drivers/staging/lustre/lustre/lmv/lmv_obd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LMV
  38 #include <linux/slab.h>
  39 #include <linux/module.h>
  40 #include <linux/init.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mm.h>
  43 #include <asm/div64.h>
  44 #include <linux/seq_file.h>
  45 #include <linux/namei.h>
  46 #include <linux/uaccess.h>
  47
  48 #include "../include/lustre/lustre_idl.h"
  49 #include "../include/obd_support.h"
  50 #include "../include/lustre_lib.h"
  51 #include "../include/lustre_net.h"
  52 #include "../include/obd_class.h"
  53 #include "../include/lprocfs_status.h"
  54 #include "../include/lustre_lite.h"
  55 #include "../include/lustre_fid.h"
  56 #include "lmv_internal.h"
  57
  58 static void lmv_activate_target(struct lmv_obd *lmv,
  59                                 struct lmv_tgt_desc *tgt,
  60                                 int activate)
  61 {
  62         if (tgt->ltd_active == activate)
  63                 return;
  64
  65         tgt->ltd_active = activate;
  66         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  67 }
  68
  69 /**
  70  * Error codes:
  71  *
  72  *  -EINVAL  : UUID can't be found in the LMV's target list
  73  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  74  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  75  */
  76 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
  77                               int activate)
  78 {
  79         struct lmv_tgt_desc    *uninitialized_var(tgt);
  80         struct obd_device      *obd;
  81         int                  i;
  82         int                  rc = 0;
  83
  84         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  85                lmv, uuid->uuid, activate);
  86
  87         spin_lock(&lmv->lmv_lock);
  88         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
  89                 tgt = lmv->tgts[i];
  90                 if (tgt == NULL || tgt->ltd_exp == NULL)
  91                         continue;
  92
  93                 CDEBUG(D_INFO, "Target idx %d is %s conn %#llx\n", i,
  94                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
  95
  96                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
  97                         break;
  98         }
  99
 100         if (i == lmv->desc.ld_tgt_count) {
 101                 rc = -EINVAL;
 102                 goto out_lmv_lock;
 103         }
 104
 105         obd = class_exp2obd(tgt->ltd_exp);
 106         if (obd == NULL) {
 107                 rc = -ENOTCONN;
 108                 goto out_lmv_lock;
 109         }
 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, LUSTRE_MDC_NAME) == 0);
 115
 116         if (tgt->ltd_active == activate) {
 117                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 118                        activate ? "" : "in");
 119                 goto out_lmv_lock;
 120         }
 121
 122         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 123                activate ? "" : "in");
 124         lmv_activate_target(lmv, tgt, activate);
 125
 126  out_lmv_lock:
 127         spin_unlock(&lmv->lmv_lock);
 128         return rc;
 129 }
 130
 131 static struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 132 {
 133         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
 134
 135         return obd_get_uuid(lmv->tgts[0]->ltd_exp);
 136 }
 137
 138 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 139                       enum obd_notify_event ev, void *data)
 140 {
 141         struct obd_connect_data *conn_data;
 142         struct lmv_obd    *lmv = &obd->u.lmv;
 143         struct obd_uuid  *uuid;
 144         int                   rc = 0;
 145
 146         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 147                 CERROR("unexpected notification of %s %s!\n",
 148                        watched->obd_type->typ_name,
 149                        watched->obd_name);
 150                 return -EINVAL;
 151         }
 152
 153         uuid = &watched->u.cli.cl_target_uuid;
 154         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 155                 /*
 156                  * Set MDC as active before notifying the observer, so the
 157                  * observer can use the MDC normally.
 158                  */
 159                 rc = lmv_set_mdc_active(lmv, uuid,
 160                                         ev == OBD_NOTIFY_ACTIVE);
 161                 if (rc) {
 162                         CERROR("%sactivation of %s failed: %d\n",
 163                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 164                                uuid->uuid, rc);
 165                         return rc;
 166                 }
 167         } else if (ev == OBD_NOTIFY_OCD) {
 168                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 169                 /*
 170                  * XXX: Make sure that ocd_connect_flags from all targets are
 171                  * the same. Otherwise one of MDTs runs wrong version or
 172                  * something like this.  --umka
 173                  */
 174                 obd->obd_self_export->exp_connect_data = *conn_data;
 175         }
 176 #if 0
 177         else if (ev == OBD_NOTIFY_DISCON) {
 178                 /*
 179                  * For disconnect event, flush fld cache for failout MDS case.
 180                  */
 181                 fld_client_flush(&lmv->lmv_fld);
 182         }
 183 #endif
 184         /*
 185          * Pass the notification up the chain.
 186          */
 187         if (obd->obd_observer)
 188                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 189
 190         return rc;
 191 }
 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  */
 197 static int lmv_connect(const struct lu_env *env,
 198                        struct obd_export **exp, struct obd_device *obd,
 199                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 200                        void *localdata)
 201 {
 202         struct lmv_obd  *lmv = &obd->u.lmv;
 203         struct lustre_handle  conn = { 0 };
 204         int                 rc = 0;
 205
 206         /*
 207          * We don't want to actually do the underlying connections more than
 208          * once, so keep track.
 209          */
 210         lmv->refcount++;
 211         if (lmv->refcount > 1) {
 212                 *exp = NULL;
 213                 return 0;
 214         }
 215
 216         rc = class_connect(&conn, obd, cluuid);
 217         if (rc) {
 218                 CERROR("class_connection() returned %d\n", rc);
 219                 return rc;
 220         }
 221
 222         *exp = class_conn2export(&conn);
 223         class_export_get(*exp);
 224
 225         lmv->exp = *exp;
 226         lmv->connected = 0;
 227         lmv->cluuid = *cluuid;
 228
 229         if (data)
 230                 lmv->conn_data = *data;
 231
 232         lmv->lmv_tgts_kobj = kobject_create_and_add("target_obds",
 233                                                     &obd->obd_kobj);
 234         /*
 235          * All real clients should perform actual connection right away, because
 236          * it is possible, that LMV will not have opportunity to connect targets
 237          * and MDC stuff will be called directly, for instance while reading
 238          * ../mdc/../kbytesfree procfs file, etc.
 239          */
 240         if (data->ocd_connect_flags & OBD_CONNECT_REAL)
 241                 rc = lmv_check_connect(obd);
 242
 243         if (rc && lmv->lmv_tgts_kobj)
 244                 kobject_put(lmv->lmv_tgts_kobj);
 245
 246         return rc;
 247 }
 248
 249 static void lmv_set_timeouts(struct obd_device *obd)
 250 {
 251         struct lmv_tgt_desc   *tgt;
 252         struct lmv_obd  *lmv;
 253         int                 i;
 254
 255         lmv = &obd->u.lmv;
 256         if (lmv->server_timeout == 0)
 257                 return;
 258
 259         if (lmv->connected == 0)
 260                 return;
 261
 262         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 263                 tgt = lmv->tgts[i];
 264                 if (tgt == NULL || tgt->ltd_exp == NULL || tgt->ltd_active == 0)
 265                         continue;
 266
 267                 obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
 268                                    KEY_INTERMDS, 0, NULL, NULL);
 269         }
 270 }
 271
 272 static int lmv_init_ea_size(struct obd_export *exp, int easize,
 273                             int def_easize, int cookiesize, int def_cookiesize)
 274 {
 275         struct obd_device   *obd = exp->exp_obd;
 276         struct lmv_obd      *lmv = &obd->u.lmv;
 277         int               i;
 278         int               rc = 0;
 279         int               change = 0;
 280
 281         if (lmv->max_easize < easize) {
 282                 lmv->max_easize = easize;
 283                 change = 1;
 284         }
 285         if (lmv->max_def_easize < def_easize) {
 286                 lmv->max_def_easize = def_easize;
 287                 change = 1;
 288         }
 289         if (lmv->max_cookiesize < cookiesize) {
 290                 lmv->max_cookiesize = cookiesize;
 291                 change = 1;
 292         }
 293         if (lmv->max_def_cookiesize < def_cookiesize) {
 294                 lmv->max_def_cookiesize = def_cookiesize;
 295                 change = 1;
 296         }
 297         if (change == 0)
 298                 return 0;
 299
 300         if (lmv->connected == 0)
 301                 return 0;
 302
 303         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 304                 if (lmv->tgts[i] == NULL ||
 305                     lmv->tgts[i]->ltd_exp == NULL ||
 306                     lmv->tgts[i]->ltd_active == 0) {
 307                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 308                         continue;
 309                 }
 310
 311                 rc = md_init_ea_size(lmv->tgts[i]->ltd_exp, easize, def_easize,
 312                                      cookiesize, def_cookiesize);
 313                 if (rc) {
 314                         CERROR("%s: obd_init_ea_size() failed on MDT target %d: rc = %d.\n",
 315                                obd->obd_name, i, rc);
 316                         break;
 317                 }
 318         }
 319         return rc;
 320 }
 321
 322 #define MAX_STRING_SIZE 128
 323
 324 static int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 325 {
 326         struct lmv_obd    *lmv = &obd->u.lmv;
 327         struct obd_uuid  *cluuid = &lmv->cluuid;
 328         struct obd_uuid   lmv_mdc_uuid = { "LMV_MDC_UUID" };
 329         struct obd_device       *mdc_obd;
 330         struct obd_export       *mdc_exp;
 331         struct lu_fld_target     target;
 332         int                   rc;
 333
 334         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 335                                         &obd->obd_uuid);
 336         if (!mdc_obd) {
 337                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 338                 return -EINVAL;
 339         }
 340
 341         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 342                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 343                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 344                 cluuid->uuid);
 345
 346         if (!mdc_obd->obd_set_up) {
 347                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 348                 return -EINVAL;
 349         }
 350
 351         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 352                          &lmv->conn_data, NULL);
 353         if (rc) {
 354                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 355                 return rc;
 356         }
 357
 358         /*
 359          * Init fid sequence client for this mdc and add new fld target.
 360          */
 361         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 362         if (rc)
 363                 return rc;
 364
 365         target.ft_srv = NULL;
 366         target.ft_exp = mdc_exp;
 367         target.ft_idx = tgt->ltd_idx;
 368
 369         fld_client_add_target(&lmv->lmv_fld, &target);
 370
 371         rc = obd_register_observer(mdc_obd, obd);
 372         if (rc) {
 373                 obd_disconnect(mdc_exp);
 374                 CERROR("target %s register_observer error %d\n",
 375                        tgt->ltd_uuid.uuid, rc);
 376                 return rc;
 377         }
 378
 379         if (obd->obd_observer) {
 380                 /*
 381                  * Tell the observer about the new target.
