358                         CERROR("target %s register_observer error %d\n",
 359                                tgts->uuid.uuid, rc);
 360                         obd_disconnect(tgts->ltd_exp, 0);
 361                         GOTO(out_disc, rc);
 362                 }
 363
 364                 lmv->desc.ld_active_tgt_count++;
 365                 tgts->active = 1;
 366
 367                 CDEBUG(D_OTHER, "connected to %s(%s) successfully (%d)\n",
 368                         tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
 369                         atomic_read(&obd->obd_refcount));
 370
 371 #ifdef __KERNEL__
 372                 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 373                 if (lmv_proc_dir) {
 374                         struct obd_device *mdc_obd = class_conn2obd(&conn);
 375                         struct proc_dir_entry *mdc_symlink;
 376                         char name[MAX_STRING_SIZE + 1];
 377
 378                         LASSERT(mdc_obd != NULL);
 379                         LASSERT(mdc_obd->obd_type != NULL);
 380                         LASSERT(mdc_obd->obd_type->typ_name != NULL);
 381                         name[MAX_STRING_SIZE] = '\0';
 382                         snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
 383                                  mdc_obd->obd_type->typ_name,
 384                                  mdc_obd->obd_name);
 385                         mdc_symlink = proc_symlink(mdc_obd->obd_name,
 386                                                    lmv_proc_dir, name);
 387                         if (mdc_symlink == NULL) {
 388                                 CERROR("could not register LMV target "
 389                                        "/proc/fs/lustre/%s/%s/target_obds/%s.",
 390                                        obd->obd_type->typ_name, obd->obd_name,
 391                                        mdc_obd->obd_name);
 392                                 lprocfs_remove(lmv_proc_dir);
 393                                 lmv_proc_dir = NULL;
 394                         }
 395                 }
 396 #endif
 397         }
 398
 399         lmv_set_timeouts(obd);
 400         class_export_put(exp);
 401         lmv->connected = 1;
 402         up(&lmv->init_sem);
 403         RETURN(0);
 404
 405  out_disc:
 406         while (i-- > 0) {
 407                 struct obd_uuid uuid;
 408                 --tgts;
 409                 --lmv->desc.ld_active_tgt_count;
 410                 tgts->active = 0;
 411                 /* save for CERROR below; (we know it's terminated) */
 412                 uuid = tgts->uuid;
 413                 rc2 = obd_disconnect(tgts->ltd_exp, 0);
 414                 if (rc2)
 415                         CERROR("error: LMV target %s disconnect on MDC idx %d: "
 416                                "error %d\n", uuid.uuid, i, rc2);
 417         }
 418         class_disconnect(exp, 0);
 419         up(&lmv->init_sem);
 420         return rc;
 421 }
 422
 423 static int lmv_disconnect(struct obd_export *exp, unsigned long flags)
 424 {
 425         struct obd_device *obd = class_exp2obd(exp);
 426         struct lmv_obd *lmv = &obd->u.lmv;
 427
 428 #ifdef __KERNEL__
 429         struct proc_dir_entry *lmv_proc_dir;
 430 #endif
 431         int rc, i;
 432         ENTRY;
 433
 434         if (!lmv->tgts)
 435                 goto out_local;
 436
 437         /* Only disconnect the underlying layers on the final disconnect. */
 438         lmv->refcount--;
 439         if (lmv->refcount != 0)
 440                 goto out_local;
 441
 442 #ifdef __KERNEL__
 443         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 444 #endif
 445
 446         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 447                 struct obd_device *mdc_obd;
 448
 449                 if (lmv->tgts[i].ltd_exp == NULL)
 450                         continue;
 451
 452                 mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
 453
 454                 if (mdc_obd)
 455                         mdc_obd->obd_no_recov = obd->obd_no_recov;
 456
 457 #ifdef __KERNEL__
 458                 if (lmv_proc_dir) {
 459                         struct proc_dir_entry *mdc_symlink;
 460
 461                         mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
 462                         if (mdc_symlink) {
 463                                 lprocfs_remove(mdc_symlink);
 464                         } else {
 465                                 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 466                                        obd->obd_type->typ_name, obd->obd_name,
 467                                        mdc_obd->obd_name);
 468                         }
 469                 }
 470 #endif
 471                 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
 472                         lmv->tgts[i].ltd_exp->exp_obd->obd_name,
 473                         lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
 474
 475                 obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL);
 476                 rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags);
 477                 if (rc) {
 478                         if (lmv->tgts[i].active) {
 479                                 CERROR("Target %s disconnect error %d\n",
 480                                        lmv->tgts[i].uuid.uuid, rc);
 481                         }
 482                         rc = 0;
 483                 }
 484
 485                 lmv_activate_target(lmv, &lmv->tgts[i], 0);
 486                 lmv->tgts[i].ltd_exp = NULL;
 487         }
 488
 489 #ifdef __KERNEL__
 490         if (lmv_proc_dir) {
 491                 lprocfs_remove(lmv_proc_dir);
 492         } else {
 493                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 494                        obd->obd_type->typ_name, obd->obd_name);
 495         }
 496 #endif
 497
 498 out_local:
 499         /* this is the case when no real connection is established by
 500          * lmv_check_connect(). */
 501         if (!lmv->connected)
 502                 class_export_put(exp);
 503         rc = class_disconnect(exp, 0);
 504         if (lmv->refcount == 0)
 505                 lmv->connected = 0;
 506         RETURN(rc);
 507 }
 508
 509 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 510                          int len, void *karg, void *uarg)
 511 {
 512         struct obd_device *obddev = class_exp2obd(exp);
 513         struct lmv_obd *lmv = &obddev->u.lmv;
 514         int i, rc = 0, set = 0;
 515         ENTRY;
 516
 517         if (lmv->desc.ld_tgt_count == 0)
 518                 RETURN(-ENOTTY);
 519
 520         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 521                 int err;
 522
 523                 if (lmv->tgts[i].ltd_exp == NULL)
 524                         continue;
 525
 526                 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
 527                 if (err) {
 528                         if (lmv->tgts[i].active) {
 529                                 CERROR("error: iocontrol MDC %s on MDT"
 530                                        "idx %d: err = %d\n",
 531                                        lmv->tgts[i].uuid.uuid, i, err);
 532                                 if (!rc)
 533                                         rc = err;
 534                         }
 535                 } else
 536                         set = 1;
 537         }
 538         if (!set && !rc)
 539                 rc = -EIO;
 540
 541         RETURN(rc);
 542 }
 543
 544 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
 545 {
 546         int i, rc = 0;
 547         struct lmv_desc *desc;
 548         struct obd_uuid *uuids;
 549         struct lmv_tgt_desc *tgts;
 550         struct obd_device *tgt_obd;
 551         struct lustre_cfg *lcfg = buf;
 552         struct lmv_obd *lmv = &obd->u.lmv;
 553         ENTRY;
 554
 555         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 556                 CERROR("LMV setup requires a descriptor\n");
 557                 RETURN(-EINVAL);
 558         }
 559
 560         if (LUSTRE_CFG_BUFLEN(lcfg, 2) < 1) {
 561                 CERROR("LMV setup requires an MDT UUID list\n");
 562                 RETURN(-EINVAL);
 563         }
 564
 565         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
 566         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 567                 CERROR("descriptor size wrong: %d > %d\n",
 568                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 569                 RETURN(-EINVAL);
 570         }
 571
 572         uuids = (struct obd_uuid *)lustre_cfg_buf(lcfg, 2);
 573         if (sizeof(*uuids) * desc->ld_tgt_count != LUSTRE_CFG_BUFLEN(lcfg, 2)) {
 574                 CERROR("UUID array size wrong: %u * %u != %u\n",
 575                        sizeof(*uuids), desc->ld_tgt_count, LUSTRE_CFG_BUFLEN(lcfg, 2));
 576                 RETURN(-EINVAL);
