lustre/lmv/lmv_obd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LMV
  38 #ifdef __KERNEL__
  39 #include <linux/slab.h>
  40 #include <linux/module.h>
  41 #include <linux/init.h>
  42 #include <linux/slab.h>
  43 #include <linux/pagemap.h>
  44 #include <linux/mm.h>
  45 #include <linux/math64.h>
  46 #include <linux/seq_file.h>
  47 #include <linux/namei.h>
  48 #else
  49 #include <liblustre.h>
  50 #endif
  51
  52 #include <lustre/lustre_idl.h>
  53 #include <obd_support.h>
  54 #include <lustre_lib.h>
  55 #include <lustre_net.h>
  56 #include <obd_class.h>
  57 #include <lustre_lmv.h>
  58 #include <lprocfs_status.h>
  59 #include <cl_object.h>
  60 #include <lclient.h>
  61 #include <lustre_lite.h>
  62 #include <lustre_fid.h>
  63 #include <lustre_ioctl.h>
  64 #include "lmv_internal.h"
  65
  66 /* This hash is only for testing purpose */
  67 static inline unsigned int
  68 lmv_hash_all_chars(unsigned int count, const char *name, int namelen)
  69 {
  70         unsigned int c = 0;
  71         const unsigned char *p = (const unsigned char *)name;
  72
  73         while (--namelen >= 0)
  74                 c += p[namelen];
  75
  76         c = c % count;
  77
  78         return c;
  79 }
  80
  81 static inline unsigned int
  82 lmv_hash_fnv1a(unsigned int count, const char *name, int namelen)
  83 {
  84         __u64   hash;
  85
  86         hash = lustre_hash_fnv_1a_64(name, namelen);
  87
  88         hash = hash % count;
  89
  90         return hash;
  91 }
  92
  93 int lmv_name_to_stripe_index(enum lmv_hash_type hashtype,
  94                              unsigned int max_mdt_index,
  95                              const char *name, int namelen)
  96 {
  97         int     idx;
  98
  99         LASSERT(namelen > 0);
 100         if (max_mdt_index <= 1)
 101                 return 0;
 102
 103         switch (hashtype) {
 104         case LMV_HASH_TYPE_ALL_CHARS:
 105                 idx = lmv_hash_all_chars(max_mdt_index, name, namelen);
 106                 break;
 107         case LMV_HASH_TYPE_FNV_1A_64:
 108                 idx = lmv_hash_fnv1a(max_mdt_index, name, namelen);
 109                 break;
 110         /* LMV_HASH_TYPE_MIGRATION means the file is being migrated,
 111          * and the file should be accessed by client, except for
 112          * lookup(see lmv_intent_lookup), return -EACCES here */
 113         case LMV_HASH_TYPE_MIGRATION:
 114                 CERROR("%.*s is being migrated: rc = %d\n", namelen,
 115                        name, -EACCES);
 116                 return -EACCES;
 117         default:
 118                 CERROR("Unknown hash type 0x%x\n", hashtype);
 119                 return -EINVAL;
 120         }
 121
 122         CDEBUG(D_INFO, "name %.*s hash_type %d idx %d\n", namelen, name,
 123                hashtype, idx);
 124
 125         LASSERT(idx < max_mdt_index);
 126         return idx;
 127 }
 128
 129 static void lmv_activate_target(struct lmv_obd *lmv,
 130                                 struct lmv_tgt_desc *tgt,
 131                                 int activate)
 132 {
 133         if (tgt->ltd_active == activate)
 134                 return;
 135
 136         tgt->ltd_active = activate;
 137         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
 138 }
 139
 140 /**
 141  * Error codes:
 142  *
 143  *  -EINVAL  : UUID can't be found in the LMV's target list
 144  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 145  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
 146  */
 147 static int lmv_set_mdc_active(struct lmv_obd *lmv,
 148                               const struct obd_uuid *uuid,
 149                               int activate)
 150 {
 151         struct lmv_tgt_desc     *tgt = NULL;
 152         struct obd_device       *obd;
 153         __u32                    i;
 154         int                      rc = 0;
 155         ENTRY;
 156
 157         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
 158                         lmv, uuid->uuid, activate);
 159
 160         spin_lock(&lmv->lmv_lock);
 161         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 162                 tgt = lmv->tgts[i];
 163                 if (tgt == NULL || tgt->ltd_exp == NULL)
 164                         continue;
 165
 166                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
 167                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 168
