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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #define DEBUG_SUBSYSTEM S_LMV
  34 #include <linux/slab.h>
  35 #include <linux/module.h>
  36 #include <linux/init.h>
  37 #include <linux/user_namespace.h>
  38 #ifdef HAVE_UIDGID_HEADER
  39 # include <linux/uidgid.h>
  40 #endif
  41 #include <linux/slab.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <linux/math64.h>
  45 #include <linux/seq_file.h>
  46 #include <linux/namei.h>
  47
  48 #include <lustre/lustre_idl.h>
  49 #include <obd_support.h>
  50 #include <lustre_lib.h>
  51 #include <lustre_net.h>
  52 #include <obd_class.h>
  53 #include <lustre_lmv.h>
  54 #include <lprocfs_status.h>
  55 #include <cl_object.h>
  56 #include <lustre_fid.h>
  57 #include <lustre_ioctl.h>
  58 #include <lustre_kernelcomm.h>
  59 #include "lmv_internal.h"
  60
  61 static int lmv_check_connect(struct obd_device *obd);
  62
  63 static void lmv_activate_target(struct lmv_obd *lmv,
  64                                 struct lmv_tgt_desc *tgt,
  65                                 int activate)
  66 {
  67         if (tgt->ltd_active == activate)
  68                 return;
  69
  70         tgt->ltd_active = activate;
  71         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  72
  73         tgt->ltd_exp->exp_obd->obd_inactive = !activate;
  74 }
  75
  76 /**
  77  * Error codes:
  78  *
  79  *  -EINVAL  : UUID can't be found in the LMV's target list
  80  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  81  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  82  */
  83 static int lmv_set_mdc_active(struct lmv_obd *lmv,
  84                               const struct obd_uuid *uuid,
  85                               int activate)
  86 {
  87         struct lmv_tgt_desc     *tgt = NULL;
  88         struct obd_device       *obd;
  89         __u32                    i;
  90         int                      rc = 0;
  91         ENTRY;
  92
  93         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  94                         lmv, uuid->uuid, activate);
  95
  96         spin_lock(&lmv->lmv_lock);
  97         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
  98                 tgt = lmv->tgts[i];
  99                 if (tgt == NULL || tgt->ltd_exp == NULL)
 100                         continue;
 101
 102                 CDEBUG(D_INFO, "Target idx %d is %s conn %#llx\n", i,
 103                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 104
 105                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 106                         break;
 107         }
 108
 109         if (i == lmv->desc.ld_tgt_count)
 110                 GOTO(out_lmv_lock, rc = -EINVAL);
 111
 112         obd = class_exp2obd(tgt->ltd_exp);
 113         if (obd == NULL)
 114                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 115
 116         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 117                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 118                obd->obd_type->typ_name, i);
 119         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 120
 121         if (tgt->ltd_active == activate) {
 122                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 123                        activate ? "" : "in");
 124                 GOTO(out_lmv_lock, rc);
 125         }
 126
 127         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 128                activate ? "" : "in");
 129         lmv_activate_target(lmv, tgt, activate);
 130         EXIT;
 131
 132  out_lmv_lock:
 133         spin_unlock(&lmv->lmv_lock);
 134         return rc;
 135 }
 136
 137 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 138 {
 139         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
 140         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
 141
 142         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
 143 }
 144
 145 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 146                       enum obd_notify_event ev, void *data)
 147 {
 148         struct obd_connect_data *conn_data;
 149         struct lmv_obd          *lmv = &obd->u.lmv;
 150         struct obd_uuid         *uuid;
 151         int                      rc = 0;
 152         ENTRY;
 153
 154         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 155                 CERROR("unexpected notification of %s %s!\n",
 156                        watched->obd_type->typ_name,
 157                        watched->obd_name);
 158                 RETURN(-EINVAL);
 159         }
 160
 161         uuid = &watched->u.cli.cl_target_uuid;
 162         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 163                 /*
 164                  * Set MDC as active before notifying the observer, so the
 165                  * observer can use the MDC normally.
 166                  */
 167                 rc = lmv_set_mdc_active(lmv, uuid,
 168                                         ev == OBD_NOTIFY_ACTIVE);
 169                 if (rc) {
 170                         CERROR("%sactivation of %s failed: %d\n",
 171                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 172                                uuid->uuid, rc);
 173                         RETURN(rc);
 174                 }
 175         } else if (ev == OBD_NOTIFY_OCD) {
 176                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 177                 /*
 178                  * XXX: Make sure that ocd_connect_flags from all targets are
 179                  * the same. Otherwise one of MDTs runs wrong version or
 180                  * something like this.  --umka
 181                  */
 182                 obd->obd_self_export->exp_connect_data = *conn_data;
 183         }
 184
 185         /*
 186          * Pass the notification up the chain.
 187          */
 188         if (obd->obd_observer)
 189                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 190
 191         RETURN(rc);
 192 }
 193
 194 static int lmv_connect(const struct lu_env *env,
 195                        struct obd_export **pexp, struct obd_device *obd,
 196                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 197                        void *localdata)
 198 {
 199         struct lmv_obd *lmv = &obd->u.lmv;
 200         struct lustre_handle conn = { 0 };
 201         struct obd_export *exp;
 202         int rc;
 203         ENTRY;
 204
 205         rc = class_connect(&conn, obd, cluuid);
 206         if (rc) {
 207                 CERROR("class_connection() returned %d\n", rc);
 208                 RETURN(rc);
 209         }
 210
 211         exp = class_conn2export(&conn);
 212
 213         lmv->connected = 0;
 214         lmv->cluuid = *cluuid;
 215         lmv->conn_data = *data;
 216
 217         if (lmv->targets_proc_entry == NULL) {
 218                 lmv->targets_proc_entry = lprocfs_register("target_obds",
 219                                                            obd->obd_proc_entry,
 220                                                            NULL, NULL);
 221                 if (IS_ERR(lmv->targets_proc_entry)) {
 222                         CERROR("%s: cannot register "
 223                                "/proc/fs/lustre/%s/%s/target_obds\n",
 224                                obd->obd_name, obd->obd_type->typ_name,
 225                                obd->obd_name);
 226                         lmv->targets_proc_entry = NULL;
 227                 }
 228         }
 229
 230         rc = lmv_check_connect(obd);
 231         if (rc != 0)
 232                 GOTO(out_proc, rc);
 233
 234         *pexp = exp;
 235
 236         RETURN(rc);
 237
 238 out_proc:
 239         if (lmv->targets_proc_entry != NULL)
 240                 lprocfs_remove(&lmv->targets_proc_entry);
 241
 242         class_disconnect(exp);
 243
 244         return rc;
 245 }
 246
 247 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
 248                             __u32 def_easize)
 249 {
 250         struct obd_device       *obd = exp->exp_obd;
 251         struct lmv_obd          *lmv = &obd->u.lmv;
 252         __u32                    i;
 253         int                      rc = 0;
 254         int                      change = 0;
 255         ENTRY;
 256
 257         if (lmv->max_easize < easize) {
 258                 lmv->max_easize = easize;
 259                 change = 1;
 260         }
 261         if (lmv->max_def_easize < def_easize) {
 262                 lmv->max_def_easize = def_easize;
 263                 change = 1;
 264         }
 265
 266         if (change == 0)
 267                 RETURN(0);
 268
 269         if (lmv->connected == 0)
 270                 RETURN(0);
 271
 272         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 273                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 274
 275                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
 276                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 277                         continue;
 278                 }
 279
 280                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
 281                 if (rc) {
 282                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 283                                " rc = %d\n", obd->obd_name, i, rc);
 284                         break;
 285                 }
 286         }
 287         RETURN(rc);
 288 }
 289
 290 #define MAX_STRING_SIZE 128
 291
 292 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 293 {
 294         struct lmv_obd          *lmv = &obd->u.lmv;
 295         struct obd_uuid         *cluuid = &lmv->cluuid;
 296         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 297         struct obd_device       *mdc_obd;
 298         struct obd_export       *mdc_exp;
 299         struct lu_fld_target     target;
 300         int                      rc;
 301         ENTRY;
 302
 303         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 304                                         &obd->obd_uuid);
 305         if (!mdc_obd) {
 306                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 307                 RETURN(-EINVAL);
 308         }
 309
 310         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 311                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 312                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 313                 cluuid->uuid);
 314
 315         if (!mdc_obd->obd_set_up) {
 316                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 317                 RETURN(-EINVAL);
 318         }
 319
 320         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 321                          &lmv->conn_data, NULL);
 322         if (rc) {
 323                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 324                 RETURN(rc);
 325         }
 326
 327         /*
 328          * Init fid sequence client for this mdc and add new fld target.
 329          */
 330         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 331         if (rc)
 332                 RETURN(rc);
 333
 334         target.ft_srv = NULL;
 335         target.ft_exp = mdc_exp;
 336         target.ft_idx = tgt->ltd_idx;
 337
 338         fld_client_add_target(&lmv->lmv_fld, &target);
 339
 340         rc = obd_register_observer(mdc_obd, obd);
 341         if (rc) {
 342                 obd_disconnect(mdc_exp);
 343                 CERROR("target %s register_observer error %d\n",
 344                        tgt->ltd_uuid.uuid, rc);
 345                 RETURN(rc);
 346         }
 347
 348         if (obd->obd_observer) {
 349                 /*
 350                  * Tell the observer about the new target.
