lustre/lmv/lmv_obd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LMV
  38 #include <linux/slab.h>
  39 #include <linux/module.h>
  40 #include <linux/init.h>
  41 #include <linux/slab.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <linux/math64.h>
  45 #include <linux/seq_file.h>
  46 #include <linux/namei.h>
  47
  48 #include <lustre/lustre_idl.h>
  49 #include <obd_support.h>
  50 #include <lustre_lib.h>
  51 #include <lustre_net.h>
  52 #include <obd_class.h>
  53 #include <lustre_lmv.h>
  54 #include <lprocfs_status.h>
  55 #include <cl_object.h>
  56 #include <lustre_fid.h>
  57 #include <lustre_ioctl.h>
  58 #include "lmv_internal.h"
  59
  60 static void lmv_activate_target(struct lmv_obd *lmv,
  61                                 struct lmv_tgt_desc *tgt,
  62                                 int activate)
  63 {
  64         if (tgt->ltd_active == activate)
  65                 return;
  66
  67         tgt->ltd_active = activate;
  68         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  69 }
  70
  71 /**
  72  * Error codes:
  73  *
  74  *  -EINVAL  : UUID can't be found in the LMV's target list
  75  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  76  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  77  */
  78 static int lmv_set_mdc_active(struct lmv_obd *lmv,
  79                               const struct obd_uuid *uuid,
  80                               int activate)
  81 {
  82         struct lmv_tgt_desc     *tgt = NULL;
  83         struct obd_device       *obd;
  84         __u32                    i;
  85         int                      rc = 0;
  86         ENTRY;
  87
  88         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  89                         lmv, uuid->uuid, activate);
  90
  91         spin_lock(&lmv->lmv_lock);
  92         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
  93                 tgt = lmv->tgts[i];
  94                 if (tgt == NULL || tgt->ltd_exp == NULL)
  95                         continue;
  96
  97                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
  98                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
  99
 100                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 101                         break;
 102         }
 103
 104         if (i == lmv->desc.ld_tgt_count)
 105                 GOTO(out_lmv_lock, rc = -EINVAL);
 106
 107         obd = class_exp2obd(tgt->ltd_exp);
 108         if (obd == NULL)
 109                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 110
 111         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 112                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 113                obd->obd_type->typ_name, i);
 114         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 115
 116         if (tgt->ltd_active == activate) {
 117                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 118                        activate ? "" : "in");
 119                 GOTO(out_lmv_lock, rc);
 120         }
 121
 122         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 123                activate ? "" : "in");
 124         lmv_activate_target(lmv, tgt, activate);
 125         EXIT;
 126
 127  out_lmv_lock:
 128         spin_unlock(&lmv->lmv_lock);
 129         return rc;
 130 }
 131
 132 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 133 {
 134         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
 135         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
 136
 137         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
 138 }
 139
 140 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 141                       enum obd_notify_event ev, void *data)
 142 {
 143         struct obd_connect_data *conn_data;
 144         struct lmv_obd          *lmv = &obd->u.lmv;
 145         struct obd_uuid         *uuid;
 146         int                      rc = 0;
 147         ENTRY;
 148
 149         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 150                 CERROR("unexpected notification of %s %s!\n",
 151                        watched->obd_type->typ_name,
 152                        watched->obd_name);
 153                 RETURN(-EINVAL);
 154         }
 155
 156         uuid = &watched->u.cli.cl_target_uuid;
 157         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 158                 /*
 159                  * Set MDC as active before notifying the observer, so the
 160                  * observer can use the MDC normally.
 161                  */
 162                 rc = lmv_set_mdc_active(lmv, uuid,
 163                                         ev == OBD_NOTIFY_ACTIVE);
 164                 if (rc) {
 165                         CERROR("%sactivation of %s failed: %d\n",
 166                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 167                                uuid->uuid, rc);
 168                         RETURN(rc);
 169                 }
 170         } else if (ev == OBD_NOTIFY_OCD) {
 171                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 172                 /*
 173                  * XXX: Make sure that ocd_connect_flags from all targets are
 174                  * the same. Otherwise one of MDTs runs wrong version or
 175                  * something like this.  --umka
 176                  */
 177                 obd->obd_self_export->exp_connect_data = *conn_data;
 178         }
 179 #if 0
 180         else if (ev == OBD_NOTIFY_DISCON) {
 181                 /*
 182                  * For disconnect event, flush fld cache for failout MDS case.
 183                  */
 184                 fld_client_flush(&lmv->lmv_fld);
 185         }
 186 #endif
 187         /*
 188          * Pass the notification up the chain.
 189          */
 190         if (obd->obd_observer)
 191                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 192
 193         RETURN(rc);
 194 }
 195
 196 /**
 197  * This is fake connect function. Its purpose is to initialize lmv and say
 198  * caller that everything is okay. Real connection will be performed later.
 199  */
 200 static int lmv_connect(const struct lu_env *env,
 201                        struct obd_export **exp, struct obd_device *obd,
 202                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 203                        void *localdata)
 204 {
 205         struct lmv_obd        *lmv = &obd->u.lmv;
 206         struct lustre_handle  conn = { 0 };
 207         int                    rc = 0;
 208         ENTRY;
 209
 210         /*
 211          * We don't want to actually do the underlying connections more than
 212          * once, so keep track.
 213          */
 214         lmv->refcount++;
 215         if (lmv->refcount > 1) {
 216                 *exp = NULL;
 217                 RETURN(0);
 218         }
 219
 220         rc = class_connect(&conn, obd, cluuid);
 221         if (rc) {
 222                 CERROR("class_connection() returned %d\n", rc);
 223                 RETURN(rc);
 224         }
 225
 226         *exp = class_conn2export(&conn);
 227         class_export_get(*exp);
 228
 229         lmv->exp = *exp;
 230         lmv->connected = 0;
 231         lmv->cluuid = *cluuid;
 232
 233         if (data)
 234                 lmv->conn_data = *data;
 235
 236         if (lmv->targets_proc_entry == NULL) {
 237                 lmv->targets_proc_entry = lprocfs_seq_register("target_obds",
 238                                                         obd->obd_proc_entry,
 239                                                         NULL, NULL);
 240                 if (IS_ERR(lmv->targets_proc_entry)) {
 241                         CERROR("%s: cannot register "
 242                                "/proc/fs/lustre/%s/%s/target_obds\n",
 243                                obd->obd_name, obd->obd_type->typ_name,
 244                                obd->obd_name);
 245                         lmv->targets_proc_entry = NULL;
 246                 }
 247         }
 248
 249         /*
 250          * All real clients should perform actual connection right away, because
 251          * it is possible, that LMV will not have opportunity to connect targets
 252          * and MDC stuff will be called directly, for instance while reading
 253          * ../mdc/../kbytesfree procfs file, etc.
 254          */
 255         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
 256                 rc = lmv_check_connect(obd);
 257
 258         if (rc && lmv->targets_proc_entry != NULL)
 259                 lprocfs_remove(&lmv->targets_proc_entry);
 260         RETURN(rc);
 261 }
 262
 263 static int lmv_init_ea_size(struct obd_export *exp,
 264                             __u32 easize, __u32 def_easize,
 265                             __u32 cookiesize, __u32 def_cookiesize)
 266 {
 267         struct obd_device       *obd = exp->exp_obd;
 268         struct lmv_obd          *lmv = &obd->u.lmv;
 269         __u32                    i;
 270         int                      rc = 0;
 271         int                      change = 0;
 272         ENTRY;
 273
 274         if (lmv->max_easize < easize) {
 275                 lmv->max_easize = easize;
 276                 change = 1;
 277         }
 278         if (lmv->max_def_easize < def_easize) {
 279                 lmv->max_def_easize = def_easize;
 280                 change = 1;
 281         }
 282         if (lmv->max_cookiesize < cookiesize) {
 283                 lmv->max_cookiesize = cookiesize;
 284                 change = 1;
 285         }
 286         if (lmv->max_def_cookiesize < def_cookiesize) {
 287                 lmv->max_def_cookiesize = def_cookiesize;
 288                 change = 1;
 289         }
 290         if (change == 0)
 291                 RETURN(0);
 292
 293         if (lmv->connected == 0)
 294                 RETURN(0);
 295
 296         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 297                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 298
 299                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
 300                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 301                         continue;
 302                 }
 303
 304                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
 305                                      cookiesize, def_cookiesize);
 306                 if (rc) {
 307                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 308                                " rc = %d\n", obd->obd_name, i, rc);
 309                         break;
 310                 }
 311         }
 312         RETURN(rc);
 313 }
 314
 315 #define MAX_STRING_SIZE 128
 316
 317 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 318 {
 319         struct lmv_obd          *lmv = &obd->u.lmv;
 320         struct obd_uuid         *cluuid = &lmv->cluuid;
 321         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 322         struct obd_device       *mdc_obd;
 323         struct obd_export       *mdc_exp;
 324         struct lu_fld_target     target;
 325         int                      rc;
 326         ENTRY;
 327
 328         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 329                                         &obd->obd_uuid);
 330         if (!mdc_obd) {
 331                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 332                 RETURN(-EINVAL);
 333         }
 334
 335         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 336                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 337                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 338                 cluuid->uuid);
 339
 340         if (!mdc_obd->obd_set_up) {
 341                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 342                 RETURN(-EINVAL);
 343         }
 344
 345         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 346                          &lmv->conn_data, NULL);
 347         if (rc) {
 348                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 349                 RETURN(rc);
 350         }
 351
 352         /*
 353          * Init fid sequence client for this mdc and add new fld target.
 354          */
 355         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 356         if (rc)
 357                 RETURN(rc);
 358
 359         target.ft_srv = NULL;
 360         target.ft_exp = mdc_exp;
 361         target.ft_idx = tgt->ltd_idx;
 362
 363         fld_client_add_target(&lmv->lmv_fld, &target);
 364
 365         rc = obd_register_observer(mdc_obd, obd);
 366         if (rc) {
 367                 obd_disconnect(mdc_exp);
 368                 CERROR("target %s register_observer error %d\n",
 369                        tgt->ltd_uuid.uuid, rc);
 370                 RETURN(rc);
 371         }
 372
 373         if (obd->obd_observer) {
 374                 /*
 375                  * Tell the observer about the new target.
