lustre/lmv/lmv_obd.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of Lustre, http://www.lustre.org.
   7  *
   8  *   Lustre is free software; you can redistribute it and/or
   9  *   modify it under the terms of version 2 of the GNU General Public
  10  *   License as published by the Free Software Foundation.
  11  *
  12  *   Lustre is distributed in the hope that it will be useful,
  13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *   GNU General Public License for more details.
  16  *
  17  *   You should have received a copy of the GNU General Public License
  18  *   along with Lustre; if not, write to the Free Software
  19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  */
  21
  22 #ifndef EXPORT_SYMTAB
  23 # define EXPORT_SYMTAB
  24 #endif
  25 #define DEBUG_SUBSYSTEM S_LMV
  26 #ifdef __KERNEL__
  27 #include <linux/slab.h>
  28 #include <linux/module.h>
  29 #include <linux/init.h>
  30 #include <linux/slab.h>
  31 #include <linux/pagemap.h>
  32 #include <asm/div64.h>
  33 #include <linux/seq_file.h>
  34 #include <linux/namei.h>
  35 #else
  36 #include <liblustre.h>
  37 #endif
  38 #include <linux/ext2_fs.h>
  39
  40 #include <linux/obd_support.h>
  41 #include <linux/lustre_lib.h>
  42 #include <linux/lustre_net.h>
  43 #include <linux/lustre_idl.h>
  44 #include <linux/lustre_dlm.h>
  45 #include <linux/lustre_mds.h>
  46 #include <linux/obd_class.h>
  47 #include <linux/obd_ost.h>
  48 #include <linux/lprocfs_status.h>
  49 #include <linux/lustre_fsfilt.h>
  50 #include <linux/obd_lmv.h>
  51 #include <linux/lustre_lite.h>
  52 #include <linux/lustre_audit.h>
  53 #include "lmv_internal.h"
  54
  55 /* not defined for liblustre building */
  56 #if !defined(ATOMIC_INIT)
  57 #define ATOMIC_INIT(val) { (val) }
  58 #endif
  59
  60 /* object cache. */
  61 kmem_cache_t *obj_cache;
  62 atomic_t obj_cache_count = ATOMIC_INIT(0);
  63
  64 static void lmv_activate_target(struct lmv_obd *lmv,
  65                                 struct lmv_tgt_desc *tgt,
  66                                 int activate)
  67 {
  68         if (tgt->active == activate)
  69                 return;
  70
  71         tgt->active = activate;
  72         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  73 }
  74
  75 /* Error codes:
  76  *
  77  *  -EINVAL  : UUID can't be found in the LMV's target list
  78  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  79  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  80  */
  81 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
  82                               int activate)
  83 {
  84         struct lmv_tgt_desc *tgt;
  85         struct obd_device *obd;
  86         int i, rc = 0;
  87         ENTRY;
  88
  89         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  90                lmv, uuid->uuid, activate);
  91
  92         spin_lock(&lmv->lmv_lock);
  93         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
  94                 if (tgt->ltd_exp == NULL)
  95                         continue;
  96
  97                 CDEBUG(D_INFO, "lmv idx %d is %s conn "LPX64"\n",
  98                        i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
  99
 100                 if (obd_uuid_equals(uuid, &tgt->uuid))
 101                         break;
 102         }
 103
 104         if (i == lmv->desc.ld_tgt_count)
 105                 GOTO(out_lmv_lock, rc = -EINVAL);
 106
 107         obd = class_exp2obd(tgt->ltd_exp);
 108         if (obd == NULL)
 109                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 110
 111         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 112                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 113                obd->obd_type->typ_name, i);
 114         LASSERT(strcmp(obd->obd_type->typ_name, OBD_MDC_DEVICENAME) == 0);
 115
 116         if (tgt->active == activate) {
 117                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 118                        activate ? "" : "in");
 119                 GOTO(out_lmv_lock, rc);
 120         }
 121
 122         CDEBUG(D_INFO, "Marking OBD %p %sactive\n",
 123                obd, activate ? "" : "in");
 124
 125         lmv_activate_target(lmv, tgt, activate);
 126
 127         EXIT;
 128
 129  out_lmv_lock:
 130         spin_unlock(&lmv->lmv_lock);
 131         return rc;
 132 }
 133
 134 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 135                       int active, void *data)
 136 {
 137         struct obd_uuid *uuid;
 138         int rc;
 139         ENTRY;
 140
 141         if (strcmp(watched->obd_type->typ_name, OBD_MDC_DEVICENAME)) {
 142                 CERROR("unexpected notification of %s %s!\n",
 143                        watched->obd_type->typ_name,
 144                        watched->obd_name);
 145                 RETURN(-EINVAL);
 146         }
 147         uuid = &watched->u.cli.cl_import->imp_target_uuid;
 148
 149         /* Set MDC as active before notifying the observer, so the observer can
 150          * use the MDC normally.
 151          */
 152         rc = lmv_set_mdc_active(&obd->u.lmv, uuid, active);
 153         if (rc) {
 154                 CERROR("%sactivation of %s failed: %d\n",
 155                        active ? "" : "de", uuid->uuid, rc);
 156                 RETURN(rc);
 157         }
 158
 159         if (obd->obd_observer)
 160                 /* Pass the notification up the chain. */
 161                 rc = obd_notify(obd->obd_observer, watched, active, data);
 162
 163         RETURN(rc);
 164 }
 165
 166 static int lmv_attach(struct obd_device *dev, obd_count len, void *data)
 167 {
 168         struct lprocfs_static_vars lvars;
 169         int rc;
 170         ENTRY;
 171
 172         lprocfs_init_vars(lmv, &lvars);
 173         rc = lprocfs_obd_attach(dev, lvars.obd_vars);
 174 #ifdef __KERNEL__
 175         if (rc == 0) {
 176                 struct proc_dir_entry *entry;
 177
 178                 entry = create_proc_entry("target_obd_status", 0444,
 179                                            dev->obd_proc_entry);
 180                 if (entry == NULL)
 181                         RETURN(-ENOMEM);
 182                 entry->proc_fops = &lmv_proc_target_fops;
 183                 entry->data = dev;
 184        }
 185 #endif
 186         RETURN (rc);
 187 }
 188
 189 static int lmv_detach(struct obd_device *dev)
 190 {
 191         return lprocfs_obd_detach(dev);
 192 }
 193
 194 /* this is fake connect function. Its purpose is to initialize lmv and say
 195  * caller that everything is okay. Real connection will be performed later. */
 196 static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd,
 197                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 198                        unsigned long flags)
 199 {
 200 #ifdef __KERNEL__
 201         struct proc_dir_entry *lmv_proc_dir;
 202 #endif
 203         struct lmv_obd *lmv = &obd->u.lmv;
 204         struct obd_export *exp;
 205         int rc = 0;
 206         ENTRY;
 207
 208         rc = class_connect(conn, obd, cluuid);
 209         if (rc) {
 210                 CERROR("class_connection() returned %d\n", rc);
 211                 RETURN(rc);
 212         }
 213
 214         exp = class_conn2export(conn);
 215
 216         /* we don't want to actually do the underlying connections more than
 217          * once, so keep track. */
 218         lmv->refcount++;
 219         if (lmv->refcount > 1) {
 220                 class_export_put(exp);
 221                 RETURN(0);
 222         }
 223
 224         lmv->exp = exp;
 225         lmv->connected = 0;
 226         lmv->cluuid = *cluuid;
 227         lmv->connect_flags = flags;
 228         sema_init(&lmv->init_sem, 1);
 229         if (data)
 230                 memcpy(&lmv->conn_data, data, sizeof(*data));
 231
 232 #ifdef __KERNEL__
 233         lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
 234                                         NULL, NULL);
 235         if (IS_ERR(lmv_proc_dir)) {
 236                 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 237                        obd->obd_type->typ_name, obd->obd_name);
 238                 lmv_proc_dir = NULL;
 239         }
 240 #endif
 241
 242         /*
 243          * all real clients shouls perform actual connection rightaway, because
 244          * it is possible, that LMV will not have opportunity to connect
 245          * targets, as MDC stuff will bit called directly, for instance while
 246          * reading ../mdc/../kbytesfree procfs file, etc.
