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, 2013, 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/slab.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <linux/math64.h>
  45 #include <linux/seq_file.h>
  46 #include <linux/namei.h>
  47
  48 #include <lustre/lustre_idl.h>
  49 #include <obd_support.h>
  50 #include <lustre_lib.h>
  51 #include <lustre_net.h>
  52 #include <obd_class.h>
  53 #include <lustre_lmv.h>
  54 #include <lprocfs_status.h>
  55 #include <cl_object.h>
  56 #include <lclient.h>
  57 #include <lustre_fid.h>
  58 #include <lustre_ioctl.h>
  59 #include "lmv_internal.h"
  60
  61 static void lmv_activate_target(struct lmv_obd *lmv,
  62                                 struct lmv_tgt_desc *tgt,
  63                                 int activate)
  64 {
  65         if (tgt->ltd_active == activate)
  66                 return;
  67
  68         tgt->ltd_active = activate;
  69         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  70 }
  71
  72 /**
  73  * Error codes:
  74  *
  75  *  -EINVAL  : UUID can't be found in the LMV's target list
  76  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  77  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  78  */
  79 static int lmv_set_mdc_active(struct lmv_obd *lmv,
  80                               const struct obd_uuid *uuid,
  81                               int activate)
  82 {
  83         struct lmv_tgt_desc     *tgt = NULL;
  84         struct obd_device       *obd;
  85         __u32                    i;
  86         int                      rc = 0;
  87         ENTRY;
  88
  89         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  90                         lmv, uuid->uuid, activate);
  91
  92         spin_lock(&lmv->lmv_lock);
  93         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
  94                 tgt = lmv->tgts[i];
  95                 if (tgt == NULL || tgt->ltd_exp == NULL)
  96                         continue;
  97
  98                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
  99                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 100
 101                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 102                         break;
 103         }
 104
 105         if (i == lmv->desc.ld_tgt_count)
 106                 GOTO(out_lmv_lock, rc = -EINVAL);
 107
 108         obd = class_exp2obd(tgt->ltd_exp);
 109         if (obd == NULL)
 110                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 111
 112         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 113                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 114                obd->obd_type->typ_name, i);
 115         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 116
 117         if (tgt->ltd_active == activate) {
 118                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 119                        activate ? "" : "in");
 120                 GOTO(out_lmv_lock, rc);
 121         }
 122
 123         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 124                activate ? "" : "in");
 125         lmv_activate_target(lmv, tgt, activate);
 126         EXIT;
 127
 128  out_lmv_lock:
 129         spin_unlock(&lmv->lmv_lock);
 130         return rc;
 131 }
 132
 133 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 134 {
 135         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
 136         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
 137
 138         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
 139 }
 140
 141 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 142                       enum obd_notify_event ev, void *data)
 143 {
 144         struct obd_connect_data *conn_data;
 145         struct lmv_obd          *lmv = &obd->u.lmv;
 146         struct obd_uuid         *uuid;
 147         int                      rc = 0;
 148         ENTRY;
 149
 150         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 151                 CERROR("unexpected notification of %s %s!\n",
 152                        watched->obd_type->typ_name,
 153                        watched->obd_name);
 154                 RETURN(-EINVAL);
 155         }
 156
 157         uuid = &watched->u.cli.cl_target_uuid;
 158         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 159                 /*
 160                  * Set MDC as active before notifying the observer, so the
 161                  * observer can use the MDC normally.
 162                  */
 163                 rc = lmv_set_mdc_active(lmv, uuid,
 164                                         ev == OBD_NOTIFY_ACTIVE);
 165                 if (rc) {
 166                         CERROR("%sactivation of %s failed: %d\n",
 167                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 168                                uuid->uuid, rc);
 169                         RETURN(rc);
 170                 }
 171         } else if (ev == OBD_NOTIFY_OCD) {
 172                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 173                 /*
 174                  * XXX: Make sure that ocd_connect_flags from all targets are
 175                  * the same. Otherwise one of MDTs runs wrong version or
 176                  * something like this.  --umka
 177                  */
 178                 obd->obd_self_export->exp_connect_data = *conn_data;
 179         }
 180 #if 0
 181         else if (ev == OBD_NOTIFY_DISCON) {
 182                 /*
 183                  * For disconnect event, flush fld cache for failout MDS case.
 184                  */
 185                 fld_client_flush(&lmv->lmv_fld);
 186         }
 187 #endif
 188         /*
 189          * Pass the notification up the chain.
 190          */
 191         if (obd->obd_observer)
 192                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 193
 194         RETURN(rc);
 195 }
 196
 197 /**
 198  * This is fake connect function. Its purpose is to initialize lmv and say
 199  * caller that everything is okay. Real connection will be performed later.
 200  */
 201 static int lmv_connect(const struct lu_env *env,
 202                        struct obd_export **exp, struct obd_device *obd,
 203                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 204                        void *localdata)
 205 {
 206         struct lmv_obd        *lmv = &obd->u.lmv;
 207         struct lustre_handle  conn = { 0 };
 208         int                    rc = 0;
 209         ENTRY;
 210
 211         /*
 212          * We don't want to actually do the underlying connections more than
 213          * once, so keep track.
 214          */
 215         lmv->refcount++;
 216         if (lmv->refcount > 1) {
 217                 *exp = NULL;
 218                 RETURN(0);
 219         }
 220
 221         rc = class_connect(&conn, obd, cluuid);
 222         if (rc) {
 223                 CERROR("class_connection() returned %d\n", rc);
 224                 RETURN(rc);
 225         }
 226
 227         *exp = class_conn2export(&conn);
 228         class_export_get(*exp);
 229
 230         lmv->exp = *exp;
 231         lmv->connected = 0;
 232         lmv->cluuid = *cluuid;
 233
 234         if (data)
 235                 lmv->conn_data = *data;
 236
 237         if (lmv->targets_proc_entry == NULL) {
 238                 lmv->targets_proc_entry = lprocfs_seq_register("target_obds",
 239                                                         obd->obd_proc_entry,
 240                                                         NULL, NULL);
 241                 if (IS_ERR(lmv->targets_proc_entry)) {
 242                         CERROR("%s: cannot register "
 243                                "/proc/fs/lustre/%s/%s/target_obds\n",
 244                                obd->obd_name, obd->obd_type->typ_name,
 245                                obd->obd_name);
 246                         lmv->targets_proc_entry = NULL;
 247                 }
 248         }
 249
 250         /*
 251          * All real clients should perform actual connection right away, because
 252          * it is possible, that LMV will not have opportunity to connect targets
 253          * and MDC stuff will be called directly, for instance while reading
 254          * ../mdc/../kbytesfree procfs file, etc.
 255          */
 256         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
 257                 rc = lmv_check_connect(obd);
 258
 259         if (rc && lmv->targets_proc_entry != NULL)
 260                 lprocfs_remove(&lmv->targets_proc_entry);
 261         RETURN(rc);
 262 }
 263
 264 static int lmv_init_ea_size(struct obd_export *exp, int easize,
 265                             int def_easize, int cookiesize, int def_cookiesize)
 266 {
 267         struct obd_device       *obd = exp->exp_obd;
 268         struct lmv_obd          *lmv = &obd->u.lmv;
 269         __u32                    i;
 270         int                      rc = 0;
 271         int                      change = 0;
 272         ENTRY;
 273
 274         if (lmv->max_easize < easize) {
 275                 lmv->max_easize = easize;
 276                 change = 1;
 277         }
 278         if (lmv->max_def_easize < def_easize) {
 279                 lmv->max_def_easize = def_easize;
 280                 change = 1;
 281         }
 282         if (lmv->max_cookiesize < cookiesize) {
 283                 lmv->max_cookiesize = cookiesize;
 284                 change = 1;
 285         }
 286         if (lmv->max_def_cookiesize < def_cookiesize) {
 287                 lmv->max_def_cookiesize = def_cookiesize;
 288                 change = 1;
 289         }
 290         if (change == 0)
 291                 RETURN(0);
 292
 293         if (lmv->connected == 0)
 294                 RETURN(0);
 295
 296         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 297                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 298
 299                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
 300                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 301                         continue;
 302                 }
 303
 304                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
 305                                      cookiesize, def_cookiesize);
 306                 if (rc) {
 307                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 308                                " rc = %d\n", obd->obd_name, i, rc);
 309                         break;
 310                 }
 311         }
 312         RETURN(rc);
 313 }
 314
 315 #define MAX_STRING_SIZE 128
 316
 317 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 318 {
 319         struct lmv_obd          *lmv = &obd->u.lmv;
 320         struct obd_uuid         *cluuid = &lmv->cluuid;
 321         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 322         struct obd_device       *mdc_obd;
 323         struct obd_export       *mdc_exp;
 324         struct lu_fld_target     target;
 325         int                      rc;
 326         ENTRY;
 327
 328         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 329                                         &obd->obd_uuid);
 330         if (!mdc_obd) {
 331                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 332                 RETURN(-EINVAL);
 333         }
 334
 335         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 336                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 337                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 338                 cluuid->uuid);
 339
 340         if (!mdc_obd->obd_set_up) {
 341                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 342                 RETURN(-EINVAL);
 343         }
 344
 345         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 346                          &lmv->conn_data, NULL);
 347         if (rc) {
 348                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 349                 RETURN(rc);
 350         }
 351
 352         /*
 353          * Init fid sequence client for this mdc and add new fld target.
 354          */
 355         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 356         if (rc)
 357                 RETURN(rc);
 358
 359         target.ft_srv = NULL;
 360         target.ft_exp = mdc_exp;
 361         target.ft_idx = tgt->ltd_idx;
 362
 363         fld_client_add_target(&lmv->lmv_fld, &target);
 364
 365         rc = obd_register_observer(mdc_obd, obd);
 366         if (rc) {
 367                 obd_disconnect(mdc_exp);
 368                 CERROR("target %s register_observer error %d\n",
 369                        tgt->ltd_uuid.uuid, rc);
 370                 RETURN(rc);
 371         }
 372
 373         if (obd->obd_observer) {
 374                 /*
 375                  * Tell the observer about the new target.
