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 *xattr_name,
 687                        const void *xattr_data, unsigned int xattr_datalen,
 688                        unsigned int ea_size, struct ptlrpc_request **request)
 689 {
 690         struct obd_device *obd = exp->exp_obd;
 691         struct lmv_obd *lmv = &obd->u.lmv;
 692         int rc, i = id_group(id);
 693         struct lmv_obj *obj;
 694         ENTRY;
 695
 696         rc = lmv_check_connect(obd);
 697         if (rc)
 698                 RETURN(rc);
 699
 700         LASSERT(i < lmv->desc.ld_tgt_count);
 701
 702
 703         rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid,
 704                         xattr_name, xattr_data, xattr_datalen,
 705                         ea_size, request);
 706         if (rc)
 707                 RETURN(rc);
 708
 709         obj = lmv_grab_obj(obd, id);
 710
 711         CDEBUG(D_OTHER, "GETATTR for "DLID4" %s\n",
 712                OLID4(id), obj ? "(splitted)" : "");
 713
 714         /*
 715          * if object is splitted, then we loop over all the slaves and gather
 716          * size attribute. In ideal world we would have to gather also mds field
 717          * from all slaves, as object is spread over the cluster and this is
 718          * definitely interesting information and it is not good to loss it,
 719          * but...
 720          */
 721         if (obj) {
 722                 struct mds_body *body;
 723
 724                 if (*request == NULL) {
 725                         lmv_put_obj(obj);
 726                         RETURN(rc);
 727                 }
 728
 729                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
 730                                       sizeof(*body));
 731                 LASSERT(body != NULL);
 732
 733                 lmv_lock_obj(obj);
 734
 735                 for (i = 0; i < obj->objcount; i++) {
 736
 737                         if (lmv->tgts[i].ltd_exp == NULL) {
 738                                 CWARN("%s: NULL export for %d\n",
 739                                       obd->obd_name, i);
 740                                 continue;
 741                         }
 742
 743                         /* skip master obj. */
 744                         if (id_equal_fid(&obj->id, &obj->objs[i].id))
 745                                 continue;
 746
 747                         body->size += obj->objs[i].size;
 748                 }
 749
 750                 lmv_unlock_obj(obj);
 751                 lmv_put_obj(obj);
 752         }
 753
 754         RETURN(rc);
 755 }
 756
 757 static int lmv_access_check(struct obd_export *exp,
 758                             struct lustre_id *id,
 759                             struct ptlrpc_request **request)
 760 {
 761         struct obd_device *obd = exp->exp_obd;
 762         struct lmv_obd *lmv = &obd->u.lmv;
 763         int rc, i = id_group(id);
 764         ENTRY;
 765
 766         rc = lmv_check_connect(obd);
 767         if (rc)
 768                 RETURN(rc);
 769
 770         LASSERT(i < lmv->desc.ld_tgt_count);
 771         rc = md_access_check(lmv->tgts[i].ltd_exp, id, request);
 772         RETURN(rc);
 773 }
 774
 775 static int lmv_change_cbdata(struct obd_export *exp,
 776                              struct lustre_id *id,
 777                              ldlm_iterator_t it,
 778                              void *data)
 779 {
 780         struct obd_device *obd = exp->exp_obd;
 781         struct lmv_obd *lmv = &obd->u.lmv;
 782         int i, rc;
 783         ENTRY;
 784
 785         rc = lmv_check_connect(obd);
 786         if (rc)
 787                 RETURN(rc);
 788
 789         CDEBUG(D_OTHER, "CBDATA for "DLID4"\n", OLID4(id));
 790         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
 791
 792         /* with CMD every object can have two locks in different
 793          * namespaces: lookup lock in space of mds storing direntry
 794          * and update/open lock in space of mds storing inode */
 795         for (i = 0; i < lmv->desc.ld_tgt_count; i++)
 796                 md_change_cbdata(lmv->tgts[i].ltd_exp, id, it, data);
 797
 798         RETURN(0);
 799 }
 800
 801 static int lmv_change_cbdata_name(struct obd_export *exp,
 802                                   struct lustre_id *pid,
 803                                   char *name, int len,
 804                                   struct lustre_id *cid,
 805                                   ldlm_iterator_t it,
 806                                   void *data)
 807 {
 808         struct obd_device *obd = exp->exp_obd;
 809         struct lmv_obd *lmv = &obd->u.lmv;
 810         struct lustre_id rcid = *cid;
 811         struct lmv_obj *obj;
 812         int rc = 0, mds;
 813         ENTRY;
 814
 815         rc = lmv_check_connect(obd);
 816         if (rc)
 817                 RETURN(rc);
 818
 819         LASSERT(id_group(pid) < lmv->desc.ld_tgt_count);
 820         LASSERT(id_group(cid) < lmv->desc.ld_tgt_count);
 821
 822         CDEBUG(D_OTHER, "CBDATA for "DLID4":%*s -> "DLID4"\n",
 823                OLID4(pid), len, name, OLID4(cid));
 824
 825         /* this is default mds for directory name belongs to. */
 826         mds = id_group(pid);
 827         obj = lmv_grab_obj(obd, pid);
 828         if (obj) {
 829                 /* directory is splitted. look for right mds for this name. */
 830                 mds = raw_name2idx(obj->hashtype, obj->objcount, name, len);
 831                 rcid = obj->objs[mds].id;
 832                 mds = id_group(&rcid);
 833                 lmv_put_obj(obj);
 834         }
 835         rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, &rcid, it, data);
 836         RETURN(rc);
 837 }
 838
 839 static int lmv_valid_attrs(struct obd_export *exp, struct lustre_id *id)
 840 {
 841         struct obd_device *obd = exp->exp_obd;
 842         struct lmv_obd *lmv = &obd->u.lmv;
 843         int rc = 0;
 844         ENTRY;
 845
 846         rc = lmv_check_connect(obd);
 847         if (rc)
 848                 RETURN(rc);
 849
 850         CDEBUG(D_OTHER, "validate "DLID4"\n", OLID4(id));
 851         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
 852         rc = md_valid_attrs(lmv->tgts[id_group(id)].ltd_exp, id);
 853         RETURN(rc);
 854 }
 855
 856 static int lmv_close(struct obd_export *exp, struct obdo *obdo,
 857                      struct obd_client_handle *och,
 858                      struct ptlrpc_request **request)
 859 {
 860         struct obd_device *obd = exp->exp_obd;
 861         struct lmv_obd *lmv = &obd->u.lmv;
 862         int rc, i = obdo->o_mds;
 863         ENTRY;
 864
 865         rc = lmv_check_connect(obd);
 866         if (rc)
 867                 RETURN(rc);
 868
 869         LASSERT(i < lmv->desc.ld_tgt_count);
 870         CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long)obdo->o_mds,
 871                (unsigned long)obdo->o_id, (unsigned long)obdo->o_generation);
 872         rc = md_close(lmv->tgts[i].ltd_exp, obdo, och, request);
 873         RETURN(rc);
 874 }
 875
 876 int lmv_get_mea_and_update_object(struct obd_export *exp,
 877                                   struct lustre_id *id)
 878 {
 879         struct obd_device *obd = exp->exp_obd;
 880         struct lmv_obd *lmv = &obd->u.lmv;
 881         struct ptlrpc_request *req = NULL;
 882         struct lmv_obj *obj;
 883         struct lustre_md md;
 884         int mealen, rc;
 885         __u64 valid;
 886         ENTRY;
 887
 888         md.mea = NULL;
 889         mealen = MEA_SIZE_LMV(lmv);
 890
 891         valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA;
 892
 893         /* time to update mea of parent id */
 894         rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp,
 895                         id, valid, NULL, NULL, 0, mealen, &req);
 896         if (rc) {
