lustre/lmv/lmv_obd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LMV
  38 #include <linux/slab.h>
  39 #include <linux/module.h>
  40 #include <linux/init.h>
  41 #include <linux/user_namespace.h>
  42 #ifdef HAVE_UIDGID_HEADER
  43 # include <linux/uidgid.h>
  44 #endif
  45 #include <linux/slab.h>
  46 #include <linux/pagemap.h>
  47 #include <linux/mm.h>
  48 #include <linux/math64.h>
  49 #include <linux/seq_file.h>
  50 #include <linux/namei.h>
  51
  52 #include <lustre/lustre_idl.h>
  53 #include <obd_support.h>
  54 #include <lustre_lib.h>
  55 #include <lustre_net.h>
  56 #include <obd_class.h>
  57 #include <lustre_lmv.h>
  58 #include <lprocfs_status.h>
  59 #include <cl_object.h>
  60 #include <lustre_fid.h>
  61 #include <lustre_ioctl.h>
  62 #include <lustre_kernelcomm.h>
  63 #include "lmv_internal.h"
  64
  65 static void lmv_activate_target(struct lmv_obd *lmv,
  66                                 struct lmv_tgt_desc *tgt,
  67                                 int activate)
  68 {
  69         if (tgt->ltd_active == activate)
  70                 return;
  71
  72         tgt->ltd_active = activate;
  73         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  74
  75         tgt->ltd_exp->exp_obd->obd_inactive = !activate;
  76 }
  77
  78 /**
  79  * Error codes:
  80  *
  81  *  -EINVAL  : UUID can't be found in the LMV's target list
  82  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  83  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  84  */
  85 static int lmv_set_mdc_active(struct lmv_obd *lmv,
  86                               const struct obd_uuid *uuid,
  87                               int activate)
  88 {
  89         struct lmv_tgt_desc     *tgt = NULL;
  90         struct obd_device       *obd;
  91         __u32                    i;
  92         int                      rc = 0;
  93         ENTRY;
  94
  95         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  96                         lmv, uuid->uuid, activate);
  97
  98         spin_lock(&lmv->lmv_lock);
  99         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 100                 tgt = lmv->tgts[i];
 101                 if (tgt == NULL || tgt->ltd_exp == NULL)
 102                         continue;
 103
 104                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
 105                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 106
 107                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 108                         break;
 109         }
 110
 111         if (i == lmv->desc.ld_tgt_count)
 112                 GOTO(out_lmv_lock, rc = -EINVAL);
 113
 114         obd = class_exp2obd(tgt->ltd_exp);
 115         if (obd == NULL)
 116                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 117
 118         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 119                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 120                obd->obd_type->typ_name, i);
 121         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 122
 123         if (tgt->ltd_active == activate) {
 124                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 125                        activate ? "" : "in");
 126                 GOTO(out_lmv_lock, rc);
 127         }
 128
 129         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 130                activate ? "" : "in");
 131         lmv_activate_target(lmv, tgt, activate);
 132         EXIT;
 133
 134  out_lmv_lock:
 135         spin_unlock(&lmv->lmv_lock);
 136         return rc;
 137 }
 138
 139 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 140 {
 141         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
 142         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
 143
 144         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
 145 }
 146
 147 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 148                       enum obd_notify_event ev, void *data)
 149 {
 150         struct obd_connect_data *conn_data;
 151         struct lmv_obd          *lmv = &obd->u.lmv;
 152         struct obd_uuid         *uuid;
 153         int                      rc = 0;
 154         ENTRY;
 155
 156         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 157                 CERROR("unexpected notification of %s %s!\n",
 158                        watched->obd_type->typ_name,
 159                        watched->obd_name);
 160                 RETURN(-EINVAL);
 161         }
 162
 163         uuid = &watched->u.cli.cl_target_uuid;
 164         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 165                 /*
 166                  * Set MDC as active before notifying the observer, so the
 167                  * observer can use the MDC normally.
 168                  */
 169                 rc = lmv_set_mdc_active(lmv, uuid,
 170                                         ev == OBD_NOTIFY_ACTIVE);
 171                 if (rc) {
 172                         CERROR("%sactivation of %s failed: %d\n",
 173                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 174                                uuid->uuid, rc);
 175                         RETURN(rc);
 176                 }
 177         } else if (ev == OBD_NOTIFY_OCD) {
 178                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 179                 /*
 180                  * XXX: Make sure that ocd_connect_flags from all targets are
 181                  * the same. Otherwise one of MDTs runs wrong version or
 182                  * something like this.  --umka
 183                  */
 184                 obd->obd_self_export->exp_connect_data = *conn_data;
 185         }
 186 #if 0
 187         else if (ev == OBD_NOTIFY_DISCON) {
 188                 /*
 189                  * For disconnect event, flush fld cache for failout MDS case.
 190                  */
 191                 fld_client_flush(&lmv->lmv_fld);
 192         }
 193 #endif
 194         /*
 195          * Pass the notification up the chain.
 196          */
 197         if (obd->obd_observer)
 198                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 199
 200         RETURN(rc);
 201 }
 202
 203 /**
 204  * This is fake connect function. Its purpose is to initialize lmv and say
 205  * caller that everything is okay. Real connection will be performed later.
 206  */
 207 static int lmv_connect(const struct lu_env *env,
 208                        struct obd_export **exp, struct obd_device *obd,
 209                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 210                        void *localdata)
 211 {
 212         struct lmv_obd        *lmv = &obd->u.lmv;
 213         struct lustre_handle  conn = { 0 };
 214         int                    rc = 0;
 215         ENTRY;
 216
 217         /*
 218          * We don't want to actually do the underlying connections more than
 219          * once, so keep track.
 220          */
 221         lmv->refcount++;
 222         if (lmv->refcount > 1) {
 223                 *exp = NULL;
 224                 RETURN(0);
 225         }
 226
 227         rc = class_connect(&conn, obd, cluuid);
 228         if (rc) {
 229                 CERROR("class_connection() returned %d\n", rc);
 230                 RETURN(rc);
 231         }
 232
 233         *exp = class_conn2export(&conn);
 234         class_export_get(*exp);
 235
 236         lmv->exp = *exp;
 237         lmv->connected = 0;
 238         lmv->cluuid = *cluuid;
 239
 240         if (data)
 241                 lmv->conn_data = *data;
 242
 243         if (lmv->targets_proc_entry == NULL) {
 244                 lmv->targets_proc_entry = lprocfs_register("target_obds",
 245                                                            obd->obd_proc_entry,
 246                                                            NULL, NULL);
 247                 if (IS_ERR(lmv->targets_proc_entry)) {
 248                         CERROR("%s: cannot register "
 249                                "/proc/fs/lustre/%s/%s/target_obds\n",
 250                                obd->obd_name, obd->obd_type->typ_name,
 251                                obd->obd_name);
 252                         lmv->targets_proc_entry = NULL;
 253                 }
 254         }
 255
 256         /*
 257          * All real clients should perform actual connection right away, because
 258          * it is possible, that LMV will not have opportunity to connect targets
 259          * and MDC stuff will be called directly, for instance while reading
 260          * ../mdc/../kbytesfree procfs file, etc.
 261          */
 262         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
 263                 rc = lmv_check_connect(obd);
 264
 265         if (rc && lmv->targets_proc_entry != NULL)
 266                 lprocfs_remove(&lmv->targets_proc_entry);
 267         RETURN(rc);
 268 }
 269
 270 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
 271                             __u32 def_easize)
 272 {
 273         struct obd_device       *obd = exp->exp_obd;
 274         struct lmv_obd          *lmv = &obd->u.lmv;
 275         __u32                    i;
 276         int                      rc = 0;
 277         int                      change = 0;
 278         ENTRY;
 279
 280         if (lmv->max_easize < easize) {
 281                 lmv->max_easize = easize;
 282                 change = 1;
 283         }
 284         if (lmv->max_def_easize < def_easize) {
 285                 lmv->max_def_easize = def_easize;
 286                 change = 1;
 287         }
 288
 289         if (change == 0)
 290                 RETURN(0);
 291
 292         if (lmv->connected == 0)
 293                 RETURN(0);
 294
 295         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 296                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 297
 298                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
 299                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 300                         continue;
 301                 }
 302
 303                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
 304                 if (rc) {
 305                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 306                                " rc = %d\n", obd->obd_name, i, rc);
 307                         break;
 308                 }
 309         }
 310         RETURN(rc);
 311 }
 312
 313 #define MAX_STRING_SIZE 128
 314
 315 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 316 {
 317         struct lmv_obd          *lmv = &obd->u.lmv;
 318         struct obd_uuid         *cluuid = &lmv->cluuid;
 319         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 320         struct obd_device       *mdc_obd;
 321         struct obd_export       *mdc_exp;
 322         struct lu_fld_target     target;
 323         int                      rc;
 324         ENTRY;
 325
 326         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 327                                         &obd->obd_uuid);
 328         if (!mdc_obd) {
 329                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 330                 RETURN(-EINVAL);
 331         }
 332
 333         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 334                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 335                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 336                 cluuid->uuid);
 337
 338         if (!mdc_obd->obd_set_up) {
 339                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 340                 RETURN(-EINVAL);
 341         }
 342
 343         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 344                          &lmv->conn_data, NULL);
 345         if (rc) {
 346                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 347                 RETURN(rc);
 348         }
 349
 350         /*
 351          * Init fid sequence client for this mdc and add new fld target.
 352          */
 353         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 354         if (rc)
 355                 RETURN(rc);
 356
 357         target.ft_srv = NULL;
 358         target.ft_exp = mdc_exp;
 359         target.ft_idx = tgt->ltd_idx;
 360
 361         fld_client_add_target(&lmv->lmv_fld, &target);
 362
 363         rc = obd_register_observer(mdc_obd, obd);
 364         if (rc) {
 365                 obd_disconnect(mdc_exp);
 366                 CERROR("target %s register_observer error %d\n",
 367                        tgt->ltd_uuid.uuid, rc);
 368                 RETURN(rc);
 369         }
 370
 371         if (obd->obd_observer) {
 372                 /*
 373                  * Tell the observer about the new target.