 376                  */
 377                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 378                                 OBD_NOTIFY_ACTIVE,
 379                                 (void *)(tgt - lmv->tgts[0]));
 380                 if (rc) {
 381                         obd_disconnect(mdc_exp);
 382                         RETURN(rc);
 383                 }
 384         }
 385
 386         tgt->ltd_active = 1;
 387         tgt->ltd_exp = mdc_exp;
 388         lmv->desc.ld_active_tgt_count++;
 389
 390         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
 391                         lmv->max_cookiesize, lmv->max_def_cookiesize);
 392
 393         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 394                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 395                 atomic_read(&obd->obd_refcount));
 396
 397         if (lmv->targets_proc_entry != NULL) {
 398                 struct proc_dir_entry *mdc_symlink;
 399
 400                 LASSERT(mdc_obd->obd_type != NULL);
 401                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 402                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 403                                                   lmv->targets_proc_entry,
 404                                                   "../../../%s/%s",
 405                                                   mdc_obd->obd_type->typ_name,
 406                                                   mdc_obd->obd_name);
 407                 if (mdc_symlink == NULL) {
 408                         CERROR("cannot register LMV target "
 409                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
 410                                obd->obd_type->typ_name, obd->obd_name,
 411                                mdc_obd->obd_name);
 412                 }
 413         }
 414         RETURN(0);
 415 }
 416
 417 static void lmv_del_target(struct lmv_obd *lmv, int index)
 418 {
 419         if (lmv->tgts[index] == NULL)
 420                 return;
 421
 422         OBD_FREE_PTR(lmv->tgts[index]);
 423         lmv->tgts[index] = NULL;
 424         return;
 425 }
 426
 427 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 428                            __u32 index, int gen)
 429 {
 430         struct lmv_obd      *lmv = &obd->u.lmv;
 431         struct lmv_tgt_desc *tgt;
 432         int                  orig_tgt_count = 0;
 433         int                  rc = 0;
 434         ENTRY;
 435
 436         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 437
 438         lmv_init_lock(lmv);
 439
 440         if (lmv->desc.ld_tgt_count == 0) {
 441                 struct obd_device *mdc_obd;
 442
 443                 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 444                                                 &obd->obd_uuid);
 445                 if (!mdc_obd) {
 446                         lmv_init_unlock(lmv);
 447                         CERROR("%s: Target %s not attached: rc = %d\n",
 448                                obd->obd_name, uuidp->uuid, -EINVAL);
 449                         RETURN(-EINVAL);
 450                 }
 451         }
 452
 453         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
 454                 tgt = lmv->tgts[index];
 455                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
 456                        " rc = %d\n", obd->obd_name,
 457                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
 458                 lmv_init_unlock(lmv);
 459                 RETURN(-EEXIST);
 460         }
 461
 462         if (index >= lmv->tgts_size) {
 463                 /* We need to reallocate the lmv target array. */
 464                 struct lmv_tgt_desc **newtgts, **old = NULL;
 465                 __u32 newsize = 1;
 466                 __u32 oldsize = 0;
 467
 468                 while (newsize < index + 1)
 469                         newsize = newsize << 1;
 470                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 471                 if (newtgts == NULL) {
 472                         lmv_init_unlock(lmv);
 473                         RETURN(-ENOMEM);
 474                 }
 475
 476                 if (lmv->tgts_size) {
 477                         memcpy(newtgts, lmv->tgts,
 478                                sizeof(*newtgts) * lmv->tgts_size);
 479                         old = lmv->tgts;
 480                         oldsize = lmv->tgts_size;
 481                 }
 482
 483                 lmv->tgts = newtgts;
 484                 lmv->tgts_size = newsize;
 485                 smp_rmb();
 486                 if (old)
 487                         OBD_FREE(old, sizeof(*old) * oldsize);
 488
 489                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
 490                        lmv->tgts_size);
 491         }
 492
 493         OBD_ALLOC_PTR(tgt);
 494         if (!tgt) {
 495                 lmv_init_unlock(lmv);
 496                 RETURN(-ENOMEM);
 497         }
 498
 499         mutex_init(&tgt->ltd_fid_mutex);
 500         tgt->ltd_idx = index;
 501         tgt->ltd_uuid = *uuidp;
 502         tgt->ltd_active = 0;
 503         lmv->tgts[index] = tgt;
 504         if (index >= lmv->desc.ld_tgt_count) {
 505                 orig_tgt_count = lmv->desc.ld_tgt_count;
 506                 lmv->desc.ld_tgt_count = index + 1;
 507         }
 508
 509         if (lmv->connected) {
 510                 rc = lmv_connect_mdc(obd, tgt);
 511                 if (rc != 0) {
 512                         spin_lock(&lmv->lmv_lock);
 513                         if (lmv->desc.ld_tgt_count == index + 1)
 514                                 lmv->desc.ld_tgt_count = orig_tgt_count;
 515                         memset(tgt, 0, sizeof(*tgt));
 516                         spin_unlock(&lmv->lmv_lock);
 517                 } else {
 518                         int easize = sizeof(struct lmv_stripe_md) +
 519                                 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
 520                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
 521                 }
 522         }
 523
 524         lmv_init_unlock(lmv);
 525         RETURN(rc);
 526 }
 527
 528 int lmv_check_connect(struct obd_device *obd)
 529 {
 530         struct lmv_obd          *lmv = &obd->u.lmv;
 531         struct lmv_tgt_desc     *tgt;
 532         __u32                    i;
 533         int                      rc;
 534         int                      easize;
 535         ENTRY;
 536
 537         if (lmv->connected)
 538                 RETURN(0);
 539
 540         lmv_init_lock(lmv);
 541         if (lmv->connected) {
 542                 lmv_init_unlock(lmv);
 543                 RETURN(0);
 544         }
 545
 546         if (lmv->desc.ld_tgt_count == 0) {
 547                 lmv_init_unlock(lmv);
 548                 CERROR("%s: no targets configured.\n", obd->obd_name);
 549                 RETURN(-EINVAL);
 550         }
 551
 552         LASSERT(lmv->tgts != NULL);
 553
 554         if (lmv->tgts[0] == NULL) {
 555                 lmv_init_unlock(lmv);
 556                 CERROR("%s: no target configured for index 0.\n",
 557                        obd->obd_name);
 558                 RETURN(-EINVAL);
 559         }
 560
 561         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 562                lmv->cluuid.uuid, obd->obd_name);
 563
 564         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 565                 tgt = lmv->tgts[i];
 566                 if (tgt == NULL)
 567                         continue;
 568                 rc = lmv_connect_mdc(obd, tgt);
 569                 if (rc)
 570                         GOTO(out_disc, rc);
 571         }
 572
 573         class_export_put(lmv->exp);
 574         lmv->connected = 1;
 575         easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
 576         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
 577         lmv_init_unlock(lmv);
 578         RETURN(0);
 579
 580  out_disc:
 581         while (i-- > 0) {
 582                 int rc2;
 583                 tgt = lmv->tgts[i];
 584                 if (tgt == NULL)
 585                         continue;
 586                 tgt->ltd_active = 0;
 587                 if (tgt->ltd_exp) {
 588                         --lmv->desc.ld_active_tgt_count;
 589                         rc2 = obd_disconnect(tgt->ltd_exp);
 590                         if (rc2) {
 591                                 CERROR("LMV target %s disconnect on "
 592                                        "MDC idx %d: error %d\n",
 593                                        tgt->ltd_uuid.uuid, i, rc2);
 594                         }
 595                 }
 596         }
 597         class_disconnect(lmv->exp);
 598         lmv_init_unlock(lmv);
 599         RETURN(rc);
 600 }
 601
 602 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 603 {
 604         struct lmv_obd         *lmv = &obd->u.lmv;
 605         struct obd_device      *mdc_obd;
 606         int                     rc;
 607         ENTRY;
 608
 609         LASSERT(tgt != NULL);
 610         LASSERT(obd != NULL);
 611
 612         mdc_obd = class_exp2obd(tgt->ltd_exp);
 613
 614         if (mdc_obd) {
 615                 mdc_obd->obd_force = obd->obd_force;
 616                 mdc_obd->obd_fail = obd->obd_fail;
 617                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 618         }
 619
 620         if (lmv->targets_proc_entry != NULL)
 621                 lprocfs_remove_proc_entry(mdc_obd->obd_name,
 622                                           lmv->targets_proc_entry);
 623
 624         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 625         if (rc)
 626                 CERROR("Can't finanize fids factory\n");
 627
 628         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 629                tgt->ltd_exp->exp_obd->obd_name,
 630                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 631
 632         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 633         rc = obd_disconnect(tgt->ltd_exp);
 634         if (rc) {
 635                 if (tgt->ltd_active) {
 636                         CERROR("Target %s disconnect error %d\n",
 637                                tgt->ltd_uuid.uuid, rc);
 638                 }
 639         }
 640
 641         lmv_activate_target(lmv, tgt, 0);
 642         tgt->ltd_exp = NULL;
 643         RETURN(0);
 644 }
 645
 646 static int lmv_disconnect(struct obd_export *exp)
 647 {
 648         struct obd_device       *obd = class_exp2obd(exp);
 649         struct lmv_obd          *lmv = &obd->u.lmv;
 650         int                      rc;
 651         __u32                    i;
 652         ENTRY;
 653
 654         if (!lmv->tgts)
 655                 goto out_local;
 656
 657         /*
 658          * Only disconnect the underlying layers on the final disconnect.
 659          */
 660         lmv->refcount--;
 661         if (lmv->refcount != 0)
 662                 goto out_local;
 663
 664         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 665                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
 666                         continue;
 667
 668                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
 669         }
 670
 671         if (lmv->targets_proc_entry != NULL)
 672                 lprocfs_remove(&lmv->targets_proc_entry);
 673         else
 674                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 675                        obd->obd_type->typ_name, obd->obd_name);
 676
 677 out_local:
 678         /*
 679          * This is the case when no real connection is established by
 680          * lmv_check_connect().
 681          */
 682         if (!lmv->connected)
 683                 class_export_put(exp);
 684         rc = class_disconnect(exp);
 685         if (lmv->refcount == 0)
 686                 lmv->connected = 0;
 687         RETURN(rc);
 688 }
 689
 690 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
 691 {
 692         struct obd_device       *obddev = class_exp2obd(exp);
 693         struct lmv_obd          *lmv = &obddev->u.lmv;
 694         struct getinfo_fid2path *gf;
 695         struct lmv_tgt_desc     *tgt;
 696         struct getinfo_fid2path *remote_gf = NULL;
 697         int                     remote_gf_size = 0;
 698         int                     rc;
 699
 700         gf = (struct getinfo_fid2path *)karg;
 701         tgt = lmv_find_target(lmv, &gf->gf_fid);
 702         if (IS_ERR(tgt))
 703                 RETURN(PTR_ERR(tgt));
 704
 705 repeat_fid2path:
 706         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 707         if (rc != 0 && rc != -EREMOTE)
 708                 GOTO(out_fid2path, rc);
 709
 710         /* If remote_gf != NULL, it means just building the
 711          * path on the remote MDT, copy this path segement to gf */
 712         if (remote_gf != NULL) {
 713                 struct getinfo_fid2path *ori_gf;
 714                 char *ptr;
 715
 716                 ori_gf = (struct getinfo_fid2path *)karg;
 717                 if (strlen(ori_gf->gf_path) +
 718                     strlen(gf->gf_path) > ori_gf->gf_pathlen)
 719                         GOTO(out_fid2path, rc = -EOVERFLOW);
 720
 721                 ptr = ori_gf->gf_path;
 722
 723                 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
 724                         strlen(ori_gf->gf_path));
 725
 726                 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
 727                 ptr += strlen(gf->gf_path);
 728                 *ptr = '/';
 729         }
 730
 731         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
 732                tgt->ltd_exp->exp_obd->obd_name,
 733                gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 734                gf->gf_linkno);
 735
 736         if (rc == 0)
 737                 GOTO(out_fid2path, rc);
 738
 739         /* sigh, has to go to another MDT to do path building further */
 740         if (remote_gf == NULL) {
 741                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 742                 OBD_ALLOC(remote_gf, remote_gf_size);
 743                 if (remote_gf == NULL)
 744                         GOTO(out_fid2path, rc = -ENOMEM);
 745                 remote_gf->gf_pathlen = PATH_MAX;
 746         }
 747
 748         if (!fid_is_sane(&gf->gf_fid)) {
 749                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 750                        tgt->ltd_exp->exp_obd->obd_name,
 751                        PFID(&gf->gf_fid), -EINVAL);
 752                 GOTO(out_fid2path, rc = -EINVAL);
 753         }
 754
 755         tgt = lmv_find_target(lmv, &gf->gf_fid);
 756         if (IS_ERR(tgt))
 757                 GOTO(out_fid2path, rc = -EINVAL);
 758
 759         remote_gf->gf_fid = gf->gf_fid;
 760         remote_gf->gf_recno = -1;
 761         remote_gf->gf_linkno = -1;
 762         memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
 763         gf = remote_gf;
 764         goto repeat_fid2path;
 765
 766 out_fid2path:
 767         if (remote_gf != NULL)
 768                 OBD_FREE(remote_gf, remote_gf_size);
 769         RETURN(rc);
 770 }
 771
 772 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 773                              const struct hsm_user_request *hur,
 774                              const struct lmv_tgt_desc *tgt_mds)
 775 {
 776         __u32                    i;
 777         int                      nr = 0;
 778         struct lmv_tgt_desc     *curr_tgt;
 779
 780         /* count how many requests must be sent to the given target */
 781         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 782                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
 783                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 784                         nr++;
 785         }
 786         return nr;
 787 }
 788
 789 static void lmv_hsm_req_build(struct lmv_obd *lmv,
 790                               struct hsm_user_request *hur_in,
 791                               const struct lmv_tgt_desc *tgt_mds,
 792                               struct hsm_user_request *hur_out)
 793 {
 794         __u32                    i, nr_out;
 795         struct lmv_tgt_desc     *curr_tgt;
 796
 797         /* build the hsm_user_request for the given target */
 798         hur_out->hur_request = hur_in->hur_request;
 799         nr_out = 0;
 800         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 801                 curr_tgt = lmv_find_target(lmv,
 802                                            &hur_in->hur_user_item[i].hui_fid);
 803                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 804                         hur_out->hur_user_item[nr_out] =
 805                                                 hur_in->hur_user_item[i];
 806                         nr_out++;
 807                 }
 808         }
 809         hur_out->hur_request.hr_itemcount = nr_out;
 810         memcpy(hur_data(hur_out), hur_data(hur_in),
 811                hur_in->hur_request.hr_data_len);
 812 }
 813
 814 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
 815                                  struct lustre_kernelcomm *lk, void *uarg)
 816 {
 817         __u32                    i;
 818         int                      rc;
 819         struct kkuc_ct_data     *kcd = NULL;
 820         ENTRY;
 821
 822         /* unregister request (call from llapi_hsm_copytool_fini) */
 823         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 824                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 825
 826                 if (tgt == NULL || tgt->ltd_exp == NULL)
 827                         continue;
 828                 /* best effort: try to clean as much as possible
 829                  * (continue on error) */
 830                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 831         }
 832
 833         /* Whatever the result, remove copytool from kuc groups.
 834          * Unreached coordinators will get EPIPE on next requests
 835          * and will unregister automatically.
 836          */
 837         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
 838         if (kcd != NULL)
 839                 OBD_FREE_PTR(kcd);
 840
 841         RETURN(rc);
 842 }
 843
 844 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
 845                                struct lustre_kernelcomm *lk, void *uarg)
 846 {
 847         struct file             *filp;
 848         __u32                    i, j;
 849         int                      err, rc;
 850         bool                     any_set = false;
 851         struct kkuc_ct_data     *kcd;
 852         ENTRY;
 853
 854         /* All or nothing: try to register to all MDS.
 855          * In case of failure, unregister from previous MDS,
 856          * except if it because of inactive target. */
 857         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 858                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 859
 860                 if (tgt == NULL || tgt->ltd_exp == NULL)
 861                         continue;
 862                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 863                 if (err) {
 864                         if (tgt->ltd_active) {
 865                                 /* permanent error */
 866                                 CERROR("%s: iocontrol MDC %s on MDT"
 867                                        " idx %d cmd %x: err = %d\n",
 868                                        class_exp2obd(lmv->exp)->obd_name,
 869                                        tgt->ltd_uuid.uuid, i, cmd, err);
 870                                 rc = err;
 871                                 lk->lk_flags |= LK_FLG_STOP;
 872                                 /* unregister from previous MDS */
 873                                 for (j = 0; j < i; j++) {
 874                                         tgt = lmv->tgts[j];
 875                                         if (tgt == NULL || tgt->ltd_exp == NULL)
 876                                                 continue;
 877                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
 878                                                       lk, uarg);
 879                                 }
 880                                 RETURN(rc);
 881                         }
 882                         /* else: transient error.