 897                 CERROR("md_getattr() failed, error %d\n", rc);
 898                 GOTO(cleanup, rc);
 899         }
 900
 901         rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
 902         if (rc) {
 903                 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
 904                 GOTO(cleanup, rc);
 905         }
 906
 907         if (md.mea == NULL)
 908                 GOTO(cleanup, rc = -ENODATA);
 909
 910         obj = lmv_create_obj(exp, id, md.mea);
 911         if (IS_ERR(obj))
 912                 rc = PTR_ERR(obj);
 913         else
 914                 lmv_put_obj(obj);
 915
 916         obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
 917
 918         EXIT;
 919 cleanup:
 920         if (req)
 921                 ptlrpc_req_finished(req);
 922         return rc;
 923 }
 924
 925 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
 926                const void *data, int datalen, int mode, __u32 uid,
 927                __u32 gid, __u64 rdev, struct ptlrpc_request **request)
 928 {
 929         struct obd_device *obd = exp->exp_obd;
 930         struct lmv_obd *lmv = &obd->u.lmv;
 931         struct mds_body *body;
 932         struct lmv_obj *obj;
 933         int rc, mds, loop = 0;
 934         ENTRY;
 935
 936         rc = lmv_check_connect(obd);
 937         if (rc)
 938                 RETURN(rc);
 939
 940         if (!lmv->desc.ld_active_tgt_count)
 941                 RETURN(-EIO);
 942 repeat:
 943         LASSERT(++loop <= 2);
 944         obj = lmv_grab_obj(obd, &op_data->id1);
 945         if (obj) {
 946                 mds = raw_name2idx(obj->hashtype, obj->objcount,
 947                                    op_data->name, op_data->namelen);
 948                 op_data->id1 = obj->objs[mds].id;
 949                 lmv_put_obj(obj);
 950         }
 951
 952         CDEBUG(D_OTHER, "CREATE '%*s' on "DLID4"\n", op_data->namelen,
 953                op_data->name, OLID4(&op_data->id1));
 954
 955         rc = md_create(lmv->tgts[id_group(&op_data->id1)].ltd_exp,
 956                        op_data, data, datalen, mode, uid, gid, rdev,
 957                        request);
 958         if (rc == 0) {
 959                 if (*request == NULL)
 960                         RETURN(rc);
 961
 962                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
 963                                       sizeof(*body));
 964                 if (body == NULL)
 965                         RETURN(-ENOMEM);
 966
 967                 CDEBUG(D_OTHER, "created. "DLID4"\n", OLID4(&op_data->id1));
 968         } else if (rc == -ERESTART) {
 969                 /*
 970                  * directory got splitted. time to update local object and
 971                  * repeat the request with proper MDS.
 972                  */
 973                 rc = lmv_get_mea_and_update_object(exp, &op_data->id1);
 974                 if (rc == 0) {
 975                         ptlrpc_req_finished(*request);
 976                         goto repeat;
 977                 }
 978         }
 979         RETURN(rc);
 980 }
 981
 982 static int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
 983 {
 984         struct obd_device *obd = exp->exp_obd;
 985         struct lmv_obd *lmv = &obd->u.lmv;
 986         int rc;
 987         ENTRY;
 988
 989         rc = lmv_check_connect(obd);
 990         if (rc)
 991                 RETURN(rc);
 992
 993         /* FIXME: choose right MDC here */
 994         CWARN("this method isn't implemented yet\n");
 995         rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
 996         RETURN(rc);
 997 }
 998
 999 static int
1000 lmv_enqueue_slaves(struct obd_export *exp, int locktype,
1001                    struct lookup_intent *it, int lockmode,
1002                    struct mdc_op_data *data, struct lustre_handle *lockh,
1003                    void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1004                    ldlm_blocking_callback cb_blocking, void *cb_data)
1005 {
1006         struct obd_device *obd = exp->exp_obd;
1007         struct lmv_obd *lmv = &obd->u.lmv;
1008         struct mea *mea = data->mea1;
1009         struct mdc_op_data *data2;
1010         int i, rc, mds;
1011         ENTRY;
1012
1013         OBD_ALLOC(data2, sizeof(*data2));
1014         if (data2 == NULL)
1015                 RETURN(-ENOMEM);
1016
1017         LASSERT(mea != NULL);
1018         for (i = 0; i < mea->mea_count; i++) {
1019                 memset(data2, 0, sizeof(*data2));
1020                 data2->id1 = mea->mea_ids[i];
1021                 mds = id_group(&data2->id1);
1022
1023                 if (lmv->tgts[mds].ltd_exp == NULL)
1024                         continue;
1025
1026                 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it,
1027                                 lockmode, data2, lockh + i, lmm, lmmsize,
1028                                 cb_compl, cb_blocking, cb_data);
1029
1030                 CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n",
1031                        OLID4(&mea->mea_ids[i]), rc, LUSTRE_IT(it)->it_status);
1032                 if (rc)
1033                         GOTO(cleanup, rc);
1034                 if (LUSTRE_IT(it)->it_data) {
1035                         struct ptlrpc_request *req;
1036                         req = (struct ptlrpc_request *) LUSTRE_IT(it)->it_data;
1037                         ptlrpc_req_finished(req);
1038                 }
1039
1040                 if (LUSTRE_IT(it)->it_status)
1041                         GOTO(cleanup, rc = LUSTRE_IT(it)->it_status);
1042         }
1043
1044         OBD_FREE(data2, sizeof(*data2));
1045         RETURN(0);
1046 cleanup:
1047         OBD_FREE(data2, sizeof(*data2));
1048
1049         /* drop all taken locks */
1050         while (--i >= 0) {
1051                 if (lockh[i].cookie)
1052                         ldlm_lock_decref(lockh + i, lockmode);
1053                 lockh[i].cookie = 0;
1054         }
1055         return rc;
1056 }
1057
1058 static int
1059 lmv_enqueue_remote(struct obd_export *exp, int lock_type,
1060                    struct lookup_intent *it, int lock_mode,
1061                    struct mdc_op_data *data, struct lustre_handle *lockh,
1062                    void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1063                    ldlm_blocking_callback cb_blocking, void *cb_data)
1064 {
1065         struct ptlrpc_request *req = LUSTRE_IT(it)->it_data;
1066         struct obd_device *obd = exp->exp_obd;
1067         struct lmv_obd *lmv = &obd->u.lmv;
1068         struct lustre_handle plock;
1069         struct mdc_op_data rdata;
1070         struct mds_body *body = NULL;
1071         int rc = 0, pmode;
1072         ENTRY;
1073
1074         body = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body));
1075         LASSERT(body != NULL);
1076
1077         if (!(body->valid & OBD_MD_MDS))
1078                 RETURN(0);
1079
1080         CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4" -> "DLID4"\n",
1081                LL_IT2STR(it), OLID4(&data->id1), OLID4(&body->id1));
1082
1083         /* we got LOOKUP lock, but we really need attrs */
1084         pmode = LUSTRE_IT(it)->it_lock_mode;
1085         LASSERT(pmode != 0);
1086         memcpy(&plock, lockh, sizeof(plock));
1087         LUSTRE_IT(it)->it_lock_mode = 0;
1088         LUSTRE_IT(it)->it_data = NULL;
1089         LASSERT((body->valid & OBD_MD_FID) != 0);
1090
1091         memcpy(&rdata, data, sizeof(rdata));
1092         rdata.id1 = body->id1;
1093         rdata.name = NULL;
1094         rdata.namelen = 0;
1095
1096         LUSTRE_IT(it)->it_disposition &= ~DISP_ENQ_COMPLETE;
1097         ptlrpc_req_finished(req);
1098
1099         rc = md_enqueue(lmv->tgts[id_group(&rdata.id1)].ltd_exp,
1100                         lock_type, it, lock_mode, &rdata, lockh, lmm,
1101                         lmmsize, cb_compl, cb_blocking, cb_data);
