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, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LMV
  38 #include <linux/slab.h>
  39 #include <linux/module.h>
  40 #include <linux/init.h>
  41 #include <linux/user_namespace.h>
  42 #ifdef HAVE_UIDGID_HEADER
  43 # include <linux/uidgid.h>
  44 #endif
  45 #include <linux/slab.h>
  46 #include <linux/pagemap.h>
  47 #include <linux/mm.h>
  48 #include <linux/math64.h>
  49 #include <linux/seq_file.h>
  50 #include <linux/namei.h>
  51
  52 #include <lustre/lustre_idl.h>
  53 #include <obd_support.h>
  54 #include <lustre_lib.h>
  55 #include <lustre_net.h>
  56 #include <obd_class.h>
  57 #include <lustre_lmv.h>
  58 #include <lprocfs_status.h>
  59 #include <cl_object.h>
  60 #include <lustre_fid.h>
  61 #include <lustre_ioctl.h>
  62 #include <lustre_kernelcomm.h>
  63 #include "lmv_internal.h"
  64
  65 static void lmv_activate_target(struct lmv_obd *lmv,
  66                                 struct lmv_tgt_desc *tgt,
  67                                 int activate)
  68 {
  69         if (tgt->ltd_active == activate)
  70                 return;
  71
  72         tgt->ltd_active = activate;
  73         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  74
  75         tgt->ltd_exp->exp_obd->obd_inactive = !activate;
  76 }
  77
  78 /**
  79  * Error codes:
  80  *
  81  *  -EINVAL  : UUID can't be found in the LMV's target list
  82  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  83  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  84  */
  85 static int lmv_set_mdc_active(struct lmv_obd *lmv,
  86                               const struct obd_uuid *uuid,
  87                               int activate)
  88 {
  89         struct lmv_tgt_desc     *tgt = NULL;
  90         struct obd_device       *obd;
  91         __u32                    i;
  92         int                      rc = 0;
  93         ENTRY;
  94
  95         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  96                         lmv, uuid->uuid, activate);
  97
  98         spin_lock(&lmv->lmv_lock);
  99         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 100                 tgt = lmv->tgts[i];
 101                 if (tgt == NULL || tgt->ltd_exp == NULL)
 102                         continue;
 103
 104                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
 105                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 106
 107                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 108                         break;
 109         }
 110
 111         if (i == lmv->desc.ld_tgt_count)
 112                 GOTO(out_lmv_lock, rc = -EINVAL);
 113
 114         obd = class_exp2obd(tgt->ltd_exp);
 115         if (obd == NULL)
 116                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 117
 118         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 119                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 120                obd->obd_type->typ_name, i);
 121         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 122
 123         if (tgt->ltd_active == activate) {
 124                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 125                        activate ? "" : "in");
 126                 GOTO(out_lmv_lock, rc);
 127         }
 128
 129         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 130                activate ? "" : "in");
 131         lmv_activate_target(lmv, tgt, activate);
 132         EXIT;
 133
 134  out_lmv_lock:
 135         spin_unlock(&lmv->lmv_lock);
 136         return rc;
 137 }
 138
 139 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 140 {
 141         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
 142         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
 143
 144         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
 145 }
 146
 147 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 148                       enum obd_notify_event ev, void *data)
 149 {
 150         struct obd_connect_data *conn_data;
 151         struct lmv_obd          *lmv = &obd->u.lmv;
 152         struct obd_uuid         *uuid;
 153         int                      rc = 0;
 154         ENTRY;
 155
 156         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 157                 CERROR("unexpected notification of %s %s!\n",
 158                        watched->obd_type->typ_name,
 159                        watched->obd_name);
 160                 RETURN(-EINVAL);
 161         }
 162
 163         uuid = &watched->u.cli.cl_target_uuid;
 164         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 165                 /*
 166                  * Set MDC as active before notifying the observer, so the
 167                  * observer can use the MDC normally.
 168                  */
 169                 rc = lmv_set_mdc_active(lmv, uuid,
 170                                         ev == OBD_NOTIFY_ACTIVE);
 171                 if (rc) {
 172                         CERROR("%sactivation of %s failed: %d\n",
 173                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 174                                uuid->uuid, rc);
 175                         RETURN(rc);
 176                 }
 177         } else if (ev == OBD_NOTIFY_OCD) {
 178                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 179                 /*
 180                  * XXX: Make sure that ocd_connect_flags from all targets are
 181                  * the same. Otherwise one of MDTs runs wrong version or
 182                  * something like this.  --umka
 183                  */
 184                 obd->obd_self_export->exp_connect_data = *conn_data;
 185         }
 186 #if 0
 187         else if (ev == OBD_NOTIFY_DISCON) {
 188                 /*
 189                  * For disconnect event, flush fld cache for failout MDS case.
 190                  */
 191                 fld_client_flush(&lmv->lmv_fld);
 192         }
 193 #endif
 194         /*
 195          * Pass the notification up the chain.
 196          */
 197         if (obd->obd_observer)
 198                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 199
 200         RETURN(rc);
 201 }
 202
 203 /**
 204  * This is fake connect function. Its purpose is to initialize lmv and say
 205  * caller that everything is okay. Real connection will be performed later.
 206  */
 207 static int lmv_connect(const struct lu_env *env,
 208                        struct obd_export **exp, struct obd_device *obd,
 209                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 210                        void *localdata)
 211 {
 212         struct lmv_obd        *lmv = &obd->u.lmv;
 213         struct lustre_handle  conn = { 0 };
 214         int                    rc = 0;
 215         ENTRY;
 216
 217         /*
 218          * We don't want to actually do the underlying connections more than
 219          * once, so keep track.
 220          */
 221         lmv->refcount++;
 222         if (lmv->refcount > 1) {
 223                 *exp = NULL;
 224                 RETURN(0);
 225         }
 226
 227         rc = class_connect(&conn, obd, cluuid);
 228         if (rc) {
 229                 CERROR("class_connection() returned %d\n", rc);
 230                 RETURN(rc);
 231         }
 232
 233         *exp = class_conn2export(&conn);
 234         class_export_get(*exp);
 235
 236         lmv->exp = *exp;
 237         lmv->connected = 0;
 238         lmv->cluuid = *cluuid;
 239
 240         if (data)
 241                 lmv->conn_data = *data;
 242
 243         if (lmv->targets_proc_entry == NULL) {
 244                 lmv->targets_proc_entry = lprocfs_register("target_obds",
 245                                                            obd->obd_proc_entry,
 246                                                            NULL, NULL);
 247                 if (IS_ERR(lmv->targets_proc_entry)) {
 248                         CERROR("%s: cannot register "
 249                                "/proc/fs/lustre/%s/%s/target_obds\n",
 250                                obd->obd_name, obd->obd_type->typ_name,
 251                                obd->obd_name);
 252                         lmv->targets_proc_entry = NULL;
 253                 }
 254         }
 255
 256         /*
 257          * All real clients should perform actual connection right away, because
 258          * it is possible, that LMV will not have opportunity to connect targets
 259          * and MDC stuff will be called directly, for instance while reading
 260          * ../mdc/../kbytesfree procfs file, etc.
 261          */
 262         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
 263                 rc = lmv_check_connect(obd);
 264
 265         if (rc && lmv->targets_proc_entry != NULL)
 266                 lprocfs_remove(&lmv->targets_proc_entry);
 267         RETURN(rc);
 268 }
 269
 270 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
 271                             __u32 def_easize)
 272 {
 273         struct obd_device       *obd = exp->exp_obd;
 274         struct lmv_obd          *lmv = &obd->u.lmv;
 275         __u32                    i;
 276         int                      rc = 0;
 277         int                      change = 0;
 278         ENTRY;
 279
 280         if (lmv->max_easize < easize) {
 281                 lmv->max_easize = easize;
 282                 change = 1;
 283         }
 284         if (lmv->max_def_easize < def_easize) {
 285                 lmv->max_def_easize = def_easize;
 286                 change = 1;
 287         }
 288
 289         if (change == 0)
 290                 RETURN(0);
 291
 292         if (lmv->connected == 0)
 293                 RETURN(0);
 294
 295         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 296                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 297
 298                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
 299                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 300                         continue;
 301                 }
 302
 303                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
 304                 if (rc) {
 305                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 306                                " rc = %d\n", obd->obd_name, i, rc);
 307                         break;
 308                 }
 309         }
 310         RETURN(rc);
 311 }
 312
 313 #define MAX_STRING_SIZE 128
 314
 315 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 316 {
 317         struct lmv_obd          *lmv = &obd->u.lmv;
 318         struct obd_uuid         *cluuid = &lmv->cluuid;
 319         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 320         struct obd_device       *mdc_obd;
 321         struct obd_export       *mdc_exp;
 322         struct lu_fld_target     target;
 323         int                      rc;
 324         ENTRY;
 325
 326         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 327                                         &obd->obd_uuid);
 328         if (!mdc_obd) {
 329                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 330                 RETURN(-EINVAL);
 331         }
 332
 333         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 334                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 335                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 336                 cluuid->uuid);
 337
 338         if (!mdc_obd->obd_set_up) {
 339                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 340                 RETURN(-EINVAL);
 341         }
 342
 343         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 344                          &lmv->conn_data, NULL);
 345         if (rc) {
 346                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 347                 RETURN(rc);
 348         }
 349
 350         /*
 351          * Init fid sequence client for this mdc and add new fld target.
 352          */
 353         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 354         if (rc)
 355                 RETURN(rc);
 356
 357         target.ft_srv = NULL;
 358         target.ft_exp = mdc_exp;
 359         target.ft_idx = tgt->ltd_idx;
 360
 361         fld_client_add_target(&lmv->lmv_fld, &target);
 362
 363         rc = obd_register_observer(mdc_obd, obd);
 364         if (rc) {
 365                 obd_disconnect(mdc_exp);
 366                 CERROR("target %s register_observer error %d\n",
 367                        tgt->ltd_uuid.uuid, rc);
 368                 RETURN(rc);
 369         }
 370
 371         if (obd->obd_observer) {
 372                 /*
 373                  * Tell the observer about the new target.