 374                  */
 375                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 376                                 OBD_NOTIFY_ACTIVE,
 377                                 (void *)(tgt - lmv->tgts[0]));
 378                 if (rc) {
 379                         obd_disconnect(mdc_exp);
 380                         RETURN(rc);
 381                 }
 382         }
 383
 384         tgt->ltd_active = 1;
 385         tgt->ltd_exp = mdc_exp;
 386         lmv->desc.ld_active_tgt_count++;
 387
 388         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
 389
 390         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 391                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 392                 atomic_read(&obd->obd_refcount));
 393
 394         if (lmv->targets_proc_entry != NULL) {
 395                 struct proc_dir_entry *mdc_symlink;
 396
 397                 LASSERT(mdc_obd->obd_type != NULL);
 398                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 399                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 400                                                   lmv->targets_proc_entry,
 401                                                   "../../../%s/%s",
 402                                                   mdc_obd->obd_type->typ_name,
 403                                                   mdc_obd->obd_name);
 404                 if (mdc_symlink == NULL) {
 405                         CERROR("cannot register LMV target "
 406                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
 407                                obd->obd_type->typ_name, obd->obd_name,
 408                                mdc_obd->obd_name);
 409                 }
 410         }
 411         RETURN(0);
 412 }
 413
 414 static void lmv_del_target(struct lmv_obd *lmv, int index)
 415 {
 416         if (lmv->tgts[index] == NULL)
 417                 return;
 418
 419         OBD_FREE_PTR(lmv->tgts[index]);
 420         lmv->tgts[index] = NULL;
 421         return;
 422 }
 423
 424 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 425                            __u32 index, int gen)
 426 {
 427         struct obd_device *mdc_obd;
 428         struct lmv_obd      *lmv = &obd->u.lmv;
 429         struct lmv_tgt_desc *tgt;
 430         int                  orig_tgt_count = 0;
 431         int                  rc = 0;
 432         ENTRY;
 433
 434         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 435         mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 436                                         &obd->obd_uuid);
 437         if (!mdc_obd) {
 438                 CERROR("%s: Target %s not attached: rc = %d\n",
 439                        obd->obd_name, uuidp->uuid, -EINVAL);
 440                 RETURN(-EINVAL);
 441         }
 442
 443         mutex_lock(&lmv->lmv_init_mutex);
 444         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
 445                 tgt = lmv->tgts[index];
 446                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
 447                        " rc = %d\n", obd->obd_name,
 448                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
 449                 mutex_unlock(&lmv->lmv_init_mutex);
 450                 RETURN(-EEXIST);
 451         }
 452
 453         if (index >= lmv->tgts_size) {
 454                 /* We need to reallocate the lmv target array. */
 455                 struct lmv_tgt_desc **newtgts, **old = NULL;
 456                 __u32 newsize = 1;
 457                 __u32 oldsize = 0;
 458
 459                 while (newsize < index + 1)
 460                         newsize = newsize << 1;
 461                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 462                 if (newtgts == NULL) {
 463                         mutex_unlock(&lmv->lmv_init_mutex);
 464                         RETURN(-ENOMEM);
 465                 }
 466
 467                 if (lmv->tgts_size) {
 468                         memcpy(newtgts, lmv->tgts,
 469                                sizeof(*newtgts) * lmv->tgts_size);
 470                         old = lmv->tgts;
 471                         oldsize = lmv->tgts_size;
 472                 }
 473
 474                 lmv->tgts = newtgts;
 475                 lmv->tgts_size = newsize;
 476                 smp_rmb();
 477                 if (old)
 478                         OBD_FREE(old, sizeof(*old) * oldsize);
 479
 480                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
 481                        lmv->tgts_size);
 482         }
 483
 484         OBD_ALLOC_PTR(tgt);
 485         if (!tgt) {
 486                 mutex_unlock(&lmv->lmv_init_mutex);
 487                 RETURN(-ENOMEM);
 488         }
 489
 490         mutex_init(&tgt->ltd_fid_mutex);
 491         tgt->ltd_idx = index;
 492         tgt->ltd_uuid = *uuidp;
 493         tgt->ltd_active = 0;
 494         lmv->tgts[index] = tgt;
 495         if (index >= lmv->desc.ld_tgt_count) {
 496                 orig_tgt_count = lmv->desc.ld_tgt_count;
 497                 lmv->desc.ld_tgt_count = index + 1;
 498         }
 499
 500         if (lmv->connected == 0) {
 501                 /* lmv_check_connect() will connect this target. */
 502                 mutex_unlock(&lmv->lmv_init_mutex);
 503                 RETURN(0);
 504         }
 505
 506         /* Otherwise let's connect it ourselves */
 507         mutex_unlock(&lmv->lmv_init_mutex);
 508         rc = lmv_connect_mdc(obd, tgt);
 509         if (rc != 0) {
 510                 spin_lock(&lmv->lmv_lock);
 511                 if (lmv->desc.ld_tgt_count == index + 1)
 512                         lmv->desc.ld_tgt_count = orig_tgt_count;
 513                 memset(tgt, 0, sizeof(*tgt));
 514                 spin_unlock(&lmv->lmv_lock);
 515         } else {
 516                 int easize = sizeof(struct lmv_stripe_md) +
 517                         lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
 518                 lmv_init_ea_size(obd->obd_self_export, easize, 0);
 519         }
 520
 521         RETURN(rc);
 522 }
 523
 524 int lmv_check_connect(struct obd_device *obd)
 525 {
 526         struct lmv_obd          *lmv = &obd->u.lmv;
 527         struct lmv_tgt_desc     *tgt;
 528         __u32                    i;
 529         int                      rc;
 530         int                      easize;
 531         ENTRY;
 532
 533         if (lmv->connected)
 534                 RETURN(0);
 535
 536         mutex_lock(&lmv->lmv_init_mutex);
 537         if (lmv->connected) {
 538                 mutex_unlock(&lmv->lmv_init_mutex);
 539                 RETURN(0);
 540         }
 541
 542         if (lmv->desc.ld_tgt_count == 0) {
 543                 mutex_unlock(&lmv->lmv_init_mutex);
 544                 CERROR("%s: no targets configured.\n", obd->obd_name);
 545                 RETURN(-EINVAL);
 546         }
 547
 548         LASSERT(lmv->tgts != NULL);
 549
 550         if (lmv->tgts[0] == NULL) {
 551                 mutex_unlock(&lmv->lmv_init_mutex);
 552                 CERROR("%s: no target configured for index 0.\n",
 553                        obd->obd_name);
 554                 RETURN(-EINVAL);
 555         }
 556
 557         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 558                lmv->cluuid.uuid, obd->obd_name);
 559
 560         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 561                 tgt = lmv->tgts[i];
 562                 if (tgt == NULL)
 563                         continue;
 564                 rc = lmv_connect_mdc(obd, tgt);
 565                 if (rc)
 566                         GOTO(out_disc, rc);
 567         }
 568
 569         class_export_put(lmv->exp);
 570         lmv->connected = 1;
 571         easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
 572         lmv_init_ea_size(obd->obd_self_export, easize, 0);
 573         mutex_unlock(&lmv->lmv_init_mutex);
 574         RETURN(0);
 575
 576  out_disc:
 577         while (i-- > 0) {
 578                 int rc2;
 579                 tgt = lmv->tgts[i];
 580                 if (tgt == NULL)
 581                         continue;
 582                 tgt->ltd_active = 0;
 583                 if (tgt->ltd_exp) {
 584                         --lmv->desc.ld_active_tgt_count;
 585                         rc2 = obd_disconnect(tgt->ltd_exp);
 586                         if (rc2) {
 587                                 CERROR("LMV target %s disconnect on "
 588                                        "MDC idx %d: error %d\n",
 589                                        tgt->ltd_uuid.uuid, i, rc2);
 590                         }
 591                 }
 592         }
 593         class_disconnect(lmv->exp);
 594         mutex_unlock(&lmv->lmv_init_mutex);
 595         RETURN(rc);
 596 }
 597
 598 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 599 {
 600         struct lmv_obd         *lmv = &obd->u.lmv;
 601         struct obd_device      *mdc_obd;
 602         int                     rc;
 603         ENTRY;
 604
 605         LASSERT(tgt != NULL);
 606         LASSERT(obd != NULL);
 607
 608         mdc_obd = class_exp2obd(tgt->ltd_exp);
 609
 610         if (mdc_obd) {
 611                 mdc_obd->obd_force = obd->obd_force;
 612                 mdc_obd->obd_fail = obd->obd_fail;
 613                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 614
 615                 if (lmv->targets_proc_entry != NULL)
 616                         lprocfs_remove_proc_entry(mdc_obd->obd_name,
 617                                                   lmv->targets_proc_entry);
 618         }
 619
 620         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 621         if (rc)
 622                 CERROR("Can't finanize fids factory\n");
 623
 624         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 625                tgt->ltd_exp->exp_obd->obd_name,
 626                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 627
 628         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 629         rc = obd_disconnect(tgt->ltd_exp);
 630         if (rc) {
 631                 if (tgt->ltd_active) {
 632                         CERROR("Target %s disconnect error %d\n",
 633                                tgt->ltd_uuid.uuid, rc);
 634                 }
 635         }
 636
 637         lmv_activate_target(lmv, tgt, 0);
 638         tgt->ltd_exp = NULL;
 639         RETURN(0);
 640 }
 641
 642 static int lmv_disconnect(struct obd_export *exp)
 643 {
 644         struct obd_device       *obd = class_exp2obd(exp);
 645         struct lmv_obd          *lmv = &obd->u.lmv;
 646         int                      rc;
 647         __u32                    i;
 648         ENTRY;
 649
 650         if (!lmv->tgts)
 651                 goto out_local;
 652
 653         /*
 654          * Only disconnect the underlying layers on the final disconnect.
 655          */
 656         lmv->refcount--;
 657         if (lmv->refcount != 0)
 658                 goto out_local;
 659
 660         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 661                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
 662                         continue;
 663
 664                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
 665         }
 666
 667         if (lmv->targets_proc_entry != NULL)
 668                 lprocfs_remove(&lmv->targets_proc_entry);
 669         else
 670                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 671                        obd->obd_type->typ_name, obd->obd_name);
 672
 673 out_local:
 674         /*
 675          * This is the case when no real connection is established by
 676          * lmv_check_connect().
 677          */
 678         if (!lmv->connected)
 679                 class_export_put(exp);
 680         rc = class_disconnect(exp);
 681         if (lmv->refcount == 0)
 682                 lmv->connected = 0;
 683         RETURN(rc);
 684 }
 685
 686 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
 687                         void __user *uarg)
 688 {
 689         struct obd_device       *obddev = class_exp2obd(exp);
 690         struct lmv_obd          *lmv = &obddev->u.lmv;
 691         struct getinfo_fid2path *gf;
 692         struct lmv_tgt_desc     *tgt;
 693         struct getinfo_fid2path *remote_gf = NULL;
 694         struct lu_fid           root_fid;
 695         int                     remote_gf_size = 0;
 696         int                     rc;
 697
 698         gf = karg;
 699         tgt = lmv_find_target(lmv, &gf->gf_fid);
 700         if (IS_ERR(tgt))
 701                 RETURN(PTR_ERR(tgt));
 702
 703         root_fid = *gf->gf_u.gf_root_fid;
 704         LASSERT(fid_is_sane(&root_fid));
 705
 706 repeat_fid2path:
 707         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 708         if (rc != 0 && rc != -EREMOTE)
 709                 GOTO(out_fid2path, rc);
 710
 711         /* If remote_gf != NULL, it means just building the
 712          * path on the remote MDT, copy this path segement to gf */
 713         if (remote_gf != NULL) {
 714                 struct getinfo_fid2path *ori_gf;
 715                 char *ptr;
 716
 717                 ori_gf = (struct getinfo_fid2path *)karg;
 718                 if (strlen(ori_gf->gf_u.gf_path) +
 719                     strlen(gf->gf_u.gf_path) > ori_gf->gf_pathlen)
 720                         GOTO(out_fid2path, rc = -EOVERFLOW);
 721
 722                 ptr = ori_gf->gf_u.gf_path;
 723
 724                 memmove(ptr + strlen(gf->gf_u.gf_path) + 1, ptr,
 725                         strlen(ori_gf->gf_u.gf_path));
 726
 727                 strncpy(ptr, gf->gf_u.gf_path,
 728                         strlen(gf->gf_u.gf_path));
 729                 ptr += strlen(gf->gf_u.gf_path);
 730                 *ptr = '/';
 731         }
 732
 733         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
 734                tgt->ltd_exp->exp_obd->obd_name,
 735                gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 736                gf->gf_linkno);
 737
 738         if (rc == 0)
 739                 GOTO(out_fid2path, rc);
 740
 741         /* sigh, has to go to another MDT to do path building further */
 742         if (remote_gf == NULL) {
 743                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 744                 OBD_ALLOC(remote_gf, remote_gf_size);
 745                 if (remote_gf == NULL)
 746                         GOTO(out_fid2path, rc = -ENOMEM);
 747                 remote_gf->gf_pathlen = PATH_MAX;
 748         }
 749
 750         if (!fid_is_sane(&gf->gf_fid)) {
 751                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 752                        tgt->ltd_exp->exp_obd->obd_name,
 753                        PFID(&gf->gf_fid), -EINVAL);
 754                 GOTO(out_fid2path, rc = -EINVAL);
 755         }
 756
 757         tgt = lmv_find_target(lmv, &gf->gf_fid);
 758         if (IS_ERR(tgt))
 759                 GOTO(out_fid2path, rc = -EINVAL);
 760
 761         remote_gf->gf_fid = gf->gf_fid;
 762         remote_gf->gf_recno = -1;
 763         remote_gf->gf_linkno = -1;
 764         memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
 765         *remote_gf->gf_u.gf_root_fid = root_fid;
 766         gf = remote_gf;
 767         goto repeat_fid2path;
 768
 769 out_fid2path:
 770         if (remote_gf != NULL)
 771                 OBD_FREE(remote_gf, remote_gf_size);
 772         RETURN(rc);
 773 }
 774
 775 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 776                              const struct hsm_user_request *hur,
 777                              const struct lmv_tgt_desc *tgt_mds)
 778 {
 779         __u32                    i;
 780         int                      nr = 0;
 781         struct lmv_tgt_desc     *curr_tgt;
 782
 783         /* count how many requests must be sent to the given target */
 784         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 785                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
 786                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 787                         nr++;
 788         }
 789         return nr;
 790 }
 791
 792 static void lmv_hsm_req_build(struct lmv_obd *lmv,
 793                               struct hsm_user_request *hur_in,
 794                               const struct lmv_tgt_desc *tgt_mds,
 795                               struct hsm_user_request *hur_out)
 796 {
 797         __u32                    i, nr_out;
 798         struct lmv_tgt_desc     *curr_tgt;
 799
 800         /* build the hsm_user_request for the given target */
 801         hur_out->hur_request = hur_in->hur_request;
 802         nr_out = 0;
 803         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 804                 curr_tgt = lmv_find_target(lmv,
 805                                            &hur_in->hur_user_item[i].hui_fid);
 806                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 807                         hur_out->hur_user_item[nr_out] =
 808                                                 hur_in->hur_user_item[i];
 809                         nr_out++;
 810                 }
 811         }
 812         hur_out->hur_request.hr_itemcount = nr_out;
 813         memcpy(hur_data(hur_out), hur_data(hur_in),
 814                hur_in->hur_request.hr_data_len);
 815 }
 816
 817 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
 818                                  struct lustre_kernelcomm *lk,
 819                                  void __user *uarg)
 820 {
 821         __u32   i;
 822         int     rc;
 823         ENTRY;
 824
 825         /* unregister request (call from llapi_hsm_copytool_fini) */
 826         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 827                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 828
 829                 if (tgt == NULL || tgt->ltd_exp == NULL)
 830                         continue;
 831                 /* best effort: try to clean as much as possible
 832                  * (continue on error) */
 833                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 834         }
 835
 836         /* Whatever the result, remove copytool from kuc groups.
 837          * Unreached coordinators will get EPIPE on next requests
 838          * and will unregister automatically.
 839          */
 840         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group);
 841
 842         RETURN(rc);
 843 }
 844
 845 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
 846                                struct lustre_kernelcomm *lk, __user void *uarg)
 847 {
 848         struct file             *filp;
 849         __u32                    i, j;
 850         int                      err, rc;
 851         bool                     any_set = false;
 852         struct kkuc_ct_data      kcd = { 0 };
 853         ENTRY;
 854
 855         /* All or nothing: try to register to all MDS.