 883                          * kuc will register to the missing MDT
 884                          * when it is back */
 885                 } else {
 886                         any_set = true;
 887                 }
 888         }
 889
 890         if (!any_set)
 891                 /* no registration done: return error */
 892                 RETURN(-ENOTCONN);
 893
 894         /* at least one registration done, with no failure */
 895         filp = fget(lk->lk_wfd);
 896         if (filp == NULL)
 897                 RETURN(-EBADF);
 898
 899         OBD_ALLOC_PTR(kcd);
 900         if (kcd == NULL) {
 901                 fput(filp);
 902                 RETURN(-ENOMEM);
 903         }
 904         kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
 905         kcd->kcd_uuid = lmv->cluuid;
 906         kcd->kcd_archive = lk->lk_data;
 907
 908         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
 909         if (rc != 0) {
 910                 if (filp != NULL)
 911                         fput(filp);
 912                 OBD_FREE_PTR(kcd);
 913         }
 914
 915         RETURN(rc);
 916 }
 917
 918
 919
 920
 921 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 922                          int len, void *karg, void *uarg)
 923 {
 924         struct obd_device       *obddev = class_exp2obd(exp);
 925         struct lmv_obd          *lmv = &obddev->u.lmv;
 926         struct lmv_tgt_desc     *tgt = NULL;
 927         __u32                    i = 0;
 928         int                      rc = 0;
 929         int                      set = 0;
 930         __u32                    count = lmv->desc.ld_tgt_count;
 931         ENTRY;
 932
 933         if (count == 0)
 934                 RETURN(-ENOTTY);
 935
 936         switch (cmd) {
 937         case IOC_OBD_STATFS: {
 938                 struct obd_ioctl_data *data = karg;
 939                 struct obd_device *mdc_obd;
 940                 struct obd_statfs stat_buf = {0};
 941                 __u32 index;
 942
 943                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 944                 if ((index >= count))
 945                         RETURN(-ENODEV);
 946
 947                 tgt = lmv->tgts[index];
 948                 if (tgt == NULL || !tgt->ltd_active)
 949                         RETURN(-ENODATA);
 950
 951                 mdc_obd = class_exp2obd(tgt->ltd_exp);
 952                 if (!mdc_obd)
 953                         RETURN(-EINVAL);
 954
 955                 /* copy UUID */
 956                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 957                                  min((int) data->ioc_plen2,
 958                                      (int) sizeof(struct obd_uuid))))
 959                         RETURN(-EFAULT);
 960
 961                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
 962                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 963                                 0);
 964                 if (rc)
 965                         RETURN(rc);
 966                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
 967                                  min((int) data->ioc_plen1,
 968                                      (int) sizeof(stat_buf))))
 969                         RETURN(-EFAULT);
 970                 break;
 971         }
 972         case OBD_IOC_QUOTACTL: {
 973                 struct if_quotactl *qctl = karg;
 974                 struct obd_quotactl *oqctl;
 975
 976                 if (qctl->qc_valid == QC_MDTIDX) {
 977                         if (count <= qctl->qc_idx)
 978                                 RETURN(-EINVAL);
 979
 980                         tgt = lmv->tgts[qctl->qc_idx];
 981                         if (tgt == NULL || tgt->ltd_exp == NULL)
 982                                 RETURN(-EINVAL);
 983                 } else if (qctl->qc_valid == QC_UUID) {
 984                         for (i = 0; i < count; i++) {
 985                                 tgt = lmv->tgts[i];
 986                                 if (tgt == NULL)
 987                                         continue;
 988                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 989                                                      &qctl->obd_uuid))
 990                                         continue;
 991
 992                                 if (tgt->ltd_exp == NULL)
 993                                         RETURN(-EINVAL);
 994
 995                                 break;
 996                         }
 997                 } else {
 998                         RETURN(-EINVAL);
 999                 }
1000
1001                 if (i >= count)
1002                         RETURN(-EAGAIN);
1003
1004                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1005                 OBD_ALLOC_PTR(oqctl);
1006                 if (!oqctl)
1007                         RETURN(-ENOMEM);
1008
1009                 QCTL_COPY(oqctl, qctl);
1010                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1011                 if (rc == 0) {
1012                         QCTL_COPY(qctl, oqctl);
1013                         qctl->qc_valid = QC_MDTIDX;
1014                         qctl->obd_uuid = tgt->ltd_uuid;
1015                 }
1016                 OBD_FREE_PTR(oqctl);
1017                 break;
1018         }
1019         case OBD_IOC_CHANGELOG_SEND:
1020         case OBD_IOC_CHANGELOG_CLEAR: {
1021                 struct ioc_changelog *icc = karg;
1022
1023                 if (icc->icc_mdtindex >= count)
1024                         RETURN(-ENODEV);
1025
1026                 tgt = lmv->tgts[icc->icc_mdtindex];
1027                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1028                         RETURN(-ENODEV);
1029                 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1030                 break;
1031         }
1032         case LL_IOC_GET_CONNECT_FLAGS: {
1033                 tgt = lmv->tgts[0];
1034                 if (tgt == NULL || tgt->ltd_exp == NULL)
1035                         RETURN(-ENODATA);
1036                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1037                 break;
1038         }
1039         case LL_IOC_FID2MDTIDX: {
1040                 struct lu_fid *fid = karg;
1041                 int             mdt_index;
1042
1043                 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1044                 if (rc != 0)
1045                         RETURN(rc);
1046
1047                 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1048                  * point to user space memory for FID2MDTIDX. */
1049                 *(__u32 *)uarg = mdt_index;
1050                 break;
1051         }
1052         case OBD_IOC_FID2PATH: {
1053                 rc = lmv_fid2path(exp, len, karg, uarg);
1054                 break;
1055         }
1056         case LL_IOC_HSM_STATE_GET:
1057         case LL_IOC_HSM_STATE_SET:
1058         case LL_IOC_HSM_ACTION: {
1059                 struct md_op_data       *op_data = karg;
1060
1061                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1062                 if (IS_ERR(tgt))
1063                         RETURN(PTR_ERR(tgt));
1064
1065                 if (tgt->ltd_exp == NULL)
1066                         RETURN(-EINVAL);
1067
1068                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1069                 break;
1070         }
1071         case LL_IOC_HSM_PROGRESS: {
1072                 const struct hsm_progress_kernel *hpk = karg;
1073
1074                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1075                 if (IS_ERR(tgt))
1076                         RETURN(PTR_ERR(tgt));
1077                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1078                 break;
1079         }
1080         case LL_IOC_HSM_REQUEST: {
1081                 struct hsm_user_request *hur = karg;
1082                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1083
1084                 if (reqcount == 0)
1085                         RETURN(0);
1086
1087                 /* if the request is about a single fid
1088                  * or if there is a single MDS, no need to split
1089                  * the request. */
1090                 if (reqcount == 1 || count == 1) {
1091                         tgt = lmv_find_target(lmv,
1092                                               &hur->hur_user_item[0].hui_fid);
1093                         if (IS_ERR(tgt))
1094                                 RETURN(PTR_ERR(tgt));
1095                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1096                 } else {
1097                         /* split fid list to their respective MDS */
1098                         for (i = 0; i < count; i++) {
1099                                 unsigned int            nr, reqlen;
1100                                 int                     rc1;
1101                                 struct hsm_user_request *req;
1102
1103                                 tgt = lmv->tgts[i];
1104                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1105                                         continue;
1106
1107                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1108                                 if (nr == 0) /* nothing for this MDS */
1109                                         continue;
1110
1111                                 /* build a request with fids for this MDS */
1112                                 reqlen = offsetof(typeof(*hur),
1113                                                   hur_user_item[nr])
1114                                                 + hur->hur_request.hr_data_len;
1115                                 OBD_ALLOC_LARGE(req, reqlen);
1116                                 if (req == NULL)
1117                                         RETURN(-ENOMEM);
1118
1119                                 lmv_hsm_req_build(lmv, hur, tgt, req);
1120
1121                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1122                                                     req, uarg);
1123                                 if (rc1 != 0 && rc == 0)
1124                                         rc = rc1;
1125                                 OBD_FREE_LARGE(req, reqlen);
1126                         }
1127                 }
1128                 break;
1129         }
1130         case LL_IOC_LOV_SWAP_LAYOUTS: {
1131                 struct md_op_data       *op_data = karg;
1132                 struct lmv_tgt_desc     *tgt1, *tgt2;
1133
1134                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1135                 if (IS_ERR(tgt1))
1136                         RETURN(PTR_ERR(tgt1));
1137
1138                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1139                 if (IS_ERR(tgt2))
1140                         RETURN(PTR_ERR(tgt2));
1141
1142                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1143                         RETURN(-EINVAL);
1144
1145                 /* only files on same MDT can have their layouts swapped */
1146                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1147                         RETURN(-EPERM);
1148
1149                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1150                 break;
1151         }
1152         case LL_IOC_HSM_CT_START: {
1153                 struct lustre_kernelcomm *lk = karg;
1154                 if (lk->lk_flags & LK_FLG_STOP)
1155                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1156                 else
1157                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1158                 break;
1159         }
1160         default:
1161                 for (i = 0; i < count; i++) {
1162                         struct obd_device *mdc_obd;
1163                         int err;
1164
1165                         tgt = lmv->tgts[i];
1166                         if (tgt == NULL || tgt->ltd_exp == NULL)
1167                                 continue;
1168                         /* ll_umount_begin() sets force flag but for lmv, not
1169                          * mdc. Let's pass it through */
1170                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1171                         mdc_obd->obd_force = obddev->obd_force;
1172                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1173                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1174                                 RETURN(err);
1175                         } else if (err) {
1176                                 if (tgt->ltd_active) {
1177                                         CERROR("error: iocontrol MDC %s on MDT"
1178                                                " idx %d cmd %x: err = %d\n",
1179                                                tgt->ltd_uuid.uuid, i, cmd, err);
1180                                         if (!rc)
1181                                                 rc = err;
1182                                 }
1183                         } else
1184                                 set = 1;
1185                 }
1186                 if (!set && !rc)
1187                         rc = -EIO;
1188         }
1189         RETURN(rc);
1190 }
1191
1192 #if 0
1193 static int lmv_all_chars_policy(int count, const char *name,
1194                                 int len)
1195 {
1196         unsigned int c = 0;
1197
1198         while (len > 0)
1199                 c += name[--len];
1200         c = c % count;
1201         return c;
1202 }
1203
1204 static int lmv_nid_policy(struct lmv_obd *lmv)
1205 {
1206         struct obd_import *imp;
1207         __u32              id;
1208
1209         /*
1210          * XXX: To get nid we assume that underlying obd device is mdc.
1211          */
1212         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1213         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1214         return id % lmv->desc.ld_tgt_count;
1215 }
1216
1217 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1218                           placement_policy_t placement)
1219 {
1220         switch (placement) {
1221         case PLACEMENT_CHAR_POLICY:
1222                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1223                                             op_data->op_name,
1224                                             op_data->op_namelen);
1225         case PLACEMENT_NID_POLICY:
1226                 return lmv_nid_policy(lmv);
1227
1228         default:
1229                 break;
1230         }
1231
1232         CERROR("Unsupported placement policy %x\n", placement);
1233         return -EINVAL;
1234 }
1235 #endif
1236
1237 /**
1238  * This is _inode_ placement policy function (not name).
1239  */
1240 static int lmv_placement_policy(struct obd_device *obd,
1241                                 struct md_op_data *op_data, u32 *mds)
1242 {
1243         struct lmv_obd          *lmv = &obd->u.lmv;
1244         ENTRY;
1245
1246         LASSERT(mds != NULL);
1247
1248         if (lmv->desc.ld_tgt_count == 1) {
1249                 *mds = 0;
1250                 RETURN(0);
1251         }
1252
1253         /**
1254          * If stripe_offset is provided during setdirstripe
1255          * (setdirstripe -i xx), xx MDS will be choosen.
1256          */
1257         if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1258                 struct lmv_user_md *lum;
1259
1260                 lum = op_data->op_data;
1261
1262                 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1263                         *mds = le32_to_cpu(lum->lum_stripe_offset);
1264                 } else {
1265                         /* -1 means default, which will be in the same MDT with
1266                          * the stripe */
1267                         *mds = op_data->op_mds;
1268                         lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1269                 }
1270         } else {
1271                 /* Allocate new fid on target according to operation type and
1272                  * parent home mds. */
1273                 *mds = op_data->op_mds;
1274         }
1275
1276         RETURN(0);
1277 }
1278
1279 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1280 {
1281         struct lmv_tgt_desc     *tgt;
1282         int                      rc;
1283         ENTRY;
1284
1285         tgt = lmv_get_target(lmv, mds, NULL);
1286         if (IS_ERR(tgt))
1287                 RETURN(PTR_ERR(tgt));
1288
1289         /*
1290          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1291          * on server that seq in new allocated fid is not yet known.
1292          */
1293         mutex_lock(&tgt->ltd_fid_mutex);
1294
1295         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1296                 GOTO(out, rc = -ENODEV);
1297
1298         /*
1299          * Asking underlying tgt layer to allocate new fid.