 374                  */
 375                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 376                                 OBD_NOTIFY_ACTIVE,
 377                                 (void *)(tgt - lmv->tgts[0]));
 378                 if (rc) {
 379                         obd_disconnect(mdc_exp);
 380                         RETURN(rc);
 381                 }
 382         }
 383
 384         tgt->ltd_active = 1;
 385         tgt->ltd_exp = mdc_exp;
 386         lmv->desc.ld_active_tgt_count++;
 387
 388         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
 389
 390         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 391                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 392                 atomic_read(&obd->obd_refcount));
 393
 394         if (lmv->targets_proc_entry != NULL) {
 395                 struct proc_dir_entry *mdc_symlink;
 396
 397                 LASSERT(mdc_obd->obd_type != NULL);
 398                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 399                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 400                                                   lmv->targets_proc_entry,
 401                                                   "../../../%s/%s",
 402                                                   mdc_obd->obd_type->typ_name,
 403                                                   mdc_obd->obd_name);
 404                 if (mdc_symlink == NULL) {
 405                         CERROR("cannot register LMV target "
 406                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
 407                                obd->obd_type->typ_name, obd->obd_name,
 408                                mdc_obd->obd_name);
 409                 }
 410         }
 411         RETURN(0);
 412 }
 413
 414 static void lmv_del_target(struct lmv_obd *lmv, int index)
 415 {
 416         if (lmv->tgts[index] == NULL)
 417                 return;
 418
 419         OBD_FREE_PTR(lmv->tgts[index]);
 420         lmv->tgts[index] = NULL;
 421         return;
 422 }
 423
 424 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 425                            __u32 index, int gen)
 426 {
 427         struct obd_device *mdc_obd;
 428         struct lmv_obd      *lmv = &obd->u.lmv;
 429         struct lmv_tgt_desc *tgt;
 430         int                  orig_tgt_count = 0;
 431         int                  rc = 0;
 432         ENTRY;
 433
 434         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 435         mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 436                                         &obd->obd_uuid);
 437         if (!mdc_obd) {
 438                 CERROR("%s: Target %s not attached: rc = %d\n",
 439                        obd->obd_name, uuidp->uuid, -EINVAL);
 440                 RETURN(-EINVAL);
 441         }
 442
 443         mutex_lock(&lmv->lmv_init_mutex);
 444         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
 445                 tgt = lmv->tgts[index];
 446                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
 447                        " rc = %d\n", obd->obd_name,
 448                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
 449                 mutex_unlock(&lmv->lmv_init_mutex);
 450                 RETURN(-EEXIST);
 451         }
 452
 453         if (index >= lmv->tgts_size) {
 454                 /* We need to reallocate the lmv target array. */
 455                 struct lmv_tgt_desc **newtgts, **old = NULL;
 456                 __u32 newsize = 1;
 457                 __u32 oldsize = 0;
 458
 459                 while (newsize < index + 1)
 460                         newsize = newsize << 1;
 461                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 462                 if (newtgts == NULL) {
 463                         mutex_unlock(&lmv->lmv_init_mutex);
 464                         RETURN(-ENOMEM);
 465                 }
 466
 467                 if (lmv->tgts_size) {
 468                         memcpy(newtgts, lmv->tgts,
 469                                sizeof(*newtgts) * lmv->tgts_size);
 470                         old = lmv->tgts;
 471                         oldsize = lmv->tgts_size;
 472                 }
 473
 474                 lmv->tgts = newtgts;
 475                 lmv->tgts_size = newsize;
 476                 smp_rmb();
 477                 if (old)
 478                         OBD_FREE(old, sizeof(*old) * oldsize);
 479
 480                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
 481                        lmv->tgts_size);
 482         }
 483
 484         OBD_ALLOC_PTR(tgt);
 485         if (!tgt) {
 486                 mutex_unlock(&lmv->lmv_init_mutex);
 487                 RETURN(-ENOMEM);
 488         }
 489
 490         mutex_init(&tgt->ltd_fid_mutex);
 491         tgt->ltd_idx = index;
 492         tgt->ltd_uuid = *uuidp;
 493         tgt->ltd_active = 0;
 494         lmv->tgts[index] = tgt;
 495         if (index >= lmv->desc.ld_tgt_count) {
 496                 orig_tgt_count = lmv->desc.ld_tgt_count;
 497                 lmv->desc.ld_tgt_count = index + 1;
 498         }
 499
 500         if (lmv->connected == 0) {
 501                 /* lmv_check_connect() will connect this target. */
 502                 mutex_unlock(&lmv->lmv_init_mutex);
 503                 RETURN(0);
 504         }
 505
 506         /* Otherwise let's connect it ourselves */
 507         mutex_unlock(&lmv->lmv_init_mutex);
 508         rc = lmv_connect_mdc(obd, tgt);
 509         if (rc != 0) {
 510                 spin_lock(&lmv->lmv_lock);
 511                 if (lmv->desc.ld_tgt_count == index + 1)
 512                         lmv->desc.ld_tgt_count = orig_tgt_count;
 513                 memset(tgt, 0, sizeof(*tgt));
 514                 spin_unlock(&lmv->lmv_lock);
 515         } else {
 516                 int easize = sizeof(struct lmv_stripe_md) +
 517                         lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
 518                 lmv_init_ea_size(obd->obd_self_export, easize, 0);
 519         }
 520
 521         RETURN(rc);
 522 }
 523
 524 int lmv_check_connect(struct obd_device *obd)
 525 {
 526         struct lmv_obd          *lmv = &obd->u.lmv;
 527         struct lmv_tgt_desc     *tgt;
 528         __u32                    i;
 529         int                      rc;
 530         int                      easize;
 531         ENTRY;
 532
 533         if (lmv->connected)
 534                 RETURN(0);
 535
 536         mutex_lock(&lmv->lmv_init_mutex);
 537         if (lmv->connected) {
 538                 mutex_unlock(&lmv->lmv_init_mutex);
 539                 RETURN(0);
 540         }
 541
 542         if (lmv->desc.ld_tgt_count == 0) {
 543                 mutex_unlock(&lmv->lmv_init_mutex);
 544                 CERROR("%s: no targets configured.\n", obd->obd_name);
 545                 RETURN(-EINVAL);
 546         }
 547
 548         LASSERT(lmv->tgts != NULL);
 549
 550         if (lmv->tgts[0] == NULL) {
 551                 mutex_unlock(&lmv->lmv_init_mutex);
 552                 CERROR("%s: no target configured for index 0.\n",
 553                        obd->obd_name);
 554                 RETURN(-EINVAL);
 555         }
 556
 557         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 558                lmv->cluuid.uuid, obd->obd_name);
 559
 560         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 561                 tgt = lmv->tgts[i];
 562                 if (tgt == NULL)
 563                         continue;
 564                 rc = lmv_connect_mdc(obd, tgt);
 565                 if (rc)
 566                         GOTO(out_disc, rc);
 567         }
 568
 569         class_export_put(lmv->exp);
 570         lmv->connected = 1;
 571         easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
 572         lmv_init_ea_size(obd->obd_self_export, easize, 0);
 573         mutex_unlock(&lmv->lmv_init_mutex);
 574         RETURN(0);
 575
 576  out_disc:
 577         while (i-- > 0) {
 578                 int rc2;
 579                 tgt = lmv->tgts[i];
 580                 if (tgt == NULL)
 581                         continue;
 582                 tgt->ltd_active = 0;
 583                 if (tgt->ltd_exp) {
 584                         --lmv->desc.ld_active_tgt_count;
 585                         rc2 = obd_disconnect(tgt->ltd_exp);
 586                         if (rc2) {
 587                                 CERROR("LMV target %s disconnect on "
 588                                        "MDC idx %d: error %d\n",
 589                                        tgt->ltd_uuid.uuid, i, rc2);
 590                         }
 591                 }
 592         }
 593         class_disconnect(lmv->exp);
 594         mutex_unlock(&lmv->lmv_init_mutex);
 595         RETURN(rc);
 596 }
 597
 598 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 599 {
 600         struct lmv_obd         *lmv = &obd->u.lmv;
 601         struct obd_device      *mdc_obd;
 602         int                     rc;
 603         ENTRY;
 604
 605         LASSERT(tgt != NULL);
 606         LASSERT(obd != NULL);
 607
 608         mdc_obd = class_exp2obd(tgt->ltd_exp);
 609
 610         if (mdc_obd) {
 611                 mdc_obd->obd_force = obd->obd_force;
 612                 mdc_obd->obd_fail = obd->obd_fail;
 613                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 614
 615                 if (lmv->targets_proc_entry != NULL)
 616                         lprocfs_remove_proc_entry(mdc_obd->obd_name,
 617                                                   lmv->targets_proc_entry);
 618         }
 619
 620         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 621         if (rc)
 622                 CERROR("Can't finanize fids factory\n");
 623
 624         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 625                tgt->ltd_exp->exp_obd->obd_name,
 626                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 627
 628         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 629         rc = obd_disconnect(tgt->ltd_exp);
 630         if (rc) {
 631                 if (tgt->ltd_active) {
 632                         CERROR("Target %s disconnect error %d\n",
 633                                tgt->ltd_uuid.uuid, rc);
 634                 }
 635         }
 636
 637         lmv_activate_target(lmv, tgt, 0);
 638         tgt->ltd_exp = NULL;
 639         RETURN(0);
 640 }
 641
 642 static int lmv_disconnect(struct obd_export *exp)
 643 {
 644         struct obd_device       *obd = class_exp2obd(exp);
 645         struct lmv_obd          *lmv = &obd->u.lmv;
 646         int                      rc;
 647         __u32                    i;
 648         ENTRY;
 649
 650         if (!lmv->tgts)
 651                 goto out_local;
 652
 653         /*
 654          * Only disconnect the underlying layers on the final disconnect.
 655          */
 656         lmv->refcount--;
 657         if (lmv->refcount != 0)
 658                 goto out_local;
 659
 660         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 661                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
 662                         continue;
 663
 664                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
 665         }
 666
 667         if (lmv->targets_proc_entry != NULL)
 668                 lprocfs_remove(&lmv->targets_proc_entry);
 669         else
 670                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 671                        obd->obd_type->typ_name, obd->obd_name);
 672
 673 out_local:
 674         /*
 675          * This is the case when no real connection is established by
 676          * lmv_check_connect().
 677          */
 678         if (!lmv->connected)
 679                 class_export_put(exp);
 680         rc = class_disconnect(exp);
 681         if (lmv->refcount == 0)
 682                 lmv->connected = 0;
 683         RETURN(rc);
 684 }
 685
 686 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
 687                         void __user *uarg)
 688 {
 689         struct obd_device       *obddev = class_exp2obd(exp);
 690         struct lmv_obd          *lmv = &obddev->u.lmv;
 691         struct getinfo_fid2path *gf;
 692         struct lmv_tgt_desc     *tgt;
 693         struct getinfo_fid2path *remote_gf = NULL;
 694         struct lu_fid           root_fid;
 695         int                     remote_gf_size = 0;
 696         int                     rc;
 697
 698         gf = karg;
 699         tgt = lmv_find_target(lmv, &gf->gf_fid);
 700         if (IS_ERR(tgt))
 701                 RETURN(PTR_ERR(tgt));
 702
 703         root_fid = *gf->gf_u.gf_root_fid;
 704         LASSERT(fid_is_sane(&root_fid));
 705
 706 repeat_fid2path:
 707         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 708         if (rc != 0 && rc != -EREMOTE)
 709                 GOTO(out_fid2path, rc);
 710
 711         /* If remote_gf != NULL, it means just building the
 712          * path on the remote MDT, copy this path segement to gf */
 713         if (remote_gf != NULL) {
 714                 struct getinfo_fid2path *ori_gf;
 715                 char *ptr;
 716
 717                 ori_gf = (struct getinfo_fid2path *)karg;
 718                 if (strlen(ori_gf->gf_u.gf_path) +
 719                     strlen(gf->gf_u.gf_path) > ori_gf->gf_pathlen)
 720                         GOTO(out_fid2path, rc = -EOVERFLOW);
 721
 722                 ptr = ori_gf->gf_u.gf_path;
 723
 724                 memmove(ptr + strlen(gf->gf_u.gf_path) + 1, ptr,
 725                         strlen(ori_gf->gf_u.gf_path));
 726
 727                 strncpy(ptr, gf->gf_u.gf_path,
 728                         strlen(gf->gf_u.gf_path));
 729                 ptr += strlen(gf->gf_u.gf_path);
 730                 *ptr = '/';
 731         }
 732
 733         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
 734                tgt->ltd_exp->exp_obd->obd_name,
 735                gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 736                gf->gf_linkno);
 737
 738         if (rc == 0)
 739                 GOTO(out_fid2path, rc);
 740
 741         /* sigh, has to go to another MDT to do path building further */
 742         if (remote_gf == NULL) {
 743                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 744                 OBD_ALLOC(remote_gf, remote_gf_size);
 745                 if (remote_gf == NULL)
 746                         GOTO(out_fid2path, rc = -ENOMEM);
 747                 remote_gf->gf_pathlen = PATH_MAX;
 748         }
 749
 750         if (!fid_is_sane(&gf->gf_fid)) {
 751                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 752                        tgt->ltd_exp->exp_obd->obd_name,
 753                        PFID(&gf->gf_fid), -EINVAL);
 754                 GOTO(out_fid2path, rc = -EINVAL);
 755         }
 756
 757         tgt = lmv_find_target(lmv, &gf->gf_fid);
 758         if (IS_ERR(tgt))
 759                 GOTO(out_fid2path, rc = -EINVAL);
 760
 761         remote_gf->gf_fid = gf->gf_fid;
 762         remote_gf->gf_recno = -1;
 763         remote_gf->gf_linkno = -1;
 764         memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
 765         *remote_gf->gf_u.gf_root_fid = root_fid;
 766         gf = remote_gf;
 767         goto repeat_fid2path;
 768
 769 out_fid2path:
 770         if (remote_gf != NULL)
 771                 OBD_FREE(remote_gf, remote_gf_size);
 772         RETURN(rc);
 773 }
 774
 775 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 776                              const struct hsm_user_request *hur,
 777                              const struct lmv_tgt_desc *tgt_mds)
 778 {
 779         __u32                    i;
 780         int                      nr = 0;
 781         struct lmv_tgt_desc     *curr_tgt;
 782
 783         /* count how many requests must be sent to the given target */
 784         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 785                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
 786                 if (IS_ERR(curr_tgt))
 787                         RETURN(PTR_ERR(curr_tgt));
 788                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 789                         nr++;
 790         }
 791         return nr;
 792 }
 793
 794 static int lmv_hsm_req_build(struct lmv_obd *lmv,
 795                               struct hsm_user_request *hur_in,
 796                               const struct lmv_tgt_desc *tgt_mds,
 797                               struct hsm_user_request *hur_out)
 798 {
 799         __u32                    i, nr_out;
 800         struct lmv_tgt_desc     *curr_tgt;
 801
 802         /* build the hsm_user_request for the given target */
 803         hur_out->hur_request = hur_in->hur_request;
 804         nr_out = 0;
 805         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 806                 curr_tgt = lmv_find_target(lmv,
 807                                            &hur_in->hur_user_item[i].hui_fid);
 808                 if (IS_ERR(curr_tgt))
 809                         RETURN(PTR_ERR(curr_tgt));
 810                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 811                         hur_out->hur_user_item[nr_out] =
 812                                                 hur_in->hur_user_item[i];
 813                         nr_out++;
 814                 }
 815         }
 816         hur_out->hur_request.hr_itemcount = nr_out;
 817         memcpy(hur_data(hur_out), hur_data(hur_in),
 818                hur_in->hur_request.hr_data_len);
 819
 820         RETURN(0);
 821 }
 822
 823 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
 824                                  struct lustre_kernelcomm *lk,
 825                                  void __user *uarg)
 826 {
 827         __u32   i;
 828         int     rc;
 829         ENTRY;
 830
 831         /* unregister request (call from llapi_hsm_copytool_fini) */
 832         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 833                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 834
 835                 if (tgt == NULL || tgt->ltd_exp == NULL)
 836                         continue;
 837                 /* best effort: try to clean as much as possible
 838                  * (continue on error) */
 839                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 840         }
 841
 842         /* Whatever the result, remove copytool from kuc groups.
