lustre/lmv/lmv_obd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LMV
  38 #include <linux/slab.h>
  39 #include <linux/module.h>
  40 #include <linux/init.h>
  41 #include <linux/user_namespace.h>
  42 #ifdef HAVE_UIDGID_HEADER
  43 # include <linux/uidgid.h>
  44 #endif
  45 #include <linux/slab.h>
  46 #include <linux/pagemap.h>
  47 #include <linux/mm.h>
  48 #include <linux/math64.h>
  49 #include <linux/seq_file.h>
  50 #include <linux/namei.h>
  51
  52 #include <lustre/lustre_idl.h>
  53 #include <obd_support.h>
  54 #include <lustre_lib.h>
  55 #include <lustre_net.h>
  56 #include <obd_class.h>
  57 #include <lustre_lmv.h>
  58 #include <lprocfs_status.h>
  59 #include <cl_object.h>
  60 #include <lustre_fid.h>
  61 #include <lustre_ioctl.h>
  62 #include <lustre_kernelcomm.h>
  63 #include "lmv_internal.h"
  64
  65 static void lmv_activate_target(struct lmv_obd *lmv,
  66                                 struct lmv_tgt_desc *tgt,
  67                                 int activate)
  68 {
  69         if (tgt->ltd_active == activate)
  70                 return;
  71
  72         tgt->ltd_active = activate;
  73         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  74
  75         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, void *uarg)
 687 {
 688         struct obd_device       *obddev = class_exp2obd(exp);
 689         struct lmv_obd          *lmv = &obddev->u.lmv;
 690         struct getinfo_fid2path *gf;
 691         struct lmv_tgt_desc     *tgt;
 692         struct getinfo_fid2path *remote_gf = NULL;
 693         int                     remote_gf_size = 0;
 694         int                     rc;
 695
 696         gf = (struct getinfo_fid2path *)karg;
 697         tgt = lmv_find_target(lmv, &gf->gf_fid);
 698         if (IS_ERR(tgt))
 699                 RETURN(PTR_ERR(tgt));
 700
 701 repeat_fid2path:
 702         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 703         if (rc != 0 && rc != -EREMOTE)
 704                 GOTO(out_fid2path, rc);
 705
 706         /* If remote_gf != NULL, it means just building the
 707          * path on the remote MDT, copy this path segement to gf */
 708         if (remote_gf != NULL) {
 709                 struct getinfo_fid2path *ori_gf;
 710                 char *ptr;
 711
 712                 ori_gf = (struct getinfo_fid2path *)karg;
 713                 if (strlen(ori_gf->gf_path) +
 714                     strlen(gf->gf_path) > ori_gf->gf_pathlen)
 715                         GOTO(out_fid2path, rc = -EOVERFLOW);
 716
 717                 ptr = ori_gf->gf_path;
 718
 719                 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
 720                         strlen(ori_gf->gf_path));
 721
 722                 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
 723                 ptr += strlen(gf->gf_path);
 724                 *ptr = '/';
 725         }
 726
 727         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
 728                tgt->ltd_exp->exp_obd->obd_name,
 729                gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 730                gf->gf_linkno);
 731
 732         if (rc == 0)
 733                 GOTO(out_fid2path, rc);
 734
 735         /* sigh, has to go to another MDT to do path building further */
 736         if (remote_gf == NULL) {
 737                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 738                 OBD_ALLOC(remote_gf, remote_gf_size);
 739                 if (remote_gf == NULL)
 740                         GOTO(out_fid2path, rc = -ENOMEM);
 741                 remote_gf->gf_pathlen = PATH_MAX;
 742         }
 743
 744         if (!fid_is_sane(&gf->gf_fid)) {
 745                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 746                        tgt->ltd_exp->exp_obd->obd_name,
 747                        PFID(&gf->gf_fid), -EINVAL);
 748                 GOTO(out_fid2path, rc = -EINVAL);
 749         }
 750
 751         tgt = lmv_find_target(lmv, &gf->gf_fid);
 752         if (IS_ERR(tgt))
 753                 GOTO(out_fid2path, rc = -EINVAL);
 754
 755         remote_gf->gf_fid = gf->gf_fid;
 756         remote_gf->gf_recno = -1;
 757         remote_gf->gf_linkno = -1;
 758         memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
 759         gf = remote_gf;
 760         goto repeat_fid2path;
 761
 762 out_fid2path:
 763         if (remote_gf != NULL)
 764                 OBD_FREE(remote_gf, remote_gf_size);
 765         RETURN(rc);
 766 }
 767
 768 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 769                              const struct hsm_user_request *hur,
 770                              const struct lmv_tgt_desc *tgt_mds)
 771 {
 772         __u32                    i;
 773         int                      nr = 0;
 774         struct lmv_tgt_desc     *curr_tgt;
 775
 776         /* count how many requests must be sent to the given target */
 777         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 778                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
 779                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 780                         nr++;
 781         }
 782         return nr;
 783 }
 784
 785 static void lmv_hsm_req_build(struct lmv_obd *lmv,
 786                               struct hsm_user_request *hur_in,
 787                               const struct lmv_tgt_desc *tgt_mds,
 788                               struct hsm_user_request *hur_out)
 789 {
 790         __u32                    i, nr_out;
 791         struct lmv_tgt_desc     *curr_tgt;
 792
 793         /* build the hsm_user_request for the given target */
 794         hur_out->hur_request = hur_in->hur_request;
 795         nr_out = 0;
 796         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 797                 curr_tgt = lmv_find_target(lmv,
 798                                            &hur_in->hur_user_item[i].hui_fid);
 799                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 800                         hur_out->hur_user_item[nr_out] =
 801                                                 hur_in->hur_user_item[i];
 802                         nr_out++;
 803                 }
 804         }
 805         hur_out->hur_request.hr_itemcount = nr_out;
 806         memcpy(hur_data(hur_out), hur_data(hur_in),
 807                hur_in->hur_request.hr_data_len);
 808 }
 809
 810 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
 811                                  struct lustre_kernelcomm *lk, void *uarg)
 812 {
 813         __u32                    i;
 814         int                      rc;
 815         struct kkuc_ct_data     *kcd = NULL;
 816         ENTRY;
 817
 818         /* unregister request (call from llapi_hsm_copytool_fini) */
 819         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 820                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 821
 822                 if (tgt == NULL || tgt->ltd_exp == NULL)
 823                         continue;
 824                 /* best effort: try to clean as much as possible
 825                  * (continue on error) */
 826                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 827         }
 828
 829         /* Whatever the result, remove copytool from kuc groups.
 830          * Unreached coordinators will get EPIPE on next requests
 831          * and will unregister automatically.
 832          */
 833         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
 834         if (kcd != NULL)
 835                 OBD_FREE_PTR(kcd);
 836
 837         RETURN(rc);
 838 }
 839
 840 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
 841                                struct lustre_kernelcomm *lk, void *uarg)
 842 {
 843         struct file             *filp;
 844         __u32                    i, j;
 845         int                      err, rc;
 846         bool                     any_set = false;
 847         struct kkuc_ct_data     *kcd;
 848         ENTRY;
 849
 850         /* All or nothing: try to register to all MDS.
 851          * In case of failure, unregister from previous MDS,
 852          * except if it because of inactive target. */
 853         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 854                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 855
 856                 if (tgt == NULL || tgt->ltd_exp == NULL)
 857                         continue;
 858                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 859                 if (err) {
 860                         if (tgt->ltd_active) {
 861                                 /* permanent error */
 862                                 CERROR("%s: iocontrol MDC %s on MDT"
 863                                        " idx %d cmd %x: err = %d\n",
 864                                        class_exp2obd(lmv->exp)->obd_name,
 865                                        tgt->ltd_uuid.uuid, i, cmd, err);
 866                                 rc = err;
 867                                 lk->lk_flags |= LK_FLG_STOP;
 868                                 /* unregister from previous MDS */
 869                                 for (j = 0; j < i; j++) {
 870                                         tgt = lmv->tgts[j];
 871                                         if (tgt == NULL || tgt->ltd_exp == NULL)
 872                                                 continue;
 873                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
 874                                                       lk, uarg);
 875                                 }
 876                                 RETURN(rc);
 877                         }
 878                         /* else: transient error.