1300          */
1301         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1302         if (rc > 0) {
1303                 LASSERT(fid_is_sane(fid));
1304                 rc = 0;
1305         }
1306
1307         EXIT;
1308 out:
1309         mutex_unlock(&tgt->ltd_fid_mutex);
1310         return rc;
1311 }
1312
1313 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1314                   struct lu_fid *fid, struct md_op_data *op_data)
1315 {
1316         struct obd_device     *obd = class_exp2obd(exp);
1317         struct lmv_obd        *lmv = &obd->u.lmv;
1318         u32                    mds = 0;
1319         int                    rc;
1320         ENTRY;
1321
1322         LASSERT(op_data != NULL);
1323         LASSERT(fid != NULL);
1324
1325         rc = lmv_placement_policy(obd, op_data, &mds);
1326         if (rc) {
1327                 CERROR("Can't get target for allocating fid, "
1328                        "rc %d\n", rc);
1329                 RETURN(rc);
1330         }
1331
1332         rc = __lmv_fid_alloc(lmv, fid, mds);
1333         if (rc) {
1334                 CERROR("Can't alloc new fid, rc %d\n", rc);
1335                 RETURN(rc);
1336         }
1337
1338         RETURN(rc);
1339 }
1340
1341 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1342 {
1343         struct lmv_obd  *lmv = &obd->u.lmv;
1344         struct lmv_desc *desc;
1345         int             rc;
1346         ENTRY;
1347
1348         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1349                 CERROR("LMV setup requires a descriptor\n");
1350                 RETURN(-EINVAL);
1351         }
1352
1353         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1354         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1355                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1356                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1357                 RETURN(-EINVAL);
1358         }
1359
1360         lmv->tgts_size = 32U;
1361         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1362         if (lmv->tgts == NULL)
1363                 RETURN(-ENOMEM);
1364
1365         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1366         lmv->desc.ld_tgt_count = 0;
1367         lmv->desc.ld_active_tgt_count = 0;
1368         lmv->max_cookiesize = 0;
1369         lmv->max_def_easize = 0;
1370         lmv->max_easize = 0;
1371         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1372
1373         spin_lock_init(&lmv->lmv_lock);
1374         mutex_init(&lmv->init_mutex);
1375
1376 #ifdef LPROCFS
1377         obd->obd_vars = lprocfs_lmv_obd_vars;
1378         lprocfs_obd_setup(obd);
1379         lprocfs_alloc_md_stats(obd, 0);
1380         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1381                                 0444, &lmv_proc_target_fops, obd);
1382         if (rc)
1383                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1384                       obd->obd_name, rc);
1385 #endif
1386         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1387                              LUSTRE_CLI_FLD_HASH_DHT);
1388         if (rc) {
1389                 CERROR("Can't init FLD, err %d\n", rc);
1390                 GOTO(out, rc);
1391         }
1392
1393         RETURN(0);
1394
1395 out:
1396         return rc;
1397 }
1398
1399 static int lmv_cleanup(struct obd_device *obd)
1400 {
1401         struct lmv_obd   *lmv = &obd->u.lmv;
1402         ENTRY;
1403
1404         fld_client_fini(&lmv->lmv_fld);
1405         if (lmv->tgts != NULL) {
1406                 int i;
1407                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1408                         if (lmv->tgts[i] == NULL)
1409                                 continue;
1410                         lmv_del_target(lmv, i);
1411                 }
1412                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1413                 lmv->tgts_size = 0;
1414         }
1415         RETURN(0);
1416 }
1417
1418 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1419 {
1420         struct lustre_cfg       *lcfg = buf;
1421         struct obd_uuid         obd_uuid;
1422         int                     gen;
1423         __u32                   index;
1424         int                     rc;
1425         ENTRY;
1426
1427         switch (lcfg->lcfg_command) {
1428         case LCFG_ADD_MDC:
1429                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1430                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1431                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1432                         GOTO(out, rc = -EINVAL);
1433
1434                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1435
1436                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1437                         GOTO(out, rc = -EINVAL);
1438                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1439                         GOTO(out, rc = -EINVAL);
1440                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1441                 GOTO(out, rc);
1442         default:
1443                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1444                 GOTO(out, rc = -EINVAL);
1445         }
1446 out:
1447         RETURN(rc);
1448 }
1449
1450 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1451                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1452 {
1453         struct obd_device       *obd = class_exp2obd(exp);
1454         struct lmv_obd          *lmv = &obd->u.lmv;
1455         struct obd_statfs       *temp;
1456         int                      rc = 0;
1457         __u32                    i;
1458         ENTRY;
1459
1460         rc = lmv_check_connect(obd);
1461         if (rc)
1462                 RETURN(rc);
1463
1464         OBD_ALLOC(temp, sizeof(*temp));
1465         if (temp == NULL)
1466                 RETURN(-ENOMEM);
1467
1468         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1469                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1470                         continue;
1471
1472                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1473                                 max_age, flags);
1474                 if (rc) {
1475                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1476                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1477                                rc);
1478                         GOTO(out_free_temp, rc);
1479                 }
1480
1481                 if (i == 0) {
1482                         *osfs = *temp;
1483                         /* If the statfs is from mount, it will needs
1484                          * retrieve necessary information from MDT0.
1485                          * i.e. mount does not need the merged osfs
1486                          * from all of MDT.
1487                          * And also clients can be mounted as long as
1488                          * MDT0 is in service*/
1489                         if (flags & OBD_STATFS_FOR_MDT0)
1490                                 GOTO(out_free_temp, rc);
1491                 } else {
1492                         osfs->os_bavail += temp->os_bavail;
1493                         osfs->os_blocks += temp->os_blocks;
1494                         osfs->os_ffree += temp->os_ffree;
1495                         osfs->os_files += temp->os_files;
1496                 }
1497         }
1498
1499         EXIT;
1500 out_free_temp:
1501         OBD_FREE(temp, sizeof(*temp));
1502         return rc;
1503 }
1504
1505 static int lmv_getstatus(struct obd_export *exp,
1506                          struct lu_fid *fid,
1507                          struct obd_capa **pc)
1508 {
1509         struct obd_device    *obd = exp->exp_obd;
1510         struct lmv_obd       *lmv = &obd->u.lmv;
1511         int                   rc;
1512         ENTRY;
1513
1514         rc = lmv_check_connect(obd);
1515         if (rc)
1516                 RETURN(rc);
1517
1518         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1519         RETURN(rc);
1520 }
1521
1522 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1523                         struct obd_capa *oc, obd_valid valid, const char *name,
1524                         const char *input, int input_size, int output_size,
1525                         int flags, struct ptlrpc_request **request)
1526 {
1527         struct obd_device      *obd = exp->exp_obd;
1528         struct lmv_obd         *lmv = &obd->u.lmv;
1529         struct lmv_tgt_desc    *tgt;
1530         int                     rc;
1531         ENTRY;
1532
1533         rc = lmv_check_connect(obd);
1534         if (rc)
1535                 RETURN(rc);
1536
1537         tgt = lmv_find_target(lmv, fid);
1538         if (IS_ERR(tgt))
1539                 RETURN(PTR_ERR(tgt));
1540
1541         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1542                          input_size, output_size, flags, request);
1543
1544         RETURN(rc);
1545 }
1546
1547 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1548                         struct obd_capa *oc, obd_valid valid, const char *name,
1549                         const char *input, int input_size, int output_size,
1550                         int flags, __u32 suppgid,
1551                         struct ptlrpc_request **request)
1552 {
1553         struct obd_device      *obd = exp->exp_obd;
1554         struct lmv_obd         *lmv = &obd->u.lmv;
1555         struct lmv_tgt_desc    *tgt;
1556         int                     rc;
1557         ENTRY;
1558
1559         rc = lmv_check_connect(obd);
1560         if (rc)
1561                 RETURN(rc);
1562
1563         tgt = lmv_find_target(lmv, fid);
1564         if (IS_ERR(tgt))
1565                 RETURN(PTR_ERR(tgt));
1566
1567         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1568                          input_size, output_size, flags, suppgid,
1569                          request);
1570
1571         RETURN(rc);
1572 }
1573
1574 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1575                        struct ptlrpc_request **request)
1576 {
1577         struct obd_device       *obd = exp->exp_obd;
1578         struct lmv_obd          *lmv = &obd->u.lmv;
1579         struct lmv_tgt_desc     *tgt;
1580         int                      rc;
1581         ENTRY;
1582
1583         rc = lmv_check_connect(obd);
1584         if (rc)
1585                 RETURN(rc);
1586
1587         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1588         if (IS_ERR(tgt))
1589                 RETURN(PTR_ERR(tgt));
1590
1591         if (op_data->op_flags & MF_GET_MDT_IDX) {
1592                 op_data->op_mds = tgt->ltd_idx;
1593                 RETURN(0);
1594         }
1595
1596         rc = md_getattr(tgt->ltd_exp, op_data, request);
1597
1598         RETURN(rc);
1599 }
1600
1601 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1602 {
1603         struct obd_device   *obd = exp->exp_obd;
1604         struct lmv_obd      *lmv = &obd->u.lmv;
1605         __u32                i;
1606         int                  rc;
1607         ENTRY;
1608
1609         rc = lmv_check_connect(obd);
1610         if (rc)
1611                 RETURN(rc);
1612
1613         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1614
1615         /*
1616          * With DNE every object can have two locks in different namespaces:
1617          * lookup lock in space of MDT storing direntry and update/open lock in
1618          * space of MDT storing inode.
1619          */
1620         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1621                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1622                         continue;
1623                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1624         }
1625
1626         RETURN(0);
1627 }
1628
1629 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1630                            ldlm_iterator_t it, void *data)
1631 {
1632         struct obd_device       *obd = exp->exp_obd;
1633         struct lmv_obd          *lmv = &obd->u.lmv;
1634         int                     i;
1635         int                     tgt;
1636         int                     rc;
1637         ENTRY;
1638
1639         rc = lmv_check_connect(obd);
1640         if (rc)
1641                 RETURN(rc);
1642
1643         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1644
1645         /*
1646          * With DNE every object can have two locks in different namespaces:
1647          * lookup lock in space of MDT storing direntry and update/open lock in
1648          * space of MDT storing inode.  Try the MDT that the FID maps to first,
1649          * since this can be easily found, and only try others if that fails.
1650          */
1651         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1652              i < lmv->desc.ld_tgt_count;
1653              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1654                 if (tgt < 0) {
1655                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1656                                obd->obd_name, PFID(fid), tgt);
1657                         tgt = 0;
1658                 }
1659
1660                 if (lmv->tgts[tgt] == NULL ||
1661                     lmv->tgts[tgt]->ltd_exp == NULL)
1662                         continue;
1663
1664                 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1665                 if (rc)
1666                         RETURN(rc);
1667         }
1668
1669         RETURN(rc);
1670 }
1671
1672
1673 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1674                      struct md_open_data *mod, struct ptlrpc_request **request)
1675 {
1676         struct obd_device     *obd = exp->exp_obd;
1677         struct lmv_obd        *lmv = &obd->u.lmv;
1678         struct lmv_tgt_desc   *tgt;
1679         int                    rc;
1680         ENTRY;
1681
1682         rc = lmv_check_connect(obd);
1683         if (rc)
1684                 RETURN(rc);
1685
1686         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1687         if (IS_ERR(tgt))
1688                 RETURN(PTR_ERR(tgt));
1689
1690         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1691         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1692         RETURN(rc);
1693 }
1694
1695 /**
1696  * Choosing the MDT by name or FID in @op_data.
1697  * For non-striped directory, it will locate MDT by fid.
1698  * For striped-directory, it will locate MDT by name. And also
1699  * it will reset op_fid1 with the FID of the choosen stripe.
1700  **/
1701 struct lmv_tgt_desc *
1702 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1703                            const char *name, int namelen, struct lu_fid *fid,
1704                            u32 *mds)
1705 {
1706         struct lmv_tgt_desc     *tgt;
1707         const struct lmv_oinfo  *oinfo;
1708
1709         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1710                 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1711         } else {
1712                 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1713                 if (IS_ERR(oinfo))
1714                         RETURN(ERR_CAST(oinfo));
1715         }
1716
1717         *fid = oinfo->lmo_fid;
1718         *mds = oinfo->lmo_mds;
1719         tgt = lmv_get_target(lmv, *mds, NULL);
1720
1721         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid));
1722         return tgt;
1723 }
1724
1725 /**
1726  * Locate mds by fid or name
1727  *
1728  * For striped directory (lsm != NULL), it will locate the stripe
1729  * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1730  * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1731  * walk through all of stripes to locate the entry.
1732  *
1733  * For normal direcotry, it will locate MDS by FID directly.
1734  * \param[in] lmv       LMV device
1735  * \param[in] op_data   client MD stack parameters, name, namelen
1736  *                      mds_num etc.
1737  * \param[in] fid       object FID used to locate MDS.
1738  *
1739  * retval               pointer to the lmv_tgt_desc if succeed.
1740  *                      ERR_PTR(errno) if failed.