1102         ldlm_lock_decref(&plock, pmode);
1103         RETURN(rc);
1104 }
1105
1106 static int
1107 lmv_enqueue(struct obd_export *exp, int lock_type,
1108             struct lookup_intent *it, int lock_mode,
1109             struct mdc_op_data *data, struct lustre_handle *lockh,
1110             void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1111             ldlm_blocking_callback cb_blocking, void *cb_data)
1112 {
1113         struct obd_device *obd = exp->exp_obd;
1114         struct lmv_obd *lmv = &obd->u.lmv;
1115         struct lmv_obj *obj;
1116         int rc, mds;
1117         ENTRY;
1118
1119         rc = lmv_check_connect(obd);
1120         if (rc)
1121                 RETURN(rc);
1122
1123         if (data->mea1 && it->it_op == IT_UNLINK) {
1124                 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
1125                                         data, lockh, lmm, lmmsize,
1126                                         cb_compl, cb_blocking, cb_data);
1127                 RETURN(rc);
1128         }
1129
1130         if (data->namelen) {
1131                 obj = lmv_grab_obj(obd, &data->id1);
1132                 if (obj) {
1133                         /* directory is splitted. look for right mds for this
1134                          * name */
1135                         mds = raw_name2idx(obj->hashtype, obj->objcount,
1136                                            (char *)data->name, data->namelen);
1137                         data->id1 = obj->objs[mds].id;
1138                         lmv_put_obj(obj);
1139                 }
1140         }
1141         CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4"\n", LL_IT2STR(it),
1142                OLID4(&data->id1));
1143
1144         rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp,
1145                         lock_type, it, lock_mode, data, lockh, lmm,
1146                         lmmsize, cb_compl, cb_blocking, cb_data);
1147         if (rc == 0 && it->it_op == IT_OPEN)
1148                 rc = lmv_enqueue_remote(exp, lock_type, it, lock_mode,
1149                                         data, lockh, lmm, lmmsize,
1150                                         cb_compl, cb_blocking, cb_data);
1151         RETURN(rc);
1152 }
1153
1154 static int
1155 lmv_getattr_lock(struct obd_export *exp, struct lustre_id *id,
1156                  char *filename, int namelen, __u64 valid,
1157                  unsigned int ea_size, struct ptlrpc_request **request)
1158 {
1159         int rc, mds = id_group(id), loop = 0;
1160         struct obd_device *obd = exp->exp_obd;
1161         struct lmv_obd *lmv = &obd->u.lmv;
1162         struct lustre_id rid = *id;
1163         struct mds_body *body;
1164         struct lmv_obj *obj;
1165         ENTRY;
1166
1167         rc = lmv_check_connect(obd);
1168         if (rc)
1169                 RETURN(rc);
1170 repeat:
1171         LASSERT(++loop <= 2);
1172         obj = lmv_grab_obj(obd, id);
1173         if (obj) {
1174                 /* directory is splitted. look for right mds for this name */
1175                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1176                                    filename, namelen - 1);
1177                 rid = obj->objs[mds].id;
1178                 lmv_put_obj(obj);
1179         }
1180
1181         CDEBUG(D_OTHER, "getattr_lock for %*s on "DLID4" -> "DLID4"\n",
1182                namelen, filename, OLID4(id), OLID4(&rid));
1183
1184         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1185                              &rid, filename, namelen,
1186                              valid == OBD_MD_FLID ? valid : valid | OBD_MD_FID,
1187                              ea_size, request);
1188         if (rc == 0) {
1189                 /*
1190                  * this could be cross-node reference. in this case all we have
1191                  * right now is lustre_id triple. we'd like to find other
1192                  * attributes.
1193                  */
1194                 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
1195                 LASSERT(body != NULL);
1196                 LASSERT((body->valid & OBD_MD_FID) != 0
1197                                 || body->valid == OBD_MD_FLID);
1198
1199                 if (body->valid & OBD_MD_MDS) {
1200                         struct ptlrpc_request *req = NULL;
1201
1202                         rid = body->id1;
1203                         CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid));
1204
1205                         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1206                                              &rid, NULL, 1, valid, ea_size, &req);
1207                         ptlrpc_req_finished(*request);
1208                         *request = req;
1209                 }
1210         } else if (rc == -ERESTART) {
1211                 /* directory got splitted. time to update local object and
1212                  * repeat the request with proper MDS */
1213                 rc = lmv_get_mea_and_update_object(exp, &rid);
1214                 if (rc == 0) {
1215                         ptlrpc_req_finished(*request);
1216                         goto repeat;
1217                 }
1218         }
1219         RETURN(rc);
1220 }
1221
1222 /*
1223  * llite passes id of an target inode in data->id1 and id of directory in
1224  * data->id2
1225  */
1226 static int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
1227                     struct ptlrpc_request **request)
1228 {
1229         struct obd_device *obd = exp->exp_obd;
1230         struct lmv_obd *lmv = &obd->u.lmv;
1231         struct lmv_obj *obj;
1232         int rc, mds;
1233         ENTRY;
1234
1235         rc = lmv_check_connect(obd);
1236         if (rc)
1237                 RETURN(rc);
1238
1239         if (data->namelen != 0) {
1240                 /* usual link request */
1241                 obj = lmv_grab_obj(obd, &data->id2);
1242                 if (obj) {
1243                         rc = raw_name2idx(obj->hashtype, obj->objcount,
1244                                           data->name, data->namelen);
1245                         data->id2 = obj->objs[rc].id;
1246                         lmv_put_obj(obj);
1247                 }
1248
1249                 mds = id_group(&data->id2);
1250
1251                 CDEBUG(D_OTHER,"link "DLID4":%*s to "DLID4"\n",
1252                        OLID4(&data->id2), data->namelen, data->name,
1253                        OLID4(&data->id1));
1254         } else {
1255                 mds = id_group(&data->id1);
1256
1257                 /* request from MDS to acquire i_links for inode by id1 */
1258                 CDEBUG(D_OTHER, "inc i_nlinks for "DLID4"\n",
1259                        OLID4(&data->id1));
1260         }
1261
1262         CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n",
1263                mds, OLID4(&data->id1));
1264         rc = md_link(lmv->tgts[mds].ltd_exp, data, request);
1265
1266         RETURN(rc);
1267 }
1268
1269 static int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1270                       const char *old, int oldlen, const char *new, int newlen,
1271                       struct ptlrpc_request **request)
1272 {
1273         struct obd_device *obd = exp->exp_obd;
1274         struct lmv_obd *lmv = &obd->u.lmv;
1275         struct lmv_obj *obj;
1276         int rc, mds;
1277         ENTRY;
1278
1279         CDEBUG(D_OTHER, "rename %*s in "DLID4" to %*s in "DLID4"\n",
1280                oldlen, old, OLID4(&data->id1), newlen, new,
1281                OLID4(&data->id2));
1282
1283         rc = lmv_check_connect(obd);
1284         if (rc)
1285                 RETURN(rc);
1286
1287         if (oldlen == 0) {
1288                 /*
1289                  * MDS with old dir entry is asking another MDS to create name
1290                  * there.