 351                  */
 352                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 353                                 OBD_NOTIFY_ACTIVE,
 354                                 (void *)(tgt - lmv->tgts[0]));
 355                 if (rc) {
 356                         obd_disconnect(mdc_exp);
 357                         RETURN(rc);
 358                 }
 359         }
 360
 361         tgt->ltd_active = 1;
 362         tgt->ltd_exp = mdc_exp;
 363         lmv->desc.ld_active_tgt_count++;
 364
 365         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
 366
 367         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 368                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 369                 atomic_read(&obd->obd_refcount));
 370
 371         if (lmv->targets_proc_entry != NULL) {
 372                 struct proc_dir_entry *mdc_symlink;
 373
 374                 LASSERT(mdc_obd->obd_type != NULL);
 375                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 376                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 377                                                   lmv->targets_proc_entry,
 378                                                   "../../../%s/%s",
 379                                                   mdc_obd->obd_type->typ_name,
 380                                                   mdc_obd->obd_name);
 381                 if (mdc_symlink == NULL) {
 382                         CERROR("cannot register LMV target "
 383                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
 384                                obd->obd_type->typ_name, obd->obd_name,
 385                                mdc_obd->obd_name);
 386                 }
 387         }
 388         RETURN(0);
 389 }
 390
 391 static void lmv_del_target(struct lmv_obd *lmv, int index)
 392 {
 393         if (lmv->tgts[index] == NULL)
 394                 return;
 395
 396         OBD_FREE_PTR(lmv->tgts[index]);
 397         lmv->tgts[index] = NULL;
 398         return;
 399 }
 400
 401 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 402                            __u32 index, int gen)
 403 {
 404         struct obd_device *mdc_obd;
 405         struct lmv_obd      *lmv = &obd->u.lmv;
 406         struct lmv_tgt_desc *tgt;
 407         int                  orig_tgt_count = 0;
 408         int                  rc = 0;
 409         ENTRY;
 410
 411         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 412         mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 413                                         &obd->obd_uuid);
 414         if (!mdc_obd) {
 415                 CERROR("%s: Target %s not attached: rc = %d\n",
 416                        obd->obd_name, uuidp->uuid, -EINVAL);
 417                 RETURN(-EINVAL);
 418         }
 419
 420         mutex_lock(&lmv->lmv_init_mutex);
 421         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
 422                 tgt = lmv->tgts[index];
 423                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
 424                        " rc = %d\n", obd->obd_name,
 425                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
 426                 mutex_unlock(&lmv->lmv_init_mutex);
 427                 RETURN(-EEXIST);
 428         }
 429
 430         if (index >= lmv->tgts_size) {
 431                 /* We need to reallocate the lmv target array. */
 432                 struct lmv_tgt_desc **newtgts, **old = NULL;
 433                 __u32 newsize = 1;
 434                 __u32 oldsize = 0;
 435
 436                 while (newsize < index + 1)
 437                         newsize = newsize << 1;
 438                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 439                 if (newtgts == NULL) {
 440                         mutex_unlock(&lmv->lmv_init_mutex);
 441                         RETURN(-ENOMEM);
 442                 }
 443
 444                 if (lmv->tgts_size) {
 445                         memcpy(newtgts, lmv->tgts,
 446                                sizeof(*newtgts) * lmv->tgts_size);
 447                         old = lmv->tgts;
 448                         oldsize = lmv->tgts_size;
 449                 }
 450
 451                 lmv->tgts = newtgts;
 452                 lmv->tgts_size = newsize;
 453                 smp_rmb();
 454                 if (old)
 455                         OBD_FREE(old, sizeof(*old) * oldsize);
 456
 457                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
 458                        lmv->tgts_size);
 459         }
 460
 461         OBD_ALLOC_PTR(tgt);
 462         if (!tgt) {
 463                 mutex_unlock(&lmv->lmv_init_mutex);
 464                 RETURN(-ENOMEM);
 465         }
 466
 467         mutex_init(&tgt->ltd_fid_mutex);
 468         tgt->ltd_idx = index;
 469         tgt->ltd_uuid = *uuidp;
 470         tgt->ltd_active = 0;
 471         lmv->tgts[index] = tgt;
 472         if (index >= lmv->desc.ld_tgt_count) {
 473                 orig_tgt_count = lmv->desc.ld_tgt_count;
 474                 lmv->desc.ld_tgt_count = index + 1;
 475         }
 476
 477         if (lmv->connected == 0) {
 478                 /* lmv_check_connect() will connect this target. */
 479                 mutex_unlock(&lmv->lmv_init_mutex);
 480                 RETURN(0);
 481         }
 482
 483         /* Otherwise let's connect it ourselves */
 484         mutex_unlock(&lmv->lmv_init_mutex);
 485         rc = lmv_connect_mdc(obd, tgt);
 486         if (rc != 0) {
 487                 spin_lock(&lmv->lmv_lock);
 488                 if (lmv->desc.ld_tgt_count == index + 1)
 489                         lmv->desc.ld_tgt_count = orig_tgt_count;
 490                 memset(tgt, 0, sizeof(*tgt));
 491                 spin_unlock(&lmv->lmv_lock);
 492         } else {
 493                 int easize = sizeof(struct lmv_stripe_md) +
 494                         lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
 495                 lmv_init_ea_size(obd->obd_self_export, easize, 0);
 496         }
 497
 498         RETURN(rc);
 499 }
 500
 501 static int lmv_check_connect(struct obd_device *obd)
 502 {
 503         struct lmv_obd          *lmv = &obd->u.lmv;
 504         struct lmv_tgt_desc     *tgt;
 505         __u32                    i;
 506         int                      rc;
 507         int                      easize;
 508         ENTRY;
 509
 510         if (lmv->connected)
 511                 RETURN(0);
 512
 513         mutex_lock(&lmv->lmv_init_mutex);
 514         if (lmv->connected) {
 515                 mutex_unlock(&lmv->lmv_init_mutex);
 516                 RETURN(0);
 517         }
 518
 519         if (lmv->desc.ld_tgt_count == 0) {
 520                 mutex_unlock(&lmv->lmv_init_mutex);
 521                 CERROR("%s: no targets configured.\n", obd->obd_name);
 522                 RETURN(-EINVAL);
 523         }
 524
 525         LASSERT(lmv->tgts != NULL);
 526
 527         if (lmv->tgts[0] == NULL) {
 528                 mutex_unlock(&lmv->lmv_init_mutex);
 529                 CERROR("%s: no target configured for index 0.\n",
 530                        obd->obd_name);
 531                 RETURN(-EINVAL);
 532         }
 533
 534         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 535                lmv->cluuid.uuid, obd->obd_name);
 536
 537         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 538                 tgt = lmv->tgts[i];
 539                 if (tgt == NULL)
 540                         continue;
 541                 rc = lmv_connect_mdc(obd, tgt);
 542                 if (rc)
 543                         GOTO(out_disc, rc);
 544         }
 545
 546         lmv->connected = 1;
 547         easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
 548         lmv_init_ea_size(obd->obd_self_export, easize, 0);
 549         mutex_unlock(&lmv->lmv_init_mutex);
 550         RETURN(0);
 551
 552  out_disc:
 553         while (i-- > 0) {
 554                 int rc2;
 555                 tgt = lmv->tgts[i];
 556                 if (tgt == NULL)
 557                         continue;
 558                 tgt->ltd_active = 0;
 559                 if (tgt->ltd_exp) {
 560                         --lmv->desc.ld_active_tgt_count;
 561                         rc2 = obd_disconnect(tgt->ltd_exp);
 562                         if (rc2) {
 563                                 CERROR("LMV target %s disconnect on "
 564                                        "MDC idx %d: error %d\n",
 565                                        tgt->ltd_uuid.uuid, i, rc2);
 566                         }
 567                 }
 568         }
 569
 570         mutex_unlock(&lmv->lmv_init_mutex);
 571
 572         RETURN(rc);
 573 }
 574
 575 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 576 {
 577         struct lmv_obd         *lmv = &obd->u.lmv;
 578         struct obd_device      *mdc_obd;
 579         int                     rc;
 580         ENTRY;
 581
 582         LASSERT(tgt != NULL);
 583         LASSERT(obd != NULL);
 584
 585         mdc_obd = class_exp2obd(tgt->ltd_exp);
 586
 587         if (mdc_obd) {
 588                 mdc_obd->obd_force = obd->obd_force;
 589                 mdc_obd->obd_fail = obd->obd_fail;
 590                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 591
 592                 if (lmv->targets_proc_entry != NULL)
 593                         lprocfs_remove_proc_entry(mdc_obd->obd_name,
 594                                                   lmv->targets_proc_entry);
 595         }
 596
 597         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 598         if (rc)
 599                 CERROR("Can't finanize fids factory\n");
 600
 601         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 602                tgt->ltd_exp->exp_obd->obd_name,
 603                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 604
 605         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 606         rc = obd_disconnect(tgt->ltd_exp);
 607         if (rc) {
 608                 if (tgt->ltd_active) {
 609                         CERROR("Target %s disconnect error %d\n",
 610                                tgt->ltd_uuid.uuid, rc);
 611                 }
 612         }
 613
 614         lmv_activate_target(lmv, tgt, 0);
 615         tgt->ltd_exp = NULL;
 616         RETURN(0);
 617 }
 618
 619 static int lmv_disconnect(struct obd_export *exp)
 620 {
 621         struct obd_device       *obd = class_exp2obd(exp);
 622         struct lmv_obd          *lmv = &obd->u.lmv;
 623         int                      rc;
 624         __u32                    i;
 625         ENTRY;
 626
 627         if (!lmv->tgts)
 628                 goto out_local;
 629
 630         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 631                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
 632                         continue;
 633
 634                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
 635         }
 636
 637         if (lmv->targets_proc_entry != NULL)
 638                 lprocfs_remove(&lmv->targets_proc_entry);
 639         else
 640                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 641                        obd->obd_type->typ_name, obd->obd_name);
 642
 643 out_local:
 644         /*
 645          * This is the case when no real connection is established by
 646          * lmv_check_connect().
 647          */
 648         if (!lmv->connected)
 649                 class_export_put(exp);
 650         rc = class_disconnect(exp);
 651         lmv->connected = 0;
 652
 653         RETURN(rc);
 654 }
 655
 656 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
 657                         void __user *uarg)
 658 {
 659         struct obd_device       *obddev = class_exp2obd(exp);
 660         struct lmv_obd          *lmv = &obddev->u.lmv;
 661         struct getinfo_fid2path *gf;
 662         struct lmv_tgt_desc     *tgt;
 663         struct getinfo_fid2path *remote_gf = NULL;
 664         struct lu_fid           root_fid;
 665         int                     remote_gf_size = 0;
 666         int                     rc;
 667
 668         gf = karg;
 669         tgt = lmv_find_target(lmv, &gf->gf_fid);
 670         if (IS_ERR(tgt))
 671                 RETURN(PTR_ERR(tgt));
 672
 673         root_fid = *gf->gf_u.gf_root_fid;
 674         LASSERT(fid_is_sane(&root_fid));
 675
 676 repeat_fid2path:
 677         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 678         if (rc != 0 && rc != -EREMOTE)
 679                 GOTO(out_fid2path, rc);
 680
 681         /* If remote_gf != NULL, it means just building the
 682          * path on the remote MDT, copy this path segement to gf */
 683         if (remote_gf != NULL) {
 684                 struct getinfo_fid2path *ori_gf;
 685                 char *ptr;
 686
 687                 ori_gf = (struct getinfo_fid2path *)karg;
 688                 if (strlen(ori_gf->gf_u.gf_path) +
 689                     strlen(gf->gf_u.gf_path) > ori_gf->gf_pathlen)
 690                         GOTO(out_fid2path, rc = -EOVERFLOW);
 691
 692                 ptr = ori_gf->gf_u.gf_path;
 693
 694                 memmove(ptr + strlen(gf->gf_u.gf_path) + 1, ptr,
 695                         strlen(ori_gf->gf_u.gf_path));
 696
 697                 strncpy(ptr, gf->gf_u.gf_path,
 698                         strlen(gf->gf_u.gf_path));
 699                 ptr += strlen(gf->gf_u.gf_path);
 700                 *ptr = '/';
 701         }
 702
 703         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: %llu ln: %u\n",
 704                tgt->ltd_exp->exp_obd->obd_name,
 705                gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 706                gf->gf_linkno);
 707
 708         if (rc == 0)
 709                 GOTO(out_fid2path, rc);
 710
 711         /* sigh, has to go to another MDT to do path building further */
 712         if (remote_gf == NULL) {
 713                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 714                 OBD_ALLOC(remote_gf, remote_gf_size);
 715                 if (remote_gf == NULL)
 716                         GOTO(out_fid2path, rc = -ENOMEM);
 717                 remote_gf->gf_pathlen = PATH_MAX;
 718         }
 719
 720         if (!fid_is_sane(&gf->gf_fid)) {
 721                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 722                        tgt->ltd_exp->exp_obd->obd_name,
 723                        PFID(&gf->gf_fid), -EINVAL);
 724                 GOTO(out_fid2path, rc = -EINVAL);
 725         }
 726
 727         tgt = lmv_find_target(lmv, &gf->gf_fid);
 728         if (IS_ERR(tgt))
 729                 GOTO(out_fid2path, rc = -EINVAL);
 730
 731         remote_gf->gf_fid = gf->gf_fid;
 732         remote_gf->gf_recno = -1;
 733         remote_gf->gf_linkno = -1;
 734         memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
 735         *remote_gf->gf_u.gf_root_fid = root_fid;
 736         gf = remote_gf;
 737         goto repeat_fid2path;
 738
 739 out_fid2path:
 740         if (remote_gf != NULL)
 741                 OBD_FREE(remote_gf, remote_gf_size);
 742         RETURN(rc);
 743 }
 744
 745 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 746                              const struct hsm_user_request *hur,
 747                              const struct lmv_tgt_desc *tgt_mds)
 748 {
 749         __u32                    i;
 750         int                      nr = 0;
 751         struct lmv_tgt_desc     *curr_tgt;
 752
 753         /* count how many requests must be sent to the given target */
 754         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 755                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
 756                 if (IS_ERR(curr_tgt))
 757                         RETURN(PTR_ERR(curr_tgt));
 758                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 759                         nr++;
 760         }
 761         return nr;
 762 }
 763
 764 static int lmv_hsm_req_build(struct lmv_obd *lmv,
 765                               struct hsm_user_request *hur_in,
 766                               const struct lmv_tgt_desc *tgt_mds,
 767                               struct hsm_user_request *hur_out)
 768 {
 769         __u32                    i, nr_out;
 770         struct lmv_tgt_desc     *curr_tgt;
 771
 772         /* build the hsm_user_request for the given target */
 773         hur_out->hur_request = hur_in->hur_request;
 774         nr_out = 0;
 775         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 776                 curr_tgt = lmv_find_target(lmv,
 777                                            &hur_in->hur_user_item[i].hui_fid);
 778                 if (IS_ERR(curr_tgt))
 779                         RETURN(PTR_ERR(curr_tgt));
 780                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 781                         hur_out->hur_user_item[nr_out] =
 782                                                 hur_in->hur_user_item[i];
 783                         nr_out++;
 784                 }
 785         }
 786         hur_out->hur_request.hr_itemcount = nr_out;
 787         memcpy(hur_data(hur_out), hur_data(hur_in),
 788                hur_in->hur_request.hr_data_len);
 789
 790         RETURN(0);
 791 }
 792
 793 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
 794                                  struct lustre_kernelcomm *lk,
 795                                  void __user *uarg)
 796 {
 797         __u32   i;
 798         int     rc;
 799         ENTRY;
 800
 801         /* unregister request (call from llapi_hsm_copytool_fini) */
 802         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 803                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 804
 805                 if (tgt == NULL || tgt->ltd_exp == NULL)
 806                         continue;
 807                 /* best effort: try to clean as much as possible
 808                  * (continue on error) */
 809                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 810         }
 811
 812         /* Whatever the result, remove copytool from kuc groups.
 813          * Unreached coordinators will get EPIPE on next requests
 814          * and will unregister automatically.