 382                  */
 383                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 384                                 OBD_NOTIFY_ACTIVE,
 385                                 (void *)(tgt - lmv->tgts[0]));
 386                 if (rc) {
 387                         obd_disconnect(mdc_exp);
 388                         return rc;
 389                 }
 390         }
 391
 392         tgt->ltd_active = 1;
 393         tgt->ltd_exp = mdc_exp;
 394         lmv->desc.ld_active_tgt_count++;
 395
 396         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
 397                         lmv->max_cookiesize, lmv->max_def_cookiesize);
 398
 399         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 400                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 401                 atomic_read(&obd->obd_refcount));
 402
 403         if (lmv->lmv_tgts_kobj)
 404                 /* Even if we failed to create the link, that's fine */
 405                 rc = sysfs_create_link(lmv->lmv_tgts_kobj, &mdc_obd->obd_kobj,
 406                                        mdc_obd->obd_name);
 407         return 0;
 408 }
 409
 410 static void lmv_del_target(struct lmv_obd *lmv, int index)
 411 {
 412         if (lmv->tgts[index] == NULL)
 413                 return;
 414
 415         kfree(lmv->tgts[index]);
 416         lmv->tgts[index] = NULL;
 417         return;
 418 }
 419
 420 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 421                            __u32 index, int gen)
 422 {
 423         struct lmv_obd      *lmv = &obd->u.lmv;
 424         struct lmv_tgt_desc *tgt;
 425         int               rc = 0;
 426
 427         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 428
 429         lmv_init_lock(lmv);
 430
 431         if (lmv->desc.ld_tgt_count == 0) {
 432                 struct obd_device *mdc_obd;
 433
 434                 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 435                                                 &obd->obd_uuid);
 436                 if (!mdc_obd) {
 437                         lmv_init_unlock(lmv);
 438                         CERROR("%s: Target %s not attached: rc = %d\n",
 439                                obd->obd_name, uuidp->uuid, -EINVAL);
 440                         return -EINVAL;
 441                 }
 442         }
 443
 444         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
 445                 tgt = lmv->tgts[index];
 446                 CERROR("%s: UUID %s already assigned at LOV target index %d: rc = %d\n",
 447                        obd->obd_name,
 448                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
 449                 lmv_init_unlock(lmv);
 450                 return -EEXIST;
 451         }
 452
 453         if (index >= lmv->tgts_size) {
 454                 /* We need to reallocate the lmv target array. */
 455                 struct lmv_tgt_desc **newtgts, **old = NULL;
 456                 __u32 newsize = 1;
 457                 __u32 oldsize = 0;
 458
 459                 while (newsize < index + 1)
 460                         newsize <<= 1;
 461                 newtgts = kcalloc(newsize, sizeof(*newtgts), GFP_NOFS);
 462                 if (newtgts == NULL) {
 463                         lmv_init_unlock(lmv);
 464                         return -ENOMEM;
 465                 }
 466
 467                 if (lmv->tgts_size) {
 468                         memcpy(newtgts, lmv->tgts,
 469                                sizeof(*newtgts) * lmv->tgts_size);
 470                         old = lmv->tgts;
 471                         oldsize = lmv->tgts_size;
 472                 }
 473
 474                 lmv->tgts = newtgts;
 475                 lmv->tgts_size = newsize;
 476                 smp_rmb();
 477                 kfree(old);
 478
 479                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
 480                        lmv->tgts_size);
 481         }
 482
 483         tgt = kzalloc(sizeof(*tgt), GFP_NOFS);
 484         if (!tgt) {
 485                 lmv_init_unlock(lmv);
 486                 return -ENOMEM;
 487         }
 488
 489         mutex_init(&tgt->ltd_fid_mutex);
 490         tgt->ltd_idx = index;
 491         tgt->ltd_uuid = *uuidp;
 492         tgt->ltd_active = 0;
 493         lmv->tgts[index] = tgt;
 494         if (index >= lmv->desc.ld_tgt_count)
 495                 lmv->desc.ld_tgt_count = index + 1;
 496
 497         if (lmv->connected) {
 498                 rc = lmv_connect_mdc(obd, tgt);
 499                 if (rc) {
 500                         spin_lock(&lmv->lmv_lock);
 501                         lmv->desc.ld_tgt_count--;
 502                         memset(tgt, 0, sizeof(*tgt));
 503                         spin_unlock(&lmv->lmv_lock);
 504                 } else {
 505                         int easize = sizeof(struct lmv_stripe_md) +
 506                                 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
 507                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
 508                 }
 509         }
 510
 511         lmv_init_unlock(lmv);
 512         return rc;
 513 }
 514
 515 int lmv_check_connect(struct obd_device *obd)
 516 {
 517         struct lmv_obd       *lmv = &obd->u.lmv;
 518         struct lmv_tgt_desc  *tgt;
 519         int                i;
 520         int                rc;
 521         int                easize;
 522
 523         if (lmv->connected)
 524                 return 0;
 525
 526         lmv_init_lock(lmv);
 527         if (lmv->connected) {
 528                 lmv_init_unlock(lmv);
 529                 return 0;
 530         }
 531
 532         if (lmv->desc.ld_tgt_count == 0) {
 533                 lmv_init_unlock(lmv);
 534                 CERROR("%s: no targets configured.\n", obd->obd_name);
 535                 return -EINVAL;
 536         }
 537
 538         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 539                lmv->cluuid.uuid, obd->obd_name);
 540
 541         LASSERT(lmv->tgts != NULL);
 542
 543         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 544                 tgt = lmv->tgts[i];
 545                 if (tgt == NULL)
 546                         continue;
 547                 rc = lmv_connect_mdc(obd, tgt);
 548                 if (rc)
 549                         goto out_disc;
 550         }
 551
 552         lmv_set_timeouts(obd);
 553         class_export_put(lmv->exp);
 554         lmv->connected = 1;
 555         easize = lmv_get_easize(lmv);
 556         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
 557         lmv_init_unlock(lmv);
 558         return 0;
 559
 560  out_disc:
 561         while (i-- > 0) {
 562                 int rc2;
 563
 564                 tgt = lmv->tgts[i];
 565                 if (tgt == NULL)
 566                         continue;
 567                 tgt->ltd_active = 0;
 568                 if (tgt->ltd_exp) {
 569                         --lmv->desc.ld_active_tgt_count;
 570                         rc2 = obd_disconnect(tgt->ltd_exp);
 571                         if (rc2) {
 572                                 CERROR("LMV target %s disconnect on MDC idx %d: error %d\n",
 573                                        tgt->ltd_uuid.uuid, i, rc2);
 574                         }
 575                 }
 576         }
 577         class_disconnect(lmv->exp);
 578         lmv_init_unlock(lmv);
 579         return rc;
 580 }
 581
 582 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 583 {
 584         struct lmv_obd   *lmv = &obd->u.lmv;
 585         struct obd_device      *mdc_obd;
 586         int                  rc;
 587
 588         LASSERT(tgt != NULL);
 589         LASSERT(obd != NULL);
 590
 591         mdc_obd = class_exp2obd(tgt->ltd_exp);
 592
 593         if (mdc_obd) {
 594                 mdc_obd->obd_force = obd->obd_force;
 595                 mdc_obd->obd_fail = obd->obd_fail;
 596                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 597
 598                 if (lmv->lmv_tgts_kobj)
 599                         sysfs_remove_link(lmv->lmv_tgts_kobj,
 600                                           mdc_obd->obd_name);
 601         }
 602
 603         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 604         if (rc)
 605                 CERROR("Can't finalize fids factory\n");
 606
 607         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 608                tgt->ltd_exp->exp_obd->obd_name,
 609                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 610
 611         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 612         rc = obd_disconnect(tgt->ltd_exp);
 613         if (rc) {
 614                 if (tgt->ltd_active) {
 615                         CERROR("Target %s disconnect error %d\n",
 616                                tgt->ltd_uuid.uuid, rc);
 617                 }
 618         }
 619
 620         lmv_activate_target(lmv, tgt, 0);
 621         tgt->ltd_exp = NULL;
 622         return 0;
 623 }
 624
 625 static int lmv_disconnect(struct obd_export *exp)
 626 {
 627         struct obd_device     *obd = class_exp2obd(exp);
 628         struct lmv_obd  *lmv = &obd->u.lmv;
 629         int                 rc;
 630         int                 i;
 631
 632         if (!lmv->tgts)
 633                 goto out_local;
 634
 635         /*
 636          * Only disconnect the underlying layers on the final disconnect.
 637          */
 638         lmv->refcount--;
 639         if (lmv->refcount != 0)
 640                 goto out_local;
 641
 642         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 643                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
 644                         continue;
 645
 646                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
 647         }
 648
 649         if (lmv->lmv_tgts_kobj)
 650                 kobject_put(lmv->lmv_tgts_kobj);
 651
 652 out_local:
 653         /*
 654          * This is the case when no real connection is established by
 655          * lmv_check_connect().
 656          */
 657         if (!lmv->connected)
 658                 class_export_put(exp);
 659         rc = class_disconnect(exp);
 660         if (lmv->refcount == 0)
 661                 lmv->connected = 0;
 662         return rc;
 663 }
 664
 665 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
 666 {
 667         struct obd_device       *obddev = class_exp2obd(exp);
 668         struct lmv_obd          *lmv = &obddev->u.lmv;
 669         struct getinfo_fid2path *gf;
 670         struct lmv_tgt_desc     *tgt;
 671         struct getinfo_fid2path *remote_gf = NULL;
 672         int                     remote_gf_size = 0;
 673         int                     rc;
 674
 675         gf = (struct getinfo_fid2path *)karg;
 676         tgt = lmv_find_target(lmv, &gf->gf_fid);
 677         if (IS_ERR(tgt))
 678                 return PTR_ERR(tgt);
 679
 680 repeat_fid2path:
 681         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 682         if (rc != 0 && rc != -EREMOTE)
 683                 goto out_fid2path;
 684
 685         /* If remote_gf != NULL, it means just building the
 686          * path on the remote MDT, copy this path segment to gf */
 687         if (remote_gf != NULL) {
 688                 struct getinfo_fid2path *ori_gf;
 689                 char *ptr;
 690
 691                 ori_gf = (struct getinfo_fid2path *)karg;
 692                 if (strlen(ori_gf->gf_path) +
 693                     strlen(gf->gf_path) > ori_gf->gf_pathlen) {
 694                         rc = -EOVERFLOW;
 695                         goto out_fid2path;
 696                 }
 697
 698                 ptr = ori_gf->gf_path;
 699
 700                 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
 701                         strlen(ori_gf->gf_path));
 702
 703                 strcpy(ptr, gf->gf_path);
 704                 ptr += strlen(gf->gf_path);
 705                 *ptr = '/';
 706         }
 707
 708         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: %llu ln: %u\n",
 709                tgt->ltd_exp->exp_obd->obd_name,
 710                gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 711                gf->gf_linkno);
 712
 713         if (rc == 0)
 714                 goto out_fid2path;
 715
 716         /* sigh, has to go to another MDT to do path building further */
 717         if (remote_gf == NULL) {
 718                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 719                 remote_gf = kzalloc(remote_gf_size, GFP_NOFS);
 720                 if (!remote_gf) {
 721                         rc = -ENOMEM;
 722                         goto out_fid2path;
 723                 }
 724                 remote_gf->gf_pathlen = PATH_MAX;
 725         }
 726
 727         if (!fid_is_sane(&gf->gf_fid)) {
 728                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 729                        tgt->ltd_exp->exp_obd->obd_name,
 730                        PFID(&gf->gf_fid), -EINVAL);
 731                 rc = -EINVAL;
 732                 goto out_fid2path;
 733         }
 734
 735         tgt = lmv_find_target(lmv, &gf->gf_fid);
 736         if (IS_ERR(tgt)) {
 737                 rc = -EINVAL;
 738                 goto out_fid2path;
 739         }
 740
 741         remote_gf->gf_fid = gf->gf_fid;
 742         remote_gf->gf_recno = -1;
 743         remote_gf->gf_linkno = -1;
 744         memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
 745         gf = remote_gf;
 746         goto repeat_fid2path;
 747
 748 out_fid2path:
 749         kfree(remote_gf);
 750         return rc;
 751 }
 752
 753 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 754                              const struct hsm_user_request *hur,
 755                              const struct lmv_tgt_desc *tgt_mds)
 756 {
 757         int                     i, nr = 0;
 758         struct lmv_tgt_desc    *curr_tgt;
 759
 760         /* count how many requests must be sent to the given target */
 761         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 762                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
 763                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 764                         nr++;
 765         }
 766         return nr;
 767 }
 768
 769 static void lmv_hsm_req_build(struct lmv_obd *lmv,
 770                               struct hsm_user_request *hur_in,
 771                               const struct lmv_tgt_desc *tgt_mds,
 772                               struct hsm_user_request *hur_out)
 773 {
 774         int                     i, nr_out;
 775         struct lmv_tgt_desc    *curr_tgt;
 776
 777         /* build the hsm_user_request for the given target */
 778         hur_out->hur_request = hur_in->hur_request;
 779         nr_out = 0;
 780         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 781                 curr_tgt = lmv_find_target(lmv,
 782                                         &hur_in->hur_user_item[i].hui_fid);
 783                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 784                         hur_out->hur_user_item[nr_out] =
 785                                 hur_in->hur_user_item[i];
 786                         nr_out++;
 787                 }
 788         }
 789         hur_out->hur_request.hr_itemcount = nr_out;
 790         memcpy(hur_data(hur_out), hur_data(hur_in),
 791                hur_in->hur_request.hr_data_len);
 792 }
 793
 794 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
 795                                  struct lustre_kernelcomm *lk, void *uarg)
 796 {
 797         int     i, rc = 0;
 798
 799         /* unregister request (call from llapi_hsm_copytool_fini) */
 800         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 801                 /* best effort: try to clean as much as possible
 802                  * (continue on error) */
 803                 obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp, len, lk, uarg);
 804         }
 805
 806         /* Whatever the result, remove copytool from kuc groups.