 577         }
 578
 579         lmv->tgts_size = sizeof(struct lmv_tgt_desc) * desc->ld_tgt_count;
 580         OBD_ALLOC(lmv->tgts, lmv->tgts_size);
 581         if (lmv->tgts == NULL) {
 582                 CERROR("Out of memory\n");
 583                 RETURN(-ENOMEM);
 584         }
 585
 586         lmv->desc = *desc;
 587         spin_lock_init(&lmv->lmv_lock);
 588
 589         for (i = 0, tgts = lmv->tgts; i < desc->ld_tgt_count; i++, tgts++)
 590                 tgts->uuid = uuids[i];
 591
 592         lmv->max_cookiesize = 0;
 593
 594         lmv->max_easize = sizeof(struct lustre_id) *
 595                 desc->ld_tgt_count + sizeof(struct mea);
 596
 597         rc = lmv_setup_mgr(obd);
 598         if (rc) {
 599                 CERROR("Can't setup LMV object manager, "
 600                        "error %d.\n", rc);
 601                 OBD_FREE(lmv->tgts, lmv->tgts_size);
 602         }
 603
 604         tgt_obd = class_find_client_obd(&lmv->tgts->uuid, OBD_MDC_DEVICENAME,
 605                                         &obd->obd_uuid);
 606         if (!tgt_obd) {
 607                 CERROR("Target %s not attached\n", lmv->tgts->uuid.uuid);
 608                 RETURN(-EINVAL);
 609         }
 610
 611         RETURN(rc);
 612 }
 613
 614 static int lmv_cleanup(struct obd_device *obd, int flags)
 615 {
 616         struct lmv_obd *lmv = &obd->u.lmv;
 617         ENTRY;
 618
 619         lmv_cleanup_mgr(obd);
 620         OBD_FREE(lmv->tgts, lmv->tgts_size);
 621
 622         RETURN(0);
 623 }
 624
 625 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 626                       unsigned long max_age)
 627 {
 628         struct lmv_obd *lmv = &obd->u.lmv;
 629         struct obd_statfs *temp;
 630         int rc = 0, i;
 631         ENTRY;
 632
 633         rc = lmv_check_connect(obd);
 634         if (rc)
 635                 RETURN(rc);
 636
 637         OBD_ALLOC(temp, sizeof(*temp));
 638         if (temp == NULL)
 639                 RETURN(-ENOMEM);
 640
 641         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 642                 if (lmv->tgts[i].ltd_exp == NULL)
 643                         continue;
 644
 645                 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp, max_age);
 646                 if (rc) {
 647                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
 648                                lmv->tgts[i].ltd_exp->exp_obd->obd_name,
 649                                rc);
 650                         GOTO(out_free_temp, rc);
 651                 }
 652                 if (i == 0) {
 653                         memcpy(osfs, temp, sizeof(*temp));
 654                 } else {
 655                         osfs->os_bavail += temp->os_bavail;
 656                         osfs->os_blocks += temp->os_blocks;
 657                         osfs->os_ffree += temp->os_ffree;
 658                         osfs->os_files += temp->os_files;
 659                 }
 660         }
 661
 662         EXIT;
 663 out_free_temp:
 664         OBD_FREE(temp, sizeof(*temp));
 665         return rc;
 666 }
 667
 668 static int lmv_getstatus(struct obd_export *exp, struct lustre_id *id)
 669 {
 670         struct obd_device *obd = exp->exp_obd;
 671         struct lmv_obd *lmv = &obd->u.lmv;
 672         int rc;
 673         ENTRY;
 674
 675         rc = lmv_check_connect(obd);
 676         if (rc)
 677                 RETURN(rc);
 678
 679         rc = md_getstatus(lmv->tgts[0].ltd_exp, id);
 680         id_group(id) = 0;
 681
 682         RETURN(rc);
 683 }
 684
 685 static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
 686                        __u64 valid, const char *ea_name, int ea_namelen,
 687                        unsigned int ea_size, struct ptlrpc_request **request)
 688 {
 689         struct obd_device *obd = exp->exp_obd;
 690         struct lmv_obd *lmv = &obd->u.lmv;
 691         int rc, i = id_group(id);
 692         struct lmv_obj *obj;
 693         ENTRY;
 694
 695         rc = lmv_check_connect(obd);
 696         if (rc)
 697                 RETURN(rc);
 698
 699         LASSERT(i < lmv->desc.ld_tgt_count);
 700
 701
 702         rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid,
 703                         ea_name, ea_namelen, ea_size, request);
 704         if (rc)
 705                 RETURN(rc);
 706
 707         obj = lmv_grab_obj(obd, id);
 708
 709         CDEBUG(D_OTHER, "GETATTR for "DLID4" %s\n",
 710                OLID4(id), obj ? "(splitted)" : "");
 711
 712         /*
 713          * if object is splitted, then we loop over all the slaves and gather
 714          * size attribute. In ideal world we would have to gather also mds field
 715          * from all slaves, as object is spread over the cluster and this is
 716          * definitely interesting information and it is not good to loss it,
 717          * but...
 718          */
 719         if (obj) {
 720                 struct mds_body *body;
 721
 722                 if (*request == NULL) {
 723                         lmv_put_obj(obj);
 724                         RETURN(rc);
 725                 }
 726
 727                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
 728                                       sizeof(*body));
 729                 LASSERT(body != NULL);
 730
 731                 lmv_lock_obj(obj);
 732
 733                 for (i = 0; i < obj->objcount; i++) {
 734
 735                         if (lmv->tgts[i].ltd_exp == NULL) {
 736                                 CWARN("%s: NULL export for %d\n",
 737                                       obd->obd_name, i);
 738                                 continue;
 739                         }
 740
 741                         /* skip master obj. */
 742                         if (id_equal_fid(&obj->id, &obj->objs[i].id))
 743                                 continue;
 744
 745                         body->size += obj->objs[i].size;
 746                 }
 747
 748                 lmv_unlock_obj(obj);
 749                 lmv_put_obj(obj);
 750         }
 751
 752         RETURN(rc);
 753 }
 754
 755 static int lmv_access_check(struct obd_export *exp,
 756                             struct lustre_id *id,
 757                             struct ptlrpc_request **request)
 758 {
 759         struct obd_device *obd = exp->exp_obd;
 760         struct lmv_obd *lmv = &obd->u.lmv;
 761         int rc, i = id_group(id);
 762         ENTRY;
 763
 764         rc = lmv_check_connect(obd);
 765         if (rc)
 766                 RETURN(rc);
 767
 768         LASSERT(i < lmv->desc.ld_tgt_count);
 769         rc = md_access_check(lmv->tgts[i].ltd_exp, id, request);
 770         RETURN(rc);
 771 }
 772
 773 static int lmv_change_cbdata(struct obd_export *exp,
 774                              struct lustre_id *id,
 775                              ldlm_iterator_t it,
 776                              void *data)
 777 {
 778         struct obd_device *obd = exp->exp_obd;
 779         struct lmv_obd *lmv = &obd->u.lmv;
 780         int rc = 0;
 781         ENTRY;
 782
 783         rc = lmv_check_connect(obd);
 784         if (rc)
 785                 RETURN(rc);
 786
 787         CDEBUG(D_OTHER, "CBDATA for "DLID4"\n", OLID4(id));
 788         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
 789
 790         rc = md_change_cbdata(lmv->tgts[id_group(id)].ltd_exp,
 791                               id, it, data);
 792
 793         RETURN(rc);
 794 }
 795
 796 static int lmv_change_cbdata_name(struct obd_export *exp,
 797                                   struct lustre_id *pid,
 798                                   char *name, int len,
 799                                   struct lustre_id *cid,
 800                                   ldlm_iterator_t it,
 801                                   void *data)
 802 {
 803         struct obd_device *obd = exp->exp_obd;
 804         struct lmv_obd *lmv = &obd->u.lmv;
 805         struct lustre_id rcid = *cid;
 806         struct lmv_obj *obj;
 807         int rc = 0, mds;
 808         ENTRY;
 809
 810         rc = lmv_check_connect(obd);
 811         if (rc)
 812                 RETURN(rc);
 813
 814         LASSERT(id_group(pid) < lmv->desc.ld_tgt_count);
 815         LASSERT(id_group(cid) < lmv->desc.ld_tgt_count);
 816
 817         CDEBUG(D_OTHER, "CBDATA for "DLID4":%*s -> "DLID4"\n",
 818                OLID4(pid), len, name, OLID4(cid));
 819
 820         /* this is default mds for directory name belongs to. */
 821         mds = id_group(pid);
 822         obj = lmv_grab_obj(obd, pid);
 823         if (obj) {
 824                 /* directory is splitted. look for right mds for this name. */
 825                 mds = raw_name2idx(obj->hashtype, obj->objcount, name, len);
 826                 rcid = obj->objs[mds].id;
 827                 mds = id_group(&rcid);
 828                 lmv_put_obj(obj);
 829         }
 830         rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, &rcid, it, data);