1291                  */
1292                 CDEBUG(D_OTHER,
1293                        "create %*s(%d/%d) in "DLID4" pointing "
1294                        "to "DLID4"\n", newlen, new, oldlen, newlen,
1295                        OLID4(&data->id2), OLID4(&data->id1));
1296
1297                 mds = id_group(&data->id2);
1298
1299                 /*
1300                  * target directory can be splitted, sowe should forward request
1301                  * to the right MDS.
1302                  */
1303                 obj = lmv_grab_obj(obd, &data->id2);
1304                 if (obj) {
1305                         mds = raw_name2idx(obj->hashtype, obj->objcount,
1306                                            (char *)new, newlen);
1307                         data->id2 = obj->objs[mds].id;
1308                         CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1309                                OLID4(&data->id2));
1310                         lmv_put_obj(obj);
1311                 }
1312                 goto request;
1313         }
1314
1315         obj = lmv_grab_obj(obd, &data->id1);
1316         if (obj) {
1317                 /*
1318                  * directory is already splitted, so we have to forward request
1319                  * to the right MDS.
1320                  */
1321                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1322                                    (char *)old, oldlen);
1323                 data->id1 = obj->objs[mds].id;
1324                 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1325                        OLID4(&data->id1));
1326                 lmv_put_obj(obj);
1327         }
1328
1329         obj = lmv_grab_obj(obd, &data->id2);
1330         if (obj) {
1331                 /*
1332                  * directory is already splitted, so we have to forward request
1333                  * to the right MDS.
1334                  */
1335                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1336                                    (char *)new, newlen);
1337
1338                 data->id2 = obj->objs[mds].id;
1339                 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1340                        OLID4(&data->id2));
1341                 lmv_put_obj(obj);
1342         }
1343
1344         mds = id_group(&data->id1);
1345
1346 request:
1347         if (id_group(&data->id1) != id_group(&data->id2)) {
1348                 CDEBUG(D_OTHER,"cross-node rename "DLID4"/%*s to "DLID4"/%*s\n",
1349                        OLID4(&data->id1), oldlen, old, OLID4(&data->id2),
1350                        newlen, new);
1351         }
1352
1353         rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1354                        new, newlen, request);
1355         RETURN(rc);
1356 }
1357
1358 static int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1359                        struct iattr *iattr, void *ea, int ealen, void *ea2,
1360                        int ea2len, void *ea3, int ea3len,
1361                        struct ptlrpc_request **request)
1362 {
1363         struct obd_device *obd = exp->exp_obd;
1364         struct lmv_obd *lmv = &obd->u.lmv;
1365         struct ptlrpc_request *req;
1366         struct mds_body *body;
1367         struct lmv_obj *obj;
1368         int rc = 0, i;
1369         ENTRY;
1370
1371         rc = lmv_check_connect(obd);
1372         if (rc)
1373                 RETURN(rc);
1374
1375         obj = lmv_grab_obj(obd, &data->id1);
1376
1377         CDEBUG(D_OTHER, "SETATTR for "DLID4", valid 0x%x%s\n",
1378                OLID4(&data->id1), iattr->ia_valid, obj ? ", splitted" : "");
1379
1380         if (obj) {
1381                 for (i = 0; i < obj->objcount; i++) {
1382                         data->id1 = obj->objs[i].id;
1383
1384                         rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1385                                         data, iattr, ea, ealen, ea2, ea2len,
1386                                         ea3, ea3len, &req);
1387
1388                         if (id_equal_fid(&obj->id, &obj->objs[i].id)) {
1389                                 /*
1390                                  * this is master object and this request should
1391                                  * be returned back to llite.
1392                                  */
1393                                 *request = req;
1394                         } else {
1395                                 ptlrpc_req_finished(req);
1396                         }
1397
1398                         if (rc)
1399                                 break;
1400                 }
1401                 lmv_put_obj(obj);
1402         } else {
1403                 LASSERT(id_group(&data->id1) < lmv->desc.ld_tgt_count);
1404                 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1405                                 data, iattr, ea, ealen, ea2, ea2len, ea3,
1406                                 ea3len, request);
1407                 if (rc == 0) {
1408                         body = lustre_msg_buf((*request)->rq_repmsg, 0,
1409                                               sizeof(*body));
1410                         LASSERT(body != NULL);
1411                         LASSERT((body->valid & OBD_MD_FID) != 0);
1412                         LASSERT(id_group(&body->id1) == id_group(&data->id1));
1413                 }
1414         }
1415         RETURN(rc);
1416 }
1417
1418 static int lmv_sync(struct obd_export *exp, struct lustre_id *id,
1419                     struct ptlrpc_request **request)
1420 {
1421         struct obd_device *obd = exp->exp_obd;
1422         struct lmv_obd *lmv = &obd->u.lmv;
1423         int rc;
1424         ENTRY;
1425
1426         rc = lmv_check_connect(obd);
1427         if (rc)
1428                 RETURN(rc);
1429
1430         rc = md_sync(lmv->tgts[id_group(id)].ltd_exp,
1431                      id, request);
1432         RETURN(rc);
1433 }
1434
1435 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock,
1436                             struct ldlm_lock_desc *desc,
1437                             void *data, int flag)
1438 {
1439         struct lustre_handle lockh;
1440         struct lmv_obj *obj;
1441         int rc;
1442         ENTRY;
1443
1444         switch (flag) {
1445         case LDLM_CB_BLOCKING:
1446                 ldlm_lock2handle(lock, &lockh);
1447                 rc = ldlm_cli_cancel(&lockh);
1448                 if (rc < 0) {
1449                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1450                         RETURN(rc);
1451                 }
1452                 break;
1453         case LDLM_CB_CANCELING:
1454                 /* time to drop cached attrs for dirobj */
1455                 obj = lock->l_ast_data;
1456                 if (obj) {
1457                         CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
1458                                ", master "DLID4"\n",
1459                                lock->l_resource->lr_name.name[3] == 1 ?