 807          * Unreached coordinators will get EPIPE on next requests
 808          * and will unregister automatically.
 809          */
 810         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group);
 811         return rc;
 812 }
 813
 814 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
 815                                struct lustre_kernelcomm *lk, void *uarg)
 816 {
 817         struct file     *filp;
 818         int              i, j, err;
 819         int              rc = 0;
 820         bool             any_set = false;
 821
 822         /* All or nothing: try to register to all MDS.
 823          * In case of failure, unregister from previous MDS,
 824          * except if it because of inactive target. */
 825         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 826                 err = obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp,
 827                                    len, lk, uarg);
 828                 if (err) {
 829                         if (lmv->tgts[i]->ltd_active) {
 830                                 /* permanent error */
 831                                 CERROR("error: iocontrol MDC %s on MDTidx %d cmd %x: err = %d\n",
 832                                        lmv->tgts[i]->ltd_uuid.uuid,
 833                                        i, cmd, err);
 834                                 rc = err;
 835                                 lk->lk_flags |= LK_FLG_STOP;
 836                                 /* unregister from previous MDS */
 837                                 for (j = 0; j < i; j++)
 838                                         obd_iocontrol(cmd,
 839                                                   lmv->tgts[j]->ltd_exp,
 840                                                   len, lk, uarg);
 841                                 return rc;
 842                         }
 843                         /* else: transient error.
 844                          * kuc will register to the missing MDT
 845                          * when it is back */
 846                 } else {
 847                         any_set = true;
 848                 }
 849         }
 850
 851         if (!any_set)
 852                 /* no registration done: return error */
 853                 return -ENOTCONN;
 854
 855         /* at least one registration done, with no failure */
 856         filp = fget(lk->lk_wfd);
 857         if (filp == NULL) {
 858                 return -EBADF;
 859         }
 860         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, lk->lk_data);
 861         if (rc != 0 && filp != NULL)
 862                 fput(filp);
 863         return rc;
 864 }
 865
 866 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 867                          int len, void *karg, void *uarg)
 868 {
 869         struct obd_device    *obddev = class_exp2obd(exp);
 870         struct lmv_obd       *lmv = &obddev->u.lmv;
 871         int                i = 0;
 872         int                rc = 0;
 873         int                set = 0;
 874         int                count = lmv->desc.ld_tgt_count;
 875
 876         if (count == 0)
 877                 return -ENOTTY;
 878
 879         switch (cmd) {
 880         case IOC_OBD_STATFS: {
 881                 struct obd_ioctl_data *data = karg;
 882                 struct obd_device *mdc_obd;
 883                 struct obd_statfs stat_buf = {0};
 884                 __u32 index;
 885
 886                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 887                 if (index >= count)
 888                         return -ENODEV;
 889
 890                 if (lmv->tgts[index] == NULL ||
 891                     lmv->tgts[index]->ltd_active == 0)
 892                         return -ENODATA;
 893
 894                 mdc_obd = class_exp2obd(lmv->tgts[index]->ltd_exp);
 895                 if (!mdc_obd)
 896                         return -EINVAL;
 897
 898                 /* copy UUID */
 899                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 900                                      min((int) data->ioc_plen2,
 901                                          (int) sizeof(struct obd_uuid))))
 902                         return -EFAULT;
 903
 904                 rc = obd_statfs(NULL, lmv->tgts[index]->ltd_exp, &stat_buf,
 905                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 906                                 0);
 907                 if (rc)
 908                         return rc;
 909                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
 910                                      min((int) data->ioc_plen1,
 911                                          (int) sizeof(stat_buf))))
 912                         return -EFAULT;
 913                 break;
 914         }
 915         case OBD_IOC_QUOTACTL: {
 916                 struct if_quotactl *qctl = karg;
 917                 struct lmv_tgt_desc *tgt = NULL;
 918                 struct obd_quotactl *oqctl;
 919
 920                 if (qctl->qc_valid == QC_MDTIDX) {
 921                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
 922                                 return -EINVAL;
 923
 924                         tgt = lmv->tgts[qctl->qc_idx];
 925                         if (tgt == NULL || tgt->ltd_exp == NULL)
 926                                 return -EINVAL;
 927                 } else if (qctl->qc_valid == QC_UUID) {
 928                         for (i = 0; i < count; i++) {
 929                                 tgt = lmv->tgts[i];
 930                                 if (tgt == NULL)
 931                                         continue;
 932                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 933                                                      &qctl->obd_uuid))
 934                                         continue;
 935
 936                                 if (tgt->ltd_exp == NULL)
 937                                         return -EINVAL;
 938
 939                                 break;
 940                         }
 941                 } else {
 942                         return -EINVAL;
 943                 }
 944
 945                 if (i >= count)
 946                         return -EAGAIN;
 947
 948                 LASSERT(tgt && tgt->ltd_exp);
 949                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
 950                 if (!oqctl)
 951                         return -ENOMEM;
 952
 953                 QCTL_COPY(oqctl, qctl);
 954                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
 955                 if (rc == 0) {
 956                         QCTL_COPY(qctl, oqctl);
 957                         qctl->qc_valid = QC_MDTIDX;
 958                         qctl->obd_uuid = tgt->ltd_uuid;
 959                 }
 960                 kfree(oqctl);
 961                 break;
 962         }
 963         case OBD_IOC_CHANGELOG_SEND:
 964         case OBD_IOC_CHANGELOG_CLEAR: {
 965                 struct ioc_changelog *icc = karg;
 966
 967                 if (icc->icc_mdtindex >= count)
 968                         return -ENODEV;
 969
 970                 if (lmv->tgts[icc->icc_mdtindex] == NULL ||
 971                     lmv->tgts[icc->icc_mdtindex]->ltd_exp == NULL ||
 972                     lmv->tgts[icc->icc_mdtindex]->ltd_active == 0)
 973                         return -ENODEV;
 974                 rc = obd_iocontrol(cmd, lmv->tgts[icc->icc_mdtindex]->ltd_exp,
 975                                    sizeof(*icc), icc, NULL);
 976                 break;
 977         }
 978         case LL_IOC_GET_CONNECT_FLAGS: {
 979                 if (lmv->tgts[0] == NULL)
 980                         return -ENODATA;
 981                 rc = obd_iocontrol(cmd, lmv->tgts[0]->ltd_exp, len, karg, uarg);
 982                 break;
 983         }
 984         case OBD_IOC_FID2PATH: {
 985                 rc = lmv_fid2path(exp, len, karg, uarg);
 986                 break;
 987         }
 988         case LL_IOC_HSM_STATE_GET:
 989         case LL_IOC_HSM_STATE_SET:
 990         case LL_IOC_HSM_ACTION: {
 991                 struct md_op_data       *op_data = karg;
 992                 struct lmv_tgt_desc     *tgt;
 993
 994                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
 995                 if (IS_ERR(tgt))
 996                                 return PTR_ERR(tgt);
 997
 998                 if (tgt->ltd_exp == NULL)
 999                                 return -EINVAL;
1000
1001                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1002                 break;
1003         }
1004         case LL_IOC_HSM_PROGRESS: {
1005                 const struct hsm_progress_kernel *hpk = karg;
1006                 struct lmv_tgt_desc     *tgt;
1007
1008                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1009                 if (IS_ERR(tgt))
1010                         return PTR_ERR(tgt);
1011                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1012                 break;
1013         }
1014         case LL_IOC_HSM_REQUEST: {
1015                 struct hsm_user_request *hur = karg;
1016                 struct lmv_tgt_desc     *tgt;
1017                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1018
1019                 if (reqcount == 0)
1020                         return 0;
1021
1022                 /* if the request is about a single fid
1023                  * or if there is a single MDS, no need to split
1024                  * the request. */
1025                 if (reqcount == 1 || count == 1) {
1026                         tgt = lmv_find_target(lmv,
1027                                               &hur->hur_user_item[0].hui_fid);
1028                         if (IS_ERR(tgt))
1029                                 return PTR_ERR(tgt);
1030                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1031                 } else {
1032                         /* split fid list to their respective MDS */
1033                         for (i = 0; i < count; i++) {
1034                                 unsigned int            nr, reqlen;
1035                                 int                     rc1;
1036                                 struct hsm_user_request *req;
1037
1038                                 nr = lmv_hsm_req_count(lmv, hur, lmv->tgts[i]);
1039                                 if (nr == 0) /* nothing for this MDS */
1040                                         continue;
1041
1042                                 /* build a request with fids for this MDS */
1043                                 reqlen = offsetof(typeof(*hur),
1044                                                   hur_user_item[nr])
1045                                          + hur->hur_request.hr_data_len;
1046                                 req = libcfs_kvzalloc(reqlen, GFP_NOFS);
1047                                 if (req == NULL)
1048                                         return -ENOMEM;
1049
1050                                 lmv_hsm_req_build(lmv, hur, lmv->tgts[i], req);
1051
1052                                 rc1 = obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp,
1053                                                     reqlen, req, uarg);
1054                                 if (rc1 != 0 && rc == 0)
1055                                         rc = rc1;
1056                                 kvfree(req);
1057                         }
1058                 }
1059                 break;
1060         }
1061         case LL_IOC_LOV_SWAP_LAYOUTS: {
1062                 struct md_op_data       *op_data = karg;
1063                 struct lmv_tgt_desc     *tgt1, *tgt2;
1064
1065                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1066                 if (IS_ERR(tgt1))
1067                         return PTR_ERR(tgt1);
1068
1069                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1070                 if (IS_ERR(tgt2))
1071                         return PTR_ERR(tgt2);
1072
1073                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1074                         return -EINVAL;
1075
1076                 /* only files on same MDT can have their layouts swapped */
1077                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1078                         return -EPERM;
1079
1080                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1081                 break;
1082         }
1083         case LL_IOC_HSM_CT_START: {
1084                 struct lustre_kernelcomm *lk = karg;
1085
1086                 if (lk->lk_flags & LK_FLG_STOP)
1087                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1088                 else
1089                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1090                 break;
1091         }
1092         default:
1093                 for (i = 0; i < count; i++) {
1094                         struct obd_device *mdc_obd;
1095                         int err;
1096
1097                         if (lmv->tgts[i] == NULL ||
1098                             lmv->tgts[i]->ltd_exp == NULL)
1099                                 continue;
1100                         /* ll_umount_begin() sets force flag but for lmv, not
1101                          * mdc. Let's pass it through */
1102                         mdc_obd = class_exp2obd(lmv->tgts[i]->ltd_exp);
1103                         mdc_obd->obd_force = obddev->obd_force;
1104                         err = obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp, len,
1105                                             karg, uarg);
1106                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1107                                 return err;
1108                         } else if (err) {
1109                                 if (lmv->tgts[i]->ltd_active) {
1110                                         CERROR("error: iocontrol MDC %s on MDTidx %d cmd %x: err = %d\n",
1111                                                lmv->tgts[i]->ltd_uuid.uuid,
1112                                                i, cmd, err);
1113                                         if (!rc)
1114                                                 rc = err;
1115                                 }
1116                         } else
1117                                 set = 1;
1118                 }
1119                 if (!set && !rc)
1120                         rc = -EIO;
1121         }
1122         return rc;
1123 }
1124
1125 #if 0
1126 static int lmv_all_chars_policy(int count, const char *name,
1127                                 int len)
1128 {
1129         unsigned int c = 0;
1130
1131         while (len > 0)
1132                 c += name[--len];
1133         c = c % count;
1134         return c;
1135 }
1136
1137 static int lmv_nid_policy(struct lmv_obd *lmv)
1138 {
1139         struct obd_import *imp;
1140         __u32         id;
1141
1142         /*
1143          * XXX: To get nid we assume that underlying obd device is mdc.