 169                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 170                         break;
 171         }
 172
 173         if (i == lmv->desc.ld_tgt_count)
 174                 GOTO(out_lmv_lock, rc = -EINVAL);
 175
 176         obd = class_exp2obd(tgt->ltd_exp);
 177         if (obd == NULL)
 178                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 179
 180         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 181                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 182                obd->obd_type->typ_name, i);
 183         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 184
 185         if (tgt->ltd_active == activate) {
 186                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 187                        activate ? "" : "in");
 188                 GOTO(out_lmv_lock, rc);
 189         }
 190
 191         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 192                activate ? "" : "in");
 193         lmv_activate_target(lmv, tgt, activate);
 194         EXIT;
 195
 196  out_lmv_lock:
 197         spin_unlock(&lmv->lmv_lock);
 198         return rc;
 199 }
 200
 201 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 202 {
 203         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
 204         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
 205
 206         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
 207 }
 208
 209 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 210                       enum obd_notify_event ev, void *data)
 211 {
 212         struct obd_connect_data *conn_data;
 213         struct lmv_obd          *lmv = &obd->u.lmv;
 214         struct obd_uuid         *uuid;
 215         int                      rc = 0;
 216         ENTRY;
 217
 218         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 219                 CERROR("unexpected notification of %s %s!\n",
 220                        watched->obd_type->typ_name,
 221                        watched->obd_name);
 222                 RETURN(-EINVAL);
 223         }
 224
 225         uuid = &watched->u.cli.cl_target_uuid;
 226         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 227                 /*
 228                  * Set MDC as active before notifying the observer, so the
 229                  * observer can use the MDC normally.
 230                  */
 231                 rc = lmv_set_mdc_active(lmv, uuid,
 232                                         ev == OBD_NOTIFY_ACTIVE);
 233                 if (rc) {
 234                         CERROR("%sactivation of %s failed: %d\n",
 235                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 236                                uuid->uuid, rc);
 237                         RETURN(rc);
 238                 }
 239         } else if (ev == OBD_NOTIFY_OCD) {
 240                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 241                 /*
 242                  * XXX: Make sure that ocd_connect_flags from all targets are
 243                  * the same. Otherwise one of MDTs runs wrong version or
 244                  * something like this.  --umka
 245                  */
 246                 obd->obd_self_export->exp_connect_data = *conn_data;
 247         }
 248 #if 0
 249         else if (ev == OBD_NOTIFY_DISCON) {
 250                 /*
 251                  * For disconnect event, flush fld cache for failout MDS case.
 252                  */
 253                 fld_client_flush(&lmv->lmv_fld);
 254         }
 255 #endif
 256         /*
 257          * Pass the notification up the chain.
 258          */
 259         if (obd->obd_observer)
 260                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 261
 262         RETURN(rc);
 263 }
 264
 265 /**
 266  * This is fake connect function. Its purpose is to initialize lmv and say
 267  * caller that everything is okay. Real connection will be performed later.
 268  */
 269 static int lmv_connect(const struct lu_env *env,
 270                        struct obd_export **exp, struct obd_device *obd,
 271                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 272                        void *localdata)
 273 {
 274         struct lmv_obd        *lmv = &obd->u.lmv;
 275         struct lustre_handle  conn = { 0 };
 276         int                    rc = 0;
 277         ENTRY;
 278
 279         /*
 280          * We don't want to actually do the underlying connections more than
 281          * once, so keep track.