 247          */
 248         if (flags & OBD_OPT_REAL_CLIENT)
 249                 rc = lmv_check_connect(obd);
 250
 251 #ifdef __KERNEL__
 252         if (rc) {
 253                 if (lmv_proc_dir)
 254                         lprocfs_remove(lmv_proc_dir);
 255         }
 256 #endif
 257
 258         RETURN(rc);
 259 }
 260
 261 static void lmv_set_timeouts(struct obd_device *obd)
 262 {
 263         struct lmv_tgt_desc *tgts;
 264         struct lmv_obd *lmv;
 265         int i;
 266
 267         lmv = &obd->u.lmv;
 268         if (lmv->server_timeout == 0)
 269                 return;
 270
 271         if (lmv->connected == 0)
 272                 return;
 273
 274         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
 275                 if (tgts->ltd_exp == NULL)
 276                         continue;
 277
 278                 obd_set_info(tgts->ltd_exp, strlen("inter_mds"),
 279                              "inter_mds", 0, NULL);
 280         }
 281 }
 282
 283 #define MAX_STRING_SIZE 128
 284
 285 /* performs a check if passed obd is connected. If no - connect it. */
 286 int lmv_check_connect(struct obd_device *obd)
 287 {
 288 #ifdef __KERNEL__
 289         struct proc_dir_entry *lmv_proc_dir;
 290 #endif
 291         struct lmv_obd *lmv = &obd->u.lmv;
 292         struct lmv_tgt_desc *tgts;
 293         struct obd_uuid *cluuid;
 294         struct obd_export *exp;
 295         int rc, rc2, i;
 296         ENTRY;
 297
 298         if (lmv->connected)
 299                 RETURN(0);
 300
 301         down(&lmv->init_sem);
 302         if (lmv->connected) {
 303                 up(&lmv->init_sem);
 304                 RETURN(0);
 305         }
 306
 307         cluuid = &lmv->cluuid;
 308         exp = lmv->exp;
 309
 310         CDEBUG(D_OTHER, "time to connect %s to %s\n",
 311                cluuid->uuid, obd->obd_name);
 312
 313         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
 314                 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
 315                 struct lustre_handle conn = {0, };
 316                 struct obd_device *tgt_obd;
 317
 318                 LASSERT(tgts != NULL);
 319
 320                 tgt_obd = class_find_client_obd(&tgts->uuid, OBD_MDC_DEVICENAME,
 321                                                 &obd->obd_uuid);
 322                 if (!tgt_obd) {
 323                         CERROR("target %s not attached\n", tgts->uuid.uuid);
 324                         GOTO(out_disc, rc = -EINVAL);
 325                 }
 326
 327                 /* for MDS: don't connect to yourself */
 328                 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
 329                         CDEBUG(D_OTHER, "don't connect back to %s\n",
 330                                cluuid->uuid);
 331                         tgts->ltd_exp = NULL;
 332                         continue;
 333                 }
 334
 335                 CDEBUG(D_OTHER, "connect to %s(%s) - %s, %s FOR %s\n",
 336                         tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
 337                         tgts->uuid.uuid, obd->obd_uuid.uuid,
 338                         cluuid->uuid);
 339
 340                 if (!tgt_obd->obd_set_up) {
 341                         CERROR("target %s not set up\n", tgts->uuid.uuid);
 342                         GOTO(out_disc, rc = -EINVAL);
 343                 }
 344
 345                 rc = obd_connect(&conn, tgt_obd, &lmv_mdc_uuid, &lmv->conn_data,
 346                                  lmv->connect_flags);
 347                 if (rc) {
 348                         CERROR("target %s connect error %d\n",
 349                                 tgts->uuid.uuid, rc);
 350                         GOTO(out_disc, rc);
 351                 }
 352                 tgts->ltd_exp = class_conn2export(&conn);
 353
 354                 obd_init_ea_size(tgts->ltd_exp, lmv->max_easize,
 355                                  lmv->max_cookiesize);
 356
 357                 rc = obd_register_observer(tgt_obd, obd);
 358                 if (rc) {
 359                         CERROR("target %s register_observer error %d\n",
 360                                tgts->uuid.uuid, rc);
 361                         obd_disconnect(tgts->ltd_exp, 0);
 362                         GOTO(out_disc, rc);
 363                 }
 364
 365                 lmv->desc.ld_active_tgt_count++;
 366                 tgts->active = 1;
 367
 368                 CDEBUG(D_OTHER, "connected to %s(%s) successfully (%d)\n",
 369                         tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
 370                         atomic_read(&obd->obd_refcount));
 371
 372 #ifdef __KERNEL__
 373                 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 374                 if (lmv_proc_dir) {
 375                         struct obd_device *mdc_obd = class_conn2obd(&conn);
 376                         struct proc_dir_entry *mdc_symlink;
 377                         char name[MAX_STRING_SIZE + 1];
 378
 379                         LASSERT(mdc_obd != NULL);
 380                         LASSERT(mdc_obd->obd_type != NULL);
 381                         LASSERT(mdc_obd->obd_type->typ_name != NULL);
 382                         name[MAX_STRING_SIZE] = '\0';
 383                         snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
 384                                  mdc_obd->obd_type->typ_name,
 385                                  mdc_obd->obd_name);
 386                         mdc_symlink = proc_symlink(mdc_obd->obd_name,
 387                                                    lmv_proc_dir, name);
 388                         if (mdc_symlink == NULL) {
 389                                 CERROR("could not register LMV target "
 390                                        "/proc/fs/lustre/%s/%s/target_obds/%s.",
 391                                        obd->obd_type->typ_name, obd->obd_name,
 392                                        mdc_obd->obd_name);
 393                                 lprocfs_remove(lmv_proc_dir);
 394                                 lmv_proc_dir = NULL;
 395                         }
 396                 }
 397 #endif
 398         }
 399
 400         lmv_set_timeouts(obd);
 401         class_export_put(exp);
 402         lmv->connected = 1;
 403         up(&lmv->init_sem);
 404         RETURN(0);
 405
 406  out_disc:
 407         while (i-- > 0) {
 408                 struct obd_uuid uuid;
 409                 --tgts;
 410                 --lmv->desc.ld_active_tgt_count;
 411                 tgts->active = 0;
 412                 /* save for CERROR below; (we know it's terminated) */
 413                 uuid = tgts->uuid;
 414                 rc2 = obd_disconnect(tgts->ltd_exp, 0);
 415                 if (rc2)
 416                         CERROR("error: LMV target %s disconnect on MDC idx %d: "
 417                                "error %d\n", uuid.uuid, i, rc2);
 418         }
 419         class_disconnect(exp, 0);
 420         up(&lmv->init_sem);
 421         return rc;
 422 }
 423
 424 static int lmv_disconnect(struct obd_export *exp, unsigned long flags)
 425 {
 426         struct obd_device *obd = class_exp2obd(exp);
 427         struct lmv_obd *lmv = &obd->u.lmv;
 428
 429 #ifdef __KERNEL__
 430         struct proc_dir_entry *lmv_proc_dir;
 431 #endif
 432         int rc, i;
 433         ENTRY;
 434
 435         if (!lmv->tgts)
 436                 goto out_local;
 437
 438         /* Only disconnect the underlying layers on the final disconnect. */
 439         lmv->refcount--;
 440         if (lmv->refcount != 0)
 441                 goto out_local;
 442
 443 #ifdef __KERNEL__
 444         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 445 #endif
 446
 447         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 448                 struct obd_device *mdc_obd;
 449
 450                 if (lmv->tgts[i].ltd_exp == NULL)
 451                         continue;
 452
 453                 mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
 454
 455                 if (mdc_obd)
 456                         mdc_obd->obd_no_recov = obd->obd_no_recov;
 457
 458 #ifdef __KERNEL__
 459                 if (lmv_proc_dir) {
 460                         struct proc_dir_entry *mdc_symlink;
 461
 462                         mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
 463                         if (mdc_symlink) {
 464                                 lprocfs_remove(mdc_symlink);
 465                         } else {
 466                                 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 467                                        obd->obd_type->typ_name, obd->obd_name,
 468                                        mdc_obd->obd_name);
 469                         }
 470                 }
 471 #endif
 472                 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
 473                         lmv->tgts[i].ltd_exp->exp_obd->obd_name,
 474                         lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
 475
 476                 obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL);
 477                 rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags);
 478                 if (rc) {
 479                         if (lmv->tgts[i].active) {
 480                                 CERROR("Target %s disconnect error %d\n",
 481                                        lmv->tgts[i].uuid.uuid, rc);
 482                         }
 483                         rc = 0;
 484                 }
 485
 486                 lmv_activate_target(lmv, &lmv->tgts[i], 0);
 487                 lmv->tgts[i].ltd_exp = NULL;
 488         }
 489
 490 #ifdef __KERNEL__
 491         if (lmv_proc_dir) {
 492                 lprocfs_remove(lmv_proc_dir);
 493         } else {
 494                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 495                        obd->obd_type->typ_name, obd->obd_name);
 496         }
 497 #endif
 498
 499 out_local:
 500         /* this is the case when no real connection is established by
 501          * lmv_check_connect(). */
 502         if (!lmv->connected)
 503                 class_export_put(exp);
 504         rc = class_disconnect(exp, 0);
 505         if (lmv->refcount == 0)
 506                 lmv->connected = 0;
 507         RETURN(rc);
 508 }
 509
 510 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 511                          int len, void *karg, void *uarg)
 512 {
 513         struct obd_device *obddev = class_exp2obd(exp);
 514         struct lmv_obd *lmv = &obddev->u.lmv;
 515         int i, rc = 0, set = 0;
 516         ENTRY;
 517
 518         if (lmv->desc.ld_tgt_count == 0)
 519                 RETURN(-ENOTTY);
 520
 521         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 522                 int err;
 523
 524                 if (lmv->tgts[i].ltd_exp == NULL)
 525                         continue;
 526
 527                 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
 528                 if (err) {
 529                         if (lmv->tgts[i].active) {
 530                                 CERROR("error: iocontrol MDC %s on MDT"
 531                                        "idx %d: err = %d\n",
 532                                        lmv->tgts[i].uuid.uuid, i, err);
 533                                 if (!rc)
 534                                         rc = err;
 535                         }
 536                 } else
 537                         set = 1;
 538         }
 539         if (!set && !rc)
 540                 rc = -EIO;
 541
 542         RETURN(rc);
 543 }
 544
 545 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
 546 {
 547         int i, rc = 0;
 548         struct lmv_desc *desc;
 549         struct obd_uuid *uuids;
 550         struct lmv_tgt_desc *tgts;
 551         struct obd_device *tgt_obd;
 552         struct lustre_cfg *lcfg = buf;
 553         struct lmv_obd *lmv = &obd->u.lmv;
 554         ENTRY;
 555
 556         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 557                 CERROR("LMV setup requires a descriptor\n");