1144          */
1145         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1146         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1147         return id % lmv->desc.ld_tgt_count;
1148 }
1149
1150 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1151                           enum placement_policy placement)
1152 {
1153         switch (placement) {
1154         case PLACEMENT_CHAR_POLICY:
1155                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1156                                             op_data->op_name,
1157                                             op_data->op_namelen);
1158         case PLACEMENT_NID_POLICY:
1159                 return lmv_nid_policy(lmv);
1160
1161         default:
1162                 break;
1163         }
1164
1165         CERROR("Unsupported placement policy %x\n", placement);
1166         return -EINVAL;
1167 }
1168 #endif
1169
1170 /**
1171  * This is _inode_ placement policy function (not name).
1172  */
1173 static int lmv_placement_policy(struct obd_device *obd,
1174                                 struct md_op_data *op_data, u32 *mds)
1175 {
1176         struct lmv_obd    *lmv = &obd->u.lmv;
1177
1178         LASSERT(mds != NULL);
1179
1180         if (lmv->desc.ld_tgt_count == 1) {
1181                 *mds = 0;
1182                 return 0;
1183         }
1184
1185         /**
1186          * If stripe_offset is provided during setdirstripe
1187          * (setdirstripe -i xx), xx MDS will be chosen.
1188          */
1189         if (op_data->op_cli_flags & CLI_SET_MEA) {
1190                 struct lmv_user_md *lum;
1191
1192                 lum = (struct lmv_user_md *)op_data->op_data;
1193                 if (lum->lum_type == LMV_STRIPE_TYPE &&
1194                     lum->lum_stripe_offset != -1) {
1195                         if (lum->lum_stripe_offset >= lmv->desc.ld_tgt_count) {
1196                                 CERROR("%s: Stripe_offset %d > MDT count %d: rc = %d\n",
1197                                        obd->obd_name,
1198                                        lum->lum_stripe_offset,
1199                                        lmv->desc.ld_tgt_count, -ERANGE);
1200                                 return -ERANGE;
1201                         }
1202                         *mds = lum->lum_stripe_offset;
1203                         return 0;
1204                 }
1205         }
1206
1207         /* Allocate new fid on target according to operation type and parent
1208          * home mds. */
1209         *mds = op_data->op_mds;
1210         return 0;
1211 }
1212
1213 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1214 {
1215         struct lmv_tgt_desc     *tgt;
1216         int                      rc;
1217
1218         tgt = lmv_get_target(lmv, mds);
1219         if (IS_ERR(tgt))
1220                 return PTR_ERR(tgt);
1221
1222         /*
1223          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1224          * on server that seq in new allocated fid is not yet known.
1225          */
1226         mutex_lock(&tgt->ltd_fid_mutex);
1227
1228         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL) {
1229                 rc = -ENODEV;
1230                 goto out;
1231         }
1232
1233         /*
1234          * Asking underlaying tgt layer to allocate new fid.
1235          */
1236         rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
1237         if (rc > 0) {
1238                 LASSERT(fid_is_sane(fid));
1239                 rc = 0;
1240         }
1241
1242 out:
1243         mutex_unlock(&tgt->ltd_fid_mutex);
1244         return rc;
1245 }
1246
1247 int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1248                   struct md_op_data *op_data)
1249 {
1250         struct obd_device     *obd = class_exp2obd(exp);
1251         struct lmv_obd  *lmv = &obd->u.lmv;
1252         u32                    mds = 0;
1253         int                 rc;
1254
1255         LASSERT(op_data != NULL);
1256         LASSERT(fid != NULL);
1257
1258         rc = lmv_placement_policy(obd, op_data, &mds);
1259         if (rc) {
1260                 CERROR("Can't get target for allocating fid, rc %d\n",
1261                        rc);
1262                 return rc;
1263         }
1264
1265         rc = __lmv_fid_alloc(lmv, fid, mds);
1266         if (rc) {
1267                 CERROR("Can't alloc new fid, rc %d\n", rc);
1268                 return rc;
1269         }
1270
1271         return rc;
1272 }
1273
1274 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1275 {
1276         struct lmv_obd       *lmv = &obd->u.lmv;
1277         struct lprocfs_static_vars  lvars = { NULL };
1278         struct lmv_desc     *desc;
1279         int                      rc;
1280
1281         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1282                 CERROR("LMV setup requires a descriptor\n");
1283                 return -EINVAL;
1284         }
1285
1286         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1287         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1288                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1289                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1290                 return -EINVAL;
1291         }
1292
1293         lmv->tgts = kcalloc(32, sizeof(*lmv->tgts), GFP_NOFS);
1294         if (lmv->tgts == NULL)
1295                 return -ENOMEM;
1296         lmv->tgts_size = 32;
1297
1298         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1299         lmv->desc.ld_tgt_count = 0;
1300         lmv->desc.ld_active_tgt_count = 0;
1301         lmv->max_cookiesize = 0;
1302         lmv->max_def_easize = 0;
1303         lmv->max_easize = 0;
1304         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1305
1306         spin_lock_init(&lmv->lmv_lock);
1307         mutex_init(&lmv->init_mutex);
1308
1309         lprocfs_lmv_init_vars(&lvars);
1310
1311         lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars);
1312         rc = ldebugfs_seq_create(obd->obd_debugfs_entry, "target_obd",
1313                                  0444, &lmv_proc_target_fops, obd);
1314         if (rc)
1315                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1316                       obd->obd_name, rc);
1317         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1318                              LUSTRE_CLI_FLD_HASH_DHT);
1319         if (rc) {
1320                 CERROR("Can't init FLD, err %d\n", rc);
1321                 goto out;
1322         }
1323
1324         return 0;
1325
1326 out:
1327         return rc;
1328 }
1329
1330 static int lmv_cleanup(struct obd_device *obd)
1331 {
1332         struct lmv_obd   *lmv = &obd->u.lmv;
1333
1334         fld_client_fini(&lmv->lmv_fld);
1335         if (lmv->tgts != NULL) {
1336                 int i;
1337
1338                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1339                         if (lmv->tgts[i] == NULL)
1340                                 continue;
1341                         lmv_del_target(lmv, i);
1342                 }
1343                 kfree(lmv->tgts);
1344                 lmv->tgts_size = 0;
1345         }
1346         return 0;
1347 }
1348
1349 static int lmv_process_config(struct obd_device *obd, u32 len, void *buf)
1350 {
1351         struct lustre_cfg       *lcfg = buf;
1352         struct obd_uuid         obd_uuid;
1353         int                     gen;
1354         __u32                   index;
1355         int                     rc;
1356
1357         switch (lcfg->lcfg_command) {
1358         case LCFG_ADD_MDC:
1359                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1360                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1361                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid)) {
1362                         rc = -EINVAL;
1363                         goto out;
1364                 }
1365
1366                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1367
1368                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1) {
1369                         rc = -EINVAL;
1370                         goto out;
1371                 }
1372                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1) {
1373                         rc = -EINVAL;
1374                         goto out;
1375                 }
1376                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1377                 goto out;
1378         default:
1379                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1380                 rc = -EINVAL;
1381                 goto out;
1382         }
1383 out:
1384         return rc;
1385 }
1386
1387 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1388                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1389 {
1390         struct obd_device     *obd = class_exp2obd(exp);
1391         struct lmv_obd  *lmv = &obd->u.lmv;
1392         struct obd_statfs     *temp;
1393         int                 rc = 0;
1394         int                 i;
1395
1396         rc = lmv_check_connect(obd);
1397         if (rc)
1398                 return rc;
1399
1400         temp = kzalloc(sizeof(*temp), GFP_NOFS);
1401         if (!temp)
1402                 return -ENOMEM;
1403
1404         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1405                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1406                         continue;
1407
1408                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1409                                 max_age, flags);
1410                 if (rc) {
1411                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1412                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1413                                rc);
1414                         goto out_free_temp;
1415                 }
1416
1417                 if (i == 0) {
1418                         *osfs = *temp;
1419                         /* If the statfs is from mount, it will needs
1420                          * retrieve necessary information from MDT0.
1421                          * i.e. mount does not need the merged osfs
1422                          * from all of MDT.
