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, 2014, 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/user_namespace.h>
  42 #ifdef HAVE_UIDGID_HEADER
  43 # include <linux/uidgid.h>
  44 #endif
  45 #include <linux/slab.h>
  46 #include <linux/pagemap.h>
  47 #include <linux/mm.h>
  48 #include <linux/math64.h>
  49 #include <linux/seq_file.h>
  50 #include <linux/namei.h>
  51
  52 #include <lustre/lustre_idl.h>
  53 #include <obd_support.h>
  54 #include <lustre_lib.h>
  55 #include <lustre_net.h>
  56 #include <obd_class.h>
  57 #include <lustre_lmv.h>
  58 #include <lprocfs_status.h>
  59 #include <cl_object.h>
  60 #include <lustre_fid.h>
  61 #include <lustre_ioctl.h>
  62 #include <lustre_kernelcomm.h>
  63 #include "lmv_internal.h"
  64
  65 static void lmv_activate_target(struct lmv_obd *lmv,
  66                                 struct lmv_tgt_desc *tgt,
  67                                 int activate)
  68 {
  69         if (tgt->ltd_active == activate)
  70                 return;
  71
  72         tgt->ltd_active = activate;
  73         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  74 }
  75
  76 /**
  77  * Error codes:
  78  *
  79  *  -EINVAL  : UUID can't be found in the LMV's target list
  80  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  81  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  82  */
  83 static int lmv_set_mdc_active(struct lmv_obd *lmv,
  84                               const struct obd_uuid *uuid,
  85                               int activate)
  86 {
  87         struct lmv_tgt_desc     *tgt = NULL;
  88         struct obd_device       *obd;
  89         __u32                    i;
  90         int                      rc = 0;
  91         ENTRY;
  92
  93         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  94                         lmv, uuid->uuid, activate);
  95
  96         spin_lock(&lmv->lmv_lock);
  97         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
  98                 tgt = lmv->tgts[i];
  99                 if (tgt == NULL || tgt->ltd_exp == NULL)
 100                         continue;
 101
 102                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
 103                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 104
 105                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 106                         break;
 107         }
 108
 109         if (i == lmv->desc.ld_tgt_count)
 110                 GOTO(out_lmv_lock, rc = -EINVAL);
 111
 112         obd = class_exp2obd(tgt->ltd_exp);
 113         if (obd == NULL)
 114                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 115
 116         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 117                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 118                obd->obd_type->typ_name, i);
 119         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 120
 121         if (tgt->ltd_active == activate) {
 122                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 123                        activate ? "" : "in");
 124                 GOTO(out_lmv_lock, rc);
 125         }
 126
 127         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 128                activate ? "" : "in");
 129         lmv_activate_target(lmv, tgt, activate);
 130         EXIT;
 131
 132  out_lmv_lock:
 133         spin_unlock(&lmv->lmv_lock);
 134         return rc;
 135 }
 136
 137 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 138 {
 139         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
 140         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
 141
 142         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
 143 }
 144
 145 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 146                       enum obd_notify_event ev, void *data)
 147 {
 148         struct obd_connect_data *conn_data;
 149         struct lmv_obd          *lmv = &obd->u.lmv;
 150         struct obd_uuid         *uuid;
 151         int                      rc = 0;
 152         ENTRY;
 153
 154         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 155                 CERROR("unexpected notification of %s %s!\n",
 156                        watched->obd_type->typ_name,
 157                        watched->obd_name);
 158                 RETURN(-EINVAL);
 159         }
 160
 161         uuid = &watched->u.cli.cl_target_uuid;
 162         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 163                 /*
 164                  * Set MDC as active before notifying the observer, so the
 165                  * observer can use the MDC normally.
 166                  */
 167                 rc = lmv_set_mdc_active(lmv, uuid,
 168                                         ev == OBD_NOTIFY_ACTIVE);
 169                 if (rc) {
 170                         CERROR("%sactivation of %s failed: %d\n",
 171                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 172                                uuid->uuid, rc);
 173                         RETURN(rc);
 174                 }
 175         } else if (ev == OBD_NOTIFY_OCD) {
 176                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 177                 /*
 178                  * XXX: Make sure that ocd_connect_flags from all targets are
 179                  * the same. Otherwise one of MDTs runs wrong version or
 180                  * something like this.  --umka
 181                  */
 182                 obd->obd_self_export->exp_connect_data = *conn_data;
 183         }
 184 #if 0
 185         else if (ev == OBD_NOTIFY_DISCON) {
 186                 /*
 187                  * For disconnect event, flush fld cache for failout MDS case.
 188                  */
 189                 fld_client_flush(&lmv->lmv_fld);
 190         }
 191 #endif
 192         /*
 193          * Pass the notification up the chain.
 194          */
 195         if (obd->obd_observer)
 196                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 197
 198         RETURN(rc);
 199 }
 200
 201 /**
 202  * This is fake connect function. Its purpose is to initialize lmv and say
 203  * caller that everything is okay. Real connection will be performed later.
 204  */
 205 static int lmv_connect(const struct lu_env *env,
 206                        struct obd_export **exp, struct obd_device *obd,
 207                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 208                        void *localdata)
 209 {
 210         struct lmv_obd        *lmv = &obd->u.lmv;
 211         struct lustre_handle  conn = { 0 };
 212         int                    rc = 0;
 213         ENTRY;
 214
 215         /*
 216          * We don't want to actually do the underlying connections more than
 217          * once, so keep track.
 218          */
 219         lmv->refcount++;
 220         if (lmv->refcount > 1) {
 221                 *exp = NULL;
 222                 RETURN(0);
 223         }
 224
 225         rc = class_connect(&conn, obd, cluuid);
 226         if (rc) {
 227                 CERROR("class_connection() returned %d\n", rc);
 228                 RETURN(rc);
 229         }
 230
 231         *exp = class_conn2export(&conn);
 232         class_export_get(*exp);
 233
 234         lmv->exp = *exp;
 235         lmv->connected = 0;
 236         lmv->cluuid = *cluuid;
 237
 238         if (data)
 239                 lmv->conn_data = *data;
 240
 241         if (lmv->targets_proc_entry == NULL) {
 242                 lmv->targets_proc_entry = lprocfs_register("target_obds",
 243                                                            obd->obd_proc_entry,
 244                                                            NULL, NULL);
 245                 if (IS_ERR(lmv->targets_proc_entry)) {
 246                         CERROR("%s: cannot register "
 247                                "/proc/fs/lustre/%s/%s/target_obds\n",
 248                                obd->obd_name, obd->obd_type->typ_name,
 249                                obd->obd_name);
 250                         lmv->targets_proc_entry = NULL;
 251                 }
 252         }
 253
 254         /*
 255          * All real clients should perform actual connection right away, because
 256          * it is possible, that LMV will not have opportunity to connect targets
 257          * and MDC stuff will be called directly, for instance while reading
 258          * ../mdc/../kbytesfree procfs file, etc.
 259          */
 260         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
 261                 rc = lmv_check_connect(obd);
 262
 263         if (rc && lmv->targets_proc_entry != NULL)
 264                 lprocfs_remove(&lmv->targets_proc_entry);
 265         RETURN(rc);
 266 }
 267
 268 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
 269                             __u32 def_easize)
 270 {
 271         struct obd_device       *obd = exp->exp_obd;
 272         struct lmv_obd          *lmv = &obd->u.lmv;
 273         __u32                    i;
 274         int                      rc = 0;
 275         int                      change = 0;
 276         ENTRY;
 277
 278         if (lmv->max_easize < easize) {
 279                 lmv->max_easize = easize;
 280                 change = 1;
 281         }
 282         if (lmv->max_def_easize < def_easize) {
 283                 lmv->max_def_easize = def_easize;
 284                 change = 1;
 285         }
 286
 287         if (change == 0)
 288                 RETURN(0);
 289
 290         if (lmv->connected == 0)
 291                 RETURN(0);
 292
 293         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 294                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 295
 296                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
 297                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 298                         continue;
 299                 }
 300
 301                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
 302                 if (rc) {
 303                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 304                                " rc = %d\n", obd->obd_name, i, rc);
 305                         break;
 306                 }
 307         }
 308         RETURN(rc);
 309 }
 310
 311 #define MAX_STRING_SIZE 128
 312
 313 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 314 {
 315         struct lmv_obd          *lmv = &obd->u.lmv;
 316         struct obd_uuid         *cluuid = &lmv->cluuid;
 317         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 318         struct obd_device       *mdc_obd;
 319         struct obd_export       *mdc_exp;
 320         struct lu_fld_target     target;
 321         int                      rc;
 322         ENTRY;
 323
 324         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 325                                         &obd->obd_uuid);
 326         if (!mdc_obd) {
 327                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 328                 RETURN(-EINVAL);
 329         }
 330
 331         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 332                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 333                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 334                 cluuid->uuid);
 335
 336         if (!mdc_obd->obd_set_up) {
 337                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 338                 RETURN(-EINVAL);
 339         }
 340
 341         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 342                          &lmv->conn_data, NULL);
 343         if (rc) {
 344                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 345                 RETURN(rc);
 346         }
 347
 348         /*
 349          * Init fid sequence client for this mdc and add new fld target.
 350          */
 351         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 352         if (rc)
 353                 RETURN(rc);
 354
 355         target.ft_srv = NULL;
 356         target.ft_exp = mdc_exp;
 357         target.ft_idx = tgt->ltd_idx;
 358
 359         fld_client_add_target(&lmv->lmv_fld, &target);
 360
 361         rc = obd_register_observer(mdc_obd, obd);
 362         if (rc) {
 363                 obd_disconnect(mdc_exp);
 364                 CERROR("target %s register_observer error %d\n",
 365                        tgt->ltd_uuid.uuid, rc);
 366                 RETURN(rc);
 367         }
 368
 369         if (obd->obd_observer) {
 370                 /*
 371                  * Tell the observer about the new target.