 879                          * kuc will register to the missing MDT
 880                          * when it is back */
 881                 } else {
 882                         any_set = true;
 883                 }
 884         }
 885
 886         if (!any_set)
 887                 /* no registration done: return error */
 888                 RETURN(-ENOTCONN);
 889
 890         /* at least one registration done, with no failure */
 891         filp = fget(lk->lk_wfd);
 892         if (filp == NULL)
 893                 RETURN(-EBADF);
 894
 895         OBD_ALLOC_PTR(kcd);
 896         if (kcd == NULL) {
 897                 fput(filp);
 898                 RETURN(-ENOMEM);
 899         }
 900         kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
 901         kcd->kcd_uuid = lmv->cluuid;
 902         kcd->kcd_archive = lk->lk_data;
 903
 904         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
 905         if (rc != 0) {
 906                 if (filp != NULL)
 907                         fput(filp);
 908                 OBD_FREE_PTR(kcd);
 909         }
 910
 911         RETURN(rc);
 912 }
 913
 914
 915
 916
 917 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 918                          int len, void *karg, void *uarg)
 919 {
 920         struct obd_device       *obddev = class_exp2obd(exp);
 921         struct lmv_obd          *lmv = &obddev->u.lmv;
 922         struct lmv_tgt_desc     *tgt = NULL;
 923         __u32                    i = 0;
 924         int                      rc = 0;
 925         int                      set = 0;
 926         __u32                    count = lmv->desc.ld_tgt_count;
 927         ENTRY;
 928
 929         if (count == 0)
 930                 RETURN(-ENOTTY);
 931
 932         switch (cmd) {
 933         case IOC_OBD_STATFS: {
 934                 struct obd_ioctl_data *data = karg;
 935                 struct obd_device *mdc_obd;
 936                 struct obd_statfs stat_buf = {0};
 937                 __u32 index;
 938
 939                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 940                 if ((index >= count))
 941                         RETURN(-ENODEV);
 942
 943                 tgt = lmv->tgts[index];
 944                 if (tgt == NULL || !tgt->ltd_active)
 945                         RETURN(-ENODATA);
 946
 947                 mdc_obd = class_exp2obd(tgt->ltd_exp);
 948                 if (!mdc_obd)
 949                         RETURN(-EINVAL);
 950
 951                 /* copy UUID */
 952                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 953                                  min((int) data->ioc_plen2,
 954                                      (int) sizeof(struct obd_uuid))))
 955                         RETURN(-EFAULT);
 956
 957                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
 958                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 959                                 0);
 960                 if (rc)
 961                         RETURN(rc);
 962                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
 963                                  min((int) data->ioc_plen1,
 964                                      (int) sizeof(stat_buf))))
 965                         RETURN(-EFAULT);
 966                 break;
 967         }
 968         case OBD_IOC_QUOTACTL: {
 969                 struct if_quotactl *qctl = karg;
 970                 struct obd_quotactl *oqctl;
 971
 972                 if (qctl->qc_valid == QC_MDTIDX) {
 973                         if (count <= qctl->qc_idx)
 974                                 RETURN(-EINVAL);
 975
 976                         tgt = lmv->tgts[qctl->qc_idx];
 977                         if (tgt == NULL || tgt->ltd_exp == NULL)
 978                                 RETURN(-EINVAL);
 979                 } else if (qctl->qc_valid == QC_UUID) {
 980                         for (i = 0; i < count; i++) {
 981                                 tgt = lmv->tgts[i];
 982                                 if (tgt == NULL)
 983                                         continue;
 984                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 985                                                      &qctl->obd_uuid))
 986                                         continue;
 987
 988                                 if (tgt->ltd_exp == NULL)
 989                                         RETURN(-EINVAL);
 990
 991                                 break;
 992                         }
 993                 } else {
 994                         RETURN(-EINVAL);
 995                 }
 996
 997                 if (i >= count)
 998                         RETURN(-EAGAIN);
 999
1000                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1001                 OBD_ALLOC_PTR(oqctl);
1002                 if (!oqctl)
1003                         RETURN(-ENOMEM);
1004
1005                 QCTL_COPY(oqctl, qctl);
1006                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1007                 if (rc == 0) {
1008                         QCTL_COPY(qctl, oqctl);
1009                         qctl->qc_valid = QC_MDTIDX;
1010                         qctl->obd_uuid = tgt->ltd_uuid;
1011                 }
1012                 OBD_FREE_PTR(oqctl);
1013                 break;
1014         }
1015         case OBD_IOC_CHANGELOG_SEND:
1016         case OBD_IOC_CHANGELOG_CLEAR: {
1017                 struct ioc_changelog *icc = karg;
1018
1019                 if (icc->icc_mdtindex >= count)
1020                         RETURN(-ENODEV);
1021
1022                 tgt = lmv->tgts[icc->icc_mdtindex];
1023                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1024                         RETURN(-ENODEV);
1025                 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1026                 break;
1027         }
1028         case LL_IOC_GET_CONNECT_FLAGS: {
1029                 tgt = lmv->tgts[0];
1030                 if (tgt == NULL || tgt->ltd_exp == NULL)
1031                         RETURN(-ENODATA);
1032                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1033                 break;
1034         }
1035         case LL_IOC_FID2MDTIDX: {
1036                 struct lu_fid *fid = karg;
1037                 int             mdt_index;
1038
1039                 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1040                 if (rc != 0)
1041                         RETURN(rc);
1042
1043                 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1044                  * point to user space memory for FID2MDTIDX. */
1045                 *(__u32 *)uarg = mdt_index;
1046                 break;
1047         }
1048         case OBD_IOC_FID2PATH: {
1049                 rc = lmv_fid2path(exp, len, karg, uarg);
1050                 break;
1051         }
1052         case LL_IOC_HSM_STATE_GET:
1053         case LL_IOC_HSM_STATE_SET:
1054         case LL_IOC_HSM_ACTION: {
1055                 struct md_op_data       *op_data = karg;
1056
1057                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1058                 if (IS_ERR(tgt))
1059                         RETURN(PTR_ERR(tgt));
1060
1061                 if (tgt->ltd_exp == NULL)
1062                         RETURN(-EINVAL);
1063
1064                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1065                 break;
1066         }
1067         case LL_IOC_HSM_PROGRESS: {
1068                 const struct hsm_progress_kernel *hpk = karg;
1069
1070                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1071                 if (IS_ERR(tgt))
1072                         RETURN(PTR_ERR(tgt));
1073                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1074                 break;
1075         }
1076         case LL_IOC_HSM_REQUEST: {
1077                 struct hsm_user_request *hur = karg;
1078                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1079
1080                 if (reqcount == 0)
1081                         RETURN(0);
1082
1083                 /* if the request is about a single fid
1084                  * or if there is a single MDS, no need to split
1085                  * the request. */
1086                 if (reqcount == 1 || count == 1) {
1087                         tgt = lmv_find_target(lmv,
1088                                               &hur->hur_user_item[0].hui_fid);
1089                         if (IS_ERR(tgt))
1090                                 RETURN(PTR_ERR(tgt));
1091                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1092                 } else {
1093                         /* split fid list to their respective MDS */
1094                         for (i = 0; i < count; i++) {
1095                                 unsigned int            nr, reqlen;
1096                                 int                     rc1;
1097                                 struct hsm_user_request *req;
1098
1099                                 tgt = lmv->tgts[i];
1100                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1101                                         continue;
1102
1103                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1104                                 if (nr == 0) /* nothing for this MDS */
1105                                         continue;
1106
1107                                 /* build a request with fids for this MDS */
1108                                 reqlen = offsetof(typeof(*hur),
1109                                                   hur_user_item[nr])
1110                                                 + hur->hur_request.hr_data_len;
1111                                 OBD_ALLOC_LARGE(req, reqlen);
1112                                 if (req == NULL)
1113                                         RETURN(-ENOMEM);
1114
1115                                 lmv_hsm_req_build(lmv, hur, tgt, req);
1116
1117                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1118                                                     req, uarg);
1119                                 if (rc1 != 0 && rc == 0)
1120                                         rc = rc1;
1121                                 OBD_FREE_LARGE(req, reqlen);
1122                         }
1123                 }
1124                 break;
1125         }
1126         case LL_IOC_LOV_SWAP_LAYOUTS: {
1127                 struct md_op_data       *op_data = karg;
1128                 struct lmv_tgt_desc     *tgt1, *tgt2;
1129
1130                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1131                 if (IS_ERR(tgt1))
1132                         RETURN(PTR_ERR(tgt1));
1133
1134                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1135                 if (IS_ERR(tgt2))
1136                         RETURN(PTR_ERR(tgt2));
1137
1138                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1139                         RETURN(-EINVAL);
1140
1141                 /* only files on same MDT can have their layouts swapped */
1142                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1143                         RETURN(-EPERM);
1144
1145                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1146                 break;
1147         }
1148         case LL_IOC_HSM_CT_START: {
1149                 struct lustre_kernelcomm *lk = karg;
1150                 if (lk->lk_flags & LK_FLG_STOP)
1151                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1152                 else
1153                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1154                 break;
1155         }
1156         default:
1157                 for (i = 0; i < count; i++) {
1158                         struct obd_device *mdc_obd;
1159                         int err;
1160
1161                         tgt = lmv->tgts[i];
1162                         if (tgt == NULL || tgt->ltd_exp == NULL)
1163                                 continue;
1164                         /* ll_umount_begin() sets force flag but for lmv, not
1165                          * mdc. Let's pass it through */
1166                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1167                         mdc_obd->obd_force = obddev->obd_force;
1168                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1169                         if (err) {
1170                                 if (tgt->ltd_active) {
1171                                         CERROR("error: iocontrol MDC %s on MDT"
1172                                                " idx %d cmd %x: err = %d\n",
1173                                                tgt->ltd_uuid.uuid, i, cmd, err);
1174                                         if (!rc)
1175                                                 rc = err;
1176                                 }
1177                         } else
1178                                 set = 1;
1179                 }
1180                 if (!set && !rc)
1181                         rc = -EIO;
1182         }
1183         RETURN(rc);
1184 }
1185
1186 #if 0
1187 static int lmv_all_chars_policy(int count, const char *name,
1188                                 int len)
1189 {
1190         unsigned int c = 0;
1191
1192         while (len > 0)
1193                 c += name[--len];
1194         c = c % count;
1195         return c;
1196 }
1197
1198 static int lmv_nid_policy(struct lmv_obd *lmv)
1199 {
1200         struct obd_import *imp;
1201         __u32              id;
1202
1203         /*
1204          * XXX: To get nid we assume that underlying obd device is mdc.
1205          */
1206         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1207         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1208         return id % lmv->desc.ld_tgt_count;
1209 }
1210
1211 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1212                           placement_policy_t placement)
1213 {
1214         switch (placement) {
1215         case PLACEMENT_CHAR_POLICY:
1216                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1217                                             op_data->op_name,
1218                                             op_data->op_namelen);
1219         case PLACEMENT_NID_POLICY:
1220                 return lmv_nid_policy(lmv);
1221
1222         default:
1223                 break;
1224         }
1225
1226         CERROR("Unsupported placement policy %x\n", placement);
1227         return -EINVAL;
1228 }
1229 #endif
1230
1231 /**
1232  * This is _inode_ placement policy function (not name).
1233  */
1234 static int lmv_placement_policy(struct obd_device *obd,
1235                                 struct md_op_data *op_data, u32 *mds)
1236 {
1237         struct lmv_obd          *lmv = &obd->u.lmv;
1238         ENTRY;
1239
1240         LASSERT(mds != NULL);
1241
1242         if (lmv->desc.ld_tgt_count == 1) {
1243                 *mds = 0;
1244                 RETURN(0);
1245         }
1246
1247         if (op_data->op_default_stripe_offset != -1) {
1248                 *mds = op_data->op_default_stripe_offset;
1249                 RETURN(0);
1250         }
1251
1252         /**
1253          * If stripe_offset is provided during setdirstripe
1254          * (setdirstripe -i xx), xx MDS will be choosen.
1255          */
1256         if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1257                 struct lmv_user_md *lum;
1258
1259                 lum = op_data->op_data;
1260
1261                 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1262                         *mds = le32_to_cpu(lum->lum_stripe_offset);
1263                 } else {
1264                         /* -1 means default, which will be in the same MDT with
1265                          * the stripe */
1266                         *mds = op_data->op_mds;
1267                         lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1268                 }
1269         } else {
1270                 /* Allocate new fid on target according to operation type and
1271                  * parent home mds. */
1272                 *mds = op_data->op_mds;
1273         }
1274
1275         RETURN(0);
1276 }
1277
1278 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1279 {
1280         struct lmv_tgt_desc     *tgt;
1281         int                      rc;
1282         ENTRY;
1283
1284         tgt = lmv_get_target(lmv, mds, NULL);
1285         if (IS_ERR(tgt))
1286                 RETURN(PTR_ERR(tgt));
1287
1288         /*
1289          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1290          * on server that seq in new allocated fid is not yet known.
1291          */
1292         mutex_lock(&tgt->ltd_fid_mutex);
1293
1294         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1295                 GOTO(out, rc = -ENODEV);
1296
1297         /*
1298          * Asking underlying tgt layer to allocate new fid.