 843          * Unreached coordinators will get EPIPE on next requests
 844          * and will unregister automatically.
 845          */
 846         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group);
 847
 848         RETURN(rc);
 849 }
 850
 851 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
 852                                struct lustre_kernelcomm *lk, __user void *uarg)
 853 {
 854         struct file             *filp;
 855         __u32                    i, j;
 856         int                      err, rc;
 857         bool                     any_set = false;
 858         struct kkuc_ct_data      kcd = { 0 };
 859         ENTRY;
 860
 861         /* All or nothing: try to register to all MDS.
 862          * In case of failure, unregister from previous MDS,
 863          * except if it because of inactive target. */
 864         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 865                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 866
 867                 if (tgt == NULL || tgt->ltd_exp == NULL)
 868                         continue;
 869                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 870                 if (err) {
 871                         if (tgt->ltd_active) {
 872                                 /* permanent error */
 873                                 CERROR("%s: iocontrol MDC %s on MDT"
 874                                        " idx %d cmd %x: err = %d\n",
 875                                        class_exp2obd(lmv->exp)->obd_name,
 876                                        tgt->ltd_uuid.uuid, i, cmd, err);
 877                                 rc = err;
 878                                 lk->lk_flags |= LK_FLG_STOP;
 879                                 /* unregister from previous MDS */
 880                                 for (j = 0; j < i; j++) {
 881                                         tgt = lmv->tgts[j];
 882                                         if (tgt == NULL || tgt->ltd_exp == NULL)
 883                                                 continue;
 884                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
 885                                                       lk, uarg);
 886                                 }
 887                                 RETURN(rc);
 888                         }
 889                         /* else: transient error.
 890                          * kuc will register to the missing MDT
 891                          * when it is back */
 892                 } else {
 893                         any_set = true;
 894                 }
 895         }
 896
 897         if (!any_set)
 898                 /* no registration done: return error */
 899                 RETURN(-ENOTCONN);
 900
 901         /* at least one registration done, with no failure */
 902         filp = fget(lk->lk_wfd);
 903         if (filp == NULL)
 904                 RETURN(-EBADF);
 905
 906         kcd.kcd_magic = KKUC_CT_DATA_MAGIC;
 907         kcd.kcd_uuid = lmv->cluuid;
 908         kcd.kcd_archive = lk->lk_data;
 909
 910         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group,
 911                                    &kcd, sizeof(kcd));
 912         if (rc != 0)
 913                 fput(filp);
 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 __user *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                                 int nr, rc1;
1100                                 size_t reqlen;
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)
1109                                         RETURN(nr);
1110                                 if (nr == 0) /* nothing for this MDS */
1111                                         continue;
1112
1113                                 /* build a request with fids for this MDS */
1114                                 reqlen = offsetof(typeof(*hur),
1115                                                   hur_user_item[nr])
1116                                                 + hur->hur_request.hr_data_len;
1117                                 OBD_ALLOC_LARGE(req, reqlen);
1118                                 if (req == NULL)
1119                                         RETURN(-ENOMEM);
1120                                 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
1121                                 if (rc1 < 0)
1122                                         GOTO(hsm_req_err, rc1);
1123                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1124                                                     req, uarg);
1125 hsm_req_err:
1126                                 if (rc1 != 0 && rc == 0)
1127                                         rc = rc1;
1128                                 OBD_FREE_LARGE(req, reqlen);
1129                         }
1130                 }
1131                 break;
1132         }
1133         case LL_IOC_LOV_SWAP_LAYOUTS: {
1134                 struct md_op_data       *op_data = karg;
1135                 struct lmv_tgt_desc     *tgt1, *tgt2;
1136
1137                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1138                 if (IS_ERR(tgt1))
1139                         RETURN(PTR_ERR(tgt1));
1140
1141                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1142                 if (IS_ERR(tgt2))
1143                         RETURN(PTR_ERR(tgt2));
1144
1145                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1146                         RETURN(-EINVAL);
1147
1148                 /* only files on same MDT can have their layouts swapped */
1149                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1150                         RETURN(-EPERM);
1151
1152                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1153                 break;
1154         }
1155         case LL_IOC_HSM_CT_START: {
1156                 struct lustre_kernelcomm *lk = karg;
1157                 if (lk->lk_flags & LK_FLG_STOP)
1158                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1159                 else
1160                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1161                 break;
1162         }
1163         default:
1164                 for (i = 0; i < count; i++) {
1165                         struct obd_device *mdc_obd;
1166                         int err;
1167
1168                         tgt = lmv->tgts[i];
1169                         if (tgt == NULL || tgt->ltd_exp == NULL)
1170                                 continue;
1171                         /* ll_umount_begin() sets force flag but for lmv, not
1172                          * mdc. Let's pass it through */
1173                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1174                         mdc_obd->obd_force = obddev->obd_force;
1175                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1176                         if (err) {
1177                                 if (tgt->ltd_active) {
1178                                         CERROR("error: iocontrol MDC %s on MDT"
1179                                                " idx %d cmd %x: err = %d\n",
1180                                                tgt->ltd_uuid.uuid, i, cmd, err);
1181                                         if (!rc)
1182                                                 rc = err;
1183                                 }
1184                         } else
1185                                 set = 1;
1186                 }
1187                 if (!set && !rc)
1188                         rc = -EIO;
1189         }
1190         RETURN(rc);
1191 }
1192
1193 /**
1194  * This is _inode_ placement policy function (not name).
1195  */
1196 static int lmv_placement_policy(struct obd_device *obd,
1197                                 struct md_op_data *op_data, u32 *mds)
1198 {
1199         struct lmv_obd          *lmv = &obd->u.lmv;
1200         ENTRY;
1201
1202         LASSERT(mds != NULL);
1203
1204         if (lmv->desc.ld_tgt_count == 1) {
1205                 *mds = 0;
1206                 RETURN(0);
1207         }
1208
1209         if (op_data->op_default_stripe_offset != -1) {
1210                 *mds = op_data->op_default_stripe_offset;
1211                 RETURN(0);
1212         }
1213
1214         /**
1215          * If stripe_offset is provided during setdirstripe
1216          * (setdirstripe -i xx), xx MDS will be choosen.
1217          */
1218         if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1219                 struct lmv_user_md *lum;
1220
1221                 lum = op_data->op_data;
1222
1223                 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1224                         *mds = le32_to_cpu(lum->lum_stripe_offset);
1225                 } else {
1226                         /* -1 means default, which will be in the same MDT with
1227                          * the stripe */
1228                         *mds = op_data->op_mds;
1229                         lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1230                 }
1231         } else {
1232                 /* Allocate new fid on target according to operation type and
1233                  * parent home mds. */
1234                 *mds = op_data->op_mds;
1235         }
1236
1237         RETURN(0);
1238 }
1239
1240 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1241 {
1242         struct lmv_tgt_desc     *tgt;
1243         int                      rc;
1244         ENTRY;
1245
1246         tgt = lmv_get_target(lmv, mds, NULL);
1247         if (IS_ERR(tgt))
1248                 RETURN(PTR_ERR(tgt));
1249
1250         /*
1251          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1252          * on server that seq in new allocated fid is not yet known.
1253          */
1254         mutex_lock(&tgt->ltd_fid_mutex);
1255
1256         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1257                 GOTO(out, rc = -ENODEV);
1258
1259         /*
1260          * Asking underlying tgt layer to allocate new fid.
1261          */
1262         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1263         if (rc > 0) {
1264                 LASSERT(fid_is_sane(fid));
1265                 rc = 0;
1266         }
1267
1268         EXIT;
1269 out:
1270         mutex_unlock(&tgt->ltd_fid_mutex);
1271         return rc;
1272 }
1273
1274 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1275                   struct lu_fid *fid, struct md_op_data *op_data)
1276 {
1277         struct obd_device     *obd = class_exp2obd(exp);
1278         struct lmv_obd        *lmv = &obd->u.lmv;
1279         u32                    mds = 0;
1280         int                    rc;
1281         ENTRY;
1282
1283         LASSERT(op_data != NULL);
1284         LASSERT(fid != NULL);
1285
1286         rc = lmv_placement_policy(obd, op_data, &mds);
1287         if (rc) {
1288                 CERROR("Can't get target for allocating fid, "
1289                        "rc %d\n", rc);
1290                 RETURN(rc);
1291         }
1292
1293         rc = __lmv_fid_alloc(lmv, fid, mds);
1294         if (rc) {
1295                 CERROR("Can't alloc new fid, rc %d\n", rc);
1296                 RETURN(rc);
1297         }
1298
1299         RETURN(rc);
1300 }
1301
1302 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1303 {
1304         struct lmv_obd  *lmv = &obd->u.lmv;
1305         struct lmv_desc *desc;
1306         int             rc;
1307         ENTRY;
1308
1309         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1310                 CERROR("LMV setup requires a descriptor\n");
1311                 RETURN(-EINVAL);
1312         }
1313
1314         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1315         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1316                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1317                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1318                 RETURN(-EINVAL);
1319         }
1320
1321         lmv->tgts_size = 32U;
1322         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1323         if (lmv->tgts == NULL)
1324                 RETURN(-ENOMEM);
1325
1326         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1327         lmv->desc.ld_tgt_count = 0;
1328         lmv->desc.ld_active_tgt_count = 0;
1329         lmv->max_def_easize = 0;
1330         lmv->max_easize = 0;
1331
1332         spin_lock_init(&lmv->lmv_lock);
1333         mutex_init(&lmv->lmv_init_mutex);
1334
1335 #ifdef CONFIG_PROC_FS
1336         obd->obd_vars = lprocfs_lmv_obd_vars;
1337         lprocfs_obd_setup(obd);
1338         lprocfs_alloc_md_stats(obd, 0);
1339         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1340                                 0444, &lmv_proc_target_fops, obd);
1341         if (rc)
1342                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1343                       obd->obd_name, rc);
1344 #endif
1345         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1346                              LUSTRE_CLI_FLD_HASH_DHT);
1347         if (rc) {
1348                 CERROR("Can't init FLD, err %d\n", rc);
1349                 GOTO(out, rc);
1350         }
1351
1352         RETURN(0);
1353
1354 out:
1355         return rc;
1356 }
1357
1358 static int lmv_cleanup(struct obd_device *obd)
1359 {
1360         struct lmv_obd   *lmv = &obd->u.lmv;
1361         ENTRY;
1362
1363         fld_client_fini(&lmv->lmv_fld);
1364         if (lmv->tgts != NULL) {
1365                 int i;
1366                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1367                         if (lmv->tgts[i] == NULL)
1368                                 continue;
1369                         lmv_del_target(lmv, i);
1370                 }
1371                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1372                 lmv->tgts_size = 0;
1373         }
1374         RETURN(0);
1375 }
1376
1377 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1378 {
1379         struct lustre_cfg       *lcfg = buf;
1380         struct obd_uuid         obd_uuid;
1381         int                     gen;
1382         __u32                   index;
1383         int                     rc;
1384         ENTRY;
1385
1386         switch (lcfg->lcfg_command) {
1387         case LCFG_ADD_MDC:
1388                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1389                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1390                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1391                         GOTO(out, rc = -EINVAL);
1392
1393                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1394
1395                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1396                         GOTO(out, rc = -EINVAL);
1397                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1398                         GOTO(out, rc = -EINVAL);
1399                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1400                 GOTO(out, rc);
1401         default:
1402                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1403                 GOTO(out, rc = -EINVAL);
1404         }
1405 out:
1406         RETURN(rc);
1407 }
1408
1409 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1410                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1411 {
1412         struct obd_device       *obd = class_exp2obd(exp);
1413         struct lmv_obd          *lmv = &obd->u.lmv;
1414         struct obd_statfs       *temp;
1415         int                      rc = 0;
1416         __u32                    i;
1417         ENTRY;
1418
1419         rc = lmv_check_connect(obd);
1420         if (rc)
1421                 RETURN(rc);
1422
1423         OBD_ALLOC(temp, sizeof(*temp));
1424         if (temp == NULL)
1425                 RETURN(-ENOMEM);
1426
1427         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1428                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1429                         continue;
1430
1431                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1432                                 max_age, flags);
1433                 if (rc) {
1434                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1435                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1436                                rc);
1437                         GOTO(out_free_temp, rc);
1438                 }
1439
1440                 if (i == 0) {
1441                         *osfs = *temp;
1442                         /* If the statfs is from mount, it will needs
1443                          * retrieve necessary information from MDT0.