 372                  */
 373                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 374                                 OBD_NOTIFY_ACTIVE,
 375                                 (void *)(tgt - lmv->tgts[0]));
 376                 if (rc) {
 377                         obd_disconnect(mdc_exp);
 378                         RETURN(rc);
 379                 }
 380         }
 381
 382         tgt->ltd_active = 1;
 383         tgt->ltd_exp = mdc_exp;
 384         lmv->desc.ld_active_tgt_count++;
 385
 386         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
 387
 388         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 389                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 390                 atomic_read(&obd->obd_refcount));
 391
 392         if (lmv->targets_proc_entry != NULL) {
 393                 struct proc_dir_entry *mdc_symlink;
 394
 395                 LASSERT(mdc_obd->obd_type != NULL);
 396                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 397                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 398                                                   lmv->targets_proc_entry,
 399                                                   "../../../%s/%s",
 400                                                   mdc_obd->obd_type->typ_name,
 401                                                   mdc_obd->obd_name);
 402                 if (mdc_symlink == NULL) {
 403                         CERROR("cannot register LMV target "
 404                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
 405                                obd->obd_type->typ_name, obd->obd_name,
 406                                mdc_obd->obd_name);
 407                 }
 408         }
 409         RETURN(0);
 410 }
 411
 412 static void lmv_del_target(struct lmv_obd *lmv, int index)
 413 {
 414         if (lmv->tgts[index] == NULL)
 415                 return;
 416
 417         OBD_FREE_PTR(lmv->tgts[index]);
 418         lmv->tgts[index] = NULL;
 419         return;
 420 }
 421
 422 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 423                            __u32 index, int gen)
 424 {
 425         struct lmv_obd      *lmv = &obd->u.lmv;
 426         struct lmv_tgt_desc *tgt;
 427         int                  orig_tgt_count = 0;
 428         int                  rc = 0;
 429         ENTRY;
 430
 431         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 432
 433         mutex_lock(&lmv->lmv_init_mutex);
 434
 435         if (lmv->desc.ld_tgt_count == 0) {
 436                 struct obd_device *mdc_obd;
 437
 438                 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 439                                                 &obd->obd_uuid);
 440                 if (!mdc_obd) {
 441                         mutex_unlock(&lmv->lmv_init_mutex);
 442                         CERROR("%s: Target %s not attached: rc = %d\n",
 443                                obd->obd_name, uuidp->uuid, -EINVAL);
 444                         RETURN(-EINVAL);
 445                 }
 446         }
 447
 448         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
 449                 tgt = lmv->tgts[index];
 450                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
 451                        " rc = %d\n", obd->obd_name,
 452                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
 453                 mutex_unlock(&lmv->lmv_init_mutex);
 454                 RETURN(-EEXIST);
 455         }
 456
 457         if (index >= lmv->tgts_size) {
 458                 /* We need to reallocate the lmv target array. */
 459                 struct lmv_tgt_desc **newtgts, **old = NULL;
 460                 __u32 newsize = 1;
 461                 __u32 oldsize = 0;
 462
 463                 while (newsize < index + 1)
 464                         newsize = newsize << 1;
 465                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 466                 if (newtgts == NULL) {
 467                         mutex_unlock(&lmv->lmv_init_mutex);
 468                         RETURN(-ENOMEM);
 469                 }
 470
 471                 if (lmv->tgts_size) {
 472                         memcpy(newtgts, lmv->tgts,
 473                                sizeof(*newtgts) * lmv->tgts_size);
 474                         old = lmv->tgts;
 475                         oldsize = lmv->tgts_size;
 476                 }
 477
 478                 lmv->tgts = newtgts;
 479                 lmv->tgts_size = newsize;
 480                 smp_rmb();
 481                 if (old)
 482                         OBD_FREE(old, sizeof(*old) * oldsize);
 483
 484                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
 485                        lmv->tgts_size);
 486         }
 487
 488         OBD_ALLOC_PTR(tgt);
 489         if (!tgt) {
 490                 mutex_unlock(&lmv->lmv_init_mutex);
 491                 RETURN(-ENOMEM);
 492         }
 493
 494         mutex_init(&tgt->ltd_fid_mutex);
 495         tgt->ltd_idx = index;
 496         tgt->ltd_uuid = *uuidp;
 497         tgt->ltd_active = 0;
 498         lmv->tgts[index] = tgt;
 499         if (index >= lmv->desc.ld_tgt_count) {
 500                 orig_tgt_count = lmv->desc.ld_tgt_count;
 501                 lmv->desc.ld_tgt_count = index + 1;
 502         }
 503
 504         if (lmv->connected) {
 505                 rc = lmv_connect_mdc(obd, tgt);
 506                 if (rc != 0) {
 507                         spin_lock(&lmv->lmv_lock);
 508                         if (lmv->desc.ld_tgt_count == index + 1)
 509                                 lmv->desc.ld_tgt_count = orig_tgt_count;
 510                         memset(tgt, 0, sizeof(*tgt));
 511                         spin_unlock(&lmv->lmv_lock);
 512                 } else {
 513                         int easize = sizeof(struct lmv_stripe_md) +
 514                                 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
 515                         lmv_init_ea_size(obd->obd_self_export, easize, 0);
 516                 }
 517         }
 518
 519         mutex_unlock(&lmv->lmv_init_mutex);
 520         RETURN(rc);
 521 }
 522
 523 int lmv_check_connect(struct obd_device *obd)
 524 {
 525         struct lmv_obd          *lmv = &obd->u.lmv;
 526         struct lmv_tgt_desc     *tgt;
 527         __u32                    i;
 528         int                      rc;
 529         int                      easize;
 530         ENTRY;
 531
 532         if (lmv->connected)
 533                 RETURN(0);
 534
 535         mutex_lock(&lmv->lmv_init_mutex);
 536         if (lmv->connected) {
 537                 mutex_unlock(&lmv->lmv_init_mutex);
 538                 RETURN(0);
 539         }
 540
 541         if (lmv->desc.ld_tgt_count == 0) {
 542                 mutex_unlock(&lmv->lmv_init_mutex);
 543                 CERROR("%s: no targets configured.\n", obd->obd_name);
 544                 RETURN(-EINVAL);
 545         }
 546
 547         LASSERT(lmv->tgts != NULL);
 548
 549         if (lmv->tgts[0] == NULL) {
 550                 mutex_unlock(&lmv->lmv_init_mutex);
 551                 CERROR("%s: no target configured for index 0.\n",
 552                        obd->obd_name);
 553                 RETURN(-EINVAL);
 554         }
 555
 556         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 557                lmv->cluuid.uuid, obd->obd_name);
 558
 559         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 560                 tgt = lmv->tgts[i];
 561                 if (tgt == NULL)
 562                         continue;
 563                 rc = lmv_connect_mdc(obd, tgt);
 564                 if (rc)
 565                         GOTO(out_disc, rc);
 566         }
 567
 568         class_export_put(lmv->exp);
 569         lmv->connected = 1;
 570         easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
 571         lmv_init_ea_size(obd->obd_self_export, easize, 0);
 572         mutex_unlock(&lmv->lmv_init_mutex);
 573         RETURN(0);
 574
 575  out_disc:
 576         while (i-- > 0) {
 577                 int rc2;
 578                 tgt = lmv->tgts[i];
 579                 if (tgt == NULL)
 580                         continue;
 581                 tgt->ltd_active = 0;
 582                 if (tgt->ltd_exp) {
 583                         --lmv->desc.ld_active_tgt_count;
 584                         rc2 = obd_disconnect(tgt->ltd_exp);
 585                         if (rc2) {
 586                                 CERROR("LMV target %s disconnect on "
 587                                        "MDC idx %d: error %d\n",
 588                                        tgt->ltd_uuid.uuid, i, rc2);
 589                         }
 590                 }
 591         }
 592         class_disconnect(lmv->exp);
 593         mutex_unlock(&lmv->lmv_init_mutex);
 594         RETURN(rc);
 595 }
 596
 597 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 598 {
 599         struct lmv_obd         *lmv = &obd->u.lmv;
 600         struct obd_device      *mdc_obd;
 601         int                     rc;
 602         ENTRY;
 603
 604         LASSERT(tgt != NULL);
 605         LASSERT(obd != NULL);
 606
 607         mdc_obd = class_exp2obd(tgt->ltd_exp);
 608
 609         if (mdc_obd) {
 610                 mdc_obd->obd_force = obd->obd_force;
 611                 mdc_obd->obd_fail = obd->obd_fail;
 612                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 613
 614                 if (lmv->targets_proc_entry != NULL)
 615                         lprocfs_remove_proc_entry(mdc_obd->obd_name,
 616                                                   lmv->targets_proc_entry);
 617         }
 618
 619         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 620         if (rc)
 621                 CERROR("Can't finanize fids factory\n");
 622
 623         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 624                tgt->ltd_exp->exp_obd->obd_name,
 625                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 626
 627         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 628         rc = obd_disconnect(tgt->ltd_exp);
 629         if (rc) {
 630                 if (tgt->ltd_active) {
 631                         CERROR("Target %s disconnect error %d\n",
 632                                tgt->ltd_uuid.uuid, rc);
 633                 }
 634         }
 635
 636         lmv_activate_target(lmv, tgt, 0);
 637         tgt->ltd_exp = NULL;
 638         RETURN(0);
 639 }
 640
 641 static int lmv_disconnect(struct obd_export *exp)
 642 {
 643         struct obd_device       *obd = class_exp2obd(exp);
 644         struct lmv_obd          *lmv = &obd->u.lmv;
 645         int                      rc;
 646         __u32                    i;
 647         ENTRY;
 648
 649         if (!lmv->tgts)
 650                 goto out_local;
 651
 652         /*
 653          * Only disconnect the underlying layers on the final disconnect.
 654          */
 655         lmv->refcount--;
 656         if (lmv->refcount != 0)
 657                 goto out_local;
 658
 659         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 660                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
 661                         continue;
 662
 663                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
 664         }
 665
 666         if (lmv->targets_proc_entry != NULL)
 667                 lprocfs_remove(&lmv->targets_proc_entry);
 668         else
 669                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 670                        obd->obd_type->typ_name, obd->obd_name);
 671
 672 out_local:
 673         /*
 674          * This is the case when no real connection is established by
 675          * lmv_check_connect().
 676          */
 677         if (!lmv->connected)
 678                 class_export_put(exp);
 679         rc = class_disconnect(exp);
 680         if (lmv->refcount == 0)
 681                 lmv->connected = 0;
 682         RETURN(rc);
 683 }
 684
 685 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
 686 {
 687         struct obd_device       *obddev = class_exp2obd(exp);
 688         struct lmv_obd          *lmv = &obddev->u.lmv;
 689         struct getinfo_fid2path *gf;
 690         struct lmv_tgt_desc     *tgt;
 691         struct getinfo_fid2path *remote_gf = NULL;
 692         int                     remote_gf_size = 0;
 693         int                     rc;
 694
 695         gf = (struct getinfo_fid2path *)karg;
 696         tgt = lmv_find_target(lmv, &gf->gf_fid);
 697         if (IS_ERR(tgt))
 698                 RETURN(PTR_ERR(tgt));
 699
 700 repeat_fid2path:
 701         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 702         if (rc != 0 && rc != -EREMOTE)
 703                 GOTO(out_fid2path, rc);
 704
 705         /* If remote_gf != NULL, it means just building the
 706          * path on the remote MDT, copy this path segement to gf */
 707         if (remote_gf != NULL) {
 708                 struct getinfo_fid2path *ori_gf;
 709                 char *ptr;
 710
 711                 ori_gf = (struct getinfo_fid2path *)karg;
 712                 if (strlen(ori_gf->gf_path) +
 713                     strlen(gf->gf_path) > ori_gf->gf_pathlen)
 714                         GOTO(out_fid2path, rc = -EOVERFLOW);
 715
 716                 ptr = ori_gf->gf_path;
 717
 718                 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
 719                         strlen(ori_gf->gf_path));
 720
 721                 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
 722                 ptr += strlen(gf->gf_path);
 723                 *ptr = '/';
 724         }
 725
 726         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
 727                tgt->ltd_exp->exp_obd->obd_name,
 728                gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 729                gf->gf_linkno);
 730
 731         if (rc == 0)
 732                 GOTO(out_fid2path, rc);
 733
 734         /* sigh, has to go to another MDT to do path building further */
 735         if (remote_gf == NULL) {
 736                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 737                 OBD_ALLOC(remote_gf, remote_gf_size);
 738                 if (remote_gf == NULL)
 739                         GOTO(out_fid2path, rc = -ENOMEM);
 740                 remote_gf->gf_pathlen = PATH_MAX;
 741         }
 742
 743         if (!fid_is_sane(&gf->gf_fid)) {
 744                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 745                        tgt->ltd_exp->exp_obd->obd_name,
 746                        PFID(&gf->gf_fid), -EINVAL);
 747                 GOTO(out_fid2path, rc = -EINVAL);
 748         }
 749
 750         tgt = lmv_find_target(lmv, &gf->gf_fid);
 751         if (IS_ERR(tgt))
 752                 GOTO(out_fid2path, rc = -EINVAL);
 753
 754         remote_gf->gf_fid = gf->gf_fid;
 755         remote_gf->gf_recno = -1;
 756         remote_gf->gf_linkno = -1;
 757         memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
 758         gf = remote_gf;
 759         goto repeat_fid2path;
 760
 761 out_fid2path:
 762         if (remote_gf != NULL)
 763                 OBD_FREE(remote_gf, remote_gf_size);
 764         RETURN(rc);
 765 }
 766
 767 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 768                              const struct hsm_user_request *hur,
 769                              const struct lmv_tgt_desc *tgt_mds)
 770 {
 771         __u32                    i;
 772         int                      nr = 0;
 773         struct lmv_tgt_desc     *curr_tgt;
 774
 775         /* count how many requests must be sent to the given target */
 776         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 777                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
 778                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 779                         nr++;
 780         }
 781         return nr;
 782 }
 783
 784 static void lmv_hsm_req_build(struct lmv_obd *lmv,
 785                               struct hsm_user_request *hur_in,
 786                               const struct lmv_tgt_desc *tgt_mds,
 787                               struct hsm_user_request *hur_out)
 788 {
 789         __u32                    i, nr_out;
 790         struct lmv_tgt_desc     *curr_tgt;
 791
 792         /* build the hsm_user_request for the given target */
 793         hur_out->hur_request = hur_in->hur_request;
 794         nr_out = 0;
 795         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 796                 curr_tgt = lmv_find_target(lmv,
 797                                            &hur_in->hur_user_item[i].hui_fid);
 798                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 799                         hur_out->hur_user_item[nr_out] =
 800                                                 hur_in->hur_user_item[i];
 801                         nr_out++;
 802                 }
 803         }
 804         hur_out->hur_request.hr_itemcount = nr_out;
 805         memcpy(hur_data(hur_out), hur_data(hur_in),
 806                hur_in->hur_request.hr_data_len);
 807 }
 808
 809 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
 810                                  struct lustre_kernelcomm *lk, void *uarg)
 811 {
 812         __u32                    i;
 813         int                      rc;
 814         struct kkuc_ct_data     *kcd = NULL;
 815         ENTRY;
 816
 817         /* unregister request (call from llapi_hsm_copytool_fini) */
 818         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 819                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 820
 821                 if (tgt == NULL || tgt->ltd_exp == NULL)
 822                         continue;
 823                 /* best effort: try to clean as much as possible
 824                  * (continue on error) */
 825                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 826         }
 827
 828         /* Whatever the result, remove copytool from kuc groups.
 829          * Unreached coordinators will get EPIPE on next requests
 830          * and will unregister automatically.
 831          */
 832         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
 833         if (kcd != NULL)
 834                 OBD_FREE_PTR(kcd);
 835
 836         RETURN(rc);
 837 }
 838
 839 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
 840                                struct lustre_kernelcomm *lk, void *uarg)
 841 {
 842         struct file             *filp;
 843         __u32                    i, j;
 844         int                      err, rc;
 845         bool                     any_set = false;
 846         struct kkuc_ct_data     *kcd;
 847         ENTRY;
 848
 849         /* All or nothing: try to register to all MDS.
 850          * In case of failure, unregister from previous MDS,
 851          * except if it because of inactive target. */
 852         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 853                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 854
 855                 if (tgt == NULL || tgt->ltd_exp == NULL)
 856                         continue;
 857                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 858                 if (err) {
 859                         if (tgt->ltd_active) {
 860                                 /* permanent error */
 861                                 CERROR("%s: iocontrol MDC %s on MDT"
 862                                        " idx %d cmd %x: err = %d\n",
 863                                        class_exp2obd(lmv->exp)->obd_name,
 864                                        tgt->ltd_uuid.uuid, i, cmd, err);
 865                                 rc = err;
 866                                 lk->lk_flags |= LK_FLG_STOP;
 867                                 /* unregister from previous MDS */
 868                                 for (j = 0; j < i; j++) {
 869                                         tgt = lmv->tgts[j];
 870                                         if (tgt == NULL || tgt->ltd_exp == NULL)
 871                                                 continue;
 872                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
 873                                                       lk, uarg);
 874                                 }
 875                                 RETURN(rc);
 876                         }
 877                         /* else: transient error.