 815          */
 816         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group);
 817
 818         RETURN(rc);
 819 }
 820
 821 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
 822                                struct lustre_kernelcomm *lk, __user void *uarg)
 823 {
 824         struct file             *filp;
 825         __u32                    i, j;
 826         int                      err, rc;
 827         bool                     any_set = false;
 828         struct kkuc_ct_data      kcd = { 0 };
 829         ENTRY;
 830
 831         /* All or nothing: try to register to all MDS.
 832          * In case of failure, unregister from previous MDS,
 833          * except if it because of inactive target. */
 834         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 835                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 836
 837                 if (tgt == NULL || tgt->ltd_exp == NULL)
 838                         continue;
 839                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 840                 if (err) {
 841                         if (tgt->ltd_active) {
 842                                 /* permanent error */
 843                                 CERROR("%s: iocontrol MDC %s on MDT"
 844                                        " idx %d cmd %x: err = %d\n",
 845                                        lmv2obd_dev(lmv)->obd_name,
 846                                        tgt->ltd_uuid.uuid, i, cmd, err);
 847                                 rc = err;
 848                                 lk->lk_flags |= LK_FLG_STOP;
 849                                 /* unregister from previous MDS */
 850                                 for (j = 0; j < i; j++) {
 851                                         tgt = lmv->tgts[j];
 852                                         if (tgt == NULL || tgt->ltd_exp == NULL)
 853                                                 continue;
 854                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
 855                                                       lk, uarg);
 856                                 }
 857                                 RETURN(rc);
 858                         }
 859                         /* else: transient error.
 860                          * kuc will register to the missing MDT
 861                          * when it is back */
 862                 } else {
 863                         any_set = true;
 864                 }
 865         }
 866
 867         if (!any_set)
 868                 /* no registration done: return error */
 869                 RETURN(-ENOTCONN);
 870
 871         /* at least one registration done, with no failure */
 872         filp = fget(lk->lk_wfd);
 873         if (filp == NULL)
 874                 RETURN(-EBADF);
 875
 876         kcd.kcd_magic = KKUC_CT_DATA_MAGIC;
 877         kcd.kcd_uuid = lmv->cluuid;
 878         kcd.kcd_archive = lk->lk_data;
 879
 880         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group,
 881                                    &kcd, sizeof(kcd));
 882         if (rc != 0)
 883                 fput(filp);
 884
 885         RETURN(rc);
 886 }
 887
 888
 889
 890
 891 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 892                          int len, void *karg, void __user *uarg)
 893 {
 894         struct obd_device       *obddev = class_exp2obd(exp);
 895         struct lmv_obd          *lmv = &obddev->u.lmv;
 896         struct lmv_tgt_desc     *tgt = NULL;
 897         __u32                    i = 0;
 898         int                      rc = 0;
 899         int                      set = 0;
 900         __u32                    count = lmv->desc.ld_tgt_count;
 901         ENTRY;
 902
 903         if (count == 0)
 904                 RETURN(-ENOTTY);
 905
 906         switch (cmd) {
 907         case IOC_OBD_STATFS: {
 908                 struct obd_ioctl_data *data = karg;
 909                 struct obd_device *mdc_obd;
 910                 struct obd_statfs stat_buf = {0};
 911                 __u32 index;
 912
 913                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 914                 if ((index >= count))
 915                         RETURN(-ENODEV);
 916
 917                 tgt = lmv->tgts[index];
 918                 if (tgt == NULL || !tgt->ltd_active)
 919                         RETURN(-ENODATA);
 920
 921                 mdc_obd = class_exp2obd(tgt->ltd_exp);
 922                 if (!mdc_obd)
 923                         RETURN(-EINVAL);
 924
 925                 /* copy UUID */
 926                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 927                                  min((int) data->ioc_plen2,
 928                                      (int) sizeof(struct obd_uuid))))
 929                         RETURN(-EFAULT);
 930
 931                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
 932                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 933                                 0);
 934                 if (rc)
 935                         RETURN(rc);
 936                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
 937                                  min((int) data->ioc_plen1,
 938                                      (int) sizeof(stat_buf))))
 939                         RETURN(-EFAULT);
 940                 break;
 941         }
 942         case OBD_IOC_QUOTACTL: {
 943                 struct if_quotactl *qctl = karg;
 944                 struct obd_quotactl *oqctl;
 945
 946                 if (qctl->qc_valid == QC_MDTIDX) {
 947                         if (count <= qctl->qc_idx)
 948                                 RETURN(-EINVAL);
 949
 950                         tgt = lmv->tgts[qctl->qc_idx];
 951                         if (tgt == NULL || tgt->ltd_exp == NULL)
 952                                 RETURN(-EINVAL);
 953                 } else if (qctl->qc_valid == QC_UUID) {
 954                         for (i = 0; i < count; i++) {
 955                                 tgt = lmv->tgts[i];
 956                                 if (tgt == NULL)
 957                                         continue;
 958                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 959                                                      &qctl->obd_uuid))
 960                                         continue;
 961
 962                                 if (tgt->ltd_exp == NULL)
 963                                         RETURN(-EINVAL);
 964
 965                                 break;
 966                         }
 967                 } else {
 968                         RETURN(-EINVAL);
 969                 }
 970
 971                 if (i >= count)
 972                         RETURN(-EAGAIN);
 973
 974                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
 975                 OBD_ALLOC_PTR(oqctl);
 976                 if (!oqctl)
 977                         RETURN(-ENOMEM);
 978
 979                 QCTL_COPY(oqctl, qctl);
 980                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
 981                 if (rc == 0) {
 982                         QCTL_COPY(qctl, oqctl);
 983                         qctl->qc_valid = QC_MDTIDX;
 984                         qctl->obd_uuid = tgt->ltd_uuid;
 985                 }
 986                 OBD_FREE_PTR(oqctl);
 987                 break;
 988         }
 989         case OBD_IOC_CHANGELOG_SEND:
 990         case OBD_IOC_CHANGELOG_CLEAR: {
 991                 struct ioc_changelog *icc = karg;
 992
 993                 if (icc->icc_mdtindex >= count)
 994                         RETURN(-ENODEV);
 995
 996                 tgt = lmv->tgts[icc->icc_mdtindex];
 997                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
 998                         RETURN(-ENODEV);
 999                 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1000                 break;
1001         }
1002         case LL_IOC_GET_CONNECT_FLAGS: {
1003                 tgt = lmv->tgts[0];
1004                 if (tgt == NULL || tgt->ltd_exp == NULL)
1005                         RETURN(-ENODATA);
1006                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1007                 break;
1008         }
1009         case LL_IOC_FID2MDTIDX: {
1010                 struct lu_fid *fid = karg;
1011                 int             mdt_index;
1012
1013                 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1014                 if (rc != 0)
1015                         RETURN(rc);
1016
1017                 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1018                  * point to user space memory for FID2MDTIDX. */
1019                 *(__u32 *)uarg = mdt_index;
1020                 break;
1021         }
1022         case OBD_IOC_FID2PATH: {
1023                 rc = lmv_fid2path(exp, len, karg, uarg);
1024                 break;
1025         }
1026         case LL_IOC_HSM_STATE_GET:
1027         case LL_IOC_HSM_STATE_SET:
1028         case LL_IOC_HSM_ACTION: {
1029                 struct md_op_data       *op_data = karg;
1030
1031                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1032                 if (IS_ERR(tgt))
1033                         RETURN(PTR_ERR(tgt));
1034
1035                 if (tgt->ltd_exp == NULL)
1036                         RETURN(-EINVAL);
1037
1038                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1039                 break;
1040         }
1041         case LL_IOC_HSM_PROGRESS: {
1042                 const struct hsm_progress_kernel *hpk = karg;
1043
1044                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1045                 if (IS_ERR(tgt))
1046                         RETURN(PTR_ERR(tgt));
1047                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1048                 break;
1049         }
1050         case LL_IOC_HSM_REQUEST: {
1051                 struct hsm_user_request *hur = karg;
1052                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1053
1054                 if (reqcount == 0)
1055                         RETURN(0);
1056
1057                 /* if the request is about a single fid
1058                  * or if there is a single MDS, no need to split
1059                  * the request. */
1060                 if (reqcount == 1 || count == 1) {
1061                         tgt = lmv_find_target(lmv,
1062                                               &hur->hur_user_item[0].hui_fid);
1063                         if (IS_ERR(tgt))
1064                                 RETURN(PTR_ERR(tgt));
1065                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1066                 } else {
1067                         /* split fid list to their respective MDS */
1068                         for (i = 0; i < count; i++) {
1069                                 int nr, rc1;
1070                                 size_t reqlen;
1071                                 struct hsm_user_request *req;
1072
1073                                 tgt = lmv->tgts[i];
1074                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1075                                         continue;
1076
1077                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1078                                 if (nr < 0)
1079                                         RETURN(nr);
1080                                 if (nr == 0) /* nothing for this MDS */
1081                                         continue;
1082
1083                                 /* build a request with fids for this MDS */
1084                                 reqlen = offsetof(typeof(*hur),
1085                                                   hur_user_item[nr])
1086                                                 + hur->hur_request.hr_data_len;
1087                                 OBD_ALLOC_LARGE(req, reqlen);
1088                                 if (req == NULL)
1089                                         RETURN(-ENOMEM);
1090                                 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
1091                                 if (rc1 < 0)
1092                                         GOTO(hsm_req_err, rc1);
1093                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1094                                                     req, uarg);
1095 hsm_req_err:
1096                                 if (rc1 != 0 && rc == 0)
1097                                         rc = rc1;
1098                                 OBD_FREE_LARGE(req, reqlen);
1099                         }
1100                 }
1101                 break;
1102         }
1103         case LL_IOC_LOV_SWAP_LAYOUTS: {
1104                 struct md_op_data       *op_data = karg;
1105                 struct lmv_tgt_desc     *tgt1, *tgt2;
1106
1107                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1108                 if (IS_ERR(tgt1))
1109                         RETURN(PTR_ERR(tgt1));
1110
1111                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1112                 if (IS_ERR(tgt2))
1113                         RETURN(PTR_ERR(tgt2));
1114
1115                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1116                         RETURN(-EINVAL);
1117
1118                 /* only files on same MDT can have their layouts swapped */
1119                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1120                         RETURN(-EPERM);
1121
1122                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1123                 break;
1124         }
1125         case LL_IOC_HSM_CT_START: {
1126                 struct lustre_kernelcomm *lk = karg;
1127                 if (lk->lk_flags & LK_FLG_STOP)
1128                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1129                 else
1130                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1131                 break;
1132         }
1133         default:
1134                 for (i = 0; i < count; i++) {
1135                         struct obd_device *mdc_obd;
1136                         int err;
1137
1138                         tgt = lmv->tgts[i];
1139                         if (tgt == NULL || tgt->ltd_exp == NULL)
1140                                 continue;
1141                         /* ll_umount_begin() sets force flag but for lmv, not
1142                          * mdc. Let's pass it through */
1143                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1144                         mdc_obd->obd_force = obddev->obd_force;
1145                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1146                         if (err) {
1147                                 if (tgt->ltd_active) {
1148                                         CERROR("error: iocontrol MDC %s on MDT"
1149                                                " idx %d cmd %x: err = %d\n",
1150                                                tgt->ltd_uuid.uuid, i, cmd, err);
1151                                         if (!rc)
1152                                                 rc = err;
1153                                 }
1154                         } else
1155                                 set = 1;
1156                 }
1157                 if (!set && !rc)
1158                         rc = -EIO;
1159         }
1160         RETURN(rc);
1161 }
1162
1163 /**
1164  * This is _inode_ placement policy function (not name).
1165  */
1166 static int lmv_placement_policy(struct obd_device *obd,
1167                                 struct md_op_data *op_data, u32 *mds)
1168 {
1169         struct lmv_obd     *lmv = &obd->u.lmv;
1170         struct lmv_user_md *lum;
1171
1172         ENTRY;
1173
1174         LASSERT(mds != NULL);
1175
1176         if (lmv->desc.ld_tgt_count == 1) {
1177                 *mds = 0;
1178                 RETURN(0);
1179         }
1180
1181         lum = op_data->op_data;
1182         /* Choose MDS by
1183          * 1. See if the stripe offset is specified by lum.
1184          * 2. Then check if there is default stripe offset.
1185          * 3. Finally choose MDS by name hash if the parent
1186          *    is striped directory. (see lmv_locate_mds()). */
1187         if (op_data->op_cli_flags & CLI_SET_MEA && lum != NULL &&
1188             le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1189                 *mds = le32_to_cpu(lum->lum_stripe_offset);
1190         } else if (op_data->op_default_stripe_offset != (__u32)-1) {
1191                 *mds = op_data->op_default_stripe_offset;
1192                 op_data->op_mds = *mds;
1193                 /* Correct the stripe offset in lum */
1194                 if (lum != NULL)
1195                         lum->lum_stripe_offset = cpu_to_le32(*mds);
1196         } else {
1197                 *mds = op_data->op_mds;
1198         }
1199
1200         RETURN(0);
1201 }
1202
1203 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1204 {
1205         struct lmv_tgt_desc     *tgt;
1206         int                      rc;
1207         ENTRY;
1208
1209         tgt = lmv_get_target(lmv, mds, NULL);
1210         if (IS_ERR(tgt))
1211                 RETURN(PTR_ERR(tgt));
1212
1213         /*
1214          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1215          * on server that seq in new allocated fid is not yet known.