1423                          * And also clients can be mounted as long as
1424                          * MDT0 is in service*/
1425                         if (flags & OBD_STATFS_FOR_MDT0)
1426                                 goto out_free_temp;
1427                 } else {
1428                         osfs->os_bavail += temp->os_bavail;
1429                         osfs->os_blocks += temp->os_blocks;
1430                         osfs->os_ffree += temp->os_ffree;
1431                         osfs->os_files += temp->os_files;
1432                 }
1433         }
1434
1435 out_free_temp:
1436         kfree(temp);
1437         return rc;
1438 }
1439
1440 static int lmv_getstatus(struct obd_export *exp,
1441                          struct lu_fid *fid)
1442 {
1443         struct obd_device    *obd = exp->exp_obd;
1444         struct lmv_obd       *lmv = &obd->u.lmv;
1445         int                rc;
1446
1447         rc = lmv_check_connect(obd);
1448         if (rc)
1449                 return rc;
1450
1451         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid);
1452         return rc;
1453 }
1454
1455 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1456                         u64 valid, const char *name,
1457                         const char *input, int input_size, int output_size,
1458                         int flags, struct ptlrpc_request **request)
1459 {
1460         struct obd_device      *obd = exp->exp_obd;
1461         struct lmv_obd   *lmv = &obd->u.lmv;
1462         struct lmv_tgt_desc    *tgt;
1463         int                  rc;
1464
1465         rc = lmv_check_connect(obd);
1466         if (rc)
1467                 return rc;
1468
1469         tgt = lmv_find_target(lmv, fid);
1470         if (IS_ERR(tgt))
1471                 return PTR_ERR(tgt);
1472
1473         rc = md_getxattr(tgt->ltd_exp, fid, valid, name, input,
1474                          input_size, output_size, flags, request);
1475
1476         return rc;
1477 }
1478
1479 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1480                         u64 valid, const char *name,
1481                         const char *input, int input_size, int output_size,
1482                         int flags, __u32 suppgid,
1483                         struct ptlrpc_request **request)
1484 {
1485         struct obd_device      *obd = exp->exp_obd;
1486         struct lmv_obd   *lmv = &obd->u.lmv;
1487         struct lmv_tgt_desc    *tgt;
1488         int                  rc;
1489
1490         rc = lmv_check_connect(obd);
1491         if (rc)
1492                 return rc;
1493
1494         tgt = lmv_find_target(lmv, fid);
1495         if (IS_ERR(tgt))
1496                 return PTR_ERR(tgt);
1497
1498         rc = md_setxattr(tgt->ltd_exp, fid, valid, name, input,
1499                          input_size, output_size, flags, suppgid,
1500                          request);
1501
1502         return rc;
1503 }
1504
1505 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1506                        struct ptlrpc_request **request)
1507 {
1508         struct obd_device       *obd = exp->exp_obd;
1509         struct lmv_obd    *lmv = &obd->u.lmv;
1510         struct lmv_tgt_desc     *tgt;
1511         int                   rc;
1512
1513         rc = lmv_check_connect(obd);
1514         if (rc)
1515                 return rc;
1516
1517         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1518         if (IS_ERR(tgt))
1519                 return PTR_ERR(tgt);
1520
1521         if (op_data->op_flags & MF_GET_MDT_IDX) {
1522                 op_data->op_mds = tgt->ltd_idx;
1523                 return 0;
1524         }
1525
1526         rc = md_getattr(tgt->ltd_exp, op_data, request);
1527
1528         return rc;
1529 }
1530
1531 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1532 {
1533         struct obd_device   *obd = exp->exp_obd;
1534         struct lmv_obd      *lmv = &obd->u.lmv;
1535         int               i;
1536         int               rc;
1537
1538         rc = lmv_check_connect(obd);
1539         if (rc)
1540                 return rc;
1541
1542         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1543
1544         /*
1545          * With DNE every object can have two locks in different namespaces:
1546          * lookup lock in space of MDT storing direntry and update/open lock in
1547          * space of MDT storing inode.
1548          */
1549         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1550                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1551                         continue;
1552                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1553         }
1554
1555         return 0;
1556 }
1557
1558 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1559                            ldlm_iterator_t it, void *data)
1560 {
1561         struct obd_device   *obd = exp->exp_obd;
1562         struct lmv_obd      *lmv = &obd->u.lmv;
1563         int               i;
1564         int               rc;
1565
1566         rc = lmv_check_connect(obd);
1567         if (rc)
1568                 return rc;
1569
1570         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1571
1572         /*
1573          * With DNE every object can have two locks in different namespaces:
1574          * lookup lock in space of MDT storing direntry and update/open lock in
1575          * space of MDT storing inode.
1576          */
1577         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1578                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1579                         continue;
1580                 rc = md_find_cbdata(lmv->tgts[i]->ltd_exp, fid, it, data);
1581                 if (rc)
1582                         return rc;
1583         }
1584
1585         return rc;
1586 }
1587
1588 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1589                      struct md_open_data *mod, struct ptlrpc_request **request)
1590 {
1591         struct obd_device     *obd = exp->exp_obd;
1592         struct lmv_obd  *lmv = &obd->u.lmv;
1593         struct lmv_tgt_desc   *tgt;
1594         int                 rc;
1595
1596         rc = lmv_check_connect(obd);
1597         if (rc)
1598                 return rc;
1599
1600         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1601         if (IS_ERR(tgt))
1602                 return PTR_ERR(tgt);
1603
1604         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1605         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1606         return rc;
1607 }
1608
1609 struct lmv_tgt_desc
1610 *lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1611                 struct lu_fid *fid)
1612 {
1613         struct lmv_tgt_desc *tgt;
1614
1615         tgt = lmv_find_target(lmv, fid);
1616         if (IS_ERR(tgt))
1617                 return tgt;
1618
1619         op_data->op_mds = tgt->ltd_idx;
1620
1621         return tgt;
1622 }
1623
1624 static int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1625                       const void *data, int datalen, int mode, __u32 uid,
1626                       __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1627                       struct ptlrpc_request **request)
1628 {
1629         struct obd_device       *obd = exp->exp_obd;
1630         struct lmv_obd    *lmv = &obd->u.lmv;
1631         struct lmv_tgt_desc     *tgt;
1632         int                   rc;
1633
1634         rc = lmv_check_connect(obd);
1635         if (rc)
1636                 return rc;
1637
1638         if (!lmv->desc.ld_active_tgt_count)
1639                 return -EIO;
1640
1641         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1642         if (IS_ERR(tgt))
1643                 return PTR_ERR(tgt);
1644
1645         rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
1646         if (rc)
1647                 return rc;
1648
1649         CDEBUG(D_INODE, "CREATE '%*s' on "DFID" -> mds #%x\n",
1650                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1651                op_data->op_mds);
1652
1653         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1654         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1655                        cap_effective, rdev, request);
1656
1657         if (rc == 0) {
1658                 if (*request == NULL)
1659                         return rc;
1660                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1661         }
1662         return rc;
1663 }
1664
1665 static int lmv_done_writing(struct obd_export *exp,
1666                             struct md_op_data *op_data,
1667                             struct md_open_data *mod)
1668 {
1669         struct obd_device     *obd = exp->exp_obd;
1670         struct lmv_obd  *lmv = &obd->u.lmv;
1671         struct lmv_tgt_desc   *tgt;
1672         int                 rc;
1673
1674         rc = lmv_check_connect(obd);
1675         if (rc)
1676                 return rc;
1677
1678         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1679         if (IS_ERR(tgt))
1680                 return PTR_ERR(tgt);
1681
1682         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1683         return rc;
1684 }
1685
1686 static int
1687 lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1688                    struct lookup_intent *it, struct md_op_data *op_data,
1689                    struct lustre_handle *lockh, void *lmm, int lmmsize,
1690                    __u64 extra_lock_flags)
1691 {
1692         struct ptlrpc_request      *req = it->d.lustre.it_data;
1693         struct obd_device         *obd = exp->exp_obd;
1694         struct lmv_obd       *lmv = &obd->u.lmv;
1695         struct lustre_handle    plock;
1696         struct lmv_tgt_desc     *tgt;
1697         struct md_op_data         *rdata;
1698         struct lu_fid          fid1;
1699         struct mdt_body     *body;
1700         int                      rc = 0;
1701         int                      pmode;
1702
1703         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1704         LASSERT(body != NULL);
1705
1706         if (!(body->valid & OBD_MD_MDS))
1707                 return 0;
1708
1709         CDEBUG(D_INODE, "REMOTE_ENQUEUE '%s' on "DFID" -> "DFID"\n",
1710                LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
1711
1712         /*
1713          * We got LOOKUP lock, but we really need attrs.
1714          */
1715         pmode = it->d.lustre.it_lock_mode;
1716         LASSERT(pmode != 0);
1717         memcpy(&plock, lockh, sizeof(plock));
1718         it->d.lustre.it_lock_mode = 0;
1719         it->d.lustre.it_data = NULL;
1720         fid1 = body->fid1;
1721
1722         ptlrpc_req_finished(req);
1723
1724         tgt = lmv_find_target(lmv, &fid1);
1725         if (IS_ERR(tgt)) {
1726                 rc = PTR_ERR(tgt);
1727                 goto out;
1728         }
1729
1730         rdata = kzalloc(sizeof(*rdata), GFP_NOFS);
1731         if (!rdata) {
1732                 rc = -ENOMEM;
1733                 goto out;
1734         }
1735
1736         rdata->op_fid1 = fid1;
1737         rdata->op_bias = MDS_CROSS_REF;
1738
1739         rc = md_enqueue(tgt->ltd_exp, einfo, it, rdata, lockh,
1740                         lmm, lmmsize, NULL, extra_lock_flags);
1741         kfree(rdata);
1742 out:
1743         ldlm_lock_decref(&plock, pmode);
1744         return rc;
1745 }
1746
1747 static int
1748 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1749             struct lookup_intent *it, struct md_op_data *op_data,
1750             struct lustre_handle *lockh, void *lmm, int lmmsize,
1751             struct ptlrpc_request **req, __u64 extra_lock_flags)
1752 {
1753         struct obd_device       *obd = exp->exp_obd;
1754         struct lmv_obd     *lmv = &obd->u.lmv;
1755         struct lmv_tgt_desc      *tgt;
1756         int                    rc;
1757
1758         rc = lmv_check_connect(obd);
1759         if (rc)
1760                 return rc;
1761
1762         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1763                LL_IT2STR(it), PFID(&op_data->op_fid1));
1764
1765         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1766         if (IS_ERR(tgt))
1767                 return PTR_ERR(tgt);
1768
1769         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1770                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1771
1772         rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh,
1773                         lmm, lmmsize, req, extra_lock_flags);
1774
1775         if (rc == 0 && it && it->it_op == IT_OPEN) {
1776                 rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
1777                                         lmm, lmmsize, extra_lock_flags);
1778         }
1779         return rc;
1780 }
1781
1782 static int
1783 lmv_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
1784                  struct ptlrpc_request **request)
1785 {
1786         struct ptlrpc_request   *req = NULL;
1787         struct obd_device       *obd = exp->exp_obd;
1788         struct lmv_obd    *lmv = &obd->u.lmv;
1789         struct lmv_tgt_desc     *tgt;
1790         struct mdt_body  *body;
1791         int                   rc;
1792
1793         rc = lmv_check_connect(obd);
1794         if (rc)
1795                 return rc;
1796
1797         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1798         if (IS_ERR(tgt))
1799                 return PTR_ERR(tgt);
1800
1801         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1802                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1803                tgt->ltd_idx);
1804
1805         rc = md_getattr_name(tgt->ltd_exp, op_data, request);
1806         if (rc != 0)
1807                 return rc;
1808
1809         body = req_capsule_server_get(&(*request)->rq_pill,
1810                                       &RMF_MDT_BODY);
1811         LASSERT(body != NULL);
1812
1813         if (body->valid & OBD_MD_MDS) {
1814                 struct lu_fid rid = body->fid1;
1815
1816                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1817                        PFID(&rid));
1818
1819                 tgt = lmv_find_target(lmv, &rid);
1820                 if (IS_ERR(tgt)) {
1821                         ptlrpc_req_finished(*request);
1822                         return PTR_ERR(tgt);
1823                 }
1824
1825                 op_data->op_fid1 = rid;
1826                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1827                 op_data->op_namelen = 0;
1828                 op_data->op_name = NULL;
1829                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1830                 ptlrpc_req_finished(*request);
1831                 *request = req;
1832         }
1833
1834         return rc;
1835 }
1836
1837 #define md_op_data_fid(op_data, fl)                  \
1838         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1839          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1840          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1841          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1842          NULL)
1843
1844 static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data,
1845                             int op_tgt, ldlm_mode_t mode, int bits, int flag)
1846 {
1847         struct lu_fid     *fid = md_op_data_fid(op_data, flag);
1848         struct obd_device      *obd = exp->exp_obd;
1849         struct lmv_obd   *lmv = &obd->u.lmv;
1850         struct lmv_tgt_desc    *tgt;
1851         ldlm_policy_data_t      policy = { {0} };
1852         int                  rc = 0;
1853
1854         if (!fid_is_sane(fid))
1855                 return 0;
1856
1857         tgt = lmv_find_target(lmv, fid);
1858         if (IS_ERR(tgt))
1859                 return PTR_ERR(tgt);
1860
1861         if (tgt->ltd_idx != op_tgt) {
1862                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1863                 policy.l_inodebits.bits = bits;
1864                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1865                                       mode, LCF_ASYNC, NULL);
1866         } else {
1867                 CDEBUG(D_INODE,
1868                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1869                        op_tgt, PFID(fid));
1870                 op_data->op_flags |= flag;
1871                 rc = 0;
1872         }
1873
1874         return rc;
1875 }
1876
1877 /*
1878  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1879  * op_data->op_fid2
1880  */
1881 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1882                     struct ptlrpc_request **request)
1883 {
1884         struct obd_device       *obd = exp->exp_obd;
1885         struct lmv_obd    *lmv = &obd->u.lmv;
1886         struct lmv_tgt_desc     *tgt;
1887         int                   rc;
1888
1889         rc = lmv_check_connect(obd);
1890         if (rc)
1891                 return rc;
1892
1893         LASSERT(op_data->op_namelen != 0);
1894
1895         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1896                PFID(&op_data->op_fid2), op_data->op_namelen,
1897                op_data->op_name, PFID(&op_data->op_fid1));
1898
1899         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
1900         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
1901         op_data->op_cap = cfs_curproc_cap_pack();
1902         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1903         if (IS_ERR(tgt))
1904                 return PTR_ERR(tgt);
1905
1906         /*
1907          * Cancel UPDATE lock on child (fid1).