1460                                "LOOKUP" : "UPDATE",
1461                                lock->l_resource->lr_name.name[0],
1462                                lock->l_resource->lr_name.name[1],
1463                                OLID4(&obj->id));
1464                         lmv_put_obj(obj);
1465                 }
1466                 break;
1467         default:
1468                 LBUG();
1469         }
1470         RETURN(0);
1471 }
1472
1473 static void lmv_remove_dots(struct page *page)
1474 {
1475         unsigned limit = PAGE_CACHE_SIZE;
1476         char *kaddr = page_address(page);
1477         struct ext2_dir_entry_2 *p;
1478         unsigned offs, rec_len;
1479
1480         for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1481                 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1482                 rec_len = le16_to_cpu(p->rec_len);
1483
1484                 if ((p->name_len == 1 && p->name[0] == '.') ||
1485                     (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1486                         p->inode = 0;
1487         }
1488 }
1489
1490 static int lmv_readpage(struct obd_export *exp, struct lustre_id *id,
1491                         __u64 offset, struct page *page,
1492                         struct ptlrpc_request **request)
1493 {
1494         struct obd_device *obd = exp->exp_obd;
1495         struct lmv_obd *lmv = &obd->u.lmv;
1496         struct lustre_id rid = *id;
1497         struct lmv_obj *obj;
1498         int rc, i;
1499         ENTRY;
1500
1501 #warning "we need well-desgined readdir() implementation"
1502         rc = lmv_check_connect(obd);
1503         if (rc)
1504                 RETURN(rc);
1505
1506         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
1507         CDEBUG(D_OTHER, "READPAGE at %llu from "DLID4"\n",
1508                offset, OLID4(&rid));
1509
1510         obj = lmv_grab_obj(obd, id);
1511         if (obj) {
1512                 lmv_lock_obj(obj);
1513
1514                 /* find dirobj containing page with requested offset. */
1515                 for (i = 0; i < obj->objcount; i++) {
1516                         if (offset < obj->objs[i].size)
1517                                 break;
1518                         offset -= obj->objs[i].size;
1519                 }
1520                 rid = obj->objs[i].id;
1521
1522                 lmv_unlock_obj(obj);
1523                 lmv_put_obj(obj);
1524
1525                 CDEBUG(D_OTHER, "forward to "DLID4" with offset %lu\n",
1526                        OLID4(&rid), (unsigned long)offset);
1527         }
1528         rc = md_readpage(lmv->tgts[id_group(&rid)].ltd_exp, &rid,
1529                          offset, page, request);
1530
1531         if (rc == 0 && !id_equal_fid(&rid, id))
1532                 /* this page isn't from master object. To avoid "." and ".."
1533                  * duplication in directory, we have to remove them from all
1534                  * slave objects */
1535                 lmv_remove_dots(page);
1536
1537         RETURN(rc);
1538 }
1539
1540 static int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1541                              struct ptlrpc_request **req)
1542 {
1543         struct obd_device *obd = exp->exp_obd;
1544         struct lmv_obd *lmv = &obd->u.lmv;
1545         struct mea *mea = data->mea1;
1546         struct mdc_op_data *data2;
1547         int i, rc = 0;
1548         ENTRY;
1549
1550         OBD_ALLOC(data2, sizeof(*data2));
1551         if (data2 == NULL)
1552                 RETURN(-ENOMEM);
1553
1554         LASSERT(mea != NULL);
1555         for (i = 0; i < mea->mea_count; i++) {
1556                 memset(data2, 0, sizeof(*data2));
1557                 data2->id1 = mea->mea_ids[i];
1558                 data2->create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1559
1560                 if (lmv->tgts[id_group(&data2->id1)].ltd_exp == NULL)
1561                         continue;
1562
1563                 rc = md_unlink(lmv->tgts[id_group(&data2->id1)].ltd_exp,
1564                                data2, req);
1565
1566                 CDEBUG(D_OTHER, "unlink slave "DLID4" -> %d\n",
1567                        OLID4(&mea->mea_ids[i]), rc);
1568
1569                 if (*req) {
1570                         ptlrpc_req_finished(*req);
1571                         *req = NULL;
1572                 }
1573                 if (rc)
1574                         RETURN(rc);
1575         }
1576         OBD_FREE(data2, sizeof(*data2));
1577         RETURN(rc);
1578 }
1579
1580 static int lmv_delete_inode(struct obd_export *exp, struct lustre_id *id)
1581 {
1582         ENTRY;
1583
1584         LASSERT(exp && id);
1585         if (lmv_delete_obj(exp, id)) {
1586                 CDEBUG(D_OTHER, "lmv object "DLID4" is destroyed.\n",
1587                        OLID4(id));
1588         }
1589         RETURN(0);
1590 }
1591
1592 static int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1593                       struct ptlrpc_request **request)
1594 {
1595         struct obd_device *obd = exp->exp_obd;
1596         struct lmv_obd *lmv = &obd->u.lmv;
1597         int rc, i = 0;
1598         ENTRY;
1599
1600         rc = lmv_check_connect(obd);
1601         if (rc)
1602                 RETURN(rc);
1603
1604         if (data->namelen == 0 && data->mea1 != NULL) {
1605                 /* mds asks to remove slave objects */
1606                 rc = lmv_unlink_slaves(exp, data, request);
1607                 RETURN(rc);
1608         }
1609
1610         if (data->namelen != 0) {
1611                 struct lmv_obj *obj;
1612
1613                 obj = lmv_grab_obj(obd, &data->id1);
1614                 if (obj) {
1615                         i = raw_name2idx(obj->hashtype, obj->objcount,
1616                                          data->name, data->namelen);
1617                         data->id1 = obj->objs[i].id;
1618                         lmv_put_obj(obj);
1619                 }
1620                 CDEBUG(D_OTHER, "unlink '%*s' in "DLID4" -> %u\n",
1621                        data->namelen, data->name, OLID4(&data->id1),
1622                        i);
1623         } else {
1624                 CDEBUG(D_OTHER, "drop i_nlink on "DLID4"\n",
1625                        OLID4(&data->id1));
1626         }
1627         rc = md_unlink(lmv->tgts[id_group(&data->id1)].ltd_exp,
1628                        data, request);
1629         RETURN(rc);
1630 }
1631
1632 static struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1633                                            struct lustre_id *id)
1634 {
1635         struct obd_device *obd = exp->exp_obd;
1636         struct lmv_obd *lmv = &obd->u.lmv;
1637         int rc;
1638         ENTRY;
1639
1640         rc = lmv_check_connect(obd);
1641         if (rc)
1642                 RETURN(ERR_PTR(rc));
1643         obd = lmv->tgts[id_group(id)].ltd_exp->exp_obd;
1644         EXIT;
1645
1646         return obd;
1647 }
1648
1649 static int lmv_init_ea_size(struct obd_export *exp, int easize,
1650                             int cookiesize)
1651 {
1652         struct obd_device *obd = exp->exp_obd;
1653         struct lmv_obd *lmv = &obd->u.lmv;
1654         int i, rc = 0, change = 0;
1655         ENTRY;
1656
1657         if (lmv->max_easize < easize) {
1658                 lmv->max_easize = easize;
1659                 change = 1;
1660         }
1661         if (lmv->max_cookiesize < cookiesize) {
1662                 lmv->max_cookiesize = cookiesize;
1663                 change = 1;
1664         }
1665         if (change == 0)
1666                 RETURN(0);
1667
1668         if (lmv->connected == 0)
1669                 RETURN(0);
1670
1671         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1672                 if (lmv->tgts[i].ltd_exp == NULL) {
1673                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
1674                         continue;
1675                 }
1676
1677                 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
1678                 if (rc) {
1679                         CERROR("obd_init_ea_size() failed on MDT target %d, "
1680                                "error %d.\n", i, rc);
1681                         break;
1682                 }
1683         }
1684         RETURN(rc);
1685 }
1686
1687 static int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1688                                  void *acl, int acl_size,
1689                                  struct lov_stripe_md **ea,
1690                                  struct obd_trans_info *oti)
1691 {
1692         struct obd_device *obd = exp->exp_obd;
1693         struct lmv_obd *lmv = &obd->u.lmv;
1694         struct lov_stripe_md obj_md;
1695         struct lov_stripe_md *obj_mdp = &obj_md;
1696         int rc = 0;
1697         ENTRY;
1698
1699         LASSERT(ea == NULL);
1700         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1701
1702         rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa,
1703                         acl, acl_size, &obj_mdp, oti);
1704
1705         RETURN(rc);
1706 }
1707
1708 /*
1709  * to be called from MDS only. @oa should have correct store cookie and o_fid
1710  * values for "master" object, as it will be used.