 878                          * kuc will register to the missing MDT
 879                          * when it is back */
 880                 } else {
 881                         any_set = true;
 882                 }
 883         }
 884
 885         if (!any_set)
 886                 /* no registration done: return error */
 887                 RETURN(-ENOTCONN);
 888
 889         /* at least one registration done, with no failure */
 890         filp = fget(lk->lk_wfd);
 891         if (filp == NULL)
 892                 RETURN(-EBADF);
 893
 894         OBD_ALLOC_PTR(kcd);
 895         if (kcd == NULL) {
 896                 fput(filp);
 897                 RETURN(-ENOMEM);
 898         }
 899         kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
 900         kcd->kcd_uuid = lmv->cluuid;
 901         kcd->kcd_archive = lk->lk_data;
 902
 903         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
 904         if (rc != 0) {
 905                 if (filp != NULL)
 906                         fput(filp);
 907                 OBD_FREE_PTR(kcd);
 908         }
 909
 910         RETURN(rc);
 911 }
 912
 913
 914
 915
 916 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 917                          int len, void *karg, void *uarg)
 918 {
 919         struct obd_device       *obddev = class_exp2obd(exp);
 920         struct lmv_obd          *lmv = &obddev->u.lmv;
 921         struct lmv_tgt_desc     *tgt = NULL;
 922         __u32                    i = 0;
 923         int                      rc = 0;
 924         int                      set = 0;
 925         __u32                    count = lmv->desc.ld_tgt_count;
 926         ENTRY;
 927
 928         if (count == 0)
 929                 RETURN(-ENOTTY);
 930
 931         switch (cmd) {
 932         case IOC_OBD_STATFS: {
 933                 struct obd_ioctl_data *data = karg;
 934                 struct obd_device *mdc_obd;
 935                 struct obd_statfs stat_buf = {0};
 936                 __u32 index;
 937
 938                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 939                 if ((index >= count))
 940                         RETURN(-ENODEV);
 941
 942                 tgt = lmv->tgts[index];
 943                 if (tgt == NULL || !tgt->ltd_active)
 944                         RETURN(-ENODATA);
 945
 946                 mdc_obd = class_exp2obd(tgt->ltd_exp);
 947                 if (!mdc_obd)
 948                         RETURN(-EINVAL);
 949
 950                 /* copy UUID */
 951                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 952                                  min((int) data->ioc_plen2,
 953                                      (int) sizeof(struct obd_uuid))))
 954                         RETURN(-EFAULT);
 955
 956                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
 957                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 958                                 0);
 959                 if (rc)
 960                         RETURN(rc);
 961                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
 962                                  min((int) data->ioc_plen1,
 963                                      (int) sizeof(stat_buf))))
 964                         RETURN(-EFAULT);
 965                 break;
 966         }
 967         case OBD_IOC_QUOTACTL: {
 968                 struct if_quotactl *qctl = karg;
 969                 struct obd_quotactl *oqctl;
 970
 971                 if (qctl->qc_valid == QC_MDTIDX) {
 972                         if (count <= qctl->qc_idx)
 973                                 RETURN(-EINVAL);
 974
 975                         tgt = lmv->tgts[qctl->qc_idx];
 976                         if (tgt == NULL || tgt->ltd_exp == NULL)
 977                                 RETURN(-EINVAL);
 978                 } else if (qctl->qc_valid == QC_UUID) {
 979                         for (i = 0; i < count; i++) {
 980                                 tgt = lmv->tgts[i];
 981                                 if (tgt == NULL)
 982                                         continue;
 983                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 984                                                      &qctl->obd_uuid))
 985                                         continue;
 986
 987                                 if (tgt->ltd_exp == NULL)
 988                                         RETURN(-EINVAL);
 989
 990                                 break;
 991                         }
 992                 } else {
 993                         RETURN(-EINVAL);
 994                 }
 995
 996                 if (i >= count)
 997                         RETURN(-EAGAIN);
 998
 999                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1000                 OBD_ALLOC_PTR(oqctl);
1001                 if (!oqctl)
1002                         RETURN(-ENOMEM);
1003
1004                 QCTL_COPY(oqctl, qctl);
1005                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1006                 if (rc == 0) {
1007                         QCTL_COPY(qctl, oqctl);
1008                         qctl->qc_valid = QC_MDTIDX;
1009                         qctl->obd_uuid = tgt->ltd_uuid;
1010                 }
1011                 OBD_FREE_PTR(oqctl);
1012                 break;
1013         }
1014         case OBD_IOC_CHANGELOG_SEND:
1015         case OBD_IOC_CHANGELOG_CLEAR: {
1016                 struct ioc_changelog *icc = karg;
1017
1018                 if (icc->icc_mdtindex >= count)
1019                         RETURN(-ENODEV);
1020
1021                 tgt = lmv->tgts[icc->icc_mdtindex];
1022                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1023                         RETURN(-ENODEV);
1024                 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1025                 break;
1026         }
1027         case LL_IOC_GET_CONNECT_FLAGS: {
1028                 tgt = lmv->tgts[0];
1029                 if (tgt == NULL || tgt->ltd_exp == NULL)
1030                         RETURN(-ENODATA);
1031                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1032                 break;
1033         }
1034         case LL_IOC_FID2MDTIDX: {
1035                 struct lu_fid *fid = karg;
1036                 int             mdt_index;
1037
1038                 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1039                 if (rc != 0)
1040                         RETURN(rc);
1041
1042                 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1043                  * point to user space memory for FID2MDTIDX. */
1044                 *(__u32 *)uarg = mdt_index;
1045                 break;
1046         }
1047         case OBD_IOC_FID2PATH: {
1048                 rc = lmv_fid2path(exp, len, karg, uarg);
1049                 break;
1050         }
1051         case LL_IOC_HSM_STATE_GET:
1052         case LL_IOC_HSM_STATE_SET:
1053         case LL_IOC_HSM_ACTION: {
1054                 struct md_op_data       *op_data = karg;
1055
1056                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1057                 if (IS_ERR(tgt))
1058                         RETURN(PTR_ERR(tgt));
1059
1060                 if (tgt->ltd_exp == NULL)
1061                         RETURN(-EINVAL);
1062
1063                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1064                 break;
1065         }
1066         case LL_IOC_HSM_PROGRESS: {
1067                 const struct hsm_progress_kernel *hpk = karg;
1068
1069                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1070                 if (IS_ERR(tgt))
1071                         RETURN(PTR_ERR(tgt));
1072                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1073                 break;
1074         }
1075         case LL_IOC_HSM_REQUEST: {
1076                 struct hsm_user_request *hur = karg;
1077                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1078
1079                 if (reqcount == 0)
1080                         RETURN(0);
1081
1082                 /* if the request is about a single fid
1083                  * or if there is a single MDS, no need to split
1084                  * the request. */
1085                 if (reqcount == 1 || count == 1) {
1086                         tgt = lmv_find_target(lmv,
1087                                               &hur->hur_user_item[0].hui_fid);
1088                         if (IS_ERR(tgt))
1089                                 RETURN(PTR_ERR(tgt));
1090                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1091                 } else {
1092                         /* split fid list to their respective MDS */
1093                         for (i = 0; i < count; i++) {
1094                                 unsigned int            nr, reqlen;
1095                                 int                     rc1;
1096                                 struct hsm_user_request *req;
1097
1098                                 tgt = lmv->tgts[i];
1099                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1100                                         continue;
1101
1102                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1103                                 if (nr == 0) /* nothing for this MDS */
1104                                         continue;
1105
1106                                 /* build a request with fids for this MDS */
1107                                 reqlen = offsetof(typeof(*hur),
1108                                                   hur_user_item[nr])
1109                                                 + hur->hur_request.hr_data_len;
1110                                 OBD_ALLOC_LARGE(req, reqlen);
1111                                 if (req == NULL)
1112                                         RETURN(-ENOMEM);
1113
1114                                 lmv_hsm_req_build(lmv, hur, tgt, req);
1115
1116                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1117                                                     req, uarg);
1118                                 if (rc1 != 0 && rc == 0)
1119                                         rc = rc1;
1120                                 OBD_FREE_LARGE(req, reqlen);
1121                         }
1122                 }
1123                 break;
1124         }
1125         case LL_IOC_LOV_SWAP_LAYOUTS: {
1126                 struct md_op_data       *op_data = karg;
1127                 struct lmv_tgt_desc     *tgt1, *tgt2;
1128
1129                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1130                 if (IS_ERR(tgt1))
1131                         RETURN(PTR_ERR(tgt1));
1132
1133                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1134                 if (IS_ERR(tgt2))
1135                         RETURN(PTR_ERR(tgt2));
1136
1137                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1138                         RETURN(-EINVAL);
1139
1140                 /* only files on same MDT can have their layouts swapped */
1141                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1142                         RETURN(-EPERM);
1143
1144                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1145                 break;
1146         }
1147         case LL_IOC_HSM_CT_START: {
1148                 struct lustre_kernelcomm *lk = karg;
1149                 if (lk->lk_flags & LK_FLG_STOP)
1150                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1151                 else
1152                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1153                 break;
1154         }
1155         default:
1156                 for (i = 0; i < count; i++) {
1157                         struct obd_device *mdc_obd;
1158                         int err;
1159
1160                         tgt = lmv->tgts[i];
1161                         if (tgt == NULL || tgt->ltd_exp == NULL)
1162                                 continue;
1163                         /* ll_umount_begin() sets force flag but for lmv, not
1164                          * mdc. Let's pass it through */
1165                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1166                         mdc_obd->obd_force = obddev->obd_force;
1167                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1168                         if (err) {
1169                                 if (tgt->ltd_active) {
1170                                         CERROR("error: iocontrol MDC %s on MDT"
1171                                                " idx %d cmd %x: err = %d\n",
1172                                                tgt->ltd_uuid.uuid, i, cmd, err);
1173                                         if (!rc)
1174                                                 rc = err;
1175                                 }
1176                         } else
1177                                 set = 1;
1178                 }
1179                 if (!set && !rc)
1180                         rc = -EIO;
1181         }
1182         RETURN(rc);
1183 }
1184
1185 #if 0
1186 static int lmv_all_chars_policy(int count, const char *name,
1187                                 int len)
1188 {
1189         unsigned int c = 0;
1190
1191         while (len > 0)
1192                 c += name[--len];
1193         c = c % count;
1194         return c;
1195 }
1196
1197 static int lmv_nid_policy(struct lmv_obd *lmv)
1198 {
1199         struct obd_import *imp;
1200         __u32              id;
1201
1202         /*
1203          * XXX: To get nid we assume that underlying obd device is mdc.
1204          */
1205         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1206         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1207         return id % lmv->desc.ld_tgt_count;
1208 }
1209
1210 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1211                           placement_policy_t placement)
1212 {
1213         switch (placement) {
1214         case PLACEMENT_CHAR_POLICY:
1215                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1216                                             op_data->op_name,
1217                                             op_data->op_namelen);
1218         case PLACEMENT_NID_POLICY:
1219                 return lmv_nid_policy(lmv);
1220
1221         default:
1222                 break;
1223         }
1224
1225         CERROR("Unsupported placement policy %x\n", placement);
1226         return -EINVAL;
1227 }
1228 #endif
1229
1230 /**
1231  * This is _inode_ placement policy function (not name).
1232  */
1233 static int lmv_placement_policy(struct obd_device *obd,
1234                                 struct md_op_data *op_data, u32 *mds)
1235 {
1236         struct lmv_obd          *lmv = &obd->u.lmv;
1237         ENTRY;
1238
1239         LASSERT(mds != NULL);
1240
1241         if (lmv->desc.ld_tgt_count == 1) {
1242                 *mds = 0;
1243                 RETURN(0);
1244         }
1245
1246         if (op_data->op_default_stripe_offset != -1) {
1247                 *mds = op_data->op_default_stripe_offset;
1248                 RETURN(0);
1249         }
1250
1251         /**
1252          * If stripe_offset is provided during setdirstripe
1253          * (setdirstripe -i xx), xx MDS will be choosen.
1254          */
1255         if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1256                 struct lmv_user_md *lum;
1257
1258                 lum = op_data->op_data;
1259
1260                 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1261                         *mds = le32_to_cpu(lum->lum_stripe_offset);
1262                 } else {
1263                         /* -1 means default, which will be in the same MDT with
1264                          * the stripe */
1265                         *mds = op_data->op_mds;
1266                         lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1267                 }
1268         } else {
1269                 /* Allocate new fid on target according to operation type and
1270                  * parent home mds. */
1271                 *mds = op_data->op_mds;
1272         }
1273
1274         RETURN(0);
1275 }
1276
1277 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1278 {
1279         struct lmv_tgt_desc     *tgt;
1280         int                      rc;
1281         ENTRY;
1282
1283         tgt = lmv_get_target(lmv, mds, NULL);
1284         if (IS_ERR(tgt))
1285                 RETURN(PTR_ERR(tgt));
1286
1287         /*
1288          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1289          * on server that seq in new allocated fid is not yet known.