 376                  */
 377                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 378                                 OBD_NOTIFY_ACTIVE,
 379                                 (void *)(tgt - lmv->tgts[0]));
 380                 if (rc) {
 381                         obd_disconnect(mdc_exp);
 382                         RETURN(rc);
 383                 }
 384         }
 385
 386         tgt->ltd_active = 1;
 387         tgt->ltd_exp = mdc_exp;
 388         lmv->desc.ld_active_tgt_count++;
 389
 390         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
 391                         lmv->max_cookiesize, lmv->max_def_cookiesize);
 392
 393         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 394                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 395                 atomic_read(&obd->obd_refcount));
 396
 397         if (lmv->targets_proc_entry != NULL) {
 398                 struct proc_dir_entry *mdc_symlink;
 399
 400                 LASSERT(mdc_obd->obd_type != NULL);
 401                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 402                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 403                                                   lmv->targets_proc_entry,
 404                                                   "../../../%s/%s",
 405                                                   mdc_obd->obd_type->typ_name,
 406                                                   mdc_obd->obd_name);
 407                 if (mdc_symlink == NULL) {
 408                         CERROR("cannot register LMV target "
 409                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
 410                                obd->obd_type->typ_name, obd->obd_name,
 411                                mdc_obd->obd_name);
 412                 }
 413         }
 414         RETURN(0);
 415 }
 416
 417 static void lmv_del_target(struct lmv_obd *lmv, int index)
 418 {
 419         if (lmv->tgts[index] == NULL)
 420                 return;
 421
 422         OBD_FREE_PTR(lmv->tgts[index]);
 423         lmv->tgts[index] = NULL;
 424         return;
 425 }
 426
 427 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 428                            __u32 index, int gen)
 429 {
 430         struct lmv_obd      *lmv = &obd->u.lmv;
 431         struct lmv_tgt_desc *tgt;
 432         int                  orig_tgt_count = 0;
 433         int                  rc = 0;
 434         ENTRY;
 435
 436         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 437
 438         lmv_init_lock(lmv);
 439
 440         if (lmv->desc.ld_tgt_count == 0) {
 441                 struct obd_device *mdc_obd;
 442
 443                 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 444                                                 &obd->obd_uuid);
 445                 if (!mdc_obd) {
 446                         lmv_init_unlock(lmv);
 447                         CERROR("%s: Target %s not attached: rc = %d\n",
 448                                obd->obd_name, uuidp->uuid, -EINVAL);
 449                         RETURN(-EINVAL);
 450                 }
 451         }
 452
 453         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
 454                 tgt = lmv->tgts[index];
 455                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
 456                        " rc = %d\n", obd->obd_name,
 457                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
 458                 lmv_init_unlock(lmv);
 459                 RETURN(-EEXIST);
 460         }
 461
 462         if (index >= lmv->tgts_size) {
 463                 /* We need to reallocate the lmv target array. */
 464                 struct lmv_tgt_desc **newtgts, **old = NULL;
 465                 __u32 newsize = 1;
 466                 __u32 oldsize = 0;
 467
 468                 while (newsize < index + 1)
 469                         newsize = newsize << 1;
 470                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 471                 if (newtgts == NULL) {
 472                         lmv_init_unlock(lmv);
 473                         RETURN(-ENOMEM);
 474                 }
 475
 476                 if (lmv->tgts_size) {
 477                         memcpy(newtgts, lmv->tgts,
 478                                sizeof(*newtgts) * lmv->tgts_size);
 479                         old = lmv->tgts;
 480                         oldsize = lmv->tgts_size;
 481                 }
 482
 483                 lmv->tgts = newtgts;
 484                 lmv->tgts_size = newsize;
 485                 smp_rmb();
 486                 if (old)
 487                         OBD_FREE(old, sizeof(*old) * oldsize);
 488
 489                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
 490                        lmv->tgts_size);
 491         }
 492
 493         OBD_ALLOC_PTR(tgt);
 494         if (!tgt) {
 495                 lmv_init_unlock(lmv);
 496                 RETURN(-ENOMEM);
 497         }
 498
 499         mutex_init(&tgt->ltd_fid_mutex);
 500         tgt->ltd_idx = index;
 501         tgt->ltd_uuid = *uuidp;
 502         tgt->ltd_active = 0;
 503         lmv->tgts[index] = tgt;
 504         if (index >= lmv->desc.ld_tgt_count) {
 505                 orig_tgt_count = lmv->desc.ld_tgt_count;
 506                 lmv->desc.ld_tgt_count = index + 1;
 507         }
 508
 509         if (lmv->connected) {
 510                 rc = lmv_connect_mdc(obd, tgt);
 511                 if (rc != 0) {
 512                         spin_lock(&lmv->lmv_lock);
 513                         if (lmv->desc.ld_tgt_count == index + 1)
 514                                 lmv->desc.ld_tgt_count = orig_tgt_count;
 515                         memset(tgt, 0, sizeof(*tgt));
 516                         spin_unlock(&lmv->lmv_lock);
 517                 } else {
 518                         int easize = sizeof(struct lmv_stripe_md) +
 519                                 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
 520                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
 521                 }
 522         }
 523
 524         lmv_init_unlock(lmv);
 525         RETURN(rc);
 526 }
 527
 528 int lmv_check_connect(struct obd_device *obd)
 529 {
 530         struct lmv_obd          *lmv = &obd->u.lmv;
 531         struct lmv_tgt_desc     *tgt;
 532         __u32                    i;
 533         int                      rc;
 534         int                      easize;
 535         ENTRY;
 536
 537         if (lmv->connected)
 538                 RETURN(0);
 539
 540         lmv_init_lock(lmv);
 541         if (lmv->connected) {
 542                 lmv_init_unlock(lmv);
 543                 RETURN(0);
 544         }
 545
 546         if (lmv->desc.ld_tgt_count == 0) {
 547                 lmv_init_unlock(lmv);
 548                 CERROR("%s: no targets configured.\n", obd->obd_name);
 549                 RETURN(-EINVAL);
 550         }
 551
 552         LASSERT(lmv->tgts != NULL);
 553
 554         if (lmv->tgts[0] == NULL) {
 555                 lmv_init_unlock(lmv);
 556                 CERROR("%s: no target configured for index 0.\n",
 557                        obd->obd_name);
 558                 RETURN(-EINVAL);
 559         }
 560
 561         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 562                lmv->cluuid.uuid, obd->obd_name);
 563
 564         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 565                 tgt = lmv->tgts[i];
 566                 if (tgt == NULL)
 567                         continue;
 568                 rc = lmv_connect_mdc(obd, tgt);
 569                 if (rc)
 570                         GOTO(out_disc, rc);
 571         }
 572
 573         class_export_put(lmv->exp);
 574         lmv->connected = 1;
 575         easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
 576         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
 577         lmv_init_unlock(lmv);
 578         RETURN(0);
 579
 580  out_disc:
 581         while (i-- > 0) {
 582                 int rc2;
 583                 tgt = lmv->tgts[i];
 584                 if (tgt == NULL)
 585                         continue;
 586                 tgt->ltd_active = 0;
 587                 if (tgt->ltd_exp) {
 588                         --lmv->desc.ld_active_tgt_count;
 589                         rc2 = obd_disconnect(tgt->ltd_exp);
 590                         if (rc2) {
 591                                 CERROR("LMV target %s disconnect on "
 592                                        "MDC idx %d: error %d\n",
 593                                        tgt->ltd_uuid.uuid, i, rc2);
 594                         }
 595                 }
 596         }
 597         class_disconnect(lmv->exp);
 598         lmv_init_unlock(lmv);
 599         RETURN(rc);
 600 }
 601
 602 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 603 {
 604         struct lmv_obd         *lmv = &obd->u.lmv;
 605         struct obd_device      *mdc_obd;
 606         int                     rc;
 607         ENTRY;
 608
 609         LASSERT(tgt != NULL);
 610         LASSERT(obd != NULL);
 611
 612         mdc_obd = class_exp2obd(tgt->ltd_exp);
 613
 614         if (mdc_obd) {
 615                 mdc_obd->obd_force = obd->obd_force;
 616                 mdc_obd->obd_fail = obd->obd_fail;
 617                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 618         }
 619
 620         if (lmv->targets_proc_entry != NULL)
 621                 lprocfs_remove_proc_entry(mdc_obd->obd_name,
 622                                           lmv->targets_proc_entry);
 623
 624         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 625         if (rc)
 626                 CERROR("Can't finanize fids factory\n");
 627
 628         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 629                tgt->ltd_exp->exp_obd->obd_name,
 630                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 631
 632         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 633         rc = obd_disconnect(tgt->ltd_exp);
 634         if (rc) {
 635                 if (tgt->ltd_active) {
 636                         CERROR("Target %s disconnect error %d\n",
 637                                tgt->ltd_uuid.uuid, rc);
 638                 }
 639         }
 640
 641         lmv_activate_target(lmv, tgt, 0);
 642         tgt->ltd_exp = NULL;
 643         RETURN(0);
 644 }
 645
 646 static int lmv_disconnect(struct obd_export *exp)
 647 {
 648         struct obd_device       *obd = class_exp2obd(exp);
 649         struct lmv_obd          *lmv = &obd->u.lmv;
 650         int                      rc;
 651         __u32                    i;
 652         ENTRY;
 653
 654         if (!lmv->tgts)
 655                 goto out_local;
 656
 657         /*
 658          * Only disconnect the underlying layers on the final disconnect.
 659          */
 660         lmv->refcount--;
 661         if (lmv->refcount != 0)
 662                 goto out_local;
 663
 664         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 665                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
 666                         continue;
 667
 668                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
 669         }
 670
 671         if (lmv->targets_proc_entry != NULL)
 672                 lprocfs_remove(&lmv->targets_proc_entry);
 673         else
 674                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 675                        obd->obd_type->typ_name, obd->obd_name);
 676
 677 out_local:
 678         /*
 679          * This is the case when no real connection is established by
 680          * lmv_check_connect().
 681          */
 682         if (!lmv->connected)
 683                 class_export_put(exp);
 684         rc = class_disconnect(exp);
 685         if (lmv->refcount == 0)
 686                 lmv->connected = 0;
 687         RETURN(rc);
 688 }
 689
 690 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
 691 {
 692         struct obd_device       *obddev = class_exp2obd(exp);
 693         struct lmv_obd          *lmv = &obddev->u.lmv;
 694         struct getinfo_fid2path *gf;
 695         struct lmv_tgt_desc     *tgt;
 696         struct getinfo_fid2path *remote_gf = NULL;
 697         int                     remote_gf_size = 0;
 698         int                     rc;
 699
 700         gf = (struct getinfo_fid2path *)karg;
 701         tgt = lmv_find_target(lmv, &gf->gf_fid);
 702         if (IS_ERR(tgt))
 703                 RETURN(PTR_ERR(tgt));
 704
 705 repeat_fid2path:
 706         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 707         if (rc != 0 && rc != -EREMOTE)
 708                 GOTO(out_fid2path, rc);
 709
 710         /* If remote_gf != NULL, it means just building the
 711          * path on the remote MDT, copy this path segement to gf */
 712         if (remote_gf != NULL) {
 713                 struct getinfo_fid2path *ori_gf;
 714                 char *ptr;
 715
 716                 ori_gf = (struct getinfo_fid2path *)karg;
 717                 if (strlen(ori_gf->gf_path) +
 718                     strlen(gf->gf_path) > ori_gf->gf_pathlen)
 719                         GOTO(out_fid2path, rc = -EOVERFLOW);
 720
 721                 ptr = ori_gf->gf_path;
 722
 723                 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
 724                         strlen(ori_gf->gf_path));
 725
 726                 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
 727                 ptr += strlen(gf->gf_path);
 728                 *ptr = '/';
 729         }
 730
 731         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
 732                tgt->ltd_exp->exp_obd->obd_name,
 733                gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 734                gf->gf_linkno);
 735
 736         if (rc == 0)
 737                 GOTO(out_fid2path, rc);
 738
 739         /* sigh, has to go to another MDT to do path building further */
 740         if (remote_gf == NULL) {
 741                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 742                 OBD_ALLOC(remote_gf, remote_gf_size);
 743                 if (remote_gf == NULL)
 744                         GOTO(out_fid2path, rc = -ENOMEM);
 745                 remote_gf->gf_pathlen = PATH_MAX;
 746         }
 747
 748         if (!fid_is_sane(&gf->gf_fid)) {
 749                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 750                        tgt->ltd_exp->exp_obd->obd_name,
 751                        PFID(&gf->gf_fid), -EINVAL);
 752                 GOTO(out_fid2path, rc = -EINVAL);
 753         }
 754
 755         tgt = lmv_find_target(lmv, &gf->gf_fid);
 756         if (IS_ERR(tgt))
 757                 GOTO(out_fid2path, rc = -EINVAL);
 758
 759         remote_gf->gf_fid = gf->gf_fid;
 760         remote_gf->gf_recno = -1;
 761         remote_gf->gf_linkno = -1;
 762         memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
 763         gf = remote_gf;
 764         goto repeat_fid2path;
 765
 766 out_fid2path:
 767         if (remote_gf != NULL)
 768                 OBD_FREE(remote_gf, remote_gf_size);
 769         RETURN(rc);
 770 }
 771
 772 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 773                              const struct hsm_user_request *hur,
 774                              const struct lmv_tgt_desc *tgt_mds)
 775 {
 776         __u32                    i;
 777         int                      nr = 0;
 778         struct lmv_tgt_desc     *curr_tgt;
 779
 780         /* count how many requests must be sent to the given target */
 781         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 782                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
 783                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 784                         nr++;
 785         }
 786         return nr;
 787 }
 788
 789 static void lmv_hsm_req_build(struct lmv_obd *lmv,
 790                               struct hsm_user_request *hur_in,
 791                               const struct lmv_tgt_desc *tgt_mds,
 792                               struct hsm_user_request *hur_out)
 793 {
 794         __u32                    i, nr_out;
 795         struct lmv_tgt_desc     *curr_tgt;
 796
 797         /* build the hsm_user_request for the given target */
 798         hur_out->hur_request = hur_in->hur_request;
 799         nr_out = 0;
 800         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 801                 curr_tgt = lmv_find_target(lmv,
 802                                            &hur_in->hur_user_item[i].hui_fid);
 803                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 804                         hur_out->hur_user_item[nr_out] =
 805                                                 hur_in->hur_user_item[i];
 806                         nr_out++;
 807                 }
 808         }
 809         hur_out->hur_request.hr_itemcount = nr_out;
 810         memcpy(hur_data(hur_out), hur_data(hur_in),
 811                hur_in->hur_request.hr_data_len);
 812 }
 813
 814 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
 815                                  struct lustre_kernelcomm *lk, void *uarg)
 816 {
 817         __u32                    i;
 818         int                      rc;
 819         struct kkuc_ct_data     *kcd = NULL;
 820         ENTRY;
 821
 822         /* unregister request (call from llapi_hsm_copytool_fini) */
 823         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 824                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 825
 826                 if (tgt == NULL || tgt->ltd_exp == NULL)
 827                         continue;
 828                 /* best effort: try to clean as much as possible
 829                  * (continue on error) */
 830                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 831         }
 832
 833         /* Whatever the result, remove copytool from kuc groups.
 834          * Unreached coordinators will get EPIPE on next requests
 835          * and will unregister automatically.
 836          */
 837         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
 838         if (kcd != NULL)
 839                 OBD_FREE_PTR(kcd);
 840
 841         RETURN(rc);
 842 }
 843
 844 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
 845                                struct lustre_kernelcomm *lk, void *uarg)
 846 {
 847         struct file             *filp;
 848         __u32                    i, j;
 849         int                      err, rc;
 850         bool                     any_set = false;
 851         struct kkuc_ct_data     *kcd;
 852         ENTRY;
 853
 854         /* All or nothing: try to register to all MDS.
 855          * In case of failure, unregister from previous MDS,
 856          * except if it because of inactive target. */
 857         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 858                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 859
 860                 if (tgt == NULL || tgt->ltd_exp == NULL)
 861                         continue;
 862                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 863                 if (err) {
 864                         if (tgt->ltd_active) {
 865                                 /* permanent error */
 866                                 CERROR("%s: iocontrol MDC %s on MDT"
 867                                        " idx %d cmd %x: err = %d\n",
 868                                        class_exp2obd(lmv->exp)->obd_name,
 869                                        tgt->ltd_uuid.uuid, i, cmd, err);
 870                                 rc = err;
 871                                 lk->lk_flags |= LK_FLG_STOP;
 872                                 /* unregister from previous MDS */
 873                                 for (j = 0; j < i; j++) {
 874                                         tgt = lmv->tgts[j];
 875                                         if (tgt == NULL || tgt->ltd_exp == NULL)
 876                                                 continue;
 877                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
 878                                                       lk, uarg);
 879                                 }
 880                                 RETURN(rc);
 881                         }
 882                         /* else: transient error.