1299          */
1300         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1301         if (rc > 0) {
1302                 LASSERT(fid_is_sane(fid));
1303                 rc = 0;
1304         }
1305
1306         EXIT;
1307 out:
1308         mutex_unlock(&tgt->ltd_fid_mutex);
1309         return rc;
1310 }
1311
1312 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1313                   struct lu_fid *fid, struct md_op_data *op_data)
1314 {
1315         struct obd_device     *obd = class_exp2obd(exp);
1316         struct lmv_obd        *lmv = &obd->u.lmv;
1317         u32                    mds = 0;
1318         int                    rc;
1319         ENTRY;
1320
1321         LASSERT(op_data != NULL);
1322         LASSERT(fid != NULL);
1323
1324         rc = lmv_placement_policy(obd, op_data, &mds);
1325         if (rc) {
1326                 CERROR("Can't get target for allocating fid, "
1327                        "rc %d\n", rc);
1328                 RETURN(rc);
1329         }
1330
1331         rc = __lmv_fid_alloc(lmv, fid, mds);
1332         if (rc) {
1333                 CERROR("Can't alloc new fid, rc %d\n", rc);
1334                 RETURN(rc);
1335         }
1336
1337         RETURN(rc);
1338 }
1339
1340 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1341 {
1342         struct lmv_obd  *lmv = &obd->u.lmv;
1343         struct lmv_desc *desc;
1344         int             rc;
1345         ENTRY;
1346
1347         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1348                 CERROR("LMV setup requires a descriptor\n");
1349                 RETURN(-EINVAL);
1350         }
1351
1352         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1353         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1354                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1355                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1356                 RETURN(-EINVAL);
1357         }
1358
1359         lmv->tgts_size = 32U;
1360         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1361         if (lmv->tgts == NULL)
1362                 RETURN(-ENOMEM);
1363
1364         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1365         lmv->desc.ld_tgt_count = 0;
1366         lmv->desc.ld_active_tgt_count = 0;
1367         lmv->max_def_easize = 0;
1368         lmv->max_easize = 0;
1369         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1370
1371         spin_lock_init(&lmv->lmv_lock);
1372         mutex_init(&lmv->lmv_init_mutex);
1373
1374 #ifdef CONFIG_PROC_FS
1375         obd->obd_vars = lprocfs_lmv_obd_vars;
1376         lprocfs_obd_setup(obd);
1377         lprocfs_alloc_md_stats(obd, 0);
1378         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1379                                 0444, &lmv_proc_target_fops, obd);
1380         if (rc)
1381                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1382                       obd->obd_name, rc);
1383 #endif
1384         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1385                              LUSTRE_CLI_FLD_HASH_DHT);
1386         if (rc) {
1387                 CERROR("Can't init FLD, err %d\n", rc);
1388                 GOTO(out, rc);
1389         }
1390
1391         RETURN(0);
1392
1393 out:
1394         return rc;
1395 }
1396
1397 static int lmv_cleanup(struct obd_device *obd)
1398 {
1399         struct lmv_obd   *lmv = &obd->u.lmv;
1400         ENTRY;
1401
1402         fld_client_fini(&lmv->lmv_fld);
1403         if (lmv->tgts != NULL) {
1404                 int i;
1405                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1406                         if (lmv->tgts[i] == NULL)
1407                                 continue;
1408                         lmv_del_target(lmv, i);
1409                 }
1410                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1411                 lmv->tgts_size = 0;
1412         }
1413         RETURN(0);
1414 }
1415
1416 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1417 {
1418         struct lustre_cfg       *lcfg = buf;
1419         struct obd_uuid         obd_uuid;
1420         int                     gen;
1421         __u32                   index;
1422         int                     rc;
1423         ENTRY;
1424
1425         switch (lcfg->lcfg_command) {
1426         case LCFG_ADD_MDC:
1427                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1428                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1429                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1430                         GOTO(out, rc = -EINVAL);
1431
1432                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1433
1434                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1435                         GOTO(out, rc = -EINVAL);
1436                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1437                         GOTO(out, rc = -EINVAL);
1438                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1439                 GOTO(out, rc);
1440         default:
1441                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1442                 GOTO(out, rc = -EINVAL);
1443         }
1444 out:
1445         RETURN(rc);
1446 }
1447
1448 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1449                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1450 {
1451         struct obd_device       *obd = class_exp2obd(exp);
1452         struct lmv_obd          *lmv = &obd->u.lmv;
1453         struct obd_statfs       *temp;
1454         int                      rc = 0;
1455         __u32                    i;
1456         ENTRY;
1457
1458         rc = lmv_check_connect(obd);
1459         if (rc)
1460                 RETURN(rc);
1461
1462         OBD_ALLOC(temp, sizeof(*temp));
1463         if (temp == NULL)
1464                 RETURN(-ENOMEM);
1465
1466         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1467                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1468                         continue;
1469
1470                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1471                                 max_age, flags);
1472                 if (rc) {
1473                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1474                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1475                                rc);
1476                         GOTO(out_free_temp, rc);
1477                 }
1478
1479                 if (i == 0) {
1480                         *osfs = *temp;
1481                         /* If the statfs is from mount, it will needs
1482                          * retrieve necessary information from MDT0.
1483                          * i.e. mount does not need the merged osfs
1484                          * from all of MDT.
1485                          * And also clients can be mounted as long as
1486                          * MDT0 is in service*/
1487                         if (flags & OBD_STATFS_FOR_MDT0)
1488                                 GOTO(out_free_temp, rc);
1489                 } else {
1490                         osfs->os_bavail += temp->os_bavail;
1491                         osfs->os_blocks += temp->os_blocks;
1492                         osfs->os_ffree += temp->os_ffree;
1493                         osfs->os_files += temp->os_files;
1494                 }
1495         }
1496
1497         EXIT;
1498 out_free_temp:
1499         OBD_FREE(temp, sizeof(*temp));
1500         return rc;
1501 }
1502
1503 static int lmv_getstatus(struct obd_export *exp, struct lu_fid *fid)
1504 {
1505         struct obd_device    *obd = exp->exp_obd;
1506         struct lmv_obd       *lmv = &obd->u.lmv;
1507         int                   rc;
1508         ENTRY;
1509
1510         rc = lmv_check_connect(obd);
1511         if (rc)
1512                 RETURN(rc);
1513
1514         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid);
1515         RETURN(rc);
1516 }
1517
1518 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1519                         u64 valid, const char *name,
1520                         const char *input, int input_size, int output_size,
1521                         int flags, struct ptlrpc_request **request)
1522 {
1523         struct obd_device      *obd = exp->exp_obd;
1524         struct lmv_obd         *lmv = &obd->u.lmv;
1525         struct lmv_tgt_desc    *tgt;
1526         int                     rc;
1527         ENTRY;
1528
1529         rc = lmv_check_connect(obd);
1530         if (rc)
1531                 RETURN(rc);
1532
1533         tgt = lmv_find_target(lmv, fid);
1534         if (IS_ERR(tgt))
1535                 RETURN(PTR_ERR(tgt));
1536
1537         rc = md_getxattr(tgt->ltd_exp, fid, valid, name, input,
1538                          input_size, output_size, flags, request);
1539
1540         RETURN(rc);
1541 }
1542
1543 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1544                         u64 valid, const char *name,
1545                         const char *input, int input_size, int output_size,
1546                         int flags, __u32 suppgid,
1547                         struct ptlrpc_request **request)
1548 {
1549         struct obd_device      *obd = exp->exp_obd;
1550         struct lmv_obd         *lmv = &obd->u.lmv;
1551         struct lmv_tgt_desc    *tgt;
1552         int                     rc;
1553         ENTRY;
1554
1555         rc = lmv_check_connect(obd);
1556         if (rc)
1557                 RETURN(rc);
1558
1559         tgt = lmv_find_target(lmv, fid);
1560         if (IS_ERR(tgt))
1561                 RETURN(PTR_ERR(tgt));
1562
1563         rc = md_setxattr(tgt->ltd_exp, fid, valid, name, input,
1564                          input_size, output_size, flags, suppgid,
1565                          request);
1566
1567         RETURN(rc);
1568 }
1569
1570 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1571                        struct ptlrpc_request **request)
1572 {
1573         struct obd_device       *obd = exp->exp_obd;
1574         struct lmv_obd          *lmv = &obd->u.lmv;
1575         struct lmv_tgt_desc     *tgt;
1576         int                      rc;
1577         ENTRY;
1578
1579         rc = lmv_check_connect(obd);
1580         if (rc)
1581                 RETURN(rc);
1582
1583         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1584         if (IS_ERR(tgt))
1585                 RETURN(PTR_ERR(tgt));
1586
1587         if (op_data->op_flags & MF_GET_MDT_IDX) {
1588                 op_data->op_mds = tgt->ltd_idx;
1589                 RETURN(0);
1590         }
1591
1592         rc = md_getattr(tgt->ltd_exp, op_data, request);
1593
1594         RETURN(rc);
1595 }
1596
1597 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1598 {
1599         struct obd_device   *obd = exp->exp_obd;
1600         struct lmv_obd      *lmv = &obd->u.lmv;
1601         __u32                i;
1602         int                  rc;
1603         ENTRY;
1604
1605         rc = lmv_check_connect(obd);
1606         if (rc)
1607                 RETURN(rc);
1608
1609         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1610
1611         /*
1612          * With DNE every object can have two locks in different namespaces:
1613          * lookup lock in space of MDT storing direntry and update/open lock in
1614          * space of MDT storing inode.
1615          */
1616         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1617                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1618                         continue;
1619                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1620         }
1621
1622         RETURN(0);
1623 }
1624
1625 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1626                            ldlm_iterator_t it, void *data)
1627 {
1628         struct obd_device       *obd = exp->exp_obd;
1629         struct lmv_obd          *lmv = &obd->u.lmv;
1630         int                     i;
1631         int                     tgt;
1632         int                     rc;
1633         ENTRY;
1634
1635         rc = lmv_check_connect(obd);
1636         if (rc)
1637                 RETURN(rc);
1638
1639         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1640
1641         /*
1642          * With DNE every object can have two locks in different namespaces:
1643          * lookup lock in space of MDT storing direntry and update/open lock in
1644          * space of MDT storing inode.  Try the MDT that the FID maps to first,
1645          * since this can be easily found, and only try others if that fails.
1646          */
1647         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1648              i < lmv->desc.ld_tgt_count;
1649              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1650                 if (tgt < 0) {
1651                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1652                                obd->obd_name, PFID(fid), tgt);
1653                         tgt = 0;
1654                 }
1655
1656                 if (lmv->tgts[tgt] == NULL ||
1657                     lmv->tgts[tgt]->ltd_exp == NULL)
1658                         continue;
1659
1660                 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1661                 if (rc)
1662                         RETURN(rc);
1663         }
1664
1665         RETURN(rc);
1666 }
1667
1668
1669 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1670                      struct md_open_data *mod, struct ptlrpc_request **request)
1671 {
1672         struct obd_device     *obd = exp->exp_obd;
1673         struct lmv_obd        *lmv = &obd->u.lmv;
1674         struct lmv_tgt_desc   *tgt;
1675         int                    rc;
1676         ENTRY;
1677
1678         rc = lmv_check_connect(obd);
1679         if (rc)
1680                 RETURN(rc);
1681
1682         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1683         if (IS_ERR(tgt))
1684                 RETURN(PTR_ERR(tgt));
1685
1686         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1687         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1688         RETURN(rc);
1689 }
1690
1691 /**
1692  * Choosing the MDT by name or FID in @op_data.
1693  * For non-striped directory, it will locate MDT by fid.
1694  * For striped-directory, it will locate MDT by name. And also
1695  * it will reset op_fid1 with the FID of the choosen stripe.