1444                          * i.e. mount does not need the merged osfs
1445                          * from all of MDT.
1446                          * And also clients can be mounted as long as
1447                          * MDT0 is in service*/
1448                         if (flags & OBD_STATFS_FOR_MDT0)
1449                                 GOTO(out_free_temp, rc);
1450                 } else {
1451                         osfs->os_bavail += temp->os_bavail;
1452                         osfs->os_blocks += temp->os_blocks;
1453                         osfs->os_ffree += temp->os_ffree;
1454                         osfs->os_files += temp->os_files;
1455                 }
1456         }
1457
1458         EXIT;
1459 out_free_temp:
1460         OBD_FREE(temp, sizeof(*temp));
1461         return rc;
1462 }
1463
1464 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1465                         struct lu_fid *fid)
1466 {
1467         struct obd_device    *obd = exp->exp_obd;
1468         struct lmv_obd       *lmv = &obd->u.lmv;
1469         int                   rc;
1470         ENTRY;
1471
1472         rc = lmv_check_connect(obd);
1473         if (rc)
1474                 RETURN(rc);
1475
1476         rc = md_get_root(lmv->tgts[0]->ltd_exp, fileset, fid);
1477         RETURN(rc);
1478 }
1479
1480 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1481                         u64 valid, const char *name,
1482                         const char *input, int input_size, int output_size,
1483                         int flags, struct ptlrpc_request **request)
1484 {
1485         struct obd_device      *obd = exp->exp_obd;
1486         struct lmv_obd         *lmv = &obd->u.lmv;
1487         struct lmv_tgt_desc    *tgt;
1488         int                     rc;
1489         ENTRY;
1490
1491         rc = lmv_check_connect(obd);
1492         if (rc)
1493                 RETURN(rc);
1494
1495         tgt = lmv_find_target(lmv, fid);
1496         if (IS_ERR(tgt))
1497                 RETURN(PTR_ERR(tgt));
1498
1499         rc = md_getxattr(tgt->ltd_exp, fid, valid, name, input,
1500                          input_size, output_size, flags, request);
1501
1502         RETURN(rc);
1503 }
1504
1505 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1506                         u64 valid, const char *name,
1507                         const char *input, int input_size, int output_size,
1508                         int flags, __u32 suppgid,
1509                         struct ptlrpc_request **request)
1510 {
1511         struct obd_device      *obd = exp->exp_obd;
1512         struct lmv_obd         *lmv = &obd->u.lmv;
1513         struct lmv_tgt_desc    *tgt;
1514         int                     rc;
1515         ENTRY;
1516
1517         rc = lmv_check_connect(obd);
1518         if (rc)
1519                 RETURN(rc);
1520
1521         tgt = lmv_find_target(lmv, fid);
1522         if (IS_ERR(tgt))
1523                 RETURN(PTR_ERR(tgt));
1524
1525         rc = md_setxattr(tgt->ltd_exp, fid, valid, name, input,
1526                          input_size, output_size, flags, suppgid,
1527                          request);
1528
1529         RETURN(rc);
1530 }
1531
1532 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1533                        struct ptlrpc_request **request)
1534 {
1535         struct obd_device       *obd = exp->exp_obd;
1536         struct lmv_obd          *lmv = &obd->u.lmv;
1537         struct lmv_tgt_desc     *tgt;
1538         int                      rc;
1539         ENTRY;
1540
1541         rc = lmv_check_connect(obd);
1542         if (rc)
1543                 RETURN(rc);
1544
1545         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1546         if (IS_ERR(tgt))
1547                 RETURN(PTR_ERR(tgt));
1548
1549         if (op_data->op_flags & MF_GET_MDT_IDX) {
1550                 op_data->op_mds = tgt->ltd_idx;
1551                 RETURN(0);
1552         }
1553
1554         rc = md_getattr(tgt->ltd_exp, op_data, request);
1555
1556         RETURN(rc);
1557 }
1558
1559 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1560 {
1561         struct obd_device   *obd = exp->exp_obd;
1562         struct lmv_obd      *lmv = &obd->u.lmv;
1563         __u32                i;
1564         int                  rc;
1565         ENTRY;
1566
1567         rc = lmv_check_connect(obd);
1568         if (rc)
1569                 RETURN(rc);
1570
1571         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1572
1573         /*
1574          * With DNE every object can have two locks in different namespaces:
1575          * lookup lock in space of MDT storing direntry and update/open lock in
1576          * space of MDT storing inode.
1577          */
1578         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1579                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1580                         continue;
1581                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1582         }
1583
1584         RETURN(0);
1585 }
1586
1587 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1588                      struct md_open_data *mod, struct ptlrpc_request **request)
1589 {
1590         struct obd_device     *obd = exp->exp_obd;
1591         struct lmv_obd        *lmv = &obd->u.lmv;
1592         struct lmv_tgt_desc   *tgt;
1593         int                    rc;
1594         ENTRY;
1595
1596         rc = lmv_check_connect(obd);
1597         if (rc)
1598                 RETURN(rc);
1599
1600         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1601         if (IS_ERR(tgt))
1602                 RETURN(PTR_ERR(tgt));
1603
1604         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1605         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1606         RETURN(rc);
1607 }
1608
1609 /**
1610  * Choosing the MDT by name or FID in @op_data.
1611  * For non-striped directory, it will locate MDT by fid.
1612  * For striped-directory, it will locate MDT by name. And also
1613  * it will reset op_fid1 with the FID of the choosen stripe.
1614  **/
1615 struct lmv_tgt_desc *
1616 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1617                            const char *name, int namelen, struct lu_fid *fid,
1618                            u32 *mds)
1619 {
1620         struct lmv_tgt_desc     *tgt;
1621         const struct lmv_oinfo  *oinfo;
1622
1623         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1624                 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1625                         RETURN(ERR_PTR(-EBADF));
1626                 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1627         } else {
1628                 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1629                 if (IS_ERR(oinfo))
1630                         RETURN(ERR_CAST(oinfo));
1631         }
1632
1633         if (fid != NULL)
1634                 *fid = oinfo->lmo_fid;
1635         if (mds != NULL)
1636                 *mds = oinfo->lmo_mds;
1637
1638         tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1639
1640         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", oinfo->lmo_mds,
1641                PFID(&oinfo->lmo_fid));
1642         return tgt;
1643 }
1644
1645 /**
1646  * Locate mds by fid or name
1647  *
1648  * For striped directory (lsm != NULL), it will locate the stripe
1649  * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1650  * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1651  * walk through all of stripes to locate the entry.
1652  *
1653  * For normal direcotry, it will locate MDS by FID directly.
1654  * \param[in] lmv       LMV device
1655  * \param[in] op_data   client MD stack parameters, name, namelen
1656  *                      mds_num etc.
1657  * \param[in] fid       object FID used to locate MDS.
1658  *
1659  * retval               pointer to the lmv_tgt_desc if succeed.
1660  *                      ERR_PTR(errno) if failed.
1661  */
1662 struct lmv_tgt_desc*
1663 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1664                struct lu_fid *fid)
1665 {
1666         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1667         struct lmv_tgt_desc     *tgt;
1668
1669         /* During creating VOLATILE file, it should honor the mdt
1670          * index if the file under striped dir is being restored, see
1671          * ct_restore(). */
1672         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1673             (int)op_data->op_mds != -1) {
1674                 int i;
1675                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1676                 if (IS_ERR(tgt))
1677                         return tgt;
1678
1679                 if (lsm != NULL) {
1680                         /* refill the right parent fid */
1681                         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1682                                 struct lmv_oinfo *oinfo;
1683
1684                                 oinfo = &lsm->lsm_md_oinfo[i];
1685                                 if (oinfo->lmo_mds == op_data->op_mds) {
1686                                         *fid = oinfo->lmo_fid;
1687                                         break;
1688                                 }
1689                         }
1690
1691                         if (i == lsm->lsm_md_stripe_count)
1692                                 *fid = lsm->lsm_md_oinfo[0].lmo_fid;
1693                 }
1694
1695                 return tgt;
1696         }
1697
1698         if (lsm == NULL || op_data->op_namelen == 0) {
1699                 tgt = lmv_find_target(lmv, fid);
1700                 if (IS_ERR(tgt))
1701                         return tgt;
1702
1703                 op_data->op_mds = tgt->ltd_idx;
1704                 return tgt;
1705         }
1706
1707         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1708                                           op_data->op_namelen, fid,
1709                                           &op_data->op_mds);
1710 }
1711
1712 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1713                 const void *data, size_t datalen, umode_t mode, uid_t uid,
1714                 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1715                 struct ptlrpc_request **request)
1716 {
1717         struct obd_device       *obd = exp->exp_obd;
1718         struct lmv_obd          *lmv = &obd->u.lmv;
1719         struct lmv_tgt_desc     *tgt;
1720         int                      rc;
1721         ENTRY;
1722
1723         rc = lmv_check_connect(obd);
1724         if (rc)
1725                 RETURN(rc);
1726
1727         if (!lmv->desc.ld_active_tgt_count)
1728                 RETURN(-EIO);
1729
1730         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1731         if (IS_ERR(tgt))
1732                 RETURN(PTR_ERR(tgt));
1733
1734         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1735                 (int)op_data->op_namelen, op_data->op_name,
1736                 PFID(&op_data->op_fid1), op_data->op_mds);
1737
1738         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1739         if (rc)
1740                 RETURN(rc);
1741         if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1742                 /* Send the create request to the MDT where the object
1743                  * will be located */
1744                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1745                 if (IS_ERR(tgt))
1746                         RETURN(PTR_ERR(tgt));
1747
1748                 op_data->op_mds = tgt->ltd_idx;
1749         } else {
1750                 CDEBUG(D_CONFIG, "Server doesn't support striped dirs\n");
1751         }
1752
1753         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1754                PFID(&op_data->op_fid2), op_data->op_mds);
1755
1756         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1757         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1758                        cap_effective, rdev, request);
1759         if (rc == 0) {
1760                 if (*request == NULL)
1761                         RETURN(rc);
1762                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1763         }
1764         RETURN(rc);
1765 }
1766
1767 static int
1768 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1769             const union ldlm_policy_data *policy,
1770             struct lookup_intent *it, struct md_op_data *op_data,
1771             struct lustre_handle *lockh, __u64 extra_lock_flags)
1772 {
1773         struct obd_device        *obd = exp->exp_obd;
1774         struct lmv_obd           *lmv = &obd->u.lmv;
1775         struct lmv_tgt_desc      *tgt;
1776         int                       rc;
1777         ENTRY;
1778
1779         rc = lmv_check_connect(obd);
1780         if (rc)
1781                 RETURN(rc);
1782
1783         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1784                LL_IT2STR(it), PFID(&op_data->op_fid1));
1785
1786         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1787         if (IS_ERR(tgt))
1788                 RETURN(PTR_ERR(tgt));
1789
1790         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1791                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1792
1793         rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1794                         extra_lock_flags);
1795
1796         RETURN(rc);
1797 }
1798
1799 static int
1800 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1801                  struct ptlrpc_request **preq)
1802 {
1803         struct ptlrpc_request   *req = NULL;
1804         struct obd_device       *obd = exp->exp_obd;
1805         struct lmv_obd          *lmv = &obd->u.lmv;
1806         struct lmv_tgt_desc     *tgt;
1807         struct mdt_body         *body;
1808         int                      rc;
1809         ENTRY;
1810
1811         rc = lmv_check_connect(obd);
1812         if (rc)
1813                 RETURN(rc);
1814
1815         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1816         if (IS_ERR(tgt))
1817                 RETURN(PTR_ERR(tgt));
1818
1819         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1820                 (int)op_data->op_namelen, op_data->op_name,
1821                 PFID(&op_data->op_fid1), tgt->ltd_idx);
1822
1823         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1824         if (rc != 0)
1825                 RETURN(rc);
1826
1827         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1828         LASSERT(body != NULL);
1829
1830         if (body->mbo_valid & OBD_MD_MDS) {
1831                 struct lu_fid rid = body->mbo_fid1;
1832                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1833                        PFID(&rid));