1711  */
1712 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
1713                    void *acl, int acl_size,
1714                    struct lov_stripe_md **ea, struct obd_trans_info *oti)
1715 {
1716         struct obd_device *obd = exp->exp_obd;
1717         struct lmv_obd *lmv = &obd->u.lmv;
1718         struct lustre_id mid;
1719         int i, c, rc = 0;
1720         struct mea *mea;
1721         ENTRY;
1722
1723         rc = lmv_check_connect(obd);
1724         if (rc)
1725                 RETURN(rc);
1726
1727         LASSERT(oa != NULL);
1728
1729         if (ea == NULL) {
1730                 rc = lmv_obd_create_single(exp, oa, acl, acl_size, NULL, oti);
1731                 if (rc)
1732                         CERROR("Can't create object, rc = %d\n", rc);
1733                 RETURN(rc);
1734         }
1735
1736         /* acl is only suppied when mds create single remote obj */
1737         LASSERT(acl == NULL && acl_size == 0);
1738
1739         if (*ea == NULL) {
1740                 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1741                 if (rc < 0) {
1742                         CERROR("obd_alloc_diskmd() failed, error %d\n",
1743                                rc);
1744                         RETURN(rc);
1745                 } else
1746                         rc = 0;
1747
1748                 if (*ea == NULL)
1749                         RETURN(-ENOMEM);
1750         }
1751
1752         /*
1753          * here we should take care about splitted dir, so store cookie and fid
1754          * for "master" object should already be allocated and passed in @oa.
1755          */
1756         LASSERT(oa->o_id != 0);
1757         LASSERT(oa->o_fid != 0);
1758
1759         /* save "master" object id */
1760         obdo2id(&mid, oa);
1761
1762         mea = (struct mea *)*ea;
1763         mea->mea_master = -1;
1764         mea->mea_magic = MEA_MAGIC_ALL_CHARS;
1765
1766         if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1767                 mea->mea_count = lmv->desc.ld_tgt_count;
1768
1769         for (i = 0, c = 0; c < mea->mea_count && i < lmv->desc.ld_tgt_count; i++) {
1770                 struct lov_stripe_md obj_md;
1771                 struct lov_stripe_md *obj_mdp = &obj_md;
1772
1773                 if (lmv->tgts[i].ltd_exp == NULL) {
1774                         /* this is "master" MDS */
1775                         mea->mea_master = i;
1776                         mea->mea_ids[c] = mid;
1777                         c++;
1778                         continue;
1779                 }
1780
1781                 /*
1782                  * "master" MDS should always be part of stripped dir,
1783                  * so scan for it.
1784                  */
1785                 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1786                         continue;
1787
1788                 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE |
1789                         OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1790
1791                 rc = obd_create(lmv->tgts[c].ltd_exp, oa, NULL, 0,
1792                                 &obj_mdp, oti);
1793                 if (rc) {
1794                         CERROR("obd_create() failed on MDT target %d, "
1795                                "error %d\n", c, rc);
1796                         RETURN(rc);
1797                 }
1798
1799                 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1800                        i, oa->o_id, oa->o_generation);
1801
1802
1803                 /*
1804                  * here, when object is created (or it is master and was passed
1805                  * from caller) on desired MDS we save its fid to local mea_ids.
1806                  */
1807                 LASSERT(oa->o_fid);
1808
1809                 /*
1810                  * store cookie should be defined here for both cases (master
1811                  * object and not master), because master is already created.