1696  **/
1697 struct lmv_tgt_desc *
1698 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1699                            const char *name, int namelen, struct lu_fid *fid,
1700                            u32 *mds)
1701 {
1702         struct lmv_tgt_desc     *tgt;
1703         const struct lmv_oinfo  *oinfo;
1704
1705         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1706                 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1707                         RETURN(ERR_PTR(-EBADF));
1708                 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1709         } else {
1710                 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1711                 if (IS_ERR(oinfo))
1712                         RETURN(ERR_CAST(oinfo));
1713         }
1714
1715         if (fid != NULL)
1716                 *fid = oinfo->lmo_fid;
1717         if (mds != NULL)
1718                 *mds = oinfo->lmo_mds;
1719
1720         tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1721
1722         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", oinfo->lmo_mds,
1723                PFID(&oinfo->lmo_fid));
1724         return tgt;
1725 }
1726
1727 /**
1728  * Locate mds by fid or name
1729  *
1730  * For striped directory (lsm != NULL), it will locate the stripe
1731  * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1732  * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1733  * walk through all of stripes to locate the entry.
1734  *
1735  * For normal direcotry, it will locate MDS by FID directly.
1736  * \param[in] lmv       LMV device
1737  * \param[in] op_data   client MD stack parameters, name, namelen
1738  *                      mds_num etc.
1739  * \param[in] fid       object FID used to locate MDS.
1740  *
1741  * retval               pointer to the lmv_tgt_desc if succeed.
1742  *                      ERR_PTR(errno) if failed.
1743  */
1744 struct lmv_tgt_desc*
1745 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1746                struct lu_fid *fid)
1747 {
1748         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1749         struct lmv_tgt_desc     *tgt;
1750
1751         /* During creating VOLATILE file, it should honor the mdt
1752          * index if the file under striped dir is being restored, see
1753          * ct_restore(). */
1754         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1755             (int)op_data->op_mds != -1 && lsm != NULL) {
1756                 int i;
1757                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1758                 if (IS_ERR(tgt))
1759                         return tgt;
1760
1761                 /* refill the right parent fid */
1762                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1763                         struct lmv_oinfo *oinfo;
1764
1765                         oinfo = &lsm->lsm_md_oinfo[i];
1766                         if (oinfo->lmo_mds == op_data->op_mds) {
1767                                 *fid = oinfo->lmo_fid;
1768                                 break;
1769                         }
1770                 }
1771
1772                 /* Hmm, can not find the stripe by mdt_index(op_mds) */
1773                 if (i == lsm->lsm_md_stripe_count)
1774                         tgt = ERR_PTR(-EINVAL);
1775
1776                 return tgt;
1777         }
1778
1779         if (lsm == NULL || op_data->op_namelen == 0) {
1780                 tgt = lmv_find_target(lmv, fid);
1781                 if (IS_ERR(tgt))
1782                         return tgt;
1783
1784                 op_data->op_mds = tgt->ltd_idx;
1785                 return tgt;
1786         }
1787
1788         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1789                                           op_data->op_namelen, fid,
1790                                           &op_data->op_mds);
1791 }
1792
1793 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1794                 const void *data, size_t datalen, umode_t mode, uid_t uid,
1795                 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1796                 struct ptlrpc_request **request)
1797 {
1798         struct obd_device       *obd = exp->exp_obd;
1799         struct lmv_obd          *lmv = &obd->u.lmv;
1800         struct lmv_tgt_desc     *tgt;
1801         int                      rc;
1802         ENTRY;
1803
1804         rc = lmv_check_connect(obd);
1805         if (rc)
1806                 RETURN(rc);
1807
1808         if (!lmv->desc.ld_active_tgt_count)
1809                 RETURN(-EIO);
1810
1811         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1812         if (IS_ERR(tgt))
1813                 RETURN(PTR_ERR(tgt));
1814
1815         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1816                 (int)op_data->op_namelen, op_data->op_name,
1817                 PFID(&op_data->op_fid1), op_data->op_mds);
1818
1819         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1820         if (rc)
1821                 RETURN(rc);
1822         if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1823                 /* Send the create request to the MDT where the object
1824                  * will be located */
1825                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1826                 if (IS_ERR(tgt))
1827                         RETURN(PTR_ERR(tgt));
1828
1829                 op_data->op_mds = tgt->ltd_idx;
1830         } else {
1831                 CDEBUG(D_CONFIG, "Server doesn't support striped dirs\n");
1832         }
1833
1834         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1835                PFID(&op_data->op_fid2), op_data->op_mds);
1836
1837         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1838         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1839                        cap_effective, rdev, request);
1840         if (rc == 0) {
1841                 if (*request == NULL)
1842                         RETURN(rc);
1843                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1844         }
1845         RETURN(rc);
1846 }
1847
1848 static int
1849 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1850             const union ldlm_policy_data *policy,
1851             struct lookup_intent *it, struct md_op_data *op_data,
1852             struct lustre_handle *lockh, __u64 extra_lock_flags)
1853 {
1854         struct obd_device        *obd = exp->exp_obd;
1855         struct lmv_obd           *lmv = &obd->u.lmv;
1856         struct lmv_tgt_desc      *tgt;
1857         int                       rc;
1858         ENTRY;
1859
1860         rc = lmv_check_connect(obd);
1861         if (rc)
1862                 RETURN(rc);
1863
1864         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1865                LL_IT2STR(it), PFID(&op_data->op_fid1));
1866
1867         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1868         if (IS_ERR(tgt))
1869                 RETURN(PTR_ERR(tgt));
1870
1871         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1872                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1873
1874         rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1875                         extra_lock_flags);
1876
1877         RETURN(rc);
1878 }
1879
1880 static int
1881 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1882                  struct ptlrpc_request **preq)
1883 {
1884         struct ptlrpc_request   *req = NULL;
1885         struct obd_device       *obd = exp->exp_obd;
1886         struct lmv_obd          *lmv = &obd->u.lmv;
1887         struct lmv_tgt_desc     *tgt;
1888         struct mdt_body         *body;
1889         int                      rc;
1890         ENTRY;
1891
1892         rc = lmv_check_connect(obd);
1893         if (rc)
1894                 RETURN(rc);
1895
1896         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1897         if (IS_ERR(tgt))
1898                 RETURN(PTR_ERR(tgt));
1899
1900         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1901                 (int)op_data->op_namelen, op_data->op_name,
1902                 PFID(&op_data->op_fid1), tgt->ltd_idx);
1903
1904         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1905         if (rc != 0)
1906                 RETURN(rc);
1907
1908         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1909         LASSERT(body != NULL);
1910
1911         if (body->mbo_valid & OBD_MD_MDS) {
1912                 struct lu_fid rid = body->mbo_fid1;
1913                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1914                        PFID(&rid));
1915
1916                 tgt = lmv_find_target(lmv, &rid);
1917                 if (IS_ERR(tgt)) {
1918                         ptlrpc_req_finished(*preq);
1919                         preq = NULL;
1920                         RETURN(PTR_ERR(tgt));
1921                 }
1922
1923                 op_data->op_fid1 = rid;
1924                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1925                 op_data->op_namelen = 0;
1926                 op_data->op_name = NULL;
1927                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1928                 ptlrpc_req_finished(*preq);
1929                 *preq = req;
1930         }
1931
1932         RETURN(rc);
1933 }
1934
1935 #define md_op_data_fid(op_data, fl)                     \
1936         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1937          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1938          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1939          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1940          NULL)
1941
1942 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1943                             struct md_op_data *op_data,
1944                             __u32 op_tgt, ldlm_mode_t mode, int bits, int flag)
1945 {
1946         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1947         struct obd_device      *obd = exp->exp_obd;
1948         struct lmv_obd         *lmv = &obd->u.lmv;
1949         ldlm_policy_data_t      policy = {{ 0 }};
1950         int                     rc = 0;
1951         ENTRY;
1952
1953         if (!fid_is_sane(fid))
1954                 RETURN(0);
1955
1956         if (tgt == NULL) {
1957                 tgt = lmv_find_target(lmv, fid);
1958                 if (IS_ERR(tgt))
1959                         RETURN(PTR_ERR(tgt));
1960         }
1961
1962         if (tgt->ltd_idx != op_tgt) {
1963                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1964                 policy.l_inodebits.bits = bits;
1965                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1966                                       mode, LCF_ASYNC, NULL);
1967         } else {
1968                 CDEBUG(D_INODE,
1969                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1970                        op_tgt, PFID(fid));
1971                 op_data->op_flags |= flag;
1972                 rc = 0;
1973         }
1974
1975         RETURN(rc);
1976 }
1977
1978 /*
1979  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1980  * op_data->op_fid2
1981  */
1982 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1983                     struct ptlrpc_request **request)
1984 {
1985         struct obd_device       *obd = exp->exp_obd;
1986         struct lmv_obd          *lmv = &obd->u.lmv;
1987         struct lmv_tgt_desc     *tgt;
1988         int                      rc;
1989         ENTRY;
1990
1991         rc = lmv_check_connect(obd);
1992         if (rc)
1993                 RETURN(rc);
1994
1995         LASSERT(op_data->op_namelen != 0);
1996
1997         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1998                PFID(&op_data->op_fid2), (int)op_data->op_namelen,
1999                op_data->op_name, PFID(&op_data->op_fid1));
2000
2001         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2002         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2003         op_data->op_cap = cfs_curproc_cap_pack();
2004         if (op_data->op_mea2 != NULL) {
2005                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
2006                 const struct lmv_oinfo  *oinfo;
2007
2008                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2009                                                 op_data->op_namelen);
2010                 if (IS_ERR(oinfo))
2011                         RETURN(PTR_ERR(oinfo));
2012
2013                 op_data->op_fid2 = oinfo->lmo_fid;
2014         }
2015
2016         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2017         if (IS_ERR(tgt))
2018                 RETURN(PTR_ERR(tgt));
2019
2020         /*
2021          * Cancel UPDATE lock on child (fid1).