1834
1835                 tgt = lmv_find_target(lmv, &rid);
1836                 if (IS_ERR(tgt)) {
1837                         ptlrpc_req_finished(*preq);
1838                         preq = NULL;
1839                         RETURN(PTR_ERR(tgt));
1840                 }
1841
1842                 op_data->op_fid1 = rid;
1843                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1844                 op_data->op_namelen = 0;
1845                 op_data->op_name = NULL;
1846                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1847                 ptlrpc_req_finished(*preq);
1848                 *preq = req;
1849         }
1850
1851         RETURN(rc);
1852 }
1853
1854 #define md_op_data_fid(op_data, fl)                     \
1855         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1856          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1857          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1858          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1859          NULL)
1860
1861 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1862                             struct md_op_data *op_data, __u32 op_tgt,
1863                             enum ldlm_mode mode, int bits, int flag)
1864 {
1865         struct lu_fid *fid = md_op_data_fid(op_data, flag);
1866         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
1867         union ldlm_policy_data policy = { { 0 } };
1868         int rc = 0;
1869         ENTRY;
1870
1871         if (!fid_is_sane(fid))
1872                 RETURN(0);
1873
1874         if (tgt == NULL) {
1875                 tgt = lmv_find_target(lmv, fid);
1876                 if (IS_ERR(tgt))
1877                         RETURN(PTR_ERR(tgt));
1878         }
1879
1880         if (tgt->ltd_idx != op_tgt) {
1881                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1882                 policy.l_inodebits.bits = bits;
1883                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1884                                       mode, LCF_ASYNC, NULL);
1885         } else {
1886                 CDEBUG(D_INODE,
1887                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1888                        op_tgt, PFID(fid));
1889                 op_data->op_flags |= flag;
1890                 rc = 0;
1891         }
1892
1893         RETURN(rc);
1894 }
1895
1896 /*
1897  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1898  * op_data->op_fid2
1899  */
1900 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1901                     struct ptlrpc_request **request)
1902 {
1903         struct obd_device       *obd = exp->exp_obd;
1904         struct lmv_obd          *lmv = &obd->u.lmv;
1905         struct lmv_tgt_desc     *tgt;
1906         int                      rc;
1907         ENTRY;
1908
1909         rc = lmv_check_connect(obd);
1910         if (rc)
1911                 RETURN(rc);
1912
1913         LASSERT(op_data->op_namelen != 0);
1914
1915         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1916                PFID(&op_data->op_fid2), (int)op_data->op_namelen,
1917                op_data->op_name, PFID(&op_data->op_fid1));
1918
1919         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
1920         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
1921         op_data->op_cap = cfs_curproc_cap_pack();
1922         if (op_data->op_mea2 != NULL) {
1923                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
1924                 const struct lmv_oinfo  *oinfo;
1925
1926                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
1927                                                 op_data->op_namelen);
1928                 if (IS_ERR(oinfo))
1929                         RETURN(PTR_ERR(oinfo));
1930
1931                 op_data->op_fid2 = oinfo->lmo_fid;
1932         }
1933
1934         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1935         if (IS_ERR(tgt))
1936                 RETURN(PTR_ERR(tgt));
1937
1938         /*
1939          * Cancel UPDATE lock on child (fid1).
1940          */
1941         op_data->op_flags |= MF_MDC_CANCEL_FID2;
1942         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
1943                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
1944         if (rc != 0)
1945                 RETURN(rc);
1946
1947         rc = md_link(tgt->ltd_exp, op_data, request);
1948
1949         RETURN(rc);
1950 }
1951
1952 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
1953                       const char *old, size_t oldlen,
1954                       const char *new, size_t newlen,
1955                       struct ptlrpc_request **request)
1956 {
1957         struct obd_device       *obd = exp->exp_obd;
1958         struct lmv_obd          *lmv = &obd->u.lmv;
1959         struct lmv_tgt_desc     *src_tgt;
1960         struct lmv_tgt_desc     *tgt_tgt;
1961         struct obd_export       *target_exp;
1962         struct mdt_body         *body;
1963         int                     rc;
1964         ENTRY;
1965
1966         LASSERT(oldlen != 0);
1967
1968         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
1969                (int)oldlen, old, PFID(&op_data->op_fid1),
1970                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
1971                (int)newlen, new, PFID(&op_data->op_fid2),
1972                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
1973
1974         rc = lmv_check_connect(obd);
1975         if (rc)
1976                 RETURN(rc);
1977
1978         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
1979         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
1980         op_data->op_cap = cfs_curproc_cap_pack();
1981         if (op_data->op_cli_flags & CLI_MIGRATE) {
1982                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
1983                          PFID(&op_data->op_fid3));
1984
1985                 if (op_data->op_mea1 != NULL) {
1986                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1987                         struct lmv_tgt_desc     *tmp;
1988
1989                         /* Fix the parent fid for striped dir */
1990                         tmp = lmv_locate_target_for_name(lmv, lsm, old,
1991                                                          oldlen,
1992                                                          &op_data->op_fid1,
1993                                                          NULL);
1994                         if (IS_ERR(tmp))
1995                                 RETURN(PTR_ERR(tmp));
1996                 }
1997
1998                 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1999                 if (rc != 0)
2000                         RETURN(rc);
2001
2002                 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2003                 if (IS_ERR(src_tgt))
2004                         RETURN(PTR_ERR(src_tgt));
2005
2006                 target_exp = src_tgt->ltd_exp;
2007         } else {
2008                 if (op_data->op_mea1 != NULL) {
2009                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2010
2011                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2012                                                              oldlen,
2013                                                              &op_data->op_fid1,
2014                                                              &op_data->op_mds);
2015                 } else {
2016                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2017                 }
2018                 if (IS_ERR(src_tgt))
2019                         RETURN(PTR_ERR(src_tgt));
2020
2021
2022                 if (op_data->op_mea2 != NULL) {
2023                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
2024
2025                         tgt_tgt = lmv_locate_target_for_name(lmv, lsm, new,
2026                                                              newlen,
2027                                                              &op_data->op_fid2,
2028                                                              &op_data->op_mds);
2029                 } else {
2030                         tgt_tgt = lmv_find_target(lmv, &op_data->op_fid2);
2031
2032                 }
2033                 if (IS_ERR(tgt_tgt))
2034                         RETURN(PTR_ERR(tgt_tgt));
2035
2036                 target_exp = tgt_tgt->ltd_exp;
2037         }
2038
2039         /*
2040          * LOOKUP lock on src child (fid3) should also be cancelled for
2041          * src_tgt in mdc_rename.
2042          */
2043         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2044
2045         /*
2046          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2047          * own target.
2048          */
2049         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2050                               LCK_EX, MDS_INODELOCK_UPDATE,
2051                               MF_MDC_CANCEL_FID2);
2052
2053         if (rc != 0)
2054                 RETURN(rc);
2055         /*
2056          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2057          */
2058         if (fid_is_sane(&op_data->op_fid3)) {
2059                 struct lmv_tgt_desc *tgt;
2060
2061                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2062                 if (IS_ERR(tgt))
2063                         RETURN(PTR_ERR(tgt));
2064
2065                 /* Cancel LOOKUP lock on its parent */
2066                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2067                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2068                                       MF_MDC_CANCEL_FID3);
2069                 if (rc != 0)
2070                         RETURN(rc);
2071
2072                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2073                                       LCK_EX, MDS_INODELOCK_FULL,
2074                                       MF_MDC_CANCEL_FID3);
2075                 if (rc != 0)
2076                         RETURN(rc);
2077         }
2078
2079 retry_rename:
2080         /*
2081          * Cancel all the locks on tgt child (fid4).
2082          */
2083         if (fid_is_sane(&op_data->op_fid4)) {
2084                 struct lmv_tgt_desc *tgt;
2085
2086                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2087                                       LCK_EX, MDS_INODELOCK_FULL,
2088                                       MF_MDC_CANCEL_FID4);
2089                 if (rc != 0)
2090                         RETURN(rc);
2091
2092                 tgt = lmv_find_target(lmv, &op_data->op_fid4);
2093                 if (IS_ERR(tgt))
2094                         RETURN(PTR_ERR(tgt));
2095
2096                 /* Since the target child might be destroyed, and it might
2097                  * become orphan, and we can only check orphan on the local
2098                  * MDT right now, so we send rename request to the MDT where
2099                  * target child is located. If target child does not exist,
2100                  * then it will send the request to the target parent */
2101                 target_exp = tgt->ltd_exp;
2102         }
2103
2104         rc = md_rename(target_exp, op_data, old, oldlen, new, newlen,
2105                        request);
2106
2107         if (rc != 0 && rc != -EXDEV)
2108                 RETURN(rc);
2109
2110         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2111         if (body == NULL)
2112                 RETURN(-EPROTO);
2113
2114         /* Not cross-ref case, just get out of here. */
2115         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2116                 RETURN(rc);
2117
2118         CDEBUG(D_INODE, "%s: try rename to another MDT for "DFID"\n",
2119                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2120
2121         op_data->op_fid4 = body->mbo_fid1;
2122         ptlrpc_req_finished(*request);
2123         *request = NULL;
2124         goto retry_rename;
2125 }
2126
2127 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2128                        void *ea, size_t ealen, struct ptlrpc_request **request)
2129 {
2130         struct obd_device       *obd = exp->exp_obd;
2131         struct lmv_obd          *lmv = &obd->u.lmv;
2132         struct lmv_tgt_desc     *tgt;
2133         int                      rc = 0;
2134         ENTRY;
2135
2136         rc = lmv_check_connect(obd);
2137         if (rc)
2138                 RETURN(rc);
2139
2140         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2141                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2142
2143         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2144         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2145         if (IS_ERR(tgt))
2146                 RETURN(PTR_ERR(tgt));
2147
2148         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2149
2150         RETURN(rc);
2151 }
2152
2153 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2154                      struct ptlrpc_request **request)
2155 {
2156         struct obd_device       *obd = exp->exp_obd;
2157         struct lmv_obd          *lmv = &obd->u.lmv;
2158         struct lmv_tgt_desc     *tgt;
2159         int                      rc;
2160         ENTRY;
2161
2162         rc = lmv_check_connect(obd);
2163         if (rc != 0)
2164                 RETURN(rc);
2165
2166         tgt = lmv_find_target(lmv, fid);
2167         if (IS_ERR(tgt))
2168                 RETURN(PTR_ERR(tgt));
2169
2170         rc = md_fsync(tgt->ltd_exp, fid, request);
2171         RETURN(rc);
2172 }
2173
2174 /**
2175  * Get current minimum entry from striped directory
2176  *
2177  * This function will search the dir entry, whose hash value is the
2178  * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2179  * only being called for striped directory.
2180  *
2181  * \param[in] exp               export of LMV
2182  * \param[in] op_data           parameters transferred beween client MD stack
2183  *                              stripe_information will be included in this
2184  *                              parameter
2185  * \param[in] cb_op             ldlm callback being used in enqueue in
2186  *                              mdc_read_page
2187  * \param[in] hash_offset       the hash value, which is used to locate
2188  *                              minum(closet) dir entry
2189  * \param[in|out] stripe_offset the caller use this to indicate the stripe
2190  *                              index of last entry, so to avoid hash conflict
2191  *                              between stripes. It will also be used to
2192  *                              return the stripe index of current dir entry.
