lustre/lmv/lmv_obd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LMV
  38 #ifdef __KERNEL__
  39 #include <linux/slab.h>
  40 #include <linux/module.h>
  41 #include <linux/init.h>
  42 #include <linux/slab.h>
  43 #include <linux/pagemap.h>
  44 #include <linux/mm.h>
  45 #include <asm/div64.h>
  46 #include <linux/seq_file.h>
  47 #include <linux/namei.h>
  48 #else
  49 #include <liblustre.h>
  50 #endif
  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 <lprocfs_status.h>
  58 #include <lustre_lite.h>
  59 #include <lustre_fid.h>
  60 #include "lmv_internal.h"
  61
  62 static void lmv_activate_target(struct lmv_obd *lmv,
  63                                 struct lmv_tgt_desc *tgt,
  64                                 int activate)
  65 {
  66         if (tgt->ltd_active == activate)
  67                 return;
  68
  69         tgt->ltd_active = activate;
  70         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  71 }
  72
  73 /**
  74  * Error codes:
  75  *
  76  *  -EINVAL  : UUID can't be found in the LMV's target list
  77  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  78  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  79  */
  80 static int lmv_set_mdc_active(struct lmv_obd *lmv,
  81                               const struct obd_uuid *uuid,
  82                               int activate)
  83 {
  84         struct lmv_tgt_desc     *tgt = NULL;
  85         struct obd_device       *obd;
  86         __u32                    i;
  87         int                      rc = 0;
  88         ENTRY;
  89
  90         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  91                         lmv, uuid->uuid, activate);
  92
  93         spin_lock(&lmv->lmv_lock);
  94         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
  95                 tgt = lmv->tgts[i];
  96                 if (tgt == NULL || tgt->ltd_exp == NULL)
  97                         continue;
  98
  99                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
 100                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 101
 102                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 103                         break;
 104         }
 105
 106         if (i == lmv->desc.ld_tgt_count)
 107                 GOTO(out_lmv_lock, rc = -EINVAL);
 108
 109         obd = class_exp2obd(tgt->ltd_exp);
 110         if (obd == NULL)
 111                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 112
 113         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 114                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 115                obd->obd_type->typ_name, i);
 116         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 117
 118         if (tgt->ltd_active == activate) {
 119                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 120                        activate ? "" : "in");
 121                 GOTO(out_lmv_lock, rc);
 122         }
 123
 124         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 125                activate ? "" : "in");
 126         lmv_activate_target(lmv, tgt, activate);
 127         EXIT;
 128
 129  out_lmv_lock:
 130         spin_unlock(&lmv->lmv_lock);
 131         return rc;
 132 }
 133
 134 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 135 {
 136         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
 137         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
 138
 139         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
 140 }
 141
 142 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 143                       enum obd_notify_event ev, void *data)
 144 {
 145         struct obd_connect_data *conn_data;
 146         struct lmv_obd          *lmv = &obd->u.lmv;
 147         struct obd_uuid         *uuid;
 148         int                      rc = 0;
 149         ENTRY;
 150
 151         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 152                 CERROR("unexpected notification of %s %s!\n",
 153                        watched->obd_type->typ_name,
 154                        watched->obd_name);
 155                 RETURN(-EINVAL);
 156         }
 157
 158         uuid = &watched->u.cli.cl_target_uuid;
 159         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 160                 /*
 161                  * Set MDC as active before notifying the observer, so the
 162                  * observer can use the MDC normally.
 163                  */
 164                 rc = lmv_set_mdc_active(lmv, uuid,
 165                                         ev == OBD_NOTIFY_ACTIVE);
 166                 if (rc) {
 167                         CERROR("%sactivation of %s failed: %d\n",
 168                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 169                                uuid->uuid, rc);
 170                         RETURN(rc);
 171                 }
 172         } else if (ev == OBD_NOTIFY_OCD) {
 173                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 174                 /*
 175                  * XXX: Make sure that ocd_connect_flags from all targets are
 176                  * the same. Otherwise one of MDTs runs wrong version or
 177                  * something like this.  --umka
 178                  */
 179                 obd->obd_self_export->exp_connect_data = *conn_data;
 180         }
 181 #if 0
 182         else if (ev == OBD_NOTIFY_DISCON) {
 183                 /*
 184                  * For disconnect event, flush fld cache for failout MDS case.
 185                  */
 186                 fld_client_flush(&lmv->lmv_fld);
 187         }
 188 #endif
 189         /*
 190          * Pass the notification up the chain.
 191          */
 192         if (obd->obd_observer)
 193                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 194
 195         RETURN(rc);
 196 }
 197
 198 /**
 199  * This is fake connect function. Its purpose is to initialize lmv and say
 200  * caller that everything is okay. Real connection will be performed later.
 201  */
 202 static int lmv_connect(const struct lu_env *env,
 203                        struct obd_export **exp, struct obd_device *obd,
 204                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 205                        void *localdata)
 206 {
 207 #ifdef __KERNEL__
 208         struct proc_dir_entry *lmv_proc_dir;
 209 #endif
 210         struct lmv_obd        *lmv = &obd->u.lmv;
 211         struct lustre_handle  conn = { 0 };
 212         int                    rc = 0;
 213         ENTRY;
 214
 215         /*
 216          * We don't want to actually do the underlying connections more than
 217          * once, so keep track.
 218          */
 219         lmv->refcount++;
 220         if (lmv->refcount > 1) {
 221                 *exp = NULL;
 222                 RETURN(0);
 223         }
 224
 225         rc = class_connect(&conn, obd, cluuid);
 226         if (rc) {
 227                 CERROR("class_connection() returned %d\n", rc);
 228                 RETURN(rc);
 229         }
 230
 231         *exp = class_conn2export(&conn);
 232         class_export_get(*exp);
 233
 234         lmv->exp = *exp;
 235         lmv->connected = 0;
 236         lmv->cluuid = *cluuid;
 237
 238         if (data)
 239                 lmv->conn_data = *data;
 240
 241 #ifdef __KERNEL__
 242         lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
 243                                         NULL, NULL);
 244         if (IS_ERR(lmv_proc_dir)) {
 245                 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 246                        obd->obd_type->typ_name, obd->obd_name);
 247                 lmv_proc_dir = NULL;
 248         }
 249 #endif
 250
 251         /*
 252          * All real clients should perform actual connection right away, because
 253          * it is possible, that LMV will not have opportunity to connect targets
 254          * and MDC stuff will be called directly, for instance while reading
 255          * ../mdc/../kbytesfree procfs file, etc.
 256          */
 257         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
 258                 rc = lmv_check_connect(obd);
 259
 260 #ifdef __KERNEL__
 261         if (rc) {
 262                 if (lmv_proc_dir)
 263                         lprocfs_remove(&lmv_proc_dir);
 264         }
 265 #endif
 266
 267         RETURN(rc);
 268 }
 269
 270 static void lmv_set_timeouts(struct obd_device *obd)
 271 {
 272         struct lmv_obd          *lmv;
 273         __u32                    i;
 274
 275         lmv = &obd->u.lmv;
 276         if (lmv->server_timeout == 0)
 277                 return;
 278
 279         if (lmv->connected == 0)
 280                 return;
 281
 282         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 283                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 284
 285                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
 286                         continue;
 287
 288                 obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
 289                                    KEY_INTERMDS, 0, NULL, NULL);
 290         }
 291 }
 292
 293 static int lmv_init_ea_size(struct obd_export *exp, int easize,
 294                             int def_easize, int cookiesize)
 295 {
 296         struct obd_device       *obd = exp->exp_obd;
 297         struct lmv_obd          *lmv = &obd->u.lmv;
 298         __u32                    i;
 299         int                      rc = 0;
 300         int                      change = 0;
 301         ENTRY;
 302
 303         if (lmv->max_easize < easize) {
 304                 lmv->max_easize = easize;
 305                 change = 1;
 306         }
 307         if (lmv->max_def_easize < def_easize) {
 308                 lmv->max_def_easize = def_easize;
 309                 change = 1;
 310         }
 311         if (lmv->max_cookiesize < cookiesize) {
 312                 lmv->max_cookiesize = cookiesize;
 313                 change = 1;
 314         }
 315         if (change == 0)
 316                 RETURN(0);
 317
 318         if (lmv->connected == 0)
 319                 RETURN(0);
 320
 321         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 322                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 323
 324                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
 325                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 326                         continue;
 327                 }
 328
 329                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
 330                                      cookiesize);
 331                 if (rc) {
 332                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 333                                " rc = %d.\n", obd->obd_name, i, rc);
 334                         break;
 335                 }
 336         }
 337         RETURN(rc);
 338 }
 339
 340 #define MAX_STRING_SIZE 128
 341
 342 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 343 {
 344 #ifdef __KERNEL__
 345         struct proc_dir_entry   *lmv_proc_dir;
 346 #endif
 347         struct lmv_obd          *lmv = &obd->u.lmv;
 348         struct obd_uuid         *cluuid = &lmv->cluuid;
 349         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 350         struct obd_device       *mdc_obd;
 351         struct obd_export       *mdc_exp;
 352         struct lu_fld_target     target;
 353         int                      rc;
 354         ENTRY;
 355
 356         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 357                                         &obd->obd_uuid);
 358         if (!mdc_obd) {
 359                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 360                 RETURN(-EINVAL);
 361         }
 362
 363         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 364                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 365                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 366                 cluuid->uuid);
 367
 368         if (!mdc_obd->obd_set_up) {
 369                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 370                 RETURN(-EINVAL);
 371         }
 372
 373         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 374                          &lmv->conn_data, NULL);
 375         if (rc) {
 376                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 377                 RETURN(rc);
 378         }
 379
 380         /*
 381          * Init fid sequence client for this mdc and add new fld target.
 382          */
 383         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 384         if (rc)
 385                 RETURN(rc);
 386
 387         target.ft_srv = NULL;
 388         target.ft_exp = mdc_exp;
 389         target.ft_idx = tgt->ltd_idx;
 390
 391         fld_client_add_target(&lmv->lmv_fld, &target);
 392
 393         rc = obd_register_observer(mdc_obd, obd);
 394         if (rc) {
 395                 obd_disconnect(mdc_exp);
 396                 CERROR("target %s register_observer error %d\n",
 397                        tgt->ltd_uuid.uuid, rc);
 398                 RETURN(rc);
 399         }
 400
 401         if (obd->obd_observer) {
 402                 /*
 403                  * Tell the observer about the new target.