2022          */
2023         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2024         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2025                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2026         if (rc != 0)
2027                 RETURN(rc);
2028
2029         rc = md_link(tgt->ltd_exp, op_data, request);
2030
2031         RETURN(rc);
2032 }
2033
2034 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2035                       const char *old, size_t oldlen,
2036                       const char *new, size_t newlen,
2037                       struct ptlrpc_request **request)
2038 {
2039         struct obd_device       *obd = exp->exp_obd;
2040         struct lmv_obd          *lmv = &obd->u.lmv;
2041         struct lmv_tgt_desc     *src_tgt;
2042         struct lmv_tgt_desc     *tgt_tgt;
2043         struct obd_export       *target_exp;
2044         struct mdt_body         *body;
2045         int                     rc;
2046         ENTRY;
2047
2048         LASSERT(oldlen != 0);
2049
2050         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2051                (int)oldlen, old, PFID(&op_data->op_fid1),
2052                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2053                (int)newlen, new, PFID(&op_data->op_fid2),
2054                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2055
2056         rc = lmv_check_connect(obd);
2057         if (rc)
2058                 RETURN(rc);
2059
2060         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2061         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2062         op_data->op_cap = cfs_curproc_cap_pack();
2063         if (op_data->op_cli_flags & CLI_MIGRATE) {
2064                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2065                          PFID(&op_data->op_fid3));
2066
2067                 if (op_data->op_mea1 != NULL) {
2068                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2069                         struct lmv_tgt_desc     *tmp;
2070
2071                         /* Fix the parent fid for striped dir */
2072                         tmp = lmv_locate_target_for_name(lmv, lsm, old,
2073                                                          oldlen,
2074                                                          &op_data->op_fid1,
2075                                                          NULL);
2076                         if (IS_ERR(tmp))
2077                                 RETURN(PTR_ERR(tmp));
2078                 }
2079
2080                 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2081                 if (rc != 0)
2082                         RETURN(rc);
2083
2084                 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2085                 if (IS_ERR(src_tgt))
2086                         RETURN(PTR_ERR(src_tgt));
2087
2088                 target_exp = src_tgt->ltd_exp;
2089         } else {
2090                 if (op_data->op_mea1 != NULL) {
2091                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2092
2093                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2094                                                              oldlen,
2095                                                              &op_data->op_fid1,
2096                                                              &op_data->op_mds);
2097                 } else {
2098                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2099                 }
2100                 if (IS_ERR(src_tgt))
2101                         RETURN(PTR_ERR(src_tgt));
2102
2103
2104                 if (op_data->op_mea2 != NULL) {
2105                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
2106
2107                         tgt_tgt = lmv_locate_target_for_name(lmv, lsm, new,
2108                                                              newlen,
2109                                                              &op_data->op_fid2,
2110                                                              &op_data->op_mds);
2111                 } else {
2112                         tgt_tgt = lmv_find_target(lmv, &op_data->op_fid2);
2113
2114                 }
2115                 if (IS_ERR(tgt_tgt))
2116                         RETURN(PTR_ERR(tgt_tgt));
2117
2118                 target_exp = tgt_tgt->ltd_exp;
2119         }
2120
2121         /*
2122          * LOOKUP lock on src child (fid3) should also be cancelled for
2123          * src_tgt in mdc_rename.
2124          */
2125         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2126
2127         /*
2128          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2129          * own target.
2130          */
2131         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2132                               LCK_EX, MDS_INODELOCK_UPDATE,
2133                               MF_MDC_CANCEL_FID2);
2134
2135         if (rc != 0)
2136                 RETURN(rc);
2137         /*
2138          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2139          */
2140         if (fid_is_sane(&op_data->op_fid3)) {
2141                 struct lmv_tgt_desc *tgt;
2142
2143                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2144                 if (IS_ERR(tgt))
2145                         RETURN(PTR_ERR(tgt));
2146
2147                 /* Cancel LOOKUP lock on its parent */
2148                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2149                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2150                                       MF_MDC_CANCEL_FID3);
2151                 if (rc != 0)
2152                         RETURN(rc);
2153
2154                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2155                                       LCK_EX, MDS_INODELOCK_FULL,
2156                                       MF_MDC_CANCEL_FID3);
2157                 if (rc != 0)
2158                         RETURN(rc);
2159         }
2160
2161 retry_rename:
2162         /*
2163          * Cancel all the locks on tgt child (fid4).
2164          */
2165         if (fid_is_sane(&op_data->op_fid4)) {
2166                 struct lmv_tgt_desc *tgt;
2167
2168                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2169                                       LCK_EX, MDS_INODELOCK_FULL,
2170                                       MF_MDC_CANCEL_FID4);
2171                 if (rc != 0)
2172                         RETURN(rc);
2173
2174                 tgt = lmv_find_target(lmv, &op_data->op_fid4);
2175                 if (IS_ERR(tgt))
2176                         RETURN(PTR_ERR(tgt));
2177
2178                 /* Since the target child might be destroyed, and it might
2179                  * become orphan, and we can only check orphan on the local
2180                  * MDT right now, so we send rename request to the MDT where
2181                  * target child is located. If target child does not exist,
2182                  * then it will send the request to the target parent */
2183                 target_exp = tgt->ltd_exp;
2184         }
2185
2186         rc = md_rename(target_exp, op_data, old, oldlen, new, newlen,
2187                        request);
2188
2189         if (rc != 0 && rc != -EXDEV)
2190                 RETURN(rc);
2191
2192         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2193         if (body == NULL)
2194                 RETURN(-EPROTO);
2195
2196         /* Not cross-ref case, just get out of here. */
2197         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2198                 RETURN(rc);
2199
2200         CDEBUG(D_INODE, "%s: try rename to another MDT for "DFID"\n",
2201                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2202
2203         op_data->op_fid4 = body->mbo_fid1;
2204         ptlrpc_req_finished(*request);
2205         *request = NULL;
2206         goto retry_rename;
2207 }
2208
2209 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2210                        void *ea, size_t ealen, struct ptlrpc_request **request)
2211 {
2212         struct obd_device       *obd = exp->exp_obd;
2213         struct lmv_obd          *lmv = &obd->u.lmv;
2214         struct lmv_tgt_desc     *tgt;
2215         int                      rc = 0;
2216         ENTRY;
2217
2218         rc = lmv_check_connect(obd);
2219         if (rc)
2220                 RETURN(rc);
2221
2222         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2223                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2224
2225         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2226         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2227         if (IS_ERR(tgt))
2228                 RETURN(PTR_ERR(tgt));
2229
2230         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2231
2232         RETURN(rc);
2233 }
2234
2235 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2236                      struct ptlrpc_request **request)
2237 {
2238         struct obd_device       *obd = exp->exp_obd;
2239         struct lmv_obd          *lmv = &obd->u.lmv;
2240         struct lmv_tgt_desc     *tgt;
2241         int                      rc;
2242         ENTRY;
2243
2244         rc = lmv_check_connect(obd);
2245         if (rc != 0)
2246                 RETURN(rc);
2247
2248         tgt = lmv_find_target(lmv, fid);
2249         if (IS_ERR(tgt))
2250                 RETURN(PTR_ERR(tgt));
2251
2252         rc = md_fsync(tgt->ltd_exp, fid, request);
2253         RETURN(rc);
2254 }
2255
2256 /**
2257  * Get current minimum entry from striped directory
2258  *
2259  * This function will search the dir entry, whose hash value is the
2260  * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2261  * only being called for striped directory.
2262  *
2263  * \param[in] exp               export of LMV
2264  * \param[in] op_data           parameters transferred beween client MD stack
2265  *                              stripe_information will be included in this
2266  *                              parameter
2267  * \param[in] cb_op             ldlm callback being used in enqueue in
2268  *                              mdc_read_page
2269  * \param[in] hash_offset       the hash value, which is used to locate
2270  *                              minum(closet) dir entry
2271  * \param[in|out] stripe_offset the caller use this to indicate the stripe
2272  *                              index of last entry, so to avoid hash conflict
2273  *                              between stripes. It will also be used to
2274  *                              return the stripe index of current dir entry.
2275  * \param[in|out] entp          the minum entry and it also is being used
2276  *                              to input the last dir entry to resolve the
2277  *                              hash conflict
2278  *
2279  * \param[out] ppage            the page which holds the minum entry
2280  *
2281  * \retval                      = 0 get the entry successfully
2282  *                              negative errno (< 0) does not get the entry
2283  */
2284 static int lmv_get_min_striped_entry(struct obd_export *exp,
2285                                      struct md_op_data *op_data,
2286                                      struct md_callback *cb_op,
2287                                      __u64 hash_offset, int *stripe_offset,
2288                                      struct lu_dirent **entp,
2289                                      struct page **ppage)
2290 {
2291         struct obd_device       *obd = exp->exp_obd;
2292         struct lmv_obd          *lmv = &obd->u.lmv;
2293         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2294         struct lmv_tgt_desc     *tgt;
2295         int                     stripe_count;
2296         struct lu_dirent        *min_ent = NULL;
2297         struct page             *min_page = NULL;
2298         int                     min_idx = 0;
2299         int                     i;
2300         int                     rc = 0;
2301         ENTRY;
2302
2303         stripe_count = lsm->lsm_md_stripe_count;
2304         for (i = 0; i < stripe_count; i++) {
2305                 struct lu_dirent        *ent = NULL;
2306                 struct page             *page = NULL;
2307                 struct lu_dirpage       *dp;
2308                 __u64                   stripe_hash = hash_offset;
2309
2310                 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2311                 if (IS_ERR(tgt))
2312                         GOTO(out, rc = PTR_ERR(tgt));
2313
2314                 /* op_data will be shared by each stripe, so we need
2315                  * reset these value for each stripe */
2316                 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2317                 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2318                 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2319 next:
2320                 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2321                                   &page);
2322                 if (rc != 0)
2323                         GOTO(out, rc);
2324
2325                 dp = page_address(page);
2326                 for (ent = lu_dirent_start(dp); ent != NULL;
2327                      ent = lu_dirent_next(ent)) {
2328                         /* Skip dummy entry */
2329                         if (le16_to_cpu(ent->lde_namelen) == 0)
2330                                 continue;
2331
2332                         if (le64_to_cpu(ent->lde_hash) < hash_offset)
2333                                 continue;
2334
2335                         if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2336                             (*entp == ent || i < *stripe_offset))
2337                                 continue;
2338
2339                         /* skip . and .. for other stripes */
2340                         if (i != 0 &&
2341                             (strncmp(ent->lde_name, ".",
2342                                      le16_to_cpu(ent->lde_namelen)) == 0 ||
2343                              strncmp(ent->lde_name, "..",
2344                                      le16_to_cpu(ent->lde_namelen)) == 0))
2345                                 continue;
2346                         break;
2347                 }
2348
2349                 if (ent == NULL) {
2350                         stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2351
2352                         kunmap(page);
2353                         page_cache_release(page);
2354                         page = NULL;
2355
2356                         /* reach the end of current stripe, go to next stripe */
2357                         if (stripe_hash == MDS_DIR_END_OFF)
2358                                 continue;
2359                         else
2360                                 goto next;
2361                 }
2362
2363                 if (min_ent != NULL) {
2364                         if (le64_to_cpu(min_ent->lde_hash) >
2365                             le64_to_cpu(ent->lde_hash)) {
2366                                 min_ent = ent;
2367                                 kunmap(min_page);
2368                                 page_cache_release(min_page);
2369                                 min_idx = i;
2370                                 min_page = page;
2371                         } else {
2372                                 kunmap(page);
2373                                 page_cache_release(page);
2374                                 page = NULL;
2375                         }
2376                 } else {
2377                         min_ent = ent;
2378                         min_page = page;
2379                         min_idx = i;
2380                 }
2381         }
2382
2383 out:
2384         if (*ppage != NULL) {
2385                 kunmap(*ppage);
2386                 page_cache_release(*ppage);
2387         }
2388         *stripe_offset = min_idx;
2389         *entp = min_ent;
2390         *ppage = min_page;
2391         RETURN(rc);
2392 }
2393
2394 /**
2395  * Build dir entry page from a striped directory
2396  *
2397  * This function gets one entry by @offset from a striped directory. It will
2398  * read entries from all of stripes, and choose one closest to the required
2399  * offset(&offset). A few notes
2400  * 1. skip . and .. for non-zero stripes, because there can only have one .