1290          */
1291         mutex_lock(&tgt->ltd_fid_mutex);
1292
1293         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1294                 GOTO(out, rc = -ENODEV);
1295
1296         /*
1297          * Asking underlying tgt layer to allocate new fid.
1298          */
1299         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1300         if (rc > 0) {
1301                 LASSERT(fid_is_sane(fid));
1302                 rc = 0;
1303         }
1304
1305         EXIT;
1306 out:
1307         mutex_unlock(&tgt->ltd_fid_mutex);
1308         return rc;
1309 }
1310
1311 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1312                   struct lu_fid *fid, struct md_op_data *op_data)
1313 {
1314         struct obd_device     *obd = class_exp2obd(exp);
1315         struct lmv_obd        *lmv = &obd->u.lmv;
1316         u32                    mds = 0;
1317         int                    rc;
1318         ENTRY;
1319
1320         LASSERT(op_data != NULL);
1321         LASSERT(fid != NULL);
1322
1323         rc = lmv_placement_policy(obd, op_data, &mds);
1324         if (rc) {
1325                 CERROR("Can't get target for allocating fid, "
1326                        "rc %d\n", rc);
1327                 RETURN(rc);
1328         }
1329
1330         rc = __lmv_fid_alloc(lmv, fid, mds);
1331         if (rc) {
1332                 CERROR("Can't alloc new fid, rc %d\n", rc);
1333                 RETURN(rc);
1334         }
1335
1336         RETURN(rc);
1337 }
1338
1339 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1340 {
1341         struct lmv_obd  *lmv = &obd->u.lmv;
1342         struct lmv_desc *desc;
1343         int             rc;
1344         ENTRY;
1345
1346         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1347                 CERROR("LMV setup requires a descriptor\n");
1348                 RETURN(-EINVAL);
1349         }
1350
1351         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1352         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1353                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1354                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1355                 RETURN(-EINVAL);
1356         }
1357
1358         lmv->tgts_size = 32U;
1359         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1360         if (lmv->tgts == NULL)
1361                 RETURN(-ENOMEM);
1362
1363         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1364         lmv->desc.ld_tgt_count = 0;
1365         lmv->desc.ld_active_tgt_count = 0;
1366         lmv->max_def_easize = 0;
1367         lmv->max_easize = 0;
1368         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1369
1370         spin_lock_init(&lmv->lmv_lock);
1371         mutex_init(&lmv->lmv_init_mutex);
1372
1373 #ifdef CONFIG_PROC_FS
1374         obd->obd_vars = lprocfs_lmv_obd_vars;
1375         lprocfs_obd_setup(obd);
1376         lprocfs_alloc_md_stats(obd, 0);
1377         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1378                                 0444, &lmv_proc_target_fops, obd);
1379         if (rc)
1380                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1381                       obd->obd_name, rc);
1382 #endif
1383         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1384                              LUSTRE_CLI_FLD_HASH_DHT);
1385         if (rc) {
1386                 CERROR("Can't init FLD, err %d\n", rc);
1387                 GOTO(out, rc);
1388         }
1389
1390         RETURN(0);
1391
1392 out:
1393         return rc;
1394 }
1395
1396 static int lmv_cleanup(struct obd_device *obd)
1397 {
1398         struct lmv_obd   *lmv = &obd->u.lmv;
1399         ENTRY;
1400
1401         fld_client_fini(&lmv->lmv_fld);
1402         if (lmv->tgts != NULL) {
1403                 int i;
1404                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1405                         if (lmv->tgts[i] == NULL)
1406                                 continue;
1407                         lmv_del_target(lmv, i);
1408                 }
1409                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1410                 lmv->tgts_size = 0;
1411         }
1412         RETURN(0);
1413 }
1414
1415 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1416 {
1417         struct lustre_cfg       *lcfg = buf;
1418         struct obd_uuid         obd_uuid;
1419         int                     gen;
1420         __u32                   index;
1421         int                     rc;
1422         ENTRY;
1423
1424         switch (lcfg->lcfg_command) {
1425         case LCFG_ADD_MDC:
1426                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1427                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1428                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1429                         GOTO(out, rc = -EINVAL);
1430
1431                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1432
1433                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1434                         GOTO(out, rc = -EINVAL);
1435                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1436                         GOTO(out, rc = -EINVAL);
1437                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1438                 GOTO(out, rc);
1439         default:
1440                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1441                 GOTO(out, rc = -EINVAL);
1442         }
1443 out:
1444         RETURN(rc);
1445 }
1446
1447 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1448                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1449 {
1450         struct obd_device       *obd = class_exp2obd(exp);
1451         struct lmv_obd          *lmv = &obd->u.lmv;
1452         struct obd_statfs       *temp;
1453         int                      rc = 0;
1454         __u32                    i;
1455         ENTRY;
1456
1457         rc = lmv_check_connect(obd);
1458         if (rc)
1459                 RETURN(rc);
1460
1461         OBD_ALLOC(temp, sizeof(*temp));
1462         if (temp == NULL)
1463                 RETURN(-ENOMEM);
1464
1465         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1466                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1467                         continue;
1468
1469                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1470                                 max_age, flags);
1471                 if (rc) {
1472                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1473                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1474                                rc);
1475                         GOTO(out_free_temp, rc);
1476                 }
1477
1478                 if (i == 0) {
1479                         *osfs = *temp;
1480                         /* If the statfs is from mount, it will needs
1481                          * retrieve necessary information from MDT0.
1482                          * i.e. mount does not need the merged osfs
1483                          * from all of MDT.
1484                          * And also clients can be mounted as long as
1485                          * MDT0 is in service*/
1486                         if (flags & OBD_STATFS_FOR_MDT0)
1487                                 GOTO(out_free_temp, rc);
1488                 } else {
1489                         osfs->os_bavail += temp->os_bavail;
1490                         osfs->os_blocks += temp->os_blocks;
1491                         osfs->os_ffree += temp->os_ffree;
1492                         osfs->os_files += temp->os_files;
1493                 }
1494         }
1495
1496         EXIT;
1497 out_free_temp:
1498         OBD_FREE(temp, sizeof(*temp));
1499         return rc;
1500 }
1501
1502 static int lmv_getstatus(struct obd_export *exp,
1503                          struct lu_fid *fid,
1504                          struct obd_capa **pc)
1505 {
1506         struct obd_device    *obd = exp->exp_obd;
1507         struct lmv_obd       *lmv = &obd->u.lmv;
1508         int                   rc;
1509         ENTRY;
1510
1511         rc = lmv_check_connect(obd);
1512         if (rc)
1513                 RETURN(rc);
1514
1515         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1516         RETURN(rc);
1517 }
1518
1519 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1520                         struct obd_capa *oc, u64 valid, const char *name,
1521                         const char *input, int input_size, int output_size,
1522                         int flags, struct ptlrpc_request **request)
1523 {
1524         struct obd_device      *obd = exp->exp_obd;
1525         struct lmv_obd         *lmv = &obd->u.lmv;
1526         struct lmv_tgt_desc    *tgt;
1527         int                     rc;
1528         ENTRY;
1529
1530         rc = lmv_check_connect(obd);
1531         if (rc)
1532                 RETURN(rc);
1533
1534         tgt = lmv_find_target(lmv, fid);
1535         if (IS_ERR(tgt))
1536                 RETURN(PTR_ERR(tgt));
1537
1538         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1539                          input_size, output_size, flags, request);
1540
1541         RETURN(rc);
1542 }
1543
1544 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1545                         struct obd_capa *oc, u64 valid, const char *name,
1546                         const char *input, int input_size, int output_size,
1547                         int flags, __u32 suppgid,
1548                         struct ptlrpc_request **request)
1549 {
1550         struct obd_device      *obd = exp->exp_obd;
1551         struct lmv_obd         *lmv = &obd->u.lmv;
1552         struct lmv_tgt_desc    *tgt;
1553         int                     rc;
1554         ENTRY;
1555
1556         rc = lmv_check_connect(obd);
1557         if (rc)
1558                 RETURN(rc);
1559
1560         tgt = lmv_find_target(lmv, fid);
1561         if (IS_ERR(tgt))
1562                 RETURN(PTR_ERR(tgt));
1563
1564         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1565                          input_size, output_size, flags, suppgid,
1566                          request);
1567
1568         RETURN(rc);
1569 }
1570
1571 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1572                        struct ptlrpc_request **request)
1573 {
1574         struct obd_device       *obd = exp->exp_obd;
1575         struct lmv_obd          *lmv = &obd->u.lmv;
1576         struct lmv_tgt_desc     *tgt;
1577         int                      rc;
1578         ENTRY;
1579
1580         rc = lmv_check_connect(obd);
1581         if (rc)
1582                 RETURN(rc);
1583
1584         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1585         if (IS_ERR(tgt))
1586                 RETURN(PTR_ERR(tgt));
1587
1588         if (op_data->op_flags & MF_GET_MDT_IDX) {
1589                 op_data->op_mds = tgt->ltd_idx;
1590                 RETURN(0);
1591         }
1592
1593         rc = md_getattr(tgt->ltd_exp, op_data, request);
1594
1595         RETURN(rc);
1596 }
1597
1598 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1599 {
1600         struct obd_device   *obd = exp->exp_obd;
1601         struct lmv_obd      *lmv = &obd->u.lmv;
1602         __u32                i;
1603         int                  rc;
1604         ENTRY;
1605
1606         rc = lmv_check_connect(obd);
1607         if (rc)
1608                 RETURN(rc);
1609
1610         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1611
1612         /*
1613          * With DNE every object can have two locks in different namespaces:
1614          * lookup lock in space of MDT storing direntry and update/open lock in
1615          * space of MDT storing inode.
1616          */
1617         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1618                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1619                         continue;
1620                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1621         }
1622
1623         RETURN(0);
1624 }
1625
1626 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1627                            ldlm_iterator_t it, void *data)
1628 {
1629         struct obd_device       *obd = exp->exp_obd;
1630         struct lmv_obd          *lmv = &obd->u.lmv;
1631         int                     i;
1632         int                     tgt;
1633         int                     rc;
1634         ENTRY;
1635
1636         rc = lmv_check_connect(obd);
1637         if (rc)
1638                 RETURN(rc);
1639
1640         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1641
1642         /*
1643          * With DNE every object can have two locks in different namespaces:
1644          * lookup lock in space of MDT storing direntry and update/open lock in
1645          * space of MDT storing inode.  Try the MDT that the FID maps to first,
1646          * since this can be easily found, and only try others if that fails.
1647          */
1648         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1649              i < lmv->desc.ld_tgt_count;
1650              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1651                 if (tgt < 0) {
1652                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1653                                obd->obd_name, PFID(fid), tgt);
1654                         tgt = 0;
1655                 }
1656
1657                 if (lmv->tgts[tgt] == NULL ||
1658                     lmv->tgts[tgt]->ltd_exp == NULL)
1659                         continue;
1660
1661                 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1662                 if (rc)
1663                         RETURN(rc);
1664         }
1665
1666         RETURN(rc);
1667 }
1668
1669
1670 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1671                      struct md_open_data *mod, struct ptlrpc_request **request)
1672 {
1673         struct obd_device     *obd = exp->exp_obd;
1674         struct lmv_obd        *lmv = &obd->u.lmv;
1675         struct lmv_tgt_desc   *tgt;
1676         int                    rc;
1677         ENTRY;
1678
1679         rc = lmv_check_connect(obd);
1680         if (rc)
1681                 RETURN(rc);
1682
1683         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1684         if (IS_ERR(tgt))
1685                 RETURN(PTR_ERR(tgt));
1686
1687         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1688         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1689         RETURN(rc);
1690 }
1691
1692 /**
1693  * Choosing the MDT by name or FID in @op_data.
1694  * For non-striped directory, it will locate MDT by fid.
1695  * For striped-directory, it will locate MDT by name. And also
1696  * it will reset op_fid1 with the FID of the choosen stripe.
1697  **/
1698 struct lmv_tgt_desc *
1699 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1700                            const char *name, int namelen, struct lu_fid *fid,
1701                            u32 *mds)
1702 {
1703         struct lmv_tgt_desc     *tgt;
1704         const struct lmv_oinfo  *oinfo;
1705
1706         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1707                 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1708                         RETURN(ERR_PTR(-EBADF));
1709                 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1710         } else {
1711                 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1712                 if (IS_ERR(oinfo))
1713                         RETURN(ERR_CAST(oinfo));
1714         }
1715
1716         if (fid != NULL)
1717                 *fid = oinfo->lmo_fid;
1718         if (mds != NULL)
1719                 *mds = oinfo->lmo_mds;
1720
1721         tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1722
1723         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", oinfo->lmo_mds,
1724                PFID(&oinfo->lmo_fid));
1725         return tgt;
1726 }
1727
1728 /**
1729  * Locate mds by fid or name
1730  *
1731  * For striped directory (lsm != NULL), it will locate the stripe
1732  * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1733  * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1734  * walk through all of stripes to locate the entry.
1735  *
1736  * For normal direcotry, it will locate MDS by FID directly.
1737  * \param[in] lmv       LMV device
1738  * \param[in] op_data   client MD stack parameters, name, namelen
1739  *                      mds_num etc.
1740  * \param[in] fid       object FID used to locate MDS.
1741  *
1742  * retval               pointer to the lmv_tgt_desc if succeed.
1743  *                      ERR_PTR(errno) if failed.