 856          * In case of failure, unregister from previous MDS,
 857          * except if it because of inactive target. */
 858         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 859                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 860
 861                 if (tgt == NULL || tgt->ltd_exp == NULL)
 862                         continue;
 863                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 864                 if (err) {
 865                         if (tgt->ltd_active) {
 866                                 /* permanent error */
 867                                 CERROR("%s: iocontrol MDC %s on MDT"
 868                                        " idx %d cmd %x: err = %d\n",
 869                                        class_exp2obd(lmv->exp)->obd_name,
 870                                        tgt->ltd_uuid.uuid, i, cmd, err);
 871                                 rc = err;
 872                                 lk->lk_flags |= LK_FLG_STOP;
 873                                 /* unregister from previous MDS */
 874                                 for (j = 0; j < i; j++) {
 875                                         tgt = lmv->tgts[j];
 876                                         if (tgt == NULL || tgt->ltd_exp == NULL)
 877                                                 continue;
 878                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
 879                                                       lk, uarg);
 880                                 }
 881                                 RETURN(rc);
 882                         }
 883                         /* else: transient error.
 884                          * kuc will register to the missing MDT
 885                          * when it is back */
 886                 } else {
 887                         any_set = true;
 888                 }
 889         }
 890
 891         if (!any_set)
 892                 /* no registration done: return error */
 893                 RETURN(-ENOTCONN);
 894
 895         /* at least one registration done, with no failure */
 896         filp = fget(lk->lk_wfd);
 897         if (filp == NULL)
 898                 RETURN(-EBADF);
 899
 900         kcd.kcd_magic = KKUC_CT_DATA_MAGIC;
 901         kcd.kcd_uuid = lmv->cluuid;
 902         kcd.kcd_archive = lk->lk_data;
 903
 904         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group,
 905                                    &kcd, sizeof(kcd));
 906         if (rc != 0)
 907                 fput(filp);
 908
 909         RETURN(rc);
 910 }
 911
 912
 913
 914
 915 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 916                          int len, void *karg, void __user *uarg)
 917 {
 918         struct obd_device       *obddev = class_exp2obd(exp);
 919         struct lmv_obd          *lmv = &obddev->u.lmv;
 920         struct lmv_tgt_desc     *tgt = NULL;
 921         __u32                    i = 0;
 922         int                      rc = 0;
 923         int                      set = 0;
 924         __u32                    count = lmv->desc.ld_tgt_count;
 925         ENTRY;
 926
 927         if (count == 0)
 928                 RETURN(-ENOTTY);
 929
 930         switch (cmd) {
 931         case IOC_OBD_STATFS: {
 932                 struct obd_ioctl_data *data = karg;
 933                 struct obd_device *mdc_obd;
 934                 struct obd_statfs stat_buf = {0};
 935                 __u32 index;
 936
 937                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 938                 if ((index >= count))
 939                         RETURN(-ENODEV);
 940
 941                 tgt = lmv->tgts[index];
 942                 if (tgt == NULL || !tgt->ltd_active)
 943                         RETURN(-ENODATA);
 944
 945                 mdc_obd = class_exp2obd(tgt->ltd_exp);
 946                 if (!mdc_obd)
 947                         RETURN(-EINVAL);
 948
 949                 /* copy UUID */
 950                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 951                                  min((int) data->ioc_plen2,
 952                                      (int) sizeof(struct obd_uuid))))
 953                         RETURN(-EFAULT);
 954
 955                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
 956                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 957                                 0);
 958                 if (rc)
 959                         RETURN(rc);
 960                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
 961                                  min((int) data->ioc_plen1,
 962                                      (int) sizeof(stat_buf))))
 963                         RETURN(-EFAULT);
 964                 break;
 965         }
 966         case OBD_IOC_QUOTACTL: {
 967                 struct if_quotactl *qctl = karg;
 968                 struct obd_quotactl *oqctl;
 969
 970                 if (qctl->qc_valid == QC_MDTIDX) {
 971                         if (count <= qctl->qc_idx)
 972                                 RETURN(-EINVAL);
 973
 974                         tgt = lmv->tgts[qctl->qc_idx];
 975                         if (tgt == NULL || tgt->ltd_exp == NULL)
 976                                 RETURN(-EINVAL);
 977                 } else if (qctl->qc_valid == QC_UUID) {
 978                         for (i = 0; i < count; i++) {
 979                                 tgt = lmv->tgts[i];
 980                                 if (tgt == NULL)
 981                                         continue;
 982                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 983                                                      &qctl->obd_uuid))
 984                                         continue;
 985
 986                                 if (tgt->ltd_exp == NULL)
 987                                         RETURN(-EINVAL);
 988
 989                                 break;
 990                         }
 991                 } else {
 992                         RETURN(-EINVAL);
 993                 }
 994
 995                 if (i >= count)
 996                         RETURN(-EAGAIN);
 997
 998                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
 999                 OBD_ALLOC_PTR(oqctl);
1000                 if (!oqctl)
1001                         RETURN(-ENOMEM);
1002
1003                 QCTL_COPY(oqctl, qctl);
1004                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1005                 if (rc == 0) {
1006                         QCTL_COPY(qctl, oqctl);
1007                         qctl->qc_valid = QC_MDTIDX;
1008                         qctl->obd_uuid = tgt->ltd_uuid;
1009                 }
1010                 OBD_FREE_PTR(oqctl);
1011                 break;
1012         }
1013         case OBD_IOC_CHANGELOG_SEND:
1014         case OBD_IOC_CHANGELOG_CLEAR: {
1015                 struct ioc_changelog *icc = karg;
1016
1017                 if (icc->icc_mdtindex >= count)
1018                         RETURN(-ENODEV);
1019
1020                 tgt = lmv->tgts[icc->icc_mdtindex];
1021                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1022                         RETURN(-ENODEV);
1023                 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1024                 break;
1025         }
1026         case LL_IOC_GET_CONNECT_FLAGS: {
1027                 tgt = lmv->tgts[0];
1028                 if (tgt == NULL || tgt->ltd_exp == NULL)
1029                         RETURN(-ENODATA);
1030                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1031                 break;
1032         }
1033         case LL_IOC_FID2MDTIDX: {
1034                 struct lu_fid *fid = karg;
1035                 int             mdt_index;
1036
1037                 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1038                 if (rc != 0)
1039                         RETURN(rc);
1040
1041                 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1042                  * point to user space memory for FID2MDTIDX. */
1043                 *(__u32 *)uarg = mdt_index;
1044                 break;
1045         }
1046         case OBD_IOC_FID2PATH: {
1047                 rc = lmv_fid2path(exp, len, karg, uarg);
1048                 break;
1049         }
1050         case LL_IOC_HSM_STATE_GET:
1051         case LL_IOC_HSM_STATE_SET:
1052         case LL_IOC_HSM_ACTION: {
1053                 struct md_op_data       *op_data = karg;
1054
1055                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1056                 if (IS_ERR(tgt))
1057                         RETURN(PTR_ERR(tgt));
1058
1059                 if (tgt->ltd_exp == NULL)
1060                         RETURN(-EINVAL);
1061
1062                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1063                 break;
1064         }
1065         case LL_IOC_HSM_PROGRESS: {
1066                 const struct hsm_progress_kernel *hpk = karg;
1067
1068                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1069                 if (IS_ERR(tgt))
1070                         RETURN(PTR_ERR(tgt));
1071                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1072                 break;
1073         }
1074         case LL_IOC_HSM_REQUEST: {
1075                 struct hsm_user_request *hur = karg;
1076                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1077
1078                 if (reqcount == 0)
1079                         RETURN(0);
1080
1081                 /* if the request is about a single fid
1082                  * or if there is a single MDS, no need to split
1083                  * the request. */
1084                 if (reqcount == 1 || count == 1) {
1085                         tgt = lmv_find_target(lmv,
1086                                               &hur->hur_user_item[0].hui_fid);
1087                         if (IS_ERR(tgt))
1088                                 RETURN(PTR_ERR(tgt));
1089                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1090                 } else {
1091                         /* split fid list to their respective MDS */
1092                         for (i = 0; i < count; i++) {
1093                                 unsigned int            nr, reqlen;
1094                                 int                     rc1;
1095                                 struct hsm_user_request *req;
1096
1097                                 tgt = lmv->tgts[i];
1098                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1099                                         continue;
1100
1101                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1102                                 if (nr == 0) /* nothing for this MDS */
1103                                         continue;
1104
1105                                 /* build a request with fids for this MDS */
1106                                 reqlen = offsetof(typeof(*hur),
1107                                                   hur_user_item[nr])
1108                                                 + hur->hur_request.hr_data_len;
1109                                 OBD_ALLOC_LARGE(req, reqlen);
1110                                 if (req == NULL)
1111                                         RETURN(-ENOMEM);
1112
1113                                 lmv_hsm_req_build(lmv, hur, tgt, req);
1114
1115                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1116                                                     req, uarg);
1117                                 if (rc1 != 0 && rc == 0)
1118                                         rc = rc1;
1119                                 OBD_FREE_LARGE(req, reqlen);
1120                         }
1121                 }
1122                 break;
1123         }
1124         case LL_IOC_LOV_SWAP_LAYOUTS: {
1125                 struct md_op_data       *op_data = karg;
1126                 struct lmv_tgt_desc     *tgt1, *tgt2;
1127
1128                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1129                 if (IS_ERR(tgt1))
1130                         RETURN(PTR_ERR(tgt1));
1131
1132                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1133                 if (IS_ERR(tgt2))
1134                         RETURN(PTR_ERR(tgt2));
1135
1136                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1137                         RETURN(-EINVAL);
1138
1139                 /* only files on same MDT can have their layouts swapped */
1140                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1141                         RETURN(-EPERM);
1142
1143                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1144                 break;
1145         }
1146         case LL_IOC_HSM_CT_START: {
1147                 struct lustre_kernelcomm *lk = karg;
1148                 if (lk->lk_flags & LK_FLG_STOP)
1149                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1150                 else
1151                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1152                 break;
1153         }
1154         default:
1155                 for (i = 0; i < count; i++) {
1156                         struct obd_device *mdc_obd;
1157                         int err;
1158
1159                         tgt = lmv->tgts[i];
1160                         if (tgt == NULL || tgt->ltd_exp == NULL)
1161                                 continue;
1162                         /* ll_umount_begin() sets force flag but for lmv, not
1163                          * mdc. Let's pass it through */
1164                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1165                         mdc_obd->obd_force = obddev->obd_force;
1166                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1167                         if (err) {
1168                                 if (tgt->ltd_active) {
1169                                         CERROR("error: iocontrol MDC %s on MDT"
1170                                                " idx %d cmd %x: err = %d\n",
1171                                                tgt->ltd_uuid.uuid, i, cmd, err);
1172                                         if (!rc)
1173                                                 rc = err;
1174                                 }
1175                         } else
1176                                 set = 1;
1177                 }
1178                 if (!set && !rc)
1179                         rc = -EIO;
1180         }
1181         RETURN(rc);
1182 }
1183
1184 #if 0
1185 static int lmv_all_chars_policy(int count, const char *name,
1186                                 int len)
1187 {
1188         unsigned int c = 0;
1189
1190         while (len > 0)
1191                 c += name[--len];
1192         c = c % count;
1193         return c;
1194 }
1195
1196 static int lmv_nid_policy(struct lmv_obd *lmv)
1197 {
1198         struct obd_import *imp;
1199         __u32              id;
1200
1201         /*
1202          * XXX: To get nid we assume that underlying obd device is mdc.
1203          */
1204         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1205         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1206         return id % lmv->desc.ld_tgt_count;
1207 }
1208
1209 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1210                           placement_policy_t placement)
1211 {
1212         switch (placement) {
1213         case PLACEMENT_CHAR_POLICY:
1214                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1215                                             op_data->op_name,
1216                                             op_data->op_namelen);
1217         case PLACEMENT_NID_POLICY:
1218                 return lmv_nid_policy(lmv);
1219
1220         default:
1221                 break;
1222         }
1223
1224         CERROR("Unsupported placement policy %x\n", placement);
1225         return -EINVAL;
1226 }
1227 #endif
1228
1229 /**
1230  * This is _inode_ placement policy function (not name).
1231  */
1232 static int lmv_placement_policy(struct obd_device *obd,
1233                                 struct md_op_data *op_data, u32 *mds)
1234 {
1235         struct lmv_obd          *lmv = &obd->u.lmv;
1236         ENTRY;
1237
1238         LASSERT(mds != NULL);
1239
1240         if (lmv->desc.ld_tgt_count == 1) {
1241                 *mds = 0;
1242                 RETURN(0);
1243         }
1244
1245         if (op_data->op_default_stripe_offset != -1) {
1246                 *mds = op_data->op_default_stripe_offset;
1247                 RETURN(0);
1248         }
1249
1250         /**
1251          * If stripe_offset is provided during setdirstripe
1252          * (setdirstripe -i xx), xx MDS will be choosen.
1253          */
1254         if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1255                 struct lmv_user_md *lum;
1256
1257                 lum = op_data->op_data;
1258
1259                 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1260                         *mds = le32_to_cpu(lum->lum_stripe_offset);
1261                 } else {
1262                         /* -1 means default, which will be in the same MDT with
1263                          * the stripe */
1264                         *mds = op_data->op_mds;
1265                         lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1266                 }
1267         } else {
1268                 /* Allocate new fid on target according to operation type and
1269                  * parent home mds. */
1270                 *mds = op_data->op_mds;
1271         }
1272
1273         RETURN(0);
1274 }
1275
1276 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1277 {
1278         struct lmv_tgt_desc     *tgt;
1279         int                      rc;
1280         ENTRY;
1281
1282         tgt = lmv_get_target(lmv, mds, NULL);
1283         if (IS_ERR(tgt))
1284                 RETURN(PTR_ERR(tgt));
1285
1286         /*
1287          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1288          * on server that seq in new allocated fid is not yet known.