2401  * and .. in a directory.
2402  * 2. op_data will be shared by all of stripes, instead of allocating new
2403  * one, so need to restore before reusing.
2404  * 3. release the entry page if that is not being chosen.
2405  *
2406  * \param[in] exp       obd export refer to LMV
2407  * \param[in] op_data   hold those MD parameters of read_entry
2408  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2409  * \param[out] ldp      the entry being read
2410  * \param[out] ppage    the page holding the entry. Note: because the entry
2411  *                      will be accessed in upper layer, so we need hold the
2412  *                      page until the usages of entry is finished, see
2413  *                      ll_dir_entry_next.
2414  *
2415  * retval               =0 if get entry successfully
2416  *                      <0 cannot get entry
2417  */
2418 static int lmv_read_striped_page(struct obd_export *exp,
2419                                  struct md_op_data *op_data,
2420                                  struct md_callback *cb_op,
2421                                  __u64 offset, struct page **ppage)
2422 {
2423         struct obd_device       *obd = exp->exp_obd;
2424         struct lu_fid           master_fid = op_data->op_fid1;
2425         struct inode            *master_inode = op_data->op_data;
2426         __u64                   hash_offset = offset;
2427         struct lu_dirpage       *dp;
2428         struct page             *min_ent_page = NULL;
2429         struct page             *ent_page = NULL;
2430         struct lu_dirent        *ent;
2431         void                    *area;
2432         int                     ent_idx = 0;
2433         struct lu_dirent        *min_ent = NULL;
2434         struct lu_dirent        *last_ent;
2435         size_t                  left_bytes;
2436         int                     rc;
2437         ENTRY;
2438
2439         rc = lmv_check_connect(obd);
2440         if (rc)
2441                 RETURN(rc);
2442
2443         /* Allocate a page and read entries from all of stripes and fill
2444          * the page by hash order */
2445         ent_page = alloc_page(GFP_KERNEL);
2446         if (ent_page == NULL)
2447                 RETURN(-ENOMEM);
2448
2449         /* Initialize the entry page */
2450         dp = kmap(ent_page);
2451         memset(dp, 0, sizeof(*dp));
2452         dp->ldp_hash_start = cpu_to_le64(offset);
2453         dp->ldp_flags |= LDF_COLLIDE;
2454
2455         area = dp + 1;
2456         left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2457         ent = area;
2458         last_ent = ent;
2459         do {
2460                 __u16   ent_size;
2461
2462                 /* Find the minum entry from all sub-stripes */
2463                 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2464                                                &ent_idx, &min_ent,
2465                                                &min_ent_page);
2466                 if (rc != 0)
2467                         GOTO(out, rc);
2468
2469                 /* If it can not get minum entry, it means it already reaches
2470                  * the end of this directory */
2471                 if (min_ent == NULL) {
2472                         last_ent->lde_reclen = 0;
2473                         hash_offset = MDS_DIR_END_OFF;
2474                         GOTO(out, rc);
2475                 }
2476
2477                 ent_size = le16_to_cpu(min_ent->lde_reclen);
2478
2479                 /* the last entry lde_reclen is 0, but it might not
2480                  * the end of this entry of this temporay entry */
2481                 if (ent_size == 0)
2482                         ent_size = lu_dirent_calc_size(
2483                                         le16_to_cpu(min_ent->lde_namelen),
2484                                         le32_to_cpu(min_ent->lde_attrs));
2485                 if (ent_size > left_bytes) {
2486                         last_ent->lde_reclen = cpu_to_le16(0);
2487                         hash_offset = le64_to_cpu(min_ent->lde_hash);
2488                         GOTO(out, rc);
2489                 }
2490
2491                 memcpy(ent, min_ent, ent_size);
2492
2493                 /* Replace . with master FID and Replace .. with the parent FID
2494                  * of master object */
2495                 if (strncmp(ent->lde_name, ".",
2496                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2497                     le16_to_cpu(ent->lde_namelen) == 1)
2498                         fid_cpu_to_le(&ent->lde_fid, &master_fid);
2499                 else if (strncmp(ent->lde_name, "..",
2500                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2501                            le16_to_cpu(ent->lde_namelen) == 2)
2502                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2503
2504                 left_bytes -= ent_size;
2505                 ent->lde_reclen = cpu_to_le16(ent_size);
2506                 last_ent = ent;
2507                 ent = (void *)ent + ent_size;
2508                 hash_offset = le64_to_cpu(min_ent->lde_hash);
2509                 if (hash_offset == MDS_DIR_END_OFF) {
2510                         last_ent->lde_reclen = 0;
2511                         break;
2512                 }
2513         } while (1);
2514 out:
2515         if (min_ent_page != NULL) {
2516                 kunmap(min_ent_page);
2517                 page_cache_release(min_ent_page);
2518         }
2519
2520         if (unlikely(rc != 0)) {
2521                 __free_page(ent_page);
2522                 ent_page = NULL;
2523         } else {
2524                 if (ent == area)
2525                         dp->ldp_flags |= LDF_EMPTY;
2526                 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2527                 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2528         }
2529
2530         /* We do not want to allocate md_op_data during each
2531          * dir entry reading, so op_data will be shared by every stripe,
2532          * then we need to restore it back to original value before
2533          * return to the upper layer */
2534         op_data->op_fid1 = master_fid;
2535         op_data->op_fid2 = master_fid;
2536         op_data->op_data = master_inode;
2537
2538         *ppage = ent_page;
2539
2540         RETURN(rc);
2541 }
2542
2543 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2544                   struct md_callback *cb_op, __u64 offset,
2545                   struct page **ppage)
2546 {
2547         struct obd_device       *obd = exp->exp_obd;
2548         struct lmv_obd          *lmv = &obd->u.lmv;
2549         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2550         struct lmv_tgt_desc     *tgt;
2551         int                     rc;
2552         ENTRY;
2553
2554         rc = lmv_check_connect(obd);
2555         if (rc != 0)
2556                 RETURN(rc);
2557
2558         if (unlikely(lsm != NULL)) {
2559                 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2560                 RETURN(rc);
2561         }
2562
2563         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2564         if (IS_ERR(tgt))
2565                 RETURN(PTR_ERR(tgt));
2566
2567         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2568
2569         RETURN(rc);
2570 }
2571
2572 /**
2573  * Unlink a file/directory
2574  *
2575  * Unlink a file or directory under the parent dir. The unlink request
2576  * usually will be sent to the MDT where the child is located, but if
2577  * the client does not have the child FID then request will be sent to the
2578  * MDT where the parent is located.
2579  *
2580  * If the parent is a striped directory then it also needs to locate which
2581  * stripe the name of the child is located, and replace the parent FID
2582  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2583  * it will walk through all of sub-stripes until the child is being
2584  * unlinked finally.
2585  *
2586  * \param[in] exp       export refer to LMV
2587  * \param[in] op_data   different parameters transferred beween client
2588  *                      MD stacks, name, namelen, FIDs etc.
2589  *                      op_fid1 is the parent FID, op_fid2 is the child
2590  *                      FID.
2591  * \param[out] request  point to the request of unlink.
2592  *
2593  * retval               0 if succeed
2594  *                      negative errno if failed.
2595  */
2596 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2597                       struct ptlrpc_request **request)
2598 {
2599         struct obd_device       *obd = exp->exp_obd;
2600         struct lmv_obd          *lmv = &obd->u.lmv;
2601         struct lmv_tgt_desc     *tgt = NULL;
2602         struct lmv_tgt_desc     *parent_tgt = NULL;
2603         struct mdt_body         *body;
2604         int                     rc;
2605         int                     stripe_index = 0;
2606         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2607         ENTRY;
2608
2609         rc = lmv_check_connect(obd);
2610         if (rc)
2611                 RETURN(rc);
2612 retry_unlink:
2613         /* For striped dir, we need to locate the parent as well */
2614         if (lsm != NULL) {
2615                 struct lmv_tgt_desc *tmp;
2616
2617                 LASSERT(op_data->op_name != NULL &&
2618                         op_data->op_namelen != 0);
2619
2620                 tmp = lmv_locate_target_for_name(lmv, lsm,
2621                                                  op_data->op_name,
2622                                                  op_data->op_namelen,
2623                                                  &op_data->op_fid1,
2624                                                  &op_data->op_mds);
2625
2626                 /* return -EBADFD means unknown hash type, might
2627                  * need try all sub-stripe here */
2628                 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2629                         RETURN(PTR_ERR(tmp));
2630
2631                 /* Note: both migrating dir and unknown hash dir need to
2632                  * try all of sub-stripes, so we need start search the
2633                  * name from stripe 0, but migrating dir is already handled
2634                  * inside lmv_locate_target_for_name(), so we only check
2635                  * unknown hash type directory here */
2636                 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2637                         struct lmv_oinfo *oinfo;
2638
2639                         oinfo = &lsm->lsm_md_oinfo[stripe_index];
2640
2641                         op_data->op_fid1 = oinfo->lmo_fid;
2642                         op_data->op_mds = oinfo->lmo_mds;
2643                 }
2644         }
2645
2646 try_next_stripe:
2647         /* Send unlink requests to the MDT where the child is located */
2648         if (likely(!fid_is_zero(&op_data->op_fid2)))
2649                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2650         else if (lsm != NULL)
2651                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2652         else
2653                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2654
2655         if (IS_ERR(tgt))
2656                 RETURN(PTR_ERR(tgt));
2657
2658         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2659         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2660         op_data->op_cap = cfs_curproc_cap_pack();
2661
2662         /*
2663          * If child's fid is given, cancel unused locks for it if it is from
2664          * another export than parent.
2665          *
2666          * LOOKUP lock for child (fid3) should also be cancelled on parent
2667          * tgt_tgt in mdc_unlink().
2668          */
2669         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2670
2671         /*
2672          * Cancel FULL locks on child (fid3).
2673          */
2674         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2675         if (IS_ERR(parent_tgt))
2676                 RETURN(PTR_ERR(parent_tgt));
2677
2678         if (parent_tgt != tgt) {
2679                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2680                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2681                                       MF_MDC_CANCEL_FID3);
2682         }
2683
2684         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2685                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2686         if (rc != 0)
2687                 RETURN(rc);
2688
2689         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2690                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2691
2692         rc = md_unlink(tgt->ltd_exp, op_data, request);
2693         if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2694                 RETURN(rc);
2695
2696         /* Try next stripe if it is needed. */
2697         if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2698                 struct lmv_oinfo *oinfo;
2699
2700                 stripe_index++;
2701                 if (stripe_index >= lsm->lsm_md_stripe_count)
2702                         RETURN(rc);
2703
2704                 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2705
2706                 op_data->op_fid1 = oinfo->lmo_fid;
2707                 op_data->op_mds = oinfo->lmo_mds;
2708
2709                 ptlrpc_req_finished(*request);
2710                 *request = NULL;
2711
2712                 goto try_next_stripe;
2713         }
2714
2715         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2716         if (body == NULL)
2717                 RETURN(-EPROTO);
2718
2719         /* Not cross-ref case, just get out of here. */
2720         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2721                 RETURN(rc);
2722
2723         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2724                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2725
2726         /* This is a remote object, try remote MDT, Note: it may
2727          * try more than 1 time here, Considering following case
2728          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2729          * 1. Initially A does not know where remote1 is, it send
2730          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2731          *    resend unlink RPC to MDT1 (retry 1st time).