1744  */
1745 struct lmv_tgt_desc*
1746 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1747                struct lu_fid *fid)
1748 {
1749         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1750         struct lmv_tgt_desc     *tgt;
1751
1752         /* During creating VOLATILE file, it should honor the mdt
1753          * index if the file under striped dir is being restored, see
1754          * ct_restore(). */
1755         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1756             (int)op_data->op_mds != -1 && lsm != NULL) {
1757                 int i;
1758                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1759                 if (IS_ERR(tgt))
1760                         return tgt;
1761
1762                 /* refill the right parent fid */
1763                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1764                         struct lmv_oinfo *oinfo;
1765
1766                         oinfo = &lsm->lsm_md_oinfo[i];
1767                         if (oinfo->lmo_mds == op_data->op_mds) {
1768                                 *fid = oinfo->lmo_fid;
1769                                 break;
1770                         }
1771                 }
1772
1773                 /* Hmm, can not find the stripe by mdt_index(op_mds) */
1774                 if (i == lsm->lsm_md_stripe_count)
1775                         tgt = ERR_PTR(-EINVAL);
1776
1777                 return tgt;
1778         }
1779
1780         if (lsm == NULL || op_data->op_namelen == 0) {
1781                 tgt = lmv_find_target(lmv, fid);
1782                 if (IS_ERR(tgt))
1783                         return tgt;
1784
1785                 op_data->op_mds = tgt->ltd_idx;
1786                 return tgt;
1787         }
1788
1789         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1790                                           op_data->op_namelen, fid,
1791                                           &op_data->op_mds);
1792 }
1793
1794 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1795                 const void *data, size_t datalen, umode_t mode, uid_t uid,
1796                 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1797                 struct ptlrpc_request **request)
1798 {
1799         struct obd_device       *obd = exp->exp_obd;
1800         struct lmv_obd          *lmv = &obd->u.lmv;
1801         struct lmv_tgt_desc     *tgt;
1802         int                      rc;
1803         ENTRY;
1804
1805         rc = lmv_check_connect(obd);
1806         if (rc)
1807                 RETURN(rc);
1808
1809         if (!lmv->desc.ld_active_tgt_count)
1810                 RETURN(-EIO);
1811
1812         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1813         if (IS_ERR(tgt))
1814                 RETURN(PTR_ERR(tgt));
1815
1816         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1817                 (int)op_data->op_namelen, op_data->op_name,
1818                 PFID(&op_data->op_fid1), op_data->op_mds);
1819
1820         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1821         if (rc)
1822                 RETURN(rc);
1823         if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1824                 /* Send the create request to the MDT where the object
1825                  * will be located */
1826                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1827                 if (IS_ERR(tgt))
1828                         RETURN(PTR_ERR(tgt));
1829
1830                 op_data->op_mds = tgt->ltd_idx;
1831         } else {
1832                 CDEBUG(D_CONFIG, "Server doesn't support striped dirs\n");
1833         }
1834
1835         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1836                PFID(&op_data->op_fid2), op_data->op_mds);
1837
1838         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1839         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1840                        cap_effective, rdev, request);
1841         if (rc == 0) {
1842                 if (*request == NULL)
1843                         RETURN(rc);
1844                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1845         }
1846         RETURN(rc);
1847 }
1848
1849 static int
1850 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1851             const union ldlm_policy_data *policy,
1852             struct lookup_intent *it, struct md_op_data *op_data,
1853             struct lustre_handle *lockh, __u64 extra_lock_flags)
1854 {
1855         struct obd_device        *obd = exp->exp_obd;
1856         struct lmv_obd           *lmv = &obd->u.lmv;
1857         struct lmv_tgt_desc      *tgt;
1858         int                       rc;
1859         ENTRY;
1860
1861         rc = lmv_check_connect(obd);
1862         if (rc)
1863                 RETURN(rc);
1864
1865         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1866                LL_IT2STR(it), PFID(&op_data->op_fid1));
1867
1868         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1869         if (IS_ERR(tgt))
1870                 RETURN(PTR_ERR(tgt));
1871
1872         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1873                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1874
1875         rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1876                         extra_lock_flags);
1877
1878         RETURN(rc);
1879 }
1880
1881 static int
1882 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1883                  struct ptlrpc_request **preq)
1884 {
1885         struct ptlrpc_request   *req = NULL;
1886         struct obd_device       *obd = exp->exp_obd;
1887         struct lmv_obd          *lmv = &obd->u.lmv;
1888         struct lmv_tgt_desc     *tgt;
1889         struct mdt_body         *body;
1890         int                      rc;
1891         ENTRY;
1892
1893         rc = lmv_check_connect(obd);
1894         if (rc)
1895                 RETURN(rc);
1896
1897         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1898         if (IS_ERR(tgt))
1899                 RETURN(PTR_ERR(tgt));
1900
1901         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1902                 (int)op_data->op_namelen, op_data->op_name,
1903                 PFID(&op_data->op_fid1), tgt->ltd_idx);
1904
1905         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1906         if (rc != 0)
1907                 RETURN(rc);
1908
1909         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1910         LASSERT(body != NULL);
1911
1912         if (body->mbo_valid & OBD_MD_MDS) {
1913                 struct lu_fid rid = body->mbo_fid1;
1914                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1915                        PFID(&rid));
1916
1917                 tgt = lmv_find_target(lmv, &rid);
1918                 if (IS_ERR(tgt)) {
1919                         ptlrpc_req_finished(*preq);
1920                         preq = NULL;
1921                         RETURN(PTR_ERR(tgt));
1922                 }
1923
1924                 op_data->op_fid1 = rid;
1925                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1926                 op_data->op_namelen = 0;
1927                 op_data->op_name = NULL;
1928                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1929                 ptlrpc_req_finished(*preq);
1930                 *preq = req;
1931         }
1932
1933         RETURN(rc);
1934 }
1935
1936 #define md_op_data_fid(op_data, fl)                     \
1937         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1938          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1939          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1940          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1941          NULL)
1942
1943 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1944                             struct md_op_data *op_data,
1945                             __u32 op_tgt, ldlm_mode_t mode, int bits, int flag)
1946 {
1947         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1948         struct obd_device      *obd = exp->exp_obd;
1949         struct lmv_obd         *lmv = &obd->u.lmv;
1950         ldlm_policy_data_t      policy = {{ 0 }};
1951         int                     rc = 0;
1952         ENTRY;
1953
1954         if (!fid_is_sane(fid))
1955                 RETURN(0);
1956
1957         if (tgt == NULL) {
1958                 tgt = lmv_find_target(lmv, fid);
1959                 if (IS_ERR(tgt))
1960                         RETURN(PTR_ERR(tgt));
1961         }
1962
1963         if (tgt->ltd_idx != op_tgt) {
1964                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1965                 policy.l_inodebits.bits = bits;
1966                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1967                                       mode, LCF_ASYNC, NULL);
1968         } else {
1969                 CDEBUG(D_INODE,
1970                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1971                        op_tgt, PFID(fid));
1972                 op_data->op_flags |= flag;
1973                 rc = 0;
1974         }
1975
1976         RETURN(rc);
1977 }
1978
1979 /*
1980  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1981  * op_data->op_fid2
1982  */
1983 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1984                     struct ptlrpc_request **request)
1985 {
1986         struct obd_device       *obd = exp->exp_obd;
1987         struct lmv_obd          *lmv = &obd->u.lmv;
1988         struct lmv_tgt_desc     *tgt;
1989         int                      rc;
1990         ENTRY;
1991
1992         rc = lmv_check_connect(obd);
1993         if (rc)
1994                 RETURN(rc);
1995
1996         LASSERT(op_data->op_namelen != 0);
1997
1998         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1999                PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2000                op_data->op_name, PFID(&op_data->op_fid1));
2001
2002         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2003         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2004         op_data->op_cap = cfs_curproc_cap_pack();
2005         if (op_data->op_mea2 != NULL) {
2006                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
2007                 const struct lmv_oinfo  *oinfo;
2008
2009                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2010                                                 op_data->op_namelen);
2011                 if (IS_ERR(oinfo))
2012                         RETURN(PTR_ERR(oinfo));
2013
2014                 op_data->op_fid2 = oinfo->lmo_fid;
2015         }
2016
2017         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2018         if (IS_ERR(tgt))
2019                 RETURN(PTR_ERR(tgt));
2020
2021         /*
2022          * Cancel UPDATE lock on child (fid1).
2023          */
2024         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2025         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2026                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2027         if (rc != 0)
2028                 RETURN(rc);
2029
2030         rc = md_link(tgt->ltd_exp, op_data, request);
2031
2032         RETURN(rc);
2033 }
2034
2035 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2036                       const char *old, size_t oldlen,
2037                       const char *new, size_t newlen,
2038                       struct ptlrpc_request **request)
2039 {
2040         struct obd_device       *obd = exp->exp_obd;
2041         struct lmv_obd          *lmv = &obd->u.lmv;
2042         struct lmv_tgt_desc     *src_tgt;
2043         int                     rc;
2044         ENTRY;
2045
2046         LASSERT(oldlen != 0);
2047
2048         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2049                (int)oldlen, old, PFID(&op_data->op_fid1),
2050                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2051                (int)newlen, new, PFID(&op_data->op_fid2),
2052                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2053
2054         rc = lmv_check_connect(obd);
2055         if (rc)
2056                 RETURN(rc);
2057
2058         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2059         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2060         op_data->op_cap = cfs_curproc_cap_pack();
2061         if (op_data->op_cli_flags & CLI_MIGRATE) {
2062                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2063                          PFID(&op_data->op_fid3));
2064
2065                 if (op_data->op_mea1 != NULL) {
2066                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2067                         struct lmv_tgt_desc     *tmp;
2068
2069                         /* Fix the parent fid for striped dir */
2070                         tmp = lmv_locate_target_for_name(lmv, lsm, old,
2071                                                          oldlen,
2072                                                          &op_data->op_fid1,
2073                                                          NULL);
2074                         if (IS_ERR(tmp))
2075                                 RETURN(PTR_ERR(tmp));
2076                 }
2077
2078                 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2079                 if (rc != 0)
2080                         RETURN(rc);
2081
2082                 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2083         } else {
2084                 if (op_data->op_mea1 != NULL) {
2085                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2086
2087                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2088                                                              oldlen,
2089                                                              &op_data->op_fid1,
2090                                                              &op_data->op_mds);
2091                         if (IS_ERR(src_tgt))
2092                                 RETURN(PTR_ERR(src_tgt));
2093                 } else {
2094                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2095                         if (IS_ERR(src_tgt))
2096                                 RETURN(PTR_ERR(src_tgt));
2097
2098                         op_data->op_mds = src_tgt->ltd_idx;
2099                 }
2100
2101                 if (op_data->op_mea2) {
2102                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
2103                         const struct lmv_oinfo  *oinfo;
2104
2105                         oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2106                         if (IS_ERR(oinfo))
2107                                 RETURN(PTR_ERR(oinfo));
2108
2109                         op_data->op_fid2 = oinfo->lmo_fid;
2110                 }
2111         }
2112         if (IS_ERR(src_tgt))
2113                 RETURN(PTR_ERR(src_tgt));
2114
2115         /*
2116          * LOOKUP lock on src child (fid3) should also be cancelled for
2117          * src_tgt in mdc_rename.
2118          */
2119         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2120
2121         /*
2122          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2123          * own target.
2124          */
2125         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2126                               LCK_EX, MDS_INODELOCK_UPDATE,
2127                               MF_MDC_CANCEL_FID2);
2128
2129         if (rc != 0)
2130                 RETURN(rc);
2131         /*
2132          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2133          */
2134         if (fid_is_sane(&op_data->op_fid3)) {
2135                 struct lmv_tgt_desc *tgt;
2136
2137                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2138                 if (IS_ERR(tgt))
2139                         RETURN(PTR_ERR(tgt));
2140
2141                 /* Cancel LOOKUP lock on its parent */
2142                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2143                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2144                                       MF_MDC_CANCEL_FID3);
2145                 if (rc != 0)
2146                         RETURN(rc);
2147
2148                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2149                                       LCK_EX, MDS_INODELOCK_FULL,
2150                                       MF_MDC_CANCEL_FID3);
2151                 if (rc != 0)
2152                         RETURN(rc);
2153         }
2154
2155         /*
2156          * Cancel all the locks on tgt child (fid4).
2157          */
2158         if (fid_is_sane(&op_data->op_fid4))
2159                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2160                                       LCK_EX, MDS_INODELOCK_FULL,
2161                                       MF_MDC_CANCEL_FID4);
2162
2163         CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2164                op_data->op_mds, PFID(&op_data->op_fid2));
2165
2166         rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2167                        request);
2168
2169         RETURN(rc);
2170 }
2171
2172 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2173                        void *ea, size_t ealen, struct ptlrpc_request **request)
2174 {
2175         struct obd_device       *obd = exp->exp_obd;
2176         struct lmv_obd          *lmv = &obd->u.lmv;
2177         struct lmv_tgt_desc     *tgt;
2178         int                      rc = 0;
2179         ENTRY;
2180
2181         rc = lmv_check_connect(obd);
2182         if (rc)
2183                 RETURN(rc);
2184
2185         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2186                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2187
2188         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2189         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2190         if (IS_ERR(tgt))
2191                 RETURN(PTR_ERR(tgt));
2192
2193         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2194
2195         RETURN(rc);
2196 }
2197
2198 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2199                      struct obd_capa *oc, struct ptlrpc_request **request)
2200 {
2201         struct obd_device       *obd = exp->exp_obd;
2202         struct lmv_obd          *lmv = &obd->u.lmv;
2203         struct lmv_tgt_desc     *tgt;
2204         int                      rc;
2205         ENTRY;
2206
2207         rc = lmv_check_connect(obd);
2208         if (rc != 0)
2209                 RETURN(rc);
2210
2211         tgt = lmv_find_target(lmv, fid);
2212         if (IS_ERR(tgt))
2213                 RETURN(PTR_ERR(tgt));
2214
2215         rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2216         RETURN(rc);
2217 }
2218
2219 /**
2220  * Get current minimum entry from striped directory
2221  *
2222  * This function will search the dir entry, whose hash value is the
2223  * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2224  * only being called for striped directory.