1741  */
1742 struct lmv_tgt_desc*
1743 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1744                struct lu_fid *fid)
1745 {
1746         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1747         struct lmv_tgt_desc     *tgt;
1748
1749         /* During creating VOLATILE file, it should honor the mdt
1750          * index if the file under striped dir is being restored, see
1751          * ct_restore(). */
1752         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1753             (int)op_data->op_mds != -1 && lsm != NULL) {
1754                 int i;
1755                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1756                 if (IS_ERR(tgt))
1757                         return tgt;
1758
1759                 /* refill the right parent fid */
1760                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1761                         struct lmv_oinfo *oinfo;
1762
1763                         oinfo = &lsm->lsm_md_oinfo[i];
1764                         if (oinfo->lmo_mds == op_data->op_mds) {
1765                                 *fid = oinfo->lmo_fid;
1766                                 break;
1767                         }
1768                 }
1769
1770                 /* Hmm, can not find the stripe by mdt_index(op_mds) */
1771                 if (i == lsm->lsm_md_stripe_count)
1772                         tgt = ERR_PTR(-EINVAL);
1773
1774                 return tgt;
1775         }
1776
1777         if (lsm == NULL || op_data->op_namelen == 0) {
1778                 tgt = lmv_find_target(lmv, fid);
1779                 if (IS_ERR(tgt))
1780                         return tgt;
1781
1782                 op_data->op_mds = tgt->ltd_idx;
1783                 return tgt;
1784         }
1785
1786         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1787                                           op_data->op_namelen, fid,
1788                                           &op_data->op_mds);
1789 }
1790
1791 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1792                 const void *data, size_t datalen, umode_t mode, uid_t uid,
1793                 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1794                 struct ptlrpc_request **request)
1795 {
1796         struct obd_device       *obd = exp->exp_obd;
1797         struct lmv_obd          *lmv = &obd->u.lmv;
1798         struct lmv_tgt_desc     *tgt;
1799         int                      rc;
1800         ENTRY;
1801
1802         rc = lmv_check_connect(obd);
1803         if (rc)
1804                 RETURN(rc);
1805
1806         if (!lmv->desc.ld_active_tgt_count)
1807                 RETURN(-EIO);
1808
1809         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1810         if (IS_ERR(tgt))
1811                 RETURN(PTR_ERR(tgt));
1812
1813         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1814                 (int)op_data->op_namelen, op_data->op_name,
1815                 PFID(&op_data->op_fid1), op_data->op_mds);
1816
1817         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1818         if (rc)
1819                 RETURN(rc);
1820
1821         /* Send the create request to the MDT where the object
1822          * will be located */
1823         tgt = lmv_find_target(lmv, &op_data->op_fid2);
1824         if (IS_ERR(tgt))
1825                 RETURN(PTR_ERR(tgt));
1826
1827         op_data->op_mds = tgt->ltd_idx;
1828
1829         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1830                PFID(&op_data->op_fid2), op_data->op_mds);
1831
1832         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1833         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1834                        cap_effective, rdev, request);
1835         if (rc == 0) {
1836                 if (*request == NULL)
1837                         RETURN(rc);
1838                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1839         }
1840         RETURN(rc);
1841 }
1842
1843 static int lmv_done_writing(struct obd_export *exp,
1844                             struct md_op_data *op_data,
1845                             struct md_open_data *mod)
1846 {
1847         struct obd_device     *obd = exp->exp_obd;
1848         struct lmv_obd        *lmv = &obd->u.lmv;
1849         struct lmv_tgt_desc   *tgt;
1850         int                    rc;
1851         ENTRY;
1852
1853         rc = lmv_check_connect(obd);
1854         if (rc)
1855                 RETURN(rc);
1856
1857         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1858         if (IS_ERR(tgt))
1859                 RETURN(PTR_ERR(tgt));
1860
1861         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1862         RETURN(rc);
1863 }
1864
1865 static int
1866 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1867             const union ldlm_policy_data *policy,
1868             struct lookup_intent *it, struct md_op_data *op_data,
1869             struct lustre_handle *lockh, __u64 extra_lock_flags)
1870 {
1871         struct obd_device        *obd = exp->exp_obd;
1872         struct lmv_obd           *lmv = &obd->u.lmv;
1873         struct lmv_tgt_desc      *tgt;
1874         int                       rc;
1875         ENTRY;
1876
1877         rc = lmv_check_connect(obd);
1878         if (rc)
1879                 RETURN(rc);
1880
1881         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1882                LL_IT2STR(it), PFID(&op_data->op_fid1));
1883
1884         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1885         if (IS_ERR(tgt))
1886                 RETURN(PTR_ERR(tgt));
1887
1888         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1889                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1890
1891         rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1892                         extra_lock_flags);
1893
1894         RETURN(rc);
1895 }
1896
1897 static int
1898 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1899                  struct ptlrpc_request **preq)
1900 {
1901         struct ptlrpc_request   *req = NULL;
1902         struct obd_device       *obd = exp->exp_obd;
1903         struct lmv_obd          *lmv = &obd->u.lmv;
1904         struct lmv_tgt_desc     *tgt;
1905         struct mdt_body         *body;
1906         int                      rc;
1907         ENTRY;
1908
1909         rc = lmv_check_connect(obd);
1910         if (rc)
1911                 RETURN(rc);
1912
1913         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1914         if (IS_ERR(tgt))
1915                 RETURN(PTR_ERR(tgt));
1916
1917         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1918                 (int)op_data->op_namelen, op_data->op_name,
1919                 PFID(&op_data->op_fid1), tgt->ltd_idx);
1920
1921         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1922         if (rc != 0)
1923                 RETURN(rc);
1924
1925         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1926         LASSERT(body != NULL);
1927
1928         if (body->mbo_valid & OBD_MD_MDS) {
1929                 struct lu_fid rid = body->mbo_fid1;
1930                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1931                        PFID(&rid));
1932
1933                 tgt = lmv_find_target(lmv, &rid);
1934                 if (IS_ERR(tgt)) {
1935                         ptlrpc_req_finished(*preq);
1936                         preq = NULL;
1937                         RETURN(PTR_ERR(tgt));
1938                 }
1939
1940                 op_data->op_fid1 = rid;
1941                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1942                 op_data->op_namelen = 0;
1943                 op_data->op_name = NULL;
1944                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1945                 ptlrpc_req_finished(*preq);
1946                 *preq = req;
1947         }
1948
1949         RETURN(rc);
1950 }
1951
1952 #define md_op_data_fid(op_data, fl)                     \
1953         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1954          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1955          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1956          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1957          NULL)
1958
1959 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1960                             struct md_op_data *op_data,
1961                             __u32 op_tgt, ldlm_mode_t mode, int bits, int flag)
1962 {
1963         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1964         struct obd_device      *obd = exp->exp_obd;
1965         struct lmv_obd         *lmv = &obd->u.lmv;
1966         ldlm_policy_data_t      policy = {{ 0 }};
1967         int                     rc = 0;
1968         ENTRY;
1969
1970         if (!fid_is_sane(fid))
1971                 RETURN(0);
1972
1973         if (tgt == NULL) {
1974                 tgt = lmv_find_target(lmv, fid);
1975                 if (IS_ERR(tgt))
1976                         RETURN(PTR_ERR(tgt));
1977         }
1978
1979         if (tgt->ltd_idx != op_tgt) {
1980                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1981                 policy.l_inodebits.bits = bits;
1982                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1983                                       mode, LCF_ASYNC, NULL);
1984         } else {
1985                 CDEBUG(D_INODE,
1986                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1987                        op_tgt, PFID(fid));
1988                 op_data->op_flags |= flag;
1989                 rc = 0;
1990         }
1991
1992         RETURN(rc);
1993 }
1994
1995 /*
1996  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1997  * op_data->op_fid2
1998  */
1999 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2000                     struct ptlrpc_request **request)
2001 {
2002         struct obd_device       *obd = exp->exp_obd;
2003         struct lmv_obd          *lmv = &obd->u.lmv;
2004         struct lmv_tgt_desc     *tgt;
2005         int                      rc;
2006         ENTRY;
2007
2008         rc = lmv_check_connect(obd);
2009         if (rc)
2010                 RETURN(rc);
2011
2012         LASSERT(op_data->op_namelen != 0);
2013
2014         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2015                PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2016                op_data->op_name, PFID(&op_data->op_fid1));
2017
2018         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2019         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2020         op_data->op_cap = cfs_curproc_cap_pack();
2021         if (op_data->op_mea2 != NULL) {
2022                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
2023                 const struct lmv_oinfo  *oinfo;
2024
2025                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2026                                                 op_data->op_namelen);
2027                 if (IS_ERR(oinfo))
2028                         RETURN(PTR_ERR(oinfo));
2029
2030                 op_data->op_fid2 = oinfo->lmo_fid;
2031         }
2032
2033         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2034         if (IS_ERR(tgt))
2035                 RETURN(PTR_ERR(tgt));
2036
2037         /*
2038          * Cancel UPDATE lock on child (fid1).
2039          */
2040         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2041         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2042                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2043         if (rc != 0)
2044                 RETURN(rc);
2045
2046         rc = md_link(tgt->ltd_exp, op_data, request);
2047
2048         RETURN(rc);
2049 }
2050
2051 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2052                       const char *old, size_t oldlen,
2053                       const char *new, size_t newlen,
2054                       struct ptlrpc_request **request)
2055 {
2056         struct obd_device       *obd = exp->exp_obd;
2057         struct lmv_obd          *lmv = &obd->u.lmv;
2058         struct lmv_tgt_desc     *src_tgt;
2059         int                     rc;
2060         ENTRY;
2061
2062         LASSERT(oldlen != 0);
2063
2064         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2065                (int)oldlen, old, PFID(&op_data->op_fid1),
2066                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2067                (int)newlen, new, PFID(&op_data->op_fid2),
2068                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2069
2070         rc = lmv_check_connect(obd);
2071         if (rc)
2072                 RETURN(rc);
2073
2074         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2075         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2076         op_data->op_cap = cfs_curproc_cap_pack();
2077         if (op_data->op_cli_flags & CLI_MIGRATE) {
2078                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2079                          PFID(&op_data->op_fid3));
2080                 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2081                 if (rc)
2082                         RETURN(rc);
2083                 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3);
2084         } else {
2085                 if (op_data->op_mea1 != NULL) {
2086                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2087
2088                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2089                                                              oldlen,
2090                                                              &op_data->op_fid1,
2091                                                              &op_data->op_mds);
2092                         if (IS_ERR(src_tgt))
2093                                 RETURN(PTR_ERR(src_tgt));
2094                 } else {
2095                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2096                         if (IS_ERR(src_tgt))
2097                                 RETURN(PTR_ERR(src_tgt));
2098
2099                         op_data->op_mds = src_tgt->ltd_idx;
2100                 }
2101
2102                 if (op_data->op_mea2) {
2103                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
2104                         const struct lmv_oinfo  *oinfo;
2105
2106                         oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2107                         if (IS_ERR(oinfo))
2108                                 RETURN(PTR_ERR(oinfo));
2109
2110                         op_data->op_fid2 = oinfo->lmo_fid;
2111                 }
2112         }
2113         if (IS_ERR(src_tgt))
2114                 RETURN(PTR_ERR(src_tgt));
2115
2116         /*
2117          * LOOKUP lock on src child (fid3) should also be cancelled for
2118          * src_tgt in mdc_rename.
2119          */
2120         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2121
2122         /*
2123          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2124          * own target.
2125          */
2126         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2127                               LCK_EX, MDS_INODELOCK_UPDATE,
2128                               MF_MDC_CANCEL_FID2);
2129
2130         if (rc != 0)
2131                 RETURN(rc);
2132         /*
2133          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2134          */
2135         if (fid_is_sane(&op_data->op_fid3)) {
2136                 struct lmv_tgt_desc *tgt;
2137
2138                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2139                 if (IS_ERR(tgt))
2140                         RETURN(PTR_ERR(tgt));
2141
2142                 /* Cancel LOOKUP lock on its parent */
2143                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2144                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2145                                       MF_MDC_CANCEL_FID3);
2146                 if (rc != 0)
2147                         RETURN(rc);
2148
2149                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2150                                       LCK_EX, MDS_INODELOCK_FULL,
2151                                       MF_MDC_CANCEL_FID3);
2152                 if (rc != 0)
2153                         RETURN(rc);
2154         }
2155
2156         /*
2157          * Cancel all the locks on tgt child (fid4).
2158          */
2159         if (fid_is_sane(&op_data->op_fid4))
2160                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2161                                       LCK_EX, MDS_INODELOCK_FULL,
2162                                       MF_MDC_CANCEL_FID4);
2163
2164         CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2165                op_data->op_mds, PFID(&op_data->op_fid2));
2166
2167         rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2168                        request);
2169
2170         RETURN(rc);
2171 }
2172
2173 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2174                         void *ea, size_t ealen, void *ea2, size_t ea2len,
2175                         struct ptlrpc_request **request,
2176                         struct md_open_data **mod)
2177 {
2178         struct obd_device       *obd = exp->exp_obd;
2179         struct lmv_obd          *lmv = &obd->u.lmv;
2180         struct lmv_tgt_desc     *tgt;
2181         int                      rc = 0;
2182         ENTRY;
2183
2184         rc = lmv_check_connect(obd);
2185         if (rc)
2186                 RETURN(rc);
2187
2188         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2189                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2190
2191         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2192         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2193         if (IS_ERR(tgt))
2194                 RETURN(PTR_ERR(tgt));
2195
2196         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2197                         ea2len, request, mod);
2198
2199         RETURN(rc);
2200 }
2201
2202 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2203                      struct obd_capa *oc, struct ptlrpc_request **request)
2204 {
2205         struct obd_device       *obd = exp->exp_obd;
2206         struct lmv_obd          *lmv = &obd->u.lmv;
2207         struct lmv_tgt_desc     *tgt;
2208         int                      rc;
2209         ENTRY;
2210
2211         rc = lmv_check_connect(obd);
2212         if (rc != 0)
2213                 RETURN(rc);
2214
2215         tgt = lmv_find_target(lmv, fid);
2216         if (IS_ERR(tgt))
2217                 RETURN(PTR_ERR(tgt));
2218
2219         rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2220         RETURN(rc);
2221 }
2222
2223 /**
2224  * Get current minimum entry from striped directory
2225  *
2226  * This function will search the dir entry, whose hash value is the
2227  * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2228  * only being called for striped directory.
2229  *
2230  * \param[in] exp               export of LMV
2231  * \param[in] op_data           parameters transferred beween client MD stack
2232  *                              stripe_information will be included in this
2233  *                              parameter
2234  * \param[in] cb_op             ldlm callback being used in enqueue in
2235  *                              mdc_read_page
2236  * \param[in] hash_offset       the hash value, which is used to locate
2237  *                              minum(closet) dir entry
2238  * \param[in|out] stripe_offset the caller use this to indicate the stripe
2239  *                              index of last entry, so to avoid hash conflict
2240  *                              between stripes. It will also be used to
2241  *                              return the stripe index of current dir entry.