 883                          * kuc will register to the missing MDT
 884                          * when it is back */
 885                 } else {
 886                         any_set = true;
 887                 }
 888         }
 889
 890         if (!any_set)
 891                 /* no registration done: return error */
 892                 RETURN(-ENOTCONN);
 893
 894         /* at least one registration done, with no failure */
 895         filp = fget(lk->lk_wfd);
 896         if (filp == NULL)
 897                 RETURN(-EBADF);
 898
 899         OBD_ALLOC_PTR(kcd);
 900         if (kcd == NULL) {
 901                 fput(filp);
 902                 RETURN(-ENOMEM);
 903         }
 904         kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
 905         kcd->kcd_uuid = lmv->cluuid;
 906         kcd->kcd_archive = lk->lk_data;
 907
 908         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
 909         if (rc != 0) {
 910                 if (filp != NULL)
 911                         fput(filp);
 912                 OBD_FREE_PTR(kcd);
 913         }
 914
 915         RETURN(rc);
 916 }
 917
 918
 919
 920
 921 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 922                          int len, void *karg, void *uarg)
 923 {
 924         struct obd_device       *obddev = class_exp2obd(exp);
 925         struct lmv_obd          *lmv = &obddev->u.lmv;
 926         struct lmv_tgt_desc     *tgt = NULL;
 927         __u32                    i = 0;
 928         int                      rc = 0;
 929         int                      set = 0;
 930         __u32                    count = lmv->desc.ld_tgt_count;
 931         ENTRY;
 932
 933         if (count == 0)
 934                 RETURN(-ENOTTY);
 935
 936         switch (cmd) {
 937         case IOC_OBD_STATFS: {
 938                 struct obd_ioctl_data *data = karg;
 939                 struct obd_device *mdc_obd;
 940                 struct obd_statfs stat_buf = {0};
 941                 __u32 index;
 942
 943                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 944                 if ((index >= count))
 945                         RETURN(-ENODEV);
 946
 947                 tgt = lmv->tgts[index];
 948                 if (tgt == NULL || !tgt->ltd_active)
 949                         RETURN(-ENODATA);
 950
 951                 mdc_obd = class_exp2obd(tgt->ltd_exp);
 952                 if (!mdc_obd)
 953                         RETURN(-EINVAL);
 954
 955                 /* copy UUID */
 956                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 957                                  min((int) data->ioc_plen2,
 958                                      (int) sizeof(struct obd_uuid))))
 959                         RETURN(-EFAULT);
 960
 961                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
 962                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 963                                 0);
 964                 if (rc)
 965                         RETURN(rc);
 966                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
 967                                  min((int) data->ioc_plen1,
 968                                      (int) sizeof(stat_buf))))
 969                         RETURN(-EFAULT);
 970                 break;
 971         }
 972         case OBD_IOC_QUOTACTL: {
 973                 struct if_quotactl *qctl = karg;
 974                 struct obd_quotactl *oqctl;
 975
 976                 if (qctl->qc_valid == QC_MDTIDX) {
 977                         if (count <= qctl->qc_idx)
 978                                 RETURN(-EINVAL);
 979
 980                         tgt = lmv->tgts[qctl->qc_idx];
 981                         if (tgt == NULL || tgt->ltd_exp == NULL)
 982                                 RETURN(-EINVAL);
 983                 } else if (qctl->qc_valid == QC_UUID) {
 984                         for (i = 0; i < count; i++) {
 985                                 tgt = lmv->tgts[i];
 986                                 if (tgt == NULL)
 987                                         continue;
 988                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 989                                                      &qctl->obd_uuid))
 990                                         continue;
 991
 992                                 if (tgt->ltd_exp == NULL)
 993                                         RETURN(-EINVAL);
 994
 995                                 break;
 996                         }
 997                 } else {
 998                         RETURN(-EINVAL);
 999                 }
1000
1001                 if (i >= count)
1002                         RETURN(-EAGAIN);
1003
1004                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1005                 OBD_ALLOC_PTR(oqctl);
1006                 if (!oqctl)
1007                         RETURN(-ENOMEM);
1008
1009                 QCTL_COPY(oqctl, qctl);
1010                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1011                 if (rc == 0) {
1012                         QCTL_COPY(qctl, oqctl);
1013                         qctl->qc_valid = QC_MDTIDX;
1014                         qctl->obd_uuid = tgt->ltd_uuid;
1015                 }
1016                 OBD_FREE_PTR(oqctl);
1017                 break;
1018         }
1019         case OBD_IOC_CHANGELOG_SEND:
1020         case OBD_IOC_CHANGELOG_CLEAR: {
1021                 struct ioc_changelog *icc = karg;
1022
1023                 if (icc->icc_mdtindex >= count)
1024                         RETURN(-ENODEV);
1025
1026                 tgt = lmv->tgts[icc->icc_mdtindex];
1027                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1028                         RETURN(-ENODEV);
1029                 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1030                 break;
1031         }
1032         case LL_IOC_GET_CONNECT_FLAGS: {
1033                 tgt = lmv->tgts[0];
1034                 if (tgt == NULL || tgt->ltd_exp == NULL)
1035                         RETURN(-ENODATA);
1036                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1037                 break;
1038         }
1039         case LL_IOC_FID2MDTIDX: {
1040                 struct lu_fid *fid = karg;
1041                 int             mdt_index;
1042
1043                 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1044                 if (rc != 0)
1045                         RETURN(rc);
1046
1047                 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1048                  * point to user space memory for FID2MDTIDX. */
1049                 *(__u32 *)uarg = mdt_index;
1050                 break;
1051         }
1052         case OBD_IOC_FID2PATH: {
1053                 rc = lmv_fid2path(exp, len, karg, uarg);
1054                 break;
1055         }
1056         case LL_IOC_HSM_STATE_GET:
1057         case LL_IOC_HSM_STATE_SET:
1058         case LL_IOC_HSM_ACTION: {
1059                 struct md_op_data       *op_data = karg;
1060
1061                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1062                 if (IS_ERR(tgt))
1063                         RETURN(PTR_ERR(tgt));
1064
1065                 if (tgt->ltd_exp == NULL)
1066                         RETURN(-EINVAL);
1067
1068                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1069                 break;
1070         }
1071         case LL_IOC_HSM_PROGRESS: {
1072                 const struct hsm_progress_kernel *hpk = karg;
1073
1074                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1075                 if (IS_ERR(tgt))
1076                         RETURN(PTR_ERR(tgt));
1077                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1078                 break;
1079         }
1080         case LL_IOC_HSM_REQUEST: {
1081                 struct hsm_user_request *hur = karg;
1082                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1083
1084                 if (reqcount == 0)
1085                         RETURN(0);
1086
1087                 /* if the request is about a single fid
1088                  * or if there is a single MDS, no need to split
1089                  * the request. */
1090                 if (reqcount == 1 || count == 1) {
1091                         tgt = lmv_find_target(lmv,
1092                                               &hur->hur_user_item[0].hui_fid);
1093                         if (IS_ERR(tgt))
1094                                 RETURN(PTR_ERR(tgt));
1095                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1096                 } else {
1097                         /* split fid list to their respective MDS */
1098                         for (i = 0; i < count; i++) {
1099                                 unsigned int            nr, reqlen;
1100                                 int                     rc1;
1101                                 struct hsm_user_request *req;
1102
1103                                 tgt = lmv->tgts[i];
1104                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1105                                         continue;
1106
1107                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1108                                 if (nr == 0) /* nothing for this MDS */
1109                                         continue;
1110
1111                                 /* build a request with fids for this MDS */
1112                                 reqlen = offsetof(typeof(*hur),
1113                                                   hur_user_item[nr])
1114                                                 + hur->hur_request.hr_data_len;
1115                                 OBD_ALLOC_LARGE(req, reqlen);
1116                                 if (req == NULL)
1117                                         RETURN(-ENOMEM);
1118
1119                                 lmv_hsm_req_build(lmv, hur, tgt, req);
1120
1121                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1122                                                     req, uarg);
1123                                 if (rc1 != 0 && rc == 0)
1124                                         rc = rc1;
1125                                 OBD_FREE_LARGE(req, reqlen);
1126                         }
1127                 }
1128                 break;
1129         }
1130         case LL_IOC_LOV_SWAP_LAYOUTS: {
1131                 struct md_op_data       *op_data = karg;
1132                 struct lmv_tgt_desc     *tgt1, *tgt2;
1133
1134                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1135                 if (IS_ERR(tgt1))
1136                         RETURN(PTR_ERR(tgt1));
1137
1138                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1139                 if (IS_ERR(tgt2))
1140                         RETURN(PTR_ERR(tgt2));
1141
1142                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1143                         RETURN(-EINVAL);
1144
1145                 /* only files on same MDT can have their layouts swapped */
1146                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1147                         RETURN(-EPERM);
1148
1149                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1150                 break;
1151         }
1152         case LL_IOC_HSM_CT_START: {
1153                 struct lustre_kernelcomm *lk = karg;
1154                 if (lk->lk_flags & LK_FLG_STOP)
1155                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1156                 else
1157                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1158                 break;
1159         }
1160         default:
1161                 for (i = 0; i < count; i++) {
1162                         struct obd_device *mdc_obd;
1163                         int err;
1164
1165                         tgt = lmv->tgts[i];
1166                         if (tgt == NULL || tgt->ltd_exp == NULL)
1167                                 continue;
1168                         /* ll_umount_begin() sets force flag but for lmv, not
1169                          * mdc. Let's pass it through */
1170                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1171                         mdc_obd->obd_force = obddev->obd_force;
1172                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1173                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1174                                 RETURN(err);
1175                         } else if (err) {
1176                                 if (tgt->ltd_active) {
1177                                         CERROR("error: iocontrol MDC %s on MDT"
1178                                                " idx %d cmd %x: err = %d\n",
1179                                                tgt->ltd_uuid.uuid, i, cmd, err);
1180                                         if (!rc)
1181                                                 rc = err;
1182                                 }
1183                         } else
1184                                 set = 1;
1185                 }
1186                 if (!set && !rc)
1187                         rc = -EIO;
1188         }
1189         RETURN(rc);
1190 }
1191
1192 #if 0
1193 static int lmv_all_chars_policy(int count, const char *name,
1194                                 int len)
1195 {
1196         unsigned int c = 0;
1197
1198         while (len > 0)
1199                 c += name[--len];
1200         c = c % count;
1201         return c;
1202 }
1203
1204 static int lmv_nid_policy(struct lmv_obd *lmv)
1205 {
1206         struct obd_import *imp;
1207         __u32              id;
1208
1209         /*
1210          * XXX: To get nid we assume that underlying obd device is mdc.
1211          */
1212         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1213         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1214         return id % lmv->desc.ld_tgt_count;
1215 }
1216
1217 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1218                           placement_policy_t placement)
1219 {
1220         switch (placement) {
1221         case PLACEMENT_CHAR_POLICY:
1222                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1223                                             op_data->op_name,
1224                                             op_data->op_namelen);
1225         case PLACEMENT_NID_POLICY:
1226                 return lmv_nid_policy(lmv);
1227
1228         default:
1229                 break;
1230         }
1231
1232         CERROR("Unsupported placement policy %x\n", placement);
1233         return -EINVAL;
1234 }
1235 #endif
1236
1237 /**
1238  * This is _inode_ placement policy function (not name).
1239  */
1240 static int lmv_placement_policy(struct obd_device *obd,
1241                                 struct md_op_data *op_data, u32 *mds)
1242 {
1243         struct lmv_obd          *lmv = &obd->u.lmv;
1244         ENTRY;
1245
1246         LASSERT(mds != NULL);
1247
1248         if (lmv->desc.ld_tgt_count == 1) {
1249                 *mds = 0;
1250                 RETURN(0);
1251         }
1252
1253         /**
1254          * If stripe_offset is provided during setdirstripe
1255          * (setdirstripe -i xx), xx MDS will be choosen.
1256          */
1257         if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1258                 struct lmv_user_md *lum;
1259
1260                 lum = op_data->op_data;
1261
1262                 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1263                         *mds = le32_to_cpu(lum->lum_stripe_offset);
1264                 } else {
1265                         /* -1 means default, which will be in the same MDT with
1266                          * the stripe */
1267                         *mds = op_data->op_mds;
1268                         lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1269                 }
1270         } else {
1271                 /* Allocate new fid on target according to operation type and
1272                  * parent home mds. */
1273                 *mds = op_data->op_mds;
1274         }
1275
1276         RETURN(0);
1277 }
1278
1279 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1280 {
1281         struct lmv_tgt_desc     *tgt;
1282         int                      rc;
1283         ENTRY;
1284
1285         tgt = lmv_get_target(lmv, mds, NULL);
1286         if (IS_ERR(tgt))
1287                 RETURN(PTR_ERR(tgt));
1288
1289         /*
1290          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1291          * on server that seq in new allocated fid is not yet known.
1292          */
1293         mutex_lock(&tgt->ltd_fid_mutex);
1294
1295         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1296                 GOTO(out, rc = -ENODEV);
1297
1298         /*
1299          * Asking underlying tgt layer to allocate new fid.