2225  *
2226  * \param[in] exp               export of LMV
2227  * \param[in] op_data           parameters transferred beween client MD stack
2228  *                              stripe_information will be included in this
2229  *                              parameter
2230  * \param[in] cb_op             ldlm callback being used in enqueue in
2231  *                              mdc_read_page
2232  * \param[in] hash_offset       the hash value, which is used to locate
2233  *                              minum(closet) dir entry
2234  * \param[in|out] stripe_offset the caller use this to indicate the stripe
2235  *                              index of last entry, so to avoid hash conflict
2236  *                              between stripes. It will also be used to
2237  *                              return the stripe index of current dir entry.
2238  * \param[in|out] entp          the minum entry and it also is being used
2239  *                              to input the last dir entry to resolve the
2240  *                              hash conflict
2241  *
2242  * \param[out] ppage            the page which holds the minum entry
2243  *
2244  * \retval                      = 0 get the entry successfully
2245  *                              negative errno (< 0) does not get the entry
2246  */
2247 static int lmv_get_min_striped_entry(struct obd_export *exp,
2248                                      struct md_op_data *op_data,
2249                                      struct md_callback *cb_op,
2250                                      __u64 hash_offset, int *stripe_offset,
2251                                      struct lu_dirent **entp,
2252                                      struct page **ppage)
2253 {
2254         struct obd_device       *obd = exp->exp_obd;
2255         struct lmv_obd          *lmv = &obd->u.lmv;
2256         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2257         struct lmv_tgt_desc     *tgt;
2258         int                     stripe_count;
2259         struct lu_dirent        *min_ent = NULL;
2260         struct page             *min_page = NULL;
2261         int                     min_idx = 0;
2262         int                     i;
2263         int                     rc = 0;
2264         ENTRY;
2265
2266         stripe_count = lsm->lsm_md_stripe_count;
2267         for (i = 0; i < stripe_count; i++) {
2268                 struct lu_dirent        *ent = NULL;
2269                 struct page             *page = NULL;
2270                 struct lu_dirpage       *dp;
2271                 __u64                   stripe_hash = hash_offset;
2272
2273                 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2274                 if (IS_ERR(tgt))
2275                         GOTO(out, rc = PTR_ERR(tgt));
2276
2277                 /* op_data will be shared by each stripe, so we need
2278                  * reset these value for each stripe */
2279                 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2280                 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2281                 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2282 next:
2283                 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2284                                   &page);
2285                 if (rc != 0)
2286                         GOTO(out, rc);
2287
2288                 dp = page_address(page);
2289                 for (ent = lu_dirent_start(dp); ent != NULL;
2290                      ent = lu_dirent_next(ent)) {
2291                         /* Skip dummy entry */
2292                         if (le16_to_cpu(ent->lde_namelen) == 0)
2293                                 continue;
2294
2295                         if (le64_to_cpu(ent->lde_hash) < hash_offset)
2296                                 continue;
2297
2298                         if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2299                             (*entp == ent || i < *stripe_offset))
2300                                 continue;
2301
2302                         /* skip . and .. for other stripes */
2303                         if (i != 0 &&
2304                             (strncmp(ent->lde_name, ".",
2305                                      le16_to_cpu(ent->lde_namelen)) == 0 ||
2306                              strncmp(ent->lde_name, "..",
2307                                      le16_to_cpu(ent->lde_namelen)) == 0))
2308                                 continue;
2309                         break;
2310                 }
2311
2312                 if (ent == NULL) {
2313                         stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2314
2315                         kunmap(page);
2316                         page_cache_release(page);
2317                         page = NULL;
2318
2319                         /* reach the end of current stripe, go to next stripe */
2320                         if (stripe_hash == MDS_DIR_END_OFF)
2321                                 continue;
2322                         else
2323                                 goto next;
2324                 }
2325
2326                 if (min_ent != NULL) {
2327                         if (le64_to_cpu(min_ent->lde_hash) >
2328                             le64_to_cpu(ent->lde_hash)) {
2329                                 min_ent = ent;
2330                                 kunmap(min_page);
2331                                 page_cache_release(min_page);
2332                                 min_idx = i;
2333                                 min_page = page;
2334                         } else {
2335                                 kunmap(page);
2336                                 page_cache_release(page);
2337                                 page = NULL;
2338                         }
2339                 } else {
2340                         min_ent = ent;
2341                         min_page = page;
2342                         min_idx = i;
2343                 }
2344         }
2345
2346 out:
2347         if (*ppage != NULL) {
2348                 kunmap(*ppage);
2349                 page_cache_release(*ppage);
2350         }
2351         *stripe_offset = min_idx;
2352         *entp = min_ent;
2353         *ppage = min_page;
2354         RETURN(rc);
2355 }
2356
2357 /**
2358  * Build dir entry page from a striped directory
2359  *
2360  * This function gets one entry by @offset from a striped directory. It will
2361  * read entries from all of stripes, and choose one closest to the required
2362  * offset(&offset). A few notes
2363  * 1. skip . and .. for non-zero stripes, because there can only have one .
2364  * and .. in a directory.
2365  * 2. op_data will be shared by all of stripes, instead of allocating new
2366  * one, so need to restore before reusing.
2367  * 3. release the entry page if that is not being chosen.
2368  *
2369  * \param[in] exp       obd export refer to LMV
2370  * \param[in] op_data   hold those MD parameters of read_entry
2371  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2372  * \param[out] ldp      the entry being read
2373  * \param[out] ppage    the page holding the entry. Note: because the entry
2374  *                      will be accessed in upper layer, so we need hold the
2375  *                      page until the usages of entry is finished, see
2376  *                      ll_dir_entry_next.
2377  *
2378  * retval               =0 if get entry successfully
2379  *                      <0 cannot get entry
2380  */
2381 static int lmv_read_striped_page(struct obd_export *exp,
2382                                  struct md_op_data *op_data,
2383                                  struct md_callback *cb_op,
2384                                  __u64 offset, struct page **ppage)
2385 {
2386         struct obd_device       *obd = exp->exp_obd;
2387         struct lu_fid           master_fid = op_data->op_fid1;
2388         struct inode            *master_inode = op_data->op_data;
2389         __u64                   hash_offset = offset;
2390         struct lu_dirpage       *dp;
2391         struct page             *min_ent_page = NULL;
2392         struct page             *ent_page = NULL;
2393         struct lu_dirent        *ent;
2394         void                    *area;
2395         int                     ent_idx = 0;
2396         struct lu_dirent        *min_ent = NULL;
2397         struct lu_dirent        *last_ent;
2398         size_t                  left_bytes;
2399         int                     rc;
2400         ENTRY;
2401
2402         rc = lmv_check_connect(obd);
2403         if (rc)
2404                 RETURN(rc);
2405
2406         /* Allocate a page and read entries from all of stripes and fill
2407          * the page by hash order */
2408         ent_page = alloc_page(GFP_KERNEL);
2409         if (ent_page == NULL)
2410                 RETURN(-ENOMEM);
2411
2412         /* Initialize the entry page */
2413         dp = kmap(ent_page);
2414         memset(dp, 0, sizeof(*dp));
2415         dp->ldp_hash_start = cpu_to_le64(offset);
2416         dp->ldp_flags |= LDF_COLLIDE;
2417
2418         area = dp + 1;
2419         left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2420         ent = area;
2421         last_ent = ent;
2422         do {
2423                 __u16   ent_size;
2424
2425                 /* Find the minum entry from all sub-stripes */
2426                 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2427                                                &ent_idx, &min_ent,
2428                                                &min_ent_page);
2429                 if (rc != 0)
2430                         GOTO(out, rc);
2431
2432                 /* If it can not get minum entry, it means it already reaches
2433                  * the end of this directory */
2434                 if (min_ent == NULL) {
2435                         last_ent->lde_reclen = 0;
2436                         hash_offset = MDS_DIR_END_OFF;
2437                         GOTO(out, rc);
2438                 }
2439
2440                 ent_size = le16_to_cpu(min_ent->lde_reclen);
2441
2442                 /* the last entry lde_reclen is 0, but it might not
2443                  * the end of this entry of this temporay entry */
2444                 if (ent_size == 0)
2445                         ent_size = lu_dirent_calc_size(
2446                                         le16_to_cpu(min_ent->lde_namelen),
2447                                         le32_to_cpu(min_ent->lde_attrs));
2448                 if (ent_size > left_bytes) {
2449                         last_ent->lde_reclen = cpu_to_le16(0);
2450                         hash_offset = le64_to_cpu(min_ent->lde_hash);
2451                         GOTO(out, rc);
2452                 }
2453
2454                 memcpy(ent, min_ent, ent_size);
2455
2456                 /* Replace . with master FID and Replace .. with the parent FID
2457                  * of master object */
2458                 if (strncmp(ent->lde_name, ".",
2459                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2460                     le16_to_cpu(ent->lde_namelen) == 1)
2461                         fid_cpu_to_le(&ent->lde_fid, &master_fid);
2462                 else if (strncmp(ent->lde_name, "..",
2463                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2464                            le16_to_cpu(ent->lde_namelen) == 2)
2465                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2466
2467                 left_bytes -= ent_size;
2468                 ent->lde_reclen = cpu_to_le16(ent_size);
2469                 last_ent = ent;
2470                 ent = (void *)ent + ent_size;
2471                 hash_offset = le64_to_cpu(min_ent->lde_hash);
2472                 if (hash_offset == MDS_DIR_END_OFF) {
2473                         last_ent->lde_reclen = 0;
2474                         break;
2475                 }
2476         } while (1);
2477 out:
2478         if (min_ent_page != NULL) {
2479                 kunmap(min_ent_page);
2480                 page_cache_release(min_ent_page);
2481         }
2482
2483         if (unlikely(rc != 0)) {
2484                 __free_page(ent_page);
2485                 ent_page = NULL;
2486         } else {
2487                 if (ent == area)
2488                         dp->ldp_flags |= LDF_EMPTY;
2489                 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2490                 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2491         }
2492
2493         /* We do not want to allocate md_op_data during each
2494          * dir entry reading, so op_data will be shared by every stripe,
2495          * then we need to restore it back to original value before
2496          * return to the upper layer */
2497         op_data->op_fid1 = master_fid;
2498         op_data->op_fid2 = master_fid;
2499         op_data->op_data = master_inode;
2500
2501         *ppage = ent_page;
2502
2503         RETURN(rc);
2504 }
2505
2506 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2507                   struct md_callback *cb_op, __u64 offset,
2508                   struct page **ppage)
2509 {
2510         struct obd_device       *obd = exp->exp_obd;
2511         struct lmv_obd          *lmv = &obd->u.lmv;
2512         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2513         struct lmv_tgt_desc     *tgt;
2514         int                     rc;
2515         ENTRY;
2516
2517         rc = lmv_check_connect(obd);
2518         if (rc != 0)
2519                 RETURN(rc);
2520
2521         if (unlikely(lsm != NULL)) {
2522                 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2523                 RETURN(rc);
2524         }
2525
2526         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2527         if (IS_ERR(tgt))
2528                 RETURN(PTR_ERR(tgt));
2529
2530         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2531
2532         RETURN(rc);
2533 }
2534
2535 /**
2536  * Unlink a file/directory
2537  *
2538  * Unlink a file or directory under the parent dir. The unlink request
2539  * usually will be sent to the MDT where the child is located, but if
2540  * the client does not have the child FID then request will be sent to the
2541  * MDT where the parent is located.
2542  *
2543  * If the parent is a striped directory then it also needs to locate which
2544  * stripe the name of the child is located, and replace the parent FID
2545  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2546  * it will walk through all of sub-stripes until the child is being
2547  * unlinked finally.
2548  *
2549  * \param[in] exp       export refer to LMV
2550  * \param[in] op_data   different parameters transferred beween client
2551  *                      MD stacks, name, namelen, FIDs etc.
2552  *                      op_fid1 is the parent FID, op_fid2 is the child
2553  *                      FID.
2554  * \param[out] request  point to the request of unlink.
2555  *
2556  * retval               0 if succeed
2557  *                      negative errno if failed.