1812                  */
1813                 LASSERT(oa->o_id);
1814
1815                 /* fill mea by store cookie and fid */
1816                 obdo2id(&mea->mea_ids[c], oa);
1817                 c++;
1818         }
1819         LASSERT(c == mea->mea_count);
1820
1821         CDEBUG(D_OTHER, "%d dirobjects created\n",
1822                (int)mea->mea_count);
1823
1824         RETURN(rc);
1825 }
1826
1827 static int lmv_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1828                          struct obd_device *tgt, int count,
1829                          struct llog_catid *logid)
1830 {
1831         struct llog_ctxt *ctxt;
1832         int rc;
1833         ENTRY;
1834
1835         rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1836                             &llog_client_ops);
1837         if (rc == 0) {
1838                 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1839                 ctxt->loc_imp = tgt->u.cli.cl_import;
1840         }
1841
1842         RETURN(rc);
1843 }
1844
1845 static int lmv_llog_finish(struct obd_device *obd,
1846                            struct obd_llogs *llogs, int count)
1847 {
1848         int rc;
1849         ENTRY;
1850
1851         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1852         RETURN(rc);
1853 }
1854
1855 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1856                         void *key, __u32 *vallen, void *val)
1857 {
1858         struct obd_device *obd;
1859         struct lmv_obd *lmv;
1860         int rc = 0;
1861         ENTRY;
1862
1863         obd = class_exp2obd(exp);
1864         if (obd == NULL) {
1865                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1866                        exp->exp_handle.h_cookie);
1867                 RETURN(-EINVAL);
1868         }
1869
1870         lmv = &obd->u.lmv;
1871         if (keylen == 6 && memcmp(key, "mdsize", 6) == 0) {
1872                 __u32 *mdsize = val;
1873                 *vallen = sizeof(__u32);
1874                 *mdsize = sizeof(struct lustre_id) * lmv->desc.ld_tgt_count
1875                         + sizeof(struct mea);
1876                 RETURN(0);
1877         } else if (keylen == 6 && memcmp(key, "mdsnum", 6) == 0) {
1878                 struct obd_uuid *cluuid = &lmv->cluuid;
1879                 struct lmv_tgt_desc *tgts;
1880                 __u32 *mdsnum = val;
1881                 int i;
1882
1883                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1884                         if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1885                                 *vallen = sizeof(__u32);
1886                                 *mdsnum = i;
1887                                 RETURN(0);
1888                         }
1889                 }
1890                 LASSERT(0);
1891         } else if (keylen == 6 && memcmp(key, "rootid", 6) == 0) {
1892                 rc = lmv_check_connect(obd);
1893                 if (rc)
1894                         RETURN(rc);
1895
1896                 /* getting rootid from first MDS. */
1897                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
1898                                   vallen, val);
1899                 RETURN(rc);
1900         } else if (keylen >= strlen("lmvdesc") && strcmp(key, "lmvdesc") == 0) {
1901                 struct lmv_desc *desc_ret = val;
1902                 *desc_ret = lmv->desc;
1903                 RETURN(0);
1904         } else if (keylen == strlen("remote_flag") &&
1905                    !strcmp(key, "remote_flag")) {
1906                 struct lmv_tgt_desc *tgts;
1907                 int i;
1908
1909                 rc = lmv_check_connect(obd);
1910                 if (rc)
1911                         RETURN(rc);
1912
1913                 LASSERT(*vallen == sizeof(__u32));
1914                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
1915                      i++, tgts++) {
1916
1917                         /* all tgts should be connected when this get called. */
1918                         if (!tgts || !tgts->ltd_exp) {
1919                                 CERROR("target not setup?\n");
1920                                 continue;
1921                         }
1922
1923                         if (!obd_get_info(tgts->ltd_exp, keylen, key,
1924                                           vallen, val))
1925                                 RETURN(0);
1926                 }
1927                 RETURN(-EINVAL);
1928         } else if (keylen >= strlen("lovdesc") && strcmp(key, "lovdesc") == 0) {
1929                 rc = lmv_check_connect(obd);
1930                 if (rc)
1931                         RETURN(rc);
1932
1933                 /* forwarding this request to first MDS, it should know LOV
1934                  * desc. */
1935                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
1936                                   vallen, val);
1937                 RETURN(rc);
1938         }
1939
1940         CDEBUG(D_IOCTL, "invalid key\n");
1941         RETURN(-EINVAL);
1942 }
1943
1944 int lmv_set_info(struct obd_export *exp, obd_count keylen,
1945                  void *key, obd_count vallen, void *val)
1946 {
1947         struct lmv_tgt_desc    *tgt;
1948         struct obd_device      *obd;
1949         struct lmv_obd         *lmv;
1950         ENTRY;
1951
1952         obd = class_exp2obd(exp);
1953         if (obd == NULL) {
1954                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1955                        exp->exp_handle.h_cookie);
1956                 RETURN(-EINVAL);
1957         }
1958         lmv = &obd->u.lmv;
1959
1960         if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
1961                 lmv->server_timeout = 1;
1962                 lmv_set_timeouts(obd);
1963                 RETURN(0);
1964         }
1965
1966         /* maybe this could be default */
1967         if ((keylen == strlen("sec") && strcmp(key, "sec") == 0) ||
1968             (keylen == strlen("sec_flags") && strcmp(key, "sec_flags") == 0) ||
1969             (keylen == strlen("nllu") && strcmp(key, "nllu") == 0)) {
1970                 struct obd_export *exp;
1971                 int rc = 0, err, i;
1972
1973                 spin_lock(&lmv->lmv_lock);
1974                 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count;
1975                      i++, tgt++) {
1976                         exp = tgt->ltd_exp;
1977                         /* during setup time the connections to mdc might
1978                          * haven't been established.
1979                          */
1980                         if (exp == NULL) {
1981                                 struct obd_device *tgt_obd;
1982
1983                                 tgt_obd = class_find_client_obd(&tgt->uuid,
1984                                                                 OBD_MDC_DEVICENAME,
1985                                                                 &obd->obd_uuid);
1986                                 if (!tgt_obd) {
1987                                         CERROR("can't set info %s, "
1988                                                "device %s not attached?\n",
1989                                                 (char *) key, tgt->uuid.uuid);
1990                                         rc = -EINVAL;
1991                                         continue;
1992                                 }
1993                                 exp = tgt_obd->obd_self_export;
1994                         }
1995
1996                         err = obd_set_info(exp, keylen, key, vallen, val);
1997                         if (!rc)
1998                                 rc = err;
1999                 }
2000                 spin_unlock(&lmv->lmv_lock);
2001
2002                 RETURN(rc);
2003         }
2004
2005         if (((keylen == strlen("flush_cred") &&
2006              strcmp(key, "flush_cred") == 0)) ||
2007              ((keylen == strlen("crypto_type") &&
2008              strcmp(key, "crypto_type") == 0))) {
2009                 int rc = 0, i;
2010
2011                 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count;
2012                      i++, tgt++) {
2013                         if (!tgt->ltd_exp)
2014                                 continue;
2015                         rc = obd_set_info(tgt->ltd_exp,
2016                                           keylen, key, vallen, val);
2017                         if (rc)
2018                                 RETURN(rc);
2019                 }
2020
2021                 RETURN(0);
2022         }
2023
2024         RETURN(-EINVAL);
2025 }
2026
2027 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2028                struct lov_stripe_md *lsm)
2029 {
2030         struct obd_device *obd = class_exp2obd(exp);
2031         struct lmv_obd *lmv = &obd->u.lmv;
2032         struct mea *meap, *lsmp;
2033         int mea_size, i;
2034         ENTRY;
2035