 282          */
 283         lmv->refcount++;
 284         if (lmv->refcount > 1) {
 285                 *exp = NULL;
 286                 RETURN(0);
 287         }
 288
 289         rc = class_connect(&conn, obd, cluuid);
 290         if (rc) {
 291                 CERROR("class_connection() returned %d\n", rc);
 292                 RETURN(rc);
 293         }
 294
 295         *exp = class_conn2export(&conn);
 296         class_export_get(*exp);
 297
 298         lmv->exp = *exp;
 299         lmv->connected = 0;
 300         lmv->cluuid = *cluuid;
 301
 302         if (data)
 303                 lmv->conn_data = *data;
 304
 305         if (obd->obd_type->typ_procsym == NULL) {
 306                 obd->obd_type->typ_procsym = lprocfs_seq_register("target_obds",
 307                                                          obd->obd_proc_entry,
 308                                                          NULL, NULL);
 309                 if (IS_ERR(obd->obd_type->typ_procsym)) {
 310                         CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 311                                obd->obd_type->typ_name, obd->obd_name);
 312                         obd->obd_type->typ_procsym = NULL;
 313                 }
 314         }
 315
 316         /*
 317          * All real clients should perform actual connection right away, because
 318          * it is possible, that LMV will not have opportunity to connect targets
 319          * and MDC stuff will be called directly, for instance while reading
 320          * ../mdc/../kbytesfree procfs file, etc.
 321          */
 322         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
 323                 rc = lmv_check_connect(obd);
 324
 325         if (rc && obd->obd_type->typ_procsym != NULL)
 326                 lprocfs_remove(&obd->obd_type->typ_procsym);
 327         RETURN(rc);
 328 }
 329
 330 static void lmv_set_timeouts(struct obd_device *obd)
 331 {
 332         struct lmv_obd          *lmv;
 333         __u32                    i;
 334
 335         lmv = &obd->u.lmv;
 336         if (lmv->server_timeout == 0)
 337                 return;
 338
 339         if (lmv->connected == 0)
 340                 return;
 341
 342         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 343                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 344
 345                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
 346                         continue;
 347
 348                 obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
 349                                    KEY_INTERMDS, 0, NULL, NULL);
 350         }
 351 }
 352
 353 static int lmv_init_ea_size(struct obd_export *exp, int easize,
 354                             int def_easize, int cookiesize, int def_cookiesize)
 355 {
 356         struct obd_device       *obd = exp->exp_obd;
 357         struct lmv_obd          *lmv = &obd->u.lmv;
 358         __u32                    i;
 359         int                      rc = 0;
 360         int                      change = 0;
 361         ENTRY;
 362
 363         if (lmv->max_easize < easize) {
 364                 lmv->max_easize = easize;
 365                 change = 1;
 366         }
 367         if (lmv->max_def_easize < def_easize) {
 368                 lmv->max_def_easize = def_easize;
 369                 change = 1;
 370         }
 371         if (lmv->max_cookiesize < cookiesize) {
 372                 lmv->max_cookiesize = cookiesize;
 373                 change = 1;
 374         }
 375         if (lmv->max_def_cookiesize < def_cookiesize) {
 376                 lmv->max_def_cookiesize = def_cookiesize;
 377                 change = 1;
 378         }
 379         if (change == 0)
 380                 RETURN(0);
 381
 382         if (lmv->connected == 0)
 383                 RETURN(0);
 384
 385         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 386                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 387
 388                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
 389                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 390                         continue;
 391                 }
 392
 393                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
 394                                      cookiesize, def_cookiesize);
 395                 if (rc) {
 396                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 397                                " rc = %d.\n", obd->obd_name, i, rc);
 398                         break;
 399                 }
 400         }
 401         RETURN(rc);
 402 }
 403
 404 #define MAX_STRING_SIZE 128
 405
 406 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 407 {
 408         struct lmv_obd          *lmv = &obd->u.lmv;
 409         struct obd_uuid         *cluuid = &lmv->cluuid;
 410         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 411         struct obd_device       *mdc_obd;
 412         struct obd_export       *mdc_exp;
 413         struct lu_fld_target     target;
 414         int                      rc;
 415         ENTRY;
 416
 417         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 418                                         &obd->obd_uuid);
 419         if (!mdc_obd) {
 420                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 421                 RETURN(-EINVAL);
 422         }
 423
 424         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 425                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 426                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 427                 cluuid->uuid);
 428
 429         if (!mdc_obd->obd_set_up) {
 430                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 431                 RETURN(-EINVAL);
 432         }
 433
 434         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 435                          &lmv->conn_data, NULL);
 436         if (rc) {
 437                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 438                 RETURN(rc);
 439         }
 440
 441         /*
 442          * Init fid sequence client for this mdc and add new fld target.