 558                 RETURN(-EINVAL);
 559         }
 560
 561         if (LUSTRE_CFG_BUFLEN(lcfg, 2) < 1) {
 562                 CERROR("LMV setup requires an MDT UUID list\n");
 563                 RETURN(-EINVAL);
 564         }
 565
 566         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
 567         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 568                 CERROR("descriptor size wrong: %d > %d\n",
 569                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 570                 RETURN(-EINVAL);
 571         }
 572
 573         uuids = (struct obd_uuid *)lustre_cfg_buf(lcfg, 2);
 574         if (sizeof(*uuids) * desc->ld_tgt_count != LUSTRE_CFG_BUFLEN(lcfg, 2)) {
 575                 CERROR("UUID array size wrong: %u * %u != %u\n",
 576                        sizeof(*uuids), desc->ld_tgt_count, LUSTRE_CFG_BUFLEN(lcfg, 2));
 577                 RETURN(-EINVAL);
 578         }
 579
 580         lmv->tgts_size = sizeof(struct lmv_tgt_desc) * desc->ld_tgt_count;
 581         OBD_ALLOC(lmv->tgts, lmv->tgts_size);
 582         if (lmv->tgts == NULL) {
 583                 CERROR("Out of memory\n");
 584                 RETURN(-ENOMEM);
 585         }
 586
 587         lmv->desc = *desc;
 588         spin_lock_init(&lmv->lmv_lock);
 589
 590         for (i = 0, tgts = lmv->tgts; i < desc->ld_tgt_count; i++, tgts++)
 591                 tgts->uuid = uuids[i];
 592
 593         lmv->max_cookiesize = 0;
 594
 595         lmv->max_easize = sizeof(struct lustre_id) *
 596                 desc->ld_tgt_count + sizeof(struct mea);
 597
 598         rc = lmv_setup_mgr(obd);
 599         if (rc) {
 600                 CERROR("Can't setup LMV object manager, "
 601                        "error %d.\n", rc);
 602                 OBD_FREE(lmv->tgts, lmv->tgts_size);
 603         }
 604
 605         tgt_obd = class_find_client_obd(&lmv->tgts->uuid, OBD_MDC_DEVICENAME,
 606                                         &obd->obd_uuid);
 607         if (!tgt_obd) {
 608                 CERROR("Target %s not attached\n", lmv->tgts->uuid.uuid);
 609                 RETURN(-EINVAL);
 610         }
 611
 612         RETURN(rc);
 613 }
 614
 615 static int lmv_cleanup(struct obd_device *obd, int flags)
 616 {
 617         struct lmv_obd *lmv = &obd->u.lmv;
 618         ENTRY;
 619
 620         lmv_cleanup_mgr(obd);
 621         OBD_FREE(lmv->tgts, lmv->tgts_size);
 622
 623         RETURN(0);
 624 }
 625
 626 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 627                       unsigned long max_age)
 628 {
 629         struct lmv_obd *lmv = &obd->u.lmv;
 630         struct obd_statfs *temp;
 631         int rc = 0, i;
 632         ENTRY;
 633
 634         rc = lmv_check_connect(obd);
 635         if (rc)
 636                 RETURN(rc);
 637
 638         OBD_ALLOC(temp, sizeof(*temp));
 639         if (temp == NULL)
 640                 RETURN(-ENOMEM);
 641
 642         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 643                 if (lmv->tgts[i].ltd_exp == NULL)
 644                         continue;
 645
 646                 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp, max_age);
 647                 if (rc) {
 648                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
 649                                lmv->tgts[i].ltd_exp->exp_obd->obd_name,
 650                                rc);
 651                         GOTO(out_free_temp, rc);
 652                 }
 653                 if (i == 0) {
 654                         memcpy(osfs, temp, sizeof(*temp));
 655                 } else {
 656                         osfs->os_bavail += temp->os_bavail;
 657                         osfs->os_blocks += temp->os_blocks;
 658                         osfs->os_ffree += temp->os_ffree;
 659                         osfs->os_files += temp->os_files;
 660                 }
 661         }
 662
 663         EXIT;
 664 out_free_temp:
 665         OBD_FREE(temp, sizeof(*temp));
 666         return rc;
 667 }
 668
 669 static int lmv_getstatus(struct obd_export *exp, struct lustre_id *id)
 670 {
 671         struct obd_device *obd = exp->exp_obd;
 672         struct lmv_obd *lmv = &obd->u.lmv;
 673         int rc;
 674         ENTRY;
 675
 676         rc = lmv_check_connect(obd);
 677         if (rc)
 678                 RETURN(rc);
 679
 680         rc = md_getstatus(lmv->tgts[0].ltd_exp, id);
 681         id_group(id) = 0;
 682
 683         RETURN(rc);
 684 }
 685
 686 static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
 687                        __u64 valid, const char *xattr_name,
 688                        const void *xattr_data, unsigned int xattr_datalen,
 689                        unsigned int ea_size, struct ptlrpc_request **request)
 690 {
 691         struct obd_device *obd = exp->exp_obd;
 692         struct lmv_obd *lmv = &obd->u.lmv;
 693         int rc, i = id_group(id);
 694         struct lmv_obj *obj;
 695         ENTRY;
 696
 697         rc = lmv_check_connect(obd);
 698         if (rc)
 699                 RETURN(rc);
 700
 701         LASSERT(i < lmv->desc.ld_tgt_count);
 702
 703
 704         rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid,
 705                         xattr_name, xattr_data, xattr_datalen,
 706                         ea_size, request);
 707         if (rc)
 708                 RETURN(rc);
 709
 710         obj = lmv_grab_obj(obd, id);
 711
 712         CDEBUG(D_OTHER, "GETATTR for "DLID4" %s\n",
 713                OLID4(id), obj ? "(splitted)" : "");
 714
 715         /*
 716          * if object is splitted, then we loop over all the slaves and gather
 717          * size attribute. In ideal world we would have to gather also mds field
 718          * from all slaves, as object is spread over the cluster and this is
 719          * definitely interesting information and it is not good to loss it,
 720          * but...
 721          */
 722         if (obj) {
 723                 struct mds_body *body;
 724
 725                 if (*request == NULL) {
 726                         lmv_put_obj(obj);
 727                         RETURN(rc);
 728                 }
 729
 730                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
 731                                       sizeof(*body));
 732                 LASSERT(body != NULL);
 733
 734                 lmv_lock_obj(obj);
 735
 736                 for (i = 0; i < obj->objcount; i++) {
 737
 738                         if (lmv->tgts[i].ltd_exp == NULL) {
 739                                 CWARN("%s: NULL export for %d\n",
 740                                       obd->obd_name, i);
 741                                 continue;
 742                         }
 743
 744                         /* skip master obj. */
 745                         if (id_equal_fid(&obj->id, &obj->objs[i].id))
 746                                 continue;
 747
 748                         body->size += obj->objs[i].size;
 749                 }
 750
 751                 lmv_unlock_obj(obj);
 752                 lmv_put_obj(obj);
 753         }
 754
 755         RETURN(rc);
 756 }
 757
 758 static int lmv_access_check(struct obd_export *exp,
 759                             struct lustre_id *id,
 760                             struct ptlrpc_request **request)
 761 {
 762         struct obd_device *obd = exp->exp_obd;
 763         struct lmv_obd *lmv = &obd->u.lmv;
 764         int rc, i = id_group(id);
 765         ENTRY;
 766
 767         rc = lmv_check_connect(obd);
 768         if (rc)
 769                 RETURN(rc);
 770
 771         LASSERT(i < lmv->desc.ld_tgt_count);
 772         rc = md_access_check(lmv->tgts[i].ltd_exp, id, request);
 773         RETURN(rc);
 774 }
 775
 776 static int lmv_change_cbdata(struct obd_export *exp,
 777                              struct lustre_id *id,
 778                              ldlm_iterator_t it,
 779                              void *data)
 780 {
 781         struct obd_device *obd = exp->exp_obd;
 782         struct lmv_obd *lmv = &obd->u.lmv;
 783         int i, rc;
 784         ENTRY;
 785
 786         rc = lmv_check_connect(obd);
 787         if (rc)
 788                 RETURN(rc);
 789
 790         CDEBUG(D_OTHER, "CBDATA for "DLID4"\n", OLID4(id));
 791         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
 792
 793         /* with CMD every object can have two locks in different
 794          * namespaces: lookup lock in space of mds storing direntry
 795          * and update/open lock in space of mds storing inode */
 796         for (i = 0; i < lmv->desc.ld_tgt_count; i++)
 797                 md_change_cbdata(lmv->tgts[i].ltd_exp, id, it, data);
 798
 799         RETURN(0);
 800 }
 801
 802 static int lmv_change_cbdata_name(struct obd_export *exp,
 803                                   struct lustre_id *pid,
 804                                   char *name, int len,
 805                                   struct lustre_id *cid,
 806                                   ldlm_iterator_t it,
 807                                   void *data)
 808 {
 809         struct obd_device *obd = exp->exp_obd;
 810         struct lmv_obd *lmv = &obd->u.lmv;
 811         struct lustre_id rcid = *cid;
 812         struct lmv_obj *obj;
 813         int rc = 0, mds;
 814         ENTRY;
 815
 816         rc = lmv_check_connect(obd);
 817         if (rc)
 818                 RETURN(rc);
 819
 820         LASSERT(id_group(pid) < lmv->desc.ld_tgt_count);
 821         LASSERT(id_group(cid) < lmv->desc.ld_tgt_count);
 822
 823         CDEBUG(D_OTHER, "CBDATA for "DLID4":%*s -> "DLID4"\n",
 824                OLID4(pid), len, name, OLID4(cid));
 825
 826         /* this is default mds for directory name belongs to. */
 827         mds = id_group(pid);
 828         obj = lmv_grab_obj(obd, pid);
 829         if (obj) {
 830                 /* directory is splitted. look for right mds for this name. */
 831                 mds = raw_name2idx(obj->hashtype, obj->objcount, name, len);
 832                 rcid = obj->objs[mds].id;
 833                 mds = id_group(&rcid);
 834                 lmv_put_obj(obj);
 835         }
 836         rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, &rcid, it, data);
 837         RETURN(rc);
 838 }
 839
 840 static int lmv_valid_attrs(struct obd_export *exp, struct lustre_id *id)
 841 {
 842         struct obd_device *obd = exp->exp_obd;
 843         struct lmv_obd *lmv = &obd->u.lmv;
 844         int rc = 0;
 845         ENTRY;
 846
 847         rc = lmv_check_connect(obd);
 848         if (rc)
 849                 RETURN(rc);
 850
 851         CDEBUG(D_OTHER, "validate "DLID4"\n", OLID4(id));
 852         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
 853         rc = md_valid_attrs(lmv->tgts[id_group(id)].ltd_exp, id);
 854         RETURN(rc);
 855 }
 856
 857 static int lmv_close(struct obd_export *exp, struct obdo *obdo,
 858                      struct obd_client_handle *och,
 859                      struct ptlrpc_request **request)
 860 {
 861         struct obd_device *obd = exp->exp_obd;
 862         struct lmv_obd *lmv = &obd->u.lmv;
 863         int rc, i = obdo->o_mds;
 864         ENTRY;
 865
 866         rc = lmv_check_connect(obd);
 867         if (rc)
 868                 RETURN(rc);
 869
 870         LASSERT(i < lmv->desc.ld_tgt_count);
 871         CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long)obdo->o_mds,
 872                (unsigned long)obdo->o_id, (unsigned long)obdo->o_generation);
 873         rc = md_close(lmv->tgts[i].ltd_exp, obdo, och, request);
 874         RETURN(rc);
 875 }
 876
 877 int lmv_get_mea_and_update_object(struct obd_export *exp,
 878                                   struct lustre_id *id)
 879 {
 880         struct obd_device *obd = exp->exp_obd;
 881         struct lmv_obd *lmv = &obd->u.lmv;
 882         struct ptlrpc_request *req = NULL;
 883         struct lmv_obj *obj;
 884         struct lustre_md md;
 885         int mealen, rc;
 886         __u64 valid;
 887         ENTRY;
 888
 889         md.mea = NULL;
 890         mealen = MEA_SIZE_LMV(lmv);
 891
 892         valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA;