2242  * \param[in|out] entp          the minum entry and it also is being used
2243  *                              to input the last dir entry to resolve the
2244  *                              hash conflict
2245  *
2246  * \param[out] ppage            the page which holds the minum entry
2247  *
2248  * \retval                      = 0 get the entry successfully
2249  *                              negative errno (< 0) does not get the entry
2250  */
2251 static int lmv_get_min_striped_entry(struct obd_export *exp,
2252                                      struct md_op_data *op_data,
2253                                      struct md_callback *cb_op,
2254                                      __u64 hash_offset, int *stripe_offset,
2255                                      struct lu_dirent **entp,
2256                                      struct page **ppage)
2257 {
2258         struct obd_device       *obd = exp->exp_obd;
2259         struct lmv_obd          *lmv = &obd->u.lmv;
2260         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2261         struct lmv_tgt_desc     *tgt;
2262         int                     stripe_count;
2263         struct lu_dirent        *min_ent = NULL;
2264         struct page             *min_page = NULL;
2265         int                     min_idx = 0;
2266         int                     i;
2267         int                     rc = 0;
2268         ENTRY;
2269
2270         stripe_count = lsm->lsm_md_stripe_count;
2271         for (i = 0; i < stripe_count; i++) {
2272                 struct lu_dirent        *ent = NULL;
2273                 struct page             *page = NULL;
2274                 struct lu_dirpage       *dp;
2275                 __u64                   stripe_hash = hash_offset;
2276
2277                 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2278                 if (IS_ERR(tgt))
2279                         GOTO(out, rc = PTR_ERR(tgt));
2280
2281                 /* op_data will be shared by each stripe, so we need
2282                  * reset these value for each stripe */
2283                 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2284                 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2285                 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2286 next:
2287                 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2288                                   &page);
2289                 if (rc != 0)
2290                         GOTO(out, rc);
2291
2292                 dp = page_address(page);
2293                 for (ent = lu_dirent_start(dp); ent != NULL;
2294                      ent = lu_dirent_next(ent)) {
2295                         /* Skip dummy entry */
2296                         if (le16_to_cpu(ent->lde_namelen) == 0)
2297                                 continue;
2298
2299                         if (le64_to_cpu(ent->lde_hash) < hash_offset)
2300                                 continue;
2301
2302                         if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2303                             (*entp == ent || i < *stripe_offset))
2304                                 continue;
2305
2306                         /* skip . and .. for other stripes */
2307                         if (i != 0 &&
2308                             (strncmp(ent->lde_name, ".",
2309                                      le16_to_cpu(ent->lde_namelen)) == 0 ||
2310                              strncmp(ent->lde_name, "..",
2311                                      le16_to_cpu(ent->lde_namelen)) == 0))
2312                                 continue;
2313                         break;
2314                 }
2315
2316                 if (ent == NULL) {
2317                         stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2318
2319                         kunmap(page);
2320                         page_cache_release(page);
2321                         page = NULL;
2322
2323                         /* reach the end of current stripe, go to next stripe */
2324                         if (stripe_hash == MDS_DIR_END_OFF)
2325                                 continue;
2326                         else
2327                                 goto next;
2328                 }
2329
2330                 if (min_ent != NULL) {
2331                         if (le64_to_cpu(min_ent->lde_hash) >
2332                             le64_to_cpu(ent->lde_hash)) {
2333                                 min_ent = ent;
2334                                 kunmap(min_page);
2335                                 page_cache_release(min_page);
2336                                 min_idx = i;
2337                                 min_page = page;
2338                         } else {
2339                                 kunmap(page);
2340                                 page_cache_release(page);
2341                                 page = NULL;
2342                         }
2343                 } else {
2344                         min_ent = ent;
2345                         min_page = page;
2346                         min_idx = i;
2347                 }
2348         }
2349
2350 out:
2351         if (*ppage != NULL) {
2352                 kunmap(*ppage);
2353                 page_cache_release(*ppage);
2354         }
2355         *stripe_offset = min_idx;
2356         *entp = min_ent;
2357         *ppage = min_page;
2358         RETURN(rc);
2359 }
2360
2361 /**
2362  * Build dir entry page from a striped directory
2363  *
2364  * This function gets one entry by @offset from a striped directory. It will
2365  * read entries from all of stripes, and choose one closest to the required
2366  * offset(&offset). A few notes
2367  * 1. skip . and .. for non-zero stripes, because there can only have one .
2368  * and .. in a directory.
2369  * 2. op_data will be shared by all of stripes, instead of allocating new
2370  * one, so need to restore before reusing.
2371  * 3. release the entry page if that is not being chosen.
2372  *
2373  * \param[in] exp       obd export refer to LMV
2374  * \param[in] op_data   hold those MD parameters of read_entry
2375  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2376  * \param[out] ldp      the entry being read
2377  * \param[out] ppage    the page holding the entry. Note: because the entry
2378  *                      will be accessed in upper layer, so we need hold the
2379  *                      page until the usages of entry is finished, see
2380  *                      ll_dir_entry_next.
2381  *
2382  * retval               =0 if get entry successfully
2383  *                      <0 cannot get entry
2384  */
2385 static int lmv_read_striped_page(struct obd_export *exp,
2386                                  struct md_op_data *op_data,
2387                                  struct md_callback *cb_op,
2388                                  __u64 offset, struct page **ppage)
2389 {
2390         struct obd_device       *obd = exp->exp_obd;
2391         struct lu_fid           master_fid = op_data->op_fid1;
2392         struct inode            *master_inode = op_data->op_data;
2393         __u64                   hash_offset = offset;
2394         struct lu_dirpage       *dp;
2395         struct page             *min_ent_page = NULL;
2396         struct page             *ent_page = NULL;
2397         struct lu_dirent        *ent;
2398         void                    *area;
2399         int                     ent_idx = 0;
2400         struct lu_dirent        *min_ent = NULL;
2401         struct lu_dirent        *last_ent;
2402         size_t                  left_bytes;
2403         int                     rc;
2404         ENTRY;
2405
2406         rc = lmv_check_connect(obd);
2407         if (rc)
2408                 RETURN(rc);
2409
2410         /* Allocate a page and read entries from all of stripes and fill
2411          * the page by hash order */
2412         ent_page = alloc_page(GFP_KERNEL);
2413         if (ent_page == NULL)
2414                 RETURN(-ENOMEM);
2415
2416         /* Initialize the entry page */
2417         dp = kmap(ent_page);
2418         memset(dp, 0, sizeof(*dp));
2419         dp->ldp_hash_start = cpu_to_le64(offset);
2420         dp->ldp_flags |= LDF_COLLIDE;
2421
2422         area = dp + 1;
2423         left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2424         ent = area;
2425         last_ent = ent;
2426         do {
2427                 __u16   ent_size;
2428
2429                 /* Find the minum entry from all sub-stripes */
2430                 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2431                                                &ent_idx, &min_ent,
2432                                                &min_ent_page);
2433                 if (rc != 0)
2434                         GOTO(out, rc);
2435
2436                 /* If it can not get minum entry, it means it already reaches
2437                  * the end of this directory */
2438                 if (min_ent == NULL) {
2439                         last_ent->lde_reclen = 0;
2440                         hash_offset = MDS_DIR_END_OFF;
2441                         GOTO(out, rc);
2442                 }
2443
2444                 ent_size = le16_to_cpu(min_ent->lde_reclen);
2445
2446                 /* the last entry lde_reclen is 0, but it might not
2447                  * the end of this entry of this temporay entry */
2448                 if (ent_size == 0)
2449                         ent_size = lu_dirent_calc_size(
2450                                         le16_to_cpu(min_ent->lde_namelen),
2451                                         le32_to_cpu(min_ent->lde_attrs));
2452                 if (ent_size > left_bytes) {
2453                         last_ent->lde_reclen = cpu_to_le16(0);
2454                         hash_offset = le64_to_cpu(min_ent->lde_hash);
2455                         GOTO(out, rc);
2456                 }
2457
2458                 memcpy(ent, min_ent, ent_size);
2459
2460                 /* Replace . with master FID and Replace .. with the parent FID
2461                  * of master object */
2462                 if (strncmp(ent->lde_name, ".",
2463                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2464                     le16_to_cpu(ent->lde_namelen) == 1)
2465                         fid_cpu_to_le(&ent->lde_fid, &master_fid);
2466                 else if (strncmp(ent->lde_name, "..",
2467                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2468                            le16_to_cpu(ent->lde_namelen) == 2)
2469                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2470
2471                 left_bytes -= ent_size;
2472                 ent->lde_reclen = cpu_to_le16(ent_size);
2473                 last_ent = ent;
2474                 ent = (void *)ent + ent_size;
2475                 hash_offset = le64_to_cpu(min_ent->lde_hash);
2476                 if (hash_offset == MDS_DIR_END_OFF) {
2477                         last_ent->lde_reclen = 0;
2478                         break;
2479                 }
2480         } while (1);
2481 out:
2482         if (min_ent_page != NULL) {
2483                 kunmap(min_ent_page);
2484                 page_cache_release(min_ent_page);
2485         }
2486
2487         if (unlikely(rc != 0)) {
2488                 __free_page(ent_page);
2489                 ent_page = NULL;
2490         } else {
2491                 if (ent == area)
2492                         dp->ldp_flags |= LDF_EMPTY;
2493                 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2494                 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2495         }
2496
2497         /* We do not want to allocate md_op_data during each
2498          * dir entry reading, so op_data will be shared by every stripe,
2499          * then we need to restore it back to original value before
2500          * return to the upper layer */
2501         op_data->op_fid1 = master_fid;
2502         op_data->op_fid2 = master_fid;
2503         op_data->op_data = master_inode;
2504
2505         *ppage = ent_page;
2506
2507         RETURN(rc);
2508 }
2509
2510 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2511                   struct md_callback *cb_op, __u64 offset,
2512                   struct page **ppage)
2513 {
2514         struct obd_device       *obd = exp->exp_obd;
2515         struct lmv_obd          *lmv = &obd->u.lmv;
2516         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2517         struct lmv_tgt_desc     *tgt;
2518         int                     rc;
2519         ENTRY;
2520
2521         rc = lmv_check_connect(obd);
2522         if (rc != 0)
2523                 RETURN(rc);
2524
2525         if (unlikely(lsm != NULL)) {
2526                 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2527                 RETURN(rc);
2528         }
2529
2530         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2531         if (IS_ERR(tgt))
2532                 RETURN(PTR_ERR(tgt));
2533
2534         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2535
2536         RETURN(rc);
2537 }
2538
2539 /**
2540  * Unlink a file/directory
2541  *
2542  * Unlink a file or directory under the parent dir. The unlink request
2543  * usually will be sent to the MDT where the child is located, but if
2544  * the client does not have the child FID then request will be sent to the
2545  * MDT where the parent is located.
2546  *
2547  * If the parent is a striped directory then it also needs to locate which
2548  * stripe the name of the child is located, and replace the parent FID
2549  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2550  * it will walk through all of sub-stripes until the child is being
2551  * unlinked finally.
2552  *
2553  * \param[in] exp       export refer to LMV
2554  * \param[in] op_data   different parameters transferred beween client
2555  *                      MD stacks, name, namelen, FIDs etc.
2556  *                      op_fid1 is the parent FID, op_fid2 is the child
2557  *                      FID.
2558  * \param[out] request  point to the request of unlink.
2559  *
2560  * retval               0 if succeed
2561  *                      negative errno if failed.
2562  */
2563 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2564                       struct ptlrpc_request **request)
2565 {
2566         struct obd_device       *obd = exp->exp_obd;
2567         struct lmv_obd          *lmv = &obd->u.lmv;
2568         struct lmv_tgt_desc     *tgt = NULL;
2569         struct lmv_tgt_desc     *parent_tgt = NULL;
2570         struct mdt_body         *body;
2571         int                     rc;
2572         int                     stripe_index = 0;
2573         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2574         ENTRY;
2575
2576         rc = lmv_check_connect(obd);
2577         if (rc)
2578                 RETURN(rc);
2579 retry_unlink:
2580         /* For striped dir, we need to locate the parent as well */
2581         if (lsm != NULL) {
2582                 struct lmv_tgt_desc *tmp;
2583
2584                 LASSERT(op_data->op_name != NULL &&
2585                         op_data->op_namelen != 0);
2586
2587                 tmp = lmv_locate_target_for_name(lmv, lsm,
2588                                                  op_data->op_name,
2589                                                  op_data->op_namelen,
2590                                                  &op_data->op_fid1,
2591                                                  &op_data->op_mds);
2592
2593                 /* return -EBADFD means unknown hash type, might
2594                  * need try all sub-stripe here */
2595                 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2596                         RETURN(PTR_ERR(tmp));
2597
2598                 /* Note: both migrating dir and unknown hash dir need to
2599                  * try all of sub-stripes, so we need start search the
2600                  * name from stripe 0, but migrating dir is already handled
2601                  * inside lmv_locate_target_for_name(), so we only check
2602                  * unknown hash type directory here */
2603                 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2604                         struct lmv_oinfo *oinfo;
2605
2606                         oinfo = &lsm->lsm_md_oinfo[stripe_index];
2607
2608                         op_data->op_fid1 = oinfo->lmo_fid;
2609                         op_data->op_mds = oinfo->lmo_mds;
2610                 }
2611         }
2612
2613 try_next_stripe:
2614         /* Send unlink requests to the MDT where the child is located */
2615         if (likely(!fid_is_zero(&op_data->op_fid2)))
2616                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2617         else if (lsm != NULL)
2618                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2619         else
2620                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2621
2622         if (IS_ERR(tgt))
2623                 RETURN(PTR_ERR(tgt));
2624
2625         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2626         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2627         op_data->op_cap = cfs_curproc_cap_pack();
2628
2629         /*
2630          * If child's fid is given, cancel unused locks for it if it is from
2631          * another export than parent.
2632          *
2633          * LOOKUP lock for child (fid3) should also be cancelled on parent
2634          * tgt_tgt in mdc_unlink().
2635          */
2636         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2637
2638         /*
2639          * Cancel FULL locks on child (fid3).