 404                  */
 405                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 406                                 OBD_NOTIFY_ACTIVE,
 407                                 (void *)(tgt - lmv->tgts[0]));
 408                 if (rc) {
 409                         obd_disconnect(mdc_exp);
 410                         RETURN(rc);
 411                 }
 412         }
 413
 414         tgt->ltd_active = 1;
 415         tgt->ltd_exp = mdc_exp;
 416         lmv->desc.ld_active_tgt_count++;
 417
 418         md_init_ea_size(tgt->ltd_exp, lmv->max_easize,
 419                         lmv->max_def_easize, lmv->max_cookiesize);
 420
 421         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 422                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 423                 cfs_atomic_read(&obd->obd_refcount));
 424
 425 #ifdef __KERNEL__
 426         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 427         if (lmv_proc_dir) {
 428                 struct proc_dir_entry *mdc_symlink;
 429
 430                 LASSERT(mdc_obd->obd_type != NULL);
 431                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 432                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 433                                                   lmv_proc_dir,
 434                                                   "../../../%s/%s",
 435                                                   mdc_obd->obd_type->typ_name,
 436                                                   mdc_obd->obd_name);
 437                 if (mdc_symlink == NULL) {
 438                         CERROR("Could not register LMV target "
 439                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 440                                obd->obd_type->typ_name, obd->obd_name,
 441                                mdc_obd->obd_name);
 442                         lprocfs_remove(&lmv_proc_dir);
 443                         lmv_proc_dir = NULL;
 444                 }
 445         }
 446 #endif
 447         RETURN(0);
 448 }
 449
 450 static void lmv_del_target(struct lmv_obd *lmv, int index)
 451 {
 452         if (lmv->tgts[index] == NULL)
 453                 return;
 454
 455         OBD_FREE_PTR(lmv->tgts[index]);
 456         lmv->tgts[index] = NULL;
 457         return;
 458 }
 459
 460 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 461                            __u32 index, int gen)
 462 {
 463         struct lmv_obd      *lmv = &obd->u.lmv;
 464         struct lmv_tgt_desc *tgt;
 465         int                  rc = 0;
 466         ENTRY;
 467
 468         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 469
 470         lmv_init_lock(lmv);
 471
 472         if (lmv->desc.ld_tgt_count == 0) {
 473                 struct obd_device *mdc_obd;
 474
 475                 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 476                                                 &obd->obd_uuid);
 477                 if (!mdc_obd) {
 478                         lmv_init_unlock(lmv);
 479                         CERROR("%s: Target %s not attached: rc = %d\n",
 480                                obd->obd_name, uuidp->uuid, -EINVAL);
 481                         RETURN(-EINVAL);
 482                 }
 483         }
 484
 485         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
 486                 tgt = lmv->tgts[index];
 487                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
 488                        " rc = %d\n", obd->obd_name,
 489                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
 490                 lmv_init_unlock(lmv);
 491                 RETURN(-EEXIST);
 492         }
 493
 494         if (index >= lmv->tgts_size) {
 495                 /* We need to reallocate the lmv target array. */
 496                 struct lmv_tgt_desc **newtgts, **old = NULL;
 497                 __u32 newsize = 1;
 498                 __u32 oldsize = 0;
 499
 500                 while (newsize < index + 1)
 501                         newsize = newsize << 1;
 502                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 503                 if (newtgts == NULL) {
 504                         lmv_init_unlock(lmv);
 505                         RETURN(-ENOMEM);
 506                 }
 507
 508                 if (lmv->tgts_size) {
 509                         memcpy(newtgts, lmv->tgts,
 510                                sizeof(*newtgts) * lmv->tgts_size);
 511                         old = lmv->tgts;
 512                         oldsize = lmv->tgts_size;
 513                 }
 514
 515                 lmv->tgts = newtgts;
 516                 lmv->tgts_size = newsize;
 517                 smp_rmb();
 518                 if (old)
 519                         OBD_FREE(old, sizeof(*old) * oldsize);
 520
 521                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
 522                        lmv->tgts_size);
 523         }
 524
 525         OBD_ALLOC_PTR(tgt);
 526         if (!tgt) {
 527                 lmv_init_unlock(lmv);
 528                 RETURN(-ENOMEM);
 529         }
 530
 531         mutex_init(&tgt->ltd_fid_mutex);
 532         tgt->ltd_idx = index;
 533         tgt->ltd_uuid = *uuidp;
 534         tgt->ltd_active = 0;
 535         lmv->tgts[index] = tgt;
 536         if (index >= lmv->desc.ld_tgt_count)
 537                 lmv->desc.ld_tgt_count = index + 1;
 538
 539         if (lmv->connected) {
 540                 rc = lmv_connect_mdc(obd, tgt);
 541                 if (rc) {
 542                         spin_lock(&lmv->lmv_lock);
 543                         lmv->desc.ld_tgt_count--;
 544                         memset(tgt, 0, sizeof(*tgt));
 545                         spin_unlock(&lmv->lmv_lock);
 546                 } else {
 547                         int easize = sizeof(struct lmv_stripe_md) +
 548                                      lmv->desc.ld_tgt_count *
 549                                      sizeof(struct lu_fid);
 550                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
 551                 }
 552         }
 553
 554         lmv_init_unlock(lmv);
 555         RETURN(rc);
 556 }
 557
 558 int lmv_check_connect(struct obd_device *obd)
 559 {
 560         struct lmv_obd          *lmv = &obd->u.lmv;
 561         struct lmv_tgt_desc     *tgt;
 562         __u32                    i;
 563         int                      rc;
 564         int                      easize;
 565         ENTRY;
 566
 567         if (lmv->connected)
 568                 RETURN(0);
 569
 570         lmv_init_lock(lmv);
 571         if (lmv->connected) {
 572                 lmv_init_unlock(lmv);
 573                 RETURN(0);
 574         }
 575
 576         if (lmv->desc.ld_tgt_count == 0) {
 577                 lmv_init_unlock(lmv);
 578                 CERROR("%s: no targets configured.\n", obd->obd_name);
 579                 RETURN(-EINVAL);
 580         }
 581
 582         LASSERT(lmv->tgts != NULL);
 583
 584         if (lmv->tgts[0] == NULL) {
 585                 lmv_init_unlock(lmv);
 586                 CERROR("%s: no target configured for index 0.\n",
 587                        obd->obd_name);
 588                 RETURN(-EINVAL);
 589         }
 590
 591         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 592                lmv->cluuid.uuid, obd->obd_name);
 593
 594         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 595                 tgt = lmv->tgts[i];
 596                 if (tgt == NULL)
 597                         continue;
 598                 rc = lmv_connect_mdc(obd, tgt);
 599                 if (rc)
 600                         GOTO(out_disc, rc);
 601         }
 602
 603         lmv_set_timeouts(obd);
 604         class_export_put(lmv->exp);
 605         lmv->connected = 1;
 606         easize = lmv_get_easize(lmv);
 607         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
 608         lmv_init_unlock(lmv);
 609         RETURN(0);
 610
 611  out_disc:
 612         while (i-- > 0) {
 613                 int rc2;
 614                 tgt = lmv->tgts[i];
 615                 if (tgt == NULL)
 616                         continue;
 617                 tgt->ltd_active = 0;
 618                 if (tgt->ltd_exp) {
 619                         --lmv->desc.ld_active_tgt_count;
 620                         rc2 = obd_disconnect(tgt->ltd_exp);
 621                         if (rc2) {
 622                                 CERROR("LMV target %s disconnect on "
 623                                        "MDC idx %d: error %d\n",
 624                                        tgt->ltd_uuid.uuid, i, rc2);
 625                         }
 626                 }
 627         }
 628         class_disconnect(lmv->exp);
 629         lmv_init_unlock(lmv);
 630         RETURN(rc);
 631 }
 632
 633 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 634 {
 635 #ifdef __KERNEL__
 636         struct proc_dir_entry  *lmv_proc_dir;
 637 #endif
 638         struct lmv_obd         *lmv = &obd->u.lmv;
 639         struct obd_device      *mdc_obd;
 640         int                     rc;
 641         ENTRY;
 642
 643         LASSERT(tgt != NULL);
 644         LASSERT(obd != NULL);
 645
 646         mdc_obd = class_exp2obd(tgt->ltd_exp);
 647
 648         if (mdc_obd) {
 649                 mdc_obd->obd_force = obd->obd_force;
 650                 mdc_obd->obd_fail = obd->obd_fail;
 651                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 652         }
 653
 654 #ifdef __KERNEL__
 655         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 656         if (lmv_proc_dir) {
 657                 struct proc_dir_entry *mdc_symlink;
 658
 659                 mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
 660                 if (mdc_symlink) {
 661                         lprocfs_remove(&mdc_symlink);
 662                 } else {
 663                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 664                                obd->obd_type->typ_name, obd->obd_name,
 665                                mdc_obd->obd_name);
 666                 }
 667         }
 668 #endif
 669         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 670         if (rc)
 671                 CERROR("Can't finanize fids factory\n");
 672
 673         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 674                tgt->ltd_exp->exp_obd->obd_name,
 675                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 676
 677         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 678         rc = obd_disconnect(tgt->ltd_exp);
 679         if (rc) {
 680                 if (tgt->ltd_active) {
 681                         CERROR("Target %s disconnect error %d\n",
 682                                tgt->ltd_uuid.uuid, rc);
 683                 }
 684         }
 685
 686         lmv_activate_target(lmv, tgt, 0);
 687         tgt->ltd_exp = NULL;
 688         RETURN(0);
 689 }
 690
 691 static int lmv_disconnect(struct obd_export *exp)
 692 {
 693         struct obd_device       *obd = class_exp2obd(exp);
 694 #ifdef __KERNEL__
 695         struct proc_dir_entry   *lmv_proc_dir;
 696 #endif
 697         struct lmv_obd          *lmv = &obd->u.lmv;
 698         int                      rc;
 699         __u32                    i;
 700         ENTRY;
 701
 702         if (!lmv->tgts)
 703                 goto out_local;
 704
 705         /*
 706          * Only disconnect the underlying layers on the final disconnect.
 707          */
 708         lmv->refcount--;
 709         if (lmv->refcount != 0)
 710                 goto out_local;
 711
 712         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 713                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
 714                         continue;
 715
 716                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
 717         }
 718
 719 #ifdef __KERNEL__
 720         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 721         if (lmv_proc_dir) {
 722                 lprocfs_remove(&lmv_proc_dir);
 723         } else {
 724                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 725                        obd->obd_type->typ_name, obd->obd_name);
 726         }
 727 #endif
 728
 729 out_local:
 730         /*
 731          * This is the case when no real connection is established by
 732          * lmv_check_connect().
 733          */
 734         if (!lmv->connected)
 735                 class_export_put(exp);
 736         rc = class_disconnect(exp);
 737         if (lmv->refcount == 0)
 738                 lmv->connected = 0;
 739         RETURN(rc);
 740 }
 741
 742 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
 743 {
 744         struct obd_device       *obddev = class_exp2obd(exp);
 745         struct lmv_obd          *lmv = &obddev->u.lmv;
 746         struct getinfo_fid2path *gf;
 747         struct lmv_tgt_desc     *tgt;
 748         struct getinfo_fid2path *remote_gf = NULL;
 749         int                     remote_gf_size = 0;
 750         int                     rc;
 751
 752         gf = (struct getinfo_fid2path *)karg;
 753         tgt = lmv_find_target(lmv, &gf->gf_fid);
 754         if (IS_ERR(tgt))
 755                 RETURN(PTR_ERR(tgt));
 756
 757 repeat_fid2path:
 758         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 759         if (rc != 0 && rc != -EREMOTE)
 760                 GOTO(out_fid2path, rc);
 761
 762         /* If remote_gf != NULL, it means just building the
 763          * path on the remote MDT, copy this path segement to gf */
 764         if (remote_gf != NULL) {
 765                 struct getinfo_fid2path *ori_gf;
 766                 char *ptr;
 767
 768                 ori_gf = (struct getinfo_fid2path *)karg;
 769                 if (strlen(ori_gf->gf_path) +
 770                     strlen(gf->gf_path) > ori_gf->gf_pathlen)
 771                         GOTO(out_fid2path, rc = -EOVERFLOW);
 772
 773                 ptr = ori_gf->gf_path;
 774
 775                 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
 776                         strlen(ori_gf->gf_path));
 777
 778                 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
 779                 ptr += strlen(gf->gf_path);
 780                 *ptr = '/';
 781         }
 782
 783         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
 784                tgt->ltd_exp->exp_obd->obd_name,
 785                gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 786                gf->gf_linkno);
 787
 788         if (rc == 0)
 789                 GOTO(out_fid2path, rc);
 790
 791         /* sigh, has to go to another MDT to do path building further */
 792         if (remote_gf == NULL) {
 793                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 794                 OBD_ALLOC(remote_gf, remote_gf_size);
 795                 if (remote_gf == NULL)
 796                         GOTO(out_fid2path, rc = -ENOMEM);
 797                 remote_gf->gf_pathlen = PATH_MAX;
 798         }
 799
 800         if (!fid_is_sane(&gf->gf_fid)) {
 801                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 802                        tgt->ltd_exp->exp_obd->obd_name,
 803                        PFID(&gf->gf_fid), -EINVAL);
 804                 GOTO(out_fid2path, rc = -EINVAL);
 805         }
 806
 807         tgt = lmv_find_target(lmv, &gf->gf_fid);
 808         if (IS_ERR(tgt))
 809                 GOTO(out_fid2path, rc = -EINVAL);
 810
 811         remote_gf->gf_fid = gf->gf_fid;
 812         remote_gf->gf_recno = -1;
 813         remote_gf->gf_linkno = -1;
 814         memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
 815         gf = remote_gf;
 816         goto repeat_fid2path;
 817
 818 out_fid2path:
 819         if (remote_gf != NULL)
 820                 OBD_FREE(remote_gf, remote_gf_size);
 821         RETURN(rc);
 822 }
 823
 824 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 825                              const struct hsm_user_request *hur,
 826                              const struct lmv_tgt_desc *tgt_mds)
 827 {
 828         __u32                    i;
 829         int                      nr = 0;
 830         struct lmv_tgt_desc     *curr_tgt;
 831
 832         /* count how many requests must be sent to the given target */
 833         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 834                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
 835                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 836                         nr++;
 837         }
 838         return nr;
 839 }
 840
 841 static void lmv_hsm_req_build(struct lmv_obd *lmv,
 842                               struct hsm_user_request *hur_in,
 843                               const struct lmv_tgt_desc *tgt_mds,
 844                               struct hsm_user_request *hur_out)
 845 {
 846         __u32                    i, nr_out;
 847         struct lmv_tgt_desc     *curr_tgt;
 848
 849         /* build the hsm_user_request for the given target */
 850         hur_out->hur_request = hur_in->hur_request;
 851         nr_out = 0;
 852         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 853                 curr_tgt = lmv_find_target(lmv,
 854                                            &hur_in->hur_user_item[i].hui_fid);
 855                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 856                         hur_out->hur_user_item[nr_out] =
 857                                                 hur_in->hur_user_item[i];
 858                         nr_out++;
 859                 }
 860         }
 861         hur_out->hur_request.hr_itemcount = nr_out;
 862         memcpy(hur_data(hur_out), hur_data(hur_in),
 863                hur_in->hur_request.hr_data_len);
 864 }
 865
 866 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
 867                                  struct lustre_kernelcomm *lk, void *uarg)
 868 {
 869         __u32                    i;
 870         int                      rc;
 871         struct kkuc_ct_data     *kcd = NULL;
 872         ENTRY;
 873
 874         /* unregister request (call from llapi_hsm_copytool_fini) */
 875         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 876                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 877
 878                 if (tgt == NULL || tgt->ltd_exp == NULL)
 879                         continue;
 880                 /* best effort: try to clean as much as possible
 881                  * (continue on error) */
 882                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 883         }
 884
 885         /* Whatever the result, remove copytool from kuc groups.