1216          */
1217         mutex_lock(&tgt->ltd_fid_mutex);
1218
1219         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1220                 GOTO(out, rc = -ENODEV);
1221
1222         /*
1223          * Asking underlying tgt layer to allocate new fid.
1224          */
1225         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1226         if (rc > 0) {
1227                 LASSERT(fid_is_sane(fid));
1228                 rc = 0;
1229         }
1230
1231         EXIT;
1232 out:
1233         mutex_unlock(&tgt->ltd_fid_mutex);
1234         return rc;
1235 }
1236
1237 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1238                   struct lu_fid *fid, struct md_op_data *op_data)
1239 {
1240         struct obd_device     *obd = class_exp2obd(exp);
1241         struct lmv_obd        *lmv = &obd->u.lmv;
1242         u32                    mds = 0;
1243         int                    rc;
1244         ENTRY;
1245
1246         LASSERT(op_data != NULL);
1247         LASSERT(fid != NULL);
1248
1249         rc = lmv_placement_policy(obd, op_data, &mds);
1250         if (rc) {
1251                 CERROR("Can't get target for allocating fid, "
1252                        "rc %d\n", rc);
1253                 RETURN(rc);
1254         }
1255
1256         rc = __lmv_fid_alloc(lmv, fid, mds);
1257         if (rc) {
1258                 CERROR("Can't alloc new fid, rc %d\n", rc);
1259                 RETURN(rc);
1260         }
1261
1262         RETURN(rc);
1263 }
1264
1265 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1266 {
1267         struct lmv_obd  *lmv = &obd->u.lmv;
1268         struct lmv_desc *desc;
1269         int             rc;
1270         ENTRY;
1271
1272         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1273                 CERROR("LMV setup requires a descriptor\n");
1274                 RETURN(-EINVAL);
1275         }
1276
1277         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1278         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1279                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1280                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1281                 RETURN(-EINVAL);
1282         }
1283
1284         lmv->tgts_size = 32U;
1285         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1286         if (lmv->tgts == NULL)
1287                 RETURN(-ENOMEM);
1288
1289         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1290         lmv->desc.ld_tgt_count = 0;
1291         lmv->desc.ld_active_tgt_count = 0;
1292         lmv->max_def_easize = 0;
1293         lmv->max_easize = 0;
1294
1295         spin_lock_init(&lmv->lmv_lock);
1296         mutex_init(&lmv->lmv_init_mutex);
1297
1298 #ifdef CONFIG_PROC_FS
1299         obd->obd_vars = lprocfs_lmv_obd_vars;
1300         lprocfs_obd_setup(obd);
1301         lprocfs_alloc_md_stats(obd, 0);
1302         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1303                                 0444, &lmv_proc_target_fops, obd);
1304         if (rc)
1305                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1306                       obd->obd_name, rc);
1307 #endif
1308         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1309                              LUSTRE_CLI_FLD_HASH_DHT);
1310         if (rc) {
1311                 CERROR("Can't init FLD, err %d\n", rc);
1312                 GOTO(out, rc);
1313         }
1314
1315         RETURN(0);
1316
1317 out:
1318         return rc;
1319 }
1320
1321 static int lmv_cleanup(struct obd_device *obd)
1322 {
1323         struct lmv_obd   *lmv = &obd->u.lmv;
1324         ENTRY;
1325
1326         fld_client_fini(&lmv->lmv_fld);
1327         if (lmv->tgts != NULL) {
1328                 int i;
1329                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1330                         if (lmv->tgts[i] == NULL)
1331                                 continue;
1332                         lmv_del_target(lmv, i);
1333                 }
1334                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1335                 lmv->tgts_size = 0;
1336         }
1337         RETURN(0);
1338 }
1339
1340 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1341 {
1342         struct lustre_cfg       *lcfg = buf;
1343         struct obd_uuid         obd_uuid;
1344         int                     gen;
1345         __u32                   index;
1346         int                     rc;
1347         ENTRY;
1348
1349         switch (lcfg->lcfg_command) {
1350         case LCFG_ADD_MDC:
1351                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1352                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1353                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1354                         GOTO(out, rc = -EINVAL);
1355
1356                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1357
1358                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1359                         GOTO(out, rc = -EINVAL);
1360                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1361                         GOTO(out, rc = -EINVAL);
1362                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1363                 GOTO(out, rc);
1364         default:
1365                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1366                 GOTO(out, rc = -EINVAL);
1367         }
1368 out:
1369         RETURN(rc);
1370 }
1371
1372 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1373                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1374 {
1375         struct obd_device       *obd = class_exp2obd(exp);
1376         struct lmv_obd          *lmv = &obd->u.lmv;
1377         struct obd_statfs       *temp;
1378         int                      rc = 0;
1379         __u32                    i;
1380         ENTRY;
1381
1382         OBD_ALLOC(temp, sizeof(*temp));
1383         if (temp == NULL)
1384                 RETURN(-ENOMEM);
1385
1386         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1387                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1388                         continue;
1389
1390                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1391                                 max_age, flags);
1392                 if (rc) {
1393                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1394                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1395                                rc);
1396                         GOTO(out_free_temp, rc);
1397                 }
1398
1399                 if (i == 0) {
1400                         *osfs = *temp;
1401                         /* If the statfs is from mount, it will needs
1402                          * retrieve necessary information from MDT0.
1403                          * i.e. mount does not need the merged osfs
1404                          * from all of MDT.
1405                          * And also clients can be mounted as long as
1406                          * MDT0 is in service*/
1407                         if (flags & OBD_STATFS_FOR_MDT0)
1408                                 GOTO(out_free_temp, rc);
1409                 } else {
1410                         osfs->os_bavail += temp->os_bavail;
1411                         osfs->os_blocks += temp->os_blocks;
1412                         osfs->os_ffree += temp->os_ffree;
1413                         osfs->os_files += temp->os_files;
1414                 }
1415         }
1416
1417         EXIT;
1418 out_free_temp:
1419         OBD_FREE(temp, sizeof(*temp));
1420         return rc;
1421 }
1422
1423 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1424                         struct lu_fid *fid)
1425 {
1426         struct obd_device    *obd = exp->exp_obd;
1427         struct lmv_obd       *lmv = &obd->u.lmv;
1428         int                   rc;
1429         ENTRY;
1430
1431         rc = md_get_root(lmv->tgts[0]->ltd_exp, fileset, fid);
1432         RETURN(rc);
1433 }
1434
1435 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1436                         u64 valid, const char *name,
1437                         const char *input, int input_size, int output_size,
1438                         int flags, struct ptlrpc_request **request)
1439 {
1440         struct obd_device      *obd = exp->exp_obd;
1441         struct lmv_obd         *lmv = &obd->u.lmv;
1442         struct lmv_tgt_desc    *tgt;
1443         int                     rc;
1444         ENTRY;
1445
1446         tgt = lmv_find_target(lmv, fid);
1447         if (IS_ERR(tgt))
1448                 RETURN(PTR_ERR(tgt));
1449
1450         rc = md_getxattr(tgt->ltd_exp, fid, valid, name, input,
1451                          input_size, output_size, flags, request);
1452
1453         RETURN(rc);
1454 }
1455
1456 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1457                         u64 valid, const char *name,
1458                         const char *input, int input_size, int output_size,
1459                         int flags, __u32 suppgid,
1460                         struct ptlrpc_request **request)
1461 {
1462         struct obd_device      *obd = exp->exp_obd;
1463         struct lmv_obd         *lmv = &obd->u.lmv;
1464         struct lmv_tgt_desc    *tgt;
1465         int                     rc;
1466         ENTRY;
1467
1468         tgt = lmv_find_target(lmv, fid);
1469         if (IS_ERR(tgt))
1470                 RETURN(PTR_ERR(tgt));
1471
1472         rc = md_setxattr(tgt->ltd_exp, fid, valid, name, input,
1473                          input_size, output_size, flags, suppgid,
1474                          request);
1475
1476         RETURN(rc);
1477 }
1478
1479 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1480                        struct ptlrpc_request **request)
1481 {
1482         struct obd_device       *obd = exp->exp_obd;
1483         struct lmv_obd          *lmv = &obd->u.lmv;
1484         struct lmv_tgt_desc     *tgt;
1485         int                      rc;
1486         ENTRY;
1487
1488         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1489         if (IS_ERR(tgt))
1490                 RETURN(PTR_ERR(tgt));
1491
1492         if (op_data->op_flags & MF_GET_MDT_IDX) {
1493                 op_data->op_mds = tgt->ltd_idx;
1494                 RETURN(0);
1495         }
1496
1497         rc = md_getattr(tgt->ltd_exp, op_data, request);
1498
1499         RETURN(rc);
1500 }
1501
1502 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1503 {
1504         struct obd_device   *obd = exp->exp_obd;
1505         struct lmv_obd      *lmv = &obd->u.lmv;
1506         __u32                i;
1507         ENTRY;
1508
1509         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1510
1511         /*
1512          * With DNE every object can have two locks in different namespaces:
1513          * lookup lock in space of MDT storing direntry and update/open lock in
1514          * space of MDT storing inode.
1515          */
1516         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1517                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1518                         continue;
1519                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1520         }
1521
1522         RETURN(0);
1523 }
1524
1525 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1526                      struct md_open_data *mod, struct ptlrpc_request **request)
1527 {
1528         struct obd_device     *obd = exp->exp_obd;
1529         struct lmv_obd        *lmv = &obd->u.lmv;
1530         struct lmv_tgt_desc   *tgt;
1531         int                    rc;
1532         ENTRY;
1533
1534         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1535         if (IS_ERR(tgt))
1536                 RETURN(PTR_ERR(tgt));
1537
1538         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1539         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1540         RETURN(rc);
1541 }
1542
1543 /**
1544  * Choosing the MDT by name or FID in @op_data.
1545  * For non-striped directory, it will locate MDT by fid.
1546  * For striped-directory, it will locate MDT by name. And also
1547  * it will reset op_fid1 with the FID of the choosen stripe.
1548  **/
1549 struct lmv_tgt_desc *
1550 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1551                            const char *name, int namelen, struct lu_fid *fid,
1552                            u32 *mds)
1553 {
1554         struct lmv_tgt_desc     *tgt;
1555         const struct lmv_oinfo  *oinfo;
1556
1557         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1558                 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1559                         RETURN(ERR_PTR(-EBADF));
1560                 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1561         } else {
1562                 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1563                 if (IS_ERR(oinfo))
1564                         RETURN(ERR_CAST(oinfo));
1565         }
1566
1567         if (fid != NULL)
1568                 *fid = oinfo->lmo_fid;
1569         if (mds != NULL)
1570                 *mds = oinfo->lmo_mds;
1571
1572         tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1573
1574         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", oinfo->lmo_mds,
1575                PFID(&oinfo->lmo_fid));
1576         return tgt;
1577 }
1578
1579 /**
1580  * Locate mds by fid or name
1581  *
1582  * For striped directory (lsm != NULL), it will locate the stripe
1583  * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1584  * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1585  * walk through all of stripes to locate the entry.
1586  *
1587  * For normal direcotry, it will locate MDS by FID directly.
1588  * \param[in] lmv       LMV device
1589  * \param[in] op_data   client MD stack parameters, name, namelen
1590  *                      mds_num etc.
1591  * \param[in] fid       object FID used to locate MDS.
1592  *
1593  * retval               pointer to the lmv_tgt_desc if succeed.
1594  *                      ERR_PTR(errno) if failed.