2558  */
2559 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2560                       struct ptlrpc_request **request)
2561 {
2562         struct obd_device       *obd = exp->exp_obd;
2563         struct lmv_obd          *lmv = &obd->u.lmv;
2564         struct lmv_tgt_desc     *tgt = NULL;
2565         struct lmv_tgt_desc     *parent_tgt = NULL;
2566         struct mdt_body         *body;
2567         int                     rc;
2568         int                     stripe_index = 0;
2569         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2570         ENTRY;
2571
2572         rc = lmv_check_connect(obd);
2573         if (rc)
2574                 RETURN(rc);
2575 retry_unlink:
2576         /* For striped dir, we need to locate the parent as well */
2577         if (lsm != NULL) {
2578                 struct lmv_tgt_desc *tmp;
2579
2580                 LASSERT(op_data->op_name != NULL &&
2581                         op_data->op_namelen != 0);
2582
2583                 tmp = lmv_locate_target_for_name(lmv, lsm,
2584                                                  op_data->op_name,
2585                                                  op_data->op_namelen,
2586                                                  &op_data->op_fid1,
2587                                                  &op_data->op_mds);
2588
2589                 /* return -EBADFD means unknown hash type, might
2590                  * need try all sub-stripe here */
2591                 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2592                         RETURN(PTR_ERR(tmp));
2593
2594                 /* Note: both migrating dir and unknown hash dir need to
2595                  * try all of sub-stripes, so we need start search the
2596                  * name from stripe 0, but migrating dir is already handled
2597                  * inside lmv_locate_target_for_name(), so we only check
2598                  * unknown hash type directory here */
2599                 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2600                         struct lmv_oinfo *oinfo;
2601
2602                         oinfo = &lsm->lsm_md_oinfo[stripe_index];
2603
2604                         op_data->op_fid1 = oinfo->lmo_fid;
2605                         op_data->op_mds = oinfo->lmo_mds;
2606                 }
2607         }
2608
2609 try_next_stripe:
2610         /* Send unlink requests to the MDT where the child is located */
2611         if (likely(!fid_is_zero(&op_data->op_fid2)))
2612                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2613         else if (lsm != NULL)
2614                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2615         else
2616                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2617
2618         if (IS_ERR(tgt))
2619                 RETURN(PTR_ERR(tgt));
2620
2621         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2622         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2623         op_data->op_cap = cfs_curproc_cap_pack();
2624
2625         /*
2626          * If child's fid is given, cancel unused locks for it if it is from
2627          * another export than parent.
2628          *
2629          * LOOKUP lock for child (fid3) should also be cancelled on parent
2630          * tgt_tgt in mdc_unlink().
2631          */
2632         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2633
2634         /*
2635          * Cancel FULL locks on child (fid3).
2636          */
2637         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2638         if (IS_ERR(parent_tgt))
2639                 RETURN(PTR_ERR(parent_tgt));
2640
2641         if (parent_tgt != tgt) {
2642                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2643                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2644                                       MF_MDC_CANCEL_FID3);
2645         }
2646
2647         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2648                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2649         if (rc != 0)
2650                 RETURN(rc);
2651
2652         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2653                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2654
2655         rc = md_unlink(tgt->ltd_exp, op_data, request);
2656         if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2657                 RETURN(rc);
2658
2659         /* Try next stripe if it is needed. */
2660         if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2661                 struct lmv_oinfo *oinfo;
2662
2663                 stripe_index++;
2664                 if (stripe_index >= lsm->lsm_md_stripe_count)
2665                         RETURN(rc);
2666
2667                 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2668
2669                 op_data->op_fid1 = oinfo->lmo_fid;
2670                 op_data->op_mds = oinfo->lmo_mds;
2671
2672                 ptlrpc_req_finished(*request);
2673                 *request = NULL;
2674
2675                 goto try_next_stripe;
2676         }
2677
2678         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2679         if (body == NULL)
2680                 RETURN(-EPROTO);
2681
2682         /* Not cross-ref case, just get out of here. */
2683         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2684                 RETURN(rc);
2685
2686         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2687                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2688
2689         /* This is a remote object, try remote MDT, Note: it may
2690          * try more than 1 time here, Considering following case
2691          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2692          * 1. Initially A does not know where remote1 is, it send
2693          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2694          *    resend unlink RPC to MDT1 (retry 1st time).
2695          *
2696          * 2. During the unlink RPC in flight,
2697          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2698          *    and create new remote1, but on MDT0
2699          *
2700          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2701          *    /mnt/lustre, then lookup get fid of remote1, and find
2702          *    it is remote dir again, and replay -EREMOTE again.
2703          *
2704          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2705          *
2706          * In theory, it might try unlimited time here, but it should
2707          * be very rare case.  */
2708         op_data->op_fid2 = body->mbo_fid1;
2709         ptlrpc_req_finished(*request);
2710         *request = NULL;
2711
2712         goto retry_unlink;
2713 }
2714
2715 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2716 {
2717         struct lmv_obd *lmv = &obd->u.lmv;
2718         int rc = 0;
2719
2720         switch (stage) {
2721         case OBD_CLEANUP_EARLY:
2722                 /* XXX: here should be calling obd_precleanup() down to
2723                  * stack. */
2724                 break;
2725         case OBD_CLEANUP_EXPORTS:
2726                 fld_client_proc_fini(&lmv->lmv_fld);
2727                 lprocfs_obd_cleanup(obd);
2728                 lprocfs_free_md_stats(obd);
2729                 break;
2730         default:
2731                 break;
2732         }
2733         RETURN(rc);
2734 }
2735
2736 /**
2737  * Get by key a value associated with a LMV device.
2738  *
2739  * Dispatch request to lower-layer devices as needed.
2740  *
2741  * \param[in] env               execution environment for this thread
2742  * \param[in] exp               export for the LMV device
2743  * \param[in] keylen            length of key identifier
2744  * \param[in] key               identifier of key to get value for
2745  * \param[in] vallen            size of \a val
2746  * \param[out] val              pointer to storage location for value
2747  * \param[in] lsm               optional striping metadata of object
2748  *
2749  * \retval 0            on success
2750  * \retval negative     negated errno on failure
2751  */
2752 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2753                         __u32 keylen, void *key, __u32 *vallen, void *val)
2754 {
2755         struct obd_device       *obd;
2756         struct lmv_obd          *lmv;
2757         int                      rc = 0;
2758         ENTRY;
2759
2760         obd = class_exp2obd(exp);
2761         if (obd == NULL) {
2762                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2763                        exp->exp_handle.h_cookie);
2764                 RETURN(-EINVAL);
2765         }
2766
2767         lmv = &obd->u.lmv;
2768         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2769                 int i;
2770
2771                 rc = lmv_check_connect(obd);
2772                 if (rc)
2773                         RETURN(rc);
2774
2775                 LASSERT(*vallen == sizeof(__u32));
2776                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2777                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2778                         /*
2779                          * All tgts should be connected when this gets called.
2780                          */
2781                         if (tgt == NULL || tgt->ltd_exp == NULL)
2782                                 continue;
2783
2784                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2785                                           vallen, val))
2786                                 RETURN(0);
2787                 }
2788                 RETURN(-EINVAL);
2789         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2790                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2791                    KEY_IS(KEY_CONN_DATA)) {
2792                 rc = lmv_check_connect(obd);
2793                 if (rc)
2794                         RETURN(rc);
2795
2796                 /*
2797                  * Forwarding this request to first MDS, it should know LOV
2798                  * desc.
2799                  */
2800                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2801                                   vallen, val);
2802                 if (!rc && KEY_IS(KEY_CONN_DATA))
2803                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2804                 RETURN(rc);
2805         } else if (KEY_IS(KEY_TGT_COUNT)) {
2806                 *((int *)val) = lmv->desc.ld_tgt_count;
2807                 RETURN(0);
2808         }
2809
2810         CDEBUG(D_IOCTL, "Invalid key\n");
2811         RETURN(-EINVAL);
2812 }
2813
2814 /**
2815  * Asynchronously set by key a value associated with a LMV device.
2816  *
2817  * Dispatch request to lower-layer devices as needed.
2818  *
2819  * \param[in] env       execution environment for this thread
2820  * \param[in] exp       export for the LMV device
2821  * \param[in] keylen    length of key identifier
2822  * \param[in] key       identifier of key to store value for
2823  * \param[in] vallen    size of value to store
2824  * \param[in] val       pointer to data to be stored
2825  * \param[in] set       optional list of related ptlrpc requests
2826  *
2827  * \retval 0            on success
2828  * \retval negative     negated errno on failure
2829  */
2830 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2831                         __u32 keylen, void *key, __u32 vallen, void *val,
2832                         struct ptlrpc_request_set *set)
2833 {
2834         struct lmv_tgt_desc     *tgt = NULL;
2835         struct obd_device       *obd;
2836         struct lmv_obd          *lmv;
2837         int rc = 0;
2838         ENTRY;
2839
2840         obd = class_exp2obd(exp);
2841         if (obd == NULL) {
2842                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2843                        exp->exp_handle.h_cookie);
2844                 RETURN(-EINVAL);
2845         }
2846         lmv = &obd->u.lmv;
2847
2848         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2849             KEY_IS(KEY_DEFAULT_EASIZE)) {
2850                 int i, err = 0;
2851
2852                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2853                         tgt = lmv->tgts[i];
2854
2855                         if (tgt == NULL || tgt->ltd_exp == NULL)
2856                                 continue;
2857
2858                         err = obd_set_info_async(env, tgt->ltd_exp,
2859                                                  keylen, key, vallen, val, set);
2860                         if (err && rc == 0)
2861                                 rc = err;
2862                 }
2863
2864                 RETURN(rc);
2865         }
2866
2867         RETURN(-EINVAL);
2868 }
2869
2870 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2871                             const struct lmv_mds_md_v1 *lmm1)
2872 {
2873         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
2874         int             stripe_count;
2875         int             cplen;
2876         int             i;
2877         int             rc = 0;
2878         ENTRY;
2879
2880         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2881         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2882         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2883         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2884                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2885         else
2886                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2887         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2888         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2889                         sizeof(lsm->lsm_md_pool_name));
2890
2891         if (cplen >= sizeof(lsm->lsm_md_pool_name))
2892                 RETURN(-E2BIG);
2893
2894         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2895                "layout_version %d\n", lsm->lsm_md_stripe_count,
2896                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2897                lsm->lsm_md_layout_version);
2898
2899         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2900         for (i = 0; i < stripe_count; i++) {
2901                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2902                               &lmm1->lmv_stripe_fids[i]);
2903                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2904                                     &lsm->lsm_md_oinfo[i].lmo_mds);
2905                 if (rc != 0)
2906                         RETURN(rc);
2907                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2908                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2909         }
2910
2911         RETURN(rc);
2912 }
2913
2914 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2915                   const union lmv_mds_md *lmm, int stripe_count)
2916 {
2917         struct lmv_stripe_md     *lsm;
2918         int                      lsm_size;
2919         int                      rc;
2920         bool                     allocated = false;
2921         ENTRY;
2922
2923         LASSERT(lsmp != NULL);
2924
2925         lsm = *lsmp;
2926         /* Free memmd */
2927         if (lsm != NULL && lmm == NULL) {
2928                 int i;
2929                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2930                         /* For migrating inode, the master stripe and master
2931                          * object will be the same, so do not need iput, see
2932                          * ll_update_lsm_md */
2933                         if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2934                               i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2935                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
2936                 }
2937                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2938                 OBD_FREE(lsm, lsm_size);
2939                 *lsmp = NULL;
2940                 RETURN(0);
2941         }
2942
2943         /* Alloc memmd */
2944         if (lsm == NULL && lmm == NULL) {
2945                 lsm_size = lmv_stripe_md_size(stripe_count);
2946                 OBD_ALLOC(lsm, lsm_size);
2947                 if (lsm == NULL)
2948                         RETURN(-ENOMEM);
2949                 lsm->lsm_md_stripe_count = stripe_count;
2950                 *lsmp = lsm;
2951                 RETURN(0);
2952         }
2953
2954         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
2955                 RETURN(-EPERM);
2956
2957         /* Unpack memmd */
2958         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
2959             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
2960                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
2961                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
2962                        -EIO);
2963                 RETURN(-EIO);
2964         }
2965
2966         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
2967                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2968         else
2969                 /**
2970                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
2971                  * stripecount should be 0 then.
2972                  */
2973                 lsm_size = lmv_stripe_md_size(0);
2974
2975         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2976         if (lsm == NULL) {
2977                 OBD_ALLOC(lsm, lsm_size);
2978                 if (lsm == NULL)
2979                         RETURN(-ENOMEM);
2980                 allocated = true;
2981                 *lsmp = lsm;
2982         }
2983
2984         switch (le32_to_cpu(lmm->lmv_magic)) {
2985         case LMV_MAGIC_V1:
2986                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
2987                 break;
2988         default:
2989                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
2990                        le32_to_cpu(lmm->lmv_magic));
2991                 rc = -EINVAL;
2992                 break;
2993         }
2994
2995         if (rc != 0 && allocated) {
2996                 OBD_FREE(lsm, lsm_size);
2997                 *lsmp = NULL;
2998                 lsm_size = rc;
2999         }
3000         RETURN(lsm_size);
3001 }
3002
3003 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3004 {
3005         return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3006 }
3007
3008 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3009 {
3010         lmv_unpack_md(NULL, &lsm, NULL, 0);
3011 }
3012 EXPORT_SYMBOL(lmv_free_memmd);
3013
3014 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3015                  struct lov_mds_md *lmm, int disk_len)
3016 {
3017         return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3018                              (union lmv_mds_md *)lmm, disk_len);
3019 }
3020
3021 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3022                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
3023                              ldlm_cancel_flags_t flags, void *opaque)
3024 {
3025         struct obd_device       *obd = exp->exp_obd;
3026         struct lmv_obd          *lmv = &obd->u.lmv;
3027         int                      rc = 0;
3028         int                      err;
3029         __u32                    i;
3030         ENTRY;
3031
3032         LASSERT(fid != NULL);
3033
3034         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3035                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3036
3037                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3038                         continue;
3039
3040                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3041                                        opaque);
3042                 if (!rc)
3043                         rc = err;
3044         }
3045         RETURN(rc);
3046 }
3047
3048 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3049                       __u64 *bits)
3050 {
3051         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3052         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3053         int                      rc;
3054         ENTRY;
3055
3056         if (tgt == NULL || tgt->ltd_exp == NULL)
3057                 RETURN(-EINVAL);
3058         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3059         RETURN(rc);
3060 }
3061
3062 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3063                            const struct lu_fid *fid, ldlm_type_t type,
3064                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
3065                            struct lustre_handle *lockh)
3066 {
3067         struct obd_device       *obd = exp->exp_obd;
3068         struct lmv_obd          *lmv = &obd->u.lmv;
3069         ldlm_mode_t             rc;
3070         int                     tgt;
3071         int                     i;
3072         ENTRY;
3073
3074         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3075
3076         /*
3077          * With DNE every object can have two locks in different namespaces:
3078          * lookup lock in space of MDT storing direntry and update/open lock in
3079          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3080          * since this can be easily found, and only try others if that fails.