 443          */
 444         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 445         if (rc)
 446                 RETURN(rc);
 447
 448         target.ft_srv = NULL;
 449         target.ft_exp = mdc_exp;
 450         target.ft_idx = tgt->ltd_idx;
 451
 452         fld_client_add_target(&lmv->lmv_fld, &target);
 453
 454         rc = obd_register_observer(mdc_obd, obd);
 455         if (rc) {
 456                 obd_disconnect(mdc_exp);
 457                 CERROR("target %s register_observer error %d\n",
 458                        tgt->ltd_uuid.uuid, rc);
 459                 RETURN(rc);
 460         }
 461
 462         if (obd->obd_observer) {
 463                 /*
 464                  * Tell the observer about the new target.
 465                  */
 466                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 467                                 OBD_NOTIFY_ACTIVE,
 468                                 (void *)(tgt - lmv->tgts[0]));
 469                 if (rc) {
 470                         obd_disconnect(mdc_exp);
 471                         RETURN(rc);
 472                 }
 473         }
 474
 475         tgt->ltd_active = 1;
 476         tgt->ltd_exp = mdc_exp;
 477         lmv->desc.ld_active_tgt_count++;
 478
 479         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
 480                         lmv->max_cookiesize, lmv->max_def_cookiesize);
 481
 482         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 483                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 484                 atomic_read(&obd->obd_refcount));
 485
 486         if (obd->obd_type->typ_procsym != NULL) {
 487                 struct proc_dir_entry *mdc_symlink;
 488
 489                 LASSERT(mdc_obd->obd_type != NULL);
 490                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 491                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 492                                                   obd->obd_type->typ_procsym,
 493                                                   "../../../%s/%s",
 494                                                   mdc_obd->obd_type->typ_name,
 495                                                   mdc_obd->obd_name);
 496                 if (mdc_symlink == NULL) {
 497                         CERROR("Could not register LMV target "
 498                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 499                                obd->obd_type->typ_name, obd->obd_name,
 500                                mdc_obd->obd_name);
 501                         lprocfs_remove(&obd->obd_type->typ_procsym);
 502                 }
 503         }
 504         RETURN(0);
 505 }
 506
 507 static void lmv_del_target(struct lmv_obd *lmv, int index)
 508 {
 509         if (lmv->tgts[index] == NULL)
 510                 return;
 511
 512         OBD_FREE_PTR(lmv->tgts[index]);
 513         lmv->tgts[index] = NULL;
 514         return;
 515 }
 516
 517 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 518                            __u32 index, int gen)
 519 {
 520         struct lmv_obd      *lmv = &obd->u.lmv;
 521         struct lmv_tgt_desc *tgt;
 522         int                  rc = 0;
 523         ENTRY;
 524
 525         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 526
 527         lmv_init_lock(lmv);
 528
 529         if (lmv->desc.ld_tgt_count == 0) {
 530                 struct obd_device *mdc_obd;
 531
 532                 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 533                                                 &obd->obd_uuid);
 534                 if (!mdc_obd) {
 535                         lmv_init_unlock(lmv);
 536                         CERROR("%s: Target %s not attached: rc = %d\n",
 537                                obd->obd_name, uuidp->uuid, -EINVAL);
 538                         RETURN(-EINVAL);
 539                 }
 540         }
 541
 542         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
 543                 tgt = lmv->tgts[index];
 544                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