1595  */
1596 struct lmv_tgt_desc*
1597 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1598                struct lu_fid *fid)
1599 {
1600         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1601         struct lmv_tgt_desc     *tgt;
1602
1603         /* During creating VOLATILE file, it should honor the mdt
1604          * index if the file under striped dir is being restored, see
1605          * ct_restore(). */
1606         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1607             (int)op_data->op_mds != -1) {
1608                 int i;
1609                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1610                 if (IS_ERR(tgt))
1611                         return tgt;
1612
1613                 if (lsm != NULL) {
1614                         /* refill the right parent fid */
1615                         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1616                                 struct lmv_oinfo *oinfo;
1617
1618                                 oinfo = &lsm->lsm_md_oinfo[i];
1619                                 if (oinfo->lmo_mds == op_data->op_mds) {
1620                                         *fid = oinfo->lmo_fid;
1621                                         break;
1622                                 }
1623                         }
1624
1625                         if (i == lsm->lsm_md_stripe_count)
1626                                 *fid = lsm->lsm_md_oinfo[0].lmo_fid;
1627                 }
1628
1629                 return tgt;
1630         }
1631
1632         if (lsm == NULL || op_data->op_namelen == 0) {
1633                 tgt = lmv_find_target(lmv, fid);
1634                 if (IS_ERR(tgt))
1635                         return tgt;
1636
1637                 op_data->op_mds = tgt->ltd_idx;
1638                 return tgt;
1639         }
1640
1641         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1642                                           op_data->op_namelen, fid,
1643                                           &op_data->op_mds);
1644 }
1645
1646 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1647                 const void *data, size_t datalen, umode_t mode, uid_t uid,
1648                 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1649                 struct ptlrpc_request **request)
1650 {
1651         struct obd_device       *obd = exp->exp_obd;
1652         struct lmv_obd          *lmv = &obd->u.lmv;
1653         struct lmv_tgt_desc     *tgt;
1654         int                      rc;
1655         ENTRY;
1656
1657         if (!lmv->desc.ld_active_tgt_count)
1658                 RETURN(-EIO);
1659
1660         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1661         if (IS_ERR(tgt))
1662                 RETURN(PTR_ERR(tgt));
1663
1664         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1665                 (int)op_data->op_namelen, op_data->op_name,
1666                 PFID(&op_data->op_fid1), op_data->op_mds);
1667
1668         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1669         if (rc)
1670                 RETURN(rc);
1671         if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1672                 /* Send the create request to the MDT where the object
1673                  * will be located */
1674                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1675                 if (IS_ERR(tgt))
1676                         RETURN(PTR_ERR(tgt));
1677
1678                 op_data->op_mds = tgt->ltd_idx;
1679         } else {
1680                 CDEBUG(D_CONFIG, "Server doesn't support striped dirs\n");
1681         }
1682
1683         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1684                PFID(&op_data->op_fid2), op_data->op_mds);
1685
1686         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1687         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1688                        cap_effective, rdev, request);
1689         if (rc == 0) {
1690                 if (*request == NULL)
1691                         RETURN(rc);
1692                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1693         }
1694         RETURN(rc);
1695 }
1696
1697 static int
1698 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1699             const union ldlm_policy_data *policy, struct md_op_data *op_data,
1700             struct lustre_handle *lockh, __u64 extra_lock_flags)
1701 {
1702         struct obd_device        *obd = exp->exp_obd;
1703         struct lmv_obd           *lmv = &obd->u.lmv;
1704         struct lmv_tgt_desc      *tgt;
1705         int                       rc;
1706         ENTRY;
1707
1708         CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
1709
1710         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1711         if (IS_ERR(tgt))
1712                 RETURN(PTR_ERR(tgt));
1713
1714         CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
1715                PFID(&op_data->op_fid1), tgt->ltd_idx);
1716
1717         rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
1718                         extra_lock_flags);
1719
1720         RETURN(rc);
1721 }
1722
1723 static int
1724 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1725                  struct ptlrpc_request **preq)
1726 {
1727         struct ptlrpc_request   *req = NULL;
1728         struct obd_device       *obd = exp->exp_obd;
1729         struct lmv_obd          *lmv = &obd->u.lmv;
1730         struct lmv_tgt_desc     *tgt;
1731         struct mdt_body         *body;
1732         int                      rc;
1733         ENTRY;
1734
1735         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1736         if (IS_ERR(tgt))
1737                 RETURN(PTR_ERR(tgt));
1738
1739         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1740                 (int)op_data->op_namelen, op_data->op_name,
1741                 PFID(&op_data->op_fid1), tgt->ltd_idx);
1742
1743         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1744         if (rc != 0)
1745                 RETURN(rc);
1746
1747         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1748         LASSERT(body != NULL);
1749
1750         if (body->mbo_valid & OBD_MD_MDS) {
1751                 struct lu_fid rid = body->mbo_fid1;
1752                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1753                        PFID(&rid));
1754
1755                 tgt = lmv_find_target(lmv, &rid);
1756                 if (IS_ERR(tgt)) {
1757                         ptlrpc_req_finished(*preq);
1758                         preq = NULL;
1759                         RETURN(PTR_ERR(tgt));
1760                 }
1761
1762                 op_data->op_fid1 = rid;
1763                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1764                 op_data->op_namelen = 0;
1765                 op_data->op_name = NULL;
1766                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1767                 ptlrpc_req_finished(*preq);
1768                 *preq = req;
1769         }
1770
1771         RETURN(rc);
1772 }
1773
1774 #define md_op_data_fid(op_data, fl)                     \
1775         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1776          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1777          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1778          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1779          NULL)
1780
1781 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1782                             struct md_op_data *op_data, __u32 op_tgt,
1783                             enum ldlm_mode mode, int bits, int flag)
1784 {
1785         struct lu_fid *fid = md_op_data_fid(op_data, flag);
1786         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
1787         union ldlm_policy_data policy = { { 0 } };
1788         int rc = 0;
1789         ENTRY;
1790
1791         if (!fid_is_sane(fid))
1792                 RETURN(0);
1793
1794         if (tgt == NULL) {
1795                 tgt = lmv_find_target(lmv, fid);
1796                 if (IS_ERR(tgt))
1797                         RETURN(PTR_ERR(tgt));
1798         }
1799
1800         if (tgt->ltd_idx != op_tgt) {
1801                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1802                 policy.l_inodebits.bits = bits;
1803                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1804                                       mode, LCF_ASYNC, NULL);
1805         } else {
1806                 CDEBUG(D_INODE,
1807                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1808                        op_tgt, PFID(fid));
1809                 op_data->op_flags |= flag;
1810                 rc = 0;
1811         }
1812
1813         RETURN(rc);
1814 }
1815
1816 /*
1817  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1818  * op_data->op_fid2
1819  */
1820 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1821                     struct ptlrpc_request **request)
1822 {
1823         struct obd_device       *obd = exp->exp_obd;
1824         struct lmv_obd          *lmv = &obd->u.lmv;
1825         struct lmv_tgt_desc     *tgt;
1826         int                      rc;
1827         ENTRY;
1828
1829         LASSERT(op_data->op_namelen != 0);
1830
1831         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1832                PFID(&op_data->op_fid2), (int)op_data->op_namelen,
1833                op_data->op_name, PFID(&op_data->op_fid1));
1834
1835         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
1836         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
1837         op_data->op_cap = cfs_curproc_cap_pack();
1838         if (op_data->op_mea2 != NULL) {
1839                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
1840                 const struct lmv_oinfo  *oinfo;
1841
1842                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
1843                                                 op_data->op_namelen);
1844                 if (IS_ERR(oinfo))
1845                         RETURN(PTR_ERR(oinfo));
1846
1847                 op_data->op_fid2 = oinfo->lmo_fid;
1848         }
1849
1850         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1851         if (IS_ERR(tgt))
1852                 RETURN(PTR_ERR(tgt));
1853
1854         /*
1855          * Cancel UPDATE lock on child (fid1).
1856          */
1857         op_data->op_flags |= MF_MDC_CANCEL_FID2;
1858         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
1859                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
1860         if (rc != 0)
1861                 RETURN(rc);
1862
1863         rc = md_link(tgt->ltd_exp, op_data, request);
1864
1865         RETURN(rc);
1866 }
1867
1868 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
1869                       const char *old, size_t oldlen,
1870                       const char *new, size_t newlen,
1871                       struct ptlrpc_request **request)
1872 {
1873         struct obd_device       *obd = exp->exp_obd;
1874         struct lmv_obd          *lmv = &obd->u.lmv;
1875         struct lmv_tgt_desc     *src_tgt;
1876         struct lmv_tgt_desc     *tgt_tgt;
1877         struct obd_export       *target_exp;
1878         struct mdt_body         *body;
1879         int                     rc;
1880         ENTRY;
1881
1882         LASSERT(oldlen != 0);
1883
1884         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
1885                (int)oldlen, old, PFID(&op_data->op_fid1),
1886                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
1887                (int)newlen, new, PFID(&op_data->op_fid2),
1888                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
1889
1890         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
1891         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
1892         op_data->op_cap = cfs_curproc_cap_pack();
1893         if (op_data->op_cli_flags & CLI_MIGRATE) {
1894                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
1895                          PFID(&op_data->op_fid3));
1896
1897                 if (op_data->op_mea1 != NULL) {
1898                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1899                         struct lmv_tgt_desc     *tmp;
1900
1901                         /* Fix the parent fid for striped dir */
1902                         tmp = lmv_locate_target_for_name(lmv, lsm, old,
1903                                                          oldlen,
1904                                                          &op_data->op_fid1,
1905                                                          NULL);
1906                         if (IS_ERR(tmp))
1907                                 RETURN(PTR_ERR(tmp));
1908                 }
1909
1910                 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1911                 if (rc != 0)
1912                         RETURN(rc);
1913
1914                 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
1915                 if (IS_ERR(src_tgt))
1916                         RETURN(PTR_ERR(src_tgt));
1917
1918                 target_exp = src_tgt->ltd_exp;
1919         } else {
1920                 if (op_data->op_mea1 != NULL) {
1921                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1922
1923                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
1924                                                              oldlen,
1925                                                              &op_data->op_fid1,
1926                                                              &op_data->op_mds);
1927                 } else {
1928                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
1929                 }
1930                 if (IS_ERR(src_tgt))
1931                         RETURN(PTR_ERR(src_tgt));
1932
1933
1934                 if (op_data->op_mea2 != NULL) {
1935                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
1936
1937                         tgt_tgt = lmv_locate_target_for_name(lmv, lsm, new,
1938                                                              newlen,
1939                                                              &op_data->op_fid2,
1940                                                              &op_data->op_mds);
1941                 } else {
1942                         tgt_tgt = lmv_find_target(lmv, &op_data->op_fid2);
1943
1944                 }
1945                 if (IS_ERR(tgt_tgt))
1946                         RETURN(PTR_ERR(tgt_tgt));
1947
1948                 target_exp = tgt_tgt->ltd_exp;
1949         }
1950
1951         /*
1952          * LOOKUP lock on src child (fid3) should also be cancelled for
1953          * src_tgt in mdc_rename.
1954          */
1955         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
1956
1957         /*
1958          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
1959          * own target.
1960          */
1961         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
1962                               LCK_EX, MDS_INODELOCK_UPDATE,
1963                               MF_MDC_CANCEL_FID2);
1964
1965         if (rc != 0)
1966                 RETURN(rc);
1967         /*
1968          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
1969          */
1970         if (fid_is_sane(&op_data->op_fid3)) {
1971                 struct lmv_tgt_desc *tgt;
1972
1973                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1974                 if (IS_ERR(tgt))
1975                         RETURN(PTR_ERR(tgt));
1976
1977                 /* Cancel LOOKUP lock on its parent */
1978                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
1979                                       LCK_EX, MDS_INODELOCK_LOOKUP,
1980                                       MF_MDC_CANCEL_FID3);
1981                 if (rc != 0)
1982                         RETURN(rc);
1983
1984                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
1985                                       LCK_EX, MDS_INODELOCK_FULL,
1986                                       MF_MDC_CANCEL_FID3);
1987                 if (rc != 0)
1988                         RETURN(rc);
1989         }
1990
1991 retry_rename:
1992         /*
1993          * Cancel all the locks on tgt child (fid4).
1994          */
1995         if (fid_is_sane(&op_data->op_fid4)) {
1996                 struct lmv_tgt_desc *tgt;
1997
1998                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
1999                                       LCK_EX, MDS_INODELOCK_FULL,
2000                                       MF_MDC_CANCEL_FID4);
2001                 if (rc != 0)
2002                         RETURN(rc);
2003
2004                 tgt = lmv_find_target(lmv, &op_data->op_fid4);
2005                 if (IS_ERR(tgt))
2006                         RETURN(PTR_ERR(tgt));
2007
2008                 /* Since the target child might be destroyed, and it might
2009                  * become orphan, and we can only check orphan on the local
2010                  * MDT right now, so we send rename request to the MDT where
2011                  * target child is located. If target child does not exist,
2012                  * then it will send the request to the target parent */
2013                 target_exp = tgt->ltd_exp;
2014         }
2015
2016         rc = md_rename(target_exp, op_data, old, oldlen, new, newlen,
2017                        request);
2018
2019         if (rc != 0 && rc != -EXDEV)
2020                 RETURN(rc);
2021
2022         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2023         if (body == NULL)
2024                 RETURN(-EPROTO);
2025
2026         /* Not cross-ref case, just get out of here. */
2027         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2028                 RETURN(rc);
2029
2030         CDEBUG(D_INODE, "%s: try rename to another MDT for "DFID"\n",
2031                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2032
2033         op_data->op_fid4 = body->mbo_fid1;
2034         ptlrpc_req_finished(*request);
2035         *request = NULL;
2036         goto retry_rename;
2037 }
2038
2039 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2040                        void *ea, size_t ealen, struct ptlrpc_request **request)
2041 {
2042         struct obd_device       *obd = exp->exp_obd;
2043         struct lmv_obd          *lmv = &obd->u.lmv;
2044         struct lmv_tgt_desc     *tgt;
2045         int                      rc = 0;
2046         ENTRY;
2047
2048         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2049                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2050
2051         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2052         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2053         if (IS_ERR(tgt))
2054                 RETURN(PTR_ERR(tgt));
2055
2056         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2057
2058         RETURN(rc);
2059 }
2060
2061 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2062                      struct ptlrpc_request **request)
2063 {
2064         struct obd_device       *obd = exp->exp_obd;
2065         struct lmv_obd          *lmv = &obd->u.lmv;
2066         struct lmv_tgt_desc     *tgt;
2067         int                      rc;
2068         ENTRY;
2069
2070         tgt = lmv_find_target(lmv, fid);
2071         if (IS_ERR(tgt))
2072                 RETURN(PTR_ERR(tgt));
2073
2074         rc = md_fsync(tgt->ltd_exp, fid, request);
2075         RETURN(rc);
2076 }
2077
2078 /**
2079  * Get current minimum entry from striped directory
2080  *
2081  * This function will search the dir entry, whose hash value is the
2082  * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2083  * only being called for striped directory.