2193  * \param[in|out] entp          the minum entry and it also is being used
2194  *                              to input the last dir entry to resolve the
2195  *                              hash conflict
2196  *
2197  * \param[out] ppage            the page which holds the minum entry
2198  *
2199  * \retval                      = 0 get the entry successfully
2200  *                              negative errno (< 0) does not get the entry
2201  */
2202 static int lmv_get_min_striped_entry(struct obd_export *exp,
2203                                      struct md_op_data *op_data,
2204                                      struct md_callback *cb_op,
2205                                      __u64 hash_offset, int *stripe_offset,
2206                                      struct lu_dirent **entp,
2207                                      struct page **ppage)
2208 {
2209         struct obd_device       *obd = exp->exp_obd;
2210         struct lmv_obd          *lmv = &obd->u.lmv;
2211         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2212         struct lmv_tgt_desc     *tgt;
2213         int                     stripe_count;
2214         struct lu_dirent        *min_ent = NULL;
2215         struct page             *min_page = NULL;
2216         int                     min_idx = 0;
2217         int                     i;
2218         int                     rc = 0;
2219         ENTRY;
2220
2221         stripe_count = lsm->lsm_md_stripe_count;
2222         for (i = 0; i < stripe_count; i++) {
2223                 struct lu_dirent        *ent = NULL;
2224                 struct page             *page = NULL;
2225                 struct lu_dirpage       *dp;
2226                 __u64                   stripe_hash = hash_offset;
2227
2228                 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2229                 if (IS_ERR(tgt))
2230                         GOTO(out, rc = PTR_ERR(tgt));
2231
2232                 /* op_data will be shared by each stripe, so we need
2233                  * reset these value for each stripe */
2234                 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2235                 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2236                 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2237 next:
2238                 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2239                                   &page);
2240                 if (rc != 0)
2241                         GOTO(out, rc);
2242
2243                 dp = page_address(page);
2244                 for (ent = lu_dirent_start(dp); ent != NULL;
2245                      ent = lu_dirent_next(ent)) {
2246                         /* Skip dummy entry */
2247                         if (le16_to_cpu(ent->lde_namelen) == 0)
2248                                 continue;
2249
2250                         if (le64_to_cpu(ent->lde_hash) < hash_offset)
2251                                 continue;
2252
2253                         if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2254                             (*entp == ent || i < *stripe_offset))
2255                                 continue;
2256
2257                         /* skip . and .. for other stripes */
2258                         if (i != 0 &&
2259                             (strncmp(ent->lde_name, ".",
2260                                      le16_to_cpu(ent->lde_namelen)) == 0 ||
2261                              strncmp(ent->lde_name, "..",
2262                                      le16_to_cpu(ent->lde_namelen)) == 0))
2263                                 continue;
2264                         break;
2265                 }
2266
2267                 if (ent == NULL) {
2268                         stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2269
2270                         kunmap(page);
2271                         page_cache_release(page);
2272                         page = NULL;
2273
2274                         /* reach the end of current stripe, go to next stripe */
2275                         if (stripe_hash == MDS_DIR_END_OFF)
2276                                 continue;
2277                         else
2278                                 goto next;
2279                 }
2280
2281                 if (min_ent != NULL) {
2282                         if (le64_to_cpu(min_ent->lde_hash) >
2283                             le64_to_cpu(ent->lde_hash)) {
2284                                 min_ent = ent;
2285                                 kunmap(min_page);
2286                                 page_cache_release(min_page);
2287                                 min_idx = i;
2288                                 min_page = page;
2289                         } else {
2290                                 kunmap(page);
2291                                 page_cache_release(page);
2292                                 page = NULL;
2293                         }
2294                 } else {
2295                         min_ent = ent;
2296                         min_page = page;
2297                         min_idx = i;
2298                 }
2299         }
2300
2301 out:
2302         if (*ppage != NULL) {
2303                 kunmap(*ppage);
2304                 page_cache_release(*ppage);
2305         }
2306         *stripe_offset = min_idx;
2307         *entp = min_ent;
2308         *ppage = min_page;
2309         RETURN(rc);
2310 }
2311
2312 /**
2313  * Build dir entry page from a striped directory
2314  *
2315  * This function gets one entry by @offset from a striped directory. It will
2316  * read entries from all of stripes, and choose one closest to the required
2317  * offset(&offset). A few notes
2318  * 1. skip . and .. for non-zero stripes, because there can only have one .
2319  * and .. in a directory.
2320  * 2. op_data will be shared by all of stripes, instead of allocating new
2321  * one, so need to restore before reusing.
2322  * 3. release the entry page if that is not being chosen.
2323  *
2324  * \param[in] exp       obd export refer to LMV
2325  * \param[in] op_data   hold those MD parameters of read_entry
2326  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2327  * \param[out] ldp      the entry being read
2328  * \param[out] ppage    the page holding the entry. Note: because the entry
2329  *                      will be accessed in upper layer, so we need hold the
2330  *                      page until the usages of entry is finished, see
2331  *                      ll_dir_entry_next.
2332  *
2333  * retval               =0 if get entry successfully
2334  *                      <0 cannot get entry
2335  */
2336 static int lmv_read_striped_page(struct obd_export *exp,
2337                                  struct md_op_data *op_data,
2338                                  struct md_callback *cb_op,
2339                                  __u64 offset, struct page **ppage)
2340 {
2341         struct obd_device       *obd = exp->exp_obd;
2342         struct lu_fid           master_fid = op_data->op_fid1;
2343         struct inode            *master_inode = op_data->op_data;
2344         __u64                   hash_offset = offset;
2345         struct lu_dirpage       *dp;
2346         struct page             *min_ent_page = NULL;
2347         struct page             *ent_page = NULL;
2348         struct lu_dirent        *ent;
2349         void                    *area;
2350         int                     ent_idx = 0;
2351         struct lu_dirent        *min_ent = NULL;
2352         struct lu_dirent        *last_ent;
2353         size_t                  left_bytes;
2354         int                     rc;
2355         ENTRY;
2356
2357         rc = lmv_check_connect(obd);
2358         if (rc)
2359                 RETURN(rc);
2360
2361         /* Allocate a page and read entries from all of stripes and fill
2362          * the page by hash order */
2363         ent_page = alloc_page(GFP_KERNEL);
2364         if (ent_page == NULL)
2365                 RETURN(-ENOMEM);
2366
2367         /* Initialize the entry page */
2368         dp = kmap(ent_page);
2369         memset(dp, 0, sizeof(*dp));
2370         dp->ldp_hash_start = cpu_to_le64(offset);
2371         dp->ldp_flags |= LDF_COLLIDE;
2372
2373         area = dp + 1;
2374         left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2375         ent = area;
2376         last_ent = ent;
2377         do {
2378                 __u16   ent_size;
2379
2380                 /* Find the minum entry from all sub-stripes */
2381                 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2382                                                &ent_idx, &min_ent,
2383                                                &min_ent_page);
2384                 if (rc != 0)
2385                         GOTO(out, rc);
2386
2387                 /* If it can not get minum entry, it means it already reaches
2388                  * the end of this directory */
2389                 if (min_ent == NULL) {
2390                         last_ent->lde_reclen = 0;
2391                         hash_offset = MDS_DIR_END_OFF;
2392                         GOTO(out, rc);
2393                 }
2394
2395                 ent_size = le16_to_cpu(min_ent->lde_reclen);
2396
2397                 /* the last entry lde_reclen is 0, but it might not
2398                  * the end of this entry of this temporay entry */
2399                 if (ent_size == 0)
2400                         ent_size = lu_dirent_calc_size(
2401                                         le16_to_cpu(min_ent->lde_namelen),
2402                                         le32_to_cpu(min_ent->lde_attrs));
2403                 if (ent_size > left_bytes) {
2404                         last_ent->lde_reclen = cpu_to_le16(0);
2405                         hash_offset = le64_to_cpu(min_ent->lde_hash);
2406                         GOTO(out, rc);
2407                 }
2408
2409                 memcpy(ent, min_ent, ent_size);
2410
2411                 /* Replace . with master FID and Replace .. with the parent FID
2412                  * of master object */
2413                 if (strncmp(ent->lde_name, ".",
2414                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2415                     le16_to_cpu(ent->lde_namelen) == 1)
2416                         fid_cpu_to_le(&ent->lde_fid, &master_fid);
2417                 else if (strncmp(ent->lde_name, "..",
2418                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2419                            le16_to_cpu(ent->lde_namelen) == 2)
2420                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2421
2422                 left_bytes -= ent_size;
2423                 ent->lde_reclen = cpu_to_le16(ent_size);
2424                 last_ent = ent;
2425                 ent = (void *)ent + ent_size;
2426                 hash_offset = le64_to_cpu(min_ent->lde_hash);
2427                 if (hash_offset == MDS_DIR_END_OFF) {
2428                         last_ent->lde_reclen = 0;
2429                         break;
2430                 }
2431         } while (1);
2432 out:
2433         if (min_ent_page != NULL) {
2434                 kunmap(min_ent_page);
2435                 page_cache_release(min_ent_page);
2436         }
2437
2438         if (unlikely(rc != 0)) {
2439                 __free_page(ent_page);
2440                 ent_page = NULL;
2441         } else {
2442                 if (ent == area)
2443                         dp->ldp_flags |= LDF_EMPTY;
2444                 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2445                 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2446         }
2447
2448         /* We do not want to allocate md_op_data during each
2449          * dir entry reading, so op_data will be shared by every stripe,
2450          * then we need to restore it back to original value before
2451          * return to the upper layer */
2452         op_data->op_fid1 = master_fid;
2453         op_data->op_fid2 = master_fid;
2454         op_data->op_data = master_inode;
2455
2456         *ppage = ent_page;
2457
2458         RETURN(rc);
2459 }
2460
2461 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2462                   struct md_callback *cb_op, __u64 offset,
2463                   struct page **ppage)
2464 {
2465         struct obd_device       *obd = exp->exp_obd;
2466         struct lmv_obd          *lmv = &obd->u.lmv;
2467         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2468         struct lmv_tgt_desc     *tgt;
2469         int                     rc;
2470         ENTRY;
2471
2472         rc = lmv_check_connect(obd);
2473         if (rc != 0)
2474                 RETURN(rc);
2475
2476         if (unlikely(lsm != NULL)) {
2477                 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2478                 RETURN(rc);
2479         }
2480
2481         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2482         if (IS_ERR(tgt))
2483                 RETURN(PTR_ERR(tgt));
2484
2485         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2486
2487         RETURN(rc);
2488 }
2489
2490 /**
2491  * Unlink a file/directory
2492  *
2493  * Unlink a file or directory under the parent dir. The unlink request
2494  * usually will be sent to the MDT where the child is located, but if
2495  * the client does not have the child FID then request will be sent to the
2496  * MDT where the parent is located.
2497  *
2498  * If the parent is a striped directory then it also needs to locate which
2499  * stripe the name of the child is located, and replace the parent FID
2500  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2501  * it will walk through all of sub-stripes until the child is being
2502  * unlinked finally.
2503  *
2504  * \param[in] exp       export refer to LMV
2505  * \param[in] op_data   different parameters transferred beween client
2506  *                      MD stacks, name, namelen, FIDs etc.
2507  *                      op_fid1 is the parent FID, op_fid2 is the child
2508  *                      FID.
2509  * \param[out] request  point to the request of unlink.
2510  *
2511  * retval               0 if succeed
2512  *                      negative errno if failed.