 545                        " rc = %d\n", obd->obd_name,
 546                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
 547                 lmv_init_unlock(lmv);
 548                 RETURN(-EEXIST);
 549         }
 550
 551         if (index >= lmv->tgts_size) {
 552                 /* We need to reallocate the lmv target array. */
 553                 struct lmv_tgt_desc **newtgts, **old = NULL;
 554                 __u32 newsize = 1;
 555                 __u32 oldsize = 0;
 556
 557                 while (newsize < index + 1)
 558                         newsize = newsize << 1;
 559                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 560                 if (newtgts == NULL) {
 561                         lmv_init_unlock(lmv);
 562                         RETURN(-ENOMEM);
 563                 }
 564
 565                 if (lmv->tgts_size) {
 566                         memcpy(newtgts, lmv->tgts,
 567                                sizeof(*newtgts) * lmv->tgts_size);
 568                         old = lmv->tgts;
 569                         oldsize = lmv->tgts_size;
 570                 }
 571
 572                 lmv->tgts = newtgts;
 573                 lmv->tgts_size = newsize;
 574                 smp_rmb();
 575                 if (old)
 576                         OBD_FREE(old, sizeof(*old) * oldsize);
 577
 578                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
 579                        lmv->tgts_size);
 580         }
 581
 582         OBD_ALLOC_PTR(tgt);
 583         if (!tgt) {
 584                 lmv_init_unlock(lmv);
 585                 RETURN(-ENOMEM);
 586         }
 587
 588         mutex_init(&tgt->ltd_fid_mutex);
 589         tgt->ltd_idx = index;
 590         tgt->ltd_uuid = *uuidp;
 591         tgt->ltd_active = 0;
 592         lmv->tgts[index] = tgt;
 593         if (index >= lmv->desc.ld_tgt_count)
 594                 lmv->desc.ld_tgt_count = index + 1;
 595
 596         if (lmv->connected) {
 597                 rc = lmv_connect_mdc(obd, tgt);
 598                 if (rc) {
 599                         spin_lock(&lmv->lmv_lock);
 600                         lmv->desc.ld_tgt_count--;
 601                         memset(tgt, 0, sizeof(*tgt));
 602                         spin_unlock(&lmv->lmv_lock);
 603                 } else {
 604                         int easize = sizeof(struct lmv_stripe_md) +
 605                                 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
 606                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
 607                 }
 608         }
 609
 610         lmv_init_unlock(lmv);
 611         RETURN(rc);
 612 }
 613
 614 int lmv_check_connect(struct obd_device *obd)
 615 {
 616         struct lmv_obd          *lmv = &obd->u.lmv;
 617         struct lmv_tgt_desc     *tgt;
 618         __u32                    i;
 619         int                      rc;
 620         int                      easize;
 621         ENTRY;
 622
 623         if (lmv->connected)
 624                 RETURN(0);
 625
 626         lmv_init_lock(lmv);
 627         if (lmv->connected) {
 628                 lmv_init_unlock(lmv);
 629                 RETURN(0);
 630         }
 631
 632         if (lmv->desc.ld_tgt_count == 0) {
 633                 lmv_init_unlock(lmv);
 634                 CERROR("%s: no targets configured.\n", obd->obd_name);
 635                 RETURN(-EINVAL);
 636         }
 637
 638         LASSERT(lmv->tgts != NULL);
 639
 640         if (lmv->tgts[0] == NULL) {
 641                 lmv_init_unlock(lmv);
 642                 CERROR("%s: no target configured for index 0.\n",
 643                        obd->obd_name);
 644                 RETURN(-EINVAL);
 645         }
 646
 647         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 648                lmv->cluuid.uuid, obd->obd_name);
 649
 650         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 651                 tgt = lmv->tgts[i];
 652                 if (tgt == NULL)
 653                         continue;
 654                 rc = lmv_connect_mdc(obd, tgt);
 655                 if (rc)
