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