 831         RETURN(rc);
 832 }
 833
 834 static int lmv_valid_attrs(struct obd_export *exp, struct lustre_id *id)
 835 {
 836         struct obd_device *obd = exp->exp_obd;
 837         struct lmv_obd *lmv = &obd->u.lmv;
 838         int rc = 0;
 839         ENTRY;
 840
 841         rc = lmv_check_connect(obd);
 842         if (rc)
 843                 RETURN(rc);
 844
 845         CDEBUG(D_OTHER, "validate "DLID4"\n", OLID4(id));
 846         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
 847         rc = md_valid_attrs(lmv->tgts[id_group(id)].ltd_exp, id);
 848         RETURN(rc);
 849 }
 850
 851 static int lmv_close(struct obd_export *exp, struct obdo *obdo,
 852                      struct obd_client_handle *och,
 853                      struct ptlrpc_request **request)
 854 {
 855         struct obd_device *obd = exp->exp_obd;
 856         struct lmv_obd *lmv = &obd->u.lmv;
 857         int rc, i = obdo->o_mds;
 858         ENTRY;
 859
 860         rc = lmv_check_connect(obd);
 861         if (rc)
 862                 RETURN(rc);
 863
 864         LASSERT(i < lmv->desc.ld_tgt_count);
 865         CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long)obdo->o_mds,
 866                (unsigned long)obdo->o_id, (unsigned long)obdo->o_generation);
 867         rc = md_close(lmv->tgts[i].ltd_exp, obdo, och, request);
 868         RETURN(rc);
 869 }
 870
 871 int lmv_get_mea_and_update_object(struct obd_export *exp,
 872                                   struct lustre_id *id)
 873 {
 874         struct obd_device *obd = exp->exp_obd;
 875         struct lmv_obd *lmv = &obd->u.lmv;
 876         struct ptlrpc_request *req = NULL;
 877         struct lmv_obj *obj;
 878         struct lustre_md md;
 879         int mealen, rc;
 880         __u64 valid;
 881         ENTRY;
 882
 883         md.mea = NULL;
 884         mealen = MEA_SIZE_LMV(lmv);
 885
 886         valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA;
 887
 888         /* time to update mea of parent id */
 889         rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp,
 890                         id, valid, NULL, 0, mealen, &req);
 891         if (rc) {
 892                 CERROR("md_getattr() failed, error %d\n", rc);
 893                 GOTO(cleanup, rc);
 894         }
 895
 896         rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
 897         if (rc) {
 898                 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
 899                 GOTO(cleanup, rc);
 900         }
 901
 902         if (md.mea == NULL)
 903                 GOTO(cleanup, rc = -ENODATA);
 904
 905         obj = lmv_create_obj(exp, id, md.mea);
 906         if (IS_ERR(obj))
 907                 rc = PTR_ERR(obj);
 908         else
 909                 lmv_put_obj(obj);
 910
 911         obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
 912
 913         EXIT;
 914 cleanup:
 915         if (req)
 916                 ptlrpc_req_finished(req);
 917         return rc;
 918 }
 919
 920 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
 921                const void *data, int datalen, int mode, __u32 uid,
 922                __u32 gid, __u64 rdev, struct ptlrpc_request **request)
 923 {
 924         struct obd_device *obd = exp->exp_obd;
 925         struct lmv_obd *lmv = &obd->u.lmv;
 926         struct mds_body *body;
 927         struct lmv_obj *obj;
 928         int rc, mds, loop = 0;
 929         ENTRY;
 930
 931         rc = lmv_check_connect(obd);
 932         if (rc)
 933                 RETURN(rc);
 934
 935         if (!lmv->desc.ld_active_tgt_count)
 936                 RETURN(-EIO);
 937 repeat:
 938         LASSERT(++loop <= 2);
 939         obj = lmv_grab_obj(obd, &op_data->id1);
 940         if (obj) {
 941                 mds = raw_name2idx(obj->hashtype, obj->objcount,
 942                                    op_data->name, op_data->namelen);
 943                 op_data->id1 = obj->objs[mds].id;
 944                 lmv_put_obj(obj);
 945         }
 946
 947         CDEBUG(D_OTHER, "CREATE '%*s' on "DLID4"\n", op_data->namelen,
 948                op_data->name, OLID4(&op_data->id1));
 949
 950         rc = md_create(lmv->tgts[id_group(&op_data->id1)].ltd_exp,
 951                        op_data, data, datalen, mode, uid, gid, rdev,
 952                        request);
 953         if (rc == 0) {
 954                 if (*request == NULL)
 955                         RETURN(rc);
 956
 957                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
 958                                       sizeof(*body));
 959                 if (body == NULL)
 960                         RETURN(-ENOMEM);
 961
 962                 CDEBUG(D_OTHER, "created. "DLID4"\n", OLID4(&op_data->id1));
 963         } else if (rc == -ERESTART) {
 964                 /*
 965                  * directory got splitted. time to update local object and
 966                  * repeat the request with proper MDS.
 967                  */
 968                 rc = lmv_get_mea_and_update_object(exp, &op_data->id1);
 969                 if (rc == 0) {
 970                         ptlrpc_req_finished(*request);
 971                         goto repeat;
 972                 }
 973         }
 974         RETURN(rc);
 975 }
 976
 977 static int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
 978 {
 979         struct obd_device *obd = exp->exp_obd;
 980         struct lmv_obd *lmv = &obd->u.lmv;
 981         int rc;
 982         ENTRY;
 983
 984         rc = lmv_check_connect(obd);
 985         if (rc)
 986                 RETURN(rc);
 987
 988         /* FIXME: choose right MDC here */
 989         CWARN("this method isn't implemented yet\n");
 990         rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
 991         RETURN(rc);
 992 }
 993
 994 static int
 995 lmv_enqueue_slaves(struct obd_export *exp, int locktype,
 996                    struct lookup_intent *it, int lockmode,
 997                    struct mdc_op_data *data, struct lustre_handle *lockh,
 998                    void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
 999                    ldlm_blocking_callback cb_blocking, void *cb_data)
1000 {
1001         struct obd_device *obd = exp->exp_obd;
1002         struct lmv_obd *lmv = &obd->u.lmv;
1003         struct mea *mea = data->mea1;
1004         struct mdc_op_data *data2;
1005         int i, rc, mds;
1006         ENTRY;
1007
1008         OBD_ALLOC(data2, sizeof(*data2));
1009         if (data2 == NULL)
1010                 RETURN(-ENOMEM);
1011
1012         LASSERT(mea != NULL);
1013         for (i = 0; i < mea->mea_count; i++) {
1014                 memset(data2, 0, sizeof(*data2));
1015                 data2->id1 = mea->mea_ids[i];
1016                 mds = id_group(&data2->id1);
1017
1018                 if (lmv->tgts[mds].ltd_exp == NULL)
1019                         continue;
1020
1021                 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it,
1022                                 lockmode, data2, lockh + i, lmm, lmmsize,
1023                                 cb_compl, cb_blocking, cb_data);
1024
1025                 CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n",
1026                        OLID4(&mea->mea_ids[i]), rc, LUSTRE_IT(it)->it_status);
1027                 if (rc)
1028                         GOTO(cleanup, rc);
1029                 if (LUSTRE_IT(it)->it_data) {
1030                         struct ptlrpc_request *req;
1031                         req = (struct ptlrpc_request *) LUSTRE_IT(it)->it_data;
1032                         ptlrpc_req_finished(req);
1033                 }
1034
1035                 if (LUSTRE_IT(it)->it_status)
1036                         GOTO(cleanup, rc = LUSTRE_IT(it)->it_status);
1037         }
1038
1039         OBD_FREE(data2, sizeof(*data2));
1040         RETURN(0);
1041 cleanup:
1042         OBD_FREE(data2, sizeof(*data2));
1043
1044         /* drop all taken locks */
1045         while (--i >= 0) {
1046                 if (lockh[i].cookie)
1047                         ldlm_lock_decref(lockh + i, lockmode);
1048                 lockh[i].cookie = 0;
1049         }
1050         return rc;
1051 }
1052
1053 static int
1054 lmv_enqueue_remote(struct obd_export *exp, int lock_type,
1055                    struct lookup_intent *it, int lock_mode,
1056                    struct mdc_op_data *data, struct lustre_handle *lockh,
1057                    void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1058                    ldlm_blocking_callback cb_blocking, void *cb_data)
1059 {
1060         struct ptlrpc_request *req = LUSTRE_IT(it)->it_data;
1061         struct obd_device *obd = exp->exp_obd;
1062         struct lmv_obd *lmv = &obd->u.lmv;
1063         struct lustre_handle plock;
1064         struct mdc_op_data rdata;
1065         struct mds_body *body = NULL;
1066         int rc = 0, pmode;
1067         ENTRY;
1068
1069         body = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body));