2036         mea_size = (sizeof(struct lustre_id) *
2037                     lmv->desc.ld_tgt_count) + sizeof(struct mea);
2038         if (!lmmp)
2039                 RETURN(mea_size);
2040
2041         if (*lmmp && !lsm) {
2042                 OBD_FREE(*lmmp, mea_size);
2043                 *lmmp = NULL;
2044                 RETURN(0);
2045         }
2046
2047         if (*lmmp == NULL) {
2048                 OBD_ALLOC(*lmmp, mea_size);
2049                 if (*lmmp == NULL)
2050                         RETURN(-ENOMEM);
2051         }
2052
2053         if (!lsm)
2054                 RETURN(mea_size);
2055
2056         lsmp = (struct mea *)lsm;
2057         meap = (struct mea *)*lmmp;
2058
2059         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2060             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2061                 RETURN(-EINVAL);
2062
2063         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2064         meap->mea_count = cpu_to_le32(lsmp->mea_count);
2065         meap->mea_master = cpu_to_le32(lsmp->mea_master);
2066
2067         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2068                 meap->mea_ids[i] = meap->mea_ids[i];
2069                 id_cpu_to_le(&meap->mea_ids[i]);
2070         }
2071
2072         RETURN(mea_size);
2073 }
2074
2075 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2076                  struct lov_mds_md *lmm, int lmm_size)
2077 {
2078         struct obd_device *obd = class_exp2obd(exp);
2079         struct mea **tmea = (struct mea **)lsmp;
2080         struct mea *mea = (struct mea *)lmm;
2081         struct lmv_obd *lmv = &obd->u.lmv;
2082         int mea_size, i, rc = 0;
2083         __u32 magic;
2084         ENTRY;
2085
2086         mea_size = sizeof(struct lustre_id) *
2087                 lmv->desc.ld_tgt_count + sizeof(struct mea);
2088
2089         if (lsmp == NULL)
2090                 return mea_size;
2091
2092         if (*lsmp != NULL && lmm == NULL) {
2093                 OBD_FREE(*tmea, mea_size);
2094                 RETURN(0);
2095         }
2096
2097         LASSERT(mea_size == lmm_size);
2098
2099         OBD_ALLOC(*tmea, mea_size);
2100         if (*tmea == NULL)
2101                 RETURN(-ENOMEM);
2102
2103         if (!lmm)
2104                 RETURN(mea_size);
2105
2106         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2107             mea->mea_magic == MEA_MAGIC_ALL_CHARS)
2108         {
2109                 magic = le32_to_cpu(mea->mea_magic);
2110         } else {
2111                 struct mea_old *old = (struct mea_old *)lmm;
2112
2113                 mea_size = sizeof(struct lustre_id) * old->mea_count +
2114                         sizeof(struct mea_old);
2115
2116                 if (old->mea_count > 256 || old->mea_master > 256 ||
2117                     lmm_size < mea_size || old->mea_master > old->mea_count) {
2118                         CWARN("bad MEA: count %u, master %u, size %u\n",
2119                               old->mea_count, old->mea_master, mea_size);
2120                         GOTO(out_free_mea, rc = -EINVAL);
2121                 }
2122                 magic = MEA_MAGIC_LAST_CHAR;
2123         }
2124
2125         (*tmea)->mea_magic = magic;
2126         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2127         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2128
2129         for (i = 0; i < (*tmea)->mea_count; i++) {
2130                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2131                 id_le_to_cpu(&(*tmea)->mea_ids[i]);
2132         }
2133         RETURN(mea_size);
2134
2135 out_free_mea:
2136         OBD_FREE(*tmea, mea_size);
2137         return rc;
2138 }
2139
2140 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
2141             struct lov_stripe_md *ea, obd_count oa_bufs,
2142             struct brw_page *pgarr, struct obd_trans_info *oti)
2143 {
2144         struct obd_device *obd = exp->exp_obd;
2145         struct lmv_obd *lmv = &obd->u.lmv;
2146         struct mea *mea = (struct mea *) ea;
2147         int err;
2148
2149         LASSERT(oa != NULL);
2150         LASSERT(ea != NULL);
2151         LASSERT(pgarr != NULL);
2152         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
2153
2154         oa->o_gr = id_gen(&mea->mea_ids[oa->o_mds]);
2155         oa->o_id = id_ino(&mea->mea_ids[oa->o_mds]);
2156         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2157
2158         err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp,
2159                       oa, NULL, oa_bufs, pgarr, oti);
2160         RETURN(err);
2161 }
2162
2163 static int lmv_cancel_unused(struct obd_export *exp,
2164                              struct lov_stripe_md *lsm,
2165                              int flags, void *opaque)
2166 {
2167         struct obd_device *obd = exp->exp_obd;
2168         struct lmv_obd *lmv = &obd->u.lmv;
2169         int rc = 0, err, i;
2170         ENTRY;
2171
2172         LASSERT(lsm == NULL);
2173
2174         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2175                 if (!lmv->tgts[i].ltd_exp || !lmv->tgts[i].active)
2176                         continue;
2177
2178                 err = obd_cancel_unused(lmv->tgts[i].ltd_exp,
2179                                         NULL, flags, opaque);
2180                 if (!rc)
2181                         rc = err;
2182         }
2183         RETURN(rc);
2184 }
2185
2186 struct obd_ops lmv_obd_ops = {
2187         .o_owner                = THIS_MODULE,
2188         .o_attach               = lmv_attach,
2189         .o_detach               = lmv_detach,
2190         .o_setup                = lmv_setup,
2191         .o_cleanup              = lmv_cleanup,
2192         .o_connect              = lmv_connect,
2193         .o_disconnect           = lmv_disconnect,
2194         .o_statfs               = lmv_statfs,
2195         .o_llog_init            = lmv_llog_init,
2196         .o_llog_finish          = lmv_llog_finish,
2197         .o_get_info             = lmv_get_info,
2198         .o_set_info             = lmv_set_info,
2199         .o_create               = lmv_obd_create,
2200         .o_packmd               = lmv_packmd,
2201         .o_unpackmd             = lmv_unpackmd,
2202         .o_brw                  = lmv_brw,
2203         .o_init_ea_size         = lmv_init_ea_size,
2204         .o_notify               = lmv_notify,
2205         .o_iocontrol            = lmv_iocontrol,
2206         .o_cancel_unused        = lmv_cancel_unused,
2207 };
2208
2209 struct md_ops lmv_md_ops = {
2210         .m_getstatus           = lmv_getstatus,
2211         .m_getattr             = lmv_getattr,
2212         .m_change_cbdata       = lmv_change_cbdata,
2213         .m_change_cbdata_name  = lmv_change_cbdata_name,
2214         .m_close               = lmv_close,
2215         .m_create              = lmv_create,
2216         .m_done_writing        = lmv_done_writing,
2217         .m_enqueue             = lmv_enqueue,
2218         .m_getattr_lock        = lmv_getattr_lock,
2219         .m_intent_lock         = lmv_intent_lock,
2220         .m_link                = lmv_link,
2221         .m_rename              = lmv_rename,
2222         .m_setattr             = lmv_setattr,
2223         .m_sync                = lmv_sync,
2224         .m_readpage            = lmv_readpage,
2225         .m_unlink              = lmv_unlink,
2226         .m_get_real_obd        = lmv_get_real_obd,
2227         .m_valid_attrs         = lmv_valid_attrs,
2228         .m_delete_inode        = lmv_delete_inode,
2229         .m_access_check        = lmv_access_check,
2230 };
2231
2232 int __init lmv_init(void)
2233 {
2234         struct lprocfs_static_vars lvars;
2235         int rc;
2236
2237         obj_cache = kmem_cache_create("lmv_objects",
2238                                       sizeof(struct lmv_obj),
2239                                       0, 0, NULL, NULL);
2240         if (!obj_cache) {
2241                 CERROR("error allocating lmv objects cache\n");
2242                 return -ENOMEM;
2243         }
2244
2245         lprocfs_init_vars(lmv, &lvars);
2246         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2247                                  lvars.module_vars,
2248                                  OBD_LMV_DEVICENAME);
2249         if (rc)
2250                 kmem_cache_destroy(obj_cache);
2251
2252         return rc;
2253 }
2254
2255 #ifdef __KERNEL__
2256 static void lmv_exit(void)
2257 {
2258         class_unregister_type(OBD_LMV_DEVICENAME);
2259
2260         LASSERTF(kmem_cache_destroy(obj_cache) == 0,
2261                  "can't free lmv objects cache, %d object(s)"
2262                  "still in use\n", atomic_read(&obj_cache_count));
2263 }
2264
2265 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2266 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2267 MODULE_LICENSE("GPL");
2268
2269 module_init(lmv_init);
2270 module_exit(lmv_exit);
2271 #endif