2732          *
2733          * 2. During the unlink RPC in flight,
2734          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2735          *    and create new remote1, but on MDT0
2736          *
2737          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2738          *    /mnt/lustre, then lookup get fid of remote1, and find
2739          *    it is remote dir again, and replay -EREMOTE again.
2740          *
2741          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2742          *
2743          * In theory, it might try unlimited time here, but it should
2744          * be very rare case.  */
2745         op_data->op_fid2 = body->mbo_fid1;
2746         ptlrpc_req_finished(*request);
2747         *request = NULL;
2748
2749         goto retry_unlink;
2750 }
2751
2752 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2753 {
2754         struct lmv_obd *lmv = &obd->u.lmv;
2755         int rc = 0;
2756
2757         switch (stage) {
2758         case OBD_CLEANUP_EARLY:
2759                 /* XXX: here should be calling obd_precleanup() down to
2760                  * stack. */
2761                 break;
2762         case OBD_CLEANUP_EXPORTS:
2763                 fld_client_proc_fini(&lmv->lmv_fld);
2764                 lprocfs_obd_cleanup(obd);
2765                 lprocfs_free_md_stats(obd);
2766                 break;
2767         default:
2768                 break;
2769         }
2770         RETURN(rc);
2771 }
2772
2773 /**
2774  * Get by key a value associated with a LMV device.
2775  *
2776  * Dispatch request to lower-layer devices as needed.
2777  *
2778  * \param[in] env               execution environment for this thread
2779  * \param[in] exp               export for the LMV device
2780  * \param[in] keylen            length of key identifier
2781  * \param[in] key               identifier of key to get value for
2782  * \param[in] vallen            size of \a val
2783  * \param[out] val              pointer to storage location for value
2784  * \param[in] lsm               optional striping metadata of object
2785  *
2786  * \retval 0            on success
2787  * \retval negative     negated errno on failure
2788  */
2789 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2790                         __u32 keylen, void *key, __u32 *vallen, void *val)
2791 {
2792         struct obd_device       *obd;
2793         struct lmv_obd          *lmv;
2794         int                      rc = 0;
2795         ENTRY;
2796
2797         obd = class_exp2obd(exp);
2798         if (obd == NULL) {
2799                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2800                        exp->exp_handle.h_cookie);
2801                 RETURN(-EINVAL);
2802         }
2803
2804         lmv = &obd->u.lmv;
2805         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2806                 int i;
2807
2808                 rc = lmv_check_connect(obd);
2809                 if (rc)
2810                         RETURN(rc);
2811
2812                 LASSERT(*vallen == sizeof(__u32));
2813                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2814                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2815                         /*
2816                          * All tgts should be connected when this gets called.
2817                          */
2818                         if (tgt == NULL || tgt->ltd_exp == NULL)
2819                                 continue;
2820
2821                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2822                                           vallen, val))
2823                                 RETURN(0);
2824                 }
2825                 RETURN(-EINVAL);
2826         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2827                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2828                    KEY_IS(KEY_CONN_DATA)) {
2829                 rc = lmv_check_connect(obd);
2830                 if (rc)
2831                         RETURN(rc);
2832
2833                 /*
2834                  * Forwarding this request to first MDS, it should know LOV
2835                  * desc.
2836                  */
2837                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2838                                   vallen, val);
2839                 if (!rc && KEY_IS(KEY_CONN_DATA))
2840                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2841                 RETURN(rc);
2842         } else if (KEY_IS(KEY_TGT_COUNT)) {
2843                 *((int *)val) = lmv->desc.ld_tgt_count;
2844                 RETURN(0);
2845         }
2846
2847         CDEBUG(D_IOCTL, "Invalid key\n");
2848         RETURN(-EINVAL);
2849 }
2850
2851 /**
2852  * Asynchronously set by key a value associated with a LMV device.
2853  *
2854  * Dispatch request to lower-layer devices as needed.
2855  *
2856  * \param[in] env       execution environment for this thread
2857  * \param[in] exp       export for the LMV device
2858  * \param[in] keylen    length of key identifier
2859  * \param[in] key       identifier of key to store value for
2860  * \param[in] vallen    size of value to store
2861  * \param[in] val       pointer to data to be stored
2862  * \param[in] set       optional list of related ptlrpc requests
2863  *
2864  * \retval 0            on success
2865  * \retval negative     negated errno on failure
2866  */
2867 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2868                         __u32 keylen, void *key, __u32 vallen, void *val,
2869                         struct ptlrpc_request_set *set)
2870 {
2871         struct lmv_tgt_desc     *tgt = NULL;
2872         struct obd_device       *obd;
2873         struct lmv_obd          *lmv;
2874         int rc = 0;
2875         ENTRY;
2876
2877         obd = class_exp2obd(exp);
2878         if (obd == NULL) {
2879                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2880                        exp->exp_handle.h_cookie);
2881                 RETURN(-EINVAL);
2882         }
2883         lmv = &obd->u.lmv;
2884
2885         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2886             KEY_IS(KEY_DEFAULT_EASIZE)) {
2887                 int i, err = 0;
2888
2889                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2890                         tgt = lmv->tgts[i];
2891
2892                         if (tgt == NULL || tgt->ltd_exp == NULL)
2893                                 continue;
2894
2895                         err = obd_set_info_async(env, tgt->ltd_exp,
2896                                                  keylen, key, vallen, val, set);
2897                         if (err && rc == 0)
2898                                 rc = err;
2899                 }
2900
2901                 RETURN(rc);
2902         }
2903
2904         RETURN(-EINVAL);
2905 }
2906
2907 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2908                             const struct lmv_mds_md_v1 *lmm1)
2909 {
2910         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
2911         int             stripe_count;
2912         int             cplen;
2913         int             i;
2914         int             rc = 0;
2915         ENTRY;
2916
2917         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2918         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2919         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2920         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2921                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2922         else
2923                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2924         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2925         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2926                         sizeof(lsm->lsm_md_pool_name));
2927
2928         if (cplen >= sizeof(lsm->lsm_md_pool_name))
2929                 RETURN(-E2BIG);
2930
2931         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2932                "layout_version %d\n", lsm->lsm_md_stripe_count,
2933                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2934                lsm->lsm_md_layout_version);
2935
2936         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2937         for (i = 0; i < stripe_count; i++) {
2938                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2939                               &lmm1->lmv_stripe_fids[i]);
2940                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2941                                     &lsm->lsm_md_oinfo[i].lmo_mds);
2942                 if (rc != 0)
2943                         RETURN(rc);
2944                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2945                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2946         }
2947
2948         RETURN(rc);
2949 }
2950
2951 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2952                         const union lmv_mds_md *lmm, size_t lmm_size)
2953 {
2954         struct lmv_stripe_md     *lsm;
2955         int                      lsm_size;
2956         int                      rc;
2957         bool                     allocated = false;
2958         ENTRY;
2959
2960         LASSERT(lsmp != NULL);
2961
2962         lsm = *lsmp;
2963         /* Free memmd */
2964         if (lsm != NULL && lmm == NULL) {
2965                 int i;
2966                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2967                         /* For migrating inode, the master stripe and master
2968                          * object will be the same, so do not need iput, see
2969                          * ll_update_lsm_md */
2970                         if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2971                               i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2972                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
2973                 }
2974                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2975                 OBD_FREE(lsm, lsm_size);
2976                 *lsmp = NULL;
2977                 RETURN(0);
2978         }
2979
2980         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
2981                 RETURN(-EPERM);
2982
2983         /* Unpack memmd */
2984         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
2985             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
2986                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
2987                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
2988                        -EIO);
2989                 RETURN(-EIO);
2990         }
2991
2992         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
2993                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2994         else
2995                 /**
2996                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
2997                  * stripecount should be 0 then.
2998                  */
2999                 lsm_size = lmv_stripe_md_size(0);
3000
3001         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3002         if (lsm == NULL) {
3003                 OBD_ALLOC(lsm, lsm_size);
3004                 if (lsm == NULL)
3005                         RETURN(-ENOMEM);
3006                 allocated = true;
3007                 *lsmp = lsm;
3008         }
3009
3010         switch (le32_to_cpu(lmm->lmv_magic)) {
3011         case LMV_MAGIC_V1:
3012                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3013                 break;
3014         default:
3015                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3016                        le32_to_cpu(lmm->lmv_magic));
3017                 rc = -EINVAL;
3018                 break;
3019         }
3020
3021         if (rc != 0 && allocated) {
3022                 OBD_FREE(lsm, lsm_size);
3023                 *lsmp = NULL;
3024                 lsm_size = rc;
3025         }
3026         RETURN(lsm_size);
3027 }
3028
3029 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3030 {
3031         lmv_unpackmd(NULL, &lsm, NULL, 0);
3032 }
3033 EXPORT_SYMBOL(lmv_free_memmd);
3034
3035 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3036                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
3037                              ldlm_cancel_flags_t flags, void *opaque)
3038 {
3039         struct obd_device       *obd = exp->exp_obd;
3040         struct lmv_obd          *lmv = &obd->u.lmv;
3041         int                      rc = 0;
3042         int                      err;
3043         __u32                    i;
3044         ENTRY;
3045
3046         LASSERT(fid != NULL);
3047
3048         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3049                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3050
3051                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3052                         continue;
3053
3054                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3055                                        opaque);
3056                 if (!rc)
3057                         rc = err;
3058         }
3059         RETURN(rc);
3060 }
3061
3062 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3063                       __u64 *bits)
3064 {
3065         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3066         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3067         int                      rc;
3068         ENTRY;
3069
3070         if (tgt == NULL || tgt->ltd_exp == NULL)
3071                 RETURN(-EINVAL);
3072         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3073         RETURN(rc);
3074 }
3075
3076 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3077                            const struct lu_fid *fid, ldlm_type_t type,
3078                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
3079                            struct lustre_handle *lockh)
3080 {
3081         struct obd_device       *obd = exp->exp_obd;
3082         struct lmv_obd          *lmv = &obd->u.lmv;
3083         ldlm_mode_t             rc;
3084         int                     tgt;
3085         int                     i;
3086         ENTRY;
3087
3088         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3089
3090         /*
3091          * With DNE every object can have two locks in different namespaces:
3092          * lookup lock in space of MDT storing direntry and update/open lock in
3093          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3094          * since this can be easily found, and only try others if that fails.