1289          */
1290         mutex_lock(&tgt->ltd_fid_mutex);
1291
1292         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1293                 GOTO(out, rc = -ENODEV);
1294
1295         /*
1296          * Asking underlying tgt layer to allocate new fid.
1297          */
1298         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1299         if (rc > 0) {
1300                 LASSERT(fid_is_sane(fid));
1301                 rc = 0;
1302         }
1303
1304         EXIT;
1305 out:
1306         mutex_unlock(&tgt->ltd_fid_mutex);
1307         return rc;
1308 }
1309
1310 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1311                   struct lu_fid *fid, struct md_op_data *op_data)
1312 {
1313         struct obd_device     *obd = class_exp2obd(exp);
1314         struct lmv_obd        *lmv = &obd->u.lmv;
1315         u32                    mds = 0;
1316         int                    rc;
1317         ENTRY;
1318
1319         LASSERT(op_data != NULL);
1320         LASSERT(fid != NULL);
1321
1322         rc = lmv_placement_policy(obd, op_data, &mds);
1323         if (rc) {
1324                 CERROR("Can't get target for allocating fid, "
1325                        "rc %d\n", rc);
1326                 RETURN(rc);
1327         }
1328
1329         rc = __lmv_fid_alloc(lmv, fid, mds);
1330         if (rc) {
1331                 CERROR("Can't alloc new fid, rc %d\n", rc);
1332                 RETURN(rc);
1333         }
1334
1335         RETURN(rc);
1336 }
1337
1338 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1339 {
1340         struct lmv_obd  *lmv = &obd->u.lmv;
1341         struct lmv_desc *desc;
1342         int             rc;
1343         ENTRY;
1344
1345         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1346                 CERROR("LMV setup requires a descriptor\n");
1347                 RETURN(-EINVAL);
1348         }
1349
1350         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1351         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1352                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1353                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1354                 RETURN(-EINVAL);
1355         }
1356
1357         lmv->tgts_size = 32U;
1358         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1359         if (lmv->tgts == NULL)
1360                 RETURN(-ENOMEM);
1361
1362         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1363         lmv->desc.ld_tgt_count = 0;
1364         lmv->desc.ld_active_tgt_count = 0;
1365         lmv->max_def_easize = 0;
1366         lmv->max_easize = 0;
1367         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1368
1369         spin_lock_init(&lmv->lmv_lock);
1370         mutex_init(&lmv->lmv_init_mutex);
1371
1372 #ifdef CONFIG_PROC_FS
1373         obd->obd_vars = lprocfs_lmv_obd_vars;
1374         lprocfs_obd_setup(obd);
1375         lprocfs_alloc_md_stats(obd, 0);
1376         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1377                                 0444, &lmv_proc_target_fops, obd);
1378         if (rc)
1379                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1380                       obd->obd_name, rc);
1381 #endif
1382         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1383                              LUSTRE_CLI_FLD_HASH_DHT);
1384         if (rc) {
1385                 CERROR("Can't init FLD, err %d\n", rc);
1386                 GOTO(out, rc);
1387         }
1388
1389         RETURN(0);
1390
1391 out:
1392         return rc;
1393 }
1394
1395 static int lmv_cleanup(struct obd_device *obd)
1396 {
1397         struct lmv_obd   *lmv = &obd->u.lmv;
1398         ENTRY;
1399
1400         fld_client_fini(&lmv->lmv_fld);
1401         if (lmv->tgts != NULL) {
1402                 int i;
1403                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1404                         if (lmv->tgts[i] == NULL)
1405                                 continue;
1406                         lmv_del_target(lmv, i);
1407                 }
1408                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1409                 lmv->tgts_size = 0;
1410         }
1411         RETURN(0);
1412 }
1413
1414 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1415 {
1416         struct lustre_cfg       *lcfg = buf;
1417         struct obd_uuid         obd_uuid;
1418         int                     gen;
1419         __u32                   index;
1420         int                     rc;
1421         ENTRY;
1422
1423         switch (lcfg->lcfg_command) {
1424         case LCFG_ADD_MDC:
1425                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1426                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1427                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1428                         GOTO(out, rc = -EINVAL);
1429
1430                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1431
1432                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1433                         GOTO(out, rc = -EINVAL);
1434                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1435                         GOTO(out, rc = -EINVAL);
1436                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1437                 GOTO(out, rc);
1438         default:
1439                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1440                 GOTO(out, rc = -EINVAL);
1441         }
1442 out:
1443         RETURN(rc);
1444 }
1445
1446 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1447                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1448 {
1449         struct obd_device       *obd = class_exp2obd(exp);
1450         struct lmv_obd          *lmv = &obd->u.lmv;
1451         struct obd_statfs       *temp;
1452         int                      rc = 0;
1453         __u32                    i;
1454         ENTRY;
1455
1456         rc = lmv_check_connect(obd);
1457         if (rc)
1458                 RETURN(rc);
1459
1460         OBD_ALLOC(temp, sizeof(*temp));
1461         if (temp == NULL)
1462                 RETURN(-ENOMEM);
1463
1464         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1465                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1466                         continue;
1467
1468                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1469                                 max_age, flags);
1470                 if (rc) {
1471                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1472                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1473                                rc);
1474                         GOTO(out_free_temp, rc);
1475                 }
1476
1477                 if (i == 0) {
1478                         *osfs = *temp;
1479                         /* If the statfs is from mount, it will needs
1480                          * retrieve necessary information from MDT0.
1481                          * i.e. mount does not need the merged osfs
1482                          * from all of MDT.
1483                          * And also clients can be mounted as long as
1484                          * MDT0 is in service*/
1485                         if (flags & OBD_STATFS_FOR_MDT0)
1486                                 GOTO(out_free_temp, rc);
1487                 } else {
1488                         osfs->os_bavail += temp->os_bavail;
1489                         osfs->os_blocks += temp->os_blocks;
1490                         osfs->os_ffree += temp->os_ffree;
1491                         osfs->os_files += temp->os_files;
1492                 }
1493         }
1494
1495         EXIT;
1496 out_free_temp:
1497         OBD_FREE(temp, sizeof(*temp));
1498         return rc;
1499 }
1500
1501 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1502                         struct lu_fid *fid)
1503 {
1504         struct obd_device    *obd = exp->exp_obd;
1505         struct lmv_obd       *lmv = &obd->u.lmv;
1506         int                   rc;
1507         ENTRY;
1508
1509         rc = lmv_check_connect(obd);
1510         if (rc)
1511                 RETURN(rc);
1512
1513         rc = md_get_root(lmv->tgts[0]->ltd_exp, fileset, fid);
1514         RETURN(rc);
1515 }
1516
1517 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1518                         u64 valid, const char *name,
1519                         const char *input, int input_size, int output_size,
1520                         int flags, struct ptlrpc_request **request)
1521 {
1522         struct obd_device      *obd = exp->exp_obd;
1523         struct lmv_obd         *lmv = &obd->u.lmv;
1524         struct lmv_tgt_desc    *tgt;
1525         int                     rc;
1526         ENTRY;
1527
1528         rc = lmv_check_connect(obd);
1529         if (rc)
1530                 RETURN(rc);
1531
1532         tgt = lmv_find_target(lmv, fid);
1533         if (IS_ERR(tgt))
1534                 RETURN(PTR_ERR(tgt));
1535
1536         rc = md_getxattr(tgt->ltd_exp, fid, valid, name, input,
1537                          input_size, output_size, flags, request);
1538
1539         RETURN(rc);
1540 }
1541
1542 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1543                         u64 valid, const char *name,
1544                         const char *input, int input_size, int output_size,
1545                         int flags, __u32 suppgid,
1546                         struct ptlrpc_request **request)
1547 {
1548         struct obd_device      *obd = exp->exp_obd;
1549         struct lmv_obd         *lmv = &obd->u.lmv;
1550         struct lmv_tgt_desc    *tgt;
1551         int                     rc;
1552         ENTRY;
1553
1554         rc = lmv_check_connect(obd);
1555         if (rc)
1556                 RETURN(rc);
1557
1558         tgt = lmv_find_target(lmv, fid);
1559         if (IS_ERR(tgt))
1560                 RETURN(PTR_ERR(tgt));
1561
1562         rc = md_setxattr(tgt->ltd_exp, fid, valid, name, input,
1563                          input_size, output_size, flags, suppgid,
1564                          request);
1565
1566         RETURN(rc);
1567 }
1568
1569 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1570                        struct ptlrpc_request **request)
1571 {
1572         struct obd_device       *obd = exp->exp_obd;
1573         struct lmv_obd          *lmv = &obd->u.lmv;
1574         struct lmv_tgt_desc     *tgt;
1575         int                      rc;
1576         ENTRY;
1577
1578         rc = lmv_check_connect(obd);
1579         if (rc)
1580                 RETURN(rc);
1581
1582         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1583         if (IS_ERR(tgt))
1584                 RETURN(PTR_ERR(tgt));
1585
1586         if (op_data->op_flags & MF_GET_MDT_IDX) {
1587                 op_data->op_mds = tgt->ltd_idx;
1588                 RETURN(0);
1589         }
1590
1591         rc = md_getattr(tgt->ltd_exp, op_data, request);
1592
1593         RETURN(rc);
1594 }
1595
1596 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1597 {
1598         struct obd_device   *obd = exp->exp_obd;
1599         struct lmv_obd      *lmv = &obd->u.lmv;
1600         __u32                i;
1601         int                  rc;
1602         ENTRY;
1603
1604         rc = lmv_check_connect(obd);
1605         if (rc)
1606                 RETURN(rc);
1607
1608         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1609
1610         /*
1611          * With DNE every object can have two locks in different namespaces:
1612          * lookup lock in space of MDT storing direntry and update/open lock in
1613          * space of MDT storing inode.
1614          */
1615         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1616                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1617                         continue;
1618                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1619         }
1620
1621         RETURN(0);
1622 }
1623
1624 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1625                      struct md_open_data *mod, struct ptlrpc_request **request)
1626 {
1627         struct obd_device     *obd = exp->exp_obd;
1628         struct lmv_obd        *lmv = &obd->u.lmv;
1629         struct lmv_tgt_desc   *tgt;
1630         int                    rc;
1631         ENTRY;
1632
1633         rc = lmv_check_connect(obd);
1634         if (rc)
1635                 RETURN(rc);
1636
1637         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1638         if (IS_ERR(tgt))
1639                 RETURN(PTR_ERR(tgt));
1640
1641         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1642         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1643         RETURN(rc);
1644 }
1645
1646 /**
1647  * Choosing the MDT by name or FID in @op_data.
1648  * For non-striped directory, it will locate MDT by fid.
1649  * For striped-directory, it will locate MDT by name. And also
1650  * it will reset op_fid1 with the FID of the choosen stripe.
1651  **/
1652 struct lmv_tgt_desc *
1653 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1654                            const char *name, int namelen, struct lu_fid *fid,
1655                            u32 *mds)
1656 {
1657         struct lmv_tgt_desc     *tgt;
1658         const struct lmv_oinfo  *oinfo;
1659
1660         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1661                 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1662                         RETURN(ERR_PTR(-EBADF));
1663                 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1664         } else {
1665                 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1666                 if (IS_ERR(oinfo))
1667                         RETURN(ERR_CAST(oinfo));
1668         }
1669
1670         if (fid != NULL)
1671                 *fid = oinfo->lmo_fid;
1672         if (mds != NULL)
1673                 *mds = oinfo->lmo_mds;
1674
1675         tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1676
1677         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", oinfo->lmo_mds,
1678                PFID(&oinfo->lmo_fid));
1679         return tgt;
1680 }
1681
1682 /**
1683  * Locate mds by fid or name
1684  *
1685  * For striped directory (lsm != NULL), it will locate the stripe
1686  * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1687  * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1688  * walk through all of stripes to locate the entry.
1689  *
1690  * For normal direcotry, it will locate MDS by FID directly.
1691  * \param[in] lmv       LMV device
1692  * \param[in] op_data   client MD stack parameters, name, namelen
1693  *                      mds_num etc.
1694  * \param[in] fid       object FID used to locate MDS.
1695  *
1696  * retval               pointer to the lmv_tgt_desc if succeed.
1697  *                      ERR_PTR(errno) if failed.