2084  *
2085  * \param[in] exp               export of LMV
2086  * \param[in] op_data           parameters transferred beween client MD stack
2087  *                              stripe_information will be included in this
2088  *                              parameter
2089  * \param[in] cb_op             ldlm callback being used in enqueue in
2090  *                              mdc_read_page
2091  * \param[in] hash_offset       the hash value, which is used to locate
2092  *                              minum(closet) dir entry
2093  * \param[in|out] stripe_offset the caller use this to indicate the stripe
2094  *                              index of last entry, so to avoid hash conflict
2095  *                              between stripes. It will also be used to
2096  *                              return the stripe index of current dir entry.
2097  * \param[in|out] entp          the minum entry and it also is being used
2098  *                              to input the last dir entry to resolve the
2099  *                              hash conflict
2100  *
2101  * \param[out] ppage            the page which holds the minum entry
2102  *
2103  * \retval                      = 0 get the entry successfully
2104  *                              negative errno (< 0) does not get the entry
2105  */
2106 static int lmv_get_min_striped_entry(struct obd_export *exp,
2107                                      struct md_op_data *op_data,
2108                                      struct md_callback *cb_op,
2109                                      __u64 hash_offset, int *stripe_offset,
2110                                      struct lu_dirent **entp,
2111                                      struct page **ppage)
2112 {
2113         struct obd_device       *obd = exp->exp_obd;
2114         struct lmv_obd          *lmv = &obd->u.lmv;
2115         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2116         struct lmv_tgt_desc     *tgt;
2117         int                     stripe_count;
2118         struct lu_dirent        *min_ent = NULL;
2119         struct page             *min_page = NULL;
2120         int                     min_idx = 0;
2121         int                     i;
2122         int                     rc = 0;
2123         ENTRY;
2124
2125         stripe_count = lsm->lsm_md_stripe_count;
2126         for (i = 0; i < stripe_count; i++) {
2127                 struct lu_dirent        *ent = NULL;
2128                 struct page             *page = NULL;
2129                 struct lu_dirpage       *dp;
2130                 __u64                   stripe_hash = hash_offset;
2131
2132                 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2133                 if (IS_ERR(tgt))
2134                         GOTO(out, rc = PTR_ERR(tgt));
2135
2136                 /* op_data will be shared by each stripe, so we need
2137                  * reset these value for each stripe */
2138                 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2139                 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2140                 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2141 next:
2142                 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2143                                   &page);
2144                 if (rc != 0)
2145                         GOTO(out, rc);
2146
2147                 dp = page_address(page);
2148                 for (ent = lu_dirent_start(dp); ent != NULL;
2149                      ent = lu_dirent_next(ent)) {
2150                         /* Skip dummy entry */
2151                         if (le16_to_cpu(ent->lde_namelen) == 0)
2152                                 continue;
2153
2154                         if (le64_to_cpu(ent->lde_hash) < hash_offset)
2155                                 continue;
2156
2157                         if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2158                             (*entp == ent || i < *stripe_offset))
2159                                 continue;
2160
2161                         /* skip . and .. for other stripes */
2162                         if (i != 0 &&
2163                             (strncmp(ent->lde_name, ".",
2164                                      le16_to_cpu(ent->lde_namelen)) == 0 ||
2165                              strncmp(ent->lde_name, "..",
2166                                      le16_to_cpu(ent->lde_namelen)) == 0))
2167                                 continue;
2168                         break;
2169                 }
2170
2171                 if (ent == NULL) {
2172                         stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2173
2174                         kunmap(page);
2175                         put_page(page);
2176                         page = NULL;
2177
2178                         /* reach the end of current stripe, go to next stripe */
2179                         if (stripe_hash == MDS_DIR_END_OFF)
2180                                 continue;
2181                         else
2182                                 goto next;
2183                 }
2184
2185                 if (min_ent != NULL) {
2186                         if (le64_to_cpu(min_ent->lde_hash) >
2187                             le64_to_cpu(ent->lde_hash)) {
2188                                 min_ent = ent;
2189                                 kunmap(min_page);
2190                                 put_page(min_page);
2191                                 min_idx = i;
2192                                 min_page = page;
2193                         } else {
2194                                 kunmap(page);
2195                                 put_page(page);
2196                                 page = NULL;
2197                         }
2198                 } else {
2199                         min_ent = ent;
2200                         min_page = page;
2201                         min_idx = i;
2202                 }
2203         }
2204
2205 out:
2206         if (*ppage != NULL) {
2207                 kunmap(*ppage);
2208                 put_page(*ppage);
2209         }
2210         *stripe_offset = min_idx;
2211         *entp = min_ent;
2212         *ppage = min_page;
2213         RETURN(rc);
2214 }
2215
2216 /**
2217  * Build dir entry page from a striped directory
2218  *
2219  * This function gets one entry by @offset from a striped directory. It will
2220  * read entries from all of stripes, and choose one closest to the required
2221  * offset(&offset). A few notes
2222  * 1. skip . and .. for non-zero stripes, because there can only have one .
2223  * and .. in a directory.
2224  * 2. op_data will be shared by all of stripes, instead of allocating new
2225  * one, so need to restore before reusing.
2226  * 3. release the entry page if that is not being chosen.
2227  *
2228  * \param[in] exp       obd export refer to LMV
2229  * \param[in] op_data   hold those MD parameters of read_entry
2230  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2231  * \param[out] ldp      the entry being read
2232  * \param[out] ppage    the page holding the entry. Note: because the entry
2233  *                      will be accessed in upper layer, so we need hold the
2234  *                      page until the usages of entry is finished, see
2235  *                      ll_dir_entry_next.
2236  *
2237  * retval               =0 if get entry successfully
2238  *                      <0 cannot get entry
2239  */
2240 static int lmv_read_striped_page(struct obd_export *exp,
2241                                  struct md_op_data *op_data,
2242                                  struct md_callback *cb_op,
2243                                  __u64 offset, struct page **ppage)
2244 {
2245         struct lu_fid           master_fid = op_data->op_fid1;
2246         struct inode            *master_inode = op_data->op_data;
2247         __u64                   hash_offset = offset;
2248         struct lu_dirpage       *dp;
2249         struct page             *min_ent_page = NULL;
2250         struct page             *ent_page = NULL;
2251         struct lu_dirent        *ent;
2252         void                    *area;
2253         int                     ent_idx = 0;
2254         struct lu_dirent        *min_ent = NULL;
2255         struct lu_dirent        *last_ent;
2256         size_t                  left_bytes;
2257         int                     rc;
2258         ENTRY;
2259
2260         /* Allocate a page and read entries from all of stripes and fill
2261          * the page by hash order */
2262         ent_page = alloc_page(GFP_KERNEL);
2263         if (ent_page == NULL)
2264                 RETURN(-ENOMEM);
2265
2266         /* Initialize the entry page */
2267         dp = kmap(ent_page);
2268         memset(dp, 0, sizeof(*dp));
2269         dp->ldp_hash_start = cpu_to_le64(offset);
2270         dp->ldp_flags |= LDF_COLLIDE;
2271
2272         area = dp + 1;
2273         left_bytes = PAGE_SIZE - sizeof(*dp);
2274         ent = area;
2275         last_ent = ent;
2276         do {
2277                 __u16   ent_size;
2278
2279                 /* Find the minum entry from all sub-stripes */
2280                 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2281                                                &ent_idx, &min_ent,
2282                                                &min_ent_page);
2283                 if (rc != 0)
2284                         GOTO(out, rc);
2285
2286                 /* If it can not get minum entry, it means it already reaches
2287                  * the end of this directory */
2288                 if (min_ent == NULL) {
2289                         last_ent->lde_reclen = 0;
2290                         hash_offset = MDS_DIR_END_OFF;
2291                         GOTO(out, rc);
2292                 }
2293
2294                 ent_size = le16_to_cpu(min_ent->lde_reclen);
2295
2296                 /* the last entry lde_reclen is 0, but it might not
2297                  * the end of this entry of this temporay entry */
2298                 if (ent_size == 0)
2299                         ent_size = lu_dirent_calc_size(
2300                                         le16_to_cpu(min_ent->lde_namelen),
2301                                         le32_to_cpu(min_ent->lde_attrs));
2302                 if (ent_size > left_bytes) {
2303                         last_ent->lde_reclen = cpu_to_le16(0);
2304                         hash_offset = le64_to_cpu(min_ent->lde_hash);
2305                         GOTO(out, rc);
2306                 }
2307
2308                 memcpy(ent, min_ent, ent_size);
2309
2310                 /* Replace . with master FID and Replace .. with the parent FID
2311                  * of master object */
2312                 if (strncmp(ent->lde_name, ".",
2313                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2314                     le16_to_cpu(ent->lde_namelen) == 1)
2315                         fid_cpu_to_le(&ent->lde_fid, &master_fid);
2316                 else if (strncmp(ent->lde_name, "..",
2317                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2318                            le16_to_cpu(ent->lde_namelen) == 2)
2319                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2320
2321                 left_bytes -= ent_size;
2322                 ent->lde_reclen = cpu_to_le16(ent_size);
2323                 last_ent = ent;
2324                 ent = (void *)ent + ent_size;
2325                 hash_offset = le64_to_cpu(min_ent->lde_hash);
2326                 if (hash_offset == MDS_DIR_END_OFF) {
2327                         last_ent->lde_reclen = 0;
2328                         break;
2329                 }
2330         } while (1);
2331 out:
2332         if (min_ent_page != NULL) {
2333                 kunmap(min_ent_page);
2334                 put_page(min_ent_page);
2335         }
2336
2337         if (unlikely(rc != 0)) {
2338                 __free_page(ent_page);
2339                 ent_page = NULL;
2340         } else {
2341                 if (ent == area)
2342                         dp->ldp_flags |= LDF_EMPTY;
2343                 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2344                 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2345         }
2346
2347         /* We do not want to allocate md_op_data during each
2348          * dir entry reading, so op_data will be shared by every stripe,
2349          * then we need to restore it back to original value before
2350          * return to the upper layer */
2351         op_data->op_fid1 = master_fid;
2352         op_data->op_fid2 = master_fid;
2353         op_data->op_data = master_inode;
2354
2355         *ppage = ent_page;
2356
2357         RETURN(rc);
2358 }
2359
2360 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2361                   struct md_callback *cb_op, __u64 offset,
2362                   struct page **ppage)
2363 {
2364         struct obd_device       *obd = exp->exp_obd;
2365         struct lmv_obd          *lmv = &obd->u.lmv;
2366         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2367         struct lmv_tgt_desc     *tgt;
2368         int                     rc;
2369         ENTRY;
2370
2371         if (unlikely(lsm != NULL)) {
2372                 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2373                 RETURN(rc);
2374         }
2375
2376         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2377         if (IS_ERR(tgt))
2378                 RETURN(PTR_ERR(tgt));
2379
2380         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2381
2382         RETURN(rc);
2383 }
2384
2385 /**
2386  * Unlink a file/directory
2387  *
2388  * Unlink a file or directory under the parent dir. The unlink request
2389  * usually will be sent to the MDT where the child is located, but if
2390  * the client does not have the child FID then request will be sent to the
2391  * MDT where the parent is located.
2392  *
2393  * If the parent is a striped directory then it also needs to locate which
2394  * stripe the name of the child is located, and replace the parent FID
2395  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2396  * it will walk through all of sub-stripes until the child is being
2397  * unlinked finally.