 893
 894         /* time to update mea of parent id */
 895         rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp,
 896                         id, valid, NULL, NULL, 0, mealen, &req);
 897         if (rc) {
 898                 CERROR("md_getattr() failed, error %d\n", rc);
 899                 GOTO(cleanup, rc);
 900         }
 901
 902         rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
 903         if (rc) {
 904                 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
 905                 GOTO(cleanup, rc);
 906         }
 907
 908         if (md.mea == NULL)
 909                 GOTO(cleanup, rc = -ENODATA);
 910
 911         obj = lmv_create_obj(exp, id, md.mea);
 912         if (IS_ERR(obj))
 913                 rc = PTR_ERR(obj);
 914         else
 915                 lmv_put_obj(obj);
 916
 917         obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
 918
 919         EXIT;
 920 cleanup:
 921         if (req)
 922                 ptlrpc_req_finished(req);
 923         return rc;
 924 }
 925
 926 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
 927                const void *data, int datalen, int mode, __u32 uid,
 928                __u32 gid, __u64 rdev, struct ptlrpc_request **request)
 929 {
 930         struct obd_device *obd = exp->exp_obd;
 931         struct lmv_obd *lmv = &obd->u.lmv;
 932         struct mds_body *body;
 933         struct lmv_obj *obj;
 934         int rc, mds, loop = 0;
 935         ENTRY;
 936
 937         rc = lmv_check_connect(obd);
 938         if (rc)
 939                 RETURN(rc);
 940
 941         if (!lmv->desc.ld_active_tgt_count)
 942                 RETURN(-EIO);
 943 repeat:
 944         LASSERT(++loop <= 2);
 945         obj = lmv_grab_obj(obd, &op_data->id1);
 946         if (obj) {
 947                 mds = raw_name2idx(obj->hashtype, obj->objcount,
 948                                    op_data->name, op_data->namelen);
 949                 op_data->id1 = obj->objs[mds].id;
 950                 lmv_put_obj(obj);
 951         }
 952
 953         CDEBUG(D_OTHER, "CREATE '%*s' on "DLID4"\n", op_data->namelen,
 954                op_data->name, OLID4(&op_data->id1));
 955
 956         rc = md_create(lmv->tgts[id_group(&op_data->id1)].ltd_exp,
 957                        op_data, data, datalen, mode, uid, gid, rdev,
 958                        request);
 959         if (rc == 0) {
 960                 if (*request == NULL)
 961                         RETURN(rc);
 962
 963                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
 964                                       sizeof(*body));
 965                 if (body == NULL)
 966                         RETURN(-ENOMEM);
 967
 968                 CDEBUG(D_OTHER, "created. "DLID4"\n", OLID4(&op_data->id1));
 969         } else if (rc == -ERESTART) {
 970                 /*
 971                  * directory got splitted. time to update local object and
 972                  * repeat the request with proper MDS.
 973                  */
 974                 rc = lmv_get_mea_and_update_object(exp, &op_data->id1);
 975                 if (rc == 0) {
 976                         ptlrpc_req_finished(*request);
 977                         goto repeat;
 978                 }
 979         }
 980         RETURN(rc);
 981 }
 982
 983 static int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
 984 {
 985         struct obd_device *obd = exp->exp_obd;
 986         struct lmv_obd *lmv = &obd->u.lmv;
 987         int rc;
 988         ENTRY;
 989
 990         rc = lmv_check_connect(obd);
 991         if (rc)
 992                 RETURN(rc);
 993
 994         /* FIXME: choose right MDC here */
 995         CWARN("this method isn't implemented yet\n");
 996         rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
 997         RETURN(rc);
 998 }
 999
1000 static int
1001 lmv_enqueue_slaves(struct obd_export *exp, int locktype,
1002                    struct lookup_intent *it, int lockmode,
1003                    struct mdc_op_data *data, struct lustre_handle *lockh,
1004                    void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1005                    ldlm_blocking_callback cb_blocking, void *cb_data)
1006 {
1007         struct obd_device *obd = exp->exp_obd;
1008         struct lmv_obd *lmv = &obd->u.lmv;
1009         struct mea *mea = data->mea1;
1010         struct mdc_op_data *data2;
1011         int i, rc, mds;
1012         ENTRY;
1013
1014         OBD_ALLOC(data2, sizeof(*data2));
1015         if (data2 == NULL)
1016                 RETURN(-ENOMEM);
1017
1018         LASSERT(mea != NULL);
1019         for (i = 0; i < mea->mea_count; i++) {
1020                 memset(data2, 0, sizeof(*data2));
1021                 data2->id1 = mea->mea_ids[i];
1022                 mds = id_group(&data2->id1);
1023
1024                 if (lmv->tgts[mds].ltd_exp == NULL)
1025                         continue;
1026
1027                 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it,
1028                                 lockmode, data2, lockh + i, lmm, lmmsize,
1029                                 cb_compl, cb_blocking, cb_data);
1030
1031                 CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n",
1032                        OLID4(&mea->mea_ids[i]), rc, LUSTRE_IT(it)->it_status);
1033                 if (rc)
1034                         GOTO(cleanup, rc);
1035                 if (LUSTRE_IT(it)->it_data) {
1036                         struct ptlrpc_request *req;
1037                         req = (struct ptlrpc_request *) LUSTRE_IT(it)->it_data;
1038                         ptlrpc_req_finished(req);
1039                 }
1040
1041                 if (LUSTRE_IT(it)->it_status)
1042                         GOTO(cleanup, rc = LUSTRE_IT(it)->it_status);
1043         }
1044
1045         OBD_FREE(data2, sizeof(*data2));
1046         RETURN(0);
1047 cleanup:
1048         OBD_FREE(data2, sizeof(*data2));
1049
1050         /* drop all taken locks */
1051         while (--i >= 0) {
1052                 if (lockh[i].cookie)
1053                         ldlm_lock_decref(lockh + i, lockmode);
1054                 lockh[i].cookie = 0;
1055         }
1056         return rc;
1057 }
1058
1059 static int
1060 lmv_enqueue_remote(struct obd_export *exp, int lock_type,
1061                    struct lookup_intent *it, int lock_mode,
1062                    struct mdc_op_data *data, struct lustre_handle *lockh,
1063                    void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1064                    ldlm_blocking_callback cb_blocking, void *cb_data)
1065 {
1066         struct ptlrpc_request *req = LUSTRE_IT(it)->it_data;
1067         struct obd_device *obd = exp->exp_obd;
1068         struct lmv_obd *lmv = &obd->u.lmv;
1069         struct lustre_handle plock;
1070         struct mdc_op_data rdata;
1071         struct mds_body *body = NULL;
1072         int rc = 0, pmode;
1073         ENTRY;
1074
1075         body = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body));
1076         LASSERT(body != NULL);
1077
1078         if (!(body->valid & OBD_MD_MDS))
1079                 RETURN(0);
1080
1081         CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4" -> "DLID4"\n",
1082                LL_IT2STR(it), OLID4(&data->id1), OLID4(&body->id1));
1083
1084         /* we got LOOKUP lock, but we really need attrs */
1085         pmode = LUSTRE_IT(it)->it_lock_mode;
1086         LASSERT(pmode != 0);
1087         memcpy(&plock, lockh, sizeof(plock));
1088         LUSTRE_IT(it)->it_lock_mode = 0;
1089         LUSTRE_IT(it)->it_data = NULL;
1090         LASSERT((body->valid & OBD_MD_FID) != 0);
1091
1092         memcpy(&rdata, data, sizeof(rdata));
1093         rdata.id1 = body->id1;
1094         rdata.name = NULL;
1095         rdata.namelen = 0;
1096
1097         LUSTRE_IT(it)->it_disposition &= ~DISP_ENQ_COMPLETE;
1098         ptlrpc_req_finished(req);
1099
1100         rc = md_enqueue(lmv->tgts[id_group(&rdata.id1)].ltd_exp,
1101                         lock_type, it, lock_mode, &rdata, lockh, lmm,
1102                         lmmsize, cb_compl, cb_blocking, cb_data);
1103         ldlm_lock_decref(&plock, pmode);
1104         RETURN(rc);
1105 }
1106
1107 static int
1108 lmv_enqueue(struct obd_export *exp, int lock_type,
1109             struct lookup_intent *it, int lock_mode,
1110             struct mdc_op_data *data, struct lustre_handle *lockh,
1111             void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1112             ldlm_blocking_callback cb_blocking, void *cb_data)
1113 {
1114         struct obd_device *obd = exp->exp_obd;
1115         struct lmv_obd *lmv = &obd->u.lmv;
1116         struct lmv_obj *obj;
1117         int rc, mds;
1118         ENTRY;
1119
1120         rc = lmv_check_connect(obd);
1121         if (rc)
1122                 RETURN(rc);
1123
1124         if (data->mea1 && it->it_op == IT_UNLINK) {
1125                 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
1126                                         data, lockh, lmm, lmmsize,
1127                                         cb_compl, cb_blocking, cb_data);
1128                 RETURN(rc);
1129         }
1130
1131         if (data->namelen) {
1132                 obj = lmv_grab_obj(obd, &data->id1);
1133                 if (obj) {
1134                         /* directory is splitted. look for right mds for this
1135                          * name */
1136                         mds = raw_name2idx(obj->hashtype, obj->objcount,
1137                                            (char *)data->name, data->namelen);
1138                         data->id1 = obj->objs[mds].id;
1139                         lmv_put_obj(obj);
1140                 }
1141         }
1142         CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4"\n", LL_IT2STR(it),
1143                OLID4(&data->id1));
1144
1145         rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp,
1146                         lock_type, it, lock_mode, data, lockh, lmm,
1147                         lmmsize, cb_compl, cb_blocking, cb_data);
1148         if (rc == 0 && it->it_op == IT_OPEN)
1149                 rc = lmv_enqueue_remote(exp, lock_type, it, lock_mode,
1150                                         data, lockh, lmm, lmmsize,
1151                                         cb_compl, cb_blocking, cb_data);
1152         RETURN(rc);
1153 }
1154
1155 static int
1156 lmv_getattr_lock(struct obd_export *exp, struct lustre_id *id,
1157                  char *filename, int namelen, __u64 valid,
1158                  unsigned int ea_size, struct ptlrpc_request **request)
1159 {
1160         int rc, mds = id_group(id), loop = 0;
1161         struct obd_device *obd = exp->exp_obd;
1162         struct lmv_obd *lmv = &obd->u.lmv;
1163         struct lustre_id rid = *id;
1164         struct mds_body *body;
1165         struct lmv_obj *obj;
1166         ENTRY;
1167
1168         rc = lmv_check_connect(obd);
1169         if (rc)
1170                 RETURN(rc);
1171 repeat:
1172         LASSERT(++loop <= 2);
1173         obj = lmv_grab_obj(obd, id);
1174         if (obj) {
1175                 /* directory is splitted. look for right mds for this name */
1176                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1177                                    filename, namelen - 1);
1178                 rid = obj->objs[mds].id;
1179                 lmv_put_obj(obj);
1180         }
1181
1182         CDEBUG(D_OTHER, "getattr_lock for %*s on "DLID4" -> "DLID4"\n",
1183                namelen, filename, OLID4(id), OLID4(&rid));
1184
1185         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1186                              &rid, filename, namelen,
1187                              valid == OBD_MD_FLID ? valid : valid | OBD_MD_FID,
1188                              ea_size, request);
1189         if (rc == 0) {
1190                 /*
1191                  * this could be cross-node reference. in this case all we have
1192                  * right now is lustre_id triple. we'd like to find other
1193                  * attributes.