1698  */
1699 struct lmv_tgt_desc*
1700 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1701                struct lu_fid *fid)
1702 {
1703         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1704         struct lmv_tgt_desc     *tgt;
1705
1706         /* During creating VOLATILE file, it should honor the mdt
1707          * index if the file under striped dir is being restored, see
1708          * ct_restore(). */
1709         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1710             (int)op_data->op_mds != -1) {
1711                 int i;
1712                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1713                 if (IS_ERR(tgt))
1714                         return tgt;
1715
1716                 if (lsm != NULL) {
1717                         /* refill the right parent fid */
1718                         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1719                                 struct lmv_oinfo *oinfo;
1720
1721                                 oinfo = &lsm->lsm_md_oinfo[i];
1722                                 if (oinfo->lmo_mds == op_data->op_mds) {
1723                                         *fid = oinfo->lmo_fid;
1724                                         break;
1725                                 }
1726                         }
1727
1728                         if (i == lsm->lsm_md_stripe_count)
1729                                 *fid = lsm->lsm_md_oinfo[0].lmo_fid;
1730                 }
1731
1732                 return tgt;
1733         }
1734
1735         if (lsm == NULL || op_data->op_namelen == 0) {
1736                 tgt = lmv_find_target(lmv, fid);
1737                 if (IS_ERR(tgt))
1738                         return tgt;
1739
1740                 op_data->op_mds = tgt->ltd_idx;
1741                 return tgt;
1742         }
1743
1744         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1745                                           op_data->op_namelen, fid,
1746                                           &op_data->op_mds);
1747 }
1748
1749 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1750                 const void *data, size_t datalen, umode_t mode, uid_t uid,
1751                 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1752                 struct ptlrpc_request **request)
1753 {
1754         struct obd_device       *obd = exp->exp_obd;
1755         struct lmv_obd          *lmv = &obd->u.lmv;
1756         struct lmv_tgt_desc     *tgt;
1757         int                      rc;
1758         ENTRY;
1759
1760         rc = lmv_check_connect(obd);
1761         if (rc)
1762                 RETURN(rc);
1763
1764         if (!lmv->desc.ld_active_tgt_count)
1765                 RETURN(-EIO);
1766
1767         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1768         if (IS_ERR(tgt))
1769                 RETURN(PTR_ERR(tgt));
1770
1771         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1772                 (int)op_data->op_namelen, op_data->op_name,
1773                 PFID(&op_data->op_fid1), op_data->op_mds);
1774
1775         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1776         if (rc)
1777                 RETURN(rc);
1778         if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1779                 /* Send the create request to the MDT where the object
1780                  * will be located */
1781                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1782                 if (IS_ERR(tgt))
1783                         RETURN(PTR_ERR(tgt));
1784
1785                 op_data->op_mds = tgt->ltd_idx;
1786         } else {
1787                 CDEBUG(D_CONFIG, "Server doesn't support striped dirs\n");
1788         }
1789
1790         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1791                PFID(&op_data->op_fid2), op_data->op_mds);
1792
1793         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1794         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1795                        cap_effective, rdev, request);
1796         if (rc == 0) {
1797                 if (*request == NULL)
1798                         RETURN(rc);
1799                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1800         }
1801         RETURN(rc);
1802 }
1803
1804 static int
1805 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1806             const union ldlm_policy_data *policy,
1807             struct lookup_intent *it, struct md_op_data *op_data,
1808             struct lustre_handle *lockh, __u64 extra_lock_flags)
1809 {
1810         struct obd_device        *obd = exp->exp_obd;
1811         struct lmv_obd           *lmv = &obd->u.lmv;
1812         struct lmv_tgt_desc      *tgt;
1813         int                       rc;
1814         ENTRY;
1815
1816         rc = lmv_check_connect(obd);
1817         if (rc)
1818                 RETURN(rc);
1819
1820         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1821                LL_IT2STR(it), PFID(&op_data->op_fid1));
1822
1823         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1824         if (IS_ERR(tgt))
1825                 RETURN(PTR_ERR(tgt));
1826
1827         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1828                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1829
1830         rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1831                         extra_lock_flags);
1832
1833         RETURN(rc);
1834 }
1835
1836 static int
1837 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1838                  struct ptlrpc_request **preq)
1839 {
1840         struct ptlrpc_request   *req = NULL;
1841         struct obd_device       *obd = exp->exp_obd;
1842         struct lmv_obd          *lmv = &obd->u.lmv;
1843         struct lmv_tgt_desc     *tgt;
1844         struct mdt_body         *body;
1845         int                      rc;
1846         ENTRY;
1847
1848         rc = lmv_check_connect(obd);
1849         if (rc)
1850                 RETURN(rc);
1851
1852         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1853         if (IS_ERR(tgt))
1854                 RETURN(PTR_ERR(tgt));
1855
1856         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1857                 (int)op_data->op_namelen, op_data->op_name,
1858                 PFID(&op_data->op_fid1), tgt->ltd_idx);
1859
1860         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1861         if (rc != 0)
1862                 RETURN(rc);
1863
1864         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1865         LASSERT(body != NULL);
1866
1867         if (body->mbo_valid & OBD_MD_MDS) {
1868                 struct lu_fid rid = body->mbo_fid1;
1869                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1870                        PFID(&rid));
1871
1872                 tgt = lmv_find_target(lmv, &rid);
1873                 if (IS_ERR(tgt)) {
1874                         ptlrpc_req_finished(*preq);
1875                         preq = NULL;
1876                         RETURN(PTR_ERR(tgt));
1877                 }
1878
1879                 op_data->op_fid1 = rid;
1880                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1881                 op_data->op_namelen = 0;
1882                 op_data->op_name = NULL;
1883                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1884                 ptlrpc_req_finished(*preq);
1885                 *preq = req;
1886         }
1887
1888         RETURN(rc);
1889 }
1890
1891 #define md_op_data_fid(op_data, fl)                     \
1892         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1893          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1894          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1895          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1896          NULL)
1897
1898 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1899                             struct md_op_data *op_data, __u32 op_tgt,
1900                             enum ldlm_mode mode, int bits, int flag)
1901 {
1902         struct lu_fid *fid = md_op_data_fid(op_data, flag);
1903         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
1904         union ldlm_policy_data policy = { { 0 } };
1905         int rc = 0;
1906         ENTRY;
1907
1908         if (!fid_is_sane(fid))
1909                 RETURN(0);
1910
1911         if (tgt == NULL) {
1912                 tgt = lmv_find_target(lmv, fid);
1913                 if (IS_ERR(tgt))
1914                         RETURN(PTR_ERR(tgt));
1915         }
1916
1917         if (tgt->ltd_idx != op_tgt) {
1918                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1919                 policy.l_inodebits.bits = bits;
1920                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1921                                       mode, LCF_ASYNC, NULL);
1922         } else {
1923                 CDEBUG(D_INODE,
1924                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1925                        op_tgt, PFID(fid));
1926                 op_data->op_flags |= flag;
1927                 rc = 0;
1928         }
1929
1930         RETURN(rc);
1931 }
1932
1933 /*
1934  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1935  * op_data->op_fid2
1936  */
1937 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1938                     struct ptlrpc_request **request)
1939 {
1940         struct obd_device       *obd = exp->exp_obd;
1941         struct lmv_obd          *lmv = &obd->u.lmv;
1942         struct lmv_tgt_desc     *tgt;
1943         int                      rc;
1944         ENTRY;
1945
1946         rc = lmv_check_connect(obd);
1947         if (rc)
1948                 RETURN(rc);
1949
1950         LASSERT(op_data->op_namelen != 0);
1951
1952         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1953                PFID(&op_data->op_fid2), (int)op_data->op_namelen,
1954                op_data->op_name, PFID(&op_data->op_fid1));
1955
1956         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
1957         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
1958         op_data->op_cap = cfs_curproc_cap_pack();
1959         if (op_data->op_mea2 != NULL) {
1960                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
1961                 const struct lmv_oinfo  *oinfo;
1962
1963                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
1964                                                 op_data->op_namelen);
1965                 if (IS_ERR(oinfo))
1966                         RETURN(PTR_ERR(oinfo));
1967
1968                 op_data->op_fid2 = oinfo->lmo_fid;
1969         }
1970
1971         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1972         if (IS_ERR(tgt))
1973                 RETURN(PTR_ERR(tgt));
1974
1975         /*
1976          * Cancel UPDATE lock on child (fid1).
1977          */
1978         op_data->op_flags |= MF_MDC_CANCEL_FID2;
1979         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
1980                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
1981         if (rc != 0)
1982                 RETURN(rc);
1983
1984         rc = md_link(tgt->ltd_exp, op_data, request);
1985
1986         RETURN(rc);
1987 }
1988
1989 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
1990                       const char *old, size_t oldlen,
1991                       const char *new, size_t newlen,
1992                       struct ptlrpc_request **request)
1993 {
1994         struct obd_device       *obd = exp->exp_obd;
1995         struct lmv_obd          *lmv = &obd->u.lmv;
1996         struct lmv_tgt_desc     *src_tgt;
1997         struct lmv_tgt_desc     *tgt_tgt;
1998         struct obd_export       *target_exp;
1999         struct mdt_body         *body;
2000         int                     rc;
2001         ENTRY;
2002
2003         LASSERT(oldlen != 0);
2004
2005         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2006                (int)oldlen, old, PFID(&op_data->op_fid1),
2007                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2008                (int)newlen, new, PFID(&op_data->op_fid2),
2009                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2010
2011         rc = lmv_check_connect(obd);
2012         if (rc)
2013                 RETURN(rc);
2014
2015         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2016         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2017         op_data->op_cap = cfs_curproc_cap_pack();
2018         if (op_data->op_cli_flags & CLI_MIGRATE) {
2019                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2020                          PFID(&op_data->op_fid3));
2021
2022                 if (op_data->op_mea1 != NULL) {
2023                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2024                         struct lmv_tgt_desc     *tmp;
2025
2026                         /* Fix the parent fid for striped dir */
2027                         tmp = lmv_locate_target_for_name(lmv, lsm, old,
2028                                                          oldlen,
2029                                                          &op_data->op_fid1,
2030                                                          NULL);
2031                         if (IS_ERR(tmp))
2032                                 RETURN(PTR_ERR(tmp));
2033                 }
2034
2035                 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2036                 if (rc != 0)
2037                         RETURN(rc);
2038
2039                 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2040                 if (IS_ERR(src_tgt))
2041                         RETURN(PTR_ERR(src_tgt));
2042
2043                 target_exp = src_tgt->ltd_exp;
2044         } else {
2045                 if (op_data->op_mea1 != NULL) {
2046                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2047
2048                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2049                                                              oldlen,
2050                                                              &op_data->op_fid1,
2051                                                              &op_data->op_mds);
2052                 } else {
2053                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2054                 }
2055                 if (IS_ERR(src_tgt))
2056                         RETURN(PTR_ERR(src_tgt));
2057
2058
2059                 if (op_data->op_mea2 != NULL) {
2060                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
2061
2062                         tgt_tgt = lmv_locate_target_for_name(lmv, lsm, new,
2063                                                              newlen,
2064                                                              &op_data->op_fid2,
2065                                                              &op_data->op_mds);
2066                 } else {
2067                         tgt_tgt = lmv_find_target(lmv, &op_data->op_fid2);
2068
2069                 }
2070                 if (IS_ERR(tgt_tgt))
2071                         RETURN(PTR_ERR(tgt_tgt));
2072
2073                 target_exp = tgt_tgt->ltd_exp;
2074         }
2075
2076         /*
2077          * LOOKUP lock on src child (fid3) should also be cancelled for
2078          * src_tgt in mdc_rename.
2079          */
2080         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2081
2082         /*
2083          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2084          * own target.
2085          */
2086         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2087                               LCK_EX, MDS_INODELOCK_UPDATE,
2088                               MF_MDC_CANCEL_FID2);
2089
2090         if (rc != 0)
2091                 RETURN(rc);
2092         /*
2093          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2094          */
2095         if (fid_is_sane(&op_data->op_fid3)) {
2096                 struct lmv_tgt_desc *tgt;
2097
2098                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2099                 if (IS_ERR(tgt))
2100                         RETURN(PTR_ERR(tgt));
2101
2102                 /* Cancel LOOKUP lock on its parent */
2103                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2104                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2105                                       MF_MDC_CANCEL_FID3);
2106                 if (rc != 0)
2107                         RETURN(rc);
2108
2109                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2110                                       LCK_EX, MDS_INODELOCK_FULL,
2111                                       MF_MDC_CANCEL_FID3);
2112                 if (rc != 0)
2113                         RETURN(rc);
2114         }
2115
2116 retry_rename:
2117         /*
2118          * Cancel all the locks on tgt child (fid4).