1300          */
1301         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1302         if (rc > 0) {
1303                 LASSERT(fid_is_sane(fid));
1304                 rc = 0;
1305         }
1306
1307         EXIT;
1308 out:
1309         mutex_unlock(&tgt->ltd_fid_mutex);
1310         return rc;
1311 }
1312
1313 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1314                   struct lu_fid *fid, struct md_op_data *op_data)
1315 {
1316         struct obd_device     *obd = class_exp2obd(exp);
1317         struct lmv_obd        *lmv = &obd->u.lmv;
1318         u32                    mds = 0;
1319         int                    rc;
1320         ENTRY;
1321
1322         LASSERT(op_data != NULL);
1323         LASSERT(fid != NULL);
1324
1325         rc = lmv_placement_policy(obd, op_data, &mds);
1326         if (rc) {
1327                 CERROR("Can't get target for allocating fid, "
1328                        "rc %d\n", rc);
1329                 RETURN(rc);
1330         }
1331
1332         rc = __lmv_fid_alloc(lmv, fid, mds);
1333         if (rc) {
1334                 CERROR("Can't alloc new fid, rc %d\n", rc);
1335                 RETURN(rc);
1336         }
1337
1338         RETURN(rc);
1339 }
1340
1341 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1342 {
1343         struct lmv_obd  *lmv = &obd->u.lmv;
1344         struct lmv_desc *desc;
1345         int             rc;
1346         ENTRY;
1347
1348         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1349                 CERROR("LMV setup requires a descriptor\n");
1350                 RETURN(-EINVAL);
1351         }
1352
1353         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1354         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1355                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1356                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1357                 RETURN(-EINVAL);
1358         }
1359
1360         lmv->tgts_size = 32U;
1361         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1362         if (lmv->tgts == NULL)
1363                 RETURN(-ENOMEM);
1364
1365         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1366         lmv->desc.ld_tgt_count = 0;
1367         lmv->desc.ld_active_tgt_count = 0;
1368         lmv->max_cookiesize = 0;
1369         lmv->max_def_easize = 0;
1370         lmv->max_easize = 0;
1371         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1372
1373         spin_lock_init(&lmv->lmv_lock);
1374         mutex_init(&lmv->init_mutex);
1375
1376 #ifdef LPROCFS
1377         obd->obd_vars = lprocfs_lmv_obd_vars;
1378         lprocfs_obd_setup(obd);
1379         lprocfs_alloc_md_stats(obd, 0);
1380         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1381                                 0444, &lmv_proc_target_fops, obd);
1382         if (rc)
1383                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1384                       obd->obd_name, rc);
1385 #endif
1386         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1387                              LUSTRE_CLI_FLD_HASH_DHT);
1388         if (rc) {
1389                 CERROR("Can't init FLD, err %d\n", rc);
1390                 GOTO(out, rc);
1391         }
1392
1393         RETURN(0);
1394
1395 out:
1396         return rc;
1397 }
1398
1399 static int lmv_cleanup(struct obd_device *obd)
1400 {
1401         struct lmv_obd   *lmv = &obd->u.lmv;
1402         ENTRY;
1403
1404         fld_client_fini(&lmv->lmv_fld);
1405         if (lmv->tgts != NULL) {
1406                 int i;
1407                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1408                         if (lmv->tgts[i] == NULL)
1409                                 continue;
1410                         lmv_del_target(lmv, i);
1411                 }
1412                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1413                 lmv->tgts_size = 0;
1414         }
1415         RETURN(0);
1416 }
1417
1418 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1419 {
1420         struct lustre_cfg       *lcfg = buf;
1421         struct obd_uuid         obd_uuid;
1422         int                     gen;
1423         __u32                   index;
1424         int                     rc;
1425         ENTRY;
1426
1427         switch (lcfg->lcfg_command) {
1428         case LCFG_ADD_MDC:
1429                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1430                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1431                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1432                         GOTO(out, rc = -EINVAL);
1433
1434                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1435
1436                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1437                         GOTO(out, rc = -EINVAL);
1438                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1439                         GOTO(out, rc = -EINVAL);
1440                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1441                 GOTO(out, rc);
1442         default:
1443                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1444                 GOTO(out, rc = -EINVAL);
1445         }
1446 out:
1447         RETURN(rc);
1448 }
1449
1450 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1451                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1452 {
1453         struct obd_device       *obd = class_exp2obd(exp);
1454         struct lmv_obd          *lmv = &obd->u.lmv;
1455         struct obd_statfs       *temp;
1456         int                      rc = 0;
1457         __u32                    i;
1458         ENTRY;
1459
1460         rc = lmv_check_connect(obd);
1461         if (rc)
1462                 RETURN(rc);
1463
1464         OBD_ALLOC(temp, sizeof(*temp));
1465         if (temp == NULL)
1466                 RETURN(-ENOMEM);
1467
1468         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1469                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1470                         continue;
1471
1472                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1473                                 max_age, flags);
1474                 if (rc) {
1475                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1476                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1477                                rc);
1478                         GOTO(out_free_temp, rc);
1479                 }
1480
1481                 if (i == 0) {
1482                         *osfs = *temp;
1483                         /* If the statfs is from mount, it will needs
1484                          * retrieve necessary information from MDT0.
1485                          * i.e. mount does not need the merged osfs
1486                          * from all of MDT.
1487                          * And also clients can be mounted as long as
1488                          * MDT0 is in service*/
1489                         if (flags & OBD_STATFS_FOR_MDT0)
1490                                 GOTO(out_free_temp, rc);
1491                 } else {
1492                         osfs->os_bavail += temp->os_bavail;
1493                         osfs->os_blocks += temp->os_blocks;
1494                         osfs->os_ffree += temp->os_ffree;
1495                         osfs->os_files += temp->os_files;
1496                 }
1497         }
1498
1499         EXIT;
1500 out_free_temp:
1501         OBD_FREE(temp, sizeof(*temp));
1502         return rc;
1503 }
1504
1505 static int lmv_getstatus(struct obd_export *exp,
1506                          struct lu_fid *fid,
1507                          struct obd_capa **pc)
1508 {
1509         struct obd_device    *obd = exp->exp_obd;
1510         struct lmv_obd       *lmv = &obd->u.lmv;
1511         int                   rc;
1512         ENTRY;
1513
1514         rc = lmv_check_connect(obd);
1515         if (rc)
1516                 RETURN(rc);
1517
1518         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1519         RETURN(rc);
1520 }
1521
1522 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1523                         struct obd_capa *oc, obd_valid valid, const char *name,
1524                         const char *input, int input_size, int output_size,
1525                         int flags, struct ptlrpc_request **request)
1526 {
1527         struct obd_device      *obd = exp->exp_obd;
1528         struct lmv_obd         *lmv = &obd->u.lmv;
1529         struct lmv_tgt_desc    *tgt;
1530         int                     rc;
1531         ENTRY;
1532
1533         rc = lmv_check_connect(obd);
1534         if (rc)
1535                 RETURN(rc);
1536
1537         tgt = lmv_find_target(lmv, fid);
1538         if (IS_ERR(tgt))
1539                 RETURN(PTR_ERR(tgt));
1540
1541         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1542                          input_size, output_size, flags, request);
1543
1544         RETURN(rc);
1545 }
1546
1547 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1548                         struct obd_capa *oc, obd_valid valid, const char *name,
1549                         const char *input, int input_size, int output_size,
1550                         int flags, __u32 suppgid,
1551                         struct ptlrpc_request **request)
1552 {
1553         struct obd_device      *obd = exp->exp_obd;
1554         struct lmv_obd         *lmv = &obd->u.lmv;
1555         struct lmv_tgt_desc    *tgt;
1556         int                     rc;
1557         ENTRY;
1558
1559         rc = lmv_check_connect(obd);
1560         if (rc)
1561                 RETURN(rc);
1562
1563         tgt = lmv_find_target(lmv, fid);
1564         if (IS_ERR(tgt))
1565                 RETURN(PTR_ERR(tgt));
1566
1567         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1568                          input_size, output_size, flags, suppgid,
1569                          request);
1570
1571         RETURN(rc);
1572 }
1573
1574 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1575                        struct ptlrpc_request **request)
1576 {
1577         struct obd_device       *obd = exp->exp_obd;
1578         struct lmv_obd          *lmv = &obd->u.lmv;
1579         struct lmv_tgt_desc     *tgt;
1580         int                      rc;
1581         ENTRY;
1582
1583         rc = lmv_check_connect(obd);
1584         if (rc)
1585                 RETURN(rc);
1586
1587         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1588         if (IS_ERR(tgt))
1589                 RETURN(PTR_ERR(tgt));
1590
1591         if (op_data->op_flags & MF_GET_MDT_IDX) {
1592                 op_data->op_mds = tgt->ltd_idx;
1593                 RETURN(0);
1594         }
1595
1596         rc = md_getattr(tgt->ltd_exp, op_data, request);
1597
1598         RETURN(rc);
1599 }
1600
1601 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1602 {
1603         struct obd_device   *obd = exp->exp_obd;
1604         struct lmv_obd      *lmv = &obd->u.lmv;
1605         __u32                i;
1606         int                  rc;
1607         ENTRY;
1608
1609         rc = lmv_check_connect(obd);
1610         if (rc)
1611                 RETURN(rc);
1612
1613         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1614
1615         /*
1616          * With DNE every object can have two locks in different namespaces:
1617          * lookup lock in space of MDT storing direntry and update/open lock in
1618          * space of MDT storing inode.
1619          */
1620         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1621                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1622                         continue;
1623                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1624         }
1625
1626         RETURN(0);
1627 }
1628
1629 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1630                            ldlm_iterator_t it, void *data)
1631 {
1632         struct obd_device       *obd = exp->exp_obd;
1633         struct lmv_obd          *lmv = &obd->u.lmv;
1634         int                     i;
1635         int                     tgt;
1636         int                     rc;
1637         ENTRY;
1638
1639         rc = lmv_check_connect(obd);
1640         if (rc)
1641                 RETURN(rc);
1642
1643         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1644
1645         /*
1646          * With DNE every object can have two locks in different namespaces:
1647          * lookup lock in space of MDT storing direntry and update/open lock in
1648          * space of MDT storing inode.  Try the MDT that the FID maps to first,
1649          * since this can be easily found, and only try others if that fails.
1650          */
1651         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1652              i < lmv->desc.ld_tgt_count;
1653              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1654                 if (tgt < 0) {
1655                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1656                                obd->obd_name, PFID(fid), tgt);
1657                         tgt = 0;
1658                 }
1659
1660                 if (lmv->tgts[tgt] == NULL ||
1661                     lmv->tgts[tgt]->ltd_exp == NULL)
1662                         continue;
1663
1664                 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1665                 if (rc)
1666                         RETURN(rc);
1667         }
1668
1669         RETURN(rc);
1670 }
1671
1672
1673 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1674                      struct md_open_data *mod, struct ptlrpc_request **request)
1675 {
1676         struct obd_device     *obd = exp->exp_obd;
1677         struct lmv_obd        *lmv = &obd->u.lmv;
1678         struct lmv_tgt_desc   *tgt;
1679         int                    rc;
1680         ENTRY;
1681
1682         rc = lmv_check_connect(obd);
1683         if (rc)
1684                 RETURN(rc);
1685
1686         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1687         if (IS_ERR(tgt))
1688                 RETURN(PTR_ERR(tgt));
1689
1690         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1691         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1692         RETURN(rc);
1693 }
1694
1695 /**
1696  * Choosing the MDT by name or FID in @op_data.
1697  * For non-striped directory, it will locate MDT by fid.
1698  * For striped-directory, it will locate MDT by name. And also
1699  * it will reset op_fid1 with the FID of the choosen stripe.
1700  **/
1701 struct lmv_tgt_desc *
1702 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1703                            const char *name, int namelen, struct lu_fid *fid,
1704                            u32 *mds)
1705 {
1706         struct lmv_tgt_desc     *tgt;
1707         const struct lmv_oinfo  *oinfo;
1708
1709         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1710         if (IS_ERR(oinfo))
1711                 RETURN(ERR_CAST(oinfo));
1712         *fid = oinfo->lmo_fid;
1713         *mds = oinfo->lmo_mds;
1714         tgt = lmv_get_target(lmv, *mds, NULL);
1715
1716         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid));
1717         return tgt;
1718 }
1719
1720 /**
1721  * Locate mds by fid or name
1722  *
1723  * For striped directory (lsm != NULL), it will locate the stripe
1724  * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1725  * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1726  * walk through all of stripes to locate the entry.
1727  *
1728  * For normal direcotry, it will locate MDS by FID directly.
1729  * \param[in] lmv       LMV device
1730  * \param[in] op_data   client MD stack parameters, name, namelen
1731  *                      mds_num etc.
1732  * \param[in] fid       object FID used to locate MDS.
1733  *
1734  * retval               pointer to the lmv_tgt_desc if succeed.
1735  *                      ERR_PTR(errno) if failed.