2513  */
2514 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2515                       struct ptlrpc_request **request)
2516 {
2517         struct obd_device       *obd = exp->exp_obd;
2518         struct lmv_obd          *lmv = &obd->u.lmv;
2519         struct lmv_tgt_desc     *tgt = NULL;
2520         struct lmv_tgt_desc     *parent_tgt = NULL;
2521         struct mdt_body         *body;
2522         int                     rc;
2523         int                     stripe_index = 0;
2524         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2525         ENTRY;
2526
2527         rc = lmv_check_connect(obd);
2528         if (rc)
2529                 RETURN(rc);
2530 retry_unlink:
2531         /* For striped dir, we need to locate the parent as well */
2532         if (lsm != NULL) {
2533                 struct lmv_tgt_desc *tmp;
2534
2535                 LASSERT(op_data->op_name != NULL &&
2536                         op_data->op_namelen != 0);
2537
2538                 tmp = lmv_locate_target_for_name(lmv, lsm,
2539                                                  op_data->op_name,
2540                                                  op_data->op_namelen,
2541                                                  &op_data->op_fid1,
2542                                                  &op_data->op_mds);
2543
2544                 /* return -EBADFD means unknown hash type, might
2545                  * need try all sub-stripe here */
2546                 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2547                         RETURN(PTR_ERR(tmp));
2548
2549                 /* Note: both migrating dir and unknown hash dir need to
2550                  * try all of sub-stripes, so we need start search the
2551                  * name from stripe 0, but migrating dir is already handled
2552                  * inside lmv_locate_target_for_name(), so we only check
2553                  * unknown hash type directory here */
2554                 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2555                         struct lmv_oinfo *oinfo;
2556
2557                         oinfo = &lsm->lsm_md_oinfo[stripe_index];
2558
2559                         op_data->op_fid1 = oinfo->lmo_fid;
2560                         op_data->op_mds = oinfo->lmo_mds;
2561                 }
2562         }
2563
2564 try_next_stripe:
2565         /* Send unlink requests to the MDT where the child is located */
2566         if (likely(!fid_is_zero(&op_data->op_fid2)))
2567                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2568         else if (lsm != NULL)
2569                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2570         else
2571                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2572
2573         if (IS_ERR(tgt))
2574                 RETURN(PTR_ERR(tgt));
2575
2576         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2577         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2578         op_data->op_cap = cfs_curproc_cap_pack();
2579
2580         /*
2581          * If child's fid is given, cancel unused locks for it if it is from
2582          * another export than parent.
2583          *
2584          * LOOKUP lock for child (fid3) should also be cancelled on parent
2585          * tgt_tgt in mdc_unlink().
2586          */
2587         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2588
2589         /*
2590          * Cancel FULL locks on child (fid3).
2591          */
2592         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2593         if (IS_ERR(parent_tgt))
2594                 RETURN(PTR_ERR(parent_tgt));
2595
2596         if (parent_tgt != tgt) {
2597                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2598                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2599                                       MF_MDC_CANCEL_FID3);
2600         }
2601
2602         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2603                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2604         if (rc != 0)
2605                 RETURN(rc);
2606
2607         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2608                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2609
2610         rc = md_unlink(tgt->ltd_exp, op_data, request);
2611         if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2612                 RETURN(rc);
2613
2614         /* Try next stripe if it is needed. */
2615         if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2616                 struct lmv_oinfo *oinfo;
2617
2618                 stripe_index++;
2619                 if (stripe_index >= lsm->lsm_md_stripe_count)
2620                         RETURN(rc);
2621
2622                 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2623
2624                 op_data->op_fid1 = oinfo->lmo_fid;
2625                 op_data->op_mds = oinfo->lmo_mds;
2626
2627                 ptlrpc_req_finished(*request);
2628                 *request = NULL;
2629
2630                 goto try_next_stripe;
2631         }
2632
2633         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2634         if (body == NULL)
2635                 RETURN(-EPROTO);
2636
2637         /* Not cross-ref case, just get out of here. */
2638         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2639                 RETURN(rc);
2640
2641         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2642                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2643
2644         /* This is a remote object, try remote MDT, Note: it may
2645          * try more than 1 time here, Considering following case
2646          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2647          * 1. Initially A does not know where remote1 is, it send
2648          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2649          *    resend unlink RPC to MDT1 (retry 1st time).
2650          *
2651          * 2. During the unlink RPC in flight,
2652          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2653          *    and create new remote1, but on MDT0
2654          *
2655          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2656          *    /mnt/lustre, then lookup get fid of remote1, and find
2657          *    it is remote dir again, and replay -EREMOTE again.
2658          *
2659          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2660          *
2661          * In theory, it might try unlimited time here, but it should
2662          * be very rare case.  */
2663         op_data->op_fid2 = body->mbo_fid1;
2664         ptlrpc_req_finished(*request);
2665         *request = NULL;
2666
2667         goto retry_unlink;
2668 }
2669
2670 static int lmv_precleanup(struct obd_device *obd)
2671 {
2672         ENTRY;
2673         fld_client_proc_fini(&obd->u.lmv.lmv_fld);
2674         lprocfs_obd_cleanup(obd);
2675         lprocfs_free_md_stats(obd);
2676         RETURN(0);
2677 }
2678
2679 /**
2680  * Get by key a value associated with a LMV device.
2681  *
2682  * Dispatch request to lower-layer devices as needed.
2683  *
2684  * \param[in] env               execution environment for this thread
2685  * \param[in] exp               export for the LMV device
2686  * \param[in] keylen            length of key identifier
2687  * \param[in] key               identifier of key to get value for
2688  * \param[in] vallen            size of \a val
2689  * \param[out] val              pointer to storage location for value
2690  * \param[in] lsm               optional striping metadata of object
2691  *
2692  * \retval 0            on success
2693  * \retval negative     negated errno on failure
2694  */
2695 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2696                         __u32 keylen, void *key, __u32 *vallen, void *val)
2697 {
2698         struct obd_device       *obd;
2699         struct lmv_obd          *lmv;
2700         int                      rc = 0;
2701         ENTRY;
2702
2703         obd = class_exp2obd(exp);
2704         if (obd == NULL) {
2705                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2706                        exp->exp_handle.h_cookie);
2707                 RETURN(-EINVAL);
2708         }
2709
2710         lmv = &obd->u.lmv;
2711         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2712                 int i;
2713
2714                 rc = lmv_check_connect(obd);
2715                 if (rc)
2716                         RETURN(rc);
2717
2718                 LASSERT(*vallen == sizeof(__u32));
2719                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2720                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2721                         /*
2722                          * All tgts should be connected when this gets called.
2723                          */
2724                         if (tgt == NULL || tgt->ltd_exp == NULL)
2725                                 continue;
2726
2727                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2728                                           vallen, val))
2729                                 RETURN(0);
2730                 }
2731                 RETURN(-EINVAL);
2732         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2733                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2734                    KEY_IS(KEY_CONN_DATA)) {
2735                 rc = lmv_check_connect(obd);
2736                 if (rc)
2737                         RETURN(rc);
2738
2739                 /*
2740                  * Forwarding this request to first MDS, it should know LOV
2741                  * desc.
2742                  */
2743                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2744                                   vallen, val);
2745                 if (!rc && KEY_IS(KEY_CONN_DATA))
2746                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2747                 RETURN(rc);
2748         } else if (KEY_IS(KEY_TGT_COUNT)) {
2749                 *((int *)val) = lmv->desc.ld_tgt_count;
2750                 RETURN(0);
2751         }
2752
2753         CDEBUG(D_IOCTL, "Invalid key\n");
2754         RETURN(-EINVAL);
2755 }
2756
2757 /**
2758  * Asynchronously set by key a value associated with a LMV device.
2759  *
2760  * Dispatch request to lower-layer devices as needed.
2761  *
2762  * \param[in] env       execution environment for this thread
2763  * \param[in] exp       export for the LMV device
2764  * \param[in] keylen    length of key identifier
2765  * \param[in] key       identifier of key to store value for
2766  * \param[in] vallen    size of value to store
2767  * \param[in] val       pointer to data to be stored
2768  * \param[in] set       optional list of related ptlrpc requests
2769  *
2770  * \retval 0            on success
2771  * \retval negative     negated errno on failure
2772  */
2773 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2774                         __u32 keylen, void *key, __u32 vallen, void *val,
2775                         struct ptlrpc_request_set *set)
2776 {
2777         struct lmv_tgt_desc     *tgt = NULL;
2778         struct obd_device       *obd;
2779         struct lmv_obd          *lmv;
2780         int rc = 0;
2781         ENTRY;
2782
2783         obd = class_exp2obd(exp);
2784         if (obd == NULL) {
2785                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2786                        exp->exp_handle.h_cookie);
2787                 RETURN(-EINVAL);
2788         }
2789         lmv = &obd->u.lmv;
2790
2791         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2792             KEY_IS(KEY_DEFAULT_EASIZE)) {
2793                 int i, err = 0;
2794
2795                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2796                         tgt = lmv->tgts[i];
2797
2798                         if (tgt == NULL || tgt->ltd_exp == NULL)
2799                                 continue;
2800
2801                         err = obd_set_info_async(env, tgt->ltd_exp,
2802                                                  keylen, key, vallen, val, set);
2803                         if (err && rc == 0)
2804                                 rc = err;
2805                 }
2806
2807                 RETURN(rc);
2808         }
2809
2810         RETURN(-EINVAL);
2811 }
2812
2813 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2814                             const struct lmv_mds_md_v1 *lmm1)
2815 {
2816         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
2817         int             stripe_count;
2818         int             cplen;
2819         int             i;
2820         int             rc = 0;
2821         ENTRY;
2822
2823         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2824         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2825         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2826         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2827                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2828         else
2829                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2830         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2831         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2832                         sizeof(lsm->lsm_md_pool_name));
2833
2834         if (cplen >= sizeof(lsm->lsm_md_pool_name))
2835                 RETURN(-E2BIG);
2836
2837         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2838                "layout_version %d\n", lsm->lsm_md_stripe_count,
2839                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2840                lsm->lsm_md_layout_version);
2841
2842         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2843         for (i = 0; i < stripe_count; i++) {
2844                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2845                               &lmm1->lmv_stripe_fids[i]);
2846                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2847                                     &lsm->lsm_md_oinfo[i].lmo_mds);
2848                 if (rc != 0)
2849                         RETURN(rc);
2850                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2851                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2852         }
2853
2854         RETURN(rc);
2855 }
2856
2857 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2858                         const union lmv_mds_md *lmm, size_t lmm_size)
2859 {
2860         struct lmv_stripe_md     *lsm;
2861         int                      lsm_size;
2862         int                      rc;
2863         bool                     allocated = false;
2864         ENTRY;
2865
2866         LASSERT(lsmp != NULL);
2867
2868         lsm = *lsmp;
2869         /* Free memmd */
2870         if (lsm != NULL && lmm == NULL) {
2871                 int i;
2872                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2873                         /* For migrating inode, the master stripe and master
2874                          * object will be the same, so do not need iput, see
2875                          * ll_update_lsm_md */
2876                         if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2877                               i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2878                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
2879                 }
2880                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2881                 OBD_FREE(lsm, lsm_size);
2882                 *lsmp = NULL;
2883                 RETURN(0);
2884         }
2885
2886         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
2887                 RETURN(-EPERM);
2888
2889         /* Unpack memmd */
2890         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
2891             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
2892                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
2893                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
2894                        -EIO);
2895                 RETURN(-EIO);
2896         }
2897
2898         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
2899                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2900         else
2901                 /**
2902                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
2903                  * stripecount should be 0 then.
2904                  */
2905                 lsm_size = lmv_stripe_md_size(0);
2906
2907         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2908         if (lsm == NULL) {
2909                 OBD_ALLOC(lsm, lsm_size);
2910                 if (lsm == NULL)
2911                         RETURN(-ENOMEM);
2912                 allocated = true;
2913                 *lsmp = lsm;
2914         }
2915
2916         switch (le32_to_cpu(lmm->lmv_magic)) {
2917         case LMV_MAGIC_V1:
2918                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
2919                 break;
2920         default:
2921                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
2922                        le32_to_cpu(lmm->lmv_magic));
2923                 rc = -EINVAL;
2924                 break;
2925         }
2926
2927         if (rc != 0 && allocated) {
2928                 OBD_FREE(lsm, lsm_size);
2929                 *lsmp = NULL;
2930                 lsm_size = rc;
2931         }
2932         RETURN(lsm_size);
2933 }
2934
2935 void lmv_free_memmd(struct lmv_stripe_md *lsm)
2936 {
2937         lmv_unpackmd(NULL, &lsm, NULL, 0);
2938 }
2939 EXPORT_SYMBOL(lmv_free_memmd);
2940
2941 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2942                              union ldlm_policy_data *policy,
2943                              enum ldlm_mode mode, enum ldlm_cancel_flags flags,
2944                              void *opaque)
2945 {
2946         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2947         int rc = 0;
2948         __u32 i;
2949         ENTRY;
2950
2951         LASSERT(fid != NULL);
2952
2953         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2954                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2955                 int err;
2956
2957                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
2958                         continue;
2959
2960                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
2961                                        opaque);
2962                 if (!rc)
2963                         rc = err;
2964         }
2965         RETURN(rc);
2966 }
2967
2968 static int lmv_set_lock_data(struct obd_export *exp,
2969                              const struct lustre_handle *lockh,
2970                              void *data, __u64 *bits)
2971 {
2972         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2973         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
2974         int                      rc;
2975         ENTRY;
2976
2977         if (tgt == NULL || tgt->ltd_exp == NULL)
2978                 RETURN(-EINVAL);
2979         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
2980         RETURN(rc);
2981 }
2982
2983 enum ldlm_mode lmv_lock_match(struct obd_export *exp, __u64 flags,
2984                               const struct lu_fid *fid, enum ldlm_type type,
2985                               union ldlm_policy_data *policy,
2986                               enum ldlm_mode mode, struct lustre_handle *lockh)
2987 {
2988         struct obd_device       *obd = exp->exp_obd;
2989         struct lmv_obd          *lmv = &obd->u.lmv;
2990         enum ldlm_mode          rc;
2991         int                     tgt;
2992         int                     i;
2993         ENTRY;
2994
2995         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2996
2997         /*
2998          * With DNE every object can have two locks in different namespaces:
2999          * lookup lock in space of MDT storing direntry and update/open lock in
3000          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3001          * since this can be easily found, and only try others if that fails.