2119          */
2120         if (fid_is_sane(&op_data->op_fid4)) {
2121                 struct lmv_tgt_desc *tgt;
2122
2123                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2124                                       LCK_EX, MDS_INODELOCK_FULL,
2125                                       MF_MDC_CANCEL_FID4);
2126                 if (rc != 0)
2127                         RETURN(rc);
2128
2129                 tgt = lmv_find_target(lmv, &op_data->op_fid4);
2130                 if (IS_ERR(tgt))
2131                         RETURN(PTR_ERR(tgt));
2132
2133                 /* Since the target child might be destroyed, and it might
2134                  * become orphan, and we can only check orphan on the local
2135                  * MDT right now, so we send rename request to the MDT where
2136                  * target child is located. If target child does not exist,
2137                  * then it will send the request to the target parent */
2138                 target_exp = tgt->ltd_exp;
2139         }
2140
2141         rc = md_rename(target_exp, op_data, old, oldlen, new, newlen,
2142                        request);
2143
2144         if (rc != 0 && rc != -EXDEV)
2145                 RETURN(rc);
2146
2147         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2148         if (body == NULL)
2149                 RETURN(-EPROTO);
2150
2151         /* Not cross-ref case, just get out of here. */
2152         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2153                 RETURN(rc);
2154
2155         CDEBUG(D_INODE, "%s: try rename to another MDT for "DFID"\n",
2156                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2157
2158         op_data->op_fid4 = body->mbo_fid1;
2159         ptlrpc_req_finished(*request);
2160         *request = NULL;
2161         goto retry_rename;
2162 }
2163
2164 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2165                        void *ea, size_t ealen, struct ptlrpc_request **request)
2166 {
2167         struct obd_device       *obd = exp->exp_obd;
2168         struct lmv_obd          *lmv = &obd->u.lmv;
2169         struct lmv_tgt_desc     *tgt;
2170         int                      rc = 0;
2171         ENTRY;
2172
2173         rc = lmv_check_connect(obd);
2174         if (rc)
2175                 RETURN(rc);
2176
2177         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2178                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2179
2180         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2181         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2182         if (IS_ERR(tgt))
2183                 RETURN(PTR_ERR(tgt));
2184
2185         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2186
2187         RETURN(rc);
2188 }
2189
2190 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2191                      struct ptlrpc_request **request)
2192 {
2193         struct obd_device       *obd = exp->exp_obd;
2194         struct lmv_obd          *lmv = &obd->u.lmv;
2195         struct lmv_tgt_desc     *tgt;
2196         int                      rc;
2197         ENTRY;
2198
2199         rc = lmv_check_connect(obd);
2200         if (rc != 0)
2201                 RETURN(rc);
2202
2203         tgt = lmv_find_target(lmv, fid);
2204         if (IS_ERR(tgt))
2205                 RETURN(PTR_ERR(tgt));
2206
2207         rc = md_fsync(tgt->ltd_exp, fid, request);
2208         RETURN(rc);
2209 }
2210
2211 /**
2212  * Get current minimum entry from striped directory
2213  *
2214  * This function will search the dir entry, whose hash value is the
2215  * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2216  * only being called for striped directory.
2217  *
2218  * \param[in] exp               export of LMV
2219  * \param[in] op_data           parameters transferred beween client MD stack
2220  *                              stripe_information will be included in this
2221  *                              parameter
2222  * \param[in] cb_op             ldlm callback being used in enqueue in
2223  *                              mdc_read_page
2224  * \param[in] hash_offset       the hash value, which is used to locate
2225  *                              minum(closet) dir entry
2226  * \param[in|out] stripe_offset the caller use this to indicate the stripe
2227  *                              index of last entry, so to avoid hash conflict
2228  *                              between stripes. It will also be used to
2229  *                              return the stripe index of current dir entry.
2230  * \param[in|out] entp          the minum entry and it also is being used
2231  *                              to input the last dir entry to resolve the
2232  *                              hash conflict
2233  *
2234  * \param[out] ppage            the page which holds the minum entry
2235  *
2236  * \retval                      = 0 get the entry successfully
2237  *                              negative errno (< 0) does not get the entry
2238  */
2239 static int lmv_get_min_striped_entry(struct obd_export *exp,
2240                                      struct md_op_data *op_data,
2241                                      struct md_callback *cb_op,
2242                                      __u64 hash_offset, int *stripe_offset,
2243                                      struct lu_dirent **entp,
2244                                      struct page **ppage)
2245 {
2246         struct obd_device       *obd = exp->exp_obd;
2247         struct lmv_obd          *lmv = &obd->u.lmv;
2248         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2249         struct lmv_tgt_desc     *tgt;
2250         int                     stripe_count;
2251         struct lu_dirent        *min_ent = NULL;
2252         struct page             *min_page = NULL;
2253         int                     min_idx = 0;
2254         int                     i;
2255         int                     rc = 0;
2256         ENTRY;
2257
2258         stripe_count = lsm->lsm_md_stripe_count;
2259         for (i = 0; i < stripe_count; i++) {
2260                 struct lu_dirent        *ent = NULL;
2261                 struct page             *page = NULL;
2262                 struct lu_dirpage       *dp;
2263                 __u64                   stripe_hash = hash_offset;
2264
2265                 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2266                 if (IS_ERR(tgt))
2267                         GOTO(out, rc = PTR_ERR(tgt));
2268
2269                 /* op_data will be shared by each stripe, so we need
2270                  * reset these value for each stripe */
2271                 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2272                 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2273                 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2274 next:
2275                 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2276                                   &page);
2277                 if (rc != 0)
2278                         GOTO(out, rc);
2279
2280                 dp = page_address(page);
2281                 for (ent = lu_dirent_start(dp); ent != NULL;
2282                      ent = lu_dirent_next(ent)) {
2283                         /* Skip dummy entry */
2284                         if (le16_to_cpu(ent->lde_namelen) == 0)
2285                                 continue;
2286
2287                         if (le64_to_cpu(ent->lde_hash) < hash_offset)
2288                                 continue;
2289
2290                         if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2291                             (*entp == ent || i < *stripe_offset))
2292                                 continue;
2293
2294                         /* skip . and .. for other stripes */
2295                         if (i != 0 &&
2296                             (strncmp(ent->lde_name, ".",
2297                                      le16_to_cpu(ent->lde_namelen)) == 0 ||
2298                              strncmp(ent->lde_name, "..",
2299                                      le16_to_cpu(ent->lde_namelen)) == 0))
2300                                 continue;
2301                         break;
2302                 }
2303
2304                 if (ent == NULL) {
2305                         stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2306
2307                         kunmap(page);
2308                         page_cache_release(page);
2309                         page = NULL;
2310
2311                         /* reach the end of current stripe, go to next stripe */
2312                         if (stripe_hash == MDS_DIR_END_OFF)
2313                                 continue;
2314                         else
2315                                 goto next;
2316                 }
2317
2318                 if (min_ent != NULL) {
2319                         if (le64_to_cpu(min_ent->lde_hash) >
2320                             le64_to_cpu(ent->lde_hash)) {
2321                                 min_ent = ent;
2322                                 kunmap(min_page);
2323                                 page_cache_release(min_page);
2324                                 min_idx = i;
2325                                 min_page = page;
2326                         } else {
2327                                 kunmap(page);
2328                                 page_cache_release(page);
2329                                 page = NULL;
2330                         }
2331                 } else {
2332                         min_ent = ent;
2333                         min_page = page;
2334                         min_idx = i;
2335                 }
2336         }
2337
2338 out:
2339         if (*ppage != NULL) {
2340                 kunmap(*ppage);
2341                 page_cache_release(*ppage);
2342         }
2343         *stripe_offset = min_idx;
2344         *entp = min_ent;
2345         *ppage = min_page;
2346         RETURN(rc);
2347 }
2348
2349 /**
2350  * Build dir entry page from a striped directory
2351  *
2352  * This function gets one entry by @offset from a striped directory. It will
2353  * read entries from all of stripes, and choose one closest to the required
2354  * offset(&offset). A few notes
2355  * 1. skip . and .. for non-zero stripes, because there can only have one .
2356  * and .. in a directory.
2357  * 2. op_data will be shared by all of stripes, instead of allocating new
2358  * one, so need to restore before reusing.
2359  * 3. release the entry page if that is not being chosen.
2360  *
2361  * \param[in] exp       obd export refer to LMV
2362  * \param[in] op_data   hold those MD parameters of read_entry
2363  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2364  * \param[out] ldp      the entry being read
2365  * \param[out] ppage    the page holding the entry. Note: because the entry
2366  *                      will be accessed in upper layer, so we need hold the
2367  *                      page until the usages of entry is finished, see
2368  *                      ll_dir_entry_next.
2369  *
2370  * retval               =0 if get entry successfully
2371  *                      <0 cannot get entry
2372  */
2373 static int lmv_read_striped_page(struct obd_export *exp,
2374                                  struct md_op_data *op_data,
2375                                  struct md_callback *cb_op,
2376                                  __u64 offset, struct page **ppage)
2377 {
2378         struct obd_device       *obd = exp->exp_obd;
2379         struct lu_fid           master_fid = op_data->op_fid1;
2380         struct inode            *master_inode = op_data->op_data;
2381         __u64                   hash_offset = offset;
2382         struct lu_dirpage       *dp;
2383         struct page             *min_ent_page = NULL;
2384         struct page             *ent_page = NULL;
2385         struct lu_dirent        *ent;
2386         void                    *area;
2387         int                     ent_idx = 0;
2388         struct lu_dirent        *min_ent = NULL;
2389         struct lu_dirent        *last_ent;
2390         size_t                  left_bytes;
2391         int                     rc;
2392         ENTRY;
2393
2394         rc = lmv_check_connect(obd);
2395         if (rc)
2396                 RETURN(rc);
2397
2398         /* Allocate a page and read entries from all of stripes and fill
2399          * the page by hash order */
2400         ent_page = alloc_page(GFP_KERNEL);
2401         if (ent_page == NULL)
2402                 RETURN(-ENOMEM);
2403
2404         /* Initialize the entry page */
2405         dp = kmap(ent_page);
2406         memset(dp, 0, sizeof(*dp));
2407         dp->ldp_hash_start = cpu_to_le64(offset);
2408         dp->ldp_flags |= LDF_COLLIDE;
2409
2410         area = dp + 1;
2411         left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2412         ent = area;
2413         last_ent = ent;
2414         do {
2415                 __u16   ent_size;
2416
2417                 /* Find the minum entry from all sub-stripes */
2418                 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2419                                                &ent_idx, &min_ent,
2420                                                &min_ent_page);
2421                 if (rc != 0)
2422                         GOTO(out, rc);
2423
2424                 /* If it can not get minum entry, it means it already reaches
2425                  * the end of this directory */
2426                 if (min_ent == NULL) {
2427                         last_ent->lde_reclen = 0;
2428                         hash_offset = MDS_DIR_END_OFF;
2429                         GOTO(out, rc);
2430                 }
2431
2432                 ent_size = le16_to_cpu(min_ent->lde_reclen);
2433
2434                 /* the last entry lde_reclen is 0, but it might not
2435                  * the end of this entry of this temporay entry */
2436                 if (ent_size == 0)
2437                         ent_size = lu_dirent_calc_size(
2438                                         le16_to_cpu(min_ent->lde_namelen),
2439                                         le32_to_cpu(min_ent->lde_attrs));
2440                 if (ent_size > left_bytes) {
2441                         last_ent->lde_reclen = cpu_to_le16(0);
2442                         hash_offset = le64_to_cpu(min_ent->lde_hash);
2443                         GOTO(out, rc);
2444                 }
2445
2446                 memcpy(ent, min_ent, ent_size);
2447
2448                 /* Replace . with master FID and Replace .. with the parent FID
2449                  * of master object */
2450                 if (strncmp(ent->lde_name, ".",
2451                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2452                     le16_to_cpu(ent->lde_namelen) == 1)
2453                         fid_cpu_to_le(&ent->lde_fid, &master_fid);
2454                 else if (strncmp(ent->lde_name, "..",
2455                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2456                            le16_to_cpu(ent->lde_namelen) == 2)
2457                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2458
2459                 left_bytes -= ent_size;
2460                 ent->lde_reclen = cpu_to_le16(ent_size);
2461                 last_ent = ent;
2462                 ent = (void *)ent + ent_size;
2463                 hash_offset = le64_to_cpu(min_ent->lde_hash);
2464                 if (hash_offset == MDS_DIR_END_OFF) {
2465                         last_ent->lde_reclen = 0;
2466                         break;
2467                 }
2468         } while (1);
2469 out:
2470         if (min_ent_page != NULL) {
2471                 kunmap(min_ent_page);
2472                 page_cache_release(min_ent_page);
2473         }
2474
2475         if (unlikely(rc != 0)) {
2476                 __free_page(ent_page);
2477                 ent_page = NULL;
2478         } else {
2479                 if (ent == area)
2480                         dp->ldp_flags |= LDF_EMPTY;
2481                 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2482                 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2483         }
2484
2485         /* We do not want to allocate md_op_data during each
2486          * dir entry reading, so op_data will be shared by every stripe,
2487          * then we need to restore it back to original value before
2488          * return to the upper layer */
2489         op_data->op_fid1 = master_fid;
2490         op_data->op_fid2 = master_fid;
2491         op_data->op_data = master_inode;
2492
2493         *ppage = ent_page;
2494
2495         RETURN(rc);
2496 }
2497
2498 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2499                   struct md_callback *cb_op, __u64 offset,
2500                   struct page **ppage)
2501 {
2502         struct obd_device       *obd = exp->exp_obd;
2503         struct lmv_obd          *lmv = &obd->u.lmv;
2504         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2505         struct lmv_tgt_desc     *tgt;
2506         int                     rc;
2507         ENTRY;
2508
2509         rc = lmv_check_connect(obd);
2510         if (rc != 0)
2511                 RETURN(rc);
2512
2513         if (unlikely(lsm != NULL)) {
2514                 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2515                 RETURN(rc);
2516         }
2517
2518         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2519         if (IS_ERR(tgt))
2520                 RETURN(PTR_ERR(tgt));
2521
2522         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2523
2524         RETURN(rc);
2525 }
2526
2527 /**
2528  * Unlink a file/directory
2529  *
2530  * Unlink a file or directory under the parent dir. The unlink request
2531  * usually will be sent to the MDT where the child is located, but if
2532  * the client does not have the child FID then request will be sent to the
2533  * MDT where the parent is located.
2534  *
2535  * If the parent is a striped directory then it also needs to locate which
2536  * stripe the name of the child is located, and replace the parent FID
2537  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2538  * it will walk through all of sub-stripes until the child is being
2539  * unlinked finally.