1736  */
1737 struct lmv_tgt_desc*
1738 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1739                struct lu_fid *fid)
1740 {
1741         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1742         struct lmv_tgt_desc     *tgt;
1743
1744         /* During creating VOLATILE file, it should honor the mdt
1745          * index if the file under striped dir is being restored, see
1746          * ct_restore(). */
1747         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1748             (int)op_data->op_mds != -1 && lsm != NULL) {
1749                 int i;
1750                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1751                 if (IS_ERR(tgt))
1752                         return tgt;
1753
1754                 /* refill the right parent fid */
1755                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1756                         struct lmv_oinfo *oinfo;
1757
1758                         oinfo = &lsm->lsm_md_oinfo[i];
1759                         if (oinfo->lmo_mds == op_data->op_mds) {
1760                                 *fid = oinfo->lmo_fid;
1761                                 break;
1762                         }
1763                 }
1764
1765                 /* Hmm, can not find the stripe by mdt_index(op_mds) */
1766                 if (i == lsm->lsm_md_stripe_count)
1767                         tgt = ERR_PTR(-EINVAL);
1768
1769                 return tgt;
1770         }
1771
1772         if (lsm == NULL || op_data->op_namelen == 0) {
1773                 tgt = lmv_find_target(lmv, fid);
1774                 if (IS_ERR(tgt))
1775                         return tgt;
1776
1777                 op_data->op_mds = tgt->ltd_idx;
1778                 return tgt;
1779         }
1780
1781         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1782                                           op_data->op_namelen, fid,
1783                                           &op_data->op_mds);
1784 }
1785
1786 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1787                const void *data, int datalen, int mode, __u32 uid,
1788                __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1789                struct ptlrpc_request **request)
1790 {
1791         struct obd_device       *obd = exp->exp_obd;
1792         struct lmv_obd          *lmv = &obd->u.lmv;
1793         struct lmv_tgt_desc     *tgt;
1794         int                      rc;
1795         ENTRY;
1796
1797         rc = lmv_check_connect(obd);
1798         if (rc)
1799                 RETURN(rc);
1800
1801         if (!lmv->desc.ld_active_tgt_count)
1802                 RETURN(-EIO);
1803
1804         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1805         if (IS_ERR(tgt))
1806                 RETURN(PTR_ERR(tgt));
1807
1808         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1809                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1810                op_data->op_mds);
1811
1812         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1813         if (rc)
1814                 RETURN(rc);
1815
1816         /* Send the create request to the MDT where the object
1817          * will be located */
1818         tgt = lmv_find_target(lmv, &op_data->op_fid2);
1819         if (IS_ERR(tgt))
1820                 RETURN(PTR_ERR(tgt));
1821
1822         op_data->op_mds = tgt->ltd_idx;
1823
1824         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1825                PFID(&op_data->op_fid2), op_data->op_mds);
1826
1827         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1828         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1829                        cap_effective, rdev, request);
1830         if (rc == 0) {
1831                 if (*request == NULL)
1832                         RETURN(rc);
1833                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1834         }
1835         RETURN(rc);
1836 }
1837
1838 static int lmv_done_writing(struct obd_export *exp,
1839                             struct md_op_data *op_data,
1840                             struct md_open_data *mod)
1841 {
1842         struct obd_device     *obd = exp->exp_obd;
1843         struct lmv_obd        *lmv = &obd->u.lmv;
1844         struct lmv_tgt_desc   *tgt;
1845         int                    rc;
1846         ENTRY;
1847
1848         rc = lmv_check_connect(obd);
1849         if (rc)
1850                 RETURN(rc);
1851
1852         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1853         if (IS_ERR(tgt))
1854                 RETURN(PTR_ERR(tgt));
1855
1856         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1857         RETURN(rc);
1858 }
1859
1860 static int
1861 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1862             const union ldlm_policy_data *policy,
1863             struct lookup_intent *it, struct md_op_data *op_data,
1864             struct lustre_handle *lockh, __u64 extra_lock_flags)
1865 {
1866         struct obd_device        *obd = exp->exp_obd;
1867         struct lmv_obd           *lmv = &obd->u.lmv;
1868         struct lmv_tgt_desc      *tgt;
1869         int                       rc;
1870         ENTRY;
1871
1872         rc = lmv_check_connect(obd);
1873         if (rc)
1874                 RETURN(rc);
1875
1876         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1877                LL_IT2STR(it), PFID(&op_data->op_fid1));
1878
1879         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1880         if (IS_ERR(tgt))
1881                 RETURN(PTR_ERR(tgt));
1882
1883         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1884                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1885
1886         rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1887                         extra_lock_flags);
1888
1889         RETURN(rc);
1890 }
1891
1892 static int
1893 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1894                  struct ptlrpc_request **preq)
1895 {
1896         struct ptlrpc_request   *req = NULL;
1897         struct obd_device       *obd = exp->exp_obd;
1898         struct lmv_obd          *lmv = &obd->u.lmv;
1899         struct lmv_tgt_desc     *tgt;
1900         struct mdt_body         *body;
1901         int                      rc;
1902         ENTRY;
1903
1904         rc = lmv_check_connect(obd);
1905         if (rc)
1906                 RETURN(rc);
1907
1908         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1909         if (IS_ERR(tgt))
1910                 RETURN(PTR_ERR(tgt));
1911
1912         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%u\n",
1913                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1914                tgt->ltd_idx);
1915
1916         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1917         if (rc != 0)
1918                 RETURN(rc);
1919
1920         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1921         LASSERT(body != NULL);
1922
1923         if (body->mbo_valid & OBD_MD_MDS) {
1924                 struct lu_fid rid = body->mbo_fid1;
1925                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1926                        PFID(&rid));
1927
1928                 tgt = lmv_find_target(lmv, &rid);
1929                 if (IS_ERR(tgt)) {
1930                         ptlrpc_req_finished(*preq);
1931                         preq = NULL;
1932                         RETURN(PTR_ERR(tgt));
1933                 }
1934
1935                 op_data->op_fid1 = rid;
1936                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1937                 op_data->op_namelen = 0;
1938                 op_data->op_name = NULL;
1939                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1940                 ptlrpc_req_finished(*preq);
1941                 *preq = req;
1942         }
1943
1944         RETURN(rc);
1945 }
1946
1947 #define md_op_data_fid(op_data, fl)                     \
1948         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1949          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1950          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1951          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1952          NULL)
1953
1954 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1955                             struct md_op_data *op_data,
1956                             __u32 op_tgt, ldlm_mode_t mode, int bits, int flag)
1957 {
1958         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1959         struct obd_device      *obd = exp->exp_obd;
1960         struct lmv_obd         *lmv = &obd->u.lmv;
1961         ldlm_policy_data_t      policy = {{ 0 }};
1962         int                     rc = 0;
1963         ENTRY;
1964
1965         if (!fid_is_sane(fid))
1966                 RETURN(0);
1967
1968         if (tgt == NULL) {
1969                 tgt = lmv_find_target(lmv, fid);
1970                 if (IS_ERR(tgt))
1971                         RETURN(PTR_ERR(tgt));
1972         }
1973
1974         if (tgt->ltd_idx != op_tgt) {
1975                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1976                 policy.l_inodebits.bits = bits;
1977                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1978                                       mode, LCF_ASYNC, NULL);
1979         } else {
1980                 CDEBUG(D_INODE,
1981                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1982                        op_tgt, PFID(fid));
1983                 op_data->op_flags |= flag;
1984                 rc = 0;
1985         }
1986
1987         RETURN(rc);
1988 }
1989
1990 /*
1991  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1992  * op_data->op_fid2
1993  */
1994 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1995                     struct ptlrpc_request **request)
1996 {
1997         struct obd_device       *obd = exp->exp_obd;
1998         struct lmv_obd          *lmv = &obd->u.lmv;
1999         struct lmv_tgt_desc     *tgt;
2000         int                      rc;
2001         ENTRY;
2002
2003         rc = lmv_check_connect(obd);
2004         if (rc)
2005                 RETURN(rc);
2006
2007         LASSERT(op_data->op_namelen != 0);
2008
2009         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2010                PFID(&op_data->op_fid2), op_data->op_namelen,
2011                op_data->op_name, PFID(&op_data->op_fid1));
2012
2013         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2014         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2015         op_data->op_cap = cfs_curproc_cap_pack();
2016         if (op_data->op_mea2 != NULL) {
2017                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
2018                 const struct lmv_oinfo  *oinfo;
2019
2020                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2021                                                 op_data->op_namelen);
2022                 if (IS_ERR(oinfo))
2023                         RETURN(PTR_ERR(oinfo));
2024
2025                 op_data->op_fid2 = oinfo->lmo_fid;
2026         }
2027
2028         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2029         if (IS_ERR(tgt))
2030                 RETURN(PTR_ERR(tgt));
2031
2032         /*
2033          * Cancel UPDATE lock on child (fid1).
2034          */
2035         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2036         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2037                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2038         if (rc != 0)
2039                 RETURN(rc);
2040
2041         rc = md_link(tgt->ltd_exp, op_data, request);
2042
2043         RETURN(rc);
2044 }
2045
2046 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2047                       const char *old, int oldlen, const char *new, int newlen,
2048                       struct ptlrpc_request **request)
2049 {
2050         struct obd_device       *obd = exp->exp_obd;
2051         struct lmv_obd          *lmv = &obd->u.lmv;
2052         struct lmv_tgt_desc     *src_tgt;
2053         int                     rc;
2054         ENTRY;
2055
2056         LASSERT(oldlen != 0);
2057
2058         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2059                oldlen, old, PFID(&op_data->op_fid1),
2060                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2061                newlen, new, PFID(&op_data->op_fid2),
2062                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2063
2064         rc = lmv_check_connect(obd);
2065         if (rc)
2066                 RETURN(rc);
2067
2068         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2069         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2070         op_data->op_cap = cfs_curproc_cap_pack();
2071         if (op_data->op_cli_flags & CLI_MIGRATE) {
2072                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2073                          PFID(&op_data->op_fid3));
2074                 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2075                 if (rc)
2076                         RETURN(rc);
2077                 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3);
2078         } else {
2079                 if (op_data->op_mea1 != NULL) {
2080                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2081
2082                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2083                                                              oldlen,
2084                                                              &op_data->op_fid1,
2085                                                              &op_data->op_mds);
2086                         if (IS_ERR(src_tgt))
2087                                 RETURN(PTR_ERR(src_tgt));
2088                 } else {
2089                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2090                         if (IS_ERR(src_tgt))
2091                                 RETURN(PTR_ERR(src_tgt));
2092
2093                         op_data->op_mds = src_tgt->ltd_idx;
2094                 }
2095
2096                 if (op_data->op_mea2) {
2097                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
2098                         const struct lmv_oinfo  *oinfo;
2099
2100                         oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2101                         if (IS_ERR(oinfo))
2102                                 RETURN(PTR_ERR(oinfo));
2103
2104                         op_data->op_fid2 = oinfo->lmo_fid;
2105                 }
2106         }
2107         if (IS_ERR(src_tgt))
2108                 RETURN(PTR_ERR(src_tgt));
2109
2110         /*
2111          * LOOKUP lock on src child (fid3) should also be cancelled for
2112          * src_tgt in mdc_rename.
2113          */
2114         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2115
2116         /*
2117          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2118          * own target.
2119          */
2120         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2121                               LCK_EX, MDS_INODELOCK_UPDATE,
2122                               MF_MDC_CANCEL_FID2);
2123
2124         if (rc != 0)
2125                 RETURN(rc);
2126         /*
2127          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2128          */
2129         if (fid_is_sane(&op_data->op_fid3)) {
2130                 struct lmv_tgt_desc *tgt;
2131
2132                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2133                 if (IS_ERR(tgt))
2134                         RETURN(PTR_ERR(tgt));
2135
2136                 /* Cancel LOOKUP lock on its parent */
2137                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2138                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2139                                       MF_MDC_CANCEL_FID3);
2140                 if (rc != 0)
2141                         RETURN(rc);
2142
2143                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2144                                       LCK_EX, MDS_INODELOCK_FULL,
2145                                       MF_MDC_CANCEL_FID3);
2146                 if (rc != 0)
2147                         RETURN(rc);
2148         }
2149
2150         /*
2151          * Cancel all the locks on tgt child (fid4).
2152          */
2153         if (fid_is_sane(&op_data->op_fid4))
2154                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2155                                       LCK_EX, MDS_INODELOCK_FULL,
2156                                       MF_MDC_CANCEL_FID4);
2157
2158         CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2159                op_data->op_mds, PFID(&op_data->op_fid2));
2160
2161         rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2162                        request);
2163
2164         RETURN(rc);
2165 }
2166
2167 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2168                        void *ea, int ealen, void *ea2, int ea2len,
2169                        struct ptlrpc_request **request,
2170                        struct md_open_data **mod)
2171 {
2172         struct obd_device       *obd = exp->exp_obd;
2173         struct lmv_obd          *lmv = &obd->u.lmv;
2174         struct lmv_tgt_desc     *tgt;
2175         int                      rc = 0;
2176         ENTRY;
2177
2178         rc = lmv_check_connect(obd);
2179         if (rc)
2180                 RETURN(rc);
2181
2182         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2183                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2184
2185         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2186         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2187         if (IS_ERR(tgt))
2188                 RETURN(PTR_ERR(tgt));
2189
2190         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2191                         ea2len, request, mod);
2192
2193         RETURN(rc);
2194 }
2195
2196 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2197                      struct obd_capa *oc, struct ptlrpc_request **request)
2198 {
2199         struct obd_device       *obd = exp->exp_obd;
2200         struct lmv_obd          *lmv = &obd->u.lmv;
2201         struct lmv_tgt_desc     *tgt;
2202         int                      rc;
2203         ENTRY;
2204
2205         rc = lmv_check_connect(obd);
2206         if (rc != 0)
2207                 RETURN(rc);
2208
2209         tgt = lmv_find_target(lmv, fid);
2210         if (IS_ERR(tgt))
2211                 RETURN(PTR_ERR(tgt));
2212
2213         rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2214         RETURN(rc);
2215 }
2216
2217 /**
2218  * Get current minimum entry from striped directory
2219  *
2220  * This function will search the dir entry, whose hash value is the
2221  * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2222  * only being called for striped directory.
2223  *
2224  * \param[in] exp               export of LMV
2225  * \param[in] op_data           parameters transferred beween client MD stack
2226  *                              stripe_information will be included in this
2227  *                              parameter
2228  * \param[in] cb_op             ldlm callback being used in enqueue in
2229  *                              mdc_read_page
2230  * \param[in] hash_offset       the hash value, which is used to locate
2231  *                              minum(closet) dir entry
2232  * \param[in|out] stripe_offset the caller use this to indicate the stripe
2233  *                              index of last entry, so to avoid hash conflict
2234  *                              between stripes. It will also be used to
2235  *                              return the stripe index of current dir entry.