3002          */
3003         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3004              i < lmv->desc.ld_tgt_count;
3005              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3006                 if (tgt < 0) {
3007                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3008                                obd->obd_name, PFID(fid), tgt);
3009                         tgt = 0;
3010                 }
3011
3012                 if (lmv->tgts[tgt] == NULL ||
3013                     lmv->tgts[tgt]->ltd_exp == NULL ||
3014                     lmv->tgts[tgt]->ltd_active == 0)
3015                         continue;
3016
3017                 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3018                                    type, policy, mode, lockh);
3019                 if (rc)
3020                         RETURN(rc);
3021         }
3022
3023         RETURN(0);
3024 }
3025
3026 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3027                       struct obd_export *dt_exp, struct obd_export *md_exp,
3028                       struct lustre_md *md)
3029 {
3030         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3031         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3032
3033         if (tgt == NULL || tgt->ltd_exp == NULL)
3034                 RETURN(-EINVAL);
3035
3036         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3037 }
3038
3039 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3040 {
3041         struct obd_device       *obd = exp->exp_obd;
3042         struct lmv_obd          *lmv = &obd->u.lmv;
3043         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3044         ENTRY;
3045
3046         if (md->lmv != NULL) {
3047                 lmv_free_memmd(md->lmv);
3048                 md->lmv = NULL;
3049         }
3050         if (tgt == NULL || tgt->ltd_exp == NULL)
3051                 RETURN(-EINVAL);
3052         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3053 }
3054
3055 int lmv_set_open_replay_data(struct obd_export *exp,
3056                              struct obd_client_handle *och,
3057                              struct lookup_intent *it)
3058 {
3059         struct obd_device       *obd = exp->exp_obd;
3060         struct lmv_obd          *lmv = &obd->u.lmv;
3061         struct lmv_tgt_desc     *tgt;
3062         ENTRY;
3063
3064         tgt = lmv_find_target(lmv, &och->och_fid);
3065         if (IS_ERR(tgt))
3066                 RETURN(PTR_ERR(tgt));
3067
3068         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3069 }
3070
3071 int lmv_clear_open_replay_data(struct obd_export *exp,
3072                                struct obd_client_handle *och)
3073 {
3074         struct obd_device       *obd = exp->exp_obd;
3075         struct lmv_obd          *lmv = &obd->u.lmv;
3076         struct lmv_tgt_desc     *tgt;
3077         ENTRY;
3078
3079         tgt = lmv_find_target(lmv, &och->och_fid);
3080         if (IS_ERR(tgt))
3081                 RETURN(PTR_ERR(tgt));
3082
3083         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3084 }
3085
3086 int lmv_intent_getattr_async(struct obd_export *exp,
3087                              struct md_enqueue_info *minfo)
3088 {
3089         struct md_op_data       *op_data = &minfo->mi_data;
3090         struct obd_device       *obd = exp->exp_obd;
3091         struct lmv_obd          *lmv = &obd->u.lmv;
3092         struct lmv_tgt_desc     *ptgt = NULL;
3093         struct lmv_tgt_desc     *ctgt = NULL;
3094         int                      rc;
3095         ENTRY;
3096
3097         if (!fid_is_sane(&op_data->op_fid2))
3098                 RETURN(-EINVAL);
3099
3100         rc = lmv_check_connect(obd);
3101         if (rc)
3102                 RETURN(rc);
3103
3104         ptgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3105         if (IS_ERR(ptgt))
3106                 RETURN(PTR_ERR(ptgt));
3107
3108         ctgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
3109         if (IS_ERR(ctgt))
3110                 RETURN(PTR_ERR(ctgt));
3111
3112         /*
3113          * if child is on remote MDT, we need 2 async RPCs to fetch both LOOKUP
3114          * lock on parent, and UPDATE lock on child MDT, which makes all
3115          * complicated. Considering remote dir is rare case, and not supporting
3116          * it in statahead won't cause any issue, drop its support for now.
3117          */
3118         if (ptgt != ctgt)
3119                 RETURN(-ENOTSUPP);
3120
3121         rc = md_intent_getattr_async(ptgt->ltd_exp, minfo);
3122         RETURN(rc);
3123 }
3124
3125 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3126                         struct lu_fid *fid, __u64 *bits)
3127 {
3128         struct obd_device       *obd = exp->exp_obd;
3129         struct lmv_obd          *lmv = &obd->u.lmv;
3130         struct lmv_tgt_desc     *tgt;
3131         int                      rc;
3132         ENTRY;
3133
3134         rc = lmv_check_connect(obd);
3135         if (rc)
3136                 RETURN(rc);
3137
3138         tgt = lmv_find_target(lmv, fid);
3139         if (IS_ERR(tgt))
3140                 RETURN(PTR_ERR(tgt));
3141
3142         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3143         RETURN(rc);
3144 }
3145
3146 int lmv_get_fid_from_lsm(struct obd_export *exp,
3147                          const struct lmv_stripe_md *lsm,
3148                          const char *name, int namelen, struct lu_fid *fid)
3149 {
3150         const struct lmv_oinfo *oinfo;
3151
3152         LASSERT(lsm != NULL);
3153         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3154         if (IS_ERR(oinfo))
3155                 return PTR_ERR(oinfo);
3156
3157         *fid = oinfo->lmo_fid;
3158
3159         RETURN(0);
3160 }
3161
3162 /**
3163  * For lmv, only need to send request to master MDT, and the master MDT will
3164  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3165  * we directly fetch data from the slave MDTs.
3166  */
3167 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3168                  struct obd_quotactl *oqctl)
3169 {
3170         struct obd_device   *obd = class_exp2obd(exp);
3171         struct lmv_obd      *lmv = &obd->u.lmv;
3172         struct lmv_tgt_desc *tgt = lmv->tgts[0];
3173         int                  rc = 0;
3174         __u32                i;
3175         __u64                curspace, curinodes;
3176         ENTRY;
3177
3178         if (tgt == NULL ||
3179             tgt->ltd_exp == NULL ||
3180             !tgt->ltd_active ||
3181             lmv->desc.ld_tgt_count == 0) {
3182                 CERROR("master lmv inactive\n");
3183                 RETURN(-EIO);
3184         }
3185
3186         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3187                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3188                 RETURN(rc);
3189         }
3190
3191         curspace = curinodes = 0;
3192         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3193                 int err;
3194                 tgt = lmv->tgts[i];
3195
3196                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3197                         continue;
3198
3199                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3200                 if (err) {
3201                         CERROR("getquota on mdt %d failed. %d\n", i, err);
3202                         if (!rc)
3203                                 rc = err;
3204                 } else {
3205                         curspace += oqctl->qc_dqblk.dqb_curspace;
3206                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3207                 }
3208         }
3209         oqctl->qc_dqblk.dqb_curspace = curspace;
3210         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3211
3212         RETURN(rc);
3213 }
3214
3215 static int lmv_merge_attr(struct obd_export *exp,
3216                           const struct lmv_stripe_md *lsm,
3217                           struct cl_attr *attr,
3218                           ldlm_blocking_callback cb_blocking)
3219 {
3220         int rc;
3221         int i;
3222
3223         rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3224         if (rc < 0)
3225                 return rc;
3226
3227         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3228                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3229
3230                 CDEBUG(D_INFO, ""DFID" size %llu, blocks %llu nlink %u,"
3231                        " atime %lu ctime %lu, mtime %lu.\n",
3232                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3233                        i_size_read(inode), (unsigned long long)inode->i_blocks,
3234                        inode->i_nlink, LTIME_S(inode->i_atime),
3235                        LTIME_S(inode->i_ctime), LTIME_S(inode->i_mtime));
3236
3237                 /* for slave stripe, it needs to subtract nlink for . and .. */
3238                 if (i != 0)
3239                         attr->cat_nlink += inode->i_nlink - 2;
3240                 else
3241                         attr->cat_nlink = inode->i_nlink;
3242
3243                 attr->cat_size += i_size_read(inode);
3244                 attr->cat_blocks += inode->i_blocks;
3245
3246                 if (attr->cat_atime < LTIME_S(inode->i_atime))
3247                         attr->cat_atime = LTIME_S(inode->i_atime);
3248
3249                 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3250                         attr->cat_ctime = LTIME_S(inode->i_ctime);
3251
3252                 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3253                         attr->cat_mtime = LTIME_S(inode->i_mtime);
3254         }
3255         return 0;
3256 }
3257
3258 struct obd_ops lmv_obd_ops = {
3259         .o_owner                = THIS_MODULE,
3260         .o_setup                = lmv_setup,
3261         .o_cleanup              = lmv_cleanup,
3262         .o_precleanup           = lmv_precleanup,
3263         .o_process_config       = lmv_process_config,
3264         .o_connect              = lmv_connect,
3265         .o_disconnect           = lmv_disconnect,
3266         .o_statfs               = lmv_statfs,
3267         .o_get_info             = lmv_get_info,
3268         .o_set_info_async       = lmv_set_info_async,
3269         .o_notify               = lmv_notify,
3270         .o_get_uuid             = lmv_get_uuid,
3271         .o_iocontrol            = lmv_iocontrol,
3272         .o_quotactl             = lmv_quotactl
3273 };
3274
3275 struct md_ops lmv_md_ops = {
3276         .m_get_root             = lmv_get_root,
3277         .m_null_inode           = lmv_null_inode,
3278         .m_close                = lmv_close,
3279         .m_create               = lmv_create,
3280         .m_enqueue              = lmv_enqueue,
3281         .m_getattr              = lmv_getattr,
3282         .m_getxattr             = lmv_getxattr,
3283         .m_getattr_name         = lmv_getattr_name,
3284         .m_intent_lock          = lmv_intent_lock,
3285         .m_link                 = lmv_link,
3286         .m_rename               = lmv_rename,
3287         .m_setattr              = lmv_setattr,
3288         .m_setxattr             = lmv_setxattr,
3289         .m_fsync                = lmv_fsync,
3290         .m_read_page            = lmv_read_page,
3291         .m_unlink               = lmv_unlink,
3292         .m_init_ea_size         = lmv_init_ea_size,
3293         .m_cancel_unused        = lmv_cancel_unused,
3294         .m_set_lock_data        = lmv_set_lock_data,
3295         .m_lock_match           = lmv_lock_match,
3296         .m_get_lustre_md        = lmv_get_lustre_md,
3297         .m_free_lustre_md       = lmv_free_lustre_md,
3298         .m_merge_attr           = lmv_merge_attr,
3299         .m_set_open_replay_data = lmv_set_open_replay_data,
3300         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3301         .m_intent_getattr_async = lmv_intent_getattr_async,
3302         .m_revalidate_lock      = lmv_revalidate_lock,
3303         .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
3304         .m_unpackmd             = lmv_unpackmd,
3305 };
3306
3307 static int __init lmv_init(void)
3308 {
3309         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3310                                    LUSTRE_LMV_NAME, NULL);
3311 }
3312
3313 static void __exit lmv_exit(void)
3314 {
3315         class_unregister_type(LUSTRE_LMV_NAME);
3316 }
3317
3318 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3319 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
3320 MODULE_VERSION(LUSTRE_VERSION_STRING);
3321 MODULE_LICENSE("GPL");
3322
3323 module_init(lmv_init);
3324 module_exit(lmv_exit);