 656                         GOTO(out_disc, rc);
 657         }
 658
 659         lmv_set_timeouts(obd);
 660         class_export_put(lmv->exp);
 661         lmv->connected = 1;
 662         easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
 663         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
 664         lmv_init_unlock(lmv);
 665         RETURN(0);
 666
 667  out_disc:
 668         while (i-- > 0) {
 669                 int rc2;
 670                 tgt = lmv->tgts[i];
 671                 if (tgt == NULL)
 672                         continue;
 673                 tgt->ltd_active = 0;
 674                 if (tgt->ltd_exp) {
 675                         --lmv->desc.ld_active_tgt_count;
 676                         rc2 = obd_disconnect(tgt->ltd_exp);
 677                         if (rc2) {
 678                                 CERROR("LMV target %s disconnect on "
 679                                        "MDC idx %d: error %d\n",
 680                                        tgt->ltd_uuid.uuid, i, rc2);
 681                         }
 682                 }
 683         }
 684         class_disconnect(lmv->exp);
 685         lmv_init_unlock(lmv);
 686         RETURN(rc);
 687 }
 688
 689 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 690 {
 691         struct lmv_obd         *lmv = &obd->u.lmv;
 692         struct obd_device      *mdc_obd;
 693         int                     rc;
 694         ENTRY;
 695
 696         LASSERT(tgt != NULL);
 697         LASSERT(obd != NULL);
 698
 699         mdc_obd = class_exp2obd(tgt->ltd_exp);
 700
 701         if (mdc_obd) {
 702                 mdc_obd->obd_force = obd->obd_force;
 703                 mdc_obd->obd_fail = obd->obd_fail;
 704                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 705         }
 706
 707         if (obd->obd_type->typ_procsym != NULL)
 708                 lprocfs_remove_proc_entry(mdc_obd->obd_name,
 709                                           obd->obd_type->typ_procsym);
 710
 711         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 712         if (rc)
 713                 CERROR("Can't finanize fids factory\n");
 714
 715         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 716                tgt->ltd_exp->exp_obd->obd_name,
 717                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 718
 719         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 720         rc = obd_disconnect(tgt->ltd_exp);
 721         if (rc) {
 722                 if (tgt->ltd_active) {
 723                         CERROR("Target %s disconnect error %d\n",
 724                                tgt->ltd_uuid.uuid, rc);
 725                 }
 726         }
 727
 728         lmv_activate_target(lmv, tgt, 0);
 729         tgt->ltd_exp = NULL;
 730         RETURN(0);
 731 }
 732
 733 static int lmv_disconnect(struct obd_export *exp)
 734 {
 735         struct obd_device       *obd = class_exp2obd(exp);
 736         struct lmv_obd          *lmv = &obd->u.lmv;
 737         int                      rc;
 738         __u32                    i;
 739         ENTRY;
 740
 741         if (!lmv->tgts)
 742                 goto out_local;
 743
 744         /*
 745          * Only disconnect the underlying layers on the final disconnect.
 746          */
 747         lmv->refcount--;
 748         if (lmv->refcount != 0)
 749                 goto out_local;
 750
 751         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 752                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
 753                         continue;
 754
 755                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
 756         }
 757
 758         if (obd->obd_type->typ_procsym != NULL)
 759                 lprocfs_remove(&obd->obd_type->typ_procsym);
 760         else
 761                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 762                        obd->obd_type->typ_name, obd->obd_name);
 763
 764 out_local:
 765         /*
 766          * This is the case when no real connection is established by
 767          * lmv_check_connect().