 886          * Unreached coordinators will get EPIPE on next requests
 887          * and will unregister automatically.
 888          */
 889         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
 890         if (kcd != NULL)
 891                 OBD_FREE_PTR(kcd);
 892
 893         RETURN(rc);
 894 }
 895
 896 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
 897                                struct lustre_kernelcomm *lk, void *uarg)
 898 {
 899         struct file             *filp;
 900         __u32                    i, j;
 901         int                      err, rc;
 902         bool                     any_set = false;
 903         struct kkuc_ct_data     *kcd;
 904         ENTRY;
 905
 906         /* All or nothing: try to register to all MDS.
 907          * In case of failure, unregister from previous MDS,
 908          * except if it because of inactive target. */
 909         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 910                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 911
 912                 if (tgt == NULL || tgt->ltd_exp == NULL)
 913                         continue;
 914                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 915                 if (err) {
 916                         if (tgt->ltd_active) {
 917                                 /* permanent error */
 918                                 CERROR("%s: iocontrol MDC %s on MDT"
 919                                        " idx %d cmd %x: err = %d\n",
 920                                        class_exp2obd(lmv->exp)->obd_name,
 921                                        tgt->ltd_uuid.uuid, i, cmd, err);
 922                                 rc = err;
 923                                 lk->lk_flags |= LK_FLG_STOP;
 924                                 /* unregister from previous MDS */
 925                                 for (j = 0; j < i; j++) {
 926                                         tgt = lmv->tgts[j];
 927                                         if (tgt == NULL || tgt->ltd_exp == NULL)
 928                                                 continue;
 929                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
 930                                                       lk, uarg);
 931                                 }
 932                                 RETURN(rc);
 933                         }
 934                         /* else: transient error.
 935                          * kuc will register to the missing MDT
 936                          * when it is back */
 937                 } else {
 938                         any_set = true;
 939                 }
 940         }
 941
 942         if (!any_set)
 943                 /* no registration done: return error */
 944                 RETURN(-ENOTCONN);
 945
 946         /* at least one registration done, with no failure */
 947         filp = fget(lk->lk_wfd);
 948         if (filp == NULL)
 949                 RETURN(-EBADF);
 950
 951         OBD_ALLOC_PTR(kcd);
 952         if (kcd == NULL) {
 953                 fput(filp);
 954                 RETURN(-ENOMEM);
 955         }
 956         kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
 957         kcd->kcd_uuid = lmv->cluuid;
 958         kcd->kcd_archive = lk->lk_data;
 959
 960         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
 961         if (rc != 0) {
 962                 if (filp != NULL)
 963                         fput(filp);
 964                 OBD_FREE_PTR(kcd);
 965         }
 966
 967         RETURN(rc);
 968 }
 969
 970
 971
 972
 973 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 974                          int len, void *karg, void *uarg)
 975 {
 976         struct obd_device       *obddev = class_exp2obd(exp);
 977         struct lmv_obd          *lmv = &obddev->u.lmv;
 978         struct lmv_tgt_desc     *tgt = NULL;
 979         __u32                    i = 0;
 980         int                      rc = 0;
 981         int                      set = 0;
 982         __u32                    count = lmv->desc.ld_tgt_count;
 983         ENTRY;
 984
 985         if (count == 0)
 986                 RETURN(-ENOTTY);
 987
 988         switch (cmd) {
 989         case IOC_OBD_STATFS: {
 990                 struct obd_ioctl_data *data = karg;
 991                 struct obd_device *mdc_obd;
 992                 struct obd_statfs stat_buf = {0};
 993                 __u32 index;
 994
 995                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 996                 if ((index >= count))
 997                         RETURN(-ENODEV);
 998
 999                 tgt = lmv->tgts[index];
1000                 if (tgt == NULL || !tgt->ltd_active)
1001                         RETURN(-ENODATA);
1002
1003                 mdc_obd = class_exp2obd(tgt->ltd_exp);
1004                 if (!mdc_obd)
1005                         RETURN(-EINVAL);
1006
1007                 /* copy UUID */
1008                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
1009                                  min((int) data->ioc_plen2,
1010                                      (int) sizeof(struct obd_uuid))))
1011                         RETURN(-EFAULT);
1012
1013                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
1014                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1015                                 0);
1016                 if (rc)
1017                         RETURN(rc);
1018                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1019                                  min((int) data->ioc_plen1,
1020                                      (int) sizeof(stat_buf))))
1021                         RETURN(-EFAULT);
1022                 break;
1023         }
1024         case OBD_IOC_QUOTACTL: {
1025                 struct if_quotactl *qctl = karg;
1026                 struct obd_quotactl *oqctl;
1027
1028                 if (qctl->qc_valid == QC_MDTIDX) {
1029                         if (count <= qctl->qc_idx)
1030                                 RETURN(-EINVAL);
1031
1032                         tgt = lmv->tgts[qctl->qc_idx];
1033                         if (tgt == NULL || tgt->ltd_exp == NULL)
1034                                 RETURN(-EINVAL);
1035                 } else if (qctl->qc_valid == QC_UUID) {
1036                         for (i = 0; i < count; i++) {
1037                                 tgt = lmv->tgts[i];
1038                                 if (tgt == NULL)
1039                                         continue;
1040                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
1041                                                      &qctl->obd_uuid))
1042                                         continue;
1043
1044                                 if (tgt->ltd_exp == NULL)
1045                                         RETURN(-EINVAL);
1046
1047                                 break;
1048                         }
1049                 } else {
1050                         RETURN(-EINVAL);
1051                 }
1052
1053                 if (i >= count)
1054                         RETURN(-EAGAIN);
1055
1056                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1057                 OBD_ALLOC_PTR(oqctl);
1058                 if (!oqctl)
1059                         RETURN(-ENOMEM);
1060
1061                 QCTL_COPY(oqctl, qctl);
1062                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1063                 if (rc == 0) {
1064                         QCTL_COPY(qctl, oqctl);
1065                         qctl->qc_valid = QC_MDTIDX;
1066                         qctl->obd_uuid = tgt->ltd_uuid;
1067                 }
1068                 OBD_FREE_PTR(oqctl);
1069                 break;
1070         }
1071         case OBD_IOC_CHANGELOG_SEND:
1072         case OBD_IOC_CHANGELOG_CLEAR: {
1073                 struct ioc_changelog *icc = karg;
1074
1075                 if (icc->icc_mdtindex >= count)
1076                         RETURN(-ENODEV);
1077
1078                 tgt = lmv->tgts[icc->icc_mdtindex];
1079                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1080                         RETURN(-ENODEV);
1081                 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1082                 break;
1083         }
1084         case LL_IOC_GET_CONNECT_FLAGS: {
1085                 tgt = lmv->tgts[0];
1086                 if (tgt == NULL || tgt->ltd_exp == NULL)
1087                         RETURN(-ENODATA);
1088                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1089                 break;
1090         }
1091         case OBD_IOC_FID2PATH: {
1092                 rc = lmv_fid2path(exp, len, karg, uarg);
1093                 break;
1094         }
1095         case LL_IOC_HSM_STATE_GET:
1096         case LL_IOC_HSM_STATE_SET:
1097         case LL_IOC_HSM_ACTION: {
1098                 struct md_op_data       *op_data = karg;
1099
1100                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1101                 if (IS_ERR(tgt))
1102                         RETURN(PTR_ERR(tgt));
1103
1104                 if (tgt->ltd_exp == NULL)
1105                         RETURN(-EINVAL);
1106
1107                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1108                 break;
1109         }
1110         case LL_IOC_HSM_PROGRESS: {
1111                 const struct hsm_progress_kernel *hpk = karg;
1112
1113                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1114                 if (IS_ERR(tgt))
1115                         RETURN(PTR_ERR(tgt));
1116                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1117                 break;
1118         }
1119         case LL_IOC_HSM_REQUEST: {
1120                 struct hsm_user_request *hur = karg;
1121                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1122
1123                 if (reqcount == 0)
1124                         RETURN(0);
1125
1126                 /* if the request is about a single fid
1127                  * or if there is a single MDS, no need to split
1128                  * the request. */
1129                 if (reqcount == 1 || count == 1) {
1130                         tgt = lmv_find_target(lmv,
1131                                               &hur->hur_user_item[0].hui_fid);
1132                         if (IS_ERR(tgt))
1133                                 RETURN(PTR_ERR(tgt));
1134                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1135                 } else {
1136                         /* split fid list to their respective MDS */
1137                         for (i = 0; i < count; i++) {
1138                                 unsigned int            nr, reqlen;
1139                                 int                     rc1;
1140                                 struct hsm_user_request *req;
1141
1142                                 tgt = lmv->tgts[i];
1143                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1144                                         continue;
1145
1146                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1147                                 if (nr == 0) /* nothing for this MDS */
1148                                         continue;
1149
1150                                 /* build a request with fids for this MDS */
1151                                 reqlen = offsetof(typeof(*hur),
1152                                                   hur_user_item[nr])
1153                                                 + hur->hur_request.hr_data_len;
1154                                 OBD_ALLOC_LARGE(req, reqlen);
1155                                 if (req == NULL)
1156                                         RETURN(-ENOMEM);
1157
1158                                 lmv_hsm_req_build(lmv, hur, tgt, req);
1159
1160                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1161                                                     req, uarg);
1162                                 if (rc1 != 0 && rc == 0)
1163                                         rc = rc1;
1164                                 OBD_FREE_LARGE(req, reqlen);
1165                         }
1166                 }
1167                 break;
1168         }
1169         case LL_IOC_LOV_SWAP_LAYOUTS: {
1170                 struct md_op_data       *op_data = karg;
1171                 struct lmv_tgt_desc     *tgt1, *tgt2;
1172
1173                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1174                 if (IS_ERR(tgt1))
1175                         RETURN(PTR_ERR(tgt1));
1176
1177                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1178                 if (IS_ERR(tgt2))
1179                         RETURN(PTR_ERR(tgt2));
1180
1181                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1182                         RETURN(-EINVAL);
1183
1184                 /* only files on same MDT can have their layouts swapped */
1185                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1186                         RETURN(-EPERM);
1187
1188                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1189                 break;
1190         }
1191         case LL_IOC_HSM_CT_START: {
1192                 struct lustre_kernelcomm *lk = karg;
1193                 if (lk->lk_flags & LK_FLG_STOP)
1194                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1195                 else
1196                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1197                 break;
1198         }
1199         default:
1200                 for (i = 0; i < count; i++) {
1201                         struct obd_device *mdc_obd;
1202                         int err;
1203
1204                         tgt = lmv->tgts[i];
1205                         if (tgt == NULL || tgt->ltd_exp == NULL)
1206                                 continue;
1207                         /* ll_umount_begin() sets force flag but for lmv, not
1208                          * mdc. Let's pass it through */
1209                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1210                         mdc_obd->obd_force = obddev->obd_force;
1211                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1212                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1213                                 RETURN(err);
1214                         } else if (err) {
1215                                 if (tgt->ltd_active) {
1216                                         CERROR("error: iocontrol MDC %s on MDT"
1217                                                " idx %d cmd %x: err = %d\n",
1218                                                tgt->ltd_uuid.uuid, i, cmd, err);
1219                                         if (!rc)
1220                                                 rc = err;
1221                                 }
1222                         } else
1223                                 set = 1;
1224                 }
1225                 if (!set && !rc)
1226                         rc = -EIO;
1227         }
1228         RETURN(rc);
1229 }
1230
1231 #if 0
1232 static int lmv_all_chars_policy(int count, const char *name,
1233                                 int len)
1234 {
1235         unsigned int c = 0;
1236
1237         while (len > 0)
1238                 c += name[--len];
1239         c = c % count;
1240         return c;
1241 }
1242
1243 static int lmv_nid_policy(struct lmv_obd *lmv)
1244 {
1245         struct obd_import *imp;
1246         __u32              id;
1247
1248         /*
1249          * XXX: To get nid we assume that underlying obd device is mdc.