3081          */
3082         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3083              i < lmv->desc.ld_tgt_count;
3084              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3085                 if (tgt < 0) {
3086                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3087                                obd->obd_name, PFID(fid), tgt);
3088                         tgt = 0;
3089                 }
3090
3091                 if (lmv->tgts[tgt] == NULL ||
3092                     lmv->tgts[tgt]->ltd_exp == NULL ||
3093                     lmv->tgts[tgt]->ltd_active == 0)
3094                         continue;
3095
3096                 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3097                                    type, policy, mode, lockh);
3098                 if (rc)
3099                         RETURN(rc);
3100         }
3101
3102         RETURN(0);
3103 }
3104
3105 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3106                       struct obd_export *dt_exp, struct obd_export *md_exp,
3107                       struct lustre_md *md)
3108 {
3109         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3110         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3111
3112         if (tgt == NULL || tgt->ltd_exp == NULL)
3113                 RETURN(-EINVAL);
3114
3115         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3116 }
3117
3118 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3119 {
3120         struct obd_device       *obd = exp->exp_obd;
3121         struct lmv_obd          *lmv = &obd->u.lmv;
3122         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3123         ENTRY;
3124
3125         if (md->lmv != NULL) {
3126                 lmv_free_memmd(md->lmv);
3127                 md->lmv = NULL;
3128         }
3129         if (tgt == NULL || tgt->ltd_exp == NULL)
3130                 RETURN(-EINVAL);
3131         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3132 }
3133
3134 int lmv_set_open_replay_data(struct obd_export *exp,
3135                              struct obd_client_handle *och,
3136                              struct lookup_intent *it)
3137 {
3138         struct obd_device       *obd = exp->exp_obd;
3139         struct lmv_obd          *lmv = &obd->u.lmv;
3140         struct lmv_tgt_desc     *tgt;
3141         ENTRY;
3142
3143         tgt = lmv_find_target(lmv, &och->och_fid);
3144         if (IS_ERR(tgt))
3145                 RETURN(PTR_ERR(tgt));
3146
3147         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3148 }
3149
3150 int lmv_clear_open_replay_data(struct obd_export *exp,
3151                                struct obd_client_handle *och)
3152 {
3153         struct obd_device       *obd = exp->exp_obd;
3154         struct lmv_obd          *lmv = &obd->u.lmv;
3155         struct lmv_tgt_desc     *tgt;
3156         ENTRY;
3157
3158         tgt = lmv_find_target(lmv, &och->och_fid);
3159         if (IS_ERR(tgt))
3160                 RETURN(PTR_ERR(tgt));
3161
3162         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3163 }
3164
3165 static int lmv_get_remote_perm(struct obd_export *exp,
3166                                const struct lu_fid *fid,
3167                                struct obd_capa *oc, __u32 suppgid,
3168                                struct ptlrpc_request **request)
3169 {
3170         struct obd_device       *obd = exp->exp_obd;
3171         struct lmv_obd          *lmv = &obd->u.lmv;
3172         struct lmv_tgt_desc     *tgt;
3173         int                      rc;
3174         ENTRY;
3175
3176         rc = lmv_check_connect(obd);
3177         if (rc)
3178                 RETURN(rc);
3179
3180         tgt = lmv_find_target(lmv, fid);
3181         if (IS_ERR(tgt))
3182                 RETURN(PTR_ERR(tgt));
3183
3184         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3185         RETURN(rc);
3186 }
3187
3188 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3189                           renew_capa_cb_t cb)
3190 {
3191         struct obd_device       *obd = exp->exp_obd;
3192         struct lmv_obd          *lmv = &obd->u.lmv;
3193         struct lmv_tgt_desc     *tgt;
3194         int                      rc;
3195         ENTRY;
3196
3197         rc = lmv_check_connect(obd);
3198         if (rc)
3199                 RETURN(rc);
3200
3201         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3202         if (IS_ERR(tgt))
3203                 RETURN(PTR_ERR(tgt));
3204
3205         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3206         RETURN(rc);
3207 }
3208
3209 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3210                     const struct req_msg_field *field, struct obd_capa **oc)
3211 {
3212         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3213         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3214
3215         if (tgt == NULL || tgt->ltd_exp == NULL)
3216                 RETURN(-EINVAL);
3217         return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3218 }
3219
3220 int lmv_intent_getattr_async(struct obd_export *exp,
3221                              struct md_enqueue_info *minfo,
3222                              struct ldlm_enqueue_info *einfo)
3223 {
3224         struct md_op_data       *op_data = &minfo->mi_data;
3225         struct obd_device       *obd = exp->exp_obd;
3226         struct lmv_obd          *lmv = &obd->u.lmv;
3227         struct lmv_tgt_desc     *tgt = NULL;
3228         int                      rc;
3229         ENTRY;
3230
3231         rc = lmv_check_connect(obd);
3232         if (rc)
3233                 RETURN(rc);
3234
3235         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3236         if (IS_ERR(tgt))
3237                 RETURN(PTR_ERR(tgt));
3238
3239         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3240         RETURN(rc);
3241 }
3242
3243 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3244                         struct lu_fid *fid, __u64 *bits)
3245 {
3246         struct obd_device       *obd = exp->exp_obd;
3247         struct lmv_obd          *lmv = &obd->u.lmv;
3248         struct lmv_tgt_desc     *tgt;
3249         int                      rc;
3250         ENTRY;
3251
3252         rc = lmv_check_connect(obd);
3253         if (rc)
3254                 RETURN(rc);
3255
3256         tgt = lmv_find_target(lmv, fid);
3257         if (IS_ERR(tgt))
3258                 RETURN(PTR_ERR(tgt));
3259
3260         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3261         RETURN(rc);
3262 }
3263
3264 int lmv_get_fid_from_lsm(struct obd_export *exp,
3265                          const struct lmv_stripe_md *lsm,
3266                          const char *name, int namelen, struct lu_fid *fid)
3267 {
3268         const struct lmv_oinfo *oinfo;
3269
3270         LASSERT(lsm != NULL);
3271         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3272         if (IS_ERR(oinfo))
3273                 return PTR_ERR(oinfo);
3274
3275         *fid = oinfo->lmo_fid;
3276
3277         RETURN(0);
3278 }
3279
3280 /**
3281  * For lmv, only need to send request to master MDT, and the master MDT will
3282  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3283  * we directly fetch data from the slave MDTs.
3284  */
3285 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3286                  struct obd_quotactl *oqctl)
3287 {
3288         struct obd_device   *obd = class_exp2obd(exp);
3289         struct lmv_obd      *lmv = &obd->u.lmv;
3290         struct lmv_tgt_desc *tgt = lmv->tgts[0];
3291         int                  rc = 0;
3292         __u32                i;
3293         __u64                curspace, curinodes;
3294         ENTRY;
3295
3296         if (tgt == NULL ||
3297             tgt->ltd_exp == NULL ||
3298             !tgt->ltd_active ||
3299             lmv->desc.ld_tgt_count == 0) {
3300                 CERROR("master lmv inactive\n");
3301                 RETURN(-EIO);
3302         }
3303
3304         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3305                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3306                 RETURN(rc);
3307         }
3308
3309         curspace = curinodes = 0;
3310         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3311                 int err;
3312                 tgt = lmv->tgts[i];
3313
3314                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3315                         continue;
3316
3317                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3318                 if (err) {
3319                         CERROR("getquota on mdt %d failed. %d\n", i, err);
3320                         if (!rc)
3321                                 rc = err;
3322                 } else {
3323                         curspace += oqctl->qc_dqblk.dqb_curspace;
3324                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3325                 }
3326         }
3327         oqctl->qc_dqblk.dqb_curspace = curspace;
3328         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3329
3330         RETURN(rc);
3331 }
3332
3333 static int lmv_merge_attr(struct obd_export *exp,
3334                           const struct lmv_stripe_md *lsm,
3335                           struct cl_attr *attr,
3336                           ldlm_blocking_callback cb_blocking)
3337 {
3338         int rc;
3339         int i;
3340
3341         rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3342         if (rc < 0)
3343                 return rc;
3344
3345         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3346                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3347
3348                 CDEBUG(D_INFO, ""DFID" size %llu, blocks %llu nlink %u,"
3349                        " atime %lu ctime %lu, mtime %lu.\n",
3350                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3351                        i_size_read(inode), (unsigned long long)inode->i_blocks,
3352                        inode->i_nlink, LTIME_S(inode->i_atime),
3353                        LTIME_S(inode->i_ctime), LTIME_S(inode->i_mtime));
3354
3355                 /* for slave stripe, it needs to subtract nlink for . and .. */
3356                 if (i != 0)
3357                         attr->cat_nlink += inode->i_nlink - 2;
3358                 else
3359                         attr->cat_nlink = inode->i_nlink;
3360
3361                 attr->cat_size += i_size_read(inode);
3362                 attr->cat_blocks += inode->i_blocks;
3363
3364                 if (attr->cat_atime < LTIME_S(inode->i_atime))
3365                         attr->cat_atime = LTIME_S(inode->i_atime);
3366
3367                 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3368                         attr->cat_ctime = LTIME_S(inode->i_ctime);
3369
3370                 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3371                         attr->cat_mtime = LTIME_S(inode->i_mtime);
3372         }
3373         return 0;
3374 }
3375
3376 struct obd_ops lmv_obd_ops = {
3377         .o_owner                = THIS_MODULE,
3378         .o_setup                = lmv_setup,
3379         .o_cleanup              = lmv_cleanup,
3380         .o_precleanup           = lmv_precleanup,
3381         .o_process_config       = lmv_process_config,
3382         .o_connect              = lmv_connect,
3383         .o_disconnect           = lmv_disconnect,
3384         .o_statfs               = lmv_statfs,
3385         .o_get_info             = lmv_get_info,
3386         .o_set_info_async       = lmv_set_info_async,
3387         .o_unpackmd             = lmv_unpackmd,
3388         .o_notify               = lmv_notify,
3389         .o_get_uuid             = lmv_get_uuid,
3390         .o_iocontrol            = lmv_iocontrol,
3391         .o_quotactl             = lmv_quotactl
3392 };
3393
3394 struct md_ops lmv_md_ops = {
3395         .m_getstatus            = lmv_getstatus,
3396         .m_null_inode           = lmv_null_inode,
3397         .m_find_cbdata          = lmv_find_cbdata,
3398         .m_close                = lmv_close,
3399         .m_create               = lmv_create,
3400         .m_enqueue              = lmv_enqueue,
3401         .m_getattr              = lmv_getattr,
3402         .m_getxattr             = lmv_getxattr,
3403         .m_getattr_name         = lmv_getattr_name,
3404         .m_intent_lock          = lmv_intent_lock,
3405         .m_link                 = lmv_link,
3406         .m_rename               = lmv_rename,
3407         .m_setattr              = lmv_setattr,
3408         .m_setxattr             = lmv_setxattr,
3409         .m_fsync                = lmv_fsync,
3410         .m_read_page            = lmv_read_page,
3411         .m_unlink               = lmv_unlink,
3412         .m_init_ea_size         = lmv_init_ea_size,
3413         .m_cancel_unused        = lmv_cancel_unused,
3414         .m_set_lock_data        = lmv_set_lock_data,
3415         .m_lock_match           = lmv_lock_match,
3416         .m_get_lustre_md        = lmv_get_lustre_md,
3417         .m_free_lustre_md       = lmv_free_lustre_md,
3418         .m_merge_attr           = lmv_merge_attr,
3419         .m_set_open_replay_data = lmv_set_open_replay_data,
3420         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3421         .m_renew_capa           = lmv_renew_capa,
3422         .m_unpack_capa          = lmv_unpack_capa,
3423         .m_get_remote_perm      = lmv_get_remote_perm,
3424         .m_intent_getattr_async = lmv_intent_getattr_async,
3425         .m_revalidate_lock      = lmv_revalidate_lock,
3426         .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
3427 };
3428
3429 int __init lmv_init(void)
3430 {
3431         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3432                                    LUSTRE_LMV_NAME, NULL);
3433 }
3434
3435 static void lmv_exit(void)
3436 {
3437         class_unregister_type(LUSTRE_LMV_NAME);
3438 }
3439
3440 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3441 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3442 MODULE_LICENSE("GPL");
3443
3444 module_init(lmv_init);
3445 module_exit(lmv_exit);