1908          */
1909         op_data->op_flags |= MF_MDC_CANCEL_FID2;
1910         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
1911                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
1912         if (rc != 0)
1913                 return rc;
1914
1915         rc = md_link(tgt->ltd_exp, op_data, request);
1916
1917         return rc;
1918 }
1919
1920 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
1921                       const char *old, int oldlen, const char *new, int newlen,
1922                       struct ptlrpc_request **request)
1923 {
1924         struct obd_device       *obd = exp->exp_obd;
1925         struct lmv_obd    *lmv = &obd->u.lmv;
1926         struct lmv_tgt_desc     *src_tgt;
1927         struct lmv_tgt_desc     *tgt_tgt;
1928         int                     rc;
1929
1930         LASSERT(oldlen != 0);
1931
1932         CDEBUG(D_INODE, "RENAME %*s in "DFID" to %*s in "DFID"\n",
1933                oldlen, old, PFID(&op_data->op_fid1),
1934                newlen, new, PFID(&op_data->op_fid2));
1935
1936         rc = lmv_check_connect(obd);
1937         if (rc)
1938                 return rc;
1939
1940         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
1941         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
1942         op_data->op_cap = cfs_curproc_cap_pack();
1943         src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1944         if (IS_ERR(src_tgt))
1945                 return PTR_ERR(src_tgt);
1946
1947         tgt_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1948         if (IS_ERR(tgt_tgt))
1949                 return PTR_ERR(tgt_tgt);
1950         /*
1951          * LOOKUP lock on src child (fid3) should also be cancelled for
1952          * src_tgt in mdc_rename.
1953          */
1954         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
1955
1956         /*
1957          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
1958          * own target.
1959          */
1960         rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
1961                               LCK_EX, MDS_INODELOCK_UPDATE,
1962                               MF_MDC_CANCEL_FID2);
1963
1964         /*
1965          * Cancel LOOKUP locks on tgt child (fid4) for parent tgt_tgt.
1966          */
1967         if (rc == 0) {
1968                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
1969                                       LCK_EX, MDS_INODELOCK_LOOKUP,
1970                                       MF_MDC_CANCEL_FID4);
1971         }
1972
1973         /*
1974          * Cancel all the locks on tgt child (fid4).
1975          */
1976         if (rc == 0)
1977                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
1978                                       LCK_EX, MDS_INODELOCK_FULL,
1979                                       MF_MDC_CANCEL_FID4);
1980
1981         if (rc == 0)
1982                 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen,
1983                                new, newlen, request);
1984         return rc;
1985 }
1986
1987 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
1988                        void *ea, int ealen, void *ea2, int ea2len,
1989                        struct ptlrpc_request **request,
1990                        struct md_open_data **mod)
1991 {
1992         struct obd_device       *obd = exp->exp_obd;
1993         struct lmv_obd    *lmv = &obd->u.lmv;
1994         struct lmv_tgt_desc     *tgt;
1995         int                   rc;
1996
1997         rc = lmv_check_connect(obd);
1998         if (rc)
1999                 return rc;
2000
2001         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2002                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2003
2004         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2005         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2006         if (IS_ERR(tgt))
2007                 return PTR_ERR(tgt);
2008
2009         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2010                         ea2len, request, mod);
2011
2012         return rc;
2013 }
2014
2015 static int lmv_sync(struct obd_export *exp, const struct lu_fid *fid,
2016                     struct ptlrpc_request **request)
2017 {
2018         struct obd_device        *obd = exp->exp_obd;
2019         struct lmv_obd      *lmv = &obd->u.lmv;
2020         struct lmv_tgt_desc       *tgt;
2021         int                     rc;
2022
2023         rc = lmv_check_connect(obd);
2024         if (rc)
2025                 return rc;
2026
2027         tgt = lmv_find_target(lmv, fid);
2028         if (IS_ERR(tgt))
2029                 return PTR_ERR(tgt);
2030
2031         rc = md_sync(tgt->ltd_exp, fid, request);
2032         return rc;
2033 }
2034
2035 /*
2036  * Adjust a set of pages, each page containing an array of lu_dirpages,
2037  * so that each page can be used as a single logical lu_dirpage.
2038  *
2039  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
2040  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
2041  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
2042  * value is used as a cookie to request the next lu_dirpage in a
2043  * directory listing that spans multiple pages (two in this example):
2044  *   ________
2045  *  |   |
2046  * .|--------v-------   -----.
2047  * |s|e|f|p|ent|ent| ... |ent|
2048  * '--|--------------   -----'   Each CFS_PAGE contains a single
2049  *    '------.             lu_dirpage.
2050  * .---------v-------   -----.
2051  * |s|e|f|p|ent| 0 | ... | 0 |
2052  * '-----------------   -----'
2053  *
2054  * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
2055  * larger than LU_PAGE_SIZE, a single host page may contain multiple
2056  * lu_dirpages. After reading the lu_dirpages from the MDS, the
2057  * ldp_hash_end of the first lu_dirpage refers to the one immediately
2058  * after it in the same CFS_PAGE (arrows simplified for brevity, but
2059  * in general e0==s1, e1==s2, etc.):
2060  *
2061  * .--------------------   -----.
2062  * |s0|e0|f0|p|ent|ent| ... |ent|
2063  * |---v----------------   -----|
2064  * |s1|e1|f1|p|ent|ent| ... |ent|
2065  * |---v----------------   -----|  Here, each CFS_PAGE contains
2066  *           ...                 multiple lu_dirpages.
2067  * |---v----------------   -----|
2068  * |s'|e'|f'|p|ent|ent| ... |ent|
2069  * '---|----------------   -----'
2070  *     v
2071  * .----------------------------.
2072  * |    next CFS_PAGE       |
2073  *
2074  * This structure is transformed into a single logical lu_dirpage as follows:
2075  *
2076  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
2077  *   labeled 'next CFS_PAGE'.
2078  *
2079  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
2080  *   a hash collision with the next page exists.
2081  *
2082  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
2083  *   to the first entry of the next lu_dirpage.
2084  */
2085 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
2086 static void lmv_adjust_dirpages(struct page **pages, int ncfspgs, int nlupgs)
2087 {
2088         int i;
2089
2090         for (i = 0; i < ncfspgs; i++) {
2091                 struct lu_dirpage       *dp = kmap(pages[i]);
2092                 struct lu_dirpage       *first = dp;
2093                 struct lu_dirent        *end_dirent = NULL;
2094                 struct lu_dirent        *ent;
2095                 __u64                   hash_end = dp->ldp_hash_end;
2096                 __u32                   flags = dp->ldp_flags;
2097
2098                 while (--nlupgs > 0) {
2099                         ent = lu_dirent_start(dp);
2100                         for (end_dirent = ent; ent != NULL;
2101                              end_dirent = ent, ent = lu_dirent_next(ent))
2102                                 ;
2103
2104                         /* Advance dp to next lu_dirpage. */
2105                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
2106
2107                         /* Check if we've reached the end of the CFS_PAGE. */
2108                         if (!((unsigned long)dp & ~CFS_PAGE_MASK))
2109                                 break;
2110
2111                         /* Save the hash and flags of this lu_dirpage. */
2112                         hash_end = dp->ldp_hash_end;
2113                         flags = dp->ldp_flags;
2114
2115                         /* Check if lu_dirpage contains no entries. */
2116                         if (!end_dirent)
2117                                 break;
2118
2119                         /* Enlarge the end entry lde_reclen from 0 to
2120                          * first entry of next lu_dirpage. */
2121                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
2122                         end_dirent->lde_reclen =
2123                                 cpu_to_le16((char *)(dp->ldp_entries) -
2124                                             (char *)end_dirent);
2125                 }
2126
2127                 first->ldp_hash_end = hash_end;
2128                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
2129                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
2130
2131                 kunmap(pages[i]);
2132         }
2133         LASSERTF(nlupgs == 0, "left = %d", nlupgs);
2134 }
2135 #else
2136 #define lmv_adjust_dirpages(pages, ncfspgs, nlupgs) do {} while (0)
2137 #endif  /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
2138
2139 static int lmv_readpage(struct obd_export *exp, struct md_op_data *op_data,
2140                         struct page **pages, struct ptlrpc_request **request)
2141 {
2142         struct obd_device       *obd = exp->exp_obd;
2143         struct lmv_obd          *lmv = &obd->u.lmv;
2144         __u64                   offset = op_data->op_offset;
2145         int                     rc;
2146         int                     ncfspgs; /* pages read in PAGE_CACHE_SIZE */
2147         int                     nlupgs; /* pages read in LU_PAGE_SIZE */
2148         struct lmv_tgt_desc     *tgt;
2149
2150         rc = lmv_check_connect(obd);
2151         if (rc)
2152                 return rc;
2153
2154         CDEBUG(D_INODE, "READPAGE at %#llx from "DFID"\n",
2155                offset, PFID(&op_data->op_fid1));
2156
2157         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2158         if (IS_ERR(tgt))
2159                 return PTR_ERR(tgt);
2160
2161         rc = md_readpage(tgt->ltd_exp, op_data, pages, request);
2162         if (rc != 0)
2163                 return rc;
2164
2165         ncfspgs = ((*request)->rq_bulk->bd_nob_transferred + PAGE_CACHE_SIZE - 1)
2166                  >> PAGE_CACHE_SHIFT;
2167         nlupgs = (*request)->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
2168         LASSERT(!((*request)->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
2169         LASSERT(ncfspgs > 0 && ncfspgs <= op_data->op_npages);
2170
2171         CDEBUG(D_INODE, "read %d(%d)/%d pages\n", ncfspgs, nlupgs,
2172                op_data->op_npages);
2173
2174         lmv_adjust_dirpages(pages, ncfspgs, nlupgs);
2175
2176         return rc;
2177 }
2178
2179 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2180                       struct ptlrpc_request **request)
2181 {
2182         struct obd_device       *obd = exp->exp_obd;
2183         struct lmv_obd    *lmv = &obd->u.lmv;
2184         struct lmv_tgt_desc     *tgt = NULL;
2185         struct mdt_body         *body;
2186         int                  rc;
2187
2188         rc = lmv_check_connect(obd);
2189         if (rc)
2190                 return rc;
2191 retry:
2192         /* Send unlink requests to the MDT where the child is located */
2193         if (likely(!fid_is_zero(&op_data->op_fid2)))
2194                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2195         else
2196                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2197         if (IS_ERR(tgt))
2198                 return PTR_ERR(tgt);
2199
2200         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2201         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2202         op_data->op_cap = cfs_curproc_cap_pack();
2203
2204         /*
2205          * If child's fid is given, cancel unused locks for it if it is from
2206          * another export than parent.