3095          */
3096         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3097              i < lmv->desc.ld_tgt_count;
3098              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3099                 if (tgt < 0) {
3100                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3101                                obd->obd_name, PFID(fid), tgt);
3102                         tgt = 0;
3103                 }
3104
3105                 if (lmv->tgts[tgt] == NULL ||
3106                     lmv->tgts[tgt]->ltd_exp == NULL ||
3107                     lmv->tgts[tgt]->ltd_active == 0)
3108                         continue;
3109
3110                 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3111                                    type, policy, mode, lockh);
3112                 if (rc)
3113                         RETURN(rc);
3114         }
3115
3116         RETURN(0);
3117 }
3118
3119 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3120                       struct obd_export *dt_exp, struct obd_export *md_exp,
3121                       struct lustre_md *md)
3122 {
3123         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3124         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3125
3126         if (tgt == NULL || tgt->ltd_exp == NULL)
3127                 RETURN(-EINVAL);
3128
3129         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3130 }
3131
3132 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3133 {
3134         struct obd_device       *obd = exp->exp_obd;
3135         struct lmv_obd          *lmv = &obd->u.lmv;
3136         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3137         ENTRY;
3138
3139         if (md->lmv != NULL) {
3140                 lmv_free_memmd(md->lmv);
3141                 md->lmv = NULL;
3142         }
3143         if (tgt == NULL || tgt->ltd_exp == NULL)
3144                 RETURN(-EINVAL);
3145         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3146 }
3147
3148 int lmv_set_open_replay_data(struct obd_export *exp,
3149                              struct obd_client_handle *och,
3150                              struct lookup_intent *it)
3151 {
3152         struct obd_device       *obd = exp->exp_obd;
3153         struct lmv_obd          *lmv = &obd->u.lmv;
3154         struct lmv_tgt_desc     *tgt;
3155         ENTRY;
3156
3157         tgt = lmv_find_target(lmv, &och->och_fid);
3158         if (IS_ERR(tgt))
3159                 RETURN(PTR_ERR(tgt));
3160
3161         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3162 }
3163
3164 int lmv_clear_open_replay_data(struct obd_export *exp,
3165                                struct obd_client_handle *och)
3166 {
3167         struct obd_device       *obd = exp->exp_obd;
3168         struct lmv_obd          *lmv = &obd->u.lmv;
3169         struct lmv_tgt_desc     *tgt;
3170         ENTRY;
3171
3172         tgt = lmv_find_target(lmv, &och->och_fid);
3173         if (IS_ERR(tgt))
3174                 RETURN(PTR_ERR(tgt));
3175
3176         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3177 }
3178
3179 static int lmv_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
3180                                u32 suppgid, struct ptlrpc_request **request)
3181 {
3182         struct obd_device       *obd = exp->exp_obd;
3183         struct lmv_obd          *lmv = &obd->u.lmv;
3184         struct lmv_tgt_desc     *tgt;
3185         int                      rc;
3186         ENTRY;
3187
3188         rc = lmv_check_connect(obd);
3189         if (rc)
3190                 RETURN(rc);
3191
3192         tgt = lmv_find_target(lmv, fid);
3193         if (IS_ERR(tgt))
3194                 RETURN(PTR_ERR(tgt));
3195
3196         rc = md_get_remote_perm(tgt->ltd_exp, fid, suppgid, request);
3197         RETURN(rc);
3198 }
3199
3200 int lmv_intent_getattr_async(struct obd_export *exp,
3201                              struct md_enqueue_info *minfo,
3202                              struct ldlm_enqueue_info *einfo)
3203 {
3204         struct md_op_data       *op_data = &minfo->mi_data;
3205         struct obd_device       *obd = exp->exp_obd;
3206         struct lmv_obd          *lmv = &obd->u.lmv;
3207         struct lmv_tgt_desc     *tgt = NULL;
3208         int                      rc;
3209         ENTRY;
3210
3211         rc = lmv_check_connect(obd);
3212         if (rc)
3213                 RETURN(rc);
3214
3215         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3216         if (IS_ERR(tgt))
3217                 RETURN(PTR_ERR(tgt));
3218
3219         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3220         RETURN(rc);
3221 }
3222
3223 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3224                         struct lu_fid *fid, __u64 *bits)
3225 {
3226         struct obd_device       *obd = exp->exp_obd;
3227         struct lmv_obd          *lmv = &obd->u.lmv;
3228         struct lmv_tgt_desc     *tgt;
3229         int                      rc;
3230         ENTRY;
3231
3232         rc = lmv_check_connect(obd);
3233         if (rc)
3234                 RETURN(rc);
3235
3236         tgt = lmv_find_target(lmv, fid);
3237         if (IS_ERR(tgt))
3238                 RETURN(PTR_ERR(tgt));
3239
3240         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3241         RETURN(rc);
3242 }
3243
3244 int lmv_get_fid_from_lsm(struct obd_export *exp,
3245                          const struct lmv_stripe_md *lsm,
3246                          const char *name, int namelen, struct lu_fid *fid)
3247 {
3248         const struct lmv_oinfo *oinfo;
3249
3250         LASSERT(lsm != NULL);
3251         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3252         if (IS_ERR(oinfo))
3253                 return PTR_ERR(oinfo);
3254
3255         *fid = oinfo->lmo_fid;
3256
3257         RETURN(0);
3258 }
3259
3260 /**
3261  * For lmv, only need to send request to master MDT, and the master MDT will
3262  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3263  * we directly fetch data from the slave MDTs.
3264  */
3265 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3266                  struct obd_quotactl *oqctl)
3267 {
3268         struct obd_device   *obd = class_exp2obd(exp);
3269         struct lmv_obd      *lmv = &obd->u.lmv;
3270         struct lmv_tgt_desc *tgt = lmv->tgts[0];
3271         int                  rc = 0;
3272         __u32                i;
3273         __u64                curspace, curinodes;
3274         ENTRY;
3275
3276         if (tgt == NULL ||
3277             tgt->ltd_exp == NULL ||
3278             !tgt->ltd_active ||
3279             lmv->desc.ld_tgt_count == 0) {
3280                 CERROR("master lmv inactive\n");
3281                 RETURN(-EIO);
3282         }
3283
3284         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3285                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3286                 RETURN(rc);
3287         }
3288
3289         curspace = curinodes = 0;
3290         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3291                 int err;
3292                 tgt = lmv->tgts[i];
3293
3294                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3295                         continue;
3296
3297                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3298                 if (err) {
3299                         CERROR("getquota on mdt %d failed. %d\n", i, err);
3300                         if (!rc)
3301                                 rc = err;
3302                 } else {
3303                         curspace += oqctl->qc_dqblk.dqb_curspace;
3304                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3305                 }
3306         }
3307         oqctl->qc_dqblk.dqb_curspace = curspace;
3308         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3309
3310         RETURN(rc);
3311 }
3312
3313 static int lmv_merge_attr(struct obd_export *exp,
3314                           const struct lmv_stripe_md *lsm,
3315                           struct cl_attr *attr,
3316                           ldlm_blocking_callback cb_blocking)
3317 {
3318         int rc;
3319         int i;
3320
3321         rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3322         if (rc < 0)
3323                 return rc;
3324
3325         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3326                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3327
3328                 CDEBUG(D_INFO, ""DFID" size %llu, blocks %llu nlink %u,"
3329                        " atime %lu ctime %lu, mtime %lu.\n",
3330                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3331                        i_size_read(inode), (unsigned long long)inode->i_blocks,
3332                        inode->i_nlink, LTIME_S(inode->i_atime),
3333                        LTIME_S(inode->i_ctime), LTIME_S(inode->i_mtime));
3334
3335                 /* for slave stripe, it needs to subtract nlink for . and .. */
3336                 if (i != 0)
3337                         attr->cat_nlink += inode->i_nlink - 2;
3338                 else
3339                         attr->cat_nlink = inode->i_nlink;
3340
3341                 attr->cat_size += i_size_read(inode);
3342                 attr->cat_blocks += inode->i_blocks;
3343
3344                 if (attr->cat_atime < LTIME_S(inode->i_atime))
3345                         attr->cat_atime = LTIME_S(inode->i_atime);
3346
3347                 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3348                         attr->cat_ctime = LTIME_S(inode->i_ctime);
3349
3350                 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3351                         attr->cat_mtime = LTIME_S(inode->i_mtime);
3352         }
3353         return 0;
3354 }
3355
3356 struct obd_ops lmv_obd_ops = {
3357         .o_owner                = THIS_MODULE,
3358         .o_setup                = lmv_setup,
3359         .o_cleanup              = lmv_cleanup,
3360         .o_precleanup           = lmv_precleanup,
3361         .o_process_config       = lmv_process_config,
3362         .o_connect              = lmv_connect,
3363         .o_disconnect           = lmv_disconnect,
3364         .o_statfs               = lmv_statfs,
3365         .o_get_info             = lmv_get_info,
3366         .o_set_info_async       = lmv_set_info_async,
3367         .o_notify               = lmv_notify,
3368         .o_get_uuid             = lmv_get_uuid,
3369         .o_iocontrol            = lmv_iocontrol,
3370         .o_quotactl             = lmv_quotactl
3371 };
3372
3373 struct md_ops lmv_md_ops = {
3374         .m_getstatus            = lmv_getstatus,
3375         .m_null_inode           = lmv_null_inode,
3376         .m_find_cbdata          = lmv_find_cbdata,
3377         .m_close                = lmv_close,
3378         .m_create               = lmv_create,
3379         .m_enqueue              = lmv_enqueue,
3380         .m_getattr              = lmv_getattr,
3381         .m_getxattr             = lmv_getxattr,
3382         .m_getattr_name         = lmv_getattr_name,
3383         .m_intent_lock          = lmv_intent_lock,
3384         .m_link                 = lmv_link,
3385         .m_rename               = lmv_rename,
3386         .m_setattr              = lmv_setattr,
3387         .m_setxattr             = lmv_setxattr,
3388         .m_fsync                = lmv_fsync,
3389         .m_read_page            = lmv_read_page,
3390         .m_unlink               = lmv_unlink,
3391         .m_init_ea_size         = lmv_init_ea_size,
3392         .m_cancel_unused        = lmv_cancel_unused,
3393         .m_set_lock_data        = lmv_set_lock_data,
3394         .m_lock_match           = lmv_lock_match,
3395         .m_get_lustre_md        = lmv_get_lustre_md,
3396         .m_free_lustre_md       = lmv_free_lustre_md,
3397         .m_merge_attr           = lmv_merge_attr,
3398         .m_set_open_replay_data = lmv_set_open_replay_data,
3399         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3400         .m_get_remote_perm      = lmv_get_remote_perm,
3401         .m_intent_getattr_async = lmv_intent_getattr_async,
3402         .m_revalidate_lock      = lmv_revalidate_lock,
3403         .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
3404         .m_unpackmd             = lmv_unpackmd,
3405 };
3406
3407 int __init lmv_init(void)
3408 {
3409         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3410                                    LUSTRE_LMV_NAME, NULL);
3411 }
3412
3413 static void lmv_exit(void)
3414 {
3415         class_unregister_type(LUSTRE_LMV_NAME);
3416 }
3417
3418 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3419 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3420 MODULE_LICENSE("GPL");
3421
3422 module_init(lmv_init);
3423 module_exit(lmv_exit);