1250          */
1251         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1252         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1253         return id % lmv->desc.ld_tgt_count;
1254 }
1255
1256 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1257                           placement_policy_t placement)
1258 {
1259         switch (placement) {
1260         case PLACEMENT_CHAR_POLICY:
1261                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1262                                             op_data->op_name,
1263                                             op_data->op_namelen);
1264         case PLACEMENT_NID_POLICY:
1265                 return lmv_nid_policy(lmv);
1266
1267         default:
1268                 break;
1269         }
1270
1271         CERROR("Unsupported placement policy %x\n", placement);
1272         return -EINVAL;
1273 }
1274 #endif
1275
1276 /**
1277  * This is _inode_ placement policy function (not name).
1278  */
1279 static int lmv_placement_policy(struct obd_device *obd,
1280                                 struct md_op_data *op_data,
1281                                 mdsno_t *mds)
1282 {
1283         struct lmv_obd          *lmv = &obd->u.lmv;
1284         ENTRY;
1285
1286         LASSERT(mds != NULL);
1287
1288         if (lmv->desc.ld_tgt_count == 1) {
1289                 *mds = 0;
1290                 RETURN(0);
1291         }
1292
1293         /**
1294          * If stripe_offset is provided during setdirstripe
1295          * (setdirstripe -i xx), xx MDS will be choosen.
1296          */
1297         if (op_data->op_cli_flags & CLI_SET_MEA) {
1298                 struct lmv_user_md *lum;
1299
1300                 lum = (struct lmv_user_md *)op_data->op_data;
1301                 if (lum->lum_type == LMV_STRIPE_TYPE &&
1302                     lum->lum_stripe_offset != -1) {
1303                         if (lum->lum_stripe_offset >= lmv->desc.ld_tgt_count) {
1304                                 CERROR("%s: Stripe_offset %d > MDT count %d:"
1305                                        " rc = %d\n", obd->obd_name,
1306                                        lum->lum_stripe_offset,
1307                                        lmv->desc.ld_tgt_count, -ERANGE);
1308                                 RETURN(-ERANGE);
1309                         }
1310                         *mds = lum->lum_stripe_offset;
1311                         RETURN(0);
1312                 }
1313         }
1314
1315         /* Allocate new fid on target according to operation type and parent
1316          * home mds. */
1317         *mds = op_data->op_mds;
1318         RETURN(0);
1319 }
1320
1321 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
1322                     mdsno_t mds)
1323 {
1324         struct lmv_tgt_desc     *tgt;
1325         int                      rc;
1326         ENTRY;
1327
1328         tgt = lmv_get_target(lmv, mds);
1329         if (IS_ERR(tgt))
1330                 RETURN(PTR_ERR(tgt));
1331
1332         /*
1333          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1334          * on server that seq in new allocated fid is not yet known.
1335          */
1336         mutex_lock(&tgt->ltd_fid_mutex);
1337
1338         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1339                 GOTO(out, rc = -ENODEV);
1340
1341         /*
1342          * Asking underlaying tgt layer to allocate new fid.
1343          */
1344         rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
1345         if (rc > 0) {
1346                 LASSERT(fid_is_sane(fid));
1347                 rc = 0;
1348         }
1349
1350         EXIT;
1351 out:
1352         mutex_unlock(&tgt->ltd_fid_mutex);
1353         return rc;
1354 }
1355
1356 int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1357                   struct md_op_data *op_data)
1358 {
1359         struct obd_device     *obd = class_exp2obd(exp);
1360         struct lmv_obd        *lmv = &obd->u.lmv;
1361         mdsno_t                mds = 0;
1362         int                    rc;
1363         ENTRY;
1364
1365         LASSERT(op_data != NULL);
1366         LASSERT(fid != NULL);
1367
1368         rc = lmv_placement_policy(obd, op_data, &mds);
1369         if (rc) {
1370                 CERROR("Can't get target for allocating fid, "
1371                        "rc %d\n", rc);
1372                 RETURN(rc);
1373         }
1374
1375         rc = __lmv_fid_alloc(lmv, fid, mds);
1376         if (rc) {
1377                 CERROR("Can't alloc new fid, rc %d\n", rc);
1378                 RETURN(rc);
1379         }
1380
1381         RETURN(rc);
1382 }
1383
1384 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1385 {
1386         struct lmv_obd             *lmv = &obd->u.lmv;
1387         struct lprocfs_static_vars  lvars;
1388         struct lmv_desc            *desc;
1389         int                         rc;
1390         ENTRY;
1391
1392         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1393                 CERROR("LMV setup requires a descriptor\n");
1394                 RETURN(-EINVAL);
1395         }
1396
1397         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1398         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1399                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1400                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1401                 RETURN(-EINVAL);
1402         }
1403
1404         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1405         if (lmv->tgts == NULL)
1406                 RETURN(-ENOMEM);
1407         lmv->tgts_size = 32;
1408
1409         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1410         lmv->desc.ld_tgt_count = 0;
1411         lmv->desc.ld_active_tgt_count = 0;
1412         lmv->max_cookiesize = 0;
1413         lmv->max_def_easize = 0;
1414         lmv->max_easize = 0;
1415         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1416
1417         spin_lock_init(&lmv->lmv_lock);
1418         mutex_init(&lmv->init_mutex);
1419
1420         lprocfs_lmv_init_vars(&lvars);
1421
1422         lprocfs_obd_setup(obd, lvars.obd_vars);
1423         lprocfs_alloc_md_stats(obd, 0);
1424 #ifdef LPROCFS
1425         {
1426                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1427                                         0444, &lmv_proc_target_fops, obd);
1428                 if (rc)
1429                         CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1430                               obd->obd_name, rc);
1431         }
1432 #endif
1433         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1434                              LUSTRE_CLI_FLD_HASH_DHT);
1435         if (rc) {
1436                 CERROR("Can't init FLD, err %d\n", rc);
1437                 GOTO(out, rc);
1438         }
1439
1440         RETURN(0);
1441
1442 out:
1443         return rc;
1444 }
1445
1446 static int lmv_cleanup(struct obd_device *obd)
1447 {
1448         struct lmv_obd   *lmv = &obd->u.lmv;
1449         ENTRY;
1450
1451         fld_client_fini(&lmv->lmv_fld);
1452         if (lmv->tgts != NULL) {
1453                 int i;
1454                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1455                         if (lmv->tgts[i] == NULL)
1456                                 continue;
1457                         lmv_del_target(lmv, i);
1458                 }
1459                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1460                 lmv->tgts_size = 0;
1461         }
1462         RETURN(0);
1463 }
1464
1465 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1466 {
1467         struct lustre_cfg       *lcfg = buf;
1468         struct obd_uuid         obd_uuid;
1469         int                     gen;
1470         __u32                   index;
1471         int                     rc;
1472         ENTRY;
1473
1474         switch (lcfg->lcfg_command) {
1475         case LCFG_ADD_MDC:
1476                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1477                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1478                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1479                         GOTO(out, rc = -EINVAL);
1480
1481                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1482
1483                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
1484                         GOTO(out, rc = -EINVAL);
1485                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1486                         GOTO(out, rc = -EINVAL);
1487                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1488                 GOTO(out, rc);
1489         default:
1490                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1491                 GOTO(out, rc = -EINVAL);
1492         }
1493 out:
1494         RETURN(rc);
1495 }
1496
1497 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1498                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1499 {
1500         struct obd_device       *obd = class_exp2obd(exp);
1501         struct lmv_obd          *lmv = &obd->u.lmv;
1502         struct obd_statfs       *temp;
1503         int                      rc = 0;
1504         __u32                    i;
1505         ENTRY;
1506
1507         rc = lmv_check_connect(obd);
1508         if (rc)
1509                 RETURN(rc);
1510
1511         OBD_ALLOC(temp, sizeof(*temp));
1512         if (temp == NULL)
1513                 RETURN(-ENOMEM);
1514
1515         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1516                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1517                         continue;
1518
1519                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1520                                 max_age, flags);
1521                 if (rc) {
1522                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1523                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1524                                rc);
1525                         GOTO(out_free_temp, rc);
1526                 }
1527
1528                 if (i == 0) {
1529                         *osfs = *temp;
1530                         /* If the statfs is from mount, it will needs
1531                          * retrieve necessary information from MDT0.
1532                          * i.e. mount does not need the merged osfs
1533                          * from all of MDT.