2236  * \param[in|out] entp          the minum entry and it also is being used
2237  *                              to input the last dir entry to resolve the
2238  *                              hash conflict
2239  *
2240  * \param[out] ppage            the page which holds the minum entry
2241  *
2242  * \retval                      = 0 get the entry successfully
2243  *                              negative errno (< 0) does not get the entry
2244  */
2245 static int lmv_get_min_striped_entry(struct obd_export *exp,
2246                                      struct md_op_data *op_data,
2247                                      struct md_callback *cb_op,
2248                                      __u64 hash_offset, int *stripe_offset,
2249                                      struct lu_dirent **entp,
2250                                      struct page **ppage)
2251 {
2252         struct obd_device       *obd = exp->exp_obd;
2253         struct lmv_obd          *lmv = &obd->u.lmv;
2254         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2255         struct lmv_tgt_desc     *tgt;
2256         int                     stripe_count;
2257         struct lu_dirent        *min_ent = NULL;
2258         struct page             *min_page = NULL;
2259         int                     min_idx = 0;
2260         int                     i;
2261         int                     rc = 0;
2262         ENTRY;
2263
2264         stripe_count = lsm->lsm_md_stripe_count;
2265         for (i = 0; i < stripe_count; i++) {
2266                 struct lu_dirent        *ent = NULL;
2267                 struct page             *page = NULL;
2268                 struct lu_dirpage       *dp;
2269                 __u64                   stripe_hash = hash_offset;
2270
2271                 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2272                 if (IS_ERR(tgt))
2273                         GOTO(out, rc = PTR_ERR(tgt));
2274
2275                 /* op_data will be shared by each stripe, so we need
2276                  * reset these value for each stripe */
2277                 op_data->op_stripe_offset = i;
2278                 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2279                 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2280                 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2281 next:
2282                 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2283                                   &page);
2284                 if (rc != 0)
2285                         GOTO(out, rc);
2286
2287                 dp = page_address(page);
2288                 for (ent = lu_dirent_start(dp); ent != NULL;
2289                      ent = lu_dirent_next(ent)) {
2290                         /* Skip dummy entry */
2291                         if (le16_to_cpu(ent->lde_namelen) == 0)
2292                                 continue;
2293
2294                         if (le64_to_cpu(ent->lde_hash) < hash_offset)
2295                                 continue;
2296
2297                         if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2298                             (*entp == ent || i < *stripe_offset))
2299                                 continue;
2300
2301                         /* skip . and .. for other stripes */
2302                         if (i != 0 &&
2303                             (strncmp(ent->lde_name, ".",
2304                                      le16_to_cpu(ent->lde_namelen)) == 0 ||
2305                              strncmp(ent->lde_name, "..",
2306                                      le16_to_cpu(ent->lde_namelen)) == 0))
2307                                 continue;
2308                         break;
2309                 }
2310
2311                 if (ent == NULL) {
2312                         stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2313
2314                         kunmap(page);
2315                         page_cache_release(page);
2316                         page = NULL;
2317
2318                         /* reach the end of current stripe, go to next stripe */
2319                         if (stripe_hash == MDS_DIR_END_OFF)
2320                                 continue;
2321                         else
2322                                 goto next;
2323                 }
2324
2325                 if (min_ent != NULL) {
2326                         if (le64_to_cpu(min_ent->lde_hash) >
2327                             le64_to_cpu(ent->lde_hash)) {
2328                                 min_ent = ent;
2329                                 kunmap(min_page);
2330                                 page_cache_release(min_page);
2331                                 min_idx = i;
2332                                 min_page = page;
2333                         } else {
2334                                 kunmap(page);
2335                                 page_cache_release(page);
2336                                 page = NULL;
2337                         }
2338                 } else {
2339                         min_ent = ent;
2340                         min_page = page;
2341                         min_idx = i;
2342                 }
2343         }
2344
2345 out:
2346         if (*ppage != NULL) {
2347                 kunmap(*ppage);
2348                 page_cache_release(*ppage);
2349         }
2350         *stripe_offset = min_idx;
2351         *entp = min_ent;
2352         *ppage = min_page;
2353         RETURN(rc);
2354 }
2355
2356 /**
2357  * Build dir entry page from a striped directory
2358  *
2359  * This function gets one entry by @offset from a striped directory. It will
2360  * read entries from all of stripes, and choose one closest to the required
2361  * offset(&offset). A few notes
2362  * 1. skip . and .. for non-zero stripes, because there can only have one .
2363  * and .. in a directory.
2364  * 2. op_data will be shared by all of stripes, instead of allocating new
2365  * one, so need to restore before reusing.
2366  * 3. release the entry page if that is not being chosen.
2367  *
2368  * \param[in] exp       obd export refer to LMV
2369  * \param[in] op_data   hold those MD parameters of read_entry
2370  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2371  * \param[out] ldp      the entry being read
2372  * \param[out] ppage    the page holding the entry. Note: because the entry
2373  *                      will be accessed in upper layer, so we need hold the
2374  *                      page until the usages of entry is finished, see
2375  *                      ll_dir_entry_next.
2376  *
2377  * retval               =0 if get entry successfully
2378  *                      <0 cannot get entry
2379  */
2380 static int lmv_read_striped_page(struct obd_export *exp,
2381                                  struct md_op_data *op_data,
2382                                  struct md_callback *cb_op,
2383                                  __u64 offset, struct page **ppage)
2384 {
2385         struct obd_device       *obd = exp->exp_obd;
2386         struct lu_fid           master_fid = op_data->op_fid1;
2387         struct inode            *master_inode = op_data->op_data;
2388         __u64                   hash_offset = offset;
2389         struct lu_dirpage       *dp;
2390         struct page             *min_ent_page = NULL;
2391         struct page             *ent_page = NULL;
2392         struct lu_dirent        *ent;
2393         void                    *area;
2394         int                     ent_idx = 0;
2395         struct lu_dirent        *min_ent = NULL;
2396         struct lu_dirent        *last_ent;
2397         int                     left_bytes;
2398         int                     rc;
2399         ENTRY;
2400
2401         rc = lmv_check_connect(obd);
2402         if (rc)
2403                 RETURN(rc);
2404
2405         /* Allocate a page and read entries from all of stripes and fill
2406          * the page by hash order */
2407         ent_page = alloc_page(GFP_KERNEL);
2408         if (ent_page == NULL)
2409                 RETURN(-ENOMEM);
2410
2411         /* Initialize the entry page */
2412         dp = kmap(ent_page);
2413         memset(dp, 0, sizeof(*dp));
2414         dp->ldp_hash_start = cpu_to_le64(offset);
2415         dp->ldp_flags |= LDF_COLLIDE;
2416
2417         area = dp + 1;
2418         left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2419         ent = area;
2420         last_ent = ent;
2421         do {
2422                 __u16   ent_size;
2423
2424                 /* Find the minum entry from all sub-stripes */
2425                 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2426                                                &ent_idx, &min_ent,
2427                                                &min_ent_page);
2428                 if (rc != 0)
2429                         GOTO(out, rc);
2430
2431                 /* If it can not get minum entry, it means it already reaches
2432                  * the end of this directory */
2433                 if (min_ent == NULL) {
2434                         last_ent->lde_reclen = 0;
2435                         hash_offset = MDS_DIR_END_OFF;
2436                         GOTO(out, rc);
2437                 }
2438
2439                 ent_size = le16_to_cpu(min_ent->lde_reclen);
2440
2441                 /* the last entry lde_reclen is 0, but it might not
2442                  * the end of this entry of this temporay entry */
2443                 if (ent_size == 0)
2444                         ent_size = lu_dirent_calc_size(
2445                                         le16_to_cpu(min_ent->lde_namelen),
2446                                         le32_to_cpu(min_ent->lde_attrs));
2447                 if (ent_size > left_bytes) {
2448                         last_ent->lde_reclen = cpu_to_le16(0);
2449                         hash_offset = le64_to_cpu(min_ent->lde_hash);
2450                         GOTO(out, rc);
2451                 }
2452
2453                 memcpy(ent, min_ent, ent_size);
2454
2455                 /* Replace . with master FID and Replace .. with the parent FID
2456                  * of master object */
2457                 if (strncmp(ent->lde_name, ".",
2458                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2459                     le16_to_cpu(ent->lde_namelen) == 1)
2460                         fid_cpu_to_le(&ent->lde_fid, &master_fid);
2461                 else if (strncmp(ent->lde_name, "..",
2462                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2463                            le16_to_cpu(ent->lde_namelen) == 2)
2464                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2465
2466                 left_bytes -= ent_size;
2467                 ent->lde_reclen = cpu_to_le16(ent_size);
2468                 last_ent = ent;
2469                 ent = (void *)ent + ent_size;
2470                 hash_offset = le64_to_cpu(min_ent->lde_hash);
2471                 if (hash_offset == MDS_DIR_END_OFF) {
2472                         last_ent->lde_reclen = 0;
2473                         break;
2474                 }
2475         } while (1);
2476 out:
2477         if (min_ent_page != NULL) {
2478                 kunmap(min_ent_page);
2479                 page_cache_release(min_ent_page);
2480         }
2481
2482         if (unlikely(rc != 0)) {
2483                 __free_page(ent_page);
2484                 ent_page = NULL;
2485         } else {
2486                 if (ent == area)
2487                         dp->ldp_flags |= LDF_EMPTY;
2488                 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2489                 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2490         }
2491
2492         /* We do not want to allocate md_op_data during each
2493          * dir entry reading, so op_data will be shared by every stripe,
2494          * then we need to restore it back to original value before
2495          * return to the upper layer */
2496         op_data->op_fid1 = master_fid;
2497         op_data->op_fid2 = master_fid;
2498         op_data->op_data = master_inode;
2499
2500         *ppage = ent_page;
2501
2502         RETURN(rc);
2503 }
2504
2505 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2506                   struct md_callback *cb_op, __u64 offset,
2507                   struct page **ppage)
2508 {
2509         struct obd_device       *obd = exp->exp_obd;
2510         struct lmv_obd          *lmv = &obd->u.lmv;
2511         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2512         struct lmv_tgt_desc     *tgt;
2513         int                     rc;
2514         ENTRY;
2515
2516         rc = lmv_check_connect(obd);
2517         if (rc != 0)
2518                 RETURN(rc);
2519
2520         if (unlikely(lsm != NULL)) {
2521                 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2522                 RETURN(rc);
2523         }
2524
2525         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2526         if (IS_ERR(tgt))
2527                 RETURN(PTR_ERR(tgt));
2528
2529         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2530
2531         RETURN(rc);
2532 }
2533
2534 /**
2535  * Unlink a file/directory
2536  *
2537  * Unlink a file or directory under the parent dir. The unlink request
2538  * usually will be sent to the MDT where the child is located, but if
2539  * the client does not have the child FID then request will be sent to the
2540  * MDT where the parent is located.
2541  *
2542  * If the parent is a striped directory then it also needs to locate which
2543  * stripe the name of the child is located, and replace the parent FID
2544  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2545  * it will walk through all of sub-stripes until the child is being
2546  * unlinked finally.
2547  *
2548  * \param[in] exp       export refer to LMV
2549  * \param[in] op_data   different parameters transferred beween client
2550  *                      MD stacks, name, namelen, FIDs etc.
2551  *                      op_fid1 is the parent FID, op_fid2 is the child
2552  *                      FID.
2553  * \param[out] request  point to the request of unlink.
2554  *
2555  * retval               0 if succeed
2556  *                      negative errno if failed.
2557  */
2558 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2559                       struct ptlrpc_request **request)
2560 {
2561         struct obd_device       *obd = exp->exp_obd;
2562         struct lmv_obd          *lmv = &obd->u.lmv;
2563         struct lmv_tgt_desc     *tgt = NULL;
2564         struct lmv_tgt_desc     *parent_tgt = NULL;
2565         struct mdt_body         *body;
2566         int                     rc;
2567         int                     stripe_index = 0;
2568         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2569         ENTRY;
2570
2571         rc = lmv_check_connect(obd);
2572         if (rc)
2573                 RETURN(rc);
2574 retry_unlink:
2575         /* For striped dir, we need to locate the parent as well */
2576         if (lsm != NULL) {
2577                 struct lmv_tgt_desc *tmp;
2578
2579                 LASSERT(op_data->op_name != NULL &&
2580                         op_data->op_namelen != 0);
2581
2582                 tmp = lmv_locate_target_for_name(lmv, lsm,
2583                                                  op_data->op_name,
2584                                                  op_data->op_namelen,
2585                                                  &op_data->op_fid1,
2586                                                  &op_data->op_mds);
2587
2588                 /* return -EBADFD means unknown hash type, might
2589                  * need try all sub-stripe here */
2590                 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2591                         RETURN(PTR_ERR(tmp));
2592
2593                 /* Note: both migrating dir and unknown hash dir need to
2594                  * try all of sub-stripes, so we need start search the
2595                  * name from stripe 0, but migrating dir is already handled
2596                  * inside lmv_locate_target_for_name(), so we only check
2597                  * unknown hash type directory here */
2598                 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2599                         struct lmv_oinfo *oinfo;
2600
2601                         oinfo = &lsm->lsm_md_oinfo[stripe_index];
2602
2603                         op_data->op_fid1 = oinfo->lmo_fid;
2604                         op_data->op_mds = oinfo->lmo_mds;
2605                 }
2606         }
2607
2608 try_next_stripe:
2609         /* Send unlink requests to the MDT where the child is located */
2610         if (likely(!fid_is_zero(&op_data->op_fid2)))
2611                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2612         else if (lsm != NULL)
2613                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2614         else
2615                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2616
2617         if (IS_ERR(tgt))
2618                 RETURN(PTR_ERR(tgt));
2619
2620         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2621         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2622         op_data->op_cap = cfs_curproc_cap_pack();
2623
2624         /*
2625          * If child's fid is given, cancel unused locks for it if it is from
2626          * another export than parent.
2627          *
2628          * LOOKUP lock for child (fid3) should also be cancelled on parent
2629          * tgt_tgt in mdc_unlink().
2630          */
2631         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2632
2633         /*
2634          * Cancel FULL locks on child (fid3).