2398  *
2399  * \param[in] exp       export refer to LMV
2400  * \param[in] op_data   different parameters transferred beween client
2401  *                      MD stacks, name, namelen, FIDs etc.
2402  *                      op_fid1 is the parent FID, op_fid2 is the child
2403  *                      FID.
2404  * \param[out] request  point to the request of unlink.
2405  *
2406  * retval               0 if succeed
2407  *                      negative errno if failed.
2408  */
2409 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2410                       struct ptlrpc_request **request)
2411 {
2412         struct obd_device       *obd = exp->exp_obd;
2413         struct lmv_obd          *lmv = &obd->u.lmv;
2414         struct lmv_tgt_desc     *tgt = NULL;
2415         struct lmv_tgt_desc     *parent_tgt = NULL;
2416         struct mdt_body         *body;
2417         int                     rc;
2418         int                     stripe_index = 0;
2419         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2420         ENTRY;
2421
2422 retry_unlink:
2423         /* For striped dir, we need to locate the parent as well */
2424         if (lsm != NULL) {
2425                 struct lmv_tgt_desc *tmp;
2426
2427                 LASSERT(op_data->op_name != NULL &&
2428                         op_data->op_namelen != 0);
2429
2430                 tmp = lmv_locate_target_for_name(lmv, lsm,
2431                                                  op_data->op_name,
2432                                                  op_data->op_namelen,
2433                                                  &op_data->op_fid1,
2434                                                  &op_data->op_mds);
2435
2436                 /* return -EBADFD means unknown hash type, might
2437                  * need try all sub-stripe here */
2438                 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2439                         RETURN(PTR_ERR(tmp));
2440
2441                 /* Note: both migrating dir and unknown hash dir need to
2442                  * try all of sub-stripes, so we need start search the
2443                  * name from stripe 0, but migrating dir is already handled
2444                  * inside lmv_locate_target_for_name(), so we only check
2445                  * unknown hash type directory here */
2446                 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2447                         struct lmv_oinfo *oinfo;
2448
2449                         oinfo = &lsm->lsm_md_oinfo[stripe_index];
2450
2451                         op_data->op_fid1 = oinfo->lmo_fid;
2452                         op_data->op_mds = oinfo->lmo_mds;
2453                 }
2454         }
2455
2456 try_next_stripe:
2457         /* Send unlink requests to the MDT where the child is located */
2458         if (likely(!fid_is_zero(&op_data->op_fid2)))
2459                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2460         else if (lsm != NULL)
2461                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2462         else
2463                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2464
2465         if (IS_ERR(tgt))
2466                 RETURN(PTR_ERR(tgt));
2467
2468         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2469         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2470         op_data->op_cap = cfs_curproc_cap_pack();
2471
2472         /*
2473          * If child's fid is given, cancel unused locks for it if it is from
2474          * another export than parent.
2475          *
2476          * LOOKUP lock for child (fid3) should also be cancelled on parent
2477          * tgt_tgt in mdc_unlink().
2478          */
2479         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2480
2481         /*
2482          * Cancel FULL locks on child (fid3).
2483          */
2484         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2485         if (IS_ERR(parent_tgt))
2486                 RETURN(PTR_ERR(parent_tgt));
2487
2488         if (parent_tgt != tgt) {
2489                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2490                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2491                                       MF_MDC_CANCEL_FID3);
2492         }
2493
2494         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2495                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2496         if (rc != 0)
2497                 RETURN(rc);
2498
2499         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2500                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2501
2502         rc = md_unlink(tgt->ltd_exp, op_data, request);
2503         if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2504                 RETURN(rc);
2505
2506         /* Try next stripe if it is needed. */
2507         if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2508                 struct lmv_oinfo *oinfo;
2509
2510                 stripe_index++;
2511                 if (stripe_index >= lsm->lsm_md_stripe_count)
2512                         RETURN(rc);
2513
2514                 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2515
2516                 op_data->op_fid1 = oinfo->lmo_fid;
2517                 op_data->op_mds = oinfo->lmo_mds;
2518
2519                 ptlrpc_req_finished(*request);
2520                 *request = NULL;
2521
2522                 goto try_next_stripe;
2523         }
2524
2525         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2526         if (body == NULL)
2527                 RETURN(-EPROTO);
2528
2529         /* Not cross-ref case, just get out of here. */
2530         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2531                 RETURN(rc);
2532
2533         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2534                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2535
2536         /* This is a remote object, try remote MDT, Note: it may
2537          * try more than 1 time here, Considering following case
2538          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2539          * 1. Initially A does not know where remote1 is, it send
2540          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2541          *    resend unlink RPC to MDT1 (retry 1st time).
2542          *
2543          * 2. During the unlink RPC in flight,
2544          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2545          *    and create new remote1, but on MDT0
2546          *
2547          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2548          *    /mnt/lustre, then lookup get fid of remote1, and find
2549          *    it is remote dir again, and replay -EREMOTE again.
2550          *
2551          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2552          *
2553          * In theory, it might try unlimited time here, but it should
2554          * be very rare case.  */
2555         op_data->op_fid2 = body->mbo_fid1;
2556         ptlrpc_req_finished(*request);
2557         *request = NULL;
2558
2559         goto retry_unlink;
2560 }
2561
2562 static int lmv_precleanup(struct obd_device *obd)
2563 {
2564         ENTRY;
2565         fld_client_proc_fini(&obd->u.lmv.lmv_fld);
2566         lprocfs_obd_cleanup(obd);
2567         lprocfs_free_md_stats(obd);
2568         RETURN(0);
2569 }
2570
2571 /**
2572  * Get by key a value associated with a LMV device.
2573  *
2574  * Dispatch request to lower-layer devices as needed.
2575  *
2576  * \param[in] env               execution environment for this thread
2577  * \param[in] exp               export for the LMV device
2578  * \param[in] keylen            length of key identifier
2579  * \param[in] key               identifier of key to get value for
2580  * \param[in] vallen            size of \a val
2581  * \param[out] val              pointer to storage location for value
2582  * \param[in] lsm               optional striping metadata of object
2583  *
2584  * \retval 0            on success
2585  * \retval negative     negated errno on failure
2586  */
2587 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2588                         __u32 keylen, void *key, __u32 *vallen, void *val)
2589 {
2590         struct obd_device       *obd;
2591         struct lmv_obd          *lmv;
2592         int                      rc = 0;
2593         ENTRY;
2594
2595         obd = class_exp2obd(exp);
2596         if (obd == NULL) {
2597                 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
2598                        exp->exp_handle.h_cookie);
2599                 RETURN(-EINVAL);
2600         }
2601
2602         lmv = &obd->u.lmv;
2603         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2604                 int i;
2605
2606                 LASSERT(*vallen == sizeof(__u32));
2607                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2608                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2609                         /*
2610                          * All tgts should be connected when this gets called.
2611                          */
2612                         if (tgt == NULL || tgt->ltd_exp == NULL)
2613                                 continue;
2614
2615                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2616                                           vallen, val))
2617                                 RETURN(0);
2618                 }
2619                 RETURN(-EINVAL);
2620         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2621                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2622                    KEY_IS(KEY_CONN_DATA)) {
2623                 /*
2624                  * Forwarding this request to first MDS, it should know LOV
2625                  * desc.
2626                  */
2627                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2628                                   vallen, val);
2629                 if (!rc && KEY_IS(KEY_CONN_DATA))
2630                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2631                 RETURN(rc);
2632         } else if (KEY_IS(KEY_TGT_COUNT)) {
2633                 *((int *)val) = lmv->desc.ld_tgt_count;
2634                 RETURN(0);
2635         }
2636
2637         CDEBUG(D_IOCTL, "Invalid key\n");
2638         RETURN(-EINVAL);
2639 }
2640
2641 /**
2642  * Asynchronously set by key a value associated with a LMV device.
2643  *
2644  * Dispatch request to lower-layer devices as needed.
2645  *
2646  * \param[in] env       execution environment for this thread
2647  * \param[in] exp       export for the LMV device
2648  * \param[in] keylen    length of key identifier
2649  * \param[in] key       identifier of key to store value for
2650  * \param[in] vallen    size of value to store
2651  * \param[in] val       pointer to data to be stored
2652  * \param[in] set       optional list of related ptlrpc requests
2653  *
2654  * \retval 0            on success
2655  * \retval negative     negated errno on failure
2656  */
2657 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2658                         __u32 keylen, void *key, __u32 vallen, void *val,
2659                         struct ptlrpc_request_set *set)
2660 {
2661         struct lmv_tgt_desc     *tgt = NULL;
2662         struct obd_device       *obd;
2663         struct lmv_obd          *lmv;
2664         int rc = 0;
2665         ENTRY;
2666
2667         obd = class_exp2obd(exp);
2668         if (obd == NULL) {
2669                 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
2670                        exp->exp_handle.h_cookie);
2671                 RETURN(-EINVAL);
2672         }
2673         lmv = &obd->u.lmv;
2674
2675         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2676             KEY_IS(KEY_DEFAULT_EASIZE)) {
2677                 int i, err = 0;
2678
2679                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2680                         tgt = lmv->tgts[i];
2681
2682                         if (tgt == NULL || tgt->ltd_exp == NULL)
2683                                 continue;
2684
2685                         err = obd_set_info_async(env, tgt->ltd_exp,
2686                                                  keylen, key, vallen, val, set);
2687                         if (err && rc == 0)
2688                                 rc = err;
2689                 }
2690
2691                 RETURN(rc);
2692         }
2693
2694         RETURN(-EINVAL);
2695 }
2696
2697 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2698                             const struct lmv_mds_md_v1 *lmm1)
2699 {
2700         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
2701         int             stripe_count;
2702         int             cplen;
2703         int             i;
2704         int             rc = 0;
2705         ENTRY;
2706
2707         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2708         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2709         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2710         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2711                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2712         else
2713                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2714         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2715         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2716                         sizeof(lsm->lsm_md_pool_name));
2717
2718         if (cplen >= sizeof(lsm->lsm_md_pool_name))
2719                 RETURN(-E2BIG);
2720
2721         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2722                "layout_version %d\n", lsm->lsm_md_stripe_count,
2723                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2724                lsm->lsm_md_layout_version);
2725
2726         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2727         for (i = 0; i < stripe_count; i++) {
2728                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2729                               &lmm1->lmv_stripe_fids[i]);
2730                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2731                                     &lsm->lsm_md_oinfo[i].lmo_mds);
2732                 if (rc != 0)
2733                         RETURN(rc);
2734                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2735                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2736         }
2737
2738         RETURN(rc);
2739 }
2740
2741 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2742                         const union lmv_mds_md *lmm, size_t lmm_size)
2743 {
2744         struct lmv_stripe_md     *lsm;
2745         int                      lsm_size;
2746         int                      rc;
2747         bool                     allocated = false;
2748         ENTRY;
2749
2750         LASSERT(lsmp != NULL);
2751
2752         lsm = *lsmp;
2753         /* Free memmd */
2754         if (lsm != NULL && lmm == NULL) {
2755                 int i;
2756                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2757                         /* For migrating inode, the master stripe and master
2758                          * object will be the same, so do not need iput, see
2759                          * ll_update_lsm_md */
2760                         if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2761                               i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2762                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
2763                 }
2764                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2765                 OBD_FREE(lsm, lsm_size);
2766                 *lsmp = NULL;
2767                 RETURN(0);
2768         }
2769
2770         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
2771                 RETURN(-EPERM);
2772
2773         /* Unpack memmd */
2774         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
2775             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
2776                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
2777                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
2778                        -EIO);
2779                 RETURN(-EIO);
2780         }
2781
2782         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
2783                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2784         else
2785                 /**
2786                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
2787                  * stripecount should be 0 then.