1194                  */
1195                 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
1196                 LASSERT(body != NULL);
1197                 LASSERT((body->valid & OBD_MD_FID) != 0
1198                                 || body->valid == OBD_MD_FLID);
1199
1200                 if (body->valid & OBD_MD_MDS) {
1201                         struct ptlrpc_request *req = NULL;
1202
1203                         rid = body->id1;
1204                         CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid));
1205
1206                         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1207                                              &rid, NULL, 1, valid, ea_size, &req);
1208                         ptlrpc_req_finished(*request);
1209                         *request = req;
1210                 }
1211         } else if (rc == -ERESTART) {
1212                 /* directory got splitted. time to update local object and
1213                  * repeat the request with proper MDS */
1214                 rc = lmv_get_mea_and_update_object(exp, &rid);
1215                 if (rc == 0) {
1216                         ptlrpc_req_finished(*request);
1217                         goto repeat;
1218                 }
1219         }
1220         RETURN(rc);
1221 }
1222
1223 /*
1224  * llite passes id of an target inode in data->id1 and id of directory in
1225  * data->id2
1226  */
1227 static int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
1228                     struct ptlrpc_request **request)
1229 {
1230         struct obd_device *obd = exp->exp_obd;
1231         struct lmv_obd *lmv = &obd->u.lmv;
1232         struct lmv_obj *obj;
1233         int rc, mds;
1234         ENTRY;
1235
1236         rc = lmv_check_connect(obd);
1237         if (rc)
1238                 RETURN(rc);
1239
1240         if (data->namelen != 0) {
1241                 /* usual link request */
1242                 obj = lmv_grab_obj(obd, &data->id2);
1243                 if (obj) {
1244                         rc = raw_name2idx(obj->hashtype, obj->objcount,
1245                                           data->name, data->namelen);
1246                         data->id2 = obj->objs[rc].id;
1247                         lmv_put_obj(obj);
1248                 }
1249
1250                 mds = id_group(&data->id2);
1251
1252                 CDEBUG(D_OTHER,"link "DLID4":%*s to "DLID4"\n",
1253                        OLID4(&data->id2), data->namelen, data->name,
1254                        OLID4(&data->id1));
1255         } else {
1256                 mds = id_group(&data->id1);
1257
1258                 /* request from MDS to acquire i_links for inode by id1 */
1259                 CDEBUG(D_OTHER, "inc i_nlinks for "DLID4"\n",
1260                        OLID4(&data->id1));
1261         }
1262
1263         CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n",
1264                mds, OLID4(&data->id1));
1265         rc = md_link(lmv->tgts[mds].ltd_exp, data, request);
1266
1267         RETURN(rc);
1268 }
1269
1270 static int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1271                       const char *old, int oldlen, const char *new, int newlen,
1272                       struct ptlrpc_request **request)
1273 {
1274         struct obd_device *obd = exp->exp_obd;
1275         struct lmv_obd *lmv = &obd->u.lmv;
1276         struct lmv_obj *obj;
1277         int rc, mds;
1278         ENTRY;
1279
1280         CDEBUG(D_OTHER, "rename %*s in "DLID4" to %*s in "DLID4"\n",
1281                oldlen, old, OLID4(&data->id1), newlen, new,
1282                OLID4(&data->id2));
1283
1284         rc = lmv_check_connect(obd);
1285         if (rc)
1286                 RETURN(rc);
1287
1288         if (oldlen == 0) {
1289                 /*
1290                  * MDS with old dir entry is asking another MDS to create name
1291                  * there.
1292                  */
1293                 CDEBUG(D_OTHER,
1294                        "create %*s(%d/%d) in "DLID4" pointing "
1295                        "to "DLID4"\n", newlen, new, oldlen, newlen,
1296                        OLID4(&data->id2), OLID4(&data->id1));
1297
1298                 mds = id_group(&data->id2);
1299
1300                 /*
1301                  * target directory can be splitted, sowe should forward request
1302                  * to the right MDS.
1303                  */
1304                 obj = lmv_grab_obj(obd, &data->id2);
1305                 if (obj) {
1306                         mds = raw_name2idx(obj->hashtype, obj->objcount,
1307                                            (char *)new, newlen);
1308                         data->id2 = obj->objs[mds].id;
1309                         CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1310                                OLID4(&data->id2));
1311                         lmv_put_obj(obj);
1312                 }
1313                 goto request;
1314         }
1315
1316         obj = lmv_grab_obj(obd, &data->id1);
1317         if (obj) {
1318                 /*
1319                  * directory is already splitted, so we have to forward request
1320                  * to the right MDS.
1321                  */
1322                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1323                                    (char *)old, oldlen);
1324                 data->id1 = obj->objs[mds].id;
1325                 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1326                        OLID4(&data->id1));
1327                 lmv_put_obj(obj);
1328         }
1329
1330         obj = lmv_grab_obj(obd, &data->id2);
1331         if (obj) {
1332                 /*
1333                  * directory is already splitted, so we have to forward request
1334                  * to the right MDS.
1335                  */
1336                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1337                                    (char *)new, newlen);
1338
1339                 data->id2 = obj->objs[mds].id;
1340                 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1341                        OLID4(&data->id2));
1342                 lmv_put_obj(obj);
1343         }
1344
1345         mds = id_group(&data->id1);
1346
1347 request:
1348         if (id_group(&data->id1) != id_group(&data->id2)) {
1349                 CDEBUG(D_OTHER,"cross-node rename "DLID4"/%*s to "DLID4"/%*s\n",
1350                        OLID4(&data->id1), oldlen, old, OLID4(&data->id2),
1351                        newlen, new);
1352         }
1353
1354         rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1355                        new, newlen, request);
1356         RETURN(rc);
1357 }
1358
1359 static int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1360                        struct iattr *iattr, void *ea, int ealen, void *ea2,
1361                        int ea2len, void *ea3, int ea3len,
1362                        struct ptlrpc_request **request)
1363 {
1364         struct obd_device *obd = exp->exp_obd;
1365         struct lmv_obd *lmv = &obd->u.lmv;
1366         struct ptlrpc_request *req;
1367         struct mds_body *body;
1368         struct lmv_obj *obj;
1369         int rc = 0, i;
1370         ENTRY;
1371
1372         rc = lmv_check_connect(obd);
1373         if (rc)
1374                 RETURN(rc);
1375
1376         obj = lmv_grab_obj(obd, &data->id1);
1377
1378         CDEBUG(D_OTHER, "SETATTR for "DLID4", valid 0x%x%s\n",
1379                OLID4(&data->id1), iattr->ia_valid, obj ? ", splitted" : "");
1380
1381         if (obj) {
1382                 for (i = 0; i < obj->objcount; i++) {
1383                         data->id1 = obj->objs[i].id;
1384
1385                         rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1386                                         data, iattr, ea, ealen, ea2, ea2len,
1387                                         ea3, ea3len, &req);
1388
1389                         if (id_equal_fid(&obj->id, &obj->objs[i].id)) {
1390                                 /*
1391                                  * this is master object and this request should
1392                                  * be returned back to llite.
1393                                  */
1394                                 *request = req;
1395                         } else {
1396                                 ptlrpc_req_finished(req);
1397                         }
1398
1399                         if (rc)
1400                                 break;
1401                 }
1402                 lmv_put_obj(obj);
1403         } else {
1404                 LASSERT(id_group(&data->id1) < lmv->desc.ld_tgt_count);
1405                 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1406                                 data, iattr, ea, ealen, ea2, ea2len, ea3,
1407                                 ea3len, request);
1408                 if (rc == 0) {
1409                         body = lustre_msg_buf((*request)->rq_repmsg, 0,
1410                                               sizeof(*body));
1411                         LASSERT(body != NULL);
1412                         LASSERT((body->valid & OBD_MD_FID) != 0);
1413                         LASSERT(id_group(&body->id1) == id_group(&data->id1));
1414                 }
1415         }
1416         RETURN(rc);
1417 }
1418
1419 static int lmv_sync(struct obd_export *exp, struct lustre_id *id,
1420                     struct ptlrpc_request **request)
1421 {
1422         struct obd_device *obd = exp->exp_obd;
1423         struct lmv_obd *lmv = &obd->u.lmv;
1424         int rc;
1425         ENTRY;
1426
1427         rc = lmv_check_connect(obd);
1428         if (rc)
1429                 RETURN(rc);
1430
1431         rc = md_sync(lmv->tgts[id_group(id)].ltd_exp,
1432                      id, request);
1433         RETURN(rc);
1434 }
1435
1436 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock,
1437                             struct ldlm_lock_desc *desc,
1438                             void *data, int flag)
1439 {
1440         struct lustre_handle lockh;
1441         struct lmv_obj *obj;
1442         int rc;
1443         ENTRY;
1444
1445         switch (flag) {
1446         case LDLM_CB_BLOCKING:
1447                 ldlm_lock2handle(lock, &lockh);
1448                 rc = ldlm_cli_cancel(&lockh);
1449                 if (rc < 0) {
1450                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1451                         RETURN(rc);
1452                 }
1453                 break;
1454         case LDLM_CB_CANCELING:
1455                 /* time to drop cached attrs for dirobj */
1456                 obj = lock->l_ast_data;
1457                 if (obj) {
1458                         CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
1459                                ", master "DLID4"\n",
1460                                lock->l_resource->lr_name.name[3] == 1 ?