2640          */
2641         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2642         if (IS_ERR(parent_tgt))
2643                 RETURN(PTR_ERR(parent_tgt));
2644
2645         if (parent_tgt != tgt) {
2646                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2647                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2648                                       MF_MDC_CANCEL_FID3);
2649         }
2650
2651         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2652                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2653         if (rc != 0)
2654                 RETURN(rc);
2655
2656         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2657                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2658
2659         rc = md_unlink(tgt->ltd_exp, op_data, request);
2660         if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2661                 RETURN(rc);
2662
2663         /* Try next stripe if it is needed. */
2664         if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2665                 struct lmv_oinfo *oinfo;
2666
2667                 stripe_index++;
2668                 if (stripe_index >= lsm->lsm_md_stripe_count)
2669                         RETURN(rc);
2670
2671                 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2672
2673                 op_data->op_fid1 = oinfo->lmo_fid;
2674                 op_data->op_mds = oinfo->lmo_mds;
2675
2676                 ptlrpc_req_finished(*request);
2677                 *request = NULL;
2678
2679                 goto try_next_stripe;
2680         }
2681
2682         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2683         if (body == NULL)
2684                 RETURN(-EPROTO);
2685
2686         /* Not cross-ref case, just get out of here. */
2687         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2688                 RETURN(rc);
2689
2690         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2691                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2692
2693         /* This is a remote object, try remote MDT, Note: it may
2694          * try more than 1 time here, Considering following case
2695          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2696          * 1. Initially A does not know where remote1 is, it send
2697          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2698          *    resend unlink RPC to MDT1 (retry 1st time).
2699          *
2700          * 2. During the unlink RPC in flight,
2701          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2702          *    and create new remote1, but on MDT0
2703          *
2704          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2705          *    /mnt/lustre, then lookup get fid of remote1, and find
2706          *    it is remote dir again, and replay -EREMOTE again.
2707          *
2708          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2709          *
2710          * In theory, it might try unlimited time here, but it should
2711          * be very rare case.  */
2712         op_data->op_fid2 = body->mbo_fid1;
2713         ptlrpc_req_finished(*request);
2714         *request = NULL;
2715
2716         goto retry_unlink;
2717 }
2718
2719 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2720 {
2721         struct lmv_obd *lmv = &obd->u.lmv;
2722         int rc = 0;
2723
2724         switch (stage) {
2725         case OBD_CLEANUP_EARLY:
2726                 /* XXX: here should be calling obd_precleanup() down to
2727                  * stack. */
2728                 break;
2729         case OBD_CLEANUP_EXPORTS:
2730                 fld_client_proc_fini(&lmv->lmv_fld);
2731                 lprocfs_obd_cleanup(obd);
2732                 lprocfs_free_md_stats(obd);
2733                 break;
2734         default:
2735                 break;
2736         }
2737         RETURN(rc);
2738 }
2739
2740 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2741                         __u32 keylen, void *key, __u32 *vallen, void *val,
2742                         struct lov_stripe_md *lsm)
2743 {
2744         struct obd_device       *obd;
2745         struct lmv_obd          *lmv;
2746         int                      rc = 0;
2747         ENTRY;
2748
2749         obd = class_exp2obd(exp);
2750         if (obd == NULL) {
2751                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2752                        exp->exp_handle.h_cookie);
2753                 RETURN(-EINVAL);
2754         }
2755
2756         lmv = &obd->u.lmv;
2757         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2758                 int i;
2759
2760                 rc = lmv_check_connect(obd);
2761                 if (rc)
2762                         RETURN(rc);
2763
2764                 LASSERT(*vallen == sizeof(__u32));
2765                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2766                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2767                         /*
2768                          * All tgts should be connected when this gets called.
2769                          */
2770                         if (tgt == NULL || tgt->ltd_exp == NULL)
2771                                 continue;
2772
2773                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2774                                           vallen, val, NULL))
2775                                 RETURN(0);
2776                 }
2777                 RETURN(-EINVAL);
2778         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2779                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2780                    KEY_IS(KEY_MAX_COOKIESIZE) ||
2781                    KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2782                    KEY_IS(KEY_CONN_DATA)) {
2783                 rc = lmv_check_connect(obd);
2784                 if (rc)
2785                         RETURN(rc);
2786
2787                 /*
2788                  * Forwarding this request to first MDS, it should know LOV
2789                  * desc.
2790                  */
2791                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2792                                   vallen, val, NULL);
2793                 if (!rc && KEY_IS(KEY_CONN_DATA))
2794                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2795                 RETURN(rc);
2796         } else if (KEY_IS(KEY_TGT_COUNT)) {
2797                 *((int *)val) = lmv->desc.ld_tgt_count;
2798                 RETURN(0);
2799         }
2800
2801         CDEBUG(D_IOCTL, "Invalid key\n");
2802         RETURN(-EINVAL);
2803 }
2804
2805 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2806                        obd_count keylen, void *key, obd_count vallen,
2807                        void *val, struct ptlrpc_request_set *set)
2808 {
2809         struct lmv_tgt_desc    *tgt = NULL;
2810         struct obd_device      *obd;
2811         struct lmv_obd         *lmv;
2812         int rc = 0;
2813         ENTRY;
2814
2815         obd = class_exp2obd(exp);
2816         if (obd == NULL) {
2817                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2818                        exp->exp_handle.h_cookie);
2819                 RETURN(-EINVAL);
2820         }
2821         lmv = &obd->u.lmv;
2822
2823         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2824                 int i, err = 0;
2825
2826                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2827                         tgt = lmv->tgts[i];
2828
2829                         if (tgt == NULL || tgt->ltd_exp == NULL)
2830                                 continue;
2831
2832                         err = obd_set_info_async(env, tgt->ltd_exp,
2833                                                  keylen, key, vallen, val, set);
2834                         if (err && rc == 0)
2835                                 rc = err;
2836                 }
2837
2838                 RETURN(rc);
2839         }
2840
2841         RETURN(-EINVAL);
2842 }
2843
2844 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2845                           struct lmv_mds_md_v1 *lmm1)
2846 {
2847         int     cplen;
2848         int     i;
2849
2850         lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2851         lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2852         lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2853         lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2854         cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2855                         sizeof(lmm1->lmv_pool_name));
2856         if (cplen >= sizeof(lmm1->lmv_pool_name))
2857                 return -E2BIG;
2858
2859         for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2860                 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2861                               &lsm->lsm_md_oinfo[i].lmo_fid);
2862         return 0;
2863 }
2864
2865 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2866                 int stripe_count)
2867 {
2868         int     lmm_size = 0;
2869         bool    allocated = false;
2870         int     rc = 0;
2871         ENTRY;
2872
2873         LASSERT(lmmp != NULL);
2874         /* Free lmm */
2875         if (*lmmp != NULL && lsm == NULL) {
2876                 int stripe_count;
2877
2878                 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2879                 lmm_size = lmv_mds_md_size(stripe_count,
2880                                            le32_to_cpu((*lmmp)->lmv_magic));
2881                 if (lmm_size == 0)
2882                         RETURN(-EINVAL);
2883                 OBD_FREE(*lmmp, lmm_size);
2884                 *lmmp = NULL;
2885                 RETURN(0);
2886         }
2887
2888         /* Alloc lmm */
2889         if (*lmmp == NULL && lsm == NULL) {
2890                 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2891                 LASSERT(lmm_size > 0);
2892                 OBD_ALLOC(*lmmp, lmm_size);
2893                 if (*lmmp == NULL)
2894                         RETURN(-ENOMEM);
2895                 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2896                 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2897                 RETURN(lmm_size);
2898         }
2899
2900         /* pack lmm */
2901         LASSERT(lsm != NULL);
2902         lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2903         if (*lmmp == NULL) {
2904                 OBD_ALLOC(*lmmp, lmm_size);
2905                 if (*lmmp == NULL)
2906                         RETURN(-ENOMEM);
2907                 allocated = true;
2908         }
2909
2910         switch (lsm->lsm_md_magic) {
2911         case LMV_MAGIC_V1:
2912                 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2913                 break;
2914         default:
2915                 rc = -EINVAL;
2916                 break;
2917         }
2918
2919         if (rc != 0 && allocated) {
2920                 OBD_FREE(*lmmp, lmm_size);
2921                 *lmmp = NULL;
2922         }
2923
2924         RETURN(lmm_size);
2925 }
2926 EXPORT_SYMBOL(lmv_pack_md);
2927
2928 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2929                             const struct lmv_mds_md_v1 *lmm1)
2930 {
2931         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
2932         int             stripe_count;
2933         int             cplen;
2934         int             i;
2935         int             rc = 0;
2936         ENTRY;
2937
2938         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2939         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2940         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2941         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2942                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2943         else
2944                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2945         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2946         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2947                         sizeof(lsm->lsm_md_pool_name));
2948
2949         if (cplen >= sizeof(lsm->lsm_md_pool_name))
2950                 RETURN(-E2BIG);
2951
2952         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2953                "layout_version %d\n", lsm->lsm_md_stripe_count,
2954                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2955                lsm->lsm_md_layout_version);
2956
2957         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2958         for (i = 0; i < stripe_count; i++) {
2959                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2960                               &lmm1->lmv_stripe_fids[i]);
2961                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2962                                     &lsm->lsm_md_oinfo[i].lmo_mds);
2963                 if (rc != 0)
2964                         RETURN(rc);
2965                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2966                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2967         }
2968
2969         RETURN(rc);
2970 }
2971
2972 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2973                   const union lmv_mds_md *lmm, int stripe_count)
2974 {
2975         struct lmv_stripe_md     *lsm;
2976         int                      lsm_size;
2977         int                      rc;
2978         bool                     allocated = false;
2979         ENTRY;
2980
2981         LASSERT(lsmp != NULL);
2982
2983         lsm = *lsmp;
2984         /* Free memmd */
2985         if (lsm != NULL && lmm == NULL) {
2986                 int i;
2987                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2988                         /* For migrating inode, the master stripe and master
2989                          * object will be the same, so do not need iput, see
2990                          * ll_update_lsm_md */
2991                         if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2992                               i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2993                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
2994                 }
2995                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2996                 OBD_FREE(lsm, lsm_size);
2997                 *lsmp = NULL;
2998                 RETURN(0);
2999         }
3000
3001         /* Alloc memmd */
3002         if (lsm == NULL && lmm == NULL) {
3003                 lsm_size = lmv_stripe_md_size(stripe_count);
3004                 OBD_ALLOC(lsm, lsm_size);
3005                 if (lsm == NULL)
3006                         RETURN(-ENOMEM);
3007                 lsm->lsm_md_stripe_count = stripe_count;
3008                 *lsmp = lsm;
3009                 RETURN(0);
3010         }
3011
3012         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3013                 RETURN(-EPERM);
3014
3015         /* Unpack memmd */
3016         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3017             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3018                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3019                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3020                        -EIO);
3021                 RETURN(-EIO);
3022         }
3023
3024         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3025                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3026         else
3027                 /**
3028                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3029                  * stripecount should be 0 then.
3030                  */
3031                 lsm_size = lmv_stripe_md_size(0);
3032
3033         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3034         if (lsm == NULL) {
3035                 OBD_ALLOC(lsm, lsm_size);
3036                 if (lsm == NULL)
3037                         RETURN(-ENOMEM);
3038                 allocated = true;
3039                 *lsmp = lsm;
3040         }
3041
3042         switch (le32_to_cpu(lmm->lmv_magic)) {
3043         case LMV_MAGIC_V1:
3044                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3045                 break;
3046         default:
3047                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3048                        le32_to_cpu(lmm->lmv_magic));
3049                 rc = -EINVAL;
3050                 break;
3051         }
3052
3053         if (rc != 0 && allocated) {
3054                 OBD_FREE(lsm, lsm_size);
3055                 *lsmp = NULL;
3056                 lsm_size = rc;
3057         }
3058         RETURN(lsm_size);
3059 }
3060
3061 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3062 {
3063         return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3064 }
3065 EXPORT_SYMBOL(lmv_alloc_memmd);
3066
3067 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3068 {
3069         lmv_unpack_md(NULL, &lsm, NULL, 0);
3070 }
3071 EXPORT_SYMBOL(lmv_free_memmd);
3072
3073 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3074                  struct lov_mds_md *lmm, int disk_len)
3075 {
3076         return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3077                              (union lmv_mds_md *)lmm, disk_len);
3078 }
3079
3080 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3081                struct lov_stripe_md *lsm)
3082 {
3083         struct obd_device               *obd = exp->exp_obd;
3084         struct lmv_obd                  *lmv_obd = &obd->u.lmv;
3085         const struct lmv_stripe_md      *lmv = (struct lmv_stripe_md *)lsm;
3086         int                             stripe_count;
3087
3088         if (lmmp == NULL) {
3089                 if (lsm != NULL)
3090                         stripe_count = lmv->lsm_md_stripe_count;
3091                 else
3092                         stripe_count = lmv_obd->desc.ld_tgt_count;
3093
3094                 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3095         }
3096
3097         return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3098 }
3099
3100 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3101                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
3102                              ldlm_cancel_flags_t flags, void *opaque)
3103 {
3104         struct obd_device       *obd = exp->exp_obd;
3105         struct lmv_obd          *lmv = &obd->u.lmv;
3106         int                      rc = 0;
3107         int                      err;
3108         __u32                    i;
3109         ENTRY;
3110
3111         LASSERT(fid != NULL);
3112
3113         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3114                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3115
3116                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3117                         continue;
3118
3119                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3120                                        opaque);
3121                 if (!rc)
3122                         rc = err;
3123         }
3124         RETURN(rc);
3125 }
3126
3127 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3128                       __u64 *bits)
3129 {
3130         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3131         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3132         int                      rc;
3133         ENTRY;
3134
3135         if (tgt == NULL || tgt->ltd_exp == NULL)
3136                 RETURN(-EINVAL);
3137         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3138         RETURN(rc);
3139 }
3140
3141 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3142                            const struct lu_fid *fid, ldlm_type_t type,
3143                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
3144                            struct lustre_handle *lockh)
3145 {
3146         struct obd_device       *obd = exp->exp_obd;
3147         struct lmv_obd          *lmv = &obd->u.lmv;
3148         ldlm_mode_t             rc;
3149         int                     tgt;
3150         int                     i;
3151         ENTRY;
3152
3153         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3154
3155         /*
3156          * With DNE every object can have two locks in different namespaces:
3157          * lookup lock in space of MDT storing direntry and update/open lock in
3158          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3159          * since this can be easily found, and only try others if that fails.