2207          *
2208          * LOOKUP lock for child (fid3) should also be cancelled on parent
2209          * tgt_tgt in mdc_unlink().
2210          */
2211         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2212
2213         /*
2214          * Cancel FULL locks on child (fid3).
2215          */
2216         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2217                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2218
2219         if (rc != 0)
2220                 return rc;
2221
2222         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2223                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2224
2225         rc = md_unlink(tgt->ltd_exp, op_data, request);
2226         if (rc != 0 && rc != -EREMOTE)
2227                 return rc;
2228
2229         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2230         if (body == NULL)
2231                 return -EPROTO;
2232
2233         /* Not cross-ref case, just get out of here. */
2234         if (likely(!(body->valid & OBD_MD_MDS)))
2235                 return 0;
2236
2237         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2238                exp->exp_obd->obd_name, PFID(&body->fid1));
2239
2240         /* This is a remote object, try remote MDT, Note: it may
2241          * try more than 1 time here, Considering following case
2242          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2243          * 1. Initially A does not know where remote1 is, it send
2244          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2245          *    resend unlink RPC to MDT1 (retry 1st time).
2246          *
2247          * 2. During the unlink RPC in flight,
2248          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2249          *    and create new remote1, but on MDT0
2250          *
2251          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2252          *    /mnt/lustre, then lookup get fid of remote1, and find
2253          *    it is remote dir again, and replay -EREMOTE again.
2254          *
2255          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2256          *
2257          * In theory, it might try unlimited time here, but it should
2258          * be very rare case.  */
2259         op_data->op_fid2 = body->fid1;
2260         ptlrpc_req_finished(*request);
2261         *request = NULL;
2262
2263         goto retry;
2264 }
2265
2266 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2267 {
2268         struct lmv_obd *lmv = &obd->u.lmv;
2269
2270         switch (stage) {
2271         case OBD_CLEANUP_EARLY:
2272                 /* XXX: here should be calling obd_precleanup() down to
2273                  * stack. */
2274                 break;
2275         case OBD_CLEANUP_EXPORTS:
2276                 fld_client_debugfs_fini(&lmv->lmv_fld);
2277                 lprocfs_obd_cleanup(obd);
2278                 break;
2279         default:
2280                 break;
2281         }
2282         return 0;
2283 }
2284
2285 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2286                         __u32 keylen, void *key, __u32 *vallen, void *val,
2287                         struct lov_stripe_md *lsm)
2288 {
2289         struct obd_device       *obd;
2290         struct lmv_obd    *lmv;
2291         int                   rc = 0;
2292
2293         obd = class_exp2obd(exp);
2294         if (obd == NULL) {
2295                 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
2296                        exp->exp_handle.h_cookie);
2297                 return -EINVAL;
2298         }
2299
2300         lmv = &obd->u.lmv;
2301         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2302                 struct lmv_tgt_desc *tgt;
2303                 int i;
2304
2305                 rc = lmv_check_connect(obd);
2306                 if (rc)
2307                         return rc;
2308
2309                 LASSERT(*vallen == sizeof(__u32));
2310                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2311                         tgt = lmv->tgts[i];
2312                         /*
2313                          * All tgts should be connected when this gets called.
2314                          */
2315                         if (tgt == NULL || tgt->ltd_exp == NULL)
2316                                 continue;
2317
2318                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2319                                           vallen, val, NULL))
2320                                 return 0;
2321                 }
2322                 return -EINVAL;
2323         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2324                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2325                    KEY_IS(KEY_CONN_DATA)) {
2326                 rc = lmv_check_connect(obd);
2327                 if (rc)
2328                         return rc;
2329
2330                 /*
2331                  * Forwarding this request to first MDS, it should know LOV
2332                  * desc.
2333                  */
2334                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2335                                   vallen, val, NULL);
2336                 if (!rc && KEY_IS(KEY_CONN_DATA))
2337                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2338                 return rc;
2339         } else if (KEY_IS(KEY_TGT_COUNT)) {
2340                 *((int *)val) = lmv->desc.ld_tgt_count;
2341                 return 0;
2342         }
2343
2344         CDEBUG(D_IOCTL, "Invalid key\n");
2345         return -EINVAL;
2346 }
2347
2348 static int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2349                               u32 keylen, void *key, u32 vallen,
2350                               void *val, struct ptlrpc_request_set *set)
2351 {
2352         struct lmv_tgt_desc    *tgt;
2353         struct obd_device      *obd;
2354         struct lmv_obd   *lmv;
2355         int rc = 0;
2356
2357         obd = class_exp2obd(exp);
2358         if (obd == NULL) {
2359                 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
2360                        exp->exp_handle.h_cookie);
2361                 return -EINVAL;
2362         }
2363         lmv = &obd->u.lmv;
2364
2365         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2366                 int i, err = 0;
2367
2368                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2369                         tgt = lmv->tgts[i];
2370
2371                         if (tgt == NULL || tgt->ltd_exp == NULL)
2372                                 continue;
2373
2374                         err = obd_set_info_async(env, tgt->ltd_exp,
2375                                                  keylen, key, vallen, val, set);
2376                         if (err && rc == 0)
2377                                 rc = err;
2378                 }
2379
2380                 return rc;
2381         }
2382
2383         return -EINVAL;
2384 }
2385
2386 static int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2387                       struct lov_stripe_md *lsm)
2388 {
2389         struct obd_device        *obd = class_exp2obd(exp);
2390         struct lmv_obd      *lmv = &obd->u.lmv;
2391         struct lmv_stripe_md      *meap;
2392         struct lmv_stripe_md      *lsmp;
2393         int                     mea_size;
2394         int                     i;
2395
2396         mea_size = lmv_get_easize(lmv);
2397         if (!lmmp)
2398                 return mea_size;
2399
2400         if (*lmmp && !lsm) {
2401                 kvfree(*lmmp);
2402                 *lmmp = NULL;
2403                 return 0;
2404         }
2405
2406         if (*lmmp == NULL) {
2407                 *lmmp = libcfs_kvzalloc(mea_size, GFP_NOFS);
2408                 if (*lmmp == NULL)
2409                         return -ENOMEM;
2410         }
2411
2412         if (!lsm)
2413                 return mea_size;
2414
2415         lsmp = (struct lmv_stripe_md *)lsm;
2416         meap = (struct lmv_stripe_md *)*lmmp;
2417
2418         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2419             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2420                 return -EINVAL;
2421
2422         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2423         meap->mea_count = cpu_to_le32(lsmp->mea_count);
2424         meap->mea_master = cpu_to_le32(lsmp->mea_master);
2425
2426         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2427                 meap->mea_ids[i] = lsmp->mea_ids[i];
2428                 fid_cpu_to_le(&meap->mea_ids[i], &lsmp->mea_ids[i]);
2429         }
2430
2431         return mea_size;
2432 }
2433
2434 static int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2435                         struct lov_mds_md *lmm, int lmm_size)
2436 {
2437         struct obd_device         *obd = class_exp2obd(exp);
2438         struct lmv_stripe_md      **tmea = (struct lmv_stripe_md **)lsmp;
2439         struct lmv_stripe_md       *mea = (struct lmv_stripe_md *)lmm;
2440         struct lmv_obd       *lmv = &obd->u.lmv;
2441         int                      mea_size;
2442         int                      i;
2443         __u32                  magic;
2444
2445         mea_size = lmv_get_easize(lmv);
2446         if (lsmp == NULL)
2447                 return mea_size;
2448
2449         if (*lsmp != NULL && lmm == NULL) {
2450                 kvfree(*tmea);
2451                 *lsmp = NULL;
2452                 return 0;
2453         }
2454
2455         LASSERT(mea_size == lmm_size);
2456
2457         *tmea = libcfs_kvzalloc(mea_size, GFP_NOFS);
2458         if (*tmea == NULL)
2459                 return -ENOMEM;
2460
2461         if (!lmm)
2462                 return mea_size;
2463
2464         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2465             mea->mea_magic == MEA_MAGIC_ALL_CHARS ||
2466             mea->mea_magic == MEA_MAGIC_HASH_SEGMENT) {
2467                 magic = le32_to_cpu(mea->mea_magic);
2468         } else {
2469                 /*
2470                  * Old mea is not handled here.
2471                  */
2472                 CERROR("Old not supportable EA is found\n");
2473                 LBUG();
2474         }
2475
2476         (*tmea)->mea_magic = magic;
2477         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2478         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2479
2480         for (i = 0; i < (*tmea)->mea_count; i++) {
2481                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2482                 fid_le_to_cpu(&(*tmea)->mea_ids[i], &(*tmea)->mea_ids[i]);
2483         }
2484         return mea_size;
2485 }
2486
2487 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2488                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
2489                              ldlm_cancel_flags_t flags, void *opaque)
2490 {
2491         struct obd_device       *obd = exp->exp_obd;
2492         struct lmv_obd    *lmv = &obd->u.lmv;
2493         int                   rc = 0;
2494         int                   err;
2495         int                   i;
2496
2497         LASSERT(fid != NULL);
2498
2499         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2500                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL ||
2501                     lmv->tgts[i]->ltd_active == 0)
2502                         continue;
2503
2504                 err = md_cancel_unused(lmv->tgts[i]->ltd_exp, fid,
2505                                        policy, mode, flags, opaque);
2506                 if (!rc)
2507                         rc = err;
2508         }
2509         return rc;
2510 }
2511
2512 static int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
2513                              __u64 *bits)
2514 {
2515         struct lmv_obd    *lmv = &exp->exp_obd->u.lmv;
2516         int                   rc;
2517
2518         rc =  md_set_lock_data(lmv->tgts[0]->ltd_exp, lockh, data, bits);
2519         return rc;
2520 }
2521
2522 static ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
2523                                   const struct lu_fid *fid, ldlm_type_t type,
2524                                   ldlm_policy_data_t *policy, ldlm_mode_t mode,
2525                                   struct lustre_handle *lockh)
2526 {
2527         struct obd_device       *obd = exp->exp_obd;
2528         struct lmv_obd    *lmv = &obd->u.lmv;
2529         ldlm_mode_t           rc;
2530         int                   i;
2531
2532         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2533
2534         /*
2535          * With CMD every object can have two locks in different namespaces:
2536          * lookup lock in space of mds storing direntry and update/open lock in
2537          * space of mds storing inode. Thus we check all targets, not only that
2538          * one fid was created in.