1070         LASSERT(body != NULL);
1071
1072         if (!(body->valid & OBD_MD_MDS))
1073                 RETURN(0);
1074
1075         CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4" -> "DLID4"\n",
1076                LL_IT2STR(it), OLID4(&data->id1), OLID4(&body->id1));
1077
1078         /* we got LOOKUP lock, but we really need attrs */
1079         pmode = LUSTRE_IT(it)->it_lock_mode;
1080         LASSERT(pmode != 0);
1081         memcpy(&plock, lockh, sizeof(plock));
1082         LUSTRE_IT(it)->it_lock_mode = 0;
1083         LUSTRE_IT(it)->it_data = NULL;
1084         LASSERT((body->valid & OBD_MD_FID) != 0);
1085
1086         memcpy(&rdata, data, sizeof(rdata));
1087         rdata.id1 = body->id1;
1088         rdata.name = NULL;
1089         rdata.namelen = 0;
1090
1091         LUSTRE_IT(it)->it_disposition &= ~DISP_ENQ_COMPLETE;
1092         ptlrpc_req_finished(req);
1093
1094         rc = md_enqueue(lmv->tgts[id_group(&rdata.id1)].ltd_exp,
1095                         lock_type, it, lock_mode, &rdata, lockh, lmm,
1096                         lmmsize, cb_compl, cb_blocking, cb_data);
1097         ldlm_lock_decref(&plock, pmode);
1098         RETURN(rc);
1099 }
1100
1101 static int
1102 lmv_enqueue(struct obd_export *exp, int lock_type,
1103             struct lookup_intent *it, int lock_mode,
1104             struct mdc_op_data *data, struct lustre_handle *lockh,
1105             void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1106             ldlm_blocking_callback cb_blocking, void *cb_data)
1107 {
1108         struct obd_device *obd = exp->exp_obd;
1109         struct lmv_obd *lmv = &obd->u.lmv;
1110         struct lmv_obj *obj;
1111         int rc, mds;
1112         ENTRY;
1113
1114         rc = lmv_check_connect(obd);
1115         if (rc)
1116                 RETURN(rc);
1117
1118         if (data->mea1 && it->it_op == IT_UNLINK) {
1119                 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
1120                                         data, lockh, lmm, lmmsize,
1121                                         cb_compl, cb_blocking, cb_data);
1122                 RETURN(rc);
1123         }
1124
1125         if (data->namelen) {
1126                 obj = lmv_grab_obj(obd, &data->id1);
1127                 if (obj) {
1128                         /* directory is splitted. look for right mds for this
1129                          * name */
1130                         mds = raw_name2idx(obj->hashtype, obj->objcount,
1131                                            (char *)data->name, data->namelen);
1132                         data->id1 = obj->objs[mds].id;
1133                         lmv_put_obj(obj);
1134                 }
1135         }
1136         CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4"\n", LL_IT2STR(it),
1137                OLID4(&data->id1));
1138
1139         rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp,
1140                         lock_type, it, lock_mode, data, lockh, lmm,
1141                         lmmsize, cb_compl, cb_blocking, cb_data);
1142         if (rc == 0 && it->it_op == IT_OPEN)
1143                 rc = lmv_enqueue_remote(exp, lock_type, it, lock_mode,
1144                                         data, lockh, lmm, lmmsize,
1145                                         cb_compl, cb_blocking, cb_data);
1146         RETURN(rc);
1147 }
1148
1149 static int
1150 lmv_getattr_lock(struct obd_export *exp, struct lustre_id *id,
1151                  char *filename, int namelen, __u64 valid,
1152                  unsigned int ea_size, struct ptlrpc_request **request)
1153 {
1154         int rc, mds = id_group(id), loop = 0;
1155         struct obd_device *obd = exp->exp_obd;
1156         struct lmv_obd *lmv = &obd->u.lmv;
1157         struct lustre_id rid = *id;
1158         struct mds_body *body;
1159         struct lmv_obj *obj;
1160         ENTRY;
1161
1162         rc = lmv_check_connect(obd);
1163         if (rc)
1164                 RETURN(rc);
1165 repeat:
1166         LASSERT(++loop <= 2);
1167         obj = lmv_grab_obj(obd, id);
1168         if (obj) {
1169                 /* directory is splitted. look for right mds for this name */
1170                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1171                                    filename, namelen - 1);
1172                 rid = obj->objs[mds].id;
1173                 lmv_put_obj(obj);
1174         }
1175
1176         CDEBUG(D_OTHER, "getattr_lock for %*s on "DLID4" -> "DLID4"\n",
1177                namelen, filename, OLID4(id), OLID4(&rid));
1178
1179         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1180                              &rid, filename, namelen,
1181                              valid == OBD_MD_FLID ? valid : valid | OBD_MD_FID,
1182                              ea_size, request);
1183         if (rc == 0) {
1184                 /*
1185                  * this could be cross-node reference. in this case all we have
1186                  * right now is lustre_id triple. we'd like to find other
1187                  * attributes.
1188                  */
1189                 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
1190                 LASSERT(body != NULL);
1191                 LASSERT((body->valid & OBD_MD_FID) != 0
1192                                 || body->valid == OBD_MD_FLID);
1193
1194                 if (body->valid & OBD_MD_MDS) {
1195                         struct ptlrpc_request *req = NULL;
1196
1197                         rid = body->id1;
1198                         CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid));
1199
1200                         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1201                                              &rid, NULL, 1, valid, ea_size, &req);
1202                         ptlrpc_req_finished(*request);
1203                         *request = req;
1204                 }
1205         } else if (rc == -ERESTART) {
1206                 /* directory got splitted. time to update local object and
1207                  * repeat the request with proper MDS */
1208                 rc = lmv_get_mea_and_update_object(exp, &rid);
1209                 if (rc == 0) {
1210                         ptlrpc_req_finished(*request);
1211                         goto repeat;
1212                 }
1213         }
1214         RETURN(rc);
1215 }
1216
1217 /*
1218  * llite passes id of an target inode in data->id1 and id of directory in
1219  * data->id2
1220  */
1221 static int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
1222                     struct ptlrpc_request **request)
1223 {
1224         struct obd_device *obd = exp->exp_obd;
1225         struct lmv_obd *lmv = &obd->u.lmv;
1226         struct lmv_obj *obj;
1227         int rc, mds;
1228         ENTRY;
1229
1230         rc = lmv_check_connect(obd);
1231         if (rc)
1232                 RETURN(rc);
1233
1234         if (data->namelen != 0) {
1235                 /* usual link request */
1236                 obj = lmv_grab_obj(obd, &data->id2);
1237                 if (obj) {
1238                         rc = raw_name2idx(obj->hashtype, obj->objcount,
1239                                           data->name, data->namelen);
1240                         data->id2 = obj->objs[rc].id;
1241                         lmv_put_obj(obj);
1242                 }
1243
1244                 mds = id_group(&data->id2);
1245
1246                 CDEBUG(D_OTHER,"link "DLID4":%*s to "DLID4"\n",
1247                        OLID4(&data->id2), data->namelen, data->name,
1248                        OLID4(&data->id1));
1249         } else {
1250                 mds = id_group(&data->id1);
1251
1252                 /* request from MDS to acquire i_links for inode by id1 */
1253                 CDEBUG(D_OTHER, "inc i_nlinks for "DLID4"\n",
1254                        OLID4(&data->id1));
1255         }
1256
1257         CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n",
1258                mds, OLID4(&data->id1));
1259         rc = md_link(lmv->tgts[mds].ltd_exp, data, request);
1260
1261         RETURN(rc);
1262 }
1263
1264 static int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1265                       const char *old, int oldlen, const char *new, int newlen,
1266                       struct ptlrpc_request **request)
1267 {
1268         struct obd_device *obd = exp->exp_obd;
1269         struct lmv_obd *lmv = &obd->u.lmv;
1270         struct lmv_obj *obj;
1271         int rc, mds;
1272         ENTRY;
1273
1274         CDEBUG(D_OTHER, "rename %*s in "DLID4" to %*s in "DLID4"\n",
1275                oldlen, old, OLID4(&data->id1), newlen, new,
1276                OLID4(&data->id2));
1277
1278         rc = lmv_check_connect(obd);
1279         if (rc)
1280                 RETURN(rc);
1281
1282         if (oldlen == 0) {
1283                 /*
1284                  * MDS with old dir entry is asking another MDS to create name
1285                  * there.