1461                                "LOOKUP" : "UPDATE",
1462                                lock->l_resource->lr_name.name[0],
1463                                lock->l_resource->lr_name.name[1],
1464                                OLID4(&obj->id));
1465                         lmv_put_obj(obj);
1466                 }
1467                 break;
1468         default:
1469                 LBUG();
1470         }
1471         RETURN(0);
1472 }
1473
1474 static void lmv_remove_dots(struct page *page)
1475 {
1476         unsigned limit = PAGE_CACHE_SIZE;
1477         char *kaddr = page_address(page);
1478         struct ext2_dir_entry_2 *p;
1479         unsigned offs, rec_len;
1480
1481         for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1482                 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1483                 rec_len = le16_to_cpu(p->rec_len);
1484
1485                 if ((p->name_len == 1 && p->name[0] == '.') ||
1486                     (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1487                         p->inode = 0;
1488         }
1489 }
1490
1491 static int lmv_readpage(struct obd_export *exp, struct lustre_id *id,
1492                         __u64 offset, struct page *page,
1493                         struct ptlrpc_request **request)
1494 {
1495         struct obd_device *obd = exp->exp_obd;
1496         struct lmv_obd *lmv = &obd->u.lmv;
1497         struct lustre_id rid = *id;
1498         struct lmv_obj *obj;
1499         int rc, i;
1500         ENTRY;
1501
1502 #warning "we need well-desgined readdir() implementation"
1503         rc = lmv_check_connect(obd);
1504         if (rc)
1505                 RETURN(rc);
1506
1507         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
1508         CDEBUG(D_OTHER, "READPAGE at %llu from "DLID4"\n",
1509                offset, OLID4(&rid));
1510
1511         obj = lmv_grab_obj(obd, id);
1512         if (obj) {
1513                 lmv_lock_obj(obj);
1514
1515                 /* find dirobj containing page with requested offset. */
1516                 for (i = 0; i < obj->objcount; i++) {
1517                         if (offset < obj->objs[i].size)
1518                                 break;
1519                         offset -= obj->objs[i].size;
1520                 }
1521                 rid = obj->objs[i].id;
1522
1523                 lmv_unlock_obj(obj);
1524                 lmv_put_obj(obj);
1525
1526                 CDEBUG(D_OTHER, "forward to "DLID4" with offset %lu\n",
1527                        OLID4(&rid), (unsigned long)offset);
1528         }
1529         rc = md_readpage(lmv->tgts[id_group(&rid)].ltd_exp, &rid,
1530                          offset, page, request);
1531
1532         if (rc == 0 && !id_equal_fid(&rid, id))
1533                 /* this page isn't from master object. To avoid "." and ".."
1534                  * duplication in directory, we have to remove them from all
1535                  * slave objects */
1536                 lmv_remove_dots(page);
1537
1538         RETURN(rc);
1539 }
1540
1541 static int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1542                              struct ptlrpc_request **req)
1543 {
1544         struct obd_device *obd = exp->exp_obd;
1545         struct lmv_obd *lmv = &obd->u.lmv;
1546         struct mea *mea = data->mea1;
1547         struct mdc_op_data *data2;
1548         int i, rc = 0;
1549         ENTRY;
1550
1551         OBD_ALLOC(data2, sizeof(*data2));
1552         if (data2 == NULL)
1553                 RETURN(-ENOMEM);
1554
1555         LASSERT(mea != NULL);
1556         for (i = 0; i < mea->mea_count; i++) {
1557                 memset(data2, 0, sizeof(*data2));
1558                 data2->id1 = mea->mea_ids[i];
1559                 data2->create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1560
1561                 if (lmv->tgts[id_group(&data2->id1)].ltd_exp == NULL)
1562                         continue;
1563
1564                 rc = md_unlink(lmv->tgts[id_group(&data2->id1)].ltd_exp,
1565                                data2, req);
1566
1567                 CDEBUG(D_OTHER, "unlink slave "DLID4" -> %d\n",
1568                        OLID4(&mea->mea_ids[i]), rc);
1569
1570                 if (*req) {
1571                         ptlrpc_req_finished(*req);
1572                         *req = NULL;
1573                 }
1574                 if (rc)
1575                         RETURN(rc);
1576         }
1577         OBD_FREE(data2, sizeof(*data2));
1578         RETURN(rc);
1579 }
1580
1581 static int lmv_delete_inode(struct obd_export *exp, struct lustre_id *id)
1582 {
1583         ENTRY;
1584
1585         LASSERT(exp && id);
1586         if (lmv_delete_obj(exp, id)) {
1587                 CDEBUG(D_OTHER, "lmv object "DLID4" is destroyed.\n",
1588                        OLID4(id));
1589         }
1590         RETURN(0);
1591 }
1592
1593 static int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1594                       struct ptlrpc_request **request)
1595 {
1596         struct obd_device *obd = exp->exp_obd;
1597         struct lmv_obd *lmv = &obd->u.lmv;
1598         int rc, i = 0;
1599         ENTRY;
1600
1601         rc = lmv_check_connect(obd);
1602         if (rc)
1603                 RETURN(rc);
1604
1605         if (data->namelen == 0 && data->mea1 != NULL) {
1606                 /* mds asks to remove slave objects */
1607                 rc = lmv_unlink_slaves(exp, data, request);
1608                 RETURN(rc);
1609         }
1610
1611         if (data->namelen != 0) {
1612                 struct lmv_obj *obj;
1613
1614                 obj = lmv_grab_obj(obd, &data->id1);
1615                 if (obj) {
1616                         i = raw_name2idx(obj->hashtype, obj->objcount,
1617                                          data->name, data->namelen);
1618                         data->id1 = obj->objs[i].id;
1619                         lmv_put_obj(obj);
1620                 }
1621                 CDEBUG(D_OTHER, "unlink '%*s' in "DLID4" -> %u\n",
1622                        data->namelen, data->name, OLID4(&data->id1),
1623                        i);
1624         } else {
1625                 CDEBUG(D_OTHER, "drop i_nlink on "DLID4"\n",
1626                        OLID4(&data->id1));
1627         }
1628         rc = md_unlink(lmv->tgts[id_group(&data->id1)].ltd_exp,
1629                        data, request);
1630         RETURN(rc);
1631 }
1632
1633 static struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1634                                            struct lustre_id *id)
1635 {
1636         struct obd_device *obd = exp->exp_obd;
1637         struct lmv_obd *lmv = &obd->u.lmv;
1638         int rc;
1639         ENTRY;
1640
1641         rc = lmv_check_connect(obd);
1642         if (rc)
1643                 RETURN(ERR_PTR(rc));
1644         obd = lmv->tgts[id_group(id)].ltd_exp->exp_obd;
1645         EXIT;
1646
1647         return obd;
1648 }
1649
1650 static int lmv_init_ea_size(struct obd_export *exp, int easize,
1651                             int cookiesize)
1652 {
1653         struct obd_device *obd = exp->exp_obd;
1654         struct lmv_obd *lmv = &obd->u.lmv;
1655         int i, rc = 0, change = 0;
1656         ENTRY;
1657
1658         if (lmv->max_easize < easize) {
1659                 lmv->max_easize = easize;
1660                 change = 1;
1661         }
1662         if (lmv->max_cookiesize < cookiesize) {
1663                 lmv->max_cookiesize = cookiesize;
1664                 change = 1;
1665         }
1666         if (change == 0)
1667                 RETURN(0);
1668
1669         if (lmv->connected == 0)
1670                 RETURN(0);
1671
1672         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1673                 if (lmv->tgts[i].ltd_exp == NULL) {
1674                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
1675                         continue;
1676                 }
1677
1678                 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
1679                 if (rc) {
1680                         CERROR("obd_init_ea_size() failed on MDT target %d, "
1681                                "error %d.\n", i, rc);
1682                         break;
1683                 }
1684         }
1685         RETURN(rc);
1686 }
1687
1688 static int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1689                                  void *acl, int acl_size,
1690                                  struct lov_stripe_md **ea,
1691                                  struct obd_trans_info *oti)
1692 {
1693         struct obd_device *obd = exp->exp_obd;
1694         struct lmv_obd *lmv = &obd->u.lmv;
1695         struct lov_stripe_md obj_md;
1696         struct lov_stripe_md *obj_mdp = &obj_md;
1697         int rc = 0;
1698         ENTRY;
1699
1700         LASSERT(ea == NULL);
1701         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1702
1703         rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa,
1704                         acl, acl_size, &obj_mdp, oti);
1705
1706         RETURN(rc);
1707 }
1708
1709 /*
1710  * to be called from MDS only. @oa should have correct store cookie and o_fid
1711  * values for "master" object, as it will be used.
1712  */
1713 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
1714                    void *acl, int acl_size,
1715                    struct lov_stripe_md **ea, struct obd_trans_info *oti)
1716 {
1717         struct obd_device *obd = exp->exp_obd;
1718         struct lmv_obd *lmv = &obd->u.lmv;
1719         struct lustre_id mid;
1720         int i, c, rc = 0;
1721         struct mea *mea;
1722         ENTRY;
1723
1724         rc = lmv_check_connect(obd);
1725         if (rc)
1726                 RETURN(rc);
1727
1728         LASSERT(oa != NULL);
1729
1730         if (ea == NULL) {
1731                 rc = lmv_obd_create_single(exp, oa, acl, acl_size, NULL, oti);
1732                 if (rc)
1733                         CERROR("Can't create object, rc = %d\n", rc);
1734                 RETURN(rc);
1735         }
1736
1737         /* acl is only suppied when mds create single remote obj */
1738         LASSERT(acl == NULL && acl_size == 0);
1739
1740         if (*ea == NULL) {
1741                 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1742                 if (rc < 0) {
1743                         CERROR("obd_alloc_diskmd() failed, error %d\n",
1744                                rc);
1745                         RETURN(rc);
1746                 } else
1747                         rc = 0;
1748
1749                 if (*ea == NULL)
1750                         RETURN(-ENOMEM);
1751         }
1752
1753         /*
1754          * here we should take care about splitted dir, so store cookie and fid
1755          * for "master" object should already be allocated and passed in @oa.