2788                  */
2789                 lsm_size = lmv_stripe_md_size(0);
2790
2791         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2792         if (lsm == NULL) {
2793                 OBD_ALLOC(lsm, lsm_size);
2794                 if (lsm == NULL)
2795                         RETURN(-ENOMEM);
2796                 allocated = true;
2797                 *lsmp = lsm;
2798         }
2799
2800         switch (le32_to_cpu(lmm->lmv_magic)) {
2801         case LMV_MAGIC_V1:
2802                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
2803                 break;
2804         default:
2805                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
2806                        le32_to_cpu(lmm->lmv_magic));
2807                 rc = -EINVAL;
2808                 break;
2809         }
2810
2811         if (rc != 0 && allocated) {
2812                 OBD_FREE(lsm, lsm_size);
2813                 *lsmp = NULL;
2814                 lsm_size = rc;
2815         }
2816         RETURN(lsm_size);
2817 }
2818
2819 void lmv_free_memmd(struct lmv_stripe_md *lsm)
2820 {
2821         lmv_unpackmd(NULL, &lsm, NULL, 0);
2822 }
2823 EXPORT_SYMBOL(lmv_free_memmd);
2824
2825 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2826                              union ldlm_policy_data *policy,
2827                              enum ldlm_mode mode, enum ldlm_cancel_flags flags,
2828                              void *opaque)
2829 {
2830         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2831         int rc = 0;
2832         __u32 i;
2833         ENTRY;
2834
2835         LASSERT(fid != NULL);
2836
2837         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2838                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2839                 int err;
2840
2841                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
2842                         continue;
2843
2844                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
2845                                        opaque);
2846                 if (!rc)
2847                         rc = err;
2848         }
2849         RETURN(rc);
2850 }
2851
2852 static int lmv_set_lock_data(struct obd_export *exp,
2853                              const struct lustre_handle *lockh,
2854                              void *data, __u64 *bits)
2855 {
2856         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2857         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
2858         int                      rc;
2859         ENTRY;
2860
2861         if (tgt == NULL || tgt->ltd_exp == NULL)
2862                 RETURN(-EINVAL);
2863         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
2864         RETURN(rc);
2865 }
2866
2867 enum ldlm_mode lmv_lock_match(struct obd_export *exp, __u64 flags,
2868                               const struct lu_fid *fid, enum ldlm_type type,
2869                               union ldlm_policy_data *policy,
2870                               enum ldlm_mode mode, struct lustre_handle *lockh)
2871 {
2872         struct obd_device       *obd = exp->exp_obd;
2873         struct lmv_obd          *lmv = &obd->u.lmv;
2874         enum ldlm_mode          rc;
2875         int                     tgt;
2876         int                     i;
2877         ENTRY;
2878
2879         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2880
2881         /*
2882          * With DNE every object can have two locks in different namespaces:
2883          * lookup lock in space of MDT storing direntry and update/open lock in
2884          * space of MDT storing inode.  Try the MDT that the FID maps to first,
2885          * since this can be easily found, and only try others if that fails.
2886          */
2887         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
2888              i < lmv->desc.ld_tgt_count;
2889              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
2890                 if (tgt < 0) {
2891                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
2892                                obd->obd_name, PFID(fid), tgt);
2893                         tgt = 0;
2894                 }
2895
2896                 if (lmv->tgts[tgt] == NULL ||
2897                     lmv->tgts[tgt]->ltd_exp == NULL ||
2898                     lmv->tgts[tgt]->ltd_active == 0)
2899                         continue;
2900
2901                 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
2902                                    type, policy, mode, lockh);
2903                 if (rc)
2904                         RETURN(rc);
2905         }
2906
2907         RETURN(0);
2908 }
2909
2910 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2911                       struct obd_export *dt_exp, struct obd_export *md_exp,
2912                       struct lustre_md *md)
2913 {
2914         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2915         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
2916
2917         if (tgt == NULL || tgt->ltd_exp == NULL)
2918                 RETURN(-EINVAL);
2919
2920         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
2921 }
2922
2923 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2924 {
2925         struct obd_device       *obd = exp->exp_obd;
2926         struct lmv_obd          *lmv = &obd->u.lmv;
2927         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
2928         ENTRY;
2929
2930         if (md->lmv != NULL) {
2931                 lmv_free_memmd(md->lmv);
2932                 md->lmv = NULL;
2933         }
2934         if (tgt == NULL || tgt->ltd_exp == NULL)
2935                 RETURN(-EINVAL);
2936         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
2937 }
2938
2939 int lmv_set_open_replay_data(struct obd_export *exp,
2940                              struct obd_client_handle *och,
2941                              struct lookup_intent *it)
2942 {
2943         struct obd_device       *obd = exp->exp_obd;
2944         struct lmv_obd          *lmv = &obd->u.lmv;
2945         struct lmv_tgt_desc     *tgt;
2946         ENTRY;
2947
2948         tgt = lmv_find_target(lmv, &och->och_fid);
2949         if (IS_ERR(tgt))
2950                 RETURN(PTR_ERR(tgt));
2951
2952         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
2953 }
2954
2955 int lmv_clear_open_replay_data(struct obd_export *exp,
2956                                struct obd_client_handle *och)
2957 {
2958         struct obd_device       *obd = exp->exp_obd;
2959         struct lmv_obd          *lmv = &obd->u.lmv;
2960         struct lmv_tgt_desc     *tgt;
2961         ENTRY;
2962
2963         tgt = lmv_find_target(lmv, &och->och_fid);
2964         if (IS_ERR(tgt))
2965                 RETURN(PTR_ERR(tgt));
2966
2967         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
2968 }
2969
2970 int lmv_intent_getattr_async(struct obd_export *exp,
2971                              struct md_enqueue_info *minfo)
2972 {
2973         struct md_op_data       *op_data = &minfo->mi_data;
2974         struct obd_device       *obd = exp->exp_obd;
2975         struct lmv_obd          *lmv = &obd->u.lmv;
2976         struct lmv_tgt_desc     *ptgt = NULL;
2977         struct lmv_tgt_desc     *ctgt = NULL;
2978         int                      rc;
2979         ENTRY;
2980
2981         if (!fid_is_sane(&op_data->op_fid2))
2982                 RETURN(-EINVAL);
2983
2984         ptgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2985         if (IS_ERR(ptgt))
2986                 RETURN(PTR_ERR(ptgt));
2987
2988         ctgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2989         if (IS_ERR(ctgt))
2990                 RETURN(PTR_ERR(ctgt));
2991
2992         /*
2993          * if child is on remote MDT, we need 2 async RPCs to fetch both LOOKUP
2994          * lock on parent, and UPDATE lock on child MDT, which makes all
2995          * complicated. Considering remote dir is rare case, and not supporting
2996          * it in statahead won't cause any issue, drop its support for now.
2997          */
2998         if (ptgt != ctgt)
2999                 RETURN(-ENOTSUPP);
3000
3001         rc = md_intent_getattr_async(ptgt->ltd_exp, minfo);
3002         RETURN(rc);
3003 }
3004
3005 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3006                         struct lu_fid *fid, __u64 *bits)
3007 {
3008         struct obd_device       *obd = exp->exp_obd;
3009         struct lmv_obd          *lmv = &obd->u.lmv;
3010         struct lmv_tgt_desc     *tgt;
3011         int                      rc;
3012         ENTRY;
3013
3014         tgt = lmv_find_target(lmv, fid);
3015         if (IS_ERR(tgt))
3016                 RETURN(PTR_ERR(tgt));
3017
3018         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3019         RETURN(rc);
3020 }
3021
3022 int lmv_get_fid_from_lsm(struct obd_export *exp,
3023                          const struct lmv_stripe_md *lsm,
3024                          const char *name, int namelen, struct lu_fid *fid)
3025 {
3026         const struct lmv_oinfo *oinfo;
3027
3028         LASSERT(lsm != NULL);
3029         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3030         if (IS_ERR(oinfo))
3031                 return PTR_ERR(oinfo);
3032
3033         *fid = oinfo->lmo_fid;
3034
3035         RETURN(0);
3036 }
3037
3038 /**
3039  * For lmv, only need to send request to master MDT, and the master MDT will
3040  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3041  * we directly fetch data from the slave MDTs.
3042  */
3043 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3044                  struct obd_quotactl *oqctl)
3045 {
3046         struct obd_device   *obd = class_exp2obd(exp);
3047         struct lmv_obd      *lmv = &obd->u.lmv;
3048         struct lmv_tgt_desc *tgt = lmv->tgts[0];
3049         int                  rc = 0;
3050         __u32                i;
3051         __u64                curspace, curinodes;
3052         ENTRY;
3053
3054         if (tgt == NULL ||
3055             tgt->ltd_exp == NULL ||
3056             !tgt->ltd_active ||
3057             lmv->desc.ld_tgt_count == 0) {
3058                 CERROR("master lmv inactive\n");
3059                 RETURN(-EIO);
3060         }
3061
3062         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3063                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3064                 RETURN(rc);
3065         }
3066
3067         curspace = curinodes = 0;
3068         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3069                 int err;
3070                 tgt = lmv->tgts[i];
3071
3072                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3073                         continue;
3074
3075                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3076                 if (err) {
3077                         CERROR("getquota on mdt %d failed. %d\n", i, err);
3078                         if (!rc)
3079                                 rc = err;
3080                 } else {
3081                         curspace += oqctl->qc_dqblk.dqb_curspace;
3082                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3083                 }
3084         }
3085         oqctl->qc_dqblk.dqb_curspace = curspace;
3086         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3087
3088         RETURN(rc);
3089 }
3090
3091 static int lmv_merge_attr(struct obd_export *exp,
3092                           const struct lmv_stripe_md *lsm,
3093                           struct cl_attr *attr,
3094                           ldlm_blocking_callback cb_blocking)
3095 {
3096         int rc;
3097         int i;
3098
3099         rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3100         if (rc < 0)
3101                 return rc;
3102
3103         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3104                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3105
3106                 CDEBUG(D_INFO, ""DFID" size %llu, blocks %llu nlink %u,"
3107                        " atime %lu ctime %lu, mtime %lu.\n",
3108                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3109                        i_size_read(inode), (unsigned long long)inode->i_blocks,
3110                        inode->i_nlink, LTIME_S(inode->i_atime),
3111                        LTIME_S(inode->i_ctime), LTIME_S(inode->i_mtime));
3112
3113                 /* for slave stripe, it needs to subtract nlink for . and .. */
3114                 if (i != 0)
3115                         attr->cat_nlink += inode->i_nlink - 2;
3116                 else
3117                         attr->cat_nlink = inode->i_nlink;
3118
3119                 attr->cat_size += i_size_read(inode);
3120                 attr->cat_blocks += inode->i_blocks;
3121
3122                 if (attr->cat_atime < LTIME_S(inode->i_atime))
3123                         attr->cat_atime = LTIME_S(inode->i_atime);
3124
3125                 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3126                         attr->cat_ctime = LTIME_S(inode->i_ctime);
3127
3128                 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3129                         attr->cat_mtime = LTIME_S(inode->i_mtime);
3130         }
3131         return 0;
3132 }
3133
3134 struct obd_ops lmv_obd_ops = {
3135         .o_owner                = THIS_MODULE,
3136         .o_setup                = lmv_setup,
3137         .o_cleanup              = lmv_cleanup,
3138         .o_precleanup           = lmv_precleanup,
3139         .o_process_config       = lmv_process_config,
3140         .o_connect              = lmv_connect,
3141         .o_disconnect           = lmv_disconnect,
3142         .o_statfs               = lmv_statfs,
3143         .o_get_info             = lmv_get_info,
3144         .o_set_info_async       = lmv_set_info_async,
3145         .o_notify               = lmv_notify,
3146         .o_get_uuid             = lmv_get_uuid,
3147         .o_iocontrol            = lmv_iocontrol,
3148         .o_quotactl             = lmv_quotactl
3149 };
3150
3151 struct md_ops lmv_md_ops = {
3152         .m_get_root             = lmv_get_root,
3153         .m_null_inode           = lmv_null_inode,
3154         .m_close                = lmv_close,
3155         .m_create               = lmv_create,
3156         .m_enqueue              = lmv_enqueue,
3157         .m_getattr              = lmv_getattr,
3158         .m_getxattr             = lmv_getxattr,
3159         .m_getattr_name         = lmv_getattr_name,
3160         .m_intent_lock          = lmv_intent_lock,
3161         .m_link                 = lmv_link,
3162         .m_rename               = lmv_rename,
3163         .m_setattr              = lmv_setattr,
3164         .m_setxattr             = lmv_setxattr,
3165         .m_fsync                = lmv_fsync,
3166         .m_read_page            = lmv_read_page,
3167         .m_unlink               = lmv_unlink,
3168         .m_init_ea_size         = lmv_init_ea_size,
3169         .m_cancel_unused        = lmv_cancel_unused,
3170         .m_set_lock_data        = lmv_set_lock_data,
3171         .m_lock_match           = lmv_lock_match,
3172         .m_get_lustre_md        = lmv_get_lustre_md,
3173         .m_free_lustre_md       = lmv_free_lustre_md,
3174         .m_merge_attr           = lmv_merge_attr,
3175         .m_set_open_replay_data = lmv_set_open_replay_data,
3176         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3177         .m_intent_getattr_async = lmv_intent_getattr_async,
3178         .m_revalidate_lock      = lmv_revalidate_lock,
3179         .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
3180         .m_unpackmd             = lmv_unpackmd,
3181 };
3182
3183 static int __init lmv_init(void)
3184 {
3185         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3186                                    LUSTRE_LMV_NAME, NULL);
3187 }
3188
3189 static void __exit lmv_exit(void)
3190 {
3191         class_unregister_type(LUSTRE_LMV_NAME);
3192 }
3193
3194 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3195 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
3196 MODULE_VERSION(LUSTRE_VERSION_STRING);
3197 MODULE_LICENSE("GPL");
3198
3199 module_init(lmv_init);
3200 module_exit(lmv_exit);