1286                  */
1287                 CDEBUG(D_OTHER,
1288                        "create %*s(%d/%d) in "DLID4" pointing "
1289                        "to "DLID4"\n", newlen, new, oldlen, newlen,
1290                        OLID4(&data->id2), OLID4(&data->id1));
1291
1292                 mds = id_group(&data->id2);
1293
1294                 /*
1295                  * target directory can be splitted, sowe should forward request
1296                  * to the right MDS.
1297                  */
1298                 obj = lmv_grab_obj(obd, &data->id2);
1299                 if (obj) {
1300                         mds = raw_name2idx(obj->hashtype, obj->objcount,
1301                                            (char *)new, newlen);
1302                         data->id2 = obj->objs[mds].id;
1303                         CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1304                                OLID4(&data->id2));
1305                         lmv_put_obj(obj);
1306                 }
1307                 goto request;
1308         }
1309
1310         obj = lmv_grab_obj(obd, &data->id1);
1311         if (obj) {
1312                 /*
1313                  * directory is already splitted, so we have to forward request
1314                  * to the right MDS.
1315                  */
1316                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1317                                    (char *)old, oldlen);
1318                 data->id1 = obj->objs[mds].id;
1319                 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1320                        OLID4(&data->id1));
1321                 lmv_put_obj(obj);
1322         }
1323
1324         obj = lmv_grab_obj(obd, &data->id2);
1325         if (obj) {
1326                 /*
1327                  * directory is already splitted, so we have to forward request
1328                  * to the right MDS.
1329                  */
1330                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1331                                    (char *)new, newlen);
1332
1333                 data->id2 = obj->objs[mds].id;
1334                 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1335                        OLID4(&data->id2));
1336                 lmv_put_obj(obj);
1337         }
1338
1339         mds = id_group(&data->id1);
1340
1341 request:
1342         if (id_group(&data->id1) != id_group(&data->id2)) {
1343                 CDEBUG(D_OTHER,"cross-node rename "DLID4"/%*s to "DLID4"/%*s\n",
1344                        OLID4(&data->id1), oldlen, old, OLID4(&data->id2),
1345                        newlen, new);
1346         }
1347
1348         rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1349                        new, newlen, request);
1350         RETURN(rc);
1351 }
1352
1353 static int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1354                        struct iattr *iattr, void *ea, int ealen, void *ea2,
1355                        int ea2len, struct ptlrpc_request **request)
1356 {
1357         struct obd_device *obd = exp->exp_obd;
1358         struct lmv_obd *lmv = &obd->u.lmv;
1359         struct ptlrpc_request *req;
1360         struct mds_body *body;
1361         struct lmv_obj *obj;
1362         int rc = 0, i;
1363         ENTRY;
1364
1365         rc = lmv_check_connect(obd);
1366         if (rc)
1367                 RETURN(rc);
1368
1369         obj = lmv_grab_obj(obd, &data->id1);
1370
1371         CDEBUG(D_OTHER, "SETATTR for "DLID4", valid 0x%x%s\n",
1372                OLID4(&data->id1), iattr->ia_valid, obj ? ", splitted" : "");
1373
1374         if (obj) {
1375                 for (i = 0; i < obj->objcount; i++) {
1376                         data->id1 = obj->objs[i].id;
1377
1378                         rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1379                                         data, iattr, ea, ealen, ea2, ea2len, &req);
1380
1381                         if (id_equal_fid(&obj->id, &obj->objs[i].id)) {
1382                                 /*
1383                                  * this is master object and this request should
1384                                  * be returned back to llite.
1385                                  */
1386                                 *request = req;
1387                         } else {
1388                                 ptlrpc_req_finished(req);
1389                         }
1390
1391                         if (rc)
1392                                 break;
1393                 }
1394                 lmv_put_obj(obj);
1395         } else {
1396                 LASSERT(id_group(&data->id1) < lmv->desc.ld_tgt_count);
1397                 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1398                                 data, iattr, ea, ealen, ea2, ea2len, request);
1399                 if (rc == 0) {
1400                         body = lustre_msg_buf((*request)->rq_repmsg, 0,
1401                                               sizeof(*body));
1402                         LASSERT(body != NULL);
1403                         LASSERT((body->valid & OBD_MD_FID) != 0);
1404                         LASSERT(id_group(&body->id1) == id_group(&data->id1));
1405                 }
1406         }
1407         RETURN(rc);
1408 }
1409
1410 static int lmv_sync(struct obd_export *exp, struct lustre_id *id,
1411                     struct ptlrpc_request **request)
1412 {
1413         struct obd_device *obd = exp->exp_obd;
1414         struct lmv_obd *lmv = &obd->u.lmv;
1415         int rc;
1416         ENTRY;
1417
1418         rc = lmv_check_connect(obd);
1419         if (rc)
1420                 RETURN(rc);
1421
1422         rc = md_sync(lmv->tgts[id_group(id)].ltd_exp,
1423                      id, request);
1424         RETURN(rc);
1425 }
1426
1427 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock,
1428                             struct ldlm_lock_desc *desc,
1429                             void *data, int flag)
1430 {
1431         struct lustre_handle lockh;
1432         struct lmv_obj *obj;
1433         int rc;
1434         ENTRY;
1435
1436         switch (flag) {
1437         case LDLM_CB_BLOCKING:
1438                 ldlm_lock2handle(lock, &lockh);
1439                 rc = ldlm_cli_cancel(&lockh);
1440                 if (rc < 0) {
1441                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1442                         RETURN(rc);
1443                 }
1444                 break;
1445         case LDLM_CB_CANCELING:
1446                 /* time to drop cached attrs for dirobj */
1447                 obj = lock->l_ast_data;
1448                 if (obj) {
1449                         CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
1450                                ", master "DLID4"\n",
1451                                lock->l_resource->lr_name.name[3] == 1 ?
1452                                "LOOKUP" : "UPDATE",
1453                                lock->l_resource->lr_name.name[0],
1454                                lock->l_resource->lr_name.name[1],
1455                                OLID4(&obj->id));
1456                         lmv_put_obj(obj);
1457                 }
1458                 break;
1459         default:
1460                 LBUG();
1461         }
1462         RETURN(0);
1463 }
1464
1465 static void lmv_remove_dots(struct page *page)
1466 {
1467         unsigned limit = PAGE_CACHE_SIZE;
1468         char *kaddr = page_address(page);
1469         struct ext2_dir_entry_2 *p;
1470         unsigned offs, rec_len;
1471
1472         for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1473                 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1474                 rec_len = le16_to_cpu(p->rec_len);
1475
1476                 if ((p->name_len == 1 && p->name[0] == '.') ||
1477                     (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1478                         p->inode = 0;
1479         }
1480 }
1481
1482 static int lmv_readpage(struct obd_export *exp, struct lustre_id *id,
1483                         __u64 offset, struct page *page,
1484                         struct ptlrpc_request **request)
1485 {
1486         struct obd_device *obd = exp->exp_obd;
1487         struct lmv_obd *lmv = &obd->u.lmv;
1488         struct lustre_id rid = *id;
1489         struct lmv_obj *obj;
1490         int rc, i;
1491         ENTRY;
1492
1493 #warning "we need well-desgined readdir() implementation"
1494         rc = lmv_check_connect(obd);
1495         if (rc)
1496                 RETURN(rc);
1497
1498         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
1499         CDEBUG(D_OTHER, "READPAGE at %llu from "DLID4"\n",
1500                offset, OLID4(&rid));
1501
1502         obj = lmv_grab_obj(obd, id);
1503         if (obj) {
1504                 lmv_lock_obj(obj);
1505
1506                 /* find dirobj containing page with requested offset. */
1507                 for (i = 0; i < obj->objcount; i++) {
1508                         if (offset < obj->objs[i].size)
1509                                 break;
1510                         offset -= obj->objs[i].size;
1511                 }
1512                 rid = obj->objs[i].id;
1513
1514                 lmv_unlock_obj(obj);
1515                 lmv_put_obj(obj);
1516
1517                 CDEBUG(D_OTHER, "forward to "DLID4" with offset %lu\n",
1518                        OLID4(&rid), (unsigned long)offset);
1519         }
1520         rc = md_readpage(lmv->tgts[id_group(&rid)].ltd_exp, &rid,
1521                          offset, page, request);
1522
1523         if (rc == 0 && !id_equal_fid(&rid, id))
1524                 /* this page isn't from master object. To avoid "." and ".."