 768          */
 769         if (!lmv->connected)
 770                 class_export_put(exp);
 771         rc = class_disconnect(exp);
 772         if (lmv->refcount == 0)
 773                 lmv->connected = 0;
 774         RETURN(rc);
 775 }
 776
 777 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
 778 {
 779         struct obd_device       *obddev = class_exp2obd(exp);
 780         struct lmv_obd          *lmv = &obddev->u.lmv;
 781         struct getinfo_fid2path *gf;
 782         struct lmv_tgt_desc     *tgt;
 783         struct getinfo_fid2path *remote_gf = NULL;
 784         int                     remote_gf_size = 0;
 785         int                     rc;
 786
 787         gf = (struct getinfo_fid2path *)karg;
 788         tgt = lmv_find_target(lmv, &gf->gf_fid);
 789         if (IS_ERR(tgt))
 790                 RETURN(PTR_ERR(tgt));
 791
 792 repeat_fid2path:
 793         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 794         if (rc != 0 && rc != -EREMOTE)
 795                 GOTO(out_fid2path, rc);
 796
 797         /* If remote_gf != NULL, it means just building the
 798          * path on the remote MDT, copy this path segement to gf */
 799         if (remote_gf != NULL) {
 800                 struct getinfo_fid2path *ori_gf;
 801                 char *ptr;
 802
 803                 ori_gf = (struct getinfo_fid2path *)karg;
 804                 if (strlen(ori_gf->gf_path) +
 805                     strlen(gf->gf_path) > ori_gf->gf_pathlen)
 806                         GOTO(out_fid2path, rc = -EOVERFLOW);
 807
 808                 ptr = ori_gf->gf_path;
 809
 810                 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
 811                         strlen(ori_gf->gf_path));
 812
 813                 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
 814                 ptr += strlen(gf->gf_path);
 815                 *ptr = '/';
 816         }
 817
 818         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
 819                tgt->ltd_exp->exp_obd->obd_name,
 820                gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 821                gf->gf_linkno);
 822
 823         if (rc == 0)
 824                 GOTO(out_fid2path, rc);
 825
 826         /* sigh, has to go to another MDT to do path building further */
 827         if (remote_gf == NULL) {
 828                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 829                 OBD_ALLOC(remote_gf, remote_gf_size);
 830                 if (remote_gf == NULL)
 831                         GOTO(out_fid2path, rc = -ENOMEM);
 832                 remote_gf->gf_pathlen = PATH_MAX;
 833         }
 834
 835         if (!fid_is_sane(&gf->gf_fid)) {
 836                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 837                        tgt->ltd_exp->exp_obd->obd_name,
 838                        PFID(&gf->gf_fid), -EINVAL);
 839                 GOTO(out_fid2path, rc = -EINVAL);
 840         }
 841
 842         tgt = lmv_find_target(lmv, &gf->gf_fid);
 843         if (IS_ERR(tgt))
 844                 GOTO(out_fid2path, rc = -EINVAL);
 845
 846         remote_gf->gf_fid = gf->gf_fid;
 847         remote_gf->gf_recno = -1;
 848         remote_gf->gf_linkno = -1;
 849         memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
 850         gf = remote_gf;
 851         goto repeat_fid2path;
 852
 853 out_fid2path:
 854         if (remote_gf != NULL)
 855                 OBD_FREE(remote_gf, remote_gf_size);
 856         RETURN(rc);
 857 }
 858
 859 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 860                              const struct hsm_user_request *hur,
 861                              const struct lmv_tgt_desc *tgt_mds)
 862 {
 863         __u32                    i;
 864         int                      nr = 0;
 865         struct lmv_tgt_desc     *curr_tgt;
 866
 867         /* count how many requests must be sent to the given target */
 868         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 869                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
 870                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 871                         nr++;
 872         }
 873         return nr;
 874 }
 875
 876 static void lmv_hsm_req_build(struct lmv_obd *lmv,
 877                               struct hsm_user_request *hur_in,
 878                               const struct lmv_tgt_desc *tgt_mds,
 879                               struct hsm_user_request *hur_out)
 880 {
 881         __u32                    i, nr_out;
 882         struct lmv_tgt_desc     *curr_tgt;
 883