2539          */
2540         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2541                 if (lmv->tgts[i] == NULL ||
2542                     lmv->tgts[i]->ltd_exp == NULL ||
2543                     lmv->tgts[i]->ltd_active == 0)
2544                         continue;
2545
2546                 rc = md_lock_match(lmv->tgts[i]->ltd_exp, flags, fid,
2547                                    type, policy, mode, lockh);
2548                 if (rc)
2549                         return rc;
2550         }
2551
2552         return 0;
2553 }
2554
2555 static int lmv_get_lustre_md(struct obd_export *exp,
2556                              struct ptlrpc_request *req,
2557                              struct obd_export *dt_exp,
2558                              struct obd_export *md_exp,
2559                              struct lustre_md *md)
2560 {
2561         struct lmv_obd    *lmv = &exp->exp_obd->u.lmv;
2562
2563         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
2564 }
2565
2566 static int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2567 {
2568         struct obd_device       *obd = exp->exp_obd;
2569         struct lmv_obd    *lmv = &obd->u.lmv;
2570
2571         if (md->mea)
2572                 obd_free_memmd(exp, (void *)&md->mea);
2573         return md_free_lustre_md(lmv->tgts[0]->ltd_exp, md);
2574 }
2575
2576 static int lmv_set_open_replay_data(struct obd_export *exp,
2577                                     struct obd_client_handle *och,
2578                                     struct lookup_intent *it)
2579 {
2580         struct obd_device       *obd = exp->exp_obd;
2581         struct lmv_obd    *lmv = &obd->u.lmv;
2582         struct lmv_tgt_desc     *tgt;
2583
2584         tgt = lmv_find_target(lmv, &och->och_fid);
2585         if (IS_ERR(tgt))
2586                 return PTR_ERR(tgt);
2587
2588         return md_set_open_replay_data(tgt->ltd_exp, och, it);
2589 }
2590
2591 static int lmv_clear_open_replay_data(struct obd_export *exp,
2592                                       struct obd_client_handle *och)
2593 {
2594         struct obd_device       *obd = exp->exp_obd;
2595         struct lmv_obd    *lmv = &obd->u.lmv;
2596         struct lmv_tgt_desc     *tgt;
2597
2598         tgt = lmv_find_target(lmv, &och->och_fid);
2599         if (IS_ERR(tgt))
2600                 return PTR_ERR(tgt);
2601
2602         return md_clear_open_replay_data(tgt->ltd_exp, och);
2603 }
2604
2605 static int lmv_get_remote_perm(struct obd_export *exp,
2606                                const struct lu_fid *fid,
2607                                __u32 suppgid, struct ptlrpc_request **request)
2608 {
2609         struct obd_device       *obd = exp->exp_obd;
2610         struct lmv_obd    *lmv = &obd->u.lmv;
2611         struct lmv_tgt_desc     *tgt;
2612         int                   rc;
2613
2614         rc = lmv_check_connect(obd);
2615         if (rc)
2616                 return rc;
2617
2618         tgt = lmv_find_target(lmv, fid);
2619         if (IS_ERR(tgt))
2620                 return PTR_ERR(tgt);
2621
2622         rc = md_get_remote_perm(tgt->ltd_exp, fid, suppgid, request);
2623         return rc;
2624 }
2625
2626 static int lmv_intent_getattr_async(struct obd_export *exp,
2627                                     struct md_enqueue_info *minfo,
2628                                     struct ldlm_enqueue_info *einfo)
2629 {
2630         struct md_op_data       *op_data = &minfo->mi_data;
2631         struct obd_device       *obd = exp->exp_obd;
2632         struct lmv_obd    *lmv = &obd->u.lmv;
2633         struct lmv_tgt_desc     *tgt = NULL;
2634         int                   rc;
2635
2636         rc = lmv_check_connect(obd);
2637         if (rc)
2638                 return rc;
2639
2640         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2641         if (IS_ERR(tgt))
2642                 return PTR_ERR(tgt);
2643
2644         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
2645         return rc;
2646 }
2647
2648 static int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
2649                                struct lu_fid *fid, __u64 *bits)
2650 {
2651         struct obd_device       *obd = exp->exp_obd;
2652         struct lmv_obd    *lmv = &obd->u.lmv;
2653         struct lmv_tgt_desc     *tgt;
2654         int                   rc;
2655
2656         rc = lmv_check_connect(obd);
2657         if (rc)
2658                 return rc;
2659
2660         tgt = lmv_find_target(lmv, fid);
2661         if (IS_ERR(tgt))
2662                 return PTR_ERR(tgt);
2663
2664         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
2665         return rc;
2666 }
2667
2668 /**
2669  * For lmv, only need to send request to master MDT, and the master MDT will
2670  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
2671  * we directly fetch data from the slave MDTs.
2672  */
2673 static int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
2674                         struct obd_quotactl *oqctl)
2675 {
2676         struct obd_device   *obd = class_exp2obd(exp);
2677         struct lmv_obd      *lmv = &obd->u.lmv;
2678         struct lmv_tgt_desc *tgt = lmv->tgts[0];
2679         int               rc = 0, i;
2680         __u64           curspace, curinodes;
2681
2682         if (!lmv->desc.ld_tgt_count || !tgt->ltd_active) {
2683                 CERROR("master lmv inactive\n");
2684                 return -EIO;
2685         }
2686
2687         if (oqctl->qc_cmd != Q_GETOQUOTA) {
2688                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2689                 return rc;
2690         }
2691
2692         curspace = curinodes = 0;
2693         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2694                 int err;
2695
2696                 tgt = lmv->tgts[i];
2697
2698                 if (tgt == NULL || tgt->ltd_exp == NULL || tgt->ltd_active == 0)
2699                         continue;
2700                 if (!tgt->ltd_active) {
2701                         CDEBUG(D_HA, "mdt %d is inactive.\n", i);
2702                         continue;
2703                 }
2704
2705                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2706                 if (err) {
2707                         CERROR("getquota on mdt %d failed. %d\n", i, err);
2708                         if (!rc)
2709                                 rc = err;
2710                 } else {
2711                         curspace += oqctl->qc_dqblk.dqb_curspace;
2712                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
2713                 }
2714         }
2715         oqctl->qc_dqblk.dqb_curspace = curspace;
2716         oqctl->qc_dqblk.dqb_curinodes = curinodes;
2717
2718         return rc;
2719 }
2720
2721 static int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
2722                           struct obd_quotactl *oqctl)
2723 {
2724         struct obd_device   *obd = class_exp2obd(exp);
2725         struct lmv_obd      *lmv = &obd->u.lmv;
2726         struct lmv_tgt_desc *tgt;
2727         int               i, rc = 0;
2728
2729         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2730                 int err;
2731
2732                 tgt = lmv->tgts[i];
2733                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
2734                         CERROR("lmv idx %d inactive\n", i);
2735                         return -EIO;
2736                 }
2737
2738                 err = obd_quotacheck(tgt->ltd_exp, oqctl);
2739                 if (err && !rc)
2740                         rc = err;
2741         }
2742
2743         return rc;
2744 }
2745
2746 static struct obd_ops lmv_obd_ops = {
2747         .o_owner                = THIS_MODULE,
2748         .o_setup                = lmv_setup,
2749         .o_cleanup            = lmv_cleanup,
2750         .o_precleanup      = lmv_precleanup,
2751         .o_process_config       = lmv_process_config,
2752         .o_connect            = lmv_connect,
2753         .o_disconnect      = lmv_disconnect,
2754         .o_statfs              = lmv_statfs,
2755         .o_get_info          = lmv_get_info,
2756         .o_set_info_async       = lmv_set_info_async,
2757         .o_packmd              = lmv_packmd,
2758         .o_unpackmd          = lmv_unpackmd,
2759         .o_notify              = lmv_notify,
2760         .o_get_uuid          = lmv_get_uuid,
2761         .o_iocontrol        = lmv_iocontrol,
2762         .o_quotacheck      = lmv_quotacheck,
2763         .o_quotactl          = lmv_quotactl
2764 };
2765
2766 static struct md_ops lmv_md_ops = {
2767         .m_getstatus        = lmv_getstatus,
2768         .m_null_inode           = lmv_null_inode,
2769         .m_find_cbdata    = lmv_find_cbdata,
2770         .m_close                = lmv_close,
2771         .m_create              = lmv_create,
2772         .m_done_writing  = lmv_done_writing,
2773         .m_enqueue            = lmv_enqueue,
2774         .m_getattr            = lmv_getattr,
2775         .m_getxattr          = lmv_getxattr,
2776         .m_getattr_name  = lmv_getattr_name,
2777         .m_intent_lock    = lmv_intent_lock,
2778         .m_link          = lmv_link,
2779         .m_rename              = lmv_rename,
2780         .m_setattr            = lmv_setattr,
2781         .m_setxattr          = lmv_setxattr,
2782         .m_sync          = lmv_sync,
2783         .m_readpage          = lmv_readpage,
2784         .m_unlink              = lmv_unlink,
2785         .m_init_ea_size  = lmv_init_ea_size,
2786         .m_cancel_unused        = lmv_cancel_unused,
2787         .m_set_lock_data        = lmv_set_lock_data,
2788         .m_lock_match      = lmv_lock_match,
2789         .m_get_lustre_md        = lmv_get_lustre_md,
2790         .m_free_lustre_md       = lmv_free_lustre_md,
2791         .m_set_open_replay_data = lmv_set_open_replay_data,
2792         .m_clear_open_replay_data = lmv_clear_open_replay_data,
2793         .m_get_remote_perm      = lmv_get_remote_perm,
2794         .m_intent_getattr_async = lmv_intent_getattr_async,
2795         .m_revalidate_lock      = lmv_revalidate_lock
2796 };
2797
2798 static int __init lmv_init(void)
2799 {
2800         struct lprocfs_static_vars lvars;
2801         int                     rc;
2802
2803         lprocfs_lmv_init_vars(&lvars);
2804
2805         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2806                                  LUSTRE_LMV_NAME, NULL);
2807         return rc;
2808 }
2809
2810 static void lmv_exit(void)
2811 {
2812         class_unregister_type(LUSTRE_LMV_NAME);
2813 }
2814
2815 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2816 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2817 MODULE_LICENSE("GPL");
2818
2819 module_init(lmv_init);
2820 module_exit(lmv_exit);