1534                          * And also clients can be mounted as long as
1535                          * MDT0 is in service*/
1536                         if (flags & OBD_STATFS_FOR_MDT0)
1537                                 GOTO(out_free_temp, rc);
1538                 } else {
1539                         osfs->os_bavail += temp->os_bavail;
1540                         osfs->os_blocks += temp->os_blocks;
1541                         osfs->os_ffree += temp->os_ffree;
1542                         osfs->os_files += temp->os_files;
1543                 }
1544         }
1545
1546         EXIT;
1547 out_free_temp:
1548         OBD_FREE(temp, sizeof(*temp));
1549         return rc;
1550 }
1551
1552 static int lmv_getstatus(struct obd_export *exp,
1553                          struct lu_fid *fid,
1554                          struct obd_capa **pc)
1555 {
1556         struct obd_device    *obd = exp->exp_obd;
1557         struct lmv_obd       *lmv = &obd->u.lmv;
1558         int                   rc;
1559         ENTRY;
1560
1561         rc = lmv_check_connect(obd);
1562         if (rc)
1563                 RETURN(rc);
1564
1565         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1566         RETURN(rc);
1567 }
1568
1569 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1570                         struct obd_capa *oc, obd_valid valid, const char *name,
1571                         const char *input, int input_size, int output_size,
1572                         int flags, struct ptlrpc_request **request)
1573 {
1574         struct obd_device      *obd = exp->exp_obd;
1575         struct lmv_obd         *lmv = &obd->u.lmv;
1576         struct lmv_tgt_desc    *tgt;
1577         int                     rc;
1578         ENTRY;
1579
1580         rc = lmv_check_connect(obd);
1581         if (rc)
1582                 RETURN(rc);
1583
1584         tgt = lmv_find_target(lmv, fid);
1585         if (IS_ERR(tgt))
1586                 RETURN(PTR_ERR(tgt));
1587
1588         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1589                          input_size, output_size, flags, request);
1590
1591         RETURN(rc);
1592 }
1593
1594 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1595                         struct obd_capa *oc, obd_valid valid, const char *name,
1596                         const char *input, int input_size, int output_size,
1597                         int flags, __u32 suppgid,
1598                         struct ptlrpc_request **request)
1599 {
1600         struct obd_device      *obd = exp->exp_obd;
1601         struct lmv_obd         *lmv = &obd->u.lmv;
1602         struct lmv_tgt_desc    *tgt;
1603         int                     rc;
1604         ENTRY;
1605
1606         rc = lmv_check_connect(obd);
1607         if (rc)
1608                 RETURN(rc);
1609
1610         tgt = lmv_find_target(lmv, fid);
1611         if (IS_ERR(tgt))
1612                 RETURN(PTR_ERR(tgt));
1613
1614         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1615                          input_size, output_size, flags, suppgid,
1616                          request);
1617
1618         RETURN(rc);
1619 }
1620
1621 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1622                        struct ptlrpc_request **request)
1623 {
1624         struct obd_device       *obd = exp->exp_obd;
1625         struct lmv_obd          *lmv = &obd->u.lmv;
1626         struct lmv_tgt_desc     *tgt;
1627         int                      rc;
1628         ENTRY;
1629
1630         rc = lmv_check_connect(obd);
1631         if (rc)
1632                 RETURN(rc);
1633
1634         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1635         if (IS_ERR(tgt))
1636                 RETURN(PTR_ERR(tgt));
1637
1638         if (op_data->op_flags & MF_GET_MDT_IDX) {
1639                 op_data->op_mds = tgt->ltd_idx;
1640                 RETURN(0);
1641         }
1642
1643         rc = md_getattr(tgt->ltd_exp, op_data, request);
1644
1645         RETURN(rc);
1646 }
1647
1648 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1649 {
1650         struct obd_device   *obd = exp->exp_obd;
1651         struct lmv_obd      *lmv = &obd->u.lmv;
1652         __u32                i;
1653         int                  rc;
1654         ENTRY;
1655
1656         rc = lmv_check_connect(obd);
1657         if (rc)
1658                 RETURN(rc);
1659
1660         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1661
1662         /*
1663          * With DNE every object can have two locks in different namespaces:
1664          * lookup lock in space of MDT storing direntry and update/open lock in
1665          * space of MDT storing inode.
1666          */
1667         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1668                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1669                         continue;
1670                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1671         }
1672
1673         RETURN(0);
1674 }
1675
1676 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1677                            ldlm_iterator_t it, void *data)
1678 {
1679         struct obd_device   *obd = exp->exp_obd;
1680         struct lmv_obd      *lmv = &obd->u.lmv;
1681         __u32                i;
1682         int                  rc;
1683         ENTRY;
1684
1685         rc = lmv_check_connect(obd);
1686         if (rc)
1687                 RETURN(rc);
1688
1689         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1690
1691         /*
1692          * With DNE every object can have two locks in different namespaces:
1693          * lookup lock in space of MDT storing direntry and update/open lock in
1694          * space of MDT storing inode.
1695          */
1696         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1697                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1698                         continue;
1699                 rc = md_find_cbdata(lmv->tgts[i]->ltd_exp, fid, it, data);
1700                 if (rc)
1701                         RETURN(rc);
1702         }
1703
1704         RETURN(rc);
1705 }
1706
1707
1708 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1709                      struct md_open_data *mod, struct ptlrpc_request **request)
1710 {
1711         struct obd_device     *obd = exp->exp_obd;
1712         struct lmv_obd        *lmv = &obd->u.lmv;
1713         struct lmv_tgt_desc   *tgt;
1714         int                    rc;
1715         ENTRY;
1716
1717         rc = lmv_check_connect(obd);
1718         if (rc)
1719                 RETURN(rc);
1720
1721         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1722         if (IS_ERR(tgt))
1723                 RETURN(PTR_ERR(tgt));
1724
1725         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1726         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1727         RETURN(rc);
1728 }
1729
1730 struct lmv_tgt_desc
1731 *lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1732                 struct lu_fid *fid)
1733 {
1734         struct lmv_tgt_desc *tgt;
1735
1736         tgt = lmv_find_target(lmv, fid);
1737         if (IS_ERR(tgt))
1738                 return tgt;
1739
1740         op_data->op_mds = tgt->ltd_idx;
1741
1742         return tgt;
1743 }
1744
1745 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1746                const void *data, int datalen, int mode, __u32 uid,
1747                __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1748                struct ptlrpc_request **request)
1749 {
1750         struct obd_device       *obd = exp->exp_obd;
1751         struct lmv_obd          *lmv = &obd->u.lmv;
1752         struct lmv_tgt_desc     *tgt;
1753         int                      rc;
1754         ENTRY;
1755
1756         rc = lmv_check_connect(obd);
1757         if (rc)
1758                 RETURN(rc);
1759
1760         if (!lmv->desc.ld_active_tgt_count)
1761                 RETURN(-EIO);
1762
1763         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1764         if (IS_ERR(tgt))
1765                 RETURN(PTR_ERR(tgt));
1766
1767         rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
1768         if (rc)
1769                 RETURN(rc);
1770
1771         CDEBUG(D_INODE, "CREATE '%*s' on "DFID" -> mds #%x\n",
1772                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1773                op_data->op_mds);
1774
1775         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1776         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1777                        cap_effective, rdev, request);
1778
1779         if (rc == 0) {
1780                 if (*request == NULL)
1781                         RETURN(rc);
1782                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1783         }
1784         RETURN(rc);
1785 }
1786
1787 static int lmv_done_writing(struct obd_export *exp,
1788                             struct md_op_data *op_data,
1789                             struct md_open_data *mod)
1790 {
1791         struct obd_device     *obd = exp->exp_obd;
1792         struct lmv_obd        *lmv = &obd->u.lmv;
1793         struct lmv_tgt_desc   *tgt;
1794         int                    rc;
1795         ENTRY;
1796
1797         rc = lmv_check_connect(obd);
1798         if (rc)
1799                 RETURN(rc);
1800
1801         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1802         if (IS_ERR(tgt))
1803                 RETURN(PTR_ERR(tgt));
1804
1805         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1806         RETURN(rc);
1807 }
1808
1809 static int
1810 lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1811                    struct lookup_intent *it, struct md_op_data *op_data,
1812                    struct lustre_handle *lockh, void *lmm, int lmmsize,
1813                    __u64 extra_lock_flags)
1814 {
1815         struct ptlrpc_request      *req = it->d.lustre.it_data;
1816         struct obd_device          *obd = exp->exp_obd;
1817         struct lmv_obd             *lmv = &obd->u.lmv;
1818         struct lustre_handle        plock;
1819         struct lmv_tgt_desc        *tgt;
1820         struct md_op_data          *rdata;
1821         struct lu_fid               fid1;
1822         struct mdt_body            *body;
1823         int                         rc = 0;
1824         int                         pmode;
1825         ENTRY;
1826
1827         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1828         LASSERT(body != NULL);
1829
1830         if (!(body->valid & OBD_MD_MDS))
1831                 RETURN(0);
1832
1833         CDEBUG(D_INODE, "REMOTE_ENQUEUE '%s' on "DFID" -> "DFID"\n",
1834                LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
1835
1836         /*
1837          * We got LOOKUP lock, but we really need attrs.
1838          */
1839         pmode = it->d.lustre.it_lock_mode;
1840         LASSERT(pmode != 0);
1841         memcpy(&plock, lockh, sizeof(plock));
1842         it->d.lustre.it_lock_mode = 0;
1843         it->d.lustre.it_data = NULL;
1844         fid1 = body->fid1;
1845
1846         ptlrpc_req_finished(req);
1847
1848         tgt = lmv_find_target(lmv, &fid1);
1849         if (IS_ERR(tgt))
1850                 GOTO(out, rc = PTR_ERR(tgt));
1851
1852         OBD_ALLOC_PTR(rdata);
1853         if (rdata == NULL)
1854                 GOTO(out, rc = -ENOMEM);
1855
1856         rdata->op_fid1 = fid1;
1857         rdata->op_bias = MDS_CROSS_REF;
1858
1859         rc = md_enqueue(tgt->ltd_exp, einfo, it, rdata, lockh,
1860                         lmm, lmmsize, NULL, extra_lock_flags);
1861         OBD_FREE_PTR(rdata);
1862         EXIT;
1863 out:
1864         ldlm_lock_decref(&plock, pmode);
1865         return rc;
1866 }
1867
1868 static int
1869 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1870             struct lookup_intent *it, struct md_op_data *op_data,
1871             struct lustre_handle *lockh, void *lmm, int lmmsize,
1872             struct ptlrpc_request **req, __u64 extra_lock_flags)
1873 {
1874         struct obd_device        *obd = exp->exp_obd;
1875         struct lmv_obd           *lmv = &obd->u.lmv;
1876         struct lmv_tgt_desc      *tgt;
1877         int                       rc;
1878         ENTRY;
1879
1880         rc = lmv_check_connect(obd);
1881         if (rc)
1882                 RETURN(rc);
1883
1884         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1885                LL_IT2STR(it), PFID(&op_data->op_fid1));
1886
1887         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1888         if (IS_ERR(tgt))
1889                 RETURN(PTR_ERR(tgt));
1890
1891         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1892                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1893
1894         rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh,
1895                         lmm, lmmsize, req, extra_lock_flags);
1896
1897         if (rc == 0 && it && it->it_op == IT_OPEN) {
1898                 rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
1899                                         lmm, lmmsize, extra_lock_flags);
1900         }
1901         RETURN(rc);
1902 }
1903
1904 static int
1905 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1906                  struct ptlrpc_request **request)
1907 {
1908         struct ptlrpc_request   *req = NULL;
1909         struct obd_device       *obd = exp->exp_obd;
1910         struct lmv_obd          *lmv = &obd->u.lmv;
1911         struct lmv_tgt_desc     *tgt;
1912         struct mdt_body         *body;
1913         int                      rc;
1914         ENTRY;
1915
1916         rc = lmv_check_connect(obd);
1917         if (rc)
1918                 RETURN(rc);
1919
1920         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1921         if (IS_ERR(tgt))
1922                 RETURN(PTR_ERR(tgt));
1923
1924         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1925                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1926                tgt->ltd_idx);
1927
1928         rc = md_getattr_name(tgt->ltd_exp, op_data, request);
1929         if (rc != 0)
1930                 RETURN(rc);
1931
1932         body = req_capsule_server_get(&(*request)->rq_pill,
1933                                       &RMF_MDT_BODY);
1934         LASSERT(body != NULL);
1935
1936         if (body->valid & OBD_MD_MDS) {
1937                 struct lu_fid rid = body->fid1;
1938                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1939                        PFID(&rid));
1940
1941                 tgt = lmv_find_target(lmv, &rid);
1942                 if (IS_ERR(tgt)) {
1943                         ptlrpc_req_finished(*request);
1944                         RETURN(PTR_ERR(tgt));
1945                 }
1946
1947                 op_data->op_fid1 = rid;
1948                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1949                 op_data->op_namelen = 0;
1950                 op_data->op_name = NULL;
1951                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1952                 ptlrpc_req_finished(*request);
1953                 *request = req;
1954         }
1955
1956         RETURN(rc);
1957 }
1958
1959 #define md_op_data_fid(op_data, fl)                     \
1960         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1961          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1962          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1963          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1964          NULL)
1965
1966 static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data,
1967                             int op_tgt, ldlm_mode_t mode, int bits, int flag)
1968 {
1969         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1970         struct obd_device      *obd = exp->exp_obd;
1971         struct lmv_obd         *lmv = &obd->u.lmv;
1972         struct lmv_tgt_desc    *tgt;
1973         ldlm_policy_data_t      policy = {{0}};
1974         int                     rc = 0;
1975         ENTRY;
1976
1977         if (!fid_is_sane(fid))
1978                 RETURN(0);
1979
1980         tgt = lmv_find_target(lmv, fid);
1981         if (IS_ERR(tgt))
1982                 RETURN(PTR_ERR(tgt));
1983
1984         if (tgt->ltd_idx != op_tgt) {
1985                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1986                 policy.l_inodebits.bits = bits;
1987                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1988                                       mode, LCF_ASYNC, NULL);
1989         } else {
1990                 CDEBUG(D_INODE,
1991                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1992                        op_tgt, PFID(fid));
1993                 op_data->op_flags |= flag;
1994                 rc = 0;
1995         }
1996
1997         RETURN(rc);
1998 }
1999
2000 /*
2001  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2002  * op_data->op_fid2
2003  */
2004 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2005                     struct ptlrpc_request **request)
2006 {
2007         struct obd_device       *obd = exp->exp_obd;
2008         struct lmv_obd          *lmv = &obd->u.lmv;
2009         struct lmv_tgt_desc     *tgt;
2010         int                      rc;
2011         ENTRY;
2012
2013         rc = lmv_check_connect(obd);
2014         if (rc)
2015                 RETURN(rc);
2016
2017         LASSERT(op_data->op_namelen != 0);
2018
2019         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2020                PFID(&op_data->op_fid2), op_data->op_namelen,
2021                op_data->op_name, PFID(&op_data->op_fid1));
2022
2023         op_data->op_fsuid = current_fsuid();
2024         op_data->op_fsgid = current_fsgid();
2025         op_data->op_cap = cfs_curproc_cap_pack();
2026         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2027         if (IS_ERR(tgt))
2028                 RETURN(PTR_ERR(tgt));
2029
2030         /*
2031          * Cancel UPDATE lock on child (fid1).
