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