2540  *
2541  * \param[in] exp       export refer to LMV
2542  * \param[in] op_data   different parameters transferred beween client
2543  *                      MD stacks, name, namelen, FIDs etc.
2544  *                      op_fid1 is the parent FID, op_fid2 is the child
2545  *                      FID.
2546  * \param[out] request  point to the request of unlink.
2547  *
2548  * retval               0 if succeed
2549  *                      negative errno if failed.
2550  */
2551 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2552                       struct ptlrpc_request **request)
2553 {
2554         struct obd_device       *obd = exp->exp_obd;
2555         struct lmv_obd          *lmv = &obd->u.lmv;
2556         struct lmv_tgt_desc     *tgt = NULL;
2557         struct lmv_tgt_desc     *parent_tgt = NULL;
2558         struct mdt_body         *body;
2559         int                     rc;
2560         int                     stripe_index = 0;
2561         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2562         ENTRY;
2563
2564         rc = lmv_check_connect(obd);
2565         if (rc)
2566                 RETURN(rc);
2567 retry_unlink:
2568         /* For striped dir, we need to locate the parent as well */
2569         if (lsm != NULL) {
2570                 struct lmv_tgt_desc *tmp;
2571
2572                 LASSERT(op_data->op_name != NULL &&
2573                         op_data->op_namelen != 0);
2574
2575                 tmp = lmv_locate_target_for_name(lmv, lsm,
2576                                                  op_data->op_name,
2577                                                  op_data->op_namelen,
2578                                                  &op_data->op_fid1,
2579                                                  &op_data->op_mds);
2580
2581                 /* return -EBADFD means unknown hash type, might
2582                  * need try all sub-stripe here */
2583                 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2584                         RETURN(PTR_ERR(tmp));
2585
2586                 /* Note: both migrating dir and unknown hash dir need to
2587                  * try all of sub-stripes, so we need start search the
2588                  * name from stripe 0, but migrating dir is already handled
2589                  * inside lmv_locate_target_for_name(), so we only check
2590                  * unknown hash type directory here */
2591                 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2592                         struct lmv_oinfo *oinfo;
2593
2594                         oinfo = &lsm->lsm_md_oinfo[stripe_index];
2595
2596                         op_data->op_fid1 = oinfo->lmo_fid;
2597                         op_data->op_mds = oinfo->lmo_mds;
2598                 }
2599         }
2600
2601 try_next_stripe:
2602         /* Send unlink requests to the MDT where the child is located */
2603         if (likely(!fid_is_zero(&op_data->op_fid2)))
2604                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2605         else if (lsm != NULL)
2606                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2607         else
2608                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2609
2610         if (IS_ERR(tgt))
2611                 RETURN(PTR_ERR(tgt));
2612
2613         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2614         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2615         op_data->op_cap = cfs_curproc_cap_pack();
2616
2617         /*
2618          * If child's fid is given, cancel unused locks for it if it is from
2619          * another export than parent.
2620          *
2621          * LOOKUP lock for child (fid3) should also be cancelled on parent
2622          * tgt_tgt in mdc_unlink().
2623          */
2624         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2625
2626         /*
2627          * Cancel FULL locks on child (fid3).
2628          */
2629         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2630         if (IS_ERR(parent_tgt))
2631                 RETURN(PTR_ERR(parent_tgt));
2632
2633         if (parent_tgt != tgt) {
2634                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2635                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2636                                       MF_MDC_CANCEL_FID3);
2637         }
2638
2639         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2640                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2641         if (rc != 0)
2642                 RETURN(rc);
2643
2644         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2645                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2646
2647         rc = md_unlink(tgt->ltd_exp, op_data, request);
2648         if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2649                 RETURN(rc);
2650
2651         /* Try next stripe if it is needed. */
2652         if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2653                 struct lmv_oinfo *oinfo;
2654
2655                 stripe_index++;
2656                 if (stripe_index >= lsm->lsm_md_stripe_count)
2657                         RETURN(rc);
2658
2659                 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2660
2661                 op_data->op_fid1 = oinfo->lmo_fid;
2662                 op_data->op_mds = oinfo->lmo_mds;
2663
2664                 ptlrpc_req_finished(*request);
2665                 *request = NULL;
2666
2667                 goto try_next_stripe;
2668         }
2669
2670         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2671         if (body == NULL)
2672                 RETURN(-EPROTO);
2673
2674         /* Not cross-ref case, just get out of here. */
2675         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2676                 RETURN(rc);
2677
2678         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2679                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2680
2681         /* This is a remote object, try remote MDT, Note: it may
2682          * try more than 1 time here, Considering following case
2683          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2684          * 1. Initially A does not know where remote1 is, it send
2685          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2686          *    resend unlink RPC to MDT1 (retry 1st time).
2687          *
2688          * 2. During the unlink RPC in flight,
2689          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2690          *    and create new remote1, but on MDT0
2691          *
2692          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2693          *    /mnt/lustre, then lookup get fid of remote1, and find
2694          *    it is remote dir again, and replay -EREMOTE again.
2695          *
2696          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2697          *
2698          * In theory, it might try unlimited time here, but it should
2699          * be very rare case.  */
2700         op_data->op_fid2 = body->mbo_fid1;
2701         ptlrpc_req_finished(*request);
2702         *request = NULL;
2703
2704         goto retry_unlink;
2705 }
2706
2707 static int lmv_precleanup(struct obd_device *obd)
2708 {
2709         ENTRY;
2710         fld_client_proc_fini(&obd->u.lmv.lmv_fld);
2711         lprocfs_obd_cleanup(obd);
2712         lprocfs_free_md_stats(obd);
2713         RETURN(0);
2714 }
2715
2716 /**
2717  * Get by key a value associated with a LMV device.
2718  *
2719  * Dispatch request to lower-layer devices as needed.
2720  *
2721  * \param[in] env               execution environment for this thread
2722  * \param[in] exp               export for the LMV device
2723  * \param[in] keylen            length of key identifier
2724  * \param[in] key               identifier of key to get value for
2725  * \param[in] vallen            size of \a val
2726  * \param[out] val              pointer to storage location for value
2727  * \param[in] lsm               optional striping metadata of object
2728  *
2729  * \retval 0            on success
2730  * \retval negative     negated errno on failure
2731  */
2732 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2733                         __u32 keylen, void *key, __u32 *vallen, void *val)
2734 {
2735         struct obd_device       *obd;
2736         struct lmv_obd          *lmv;
2737         int                      rc = 0;
2738         ENTRY;
2739
2740         obd = class_exp2obd(exp);
2741         if (obd == NULL) {
2742                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2743                        exp->exp_handle.h_cookie);
2744                 RETURN(-EINVAL);
2745         }
2746
2747         lmv = &obd->u.lmv;
2748         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2749                 int i;
2750
2751                 rc = lmv_check_connect(obd);
2752                 if (rc)
2753                         RETURN(rc);
2754
2755                 LASSERT(*vallen == sizeof(__u32));
2756                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2757                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2758                         /*
2759                          * All tgts should be connected when this gets called.
2760                          */
2761                         if (tgt == NULL || tgt->ltd_exp == NULL)
2762                                 continue;
2763
2764                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2765                                           vallen, val))
2766                                 RETURN(0);
2767                 }
2768                 RETURN(-EINVAL);
2769         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2770                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2771                    KEY_IS(KEY_CONN_DATA)) {
2772                 rc = lmv_check_connect(obd);
2773                 if (rc)
2774                         RETURN(rc);
2775
2776                 /*
2777                  * Forwarding this request to first MDS, it should know LOV
2778                  * desc.
2779                  */
2780                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2781                                   vallen, val);
2782                 if (!rc && KEY_IS(KEY_CONN_DATA))
2783                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2784                 RETURN(rc);
2785         } else if (KEY_IS(KEY_TGT_COUNT)) {
2786                 *((int *)val) = lmv->desc.ld_tgt_count;
2787                 RETURN(0);
2788         }
2789
2790         CDEBUG(D_IOCTL, "Invalid key\n");
2791         RETURN(-EINVAL);
2792 }
2793
2794 /**
2795  * Asynchronously set by key a value associated with a LMV device.
2796  *
2797  * Dispatch request to lower-layer devices as needed.
2798  *
2799  * \param[in] env       execution environment for this thread
2800  * \param[in] exp       export for the LMV device
2801  * \param[in] keylen    length of key identifier
2802  * \param[in] key       identifier of key to store value for
2803  * \param[in] vallen    size of value to store
2804  * \param[in] val       pointer to data to be stored
2805  * \param[in] set       optional list of related ptlrpc requests
2806  *
2807  * \retval 0            on success
2808  * \retval negative     negated errno on failure
2809  */
2810 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2811                         __u32 keylen, void *key, __u32 vallen, void *val,
2812                         struct ptlrpc_request_set *set)
2813 {
2814         struct lmv_tgt_desc     *tgt = NULL;
2815         struct obd_device       *obd;
2816         struct lmv_obd          *lmv;
2817         int rc = 0;
2818         ENTRY;
2819
2820         obd = class_exp2obd(exp);
2821         if (obd == NULL) {
2822                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2823                        exp->exp_handle.h_cookie);
2824                 RETURN(-EINVAL);
2825         }
2826         lmv = &obd->u.lmv;
2827
2828         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2829             KEY_IS(KEY_DEFAULT_EASIZE)) {
2830                 int i, err = 0;
2831
2832                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2833                         tgt = lmv->tgts[i];
2834
2835                         if (tgt == NULL || tgt->ltd_exp == NULL)
2836                                 continue;
2837
2838                         err = obd_set_info_async(env, tgt->ltd_exp,
2839                                                  keylen, key, vallen, val, set);
2840                         if (err && rc == 0)
2841                                 rc = err;
2842                 }
2843
2844                 RETURN(rc);
2845         }
2846
2847         RETURN(-EINVAL);
2848 }
2849
2850 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2851                             const struct lmv_mds_md_v1 *lmm1)
2852 {
2853         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
2854         int             stripe_count;
2855         int             cplen;
2856         int             i;
2857         int             rc = 0;
2858         ENTRY;
2859
2860         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2861         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2862         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2863         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2864                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2865         else
2866                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2867         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2868         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2869                         sizeof(lsm->lsm_md_pool_name));
2870
2871         if (cplen >= sizeof(lsm->lsm_md_pool_name))
2872                 RETURN(-E2BIG);
2873
2874         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2875                "layout_version %d\n", lsm->lsm_md_stripe_count,
2876                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2877                lsm->lsm_md_layout_version);
2878
2879         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2880         for (i = 0; i < stripe_count; i++) {
2881                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2882                               &lmm1->lmv_stripe_fids[i]);
2883                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2884                                     &lsm->lsm_md_oinfo[i].lmo_mds);
2885                 if (rc != 0)
2886                         RETURN(rc);
2887                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2888                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2889         }
2890
2891         RETURN(rc);
2892 }
2893
2894 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2895                         const union lmv_mds_md *lmm, size_t lmm_size)
2896 {
2897         struct lmv_stripe_md     *lsm;
2898         int                      lsm_size;
2899         int                      rc;
2900         bool                     allocated = false;
2901         ENTRY;
2902
2903         LASSERT(lsmp != NULL);
2904
2905         lsm = *lsmp;
2906         /* Free memmd */
2907         if (lsm != NULL && lmm == NULL) {
2908                 int i;
2909                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2910                         /* For migrating inode, the master stripe and master
2911                          * object will be the same, so do not need iput, see
2912                          * ll_update_lsm_md */
2913                         if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2914                               i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2915                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
2916                 }
2917                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2918                 OBD_FREE(lsm, lsm_size);
2919                 *lsmp = NULL;
2920                 RETURN(0);
2921         }
2922
2923         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
2924                 RETURN(-EPERM);
2925
2926         /* Unpack memmd */
2927         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
2928             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
2929                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
2930                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
2931                        -EIO);
2932                 RETURN(-EIO);
2933         }
2934
2935         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
2936                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2937         else
2938                 /**
2939                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
2940                  * stripecount should be 0 then.
2941                  */
2942                 lsm_size = lmv_stripe_md_size(0);
2943
2944         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2945         if (lsm == NULL) {
2946                 OBD_ALLOC(lsm, lsm_size);
2947                 if (lsm == NULL)
2948                         RETURN(-ENOMEM);
2949                 allocated = true;
2950                 *lsmp = lsm;
2951         }
2952
2953         switch (le32_to_cpu(lmm->lmv_magic)) {
2954         case LMV_MAGIC_V1:
2955                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
2956                 break;
2957         default:
2958                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
2959                        le32_to_cpu(lmm->lmv_magic));
2960                 rc = -EINVAL;
2961                 break;
2962         }
2963
2964         if (rc != 0 && allocated) {
2965                 OBD_FREE(lsm, lsm_size);
2966                 *lsmp = NULL;
2967                 lsm_size = rc;
2968         }
2969         RETURN(lsm_size);
2970 }
2971
2972 void lmv_free_memmd(struct lmv_stripe_md *lsm)
2973 {
2974         lmv_unpackmd(NULL, &lsm, NULL, 0);
2975 }
2976 EXPORT_SYMBOL(lmv_free_memmd);
2977
2978 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2979                              union ldlm_policy_data *policy,
2980                              enum ldlm_mode mode, enum ldlm_cancel_flags flags,
2981                              void *opaque)
2982 {
2983         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2984         int rc = 0;
2985         __u32 i;
2986         ENTRY;
2987
2988         LASSERT(fid != NULL);
2989
2990         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2991                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2992                 int err;
2993
2994                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
2995                         continue;
2996
2997                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
2998                                        opaque);
2999                 if (!rc)
3000                         rc = err;
3001         }
3002         RETURN(rc);
3003 }
3004
3005 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3006                       __u64 *bits)
3007 {
3008         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3009         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3010         int                      rc;
3011         ENTRY;
3012
3013         if (tgt == NULL || tgt->ltd_exp == NULL)
3014                 RETURN(-EINVAL);
3015         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3016         RETURN(rc);
3017 }
3018
3019 enum ldlm_mode lmv_lock_match(struct obd_export *exp, __u64 flags,
3020                               const struct lu_fid *fid, enum ldlm_type type,
3021                               union ldlm_policy_data *policy,
3022                               enum ldlm_mode mode, struct lustre_handle *lockh)
3023 {
3024         struct obd_device       *obd = exp->exp_obd;
3025         struct lmv_obd          *lmv = &obd->u.lmv;
3026         enum ldlm_mode          rc;
3027         int                     tgt;
3028         int                     i;
3029         ENTRY;
3030
3031         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3032
3033         /*
3034          * With DNE every object can have two locks in different namespaces:
3035          * lookup lock in space of MDT storing direntry and update/open lock in
3036          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3037          * since this can be easily found, and only try others if that fails.