1756          */
1757         LASSERT(oa->o_id != 0);
1758         LASSERT(oa->o_fid != 0);
1759
1760         /* save "master" object id */
1761         obdo2id(&mid, oa);
1762
1763         mea = (struct mea *)*ea;
1764         mea->mea_master = -1;
1765         mea->mea_magic = MEA_MAGIC_ALL_CHARS;
1766
1767         if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1768                 mea->mea_count = lmv->desc.ld_tgt_count;
1769
1770         for (i = 0, c = 0; c < mea->mea_count && i < lmv->desc.ld_tgt_count; i++) {
1771                 struct lov_stripe_md obj_md;
1772                 struct lov_stripe_md *obj_mdp = &obj_md;
1773
1774                 if (lmv->tgts[i].ltd_exp == NULL) {
1775                         /* this is "master" MDS */
1776                         mea->mea_master = i;
1777                         mea->mea_ids[c] = mid;
1778                         c++;
1779                         continue;
1780                 }
1781
1782                 /*
1783                  * "master" MDS should always be part of stripped dir,
1784                  * so scan for it.
1785                  */
1786                 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1787                         continue;
1788
1789                 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE |
1790                         OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1791
1792                 rc = obd_create(lmv->tgts[c].ltd_exp, oa, NULL, 0,
1793                                 &obj_mdp, oti);
1794                 if (rc) {
1795                         CERROR("obd_create() failed on MDT target %d, "
1796                                "error %d\n", c, rc);
1797                         RETURN(rc);
1798                 }
1799
1800                 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1801                        i, oa->o_id, oa->o_generation);
1802
1803
1804                 /*
1805                  * here, when object is created (or it is master and was passed
1806                  * from caller) on desired MDS we save its fid to local mea_ids.
1807                  */
1808                 LASSERT(oa->o_fid);
1809
1810                 /*
1811                  * store cookie should be defined here for both cases (master
1812                  * object and not master), because master is already created.
1813                  */
1814                 LASSERT(oa->o_id);
1815
1816                 /* fill mea by store cookie and fid */
1817                 obdo2id(&mea->mea_ids[c], oa);
1818                 c++;
1819         }
1820         LASSERT(c == mea->mea_count);
1821
1822         CDEBUG(D_OTHER, "%d dirobjects created\n",
1823                (int)mea->mea_count);
1824
1825         RETURN(rc);
1826 }
1827
1828 static int lmv_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1829                          struct obd_device *tgt, int count,
1830                          struct llog_catid *logid)
1831 {
1832         struct llog_ctxt *ctxt;
1833         int rc;
1834         ENTRY;
1835
1836         rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1837                             &llog_client_ops);
1838         if (rc == 0) {
1839                 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1840                 ctxt->loc_imp = tgt->u.cli.cl_import;
1841         }
1842
1843         RETURN(rc);
1844 }
1845
1846 static int lmv_llog_finish(struct obd_device *obd,
1847                            struct obd_llogs *llogs, int count)
1848 {
1849         int rc;
1850         ENTRY;
1851
1852         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1853         RETURN(rc);
1854 }
1855
1856 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1857                         void *key, __u32 *vallen, void *val)
1858 {
1859         struct obd_device *obd;
1860         struct lmv_obd *lmv;
1861         int rc = 0;
1862         ENTRY;
1863
1864         obd = class_exp2obd(exp);
1865         if (obd == NULL) {
1866                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1867                        exp->exp_handle.h_cookie);
1868                 RETURN(-EINVAL);
1869         }
1870
1871         lmv = &obd->u.lmv;
1872         if (keylen == strlen("mdsize") && !strcmp(key, "mdsize")) {
1873                 __u32 *mdsize = val;
1874                 *vallen = sizeof(__u32);
1875                 *mdsize = sizeof(struct lustre_id) * lmv->desc.ld_tgt_count
1876                         + sizeof(struct mea);
1877                 RETURN(0);
1878         } else if (keylen == strlen("mdsnum") && !strcmp(key, "mdsnum")) {
1879                 struct obd_uuid *cluuid = &lmv->cluuid;
1880                 struct lmv_tgt_desc *tgts;
1881                 __u32 *mdsnum = val;
1882                 int i;
1883
1884                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1885                         if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1886                                 *vallen = sizeof(__u32);
1887                                 *mdsnum = i;
1888                                 RETURN(0);
1889                         }
1890                 }
1891                 LASSERT(0);
1892         } else if (keylen == strlen("rootid") && !strcmp(key, "rootid")) {
1893                 rc = lmv_check_connect(obd);
1894                 if (rc)
1895                         RETURN(rc);
1896
1897                 /* getting rootid from first MDS. */
1898                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
1899                                   vallen, val);
1900                 RETURN(rc);
1901         } else if (keylen >= strlen("lmvdesc") && !strcmp(key, "lmvdesc")) {
1902                 struct lmv_desc *desc_ret = val;
1903                 *desc_ret = lmv->desc;
1904                 RETURN(0);
1905         } else if (keylen >= strlen("remote_flag") && !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")) {
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         } else if (keylen >= strlen("getext") && !strcmp(key, "getext")) {
1939                 struct lmv_tgt_desc *tgts;
1940                 int i;
1941
1942                 rc = lmv_check_connect(obd);
1943                 if (rc)
1944                         RETURN(rc);
1945
1946                 LASSERT(*vallen == sizeof(struct fid_extent));
1947                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
1948                      i++, tgts++) {
1949
1950                         /* all tgts should be connected when this get called. */
1951                         if (!tgts || !tgts->ltd_exp) {
1952                                 CERROR("target not setup?\n");
1953                                 continue;
1954                         }
1955
1956                         rc = obd_get_info(tgts->ltd_exp, keylen, key,
1957                                           vallen, val);
1958                         if (rc)
1959                                 RETURN(rc);
1960                 }
1961                 RETURN(0);
1962         }
1963
1964         CDEBUG(D_IOCTL, "invalid key\n");
1965         RETURN(-EINVAL);
1966 }
1967
1968 int lmv_set_info(struct obd_export *exp, obd_count keylen,
1969                  void *key, obd_count vallen, void *val)
1970 {
1971         struct lmv_tgt_desc    *tgt;
1972         struct obd_device      *obd;
1973         struct lmv_obd         *lmv;
1974         ENTRY;
1975
1976         obd = class_exp2obd(exp);
1977         if (obd == NULL) {
1978                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1979                        exp->exp_handle.h_cookie);
1980                 RETURN(-EINVAL);
1981         }
1982         lmv = &obd->u.lmv;
1983
1984         if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
1985                 lmv->server_timeout = 1;
1986                 lmv_set_timeouts(obd);
1987                 RETURN(0);
1988         }
1989
1990         /* maybe this could be default */
1991         if ((keylen == strlen("sec") && strcmp(key, "sec") == 0) ||
1992             (keylen == strlen("sec_flags") && strcmp(key, "sec_flags") == 0) ||
1993             (keylen == strlen("nllu") && strcmp(key, "nllu") == 0)) {
1994                 struct obd_export *exp;
1995                 int rc = 0, err, i;
1996
1997                 spin_lock(&lmv->lmv_lock);
1998                 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count;
1999                      i++, tgt++) {
2000                         exp = tgt->ltd_exp;
2001                         /* during setup time the connections to mdc might
2002                          * haven't been established.
2003                          */
2004                         if (exp == NULL) {
2005                                 struct obd_device *tgt_obd;
2006
2007                                 tgt_obd = class_find_client_obd(&tgt->uuid,
2008                                                                 OBD_MDC_DEVICENAME,
2009                                                                 &obd->obd_uuid);
2010                                 if (!tgt_obd) {
2011                                         CERROR("can't set info %s, "
2012                                                "device %s not attached?\n",
2013                                                 (char *) key, tgt->uuid.uuid);
2014                                         rc = -EINVAL;
2015                                         continue;
2016                                 }
2017                                 exp = tgt_obd->obd_self_export;
2018                         }
2019
2020                         err = obd_set_info(exp, keylen, key, vallen, val);
2021                         if (!rc)
2022                                 rc = err;
2023                 }
2024                 spin_unlock(&lmv->lmv_lock);
2025
2026                 RETURN(rc);
2027         }
2028         if (keylen == 5 && strcmp(key, "audit") == 0) {
2029                 struct audit_attr_msg * msg = val;
2030                 int mds = id_group(&msg->id);
2031                 int i, rc = 0;
2032                 LASSERT(mds < lmv->desc.ld_tgt_count);
2033
2034                 if (IS_AUDIT_OP(msg->attr, AUDIT_FS)) {
2035                         //FS audit, send message to all mds
2036                         for (i = 0; i < lmv->desc.ld_tgt_count;i++) {
2037                                 obd_set_info(lmv->tgts[i].ltd_exp,
2038                                                   keylen, key, vallen, val);
2039                         }
2040                 }
2041                 else if (IS_AUDIT_OP(msg->attr, AUDIT_DIR)) {
2042                         //audit for dir.