2635          */
2636         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2637         if (IS_ERR(parent_tgt))
2638                 RETURN(PTR_ERR(parent_tgt));
2639
2640         if (parent_tgt != tgt) {
2641                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2642                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2643                                       MF_MDC_CANCEL_FID3);
2644         }
2645
2646         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2647                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2648         if (rc != 0)
2649                 RETURN(rc);
2650
2651         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2652                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2653
2654         rc = md_unlink(tgt->ltd_exp, op_data, request);
2655         if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2656                 RETURN(rc);
2657
2658         /* Try next stripe if it is needed. */
2659         if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2660                 struct lmv_oinfo *oinfo;
2661
2662                 stripe_index++;
2663                 if (stripe_index >= lsm->lsm_md_stripe_count)
2664                         RETURN(rc);
2665
2666                 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2667
2668                 op_data->op_fid1 = oinfo->lmo_fid;
2669                 op_data->op_mds = oinfo->lmo_mds;
2670
2671                 ptlrpc_req_finished(*request);
2672                 *request = NULL;
2673
2674                 goto try_next_stripe;
2675         }
2676
2677         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2678         if (body == NULL)
2679                 RETURN(-EPROTO);
2680
2681         /* Not cross-ref case, just get out of here. */
2682         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2683                 RETURN(rc);
2684
2685         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2686                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2687
2688         /* This is a remote object, try remote MDT, Note: it may
2689          * try more than 1 time here, Considering following case
2690          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2691          * 1. Initially A does not know where remote1 is, it send
2692          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2693          *    resend unlink RPC to MDT1 (retry 1st time).
2694          *
2695          * 2. During the unlink RPC in flight,
2696          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2697          *    and create new remote1, but on MDT0
2698          *
2699          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2700          *    /mnt/lustre, then lookup get fid of remote1, and find
2701          *    it is remote dir again, and replay -EREMOTE again.
2702          *
2703          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2704          *
2705          * In theory, it might try unlimited time here, but it should
2706          * be very rare case.  */
2707         op_data->op_fid2 = body->mbo_fid1;
2708         ptlrpc_req_finished(*request);
2709         *request = NULL;
2710
2711         goto retry_unlink;
2712 }
2713
2714 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2715 {
2716         struct lmv_obd *lmv = &obd->u.lmv;
2717         int rc = 0;
2718
2719         switch (stage) {
2720         case OBD_CLEANUP_EARLY:
2721                 /* XXX: here should be calling obd_precleanup() down to
2722                  * stack. */
2723                 break;
2724         case OBD_CLEANUP_EXPORTS:
2725                 fld_client_proc_fini(&lmv->lmv_fld);
2726                 lprocfs_obd_cleanup(obd);
2727                 lprocfs_free_md_stats(obd);
2728                 break;
2729         default:
2730                 break;
2731         }
2732         RETURN(rc);
2733 }
2734
2735 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2736                         __u32 keylen, void *key, __u32 *vallen, void *val,
2737                         struct lov_stripe_md *lsm)
2738 {
2739         struct obd_device       *obd;
2740         struct lmv_obd          *lmv;
2741         int                      rc = 0;
2742         ENTRY;
2743
2744         obd = class_exp2obd(exp);
2745         if (obd == NULL) {
2746                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2747                        exp->exp_handle.h_cookie);
2748                 RETURN(-EINVAL);
2749         }
2750
2751         lmv = &obd->u.lmv;
2752         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2753                 int i;
2754
2755                 rc = lmv_check_connect(obd);
2756                 if (rc)
2757                         RETURN(rc);
2758
2759                 LASSERT(*vallen == sizeof(__u32));
2760                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2761                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2762                         /*
2763                          * All tgts should be connected when this gets called.
2764                          */
2765                         if (tgt == NULL || tgt->ltd_exp == NULL)
2766                                 continue;
2767
2768                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2769                                           vallen, val, NULL))
2770                                 RETURN(0);
2771                 }
2772                 RETURN(-EINVAL);
2773         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2774                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2775                    KEY_IS(KEY_MAX_COOKIESIZE) ||
2776                    KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2777                    KEY_IS(KEY_CONN_DATA)) {
2778                 rc = lmv_check_connect(obd);
2779                 if (rc)
2780                         RETURN(rc);
2781
2782                 /*
2783                  * Forwarding this request to first MDS, it should know LOV
2784                  * desc.
2785                  */
2786                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2787                                   vallen, val, NULL);
2788                 if (!rc && KEY_IS(KEY_CONN_DATA))
2789                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2790                 RETURN(rc);
2791         } else if (KEY_IS(KEY_TGT_COUNT)) {
2792                 *((int *)val) = lmv->desc.ld_tgt_count;
2793                 RETURN(0);
2794         }
2795
2796         CDEBUG(D_IOCTL, "Invalid key\n");
2797         RETURN(-EINVAL);
2798 }
2799
2800 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2801                        obd_count keylen, void *key, obd_count vallen,
2802                        void *val, struct ptlrpc_request_set *set)
2803 {
2804         struct lmv_tgt_desc    *tgt = NULL;
2805         struct obd_device      *obd;
2806         struct lmv_obd         *lmv;
2807         int rc = 0;
2808         ENTRY;
2809
2810         obd = class_exp2obd(exp);
2811         if (obd == NULL) {
2812                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2813                        exp->exp_handle.h_cookie);
2814                 RETURN(-EINVAL);
2815         }
2816         lmv = &obd->u.lmv;
2817
2818         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2819                 int i, err = 0;
2820
2821                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2822                         tgt = lmv->tgts[i];
2823
2824                         if (tgt == NULL || tgt->ltd_exp == NULL)
2825                                 continue;
2826
2827                         err = obd_set_info_async(env, tgt->ltd_exp,
2828                                                  keylen, key, vallen, val, set);
2829                         if (err && rc == 0)
2830                                 rc = err;
2831                 }
2832
2833                 RETURN(rc);
2834         }
2835
2836         RETURN(-EINVAL);
2837 }
2838
2839 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2840                           struct lmv_mds_md_v1 *lmm1)
2841 {
2842         int     cplen;
2843         int     i;
2844
2845         lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2846         lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2847         lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2848         lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2849         cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2850                         sizeof(lmm1->lmv_pool_name));
2851         if (cplen >= sizeof(lmm1->lmv_pool_name))
2852                 return -E2BIG;
2853
2854         for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2855                 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2856                               &lsm->lsm_md_oinfo[i].lmo_fid);
2857         return 0;
2858 }
2859
2860 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2861                 int stripe_count)
2862 {
2863         int     lmm_size = 0;
2864         bool    allocated = false;
2865         int     rc = 0;
2866         ENTRY;
2867
2868         LASSERT(lmmp != NULL);
2869         /* Free lmm */
2870         if (*lmmp != NULL && lsm == NULL) {
2871                 int stripe_count;
2872
2873                 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2874                 lmm_size = lmv_mds_md_size(stripe_count,
2875                                            le32_to_cpu((*lmmp)->lmv_magic));
2876                 if (lmm_size == 0)
2877                         RETURN(-EINVAL);
2878                 OBD_FREE(*lmmp, lmm_size);
2879                 *lmmp = NULL;
2880                 RETURN(0);
2881         }
2882
2883         /* Alloc lmm */
2884         if (*lmmp == NULL && lsm == NULL) {
2885                 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2886                 LASSERT(lmm_size > 0);
2887                 OBD_ALLOC(*lmmp, lmm_size);
2888                 if (*lmmp == NULL)
2889                         RETURN(-ENOMEM);
2890                 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2891                 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2892                 RETURN(lmm_size);
2893         }
2894
2895         /* pack lmm */
2896         LASSERT(lsm != NULL);
2897         lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2898         if (*lmmp == NULL) {
2899                 OBD_ALLOC(*lmmp, lmm_size);
2900                 if (*lmmp == NULL)
2901                         RETURN(-ENOMEM);
2902                 allocated = true;
2903         }
2904
2905         switch (lsm->lsm_md_magic) {
2906         case LMV_MAGIC_V1:
2907                 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2908                 break;
2909         default:
2910                 rc = -EINVAL;
2911                 break;
2912         }
2913
2914         if (rc != 0 && allocated) {
2915                 OBD_FREE(*lmmp, lmm_size);
2916                 *lmmp = NULL;
2917         }
2918
2919         RETURN(lmm_size);
2920 }
2921 EXPORT_SYMBOL(lmv_pack_md);
2922
2923 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2924                             const struct lmv_mds_md_v1 *lmm1)
2925 {
2926         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
2927         int             stripe_count;
2928         int             cplen;
2929         int             i;
2930         int             rc = 0;
2931         ENTRY;
2932
2933         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2934         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2935         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2936         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2937                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2938         else
2939                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2940         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2941         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2942                         sizeof(lsm->lsm_md_pool_name));
2943
2944         if (cplen >= sizeof(lsm->lsm_md_pool_name))
2945                 RETURN(-E2BIG);
2946
2947         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2948                "layout_version %d\n", lsm->lsm_md_stripe_count,
2949                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2950                lsm->lsm_md_layout_version);
2951
2952         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2953         for (i = 0; i < stripe_count; i++) {
2954                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2955                               &lmm1->lmv_stripe_fids[i]);
2956                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2957                                     &lsm->lsm_md_oinfo[i].lmo_mds);
2958                 if (rc != 0)
2959                         RETURN(rc);
2960                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2961                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2962         }
2963
2964         RETURN(rc);
2965 }
2966
2967 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2968                   const union lmv_mds_md *lmm, int stripe_count)
2969 {
2970         struct lmv_stripe_md     *lsm;
2971         int                      lsm_size;
2972         int                      rc;
2973         bool                     allocated = false;
2974         ENTRY;
2975
2976         LASSERT(lsmp != NULL);
2977
2978         lsm = *lsmp;
2979         /* Free memmd */
2980         if (lsm != NULL && lmm == NULL) {
2981                 int i;
2982                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2983                         /* For migrating inode, the master stripe and master
2984                          * object will be the same, so do not need iput, see
2985                          * ll_update_lsm_md */
2986                         if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2987                               i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2988                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
2989                 }
2990                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2991                 OBD_FREE(lsm, lsm_size);
2992                 *lsmp = NULL;
2993                 RETURN(0);
2994         }
2995
2996         /* Alloc memmd */
2997         if (lsm == NULL && lmm == NULL) {
2998                 lsm_size = lmv_stripe_md_size(stripe_count);
2999                 OBD_ALLOC(lsm, lsm_size);
3000                 if (lsm == NULL)
3001                         RETURN(-ENOMEM);
3002                 lsm->lsm_md_stripe_count = stripe_count;
3003                 *lsmp = lsm;
3004                 RETURN(0);
3005         }
3006
3007         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3008                 RETURN(-EPERM);
3009
3010         /* Unpack memmd */
3011         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3012             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3013                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3014                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3015                        -EIO);
3016                 RETURN(-EIO);
3017         }
3018
3019         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3020                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3021         else
3022                 /**
3023                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3024                  * stripecount should be 0 then.
3025                  */
3026                 lsm_size = lmv_stripe_md_size(0);
3027
3028         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3029         if (lsm == NULL) {
3030                 OBD_ALLOC(lsm, lsm_size);
3031                 if (lsm == NULL)
3032                         RETURN(-ENOMEM);
3033                 allocated = true;
3034                 *lsmp = lsm;
3035         }
3036
3037         switch (le32_to_cpu(lmm->lmv_magic)) {
3038         case LMV_MAGIC_V1:
3039                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3040                 break;
3041         default:
3042                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3043                        le32_to_cpu(lmm->lmv_magic));
3044                 rc = -EINVAL;
3045                 break;
3046         }
3047
3048         if (rc != 0 && allocated) {
3049                 OBD_FREE(lsm, lsm_size);
3050                 *lsmp = NULL;
3051                 lsm_size = rc;
3052         }
3053         RETURN(lsm_size);
3054 }
3055
3056 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3057 {
3058         return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3059 }
3060 EXPORT_SYMBOL(lmv_alloc_memmd);
3061
3062 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3063 {
3064         lmv_unpack_md(NULL, &lsm, NULL, 0);
3065 }
3066 EXPORT_SYMBOL(lmv_free_memmd);
3067
3068 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3069                  struct lov_mds_md *lmm, int disk_len)
3070 {
3071         return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3072                              (union lmv_mds_md *)lmm, disk_len);
3073 }
3074
3075 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3076                struct lov_stripe_md *lsm)
3077 {
3078         struct obd_device               *obd = exp->exp_obd;
3079         struct lmv_obd                  *lmv_obd = &obd->u.lmv;
3080         const struct lmv_stripe_md      *lmv = (struct lmv_stripe_md *)lsm;
3081         int                             stripe_count;
3082
3083         if (lmmp == NULL) {
3084                 if (lsm != NULL)
3085                         stripe_count = lmv->lsm_md_stripe_count;
3086                 else
3087                         stripe_count = lmv_obd->desc.ld_tgt_count;
3088
3089                 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3090         }
3091
3092         return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3093 }
3094
3095 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3096                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
3097                              ldlm_cancel_flags_t flags, void *opaque)
3098 {
3099         struct obd_device       *obd = exp->exp_obd;
3100         struct lmv_obd          *lmv = &obd->u.lmv;
3101         int                      rc = 0;
3102         int                      err;
3103         __u32                    i;
3104         ENTRY;
3105
3106         LASSERT(fid != NULL);
3107
3108         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3109                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3110
3111                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3112                         continue;
3113
3114                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3115                                        opaque);
3116                 if (!rc)
3117                         rc = err;
3118         }
3119         RETURN(rc);
3120 }
3121
3122 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3123                       __u64 *bits)
3124 {
3125         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3126         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3127         int                      rc;
3128         ENTRY;
3129
3130         if (tgt == NULL || tgt->ltd_exp == NULL)
3131                 RETURN(-EINVAL);
3132         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3133         RETURN(rc);
3134 }
3135
3136 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3137                            const struct lu_fid *fid, ldlm_type_t type,
3138                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
3139                            struct lustre_handle *lockh)
3140 {
3141         struct obd_device       *obd = exp->exp_obd;
3142         struct lmv_obd          *lmv = &obd->u.lmv;
3143         ldlm_mode_t             rc;
3144         int                     tgt;
3145         int                     i;
3146         ENTRY;
3147
3148         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3149
3150         /*
3151          * With DNE every object can have two locks in different namespaces:
3152          * lookup lock in space of MDT storing direntry and update/open lock in
3153          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3154          * since this can be easily found, and only try others if that fails.