3160          */
3161         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3162              i < lmv->desc.ld_tgt_count;
3163              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3164                 if (tgt < 0) {
3165                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3166                                obd->obd_name, PFID(fid), tgt);
3167                         tgt = 0;
3168                 }
3169
3170                 if (lmv->tgts[tgt] == NULL ||
3171                     lmv->tgts[tgt]->ltd_exp == NULL ||
3172                     lmv->tgts[tgt]->ltd_active == 0)
3173                         continue;
3174
3175                 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3176                                    type, policy, mode, lockh);
3177                 if (rc)
3178                         RETURN(rc);
3179         }
3180
3181         RETURN(0);
3182 }
3183
3184 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3185                       struct obd_export *dt_exp, struct obd_export *md_exp,
3186                       struct lustre_md *md)
3187 {
3188         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3189         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3190
3191         if (tgt == NULL || tgt->ltd_exp == NULL)
3192                 RETURN(-EINVAL);
3193
3194         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3195 }
3196
3197 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3198 {
3199         struct obd_device       *obd = exp->exp_obd;
3200         struct lmv_obd          *lmv = &obd->u.lmv;
3201         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3202         ENTRY;
3203
3204         if (md->lmv != NULL) {
3205                 lmv_free_memmd(md->lmv);
3206                 md->lmv = NULL;
3207         }
3208         if (tgt == NULL || tgt->ltd_exp == NULL)
3209                 RETURN(-EINVAL);
3210         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3211 }
3212
3213 int lmv_set_open_replay_data(struct obd_export *exp,
3214                              struct obd_client_handle *och,
3215                              struct lookup_intent *it)
3216 {
3217         struct obd_device       *obd = exp->exp_obd;
3218         struct lmv_obd          *lmv = &obd->u.lmv;
3219         struct lmv_tgt_desc     *tgt;
3220         ENTRY;
3221
3222         tgt = lmv_find_target(lmv, &och->och_fid);
3223         if (IS_ERR(tgt))
3224                 RETURN(PTR_ERR(tgt));
3225
3226         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3227 }
3228
3229 int lmv_clear_open_replay_data(struct obd_export *exp,
3230                                struct obd_client_handle *och)
3231 {
3232         struct obd_device       *obd = exp->exp_obd;
3233         struct lmv_obd          *lmv = &obd->u.lmv;
3234         struct lmv_tgt_desc     *tgt;
3235         ENTRY;
3236
3237         tgt = lmv_find_target(lmv, &och->och_fid);
3238         if (IS_ERR(tgt))
3239                 RETURN(PTR_ERR(tgt));
3240
3241         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3242 }
3243
3244 static int lmv_get_remote_perm(struct obd_export *exp,
3245                                const struct lu_fid *fid,
3246                                struct obd_capa *oc, __u32 suppgid,
3247                                struct ptlrpc_request **request)
3248 {
3249         struct obd_device       *obd = exp->exp_obd;
3250         struct lmv_obd          *lmv = &obd->u.lmv;
3251         struct lmv_tgt_desc     *tgt;
3252         int                      rc;
3253         ENTRY;
3254
3255         rc = lmv_check_connect(obd);
3256         if (rc)
3257                 RETURN(rc);
3258
3259         tgt = lmv_find_target(lmv, fid);
3260         if (IS_ERR(tgt))
3261                 RETURN(PTR_ERR(tgt));
3262
3263         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3264         RETURN(rc);
3265 }
3266
3267 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3268                           renew_capa_cb_t cb)
3269 {
3270         struct obd_device       *obd = exp->exp_obd;
3271         struct lmv_obd          *lmv = &obd->u.lmv;
3272         struct lmv_tgt_desc     *tgt;
3273         int                      rc;
3274         ENTRY;
3275
3276         rc = lmv_check_connect(obd);
3277         if (rc)
3278                 RETURN(rc);
3279
3280         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3281         if (IS_ERR(tgt))
3282                 RETURN(PTR_ERR(tgt));
3283
3284         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3285         RETURN(rc);
3286 }
3287
3288 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3289                     const struct req_msg_field *field, struct obd_capa **oc)
3290 {
3291         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3292         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3293
3294         if (tgt == NULL || tgt->ltd_exp == NULL)
3295                 RETURN(-EINVAL);
3296         return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3297 }
3298
3299 int lmv_intent_getattr_async(struct obd_export *exp,
3300                              struct md_enqueue_info *minfo,
3301                              struct ldlm_enqueue_info *einfo)
3302 {
3303         struct md_op_data       *op_data = &minfo->mi_data;
3304         struct obd_device       *obd = exp->exp_obd;
3305         struct lmv_obd          *lmv = &obd->u.lmv;
3306         struct lmv_tgt_desc     *tgt = NULL;
3307         int                      rc;
3308         ENTRY;
3309
3310         rc = lmv_check_connect(obd);
3311         if (rc)
3312                 RETURN(rc);
3313
3314         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3315         if (IS_ERR(tgt))
3316                 RETURN(PTR_ERR(tgt));
3317
3318         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3319         RETURN(rc);
3320 }
3321
3322 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3323                         struct lu_fid *fid, __u64 *bits)
3324 {
3325         struct obd_device       *obd = exp->exp_obd;
3326         struct lmv_obd          *lmv = &obd->u.lmv;
3327         struct lmv_tgt_desc     *tgt;
3328         int                      rc;
3329         ENTRY;
3330
3331         rc = lmv_check_connect(obd);
3332         if (rc)
3333                 RETURN(rc);
3334
3335         tgt = lmv_find_target(lmv, fid);
3336         if (IS_ERR(tgt))
3337                 RETURN(PTR_ERR(tgt));
3338
3339         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3340         RETURN(rc);
3341 }
3342
3343 int lmv_get_fid_from_lsm(struct obd_export *exp,
3344                          const struct lmv_stripe_md *lsm,
3345                          const char *name, int namelen, struct lu_fid *fid)
3346 {
3347         const struct lmv_oinfo *oinfo;
3348
3349         LASSERT(lsm != NULL);
3350         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3351         if (IS_ERR(oinfo))
3352                 return PTR_ERR(oinfo);
3353
3354         *fid = oinfo->lmo_fid;
3355
3356         RETURN(0);
3357 }
3358
3359 /**
3360  * For lmv, only need to send request to master MDT, and the master MDT will
3361  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3362  * we directly fetch data from the slave MDTs.
3363  */
3364 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3365                  struct obd_quotactl *oqctl)
3366 {
3367         struct obd_device   *obd = class_exp2obd(exp);
3368         struct lmv_obd      *lmv = &obd->u.lmv;
3369         struct lmv_tgt_desc *tgt = lmv->tgts[0];
3370         int                  rc = 0;
3371         __u32                i;
3372         __u64                curspace, curinodes;
3373         ENTRY;
3374
3375         if (tgt == NULL ||
3376             tgt->ltd_exp == NULL ||
3377             !tgt->ltd_active ||
3378             lmv->desc.ld_tgt_count == 0) {
3379                 CERROR("master lmv inactive\n");
3380                 RETURN(-EIO);
3381         }
3382
3383         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3384                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3385                 RETURN(rc);
3386         }
3387
3388         curspace = curinodes = 0;
3389         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3390                 int err;
3391                 tgt = lmv->tgts[i];
3392
3393                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3394                         continue;
3395
3396                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3397                 if (err) {
3398                         CERROR("getquota on mdt %d failed. %d\n", i, err);
3399                         if (!rc)
3400                                 rc = err;
3401                 } else {
3402                         curspace += oqctl->qc_dqblk.dqb_curspace;
3403                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3404                 }
3405         }
3406         oqctl->qc_dqblk.dqb_curspace = curspace;
3407         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3408
3409         RETURN(rc);
3410 }
3411
3412 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3413                    struct obd_quotactl *oqctl)
3414 {
3415         struct obd_device       *obd = class_exp2obd(exp);
3416         struct lmv_obd          *lmv = &obd->u.lmv;
3417         struct lmv_tgt_desc     *tgt;
3418         __u32                    i;
3419         int                      rc = 0;
3420         ENTRY;
3421
3422         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3423                 int err;
3424                 tgt = lmv->tgts[i];
3425                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3426                         CERROR("lmv idx %d inactive\n", i);
3427                         RETURN(-EIO);
3428                 }
3429
3430                 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3431                 if (err && !rc)
3432                         rc = err;
3433         }
3434
3435         RETURN(rc);
3436 }
3437
3438 int lmv_update_lsm_md(struct obd_export *exp, struct lmv_stripe_md *lsm,
3439                       struct mdt_body *body, ldlm_blocking_callback cb_blocking)
3440 {
3441         return lmv_revalidate_slaves(exp, body, lsm, cb_blocking, 0);
3442 }
3443
3444 int lmv_merge_attr(struct obd_export *exp, const struct lmv_stripe_md *lsm,
3445                    struct cl_attr *attr)
3446 {
3447         int i;
3448
3449         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3450                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3451
3452                 CDEBUG(D_INFO, ""DFID" size %llu, nlink %u, atime %lu ctime"
3453                        "%lu, mtime %lu.\n", PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3454                        i_size_read(inode), inode->i_nlink,
3455                        LTIME_S(inode->i_atime), LTIME_S(inode->i_ctime),
3456                        LTIME_S(inode->i_mtime));
3457
3458                 /* for slave stripe, it needs to subtract nlink for . and .. */
3459                 if (i != 0)
3460                         attr->cat_nlink += inode->i_nlink - 2;
3461                 else
3462                         attr->cat_nlink = inode->i_nlink;
3463
3464                 attr->cat_size += i_size_read(inode);
3465
3466                 if (attr->cat_atime < LTIME_S(inode->i_atime))
3467                         attr->cat_atime = LTIME_S(inode->i_atime);
3468
3469                 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3470                         attr->cat_ctime = LTIME_S(inode->i_ctime);
3471
3472                 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3473                         attr->cat_mtime = LTIME_S(inode->i_mtime);
3474         }
3475         return 0;
3476 }
3477
3478 struct obd_ops lmv_obd_ops = {
3479         .o_owner                = THIS_MODULE,
3480         .o_setup                = lmv_setup,
3481         .o_cleanup              = lmv_cleanup,
3482         .o_precleanup           = lmv_precleanup,
3483         .o_process_config       = lmv_process_config,
3484         .o_connect              = lmv_connect,
3485         .o_disconnect           = lmv_disconnect,
3486         .o_statfs               = lmv_statfs,
3487         .o_get_info             = lmv_get_info,
3488         .o_set_info_async       = lmv_set_info_async,
3489         .o_packmd               = lmv_packmd,
3490         .o_unpackmd             = lmv_unpackmd,
3491         .o_notify               = lmv_notify,
3492         .o_get_uuid             = lmv_get_uuid,
3493         .o_iocontrol            = lmv_iocontrol,
3494         .o_quotacheck           = lmv_quotacheck,
3495         .o_quotactl             = lmv_quotactl
3496 };
3497
3498 struct md_ops lmv_md_ops = {
3499         .m_getstatus            = lmv_getstatus,
3500         .m_null_inode           = lmv_null_inode,
3501         .m_find_cbdata          = lmv_find_cbdata,
3502         .m_close                = lmv_close,
3503         .m_create               = lmv_create,
3504         .m_done_writing         = lmv_done_writing,
3505         .m_enqueue              = lmv_enqueue,
3506         .m_getattr              = lmv_getattr,
3507         .m_getxattr             = lmv_getxattr,
3508         .m_getattr_name         = lmv_getattr_name,
3509         .m_intent_lock          = lmv_intent_lock,
3510         .m_link                 = lmv_link,
3511         .m_rename               = lmv_rename,
3512         .m_setattr              = lmv_setattr,
3513         .m_setxattr             = lmv_setxattr,
3514         .m_fsync                = lmv_fsync,
3515         .m_read_page            = lmv_read_page,
3516         .m_unlink               = lmv_unlink,
3517         .m_init_ea_size         = lmv_init_ea_size,
3518         .m_cancel_unused        = lmv_cancel_unused,
3519         .m_set_lock_data        = lmv_set_lock_data,
3520         .m_lock_match           = lmv_lock_match,
3521         .m_get_lustre_md        = lmv_get_lustre_md,
3522         .m_free_lustre_md       = lmv_free_lustre_md,
3523         .m_update_lsm_md        = lmv_update_lsm_md,
3524         .m_merge_attr           = lmv_merge_attr,
3525         .m_set_open_replay_data = lmv_set_open_replay_data,
3526         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3527         .m_renew_capa           = lmv_renew_capa,
3528         .m_unpack_capa          = lmv_unpack_capa,
3529         .m_get_remote_perm      = lmv_get_remote_perm,
3530         .m_intent_getattr_async = lmv_intent_getattr_async,
3531         .m_revalidate_lock      = lmv_revalidate_lock,
3532         .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
3533 };
3534
3535 int __init lmv_init(void)
3536 {
3537         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3538                                    LUSTRE_LMV_NAME, NULL);
3539 }
3540
3541 static void lmv_exit(void)
3542 {
3543         class_unregister_type(LUSTRE_LMV_NAME);
3544 }
3545
3546 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3547 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3548 MODULE_LICENSE("GPL");
3549
3550 module_init(lmv_init);
3551 module_exit(lmv_exit);