1525                  * duplication in directory, we have to remove them from all
1526                  * slave objects */
1527                 lmv_remove_dots(page);
1528
1529         RETURN(rc);
1530 }
1531
1532 static int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1533                              struct ptlrpc_request **req)
1534 {
1535         struct obd_device *obd = exp->exp_obd;
1536         struct lmv_obd *lmv = &obd->u.lmv;
1537         struct mea *mea = data->mea1;
1538         struct mdc_op_data *data2;
1539         int i, rc = 0;
1540         ENTRY;
1541
1542         OBD_ALLOC(data2, sizeof(*data2));
1543         if (data2 == NULL)
1544                 RETURN(-ENOMEM);
1545
1546         LASSERT(mea != NULL);
1547         for (i = 0; i < mea->mea_count; i++) {
1548                 memset(data2, 0, sizeof(*data2));
1549                 data2->id1 = mea->mea_ids[i];
1550                 data2->create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1551
1552                 if (lmv->tgts[id_group(&data2->id1)].ltd_exp == NULL)
1553                         continue;
1554
1555                 rc = md_unlink(lmv->tgts[id_group(&data2->id1)].ltd_exp,
1556                                data2, req);
1557
1558                 CDEBUG(D_OTHER, "unlink slave "DLID4" -> %d\n",
1559                        OLID4(&mea->mea_ids[i]), rc);
1560
1561                 if (*req) {
1562                         ptlrpc_req_finished(*req);
1563                         *req = NULL;
1564                 }
1565                 if (rc)
1566                         RETURN(rc);
1567         }
1568         OBD_FREE(data2, sizeof(*data2));
1569         RETURN(rc);
1570 }
1571
1572 static int lmv_delete_inode(struct obd_export *exp, struct lustre_id *id)
1573 {
1574         ENTRY;
1575
1576         LASSERT(exp && id);
1577         if (lmv_delete_obj(exp, id)) {
1578                 CDEBUG(D_OTHER, "lmv object "DLID4" is destroyed.\n",
1579                        OLID4(id));
1580         }
1581         RETURN(0);
1582 }
1583
1584 static int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1585                       struct ptlrpc_request **request)
1586 {
1587         struct obd_device *obd = exp->exp_obd;
1588         struct lmv_obd *lmv = &obd->u.lmv;
1589         int rc, i = 0;
1590         ENTRY;
1591
1592         rc = lmv_check_connect(obd);
1593         if (rc)
1594                 RETURN(rc);
1595
1596         if (data->namelen == 0 && data->mea1 != NULL) {
1597                 /* mds asks to remove slave objects */
1598                 rc = lmv_unlink_slaves(exp, data, request);
1599                 RETURN(rc);
1600         }
1601
1602         if (data->namelen != 0) {
1603                 struct lmv_obj *obj;
1604
1605                 obj = lmv_grab_obj(obd, &data->id1);
1606                 if (obj) {
1607                         i = raw_name2idx(obj->hashtype, obj->objcount,
1608                                          data->name, data->namelen);
1609                         data->id1 = obj->objs[i].id;
1610                         lmv_put_obj(obj);
1611                 }
1612                 CDEBUG(D_OTHER, "unlink '%*s' in "DLID4" -> %u\n",
1613                        data->namelen, data->name, OLID4(&data->id1),
1614                        i);
1615         } else {
1616                 CDEBUG(D_OTHER, "drop i_nlink on "DLID4"\n",
1617                        OLID4(&data->id1));
1618         }
1619         rc = md_unlink(lmv->tgts[id_group(&data->id1)].ltd_exp,
1620                        data, request);
1621         RETURN(rc);
1622 }
1623
1624 static struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1625                                            struct lustre_id *id)
1626 {
1627         struct obd_device *obd = exp->exp_obd;
1628         struct lmv_obd *lmv = &obd->u.lmv;
1629         int rc;
1630         ENTRY;
1631
1632         rc = lmv_check_connect(obd);
1633         if (rc)
1634                 RETURN(ERR_PTR(rc));
1635         obd = lmv->tgts[id_group(id)].ltd_exp->exp_obd;
1636         EXIT;
1637
1638         return obd;
1639 }
1640
1641 static int lmv_init_ea_size(struct obd_export *exp, int easize,
1642                             int cookiesize)
1643 {
1644         struct obd_device *obd = exp->exp_obd;
1645         struct lmv_obd *lmv = &obd->u.lmv;
1646         int i, rc = 0, change = 0;
1647         ENTRY;
1648
1649         if (lmv->max_easize < easize) {
1650                 lmv->max_easize = easize;
1651                 change = 1;
1652         }
1653         if (lmv->max_cookiesize < cookiesize) {
1654                 lmv->max_cookiesize = cookiesize;
1655                 change = 1;
1656         }
1657         if (change == 0)
1658                 RETURN(0);
1659
1660         if (lmv->connected == 0)
1661                 RETURN(0);
1662
1663         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1664                 if (lmv->tgts[i].ltd_exp == NULL) {
1665                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
1666                         continue;
1667                 }
1668
1669                 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
1670                 if (rc) {
1671                         CERROR("obd_init_ea_size() failed on MDT target %d, "
1672                                "error %d.\n", i, rc);
1673                         break;
1674                 }
1675         }
1676         RETURN(rc);
1677 }
1678
1679 static int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1680                                  void *acl, int acl_size,
1681                                  struct lov_stripe_md **ea,
1682                                  struct obd_trans_info *oti)
1683 {
1684         struct obd_device *obd = exp->exp_obd;
1685         struct lmv_obd *lmv = &obd->u.lmv;
1686         struct lov_stripe_md obj_md;
1687         struct lov_stripe_md *obj_mdp = &obj_md;
1688         int rc = 0;
1689         ENTRY;
1690
1691         LASSERT(ea == NULL);
1692         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1693
1694         rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa,
1695                         acl, acl_size, &obj_mdp, oti);
1696
1697         RETURN(rc);
1698 }
1699
1700 /*
1701  * to be called from MDS only. @oa should have correct store cookie and o_fid
1702  * values for "master" object, as it will be used.
1703  */
1704 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
1705                    void *acl, int acl_size,
1706                    struct lov_stripe_md **ea, struct obd_trans_info *oti)
1707 {
1708         struct obd_device *obd = exp->exp_obd;
1709         struct lmv_obd *lmv = &obd->u.lmv;
1710         struct lustre_id mid;
1711         int i, c, rc = 0;
1712         struct mea *mea;
1713         ENTRY;
1714
1715         rc = lmv_check_connect(obd);
1716         if (rc)
1717                 RETURN(rc);
1718
1719         LASSERT(oa != NULL);
1720
1721         if (ea == NULL) {
1722                 rc = lmv_obd_create_single(exp, oa, acl, acl_size, NULL, oti);
1723                 if (rc)
1724                         CERROR("Can't create object, rc = %d\n", rc);
1725                 RETURN(rc);
1726         }
1727
1728         /* acl is only suppied when mds create single remote obj */
1729         LASSERT(acl == NULL && acl_size == 0);
1730
1731         if (*ea == NULL) {
1732                 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1733                 if (rc < 0) {
1734                         CERROR("obd_alloc_diskmd() failed, error %d\n",
1735                                rc);
1736                         RETURN(rc);
1737                 } else
1738                         rc = 0;
1739
1740                 if (*ea == NULL)
1741                         RETURN(-ENOMEM);
1742         }
1743
1744         /*
1745          * here we should take care about splitted dir, so store cookie and fid
1746          * for "master" object should already be allocated and passed in @oa.
1747          */
1748         LASSERT(oa->o_id != 0);
1749         LASSERT(oa->o_fid != 0);
1750
1751         /* save "master" object id */
1752         obdo2id(&mid, oa);
1753
1754         mea = (struct mea *)*ea;
1755         mea->mea_master = -1;
1756         mea->mea_magic = MEA_MAGIC_ALL_CHARS;
1757
1758         if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1759                 mea->mea_count = lmv->desc.ld_tgt_count;
1760
1761         for (i = 0, c = 0; c < mea->mea_count && i < lmv->desc.ld_tgt_count; i++) {
1762                 struct lov_stripe_md obj_md;
1763                 struct lov_stripe_md *obj_mdp = &obj_md;
1764
1765                 if (lmv->tgts[i].ltd_exp == NULL) {
1766                         /* this is "master" MDS */
1767                         mea->mea_master = i;
1768                         mea->mea_ids[c] = mid;
1769                         c++;
1770                         continue;
1771                 }
1772
1773                 /*
1774                  * "master" MDS should always be part of stripped dir,
1775                  * so scan for it.