3155          */
3156         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3157              i < lmv->desc.ld_tgt_count;
3158              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3159                 if (tgt < 0) {
3160                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3161                                obd->obd_name, PFID(fid), tgt);
3162                         tgt = 0;
3163                 }
3164
3165                 if (lmv->tgts[tgt] == NULL ||
3166                     lmv->tgts[tgt]->ltd_exp == NULL ||
3167                     lmv->tgts[tgt]->ltd_active == 0)
3168                         continue;
3169
3170                 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3171                                    type, policy, mode, lockh);
3172                 if (rc)
3173                         RETURN(rc);
3174         }
3175
3176         RETURN(0);
3177 }
3178
3179 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3180                       struct obd_export *dt_exp, struct obd_export *md_exp,
3181                       struct lustre_md *md)
3182 {
3183         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3184         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3185
3186         if (tgt == NULL || tgt->ltd_exp == NULL)
3187                 RETURN(-EINVAL);
3188
3189         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3190 }
3191
3192 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3193 {
3194         struct obd_device       *obd = exp->exp_obd;
3195         struct lmv_obd          *lmv = &obd->u.lmv;
3196         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3197         ENTRY;
3198
3199         if (md->lmv != NULL) {
3200                 lmv_free_memmd(md->lmv);
3201                 md->lmv = NULL;
3202         }
3203         if (tgt == NULL || tgt->ltd_exp == NULL)
3204                 RETURN(-EINVAL);
3205         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3206 }
3207
3208 int lmv_set_open_replay_data(struct obd_export *exp,
3209                              struct obd_client_handle *och,
3210                              struct lookup_intent *it)
3211 {
3212         struct obd_device       *obd = exp->exp_obd;
3213         struct lmv_obd          *lmv = &obd->u.lmv;
3214         struct lmv_tgt_desc     *tgt;
3215         ENTRY;
3216
3217         tgt = lmv_find_target(lmv, &och->och_fid);
3218         if (IS_ERR(tgt))
3219                 RETURN(PTR_ERR(tgt));
3220
3221         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3222 }
3223
3224 int lmv_clear_open_replay_data(struct obd_export *exp,
3225                                struct obd_client_handle *och)
3226 {
3227         struct obd_device       *obd = exp->exp_obd;
3228         struct lmv_obd          *lmv = &obd->u.lmv;
3229         struct lmv_tgt_desc     *tgt;
3230         ENTRY;
3231
3232         tgt = lmv_find_target(lmv, &och->och_fid);
3233         if (IS_ERR(tgt))
3234                 RETURN(PTR_ERR(tgt));
3235
3236         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3237 }
3238
3239 static int lmv_get_remote_perm(struct obd_export *exp,
3240                                const struct lu_fid *fid,
3241                                struct obd_capa *oc, __u32 suppgid,
3242                                struct ptlrpc_request **request)
3243 {
3244         struct obd_device       *obd = exp->exp_obd;
3245         struct lmv_obd          *lmv = &obd->u.lmv;
3246         struct lmv_tgt_desc     *tgt;
3247         int                      rc;
3248         ENTRY;
3249
3250         rc = lmv_check_connect(obd);
3251         if (rc)
3252                 RETURN(rc);
3253
3254         tgt = lmv_find_target(lmv, fid);
3255         if (IS_ERR(tgt))
3256                 RETURN(PTR_ERR(tgt));
3257
3258         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3259         RETURN(rc);
3260 }
3261
3262 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3263                           renew_capa_cb_t cb)
3264 {
3265         struct obd_device       *obd = exp->exp_obd;
3266         struct lmv_obd          *lmv = &obd->u.lmv;
3267         struct lmv_tgt_desc     *tgt;
3268         int                      rc;
3269         ENTRY;
3270
3271         rc = lmv_check_connect(obd);
3272         if (rc)
3273                 RETURN(rc);
3274
3275         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3276         if (IS_ERR(tgt))
3277                 RETURN(PTR_ERR(tgt));
3278
3279         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3280         RETURN(rc);
3281 }
3282
3283 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3284                     const struct req_msg_field *field, struct obd_capa **oc)
3285 {
3286         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3287         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3288
3289         if (tgt == NULL || tgt->ltd_exp == NULL)
3290                 RETURN(-EINVAL);
3291         return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3292 }
3293
3294 int lmv_intent_getattr_async(struct obd_export *exp,
3295                              struct md_enqueue_info *minfo,
3296                              struct ldlm_enqueue_info *einfo)
3297 {
3298         struct md_op_data       *op_data = &minfo->mi_data;
3299         struct obd_device       *obd = exp->exp_obd;
3300         struct lmv_obd          *lmv = &obd->u.lmv;
3301         struct lmv_tgt_desc     *tgt = NULL;
3302         int                      rc;
3303         ENTRY;
3304
3305         rc = lmv_check_connect(obd);
3306         if (rc)
3307                 RETURN(rc);
3308
3309         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3310         if (IS_ERR(tgt))
3311                 RETURN(PTR_ERR(tgt));
3312
3313         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3314         RETURN(rc);
3315 }
3316
3317 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3318                         struct lu_fid *fid, __u64 *bits)
3319 {
3320         struct obd_device       *obd = exp->exp_obd;
3321         struct lmv_obd          *lmv = &obd->u.lmv;
3322         struct lmv_tgt_desc     *tgt;
3323         int                      rc;
3324         ENTRY;
3325
3326         rc = lmv_check_connect(obd);
3327         if (rc)
3328                 RETURN(rc);
3329
3330         tgt = lmv_find_target(lmv, fid);
3331         if (IS_ERR(tgt))
3332                 RETURN(PTR_ERR(tgt));
3333
3334         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3335         RETURN(rc);
3336 }
3337
3338 int lmv_get_fid_from_lsm(struct obd_export *exp,
3339                          const struct lmv_stripe_md *lsm,
3340                          const char *name, int namelen, struct lu_fid *fid)
3341 {
3342         const struct lmv_oinfo *oinfo;
3343
3344         LASSERT(lsm != NULL);
3345         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3346         if (IS_ERR(oinfo))
3347                 return PTR_ERR(oinfo);
3348
3349         *fid = oinfo->lmo_fid;
3350
3351         RETURN(0);
3352 }
3353
3354 /**
3355  * For lmv, only need to send request to master MDT, and the master MDT will
3356  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3357  * we directly fetch data from the slave MDTs.
3358  */
3359 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3360                  struct obd_quotactl *oqctl)
3361 {
3362         struct obd_device   *obd = class_exp2obd(exp);
3363         struct lmv_obd      *lmv = &obd->u.lmv;
3364         struct lmv_tgt_desc *tgt = lmv->tgts[0];
3365         int                  rc = 0;
3366         __u32                i;
3367         __u64                curspace, curinodes;
3368         ENTRY;
3369
3370         if (tgt == NULL ||
3371             tgt->ltd_exp == NULL ||
3372             !tgt->ltd_active ||
3373             lmv->desc.ld_tgt_count == 0) {
3374                 CERROR("master lmv inactive\n");
3375                 RETURN(-EIO);
3376         }
3377
3378         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3379                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3380                 RETURN(rc);
3381         }
3382
3383         curspace = curinodes = 0;
3384         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3385                 int err;
3386                 tgt = lmv->tgts[i];
3387
3388                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3389                         continue;
3390
3391                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3392                 if (err) {
3393                         CERROR("getquota on mdt %d failed. %d\n", i, err);
3394                         if (!rc)
3395                                 rc = err;
3396                 } else {
3397                         curspace += oqctl->qc_dqblk.dqb_curspace;
3398                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3399                 }
3400         }
3401         oqctl->qc_dqblk.dqb_curspace = curspace;
3402         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3403
3404         RETURN(rc);
3405 }
3406
3407 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3408                    struct obd_quotactl *oqctl)
3409 {
3410         struct obd_device       *obd = class_exp2obd(exp);
3411         struct lmv_obd          *lmv = &obd->u.lmv;
3412         struct lmv_tgt_desc     *tgt;
3413         __u32                    i;
3414         int                      rc = 0;
3415         ENTRY;
3416
3417         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3418                 int err;
3419                 tgt = lmv->tgts[i];
3420                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3421                         CERROR("lmv idx %d inactive\n", i);
3422                         RETURN(-EIO);
3423                 }
3424
3425                 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3426                 if (err && !rc)
3427                         rc = err;
3428         }
3429
3430         RETURN(rc);
3431 }
3432
3433 int lmv_update_lsm_md(struct obd_export *exp, struct lmv_stripe_md *lsm,
3434                       struct mdt_body *body, ldlm_blocking_callback cb_blocking)
3435 {
3436         return lmv_revalidate_slaves(exp, body, lsm, cb_blocking, 0);
3437 }
3438
3439 int lmv_merge_attr(struct obd_export *exp, const struct lmv_stripe_md *lsm,
3440                    struct cl_attr *attr)
3441 {
3442         int i;
3443
3444         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3445                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3446
3447                 CDEBUG(D_INFO, ""DFID" size %llu, nlink %u, atime %lu ctime"
3448                        "%lu, mtime %lu.\n", PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3449                        i_size_read(inode), inode->i_nlink,
3450                        LTIME_S(inode->i_atime), LTIME_S(inode->i_ctime),
3451                        LTIME_S(inode->i_mtime));
3452
3453                 /* for slave stripe, it needs to subtract nlink for . and .. */
3454                 if (i != 0)
3455                         attr->cat_nlink += inode->i_nlink - 2;
3456                 else
3457                         attr->cat_nlink = inode->i_nlink;
3458
3459                 attr->cat_size += i_size_read(inode);
3460
3461                 if (attr->cat_atime < LTIME_S(inode->i_atime))
3462                         attr->cat_atime = LTIME_S(inode->i_atime);
3463
3464                 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3465                         attr->cat_ctime = LTIME_S(inode->i_ctime);
3466
3467                 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3468                         attr->cat_mtime = LTIME_S(inode->i_mtime);
3469         }
3470         return 0;
3471 }
3472
3473 struct obd_ops lmv_obd_ops = {
3474         .o_owner                = THIS_MODULE,
3475         .o_setup                = lmv_setup,
3476         .o_cleanup              = lmv_cleanup,
3477         .o_precleanup           = lmv_precleanup,
3478         .o_process_config       = lmv_process_config,
3479         .o_connect              = lmv_connect,
3480         .o_disconnect           = lmv_disconnect,
3481         .o_statfs               = lmv_statfs,
3482         .o_get_info             = lmv_get_info,
3483         .o_set_info_async       = lmv_set_info_async,
3484         .o_packmd               = lmv_packmd,
3485         .o_unpackmd             = lmv_unpackmd,
3486         .o_notify               = lmv_notify,
3487         .o_get_uuid             = lmv_get_uuid,
3488         .o_iocontrol            = lmv_iocontrol,
3489         .o_quotacheck           = lmv_quotacheck,
3490         .o_quotactl             = lmv_quotactl
3491 };
3492
3493 struct md_ops lmv_md_ops = {
3494         .m_getstatus            = lmv_getstatus,
3495         .m_null_inode           = lmv_null_inode,
3496         .m_find_cbdata          = lmv_find_cbdata,
3497         .m_close                = lmv_close,
3498         .m_create               = lmv_create,
3499         .m_done_writing         = lmv_done_writing,
3500         .m_enqueue              = lmv_enqueue,
3501         .m_getattr              = lmv_getattr,
3502         .m_getxattr             = lmv_getxattr,
3503         .m_getattr_name         = lmv_getattr_name,
3504         .m_intent_lock          = lmv_intent_lock,
3505         .m_link                 = lmv_link,
3506         .m_rename               = lmv_rename,
3507         .m_setattr              = lmv_setattr,
3508         .m_setxattr             = lmv_setxattr,
3509         .m_fsync                = lmv_fsync,
3510         .m_read_page            = lmv_read_page,
3511         .m_unlink               = lmv_unlink,
3512         .m_init_ea_size         = lmv_init_ea_size,
3513         .m_cancel_unused        = lmv_cancel_unused,
3514         .m_set_lock_data        = lmv_set_lock_data,
3515         .m_lock_match           = lmv_lock_match,
3516         .m_get_lustre_md        = lmv_get_lustre_md,
3517         .m_free_lustre_md       = lmv_free_lustre_md,
3518         .m_update_lsm_md        = lmv_update_lsm_md,
3519         .m_merge_attr           = lmv_merge_attr,
3520         .m_set_open_replay_data = lmv_set_open_replay_data,
3521         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3522         .m_renew_capa           = lmv_renew_capa,
3523         .m_unpack_capa          = lmv_unpack_capa,
3524         .m_get_remote_perm      = lmv_get_remote_perm,
3525         .m_intent_getattr_async = lmv_intent_getattr_async,
3526         .m_revalidate_lock      = lmv_revalidate_lock,
3527         .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
3528 };
3529
3530 int __init lmv_init(void)
3531 {
3532         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3533                                    LUSTRE_LMV_NAME, NULL);
3534 }
3535
3536 static void lmv_exit(void)
3537 {
3538         class_unregister_type(LUSTRE_LMV_NAME);
3539 }
3540
3541 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3542 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3543 MODULE_LICENSE("GPL");
3544
3545 module_init(lmv_init);
3546 module_exit(lmv_exit);