2032          */
2033         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2034         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2035                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2036         if (rc != 0)
2037                 RETURN(rc);
2038
2039         rc = md_link(tgt->ltd_exp, op_data, request);
2040
2041         RETURN(rc);
2042 }
2043
2044 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2045                       const char *old, int oldlen, const char *new, int newlen,
2046                       struct ptlrpc_request **request)
2047 {
2048         struct obd_device       *obd = exp->exp_obd;
2049         struct lmv_obd          *lmv = &obd->u.lmv;
2050         struct lmv_tgt_desc     *src_tgt;
2051         struct lmv_tgt_desc     *tgt_tgt;
2052         int                     rc;
2053         ENTRY;
2054
2055         LASSERT(oldlen != 0);
2056
2057         CDEBUG(D_INODE, "RENAME %*s in "DFID" to %*s in "DFID"\n",
2058                oldlen, old, PFID(&op_data->op_fid1),
2059                newlen, new, PFID(&op_data->op_fid2));
2060
2061         rc = lmv_check_connect(obd);
2062         if (rc)
2063                 RETURN(rc);
2064
2065         op_data->op_fsuid = current_fsuid();
2066         op_data->op_fsgid = current_fsgid();
2067         op_data->op_cap = cfs_curproc_cap_pack();
2068         src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2069         if (IS_ERR(src_tgt))
2070                 RETURN(PTR_ERR(src_tgt));
2071
2072         tgt_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2073         if (IS_ERR(tgt_tgt))
2074                 RETURN(PTR_ERR(tgt_tgt));
2075         /*
2076          * LOOKUP lock on src child (fid3) should also be cancelled for
2077          * src_tgt in mdc_rename.
2078          */
2079         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2080
2081         /*
2082          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2083          * own target.
2084          */
2085         rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2086                               LCK_EX, MDS_INODELOCK_UPDATE,
2087                               MF_MDC_CANCEL_FID2);
2088
2089         /*
2090          * Cancel LOOKUP locks on tgt child (fid4) for parent tgt_tgt.
2091          */
2092         if (rc == 0) {
2093                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2094                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2095                                       MF_MDC_CANCEL_FID4);
2096         }
2097
2098         /*
2099          * Cancel all the locks on tgt child (fid4).
2100          */
2101         if (rc == 0)
2102                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2103                                       LCK_EX, MDS_INODELOCK_FULL,
2104                                       MF_MDC_CANCEL_FID4);
2105
2106         if (rc == 0)
2107                 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen,
2108                                new, newlen, request);
2109         RETURN(rc);
2110 }
2111
2112 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2113                        void *ea, int ealen, void *ea2, int ea2len,
2114                        struct ptlrpc_request **request,
2115                        struct md_open_data **mod)
2116 {
2117         struct obd_device       *obd = exp->exp_obd;
2118         struct lmv_obd          *lmv = &obd->u.lmv;
2119         struct lmv_tgt_desc     *tgt;
2120         int                      rc = 0;
2121         ENTRY;
2122
2123         rc = lmv_check_connect(obd);
2124         if (rc)
2125                 RETURN(rc);
2126
2127         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2128                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2129
2130         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2131         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2132         if (IS_ERR(tgt))
2133                 RETURN(PTR_ERR(tgt));
2134
2135         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2136                         ea2len, request, mod);
2137
2138         RETURN(rc);
2139 }
2140
2141 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2142                      struct obd_capa *oc, struct ptlrpc_request **request)
2143 {
2144         struct obd_device       *obd = exp->exp_obd;
2145         struct lmv_obd          *lmv = &obd->u.lmv;
2146         struct lmv_tgt_desc     *tgt;
2147         int                      rc;
2148         ENTRY;
2149
2150         rc = lmv_check_connect(obd);
2151         if (rc != 0)
2152                 RETURN(rc);
2153
2154         tgt = lmv_find_target(lmv, fid);
2155         if (IS_ERR(tgt))
2156                 RETURN(PTR_ERR(tgt));
2157
2158         rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2159         RETURN(rc);
2160 }
2161
2162 /*
2163  * Adjust a set of pages, each page containing an array of lu_dirpages,
2164  * so that each page can be used as a single logical lu_dirpage.
2165  *
2166  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
2167  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
2168  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
2169  * value is used as a cookie to request the next lu_dirpage in a
2170  * directory listing that spans multiple pages (two in this example):
2171  *   ________
2172  *  |        |
2173  * .|--------v-------   -----.
2174  * |s|e|f|p|ent|ent| ... |ent|
2175  * '--|--------------   -----'   Each CFS_PAGE contains a single
2176  *    '------.                   lu_dirpage.
2177  * .---------v-------   -----.
2178  * |s|e|f|p|ent| 0 | ... | 0 |
2179  * '-----------------   -----'
2180  *
2181  * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
2182  * larger than LU_PAGE_SIZE, a single host page may contain multiple
2183  * lu_dirpages. After reading the lu_dirpages from the MDS, the
2184  * ldp_hash_end of the first lu_dirpage refers to the one immediately
2185  * after it in the same CFS_PAGE (arrows simplified for brevity, but
2186  * in general e0==s1, e1==s2, etc.):
2187  *
2188  * .--------------------   -----.
2189  * |s0|e0|f0|p|ent|ent| ... |ent|
2190  * |---v----------------   -----|
2191  * |s1|e1|f1|p|ent|ent| ... |ent|
2192  * |---v----------------   -----|  Here, each CFS_PAGE contains
2193  *             ...                 multiple lu_dirpages.
2194  * |---v----------------   -----|
2195  * |s'|e'|f'|p|ent|ent| ... |ent|
2196  * '---|----------------   -----'
2197  *     v
2198  * .----------------------------.
2199  * |        next CFS_PAGE       |
2200  *
2201  * This structure is transformed into a single logical lu_dirpage as follows:
2202  *
2203  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
2204  *   labeled 'next CFS_PAGE'.
2205  *
2206  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
2207  *   a hash collision with the next page exists.
2208  *
2209  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
2210  *   to the first entry of the next lu_dirpage.
2211  */
2212 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
2213 static void lmv_adjust_dirpages(struct page **pages, int ncfspgs, int nlupgs)
2214 {
2215         int i;
2216
2217         for (i = 0; i < ncfspgs; i++) {
2218                 struct lu_dirpage       *dp = kmap(pages[i]);
2219                 struct lu_dirpage       *first = dp;
2220                 struct lu_dirent        *end_dirent = NULL;
2221                 struct lu_dirent        *ent;
2222                 __u64                   hash_end = dp->ldp_hash_end;
2223                 __u32                   flags = dp->ldp_flags;
2224
2225                 while (--nlupgs > 0) {
2226                         ent = lu_dirent_start(dp);
2227                         for (end_dirent = ent; ent != NULL;
2228                              end_dirent = ent, ent = lu_dirent_next(ent));
2229
2230                         /* Advance dp to next lu_dirpage. */
2231                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
2232
2233                         /* Check if we've reached the end of the CFS_PAGE. */
2234                         if (!((unsigned long)dp & ~CFS_PAGE_MASK))
2235                                 break;
2236
2237                         /* Save the hash and flags of this lu_dirpage. */
2238                         hash_end = dp->ldp_hash_end;
2239                         flags = dp->ldp_flags;
2240
2241                         /* Check if lu_dirpage contains no entries. */
2242                         if (!end_dirent)
2243                                 break;
2244
2245                         /* Enlarge the end entry lde_reclen from 0 to
2246                          * first entry of next lu_dirpage. */
2247                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
2248                         end_dirent->lde_reclen =
2249                                 cpu_to_le16((char *)(dp->ldp_entries) -
2250                                             (char *)end_dirent);
2251                 }
2252
2253                 first->ldp_hash_end = hash_end;
2254                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
2255                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
2256
2257                 kunmap(pages[i]);
2258         }
2259         LASSERTF(nlupgs == 0, "left = %d", nlupgs);
2260 }
2261 #else
2262 #define lmv_adjust_dirpages(pages, ncfspgs, nlupgs) do {} while (0)
2263 #endif  /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
2264
2265 static int lmv_readpage(struct obd_export *exp, struct md_op_data *op_data,
2266                         struct page **pages, struct ptlrpc_request **request)
2267 {
2268         struct obd_device       *obd = exp->exp_obd;
2269         struct lmv_obd          *lmv = &obd->u.lmv;
2270         __u64                   offset = op_data->op_offset;
2271         int                     rc;
2272         int                     ncfspgs; /* pages read in PAGE_CACHE_SIZE */
2273         int                     nlupgs; /* pages read in LU_PAGE_SIZE */
2274         struct lmv_tgt_desc     *tgt;
2275         ENTRY;
2276
2277         rc = lmv_check_connect(obd);
2278         if (rc)
2279                 RETURN(rc);
2280
2281         CDEBUG(D_INODE, "READPAGE at "LPX64" from "DFID"\n",
2282                offset, PFID(&op_data->op_fid1));
2283
2284         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2285         if (IS_ERR(tgt))
2286                 RETURN(PTR_ERR(tgt));
2287
2288         rc = md_readpage(tgt->ltd_exp, op_data, pages, request);
2289         if (rc != 0)
2290                 RETURN(rc);
2291
2292         ncfspgs = ((*request)->rq_bulk->bd_nob_transferred +
2293                    PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2294         nlupgs = (*request)->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
2295         LASSERT(!((*request)->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
2296         LASSERT(ncfspgs > 0 && ncfspgs <= op_data->op_npages);
2297
2298         CDEBUG(D_INODE, "read %d(%d)/%d pages\n", ncfspgs, nlupgs,
2299                op_data->op_npages);
2300
2301         lmv_adjust_dirpages(pages, ncfspgs, nlupgs);
2302
2303         RETURN(rc);
2304 }
2305
2306 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2307                       struct ptlrpc_request **request)
2308 {
2309         struct obd_device       *obd = exp->exp_obd;
2310         struct lmv_obd          *lmv = &obd->u.lmv;
2311         struct lmv_tgt_desc     *tgt = NULL;
2312         struct mdt_body         *body;
2313         int                     rc;
2314         ENTRY;
2315
2316         rc = lmv_check_connect(obd);
2317         if (rc)
2318                 RETURN(rc);
2319 retry:
2320         /* Send unlink requests to the MDT where the child is located */
2321         if (likely(!fid_is_zero(&op_data->op_fid2)))
2322                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2323         else
2324                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2325         if (IS_ERR(tgt))
2326                 RETURN(PTR_ERR(tgt));
2327
2328         op_data->op_fsuid = current_fsuid();
2329         op_data->op_fsgid = current_fsgid();
2330         op_data->op_cap = cfs_curproc_cap_pack();
2331
2332         /*
2333          * If child's fid is given, cancel unused locks for it if it is from
2334          * another export than parent.