1776                  */
1777                 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1778                         continue;
1779
1780                 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE |
1781                         OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1782
1783                 rc = obd_create(lmv->tgts[c].ltd_exp, oa, NULL, 0,
1784                                 &obj_mdp, oti);
1785                 if (rc) {
1786                         CERROR("obd_create() failed on MDT target %d, "
1787                                "error %d\n", c, rc);
1788                         RETURN(rc);
1789                 }
1790
1791                 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1792                        i, oa->o_id, oa->o_generation);
1793
1794
1795                 /*
1796                  * here, when object is created (or it is master and was passed
1797                  * from caller) on desired MDS we save its fid to local mea_ids.
1798                  */
1799                 LASSERT(oa->o_fid);
1800
1801                 /*
1802                  * store cookie should be defined here for both cases (master
1803                  * object and not master), because master is already created.
1804                  */
1805                 LASSERT(oa->o_id);
1806
1807                 /* fill mea by store cookie and fid */
1808                 obdo2id(&mea->mea_ids[c], oa);
1809                 c++;
1810         }
1811         LASSERT(c == mea->mea_count);
1812
1813         CDEBUG(D_OTHER, "%d dirobjects created\n",
1814                (int)mea->mea_count);
1815
1816         RETURN(rc);
1817 }
1818
1819 static int lmv_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1820                          struct obd_device *tgt, int count,
1821                          struct llog_catid *logid)
1822 {
1823         struct llog_ctxt *ctxt;
1824         int rc;
1825         ENTRY;
1826
1827         rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1828                             &llog_client_ops);
1829         if (rc == 0) {
1830                 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1831                 ctxt->loc_imp = tgt->u.cli.cl_import;
1832         }
1833
1834         RETURN(rc);
1835 }
1836
1837 static int lmv_llog_finish(struct obd_device *obd,
1838                            struct obd_llogs *llogs, int count)
1839 {
1840         int rc;
1841         ENTRY;
1842
1843         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1844         RETURN(rc);
1845 }
1846
1847 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1848                         void *key, __u32 *vallen, void *val)
1849 {
1850         struct obd_device *obd;
1851         struct lmv_obd *lmv;
1852         int rc = 0;
1853         ENTRY;
1854
1855         obd = class_exp2obd(exp);
1856         if (obd == NULL) {
1857                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1858                        exp->exp_handle.h_cookie);
1859                 RETURN(-EINVAL);
1860         }
1861
1862         lmv = &obd->u.lmv;
1863         if (keylen == 6 && memcmp(key, "mdsize", 6) == 0) {
1864                 __u32 *mdsize = val;
1865                 *vallen = sizeof(__u32);
1866                 *mdsize = sizeof(struct lustre_id) * lmv->desc.ld_tgt_count
1867                         + sizeof(struct mea);
1868                 RETURN(0);
1869         } else if (keylen == 6 && memcmp(key, "mdsnum", 6) == 0) {
1870                 struct obd_uuid *cluuid = &lmv->cluuid;
1871                 struct lmv_tgt_desc *tgts;
1872                 __u32 *mdsnum = val;
1873                 int i;
1874
1875                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1876                         if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1877                                 *vallen = sizeof(__u32);
1878                                 *mdsnum = i;
1879                                 RETURN(0);
1880                         }
1881                 }
1882                 LASSERT(0);
1883         } else if (keylen == 6 && memcmp(key, "rootid", 6) == 0) {
1884                 rc = lmv_check_connect(obd);
1885                 if (rc)
1886                         RETURN(rc);
1887
1888                 /* getting rootid from first MDS. */
1889                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
1890                                   vallen, val);
1891                 RETURN(rc);
1892         } else if (keylen >= strlen("lmvdesc") && strcmp(key, "lmvdesc") == 0) {
1893                 struct lmv_desc *desc_ret = val;
1894                 *desc_ret = lmv->desc;
1895                 RETURN(0);
1896         } else if (keylen == strlen("remote_flag") &&
1897                    !strcmp(key, "remote_flag")) {
1898                 struct lmv_tgt_desc *tgts;
1899                 int i;
1900
1901                 rc = lmv_check_connect(obd);
1902                 if (rc)
1903                         RETURN(rc);
1904
1905                 LASSERT(*vallen == sizeof(__u32));
1906                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
1907                      i++, tgts++) {
1908
1909                         /* all tgts should be connected when this get called. */
1910                         if (!tgts || !tgts->ltd_exp) {
1911                                 CERROR("target not setup?\n");
1912                                 continue;
1913                         }
1914
1915                         if (!obd_get_info(tgts->ltd_exp, keylen, key,
1916                                           vallen, val))
1917                                 RETURN(0);
1918                 }
1919                 RETURN(-EINVAL);
1920         } else if (keylen >= strlen("lovdesc") && strcmp(key, "lovdesc") == 0) {
1921                 rc = lmv_check_connect(obd);
1922                 if (rc)
1923                         RETURN(rc);
1924
1925                 /* forwarding this request to first MDS, it should know LOV
1926                  * desc. */
1927                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
1928                                   vallen, val);
1929                 RETURN(rc);
1930         }
1931
1932         CDEBUG(D_IOCTL, "invalid key\n");
1933         RETURN(-EINVAL);
1934 }
1935
1936 int lmv_set_info(struct obd_export *exp, obd_count keylen,
1937                  void *key, obd_count vallen, void *val)
1938 {
1939         struct lmv_tgt_desc    *tgt;
1940         struct obd_device      *obd;
1941         struct lmv_obd         *lmv;
1942         ENTRY;
1943
1944         obd = class_exp2obd(exp);
1945         if (obd == NULL) {
1946                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1947                        exp->exp_handle.h_cookie);
1948                 RETURN(-EINVAL);
1949         }
1950         lmv = &obd->u.lmv;
1951
1952         if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
1953                 lmv->server_timeout = 1;
1954                 lmv_set_timeouts(obd);
1955                 RETURN(0);
1956         }
1957
1958         /* maybe this could be default */
1959         if ((keylen == strlen("sec") && strcmp(key, "sec") == 0) ||
1960             (keylen == strlen("sec_flags") && strcmp(key, "sec_flags") == 0) ||
1961             (keylen == strlen("nllu") && strcmp(key, "nllu") == 0)) {
1962                 struct obd_export *exp;
1963                 int rc = 0, err, i;
1964
1965                 spin_lock(&lmv->lmv_lock);
1966                 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count;
1967                      i++, tgt++) {
1968                         exp = tgt->ltd_exp;
1969                         /* during setup time the connections to mdc might
1970                          * haven't been established.