3038          */
3039         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3040              i < lmv->desc.ld_tgt_count;
3041              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3042                 if (tgt < 0) {
3043                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3044                                obd->obd_name, PFID(fid), tgt);
3045                         tgt = 0;
3046                 }
3047
3048                 if (lmv->tgts[tgt] == NULL ||
3049                     lmv->tgts[tgt]->ltd_exp == NULL ||
3050                     lmv->tgts[tgt]->ltd_active == 0)
3051                         continue;
3052
3053                 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3054                                    type, policy, mode, lockh);
3055                 if (rc)
3056                         RETURN(rc);
3057         }
3058
3059         RETURN(0);
3060 }
3061
3062 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3063                       struct obd_export *dt_exp, struct obd_export *md_exp,
3064                       struct lustre_md *md)
3065 {
3066         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3067         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3068
3069         if (tgt == NULL || tgt->ltd_exp == NULL)
3070                 RETURN(-EINVAL);
3071
3072         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3073 }
3074
3075 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3076 {
3077         struct obd_device       *obd = exp->exp_obd;
3078         struct lmv_obd          *lmv = &obd->u.lmv;
3079         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3080         ENTRY;
3081
3082         if (md->lmv != NULL) {
3083                 lmv_free_memmd(md->lmv);
3084                 md->lmv = NULL;
3085         }
3086         if (tgt == NULL || tgt->ltd_exp == NULL)
3087                 RETURN(-EINVAL);
3088         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3089 }
3090
3091 int lmv_set_open_replay_data(struct obd_export *exp,
3092                              struct obd_client_handle *och,
3093                              struct lookup_intent *it)
3094 {
3095         struct obd_device       *obd = exp->exp_obd;
3096         struct lmv_obd          *lmv = &obd->u.lmv;
3097         struct lmv_tgt_desc     *tgt;
3098         ENTRY;
3099
3100         tgt = lmv_find_target(lmv, &och->och_fid);
3101         if (IS_ERR(tgt))
3102                 RETURN(PTR_ERR(tgt));
3103
3104         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3105 }
3106
3107 int lmv_clear_open_replay_data(struct obd_export *exp,
3108                                struct obd_client_handle *och)
3109 {
3110         struct obd_device       *obd = exp->exp_obd;
3111         struct lmv_obd          *lmv = &obd->u.lmv;
3112         struct lmv_tgt_desc     *tgt;
3113         ENTRY;
3114
3115         tgt = lmv_find_target(lmv, &och->och_fid);
3116         if (IS_ERR(tgt))
3117                 RETURN(PTR_ERR(tgt));
3118
3119         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3120 }
3121
3122 static int lmv_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
3123                                u32 suppgid, struct ptlrpc_request **request)
3124 {
3125         struct obd_device       *obd = exp->exp_obd;
3126         struct lmv_obd          *lmv = &obd->u.lmv;
3127         struct lmv_tgt_desc     *tgt;
3128         int                      rc;
3129         ENTRY;
3130
3131         rc = lmv_check_connect(obd);
3132         if (rc)
3133                 RETURN(rc);
3134
3135         tgt = lmv_find_target(lmv, fid);
3136         if (IS_ERR(tgt))
3137                 RETURN(PTR_ERR(tgt));
3138
3139         rc = md_get_remote_perm(tgt->ltd_exp, fid, suppgid, request);
3140         RETURN(rc);
3141 }
3142
3143 int lmv_intent_getattr_async(struct obd_export *exp,
3144                              struct md_enqueue_info *minfo)
3145 {
3146         struct md_op_data       *op_data = &minfo->mi_data;
3147         struct obd_device       *obd = exp->exp_obd;
3148         struct lmv_obd          *lmv = &obd->u.lmv;
3149         struct lmv_tgt_desc     *ptgt = NULL;
3150         struct lmv_tgt_desc     *ctgt = NULL;
3151         int                      rc;
3152         ENTRY;
3153
3154         if (!fid_is_sane(&op_data->op_fid2))
3155                 RETURN(-EINVAL);
3156
3157         rc = lmv_check_connect(obd);
3158         if (rc)
3159                 RETURN(rc);
3160
3161         ptgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3162         if (IS_ERR(ptgt))
3163                 RETURN(PTR_ERR(ptgt));
3164
3165         ctgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
3166         if (IS_ERR(ctgt))
3167                 RETURN(PTR_ERR(ctgt));
3168
3169         /*
3170          * if child is on remote MDT, we need 2 async RPCs to fetch both LOOKUP
3171          * lock on parent, and UPDATE lock on child MDT, which makes all
3172          * complicated. Considering remote dir is rare case, and not supporting
3173          * it in statahead won't cause any issue, drop its support for now.
3174          */
3175         if (ptgt != ctgt)
3176                 RETURN(-ENOTSUPP);
3177
3178         rc = md_intent_getattr_async(ptgt->ltd_exp, minfo);
3179         RETURN(rc);
3180 }
3181
3182 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3183                         struct lu_fid *fid, __u64 *bits)
3184 {
3185         struct obd_device       *obd = exp->exp_obd;
3186         struct lmv_obd          *lmv = &obd->u.lmv;
3187         struct lmv_tgt_desc     *tgt;
3188         int                      rc;
3189         ENTRY;
3190
3191         rc = lmv_check_connect(obd);
3192         if (rc)
3193                 RETURN(rc);
3194
3195         tgt = lmv_find_target(lmv, fid);
3196         if (IS_ERR(tgt))
3197                 RETURN(PTR_ERR(tgt));
3198
3199         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3200         RETURN(rc);
3201 }
3202
3203 int lmv_get_fid_from_lsm(struct obd_export *exp,
3204                          const struct lmv_stripe_md *lsm,
3205                          const char *name, int namelen, struct lu_fid *fid)
3206 {
3207         const struct lmv_oinfo *oinfo;
3208
3209         LASSERT(lsm != NULL);
3210         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3211         if (IS_ERR(oinfo))
3212                 return PTR_ERR(oinfo);
3213
3214         *fid = oinfo->lmo_fid;
3215
3216         RETURN(0);
3217 }
3218
3219 /**
3220  * For lmv, only need to send request to master MDT, and the master MDT will
3221  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3222  * we directly fetch data from the slave MDTs.
3223  */
3224 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3225                  struct obd_quotactl *oqctl)
3226 {
3227         struct obd_device   *obd = class_exp2obd(exp);
3228         struct lmv_obd      *lmv = &obd->u.lmv;
3229         struct lmv_tgt_desc *tgt = lmv->tgts[0];
3230         int                  rc = 0;
3231         __u32                i;
3232         __u64                curspace, curinodes;
3233         ENTRY;
3234
3235         if (tgt == NULL ||
3236             tgt->ltd_exp == NULL ||
3237             !tgt->ltd_active ||
3238             lmv->desc.ld_tgt_count == 0) {
3239                 CERROR("master lmv inactive\n");
3240                 RETURN(-EIO);
3241         }
3242
3243         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3244                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3245                 RETURN(rc);
3246         }
3247
3248         curspace = curinodes = 0;
3249         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3250                 int err;
3251                 tgt = lmv->tgts[i];
3252
3253                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3254                         continue;
3255
3256                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3257                 if (err) {
3258                         CERROR("getquota on mdt %d failed. %d\n", i, err);
3259                         if (!rc)
3260                                 rc = err;
3261                 } else {
3262                         curspace += oqctl->qc_dqblk.dqb_curspace;
3263                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3264                 }
3265         }
3266         oqctl->qc_dqblk.dqb_curspace = curspace;
3267         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3268
3269         RETURN(rc);
3270 }
3271
3272 static int lmv_merge_attr(struct obd_export *exp,
3273                           const struct lmv_stripe_md *lsm,
3274                           struct cl_attr *attr,
3275                           ldlm_blocking_callback cb_blocking)
3276 {
3277         int rc;
3278         int i;
3279
3280         rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3281         if (rc < 0)
3282                 return rc;
3283
3284         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3285                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3286
3287                 CDEBUG(D_INFO, ""DFID" size %llu, blocks %llu nlink %u,"
3288                        " atime %lu ctime %lu, mtime %lu.\n",
3289                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3290                        i_size_read(inode), (unsigned long long)inode->i_blocks,
3291                        inode->i_nlink, LTIME_S(inode->i_atime),
3292                        LTIME_S(inode->i_ctime), LTIME_S(inode->i_mtime));
3293
3294                 /* for slave stripe, it needs to subtract nlink for . and .. */
3295                 if (i != 0)
3296                         attr->cat_nlink += inode->i_nlink - 2;
3297                 else
3298                         attr->cat_nlink = inode->i_nlink;
3299
3300                 attr->cat_size += i_size_read(inode);
3301                 attr->cat_blocks += inode->i_blocks;
3302
3303                 if (attr->cat_atime < LTIME_S(inode->i_atime))
3304                         attr->cat_atime = LTIME_S(inode->i_atime);
3305
3306                 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3307                         attr->cat_ctime = LTIME_S(inode->i_ctime);
3308
3309                 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3310                         attr->cat_mtime = LTIME_S(inode->i_mtime);
3311         }
3312         return 0;
3313 }
3314
3315 struct obd_ops lmv_obd_ops = {
3316         .o_owner                = THIS_MODULE,
3317         .o_setup                = lmv_setup,
3318         .o_cleanup              = lmv_cleanup,
3319         .o_precleanup           = lmv_precleanup,
3320         .o_process_config       = lmv_process_config,
3321         .o_connect              = lmv_connect,
3322         .o_disconnect           = lmv_disconnect,
3323         .o_statfs               = lmv_statfs,
3324         .o_get_info             = lmv_get_info,
3325         .o_set_info_async       = lmv_set_info_async,
3326         .o_notify               = lmv_notify,
3327         .o_get_uuid             = lmv_get_uuid,
3328         .o_iocontrol            = lmv_iocontrol,
3329         .o_quotactl             = lmv_quotactl
3330 };
3331
3332 struct md_ops lmv_md_ops = {
3333         .m_get_root             = lmv_get_root,
3334         .m_null_inode           = lmv_null_inode,
3335         .m_close                = lmv_close,
3336         .m_create               = lmv_create,
3337         .m_enqueue              = lmv_enqueue,
3338         .m_getattr              = lmv_getattr,
3339         .m_getxattr             = lmv_getxattr,
3340         .m_getattr_name         = lmv_getattr_name,
3341         .m_intent_lock          = lmv_intent_lock,
3342         .m_link                 = lmv_link,
3343         .m_rename               = lmv_rename,
3344         .m_setattr              = lmv_setattr,
3345         .m_setxattr             = lmv_setxattr,
3346         .m_fsync                = lmv_fsync,
3347         .m_read_page            = lmv_read_page,
3348         .m_unlink               = lmv_unlink,
3349         .m_init_ea_size         = lmv_init_ea_size,
3350         .m_cancel_unused        = lmv_cancel_unused,
3351         .m_set_lock_data        = lmv_set_lock_data,
3352         .m_lock_match           = lmv_lock_match,
3353         .m_get_lustre_md        = lmv_get_lustre_md,
3354         .m_free_lustre_md       = lmv_free_lustre_md,
3355         .m_merge_attr           = lmv_merge_attr,
3356         .m_set_open_replay_data = lmv_set_open_replay_data,
3357         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3358         .m_get_remote_perm      = lmv_get_remote_perm,
3359         .m_intent_getattr_async = lmv_intent_getattr_async,
3360         .m_revalidate_lock      = lmv_revalidate_lock,
3361         .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
3362         .m_unpackmd             = lmv_unpackmd,
3363 };
3364
3365 static int __init lmv_init(void)
3366 {
3367         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3368                                    LUSTRE_LMV_NAME, NULL);
3369 }
3370
3371 static void __exit lmv_exit(void)
3372 {
3373         class_unregister_type(LUSTRE_LMV_NAME);
3374 }
3375
3376 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3377 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
3378 MODULE_VERSION(LUSTRE_VERSION_STRING);
3379 MODULE_LICENSE("GPL");
3380
3381 module_init(lmv_init);
3382 module_exit(lmv_exit);