2335          *
2336          * LOOKUP lock for child (fid3) should also be cancelled on parent
2337          * tgt_tgt in mdc_unlink().
2338          */
2339         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2340
2341         /*
2342          * Cancel FULL locks on child (fid3).
2343          */
2344         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2345                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2346
2347         if (rc != 0)
2348                 RETURN(rc);
2349
2350         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2351                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2352
2353         rc = md_unlink(tgt->ltd_exp, op_data, request);
2354         if (rc != 0 && rc != -EREMOTE)
2355                 RETURN(rc);
2356
2357         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2358         if (body == NULL)
2359                 RETURN(-EPROTO);
2360
2361         /* Not cross-ref case, just get out of here. */
2362         if (likely(!(body->valid & OBD_MD_MDS)))
2363                 RETURN(0);
2364
2365         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2366                exp->exp_obd->obd_name, PFID(&body->fid1));
2367
2368         /* This is a remote object, try remote MDT, Note: it may
2369          * try more than 1 time here, Considering following case
2370          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2371          * 1. Initially A does not know where remote1 is, it send
2372          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2373          *    resend unlink RPC to MDT1 (retry 1st time).
2374          *
2375          * 2. During the unlink RPC in flight,
2376          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2377          *    and create new remote1, but on MDT0
2378          *
2379          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2380          *    /mnt/lustre, then lookup get fid of remote1, and find
2381          *    it is remote dir again, and replay -EREMOTE again.
2382          *
2383          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2384          *
2385          * In theory, it might try unlimited time here, but it should
2386          * be very rare case.  */
2387         op_data->op_fid2 = body->fid1;
2388         ptlrpc_req_finished(*request);
2389         *request = NULL;
2390
2391         goto retry;
2392 }
2393
2394 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2395 {
2396         struct lmv_obd *lmv = &obd->u.lmv;
2397         int rc = 0;
2398
2399         switch (stage) {
2400         case OBD_CLEANUP_EARLY:
2401                 /* XXX: here should be calling obd_precleanup() down to
2402                  * stack. */
2403                 break;
2404         case OBD_CLEANUP_EXPORTS:
2405                 fld_client_proc_fini(&lmv->lmv_fld);
2406                 lprocfs_obd_cleanup(obd);
2407                 lprocfs_free_md_stats(obd);
2408                 break;
2409         default:
2410                 break;
2411         }
2412         RETURN(rc);
2413 }
2414
2415 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2416                         __u32 keylen, void *key, __u32 *vallen, void *val,
2417                         struct lov_stripe_md *lsm)
2418 {
2419         struct obd_device       *obd;
2420         struct lmv_obd          *lmv;
2421         int                      rc = 0;
2422         ENTRY;
2423
2424         obd = class_exp2obd(exp);
2425         if (obd == NULL) {
2426                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2427                        exp->exp_handle.h_cookie);
2428                 RETURN(-EINVAL);
2429         }
2430
2431         lmv = &obd->u.lmv;
2432         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2433                 int i;
2434
2435                 rc = lmv_check_connect(obd);
2436                 if (rc)
2437                         RETURN(rc);
2438
2439                 LASSERT(*vallen == sizeof(__u32));
2440                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2441                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2442                         /*
2443                          * All tgts should be connected when this gets called.
2444                          */
2445                         if (tgt == NULL || tgt->ltd_exp == NULL)
2446                                 continue;
2447
2448                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2449                                           vallen, val, NULL))
2450                                 RETURN(0);
2451                 }
2452                 RETURN(-EINVAL);
2453         } else if (KEY_IS(KEY_MAX_EASIZE) || KEY_IS(KEY_CONN_DATA)) {
2454                 rc = lmv_check_connect(obd);
2455                 if (rc)
2456                         RETURN(rc);
2457
2458                 /*
2459                  * Forwarding this request to first MDS, it should know LOV
2460                  * desc.
2461                  */
2462                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2463                                   vallen, val, NULL);
2464                 if (!rc && KEY_IS(KEY_CONN_DATA))
2465                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2466                 RETURN(rc);
2467         } else if (KEY_IS(KEY_TGT_COUNT)) {
2468                 *((int *)val) = lmv->desc.ld_tgt_count;
2469                 RETURN(0);
2470         }
2471
2472         CDEBUG(D_IOCTL, "Invalid key\n");
2473         RETURN(-EINVAL);
2474 }
2475
2476 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2477                        obd_count keylen, void *key, obd_count vallen,
2478                        void *val, struct ptlrpc_request_set *set)
2479 {
2480         struct lmv_tgt_desc    *tgt;
2481         struct obd_device      *obd;
2482         struct lmv_obd         *lmv;
2483         int rc = 0;
2484         ENTRY;
2485
2486         obd = class_exp2obd(exp);
2487         if (obd == NULL) {
2488                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2489                        exp->exp_handle.h_cookie);
2490                 RETURN(-EINVAL);
2491         }
2492         lmv = &obd->u.lmv;
2493
2494         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2495                 int i, err = 0;
2496
2497                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2498                         tgt = lmv->tgts[i];
2499
2500                         if (tgt == NULL || tgt->ltd_exp == NULL)
2501                                 continue;
2502
2503                         err = obd_set_info_async(env, tgt->ltd_exp,
2504                                                  keylen, key, vallen, val, set);
2505                         if (err && rc == 0)
2506                                 rc = err;
2507                 }
2508
2509                 RETURN(rc);
2510         }
2511
2512         RETURN(-EINVAL);
2513 }
2514
2515 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2516                struct lov_stripe_md *lsm)
2517 {
2518         struct obd_device         *obd = class_exp2obd(exp);
2519         struct lmv_obd            *lmv = &obd->u.lmv;
2520         struct lmv_stripe_md      *meap;
2521         struct lmv_stripe_md      *lsmp;
2522         int                        mea_size;
2523         __u32                      i;
2524         ENTRY;
2525
2526         mea_size = lmv_get_easize(lmv);
2527         if (!lmmp)
2528                 RETURN(mea_size);
2529
2530         if (*lmmp && !lsm) {
2531                 OBD_FREE_LARGE(*lmmp, mea_size);
2532                 *lmmp = NULL;
2533                 RETURN(0);
2534         }
2535
2536         if (*lmmp == NULL) {
2537                 OBD_ALLOC_LARGE(*lmmp, mea_size);
2538                 if (*lmmp == NULL)
2539                         RETURN(-ENOMEM);
2540         }
2541
2542         if (!lsm)
2543                 RETURN(mea_size);
2544
2545         lsmp = (struct lmv_stripe_md *)lsm;
2546         meap = (struct lmv_stripe_md *)*lmmp;
2547
2548         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2549             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2550                 RETURN(-EINVAL);
2551
2552         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2553         meap->mea_count = cpu_to_le32(lsmp->mea_count);
2554         meap->mea_master = cpu_to_le32(lsmp->mea_master);
2555
2556         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2557                 meap->mea_ids[i] = lsmp->mea_ids[i];
2558                 fid_cpu_to_le(&meap->mea_ids[i], &lsmp->mea_ids[i]);
2559         }
2560
2561         RETURN(mea_size);
2562 }
2563
2564 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2565                  struct lov_mds_md *lmm, int lmm_size)
2566 {
2567         struct obd_device          *obd = class_exp2obd(exp);
2568         struct lmv_stripe_md      **tmea = (struct lmv_stripe_md **)lsmp;
2569         struct lmv_stripe_md       *mea = (struct lmv_stripe_md *)lmm;
2570         struct lmv_obd             *lmv = &obd->u.lmv;
2571         int                         mea_size;
2572         __u32                       i;
2573         __u32                       magic;
2574         ENTRY;
2575
2576         mea_size = lmv_get_easize(lmv);
2577         if (lsmp == NULL)
2578                 return mea_size;
2579
2580         if (*lsmp != NULL && lmm == NULL) {
2581                 OBD_FREE_LARGE(*tmea, mea_size);
2582                 *lsmp = NULL;
2583                 RETURN(0);
2584         }
2585
2586         LASSERT(mea_size == lmm_size);
2587
2588         OBD_ALLOC_LARGE(*tmea, mea_size);
2589         if (*tmea == NULL)
2590                 RETURN(-ENOMEM);
2591
2592         if (!lmm)
2593                 RETURN(mea_size);
2594
2595         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2596             mea->mea_magic == MEA_MAGIC_ALL_CHARS ||
2597             mea->mea_magic == MEA_MAGIC_HASH_SEGMENT)
2598         {
2599                 magic = le32_to_cpu(mea->mea_magic);
2600         } else {
2601                 /*
2602                  * Old mea is not handled here.
2603                  */
2604                 CERROR("Old not supportable EA is found\n");
2605                 LBUG();
2606         }
2607
2608         (*tmea)->mea_magic = magic;
2609         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2610         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2611
2612         for (i = 0; i < (*tmea)->mea_count; i++) {
2613                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2614                 fid_le_to_cpu(&(*tmea)->mea_ids[i], &(*tmea)->mea_ids[i]);
2615         }
2616         RETURN(mea_size);
2617 }
2618
2619 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2620                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
2621                              ldlm_cancel_flags_t flags, void *opaque)
2622 {
2623         struct obd_device       *obd = exp->exp_obd;
2624         struct lmv_obd          *lmv = &obd->u.lmv;
2625         int                      rc = 0;
2626         int                      err;
2627         __u32                    i;
2628         ENTRY;
2629
2630         LASSERT(fid != NULL);
2631
2632         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2633                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2634
2635                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
2636                         continue;
2637
2638                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
2639                                        opaque);
2640                 if (!rc)
2641                         rc = err;
2642         }
2643         RETURN(rc);
2644 }
2645
2646 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
2647                       __u64 *bits)
2648 {
2649         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2650         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
2651         int                      rc;
2652         ENTRY;
2653
2654         if (tgt == NULL || tgt->ltd_exp == NULL)
2655                 RETURN(-EINVAL);
2656         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
2657         RETURN(rc);
2658 }
2659
2660 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
2661                            const struct lu_fid *fid, ldlm_type_t type,
2662                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
2663                            struct lustre_handle *lockh)
2664 {
2665         struct obd_device       *obd = exp->exp_obd;
2666         struct lmv_obd          *lmv = &obd->u.lmv;
2667         ldlm_mode_t              rc;
2668         __u32                    i;
2669         ENTRY;
2670
2671         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2672
2673         /*
2674          * With CMD every object can have two locks in different namespaces:
2675          * lookup lock in space of mds storing direntry and update/open lock in
2676          * space of mds storing inode. Thus we check all targets, not only that
2677          * one fid was created in.