1971                          */
1972                         if (exp == NULL) {
1973                                 struct obd_device *tgt_obd;
1974
1975                                 tgt_obd = class_find_client_obd(&tgt->uuid,
1976                                                                 OBD_MDC_DEVICENAME,
1977                                                                 &obd->obd_uuid);
1978                                 if (!tgt_obd) {
1979                                         CERROR("can't set info %s, "
1980                                                "device %s not attached?\n",
1981                                                 (char *) key, tgt->uuid.uuid);
1982                                         rc = -EINVAL;
1983                                         continue;
1984                                 }
1985                                 exp = tgt_obd->obd_self_export;
1986                         }
1987
1988                         err = obd_set_info(exp, keylen, key, vallen, val);
1989                         if (!rc)
1990                                 rc = err;
1991                 }
1992                 spin_unlock(&lmv->lmv_lock);
1993
1994                 RETURN(rc);
1995         }
1996
1997         if ((keylen == strlen("flush_cred") &&
1998              strcmp(key, "flush_cred") == 0)) {
1999                 int rc = 0, i;
2000
2001                 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count;
2002                      i++, tgt++) {
2003                         if (!tgt->ltd_exp)
2004                                 continue;
2005                         rc = obd_set_info(tgt->ltd_exp,
2006                                           keylen, key, vallen, val);
2007                         if (rc)
2008                                 RETURN(rc);
2009                 }
2010
2011                 RETURN(0);
2012         }
2013
2014         RETURN(-EINVAL);
2015 }
2016
2017 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2018                struct lov_stripe_md *lsm)
2019 {
2020         struct obd_device *obd = class_exp2obd(exp);
2021         struct lmv_obd *lmv = &obd->u.lmv;
2022         struct mea *meap, *lsmp;
2023         int mea_size, i;
2024         ENTRY;
2025
2026         mea_size = (sizeof(struct lustre_id) *
2027                     lmv->desc.ld_tgt_count) + sizeof(struct mea);
2028         if (!lmmp)
2029                 RETURN(mea_size);
2030
2031         if (*lmmp && !lsm) {
2032                 OBD_FREE(*lmmp, mea_size);
2033                 *lmmp = NULL;
2034                 RETURN(0);
2035         }
2036
2037         if (*lmmp == NULL) {
2038                 OBD_ALLOC(*lmmp, mea_size);
2039                 if (*lmmp == NULL)
2040                         RETURN(-ENOMEM);
2041         }
2042
2043         if (!lsm)
2044                 RETURN(mea_size);
2045
2046         lsmp = (struct mea *)lsm;
2047         meap = (struct mea *)*lmmp;
2048
2049         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2050             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2051                 RETURN(-EINVAL);
2052
2053         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2054         meap->mea_count = cpu_to_le32(lsmp->mea_count);
2055         meap->mea_master = cpu_to_le32(lsmp->mea_master);
2056
2057         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2058                 meap->mea_ids[i] = meap->mea_ids[i];
2059                 id_cpu_to_le(&meap->mea_ids[i]);
2060         }
2061
2062         RETURN(mea_size);
2063 }
2064
2065 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2066                  struct lov_mds_md *lmm, int lmm_size)
2067 {
2068         struct obd_device *obd = class_exp2obd(exp);
2069         struct mea **tmea = (struct mea **)lsmp;
2070         struct mea *mea = (struct mea *)lmm;
2071         struct lmv_obd *lmv = &obd->u.lmv;
2072         int mea_size, i, rc = 0;
2073         __u32 magic;
2074         ENTRY;
2075
2076         mea_size = sizeof(struct lustre_id) *
2077                 lmv->desc.ld_tgt_count + sizeof(struct mea);
2078
2079         if (lsmp == NULL)
2080                 return mea_size;
2081
2082         if (*lsmp != NULL && lmm == NULL) {
2083                 OBD_FREE(*tmea, mea_size);
2084                 RETURN(0);
2085         }
2086
2087         LASSERT(mea_size == lmm_size);
2088
2089         OBD_ALLOC(*tmea, mea_size);
2090         if (*tmea == NULL)
2091                 RETURN(-ENOMEM);
2092
2093         if (!lmm)
2094                 RETURN(mea_size);
2095
2096         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2097             mea->mea_magic == MEA_MAGIC_ALL_CHARS)
2098         {
2099                 magic = le32_to_cpu(mea->mea_magic);
2100         } else {
2101                 struct mea_old *old = (struct mea_old *)lmm;
2102
2103                 mea_size = sizeof(struct lustre_id) * old->mea_count +
2104                         sizeof(struct mea_old);
2105
2106                 if (old->mea_count > 256 || old->mea_master > 256 ||
2107                     lmm_size < mea_size || old->mea_master > old->mea_count) {
2108                         CWARN("bad MEA: count %u, master %u, size %u\n",
2109                               old->mea_count, old->mea_master, mea_size);
2110                         GOTO(out_free_mea, rc = -EINVAL);
2111                 }
2112                 magic = MEA_MAGIC_LAST_CHAR;
2113         }
2114
2115         (*tmea)->mea_magic = magic;
2116         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2117         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2118
2119         for (i = 0; i < (*tmea)->mea_count; i++) {
2120                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2121                 id_le_to_cpu(&(*tmea)->mea_ids[i]);
2122         }
2123         RETURN(mea_size);
2124
2125 out_free_mea:
2126         OBD_FREE(*tmea, mea_size);
2127         return rc;
2128 }
2129
2130 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
2131             struct lov_stripe_md *ea, obd_count oa_bufs,
2132             struct brw_page *pgarr, struct obd_trans_info *oti)
2133 {
2134         struct obd_device *obd = exp->exp_obd;
2135         struct lmv_obd *lmv = &obd->u.lmv;
2136         struct mea *mea = (struct mea *) ea;
2137         int err;
2138
2139         LASSERT(oa != NULL);
2140         LASSERT(ea != NULL);
2141         LASSERT(pgarr != NULL);
2142         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
2143
2144         oa->o_gr = id_gen(&mea->mea_ids[oa->o_mds]);
2145         oa->o_id = id_ino(&mea->mea_ids[oa->o_mds]);
2146         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2147
2148         err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp,
2149                       oa, NULL, oa_bufs, pgarr, oti);
2150         RETURN(err);
2151 }
2152
2153 static int lmv_cancel_unused(struct obd_export *exp,
2154                              struct lov_stripe_md *lsm,
2155                              int flags, void *opaque)
2156 {
2157         struct obd_device *obd = exp->exp_obd;
2158         struct lmv_obd *lmv = &obd->u.lmv;
2159         int rc = 0, err, i;
2160         ENTRY;
2161
2162         LASSERT(lsm == NULL);
2163
2164         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2165                 if (!lmv->tgts[i].ltd_exp || !lmv->tgts[i].active)
2166                         continue;
2167
2168                 err = obd_cancel_unused(lmv->tgts[i].ltd_exp,
2169                                         NULL, flags, opaque);
2170                 if (!rc)
2171                         rc = err;
2172         }
2173         RETURN(rc);
2174 }
2175
2176 struct obd_ops lmv_obd_ops = {
2177         .o_owner                = THIS_MODULE,
2178         .o_attach               = lmv_attach,
2179         .o_detach               = lmv_detach,
2180         .o_setup                = lmv_setup,
2181         .o_cleanup              = lmv_cleanup,
2182         .o_connect              = lmv_connect,
2183         .o_disconnect           = lmv_disconnect,
2184         .o_statfs               = lmv_statfs,
2185         .o_llog_init            = lmv_llog_init,
2186         .o_llog_finish          = lmv_llog_finish,
2187         .o_get_info             = lmv_get_info,
2188         .o_set_info             = lmv_set_info,
2189         .o_create               = lmv_obd_create,
2190         .o_packmd               = lmv_packmd,
2191         .o_unpackmd             = lmv_unpackmd,
2192         .o_brw                  = lmv_brw,
2193         .o_init_ea_size         = lmv_init_ea_size,
2194         .o_notify               = lmv_notify,
2195         .o_iocontrol            = lmv_iocontrol,
2196         .o_cancel_unused        = lmv_cancel_unused,
2197 };
2198
2199 struct md_ops lmv_md_ops = {
2200         .m_getstatus           = lmv_getstatus,
2201         .m_getattr             = lmv_getattr,
2202         .m_change_cbdata       = lmv_change_cbdata,
2203         .m_change_cbdata_name  = lmv_change_cbdata_name,
2204         .m_close               = lmv_close,
2205         .m_create              = lmv_create,
2206         .m_done_writing        = lmv_done_writing,
2207         .m_enqueue             = lmv_enqueue,
2208         .m_getattr_lock        = lmv_getattr_lock,
2209         .m_intent_lock         = lmv_intent_lock,
2210         .m_link                = lmv_link,
2211         .m_rename              = lmv_rename,
2212         .m_setattr             = lmv_setattr,
2213         .m_sync                = lmv_sync,
2214         .m_readpage            = lmv_readpage,
2215         .m_unlink              = lmv_unlink,
2216         .m_get_real_obd        = lmv_get_real_obd,
2217         .m_valid_attrs         = lmv_valid_attrs,
2218         .m_delete_inode        = lmv_delete_inode,
2219         .m_access_check        = lmv_access_check,
2220 };
2221
2222 int __init lmv_init(void)
2223 {
2224         struct lprocfs_static_vars lvars;
2225         int rc;
2226
2227         obj_cache = kmem_cache_create("lmv_objects",
2228                                       sizeof(struct lmv_obj),
2229                                       0, 0, NULL, NULL);
2230         if (!obj_cache) {
2231                 CERROR("error allocating lmv objects cache\n");
2232                 return -ENOMEM;
2233         }
2234
2235         lprocfs_init_vars(lmv, &lvars);
2236         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2237                                  lvars.module_vars,
2238                                  OBD_LMV_DEVICENAME);
2239         if (rc)
2240                 kmem_cache_destroy(obj_cache);
2241
2242         return rc;
2243 }
2244
2245 #ifdef __KERNEL__
2246 static void lmv_exit(void)
2247 {
2248         class_unregister_type(OBD_LMV_DEVICENAME);
2249
2250         LASSERTF(kmem_cache_destroy(obj_cache) == 0,
2251                  "can't free lmv objects cache, %d object(s)"
2252                  "still in use\n", atomic_read(&obj_cache_count));
2253 }
2254
2255 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2256 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2257 MODULE_LICENSE("GPL");
2258
2259 module_init(lmv_init);
2260 module_exit(lmv_exit);
2261 #endif