2043                         //if dir is splitted, send RPC to all mds involved
2044                         struct lmv_obj *obj;
2045                         struct lustre_id rid;
2046                         int i;
2047
2048                         obj = lmv_grab_obj(obd, &msg->id);
2049                         if (obj) {
2050                                 lmv_lock_obj(obj);
2051                                 for (i = 0; i < obj->objcount; i++) {
2052                                         rid = obj->objs[i].id;
2053                                         mds = id_group(&rid);
2054                                         obd_set_info(lmv->tgts[mds].ltd_exp,
2055                                                           keylen, key,
2056                                                           vallen, val);
2057                                 }
2058                                 lmv_unlock_obj(obj);
2059                                 lmv_put_obj(obj);
2060                         }
2061                         else {
2062                                 rc = obd_set_info(lmv->tgts[mds].ltd_exp,
2063                                                  keylen, key, vallen, val);
2064                         }
2065                 }
2066                 else {
2067                         //set audit for file
2068                         rc = obd_set_info(lmv->tgts[mds].ltd_exp,
2069                                           keylen, key, vallen, val);
2070                 }
2071                 RETURN(rc);
2072         }
2073         if (((keylen == strlen("flush_cred") &&
2074              strcmp(key, "flush_cred") == 0)) ||
2075              ((keylen == strlen("crypto_type") &&
2076              strcmp(key, "crypto_type") == 0))) {
2077                 int rc = 0, i;
2078
2079                 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count;
2080                      i++, tgt++) {
2081                         if (!tgt->ltd_exp)
2082                                 continue;
2083                         rc = obd_set_info(tgt->ltd_exp,
2084                                           keylen, key, vallen, val);
2085                         if (rc)
2086                                 RETURN(rc);
2087                 }
2088
2089                 RETURN(0);
2090         }
2091
2092         RETURN(-EINVAL);
2093 }
2094
2095 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2096                struct lov_stripe_md *lsm)
2097 {
2098         struct obd_device *obd = class_exp2obd(exp);
2099         struct lmv_obd *lmv = &obd->u.lmv;
2100         struct mea *meap, *lsmp;
2101         int mea_size, i;
2102         ENTRY;
2103
2104         mea_size = (sizeof(struct lustre_id) *
2105                     lmv->desc.ld_tgt_count) + sizeof(struct mea);
2106         if (!lmmp)
2107                 RETURN(mea_size);
2108
2109         if (*lmmp && !lsm) {
2110                 OBD_FREE(*lmmp, mea_size);
2111                 *lmmp = NULL;
2112                 RETURN(0);
2113         }
2114
2115         if (*lmmp == NULL) {
2116                 OBD_ALLOC(*lmmp, mea_size);
2117                 if (*lmmp == NULL)
2118                         RETURN(-ENOMEM);
2119         }
2120
2121         if (!lsm)
2122                 RETURN(mea_size);
2123
2124         lsmp = (struct mea *)lsm;
2125         meap = (struct mea *)*lmmp;
2126
2127         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2128             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2129                 RETURN(-EINVAL);
2130
2131         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2132         meap->mea_count = cpu_to_le32(lsmp->mea_count);
2133         meap->mea_master = cpu_to_le32(lsmp->mea_master);
2134
2135         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2136                 meap->mea_ids[i] = meap->mea_ids[i];
2137                 id_cpu_to_le(&meap->mea_ids[i]);
2138         }
2139
2140         RETURN(mea_size);
2141 }
2142
2143 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2144                  struct lov_mds_md *lmm, int lmm_size)
2145 {
2146         struct obd_device *obd = class_exp2obd(exp);
2147         struct mea **tmea = (struct mea **)lsmp;
2148         struct mea *mea = (struct mea *)lmm;
2149         struct lmv_obd *lmv = &obd->u.lmv;
2150         int mea_size, i, rc = 0;
2151         __u32 magic;
2152         ENTRY;
2153
2154         mea_size = sizeof(struct lustre_id) *
2155                 lmv->desc.ld_tgt_count + sizeof(struct mea);
2156
2157         if (lsmp == NULL)
2158                 return mea_size;
2159
2160         if (*lsmp != NULL && lmm == NULL) {
2161                 OBD_FREE(*tmea, mea_size);
2162                 RETURN(0);
2163         }
2164
2165         LASSERT(mea_size == lmm_size);
2166
2167         OBD_ALLOC(*tmea, mea_size);
2168         if (*tmea == NULL)
2169                 RETURN(-ENOMEM);
2170
2171         if (!lmm)
2172                 RETURN(mea_size);
2173
2174         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2175             mea->mea_magic == MEA_MAGIC_ALL_CHARS)
2176         {
2177                 magic = le32_to_cpu(mea->mea_magic);
2178         } else {
2179                 struct mea_old *old = (struct mea_old *)lmm;
2180
2181                 mea_size = sizeof(struct lustre_id) * old->mea_count +
2182                         sizeof(struct mea_old);
2183
2184                 if (old->mea_count > 256 || old->mea_master > 256 ||
2185                     lmm_size < mea_size || old->mea_master > old->mea_count) {
2186                         CWARN("bad MEA: count %u, master %u, size %u\n",
2187                               old->mea_count, old->mea_master, mea_size);
2188                         GOTO(out_free_mea, rc = -EINVAL);
2189                 }
2190                 magic = MEA_MAGIC_LAST_CHAR;
2191         }
2192
2193         (*tmea)->mea_magic = magic;
2194         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2195         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2196
2197         for (i = 0; i < (*tmea)->mea_count; i++) {
2198                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2199                 id_le_to_cpu(&(*tmea)->mea_ids[i]);
2200         }
2201         RETURN(mea_size);
2202
2203 out_free_mea:
2204         OBD_FREE(*tmea, mea_size);
2205         return rc;
2206 }
2207
2208 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
2209             struct lov_stripe_md *ea, obd_count oa_bufs,
2210             struct brw_page *pgarr, struct obd_trans_info *oti)
2211 {
2212         struct obd_device *obd = exp->exp_obd;
2213         struct lmv_obd *lmv = &obd->u.lmv;
2214         struct mea *mea = (struct mea *) ea;
2215         int err;
2216
2217         LASSERT(oa != NULL);
2218         LASSERT(ea != NULL);
2219         LASSERT(pgarr != NULL);
2220         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
2221
2222         oa->o_gr = id_gen(&mea->mea_ids[oa->o_mds]);
2223         oa->o_id = id_ino(&mea->mea_ids[oa->o_mds]);
2224         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2225
2226         err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp,
2227                       oa, NULL, oa_bufs, pgarr, oti);
2228         RETURN(err);
2229 }
2230
2231 static int lmv_cancel_unused(struct obd_export *exp,
2232                              struct lov_stripe_md *lsm,
2233                              int flags, void *opaque)
2234 {
2235         struct obd_device *obd = exp->exp_obd;
2236         struct lmv_obd *lmv = &obd->u.lmv;
2237         int rc = 0, err, i;
2238         ENTRY;
2239
2240         LASSERT(lsm == NULL);
2241
2242         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2243                 if (!lmv->tgts[i].ltd_exp || !lmv->tgts[i].active)
2244                         continue;
2245
2246                 err = obd_cancel_unused(lmv->tgts[i].ltd_exp,
2247                                         NULL, flags, opaque);
2248                 if (!rc)
2249                         rc = err;
2250         }
2251         RETURN(rc);
2252 }
2253
2254 struct obd_ops lmv_obd_ops = {
2255         .o_owner                = THIS_MODULE,
2256         .o_attach               = lmv_attach,
2257         .o_detach               = lmv_detach,
2258         .o_setup                = lmv_setup,
2259         .o_cleanup              = lmv_cleanup,
2260         .o_connect              = lmv_connect,
2261         .o_disconnect           = lmv_disconnect,
2262         .o_statfs               = lmv_statfs,
2263         .o_llog_init            = lmv_llog_init,
2264         .o_llog_finish          = lmv_llog_finish,
2265         .o_get_info             = lmv_get_info,
2266         .o_set_info             = lmv_set_info,
2267         .o_create               = lmv_obd_create,
2268         .o_packmd               = lmv_packmd,
2269         .o_unpackmd             = lmv_unpackmd,
2270         .o_brw                  = lmv_brw,
2271         .o_init_ea_size         = lmv_init_ea_size,
2272         .o_notify               = lmv_notify,
2273         .o_iocontrol            = lmv_iocontrol,
2274         .o_cancel_unused        = lmv_cancel_unused,
2275 };
2276
2277 struct md_ops lmv_md_ops = {
2278         .m_getstatus           = lmv_getstatus,
2279         .m_getattr             = lmv_getattr,
2280         .m_change_cbdata       = lmv_change_cbdata,
2281         .m_change_cbdata_name  = lmv_change_cbdata_name,
2282         .m_close               = lmv_close,
2283         .m_create              = lmv_create,
2284         .m_done_writing        = lmv_done_writing,
2285         .m_enqueue             = lmv_enqueue,
2286         .m_getattr_lock        = lmv_getattr_lock,
2287         .m_intent_lock         = lmv_intent_lock,
2288         .m_link                = lmv_link,
2289         .m_rename              = lmv_rename,
2290         .m_setattr             = lmv_setattr,
2291         .m_sync                = lmv_sync,
2292         .m_readpage            = lmv_readpage,
2293         .m_unlink              = lmv_unlink,
2294         .m_get_real_obd        = lmv_get_real_obd,
2295         .m_valid_attrs         = lmv_valid_attrs,
2296         .m_delete_inode        = lmv_delete_inode,
2297         .m_access_check        = lmv_access_check,
2298 };
2299
2300 int __init lmv_init(void)
2301 {
2302         struct lprocfs_static_vars lvars;
2303         int rc;
2304
2305         obj_cache = kmem_cache_create("lmv_objects",
2306                                       sizeof(struct lmv_obj),
2307                                       0, 0, NULL, NULL);
2308         if (!obj_cache) {
2309                 CERROR("error allocating lmv objects cache\n");
2310                 return -ENOMEM;
2311         }
2312
2313         lprocfs_init_vars(lmv, &lvars);
2314         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2315                                  lvars.module_vars,
2316                                  OBD_LMV_DEVICENAME);
2317         if (rc)
2318                 kmem_cache_destroy(obj_cache);
2319
2320         return rc;
2321 }
2322
2323 #ifdef __KERNEL__
2324 static void lmv_exit(void)
2325 {
2326         class_unregister_type(OBD_LMV_DEVICENAME);
2327
2328         LASSERTF(kmem_cache_destroy(obj_cache) == 0,
2329                  "can't free lmv objects cache, %d object(s)"
2330                  "still in use\n", atomic_read(&obj_cache_count));
2331 }
2332
2333 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2334 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2335 MODULE_LICENSE("GPL");
2336
2337 module_init(lmv_init);
2338 module_exit(lmv_exit);
2339 #endif