2678          */
2679         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2680                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2681
2682                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
2683                         continue;
2684
2685                 rc = md_lock_match(tgt->ltd_exp, flags, fid, type, policy, mode,
2686                                    lockh);
2687                 if (rc)
2688                         RETURN(rc);
2689         }
2690
2691         RETURN(0);
2692 }
2693
2694 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2695                       struct obd_export *dt_exp, struct obd_export *md_exp,
2696                       struct lustre_md *md)
2697 {
2698         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2699         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
2700
2701         if (tgt == NULL || tgt->ltd_exp == NULL)
2702                 RETURN(-EINVAL);
2703         return md_get_lustre_md(tgt->ltd_exp, req, dt_exp, md_exp, md);
2704 }
2705
2706 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2707 {
2708         struct obd_device       *obd = exp->exp_obd;
2709         struct lmv_obd          *lmv = &obd->u.lmv;
2710         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
2711         ENTRY;
2712
2713         if (md->mea)
2714                 obd_free_memmd(exp, (void *)&md->mea);
2715         if (tgt == NULL || tgt->ltd_exp == NULL)
2716                 RETURN(-EINVAL);
2717         RETURN(md_free_lustre_md(tgt->ltd_exp, md));
2718 }
2719
2720 int lmv_set_open_replay_data(struct obd_export *exp,
2721                              struct obd_client_handle *och,
2722                              struct lookup_intent *it)
2723 {
2724         struct obd_device       *obd = exp->exp_obd;
2725         struct lmv_obd          *lmv = &obd->u.lmv;
2726         struct lmv_tgt_desc     *tgt;
2727         ENTRY;
2728
2729         tgt = lmv_find_target(lmv, &och->och_fid);
2730         if (IS_ERR(tgt))
2731                 RETURN(PTR_ERR(tgt));
2732
2733         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
2734 }
2735
2736 int lmv_clear_open_replay_data(struct obd_export *exp,
2737                                struct obd_client_handle *och)
2738 {
2739         struct obd_device       *obd = exp->exp_obd;
2740         struct lmv_obd          *lmv = &obd->u.lmv;
2741         struct lmv_tgt_desc     *tgt;
2742         ENTRY;
2743
2744         tgt = lmv_find_target(lmv, &och->och_fid);
2745         if (IS_ERR(tgt))
2746                 RETURN(PTR_ERR(tgt));
2747
2748         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
2749 }
2750
2751 static int lmv_get_remote_perm(struct obd_export *exp,
2752                                const struct lu_fid *fid,
2753                                struct obd_capa *oc, __u32 suppgid,
2754                                struct ptlrpc_request **request)
2755 {
2756         struct obd_device       *obd = exp->exp_obd;
2757         struct lmv_obd          *lmv = &obd->u.lmv;
2758         struct lmv_tgt_desc     *tgt;
2759         int                      rc;
2760         ENTRY;
2761
2762         rc = lmv_check_connect(obd);
2763         if (rc)
2764                 RETURN(rc);
2765
2766         tgt = lmv_find_target(lmv, fid);
2767         if (IS_ERR(tgt))
2768                 RETURN(PTR_ERR(tgt));
2769
2770         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
2771         RETURN(rc);
2772 }
2773
2774 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2775                           renew_capa_cb_t cb)
2776 {
2777         struct obd_device       *obd = exp->exp_obd;
2778         struct lmv_obd          *lmv = &obd->u.lmv;
2779         struct lmv_tgt_desc     *tgt;
2780         int                      rc;
2781         ENTRY;
2782
2783         rc = lmv_check_connect(obd);
2784         if (rc)
2785                 RETURN(rc);
2786
2787         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
2788         if (IS_ERR(tgt))
2789                 RETURN(PTR_ERR(tgt));
2790
2791         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
2792         RETURN(rc);
2793 }
2794
2795 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
2796                     const struct req_msg_field *field, struct obd_capa **oc)
2797 {
2798         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2799         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
2800
2801         if (tgt == NULL || tgt->ltd_exp == NULL)
2802                 RETURN(-EINVAL);
2803         return md_unpack_capa(tgt->ltd_exp, req, field, oc);
2804 }
2805
2806 int lmv_intent_getattr_async(struct obd_export *exp,
2807                              struct md_enqueue_info *minfo,
2808                              struct ldlm_enqueue_info *einfo)
2809 {
2810         struct md_op_data       *op_data = &minfo->mi_data;
2811         struct obd_device       *obd = exp->exp_obd;
2812         struct lmv_obd          *lmv = &obd->u.lmv;
2813         struct lmv_tgt_desc     *tgt = NULL;
2814         int                      rc;
2815         ENTRY;
2816
2817         rc = lmv_check_connect(obd);
2818         if (rc)
2819                 RETURN(rc);
2820
2821         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2822         if (IS_ERR(tgt))
2823                 RETURN(PTR_ERR(tgt));
2824
2825         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
2826         RETURN(rc);
2827 }
2828
2829 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
2830                         struct lu_fid *fid, __u64 *bits)
2831 {
2832         struct obd_device       *obd = exp->exp_obd;
2833         struct lmv_obd          *lmv = &obd->u.lmv;
2834         struct lmv_tgt_desc     *tgt;
2835         int                      rc;
2836         ENTRY;
2837
2838         rc = lmv_check_connect(obd);
2839         if (rc)
2840                 RETURN(rc);
2841
2842         tgt = lmv_find_target(lmv, fid);
2843         if (IS_ERR(tgt))
2844                 RETURN(PTR_ERR(tgt));
2845
2846         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
2847         RETURN(rc);
2848 }
2849
2850 /**
2851  * For lmv, only need to send request to master MDT, and the master MDT will
2852  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
2853  * we directly fetch data from the slave MDTs.
2854  */
2855 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
2856                  struct obd_quotactl *oqctl)
2857 {
2858         struct obd_device   *obd = class_exp2obd(exp);
2859         struct lmv_obd      *lmv = &obd->u.lmv;
2860         struct lmv_tgt_desc *tgt = lmv->tgts[0];
2861         int                  rc = 0;
2862         __u32                i;
2863         __u64                curspace, curinodes;
2864         ENTRY;
2865
2866         if (tgt == NULL ||
2867             tgt->ltd_exp == NULL ||
2868             !tgt->ltd_active ||
2869             lmv->desc.ld_tgt_count == 0) {
2870                 CERROR("master lmv inactive\n");
2871                 RETURN(-EIO);
2872         }
2873
2874         if (oqctl->qc_cmd != Q_GETOQUOTA) {
2875                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2876                 RETURN(rc);
2877         }
2878
2879         curspace = curinodes = 0;
2880         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2881                 int err;
2882                 tgt = lmv->tgts[i];
2883
2884                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
2885                         continue;
2886
2887                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2888                 if (err) {
2889                         CERROR("getquota on mdt %d failed. %d\n", i, err);
2890                         if (!rc)
2891                                 rc = err;
2892                 } else {
2893                         curspace += oqctl->qc_dqblk.dqb_curspace;
2894                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
2895                 }
2896         }
2897         oqctl->qc_dqblk.dqb_curspace = curspace;
2898         oqctl->qc_dqblk.dqb_curinodes = curinodes;
2899
2900         RETURN(rc);
2901 }
2902
2903 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
2904                    struct obd_quotactl *oqctl)
2905 {
2906         struct obd_device       *obd = class_exp2obd(exp);
2907         struct lmv_obd          *lmv = &obd->u.lmv;
2908         struct lmv_tgt_desc     *tgt;
2909         __u32                    i;
2910         int                      rc = 0;
2911         ENTRY;
2912
2913         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2914                 int err;
2915                 tgt = lmv->tgts[i];
2916                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
2917                         CERROR("lmv idx %d inactive\n", i);
2918                         RETURN(-EIO);
2919                 }
2920
2921                 err = obd_quotacheck(tgt->ltd_exp, oqctl);
2922                 if (err && !rc)
2923                         rc = err;
2924         }
2925
2926         RETURN(rc);
2927 }
2928
2929 struct obd_ops lmv_obd_ops = {
2930         .o_owner                = THIS_MODULE,
2931         .o_setup                = lmv_setup,
2932         .o_cleanup              = lmv_cleanup,
2933         .o_precleanup           = lmv_precleanup,
2934         .o_process_config       = lmv_process_config,
2935         .o_connect              = lmv_connect,
2936         .o_disconnect           = lmv_disconnect,
2937         .o_statfs               = lmv_statfs,
2938         .o_get_info             = lmv_get_info,
2939         .o_set_info_async       = lmv_set_info_async,
2940         .o_packmd               = lmv_packmd,
2941         .o_unpackmd             = lmv_unpackmd,
2942         .o_notify               = lmv_notify,
2943         .o_get_uuid             = lmv_get_uuid,
2944         .o_iocontrol            = lmv_iocontrol,
2945         .o_quotacheck           = lmv_quotacheck,
2946         .o_quotactl             = lmv_quotactl
2947 };
2948
2949 struct md_ops lmv_md_ops = {
2950         .m_getstatus            = lmv_getstatus,
2951         .m_null_inode           = lmv_null_inode,
2952         .m_find_cbdata          = lmv_find_cbdata,
2953         .m_close                = lmv_close,
2954         .m_create               = lmv_create,
2955         .m_done_writing         = lmv_done_writing,
2956         .m_enqueue              = lmv_enqueue,
2957         .m_getattr              = lmv_getattr,
2958         .m_getxattr             = lmv_getxattr,
2959         .m_getattr_name         = lmv_getattr_name,
2960         .m_intent_lock          = lmv_intent_lock,
2961         .m_link                 = lmv_link,
2962         .m_rename               = lmv_rename,
2963         .m_setattr              = lmv_setattr,
2964         .m_setxattr             = lmv_setxattr,
2965         .m_fsync                = lmv_fsync,
2966         .m_readpage             = lmv_readpage,
2967         .m_unlink               = lmv_unlink,
2968         .m_init_ea_size         = lmv_init_ea_size,
2969         .m_cancel_unused        = lmv_cancel_unused,
2970         .m_set_lock_data        = lmv_set_lock_data,
2971         .m_lock_match           = lmv_lock_match,
2972         .m_get_lustre_md        = lmv_get_lustre_md,
2973         .m_free_lustre_md       = lmv_free_lustre_md,
2974         .m_set_open_replay_data = lmv_set_open_replay_data,
2975         .m_clear_open_replay_data = lmv_clear_open_replay_data,
2976         .m_renew_capa           = lmv_renew_capa,
2977         .m_unpack_capa          = lmv_unpack_capa,
2978         .m_get_remote_perm      = lmv_get_remote_perm,
2979         .m_intent_getattr_async = lmv_intent_getattr_async,
2980         .m_revalidate_lock      = lmv_revalidate_lock
2981 };
2982
2983 int __init lmv_init(void)
2984 {
2985         struct lprocfs_static_vars lvars;
2986
2987         lprocfs_lmv_init_vars(&lvars);
2988
2989         return class_register_type(&lmv_obd_ops, &lmv_md_ops, NULL,
2990 #ifndef HAVE_ONLY_PROCFS_SEQ
2991                                         lvars.module_vars,
2992 #endif
2993                                         LUSTRE_LMV_NAME, NULL);
2994 }
2995
2996 #ifdef __KERNEL__
2997 static void lmv_exit(void)
2998 {
2999         class_unregister_type(LUSTRE_LMV_NAME);
3000 }
3001
3002 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3003 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3004 MODULE_LICENSE("GPL");
3005
3006 module_init(lmv_init);
3007 module_exit(lmv_exit);
3008 #endif