 884         /* build the hsm_user_request for the given target */
 885         hur_out->hur_request = hur_in->hur_request;
 886         nr_out = 0;
 887         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 888                 curr_tgt = lmv_find_target(lmv,
 889                                            &hur_in->hur_user_item[i].hui_fid);
 890                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 891                         hur_out->hur_user_item[nr_out] =
 892                                                 hur_in->hur_user_item[i];
 893                         nr_out++;
 894                 }
 895         }
 896         hur_out->hur_request.hr_itemcount = nr_out;
 897         memcpy(hur_data(hur_out), hur_data(hur_in),
 898                hur_in->hur_request.hr_data_len);
 899 }
 900
 901 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
 902                                  struct lustre_kernelcomm *lk, void *uarg)
 903 {
 904         __u32                    i;
 905         int                      rc;
 906         struct kkuc_ct_data     *kcd = NULL;
 907         ENTRY;
 908
 909         /* unregister request (call from llapi_hsm_copytool_fini) */
 910         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 911                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 912
 913                 if (tgt == NULL || tgt->ltd_exp == NULL)
 914                         continue;
 915                 /* best effort: try to clean as much as possible
 916                  * (continue on error) */
 917                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 918         }
 919
 920         /* Whatever the result, remove copytool from kuc groups.
 921          * Unreached coordinators will get EPIPE on next requests
 922          * and will unregister automatically.
 923          */
 924         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
 925         if (kcd != NULL)
 926                 OBD_FREE_PTR(kcd);
 927
 928         RETURN(rc);
 929 }
 930
 931 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
 932                                struct lustre_kernelcomm *lk, void *uarg)
 933 {
 934         struct file             *filp;
 935         __u32                    i, j;
 936         int                      err, rc;
 937         bool                     any_set = false;
 938         struct kkuc_ct_data     *kcd;
 939         ENTRY;
 940
 941         /* All or nothing: try to register to all MDS.
 942          * In case of failure, unregister from previous MDS,
 943          * except if it because of inactive target. */
 944         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 945                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 946
 947                 if (tgt == NULL || tgt->ltd_exp == NULL)
 948                         continue;
 949                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 950                 if (err) {
 951                         if (tgt->ltd_active) {
 952                                 /* permanent error */
 953                                 CERROR("%s: iocontrol MDC %s on MDT"
 954                                        " idx %d cmd %x: err = %d\n",
 955                                        class_exp2obd(lmv->exp)->obd_name,
 956                                        tgt->ltd_uuid.uuid, i, cmd, err);
 957                                 rc = err;
 958                                 lk->lk_flags |= LK_FLG_STOP;
 959                                 /* unregister from previous MDS */
 960                                 for (j = 0; j < i; j++) {
 961                                         tgt = lmv->tgts[j];
 962                                         if (tgt == NULL || tgt->ltd_exp == NULL)
 963                                                 continue;
 964                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
 965                                                       lk, uarg);
 966                                 }
 967                                 RETURN(rc);
 968                         }
 969                         /* else: transient error.
 970                          * kuc will register to the missing MDT
 971                          * when it is back */
 972                 } else {
 973                         any_set = true;
 974                 }
 975         }
 976
 977         if (!any_set)
 978                 /* no registration done: return error */
 979                 RETURN(-ENOTCONN);
 980
 981         /* at least one registration done, with no failure */
 982         filp = fget(lk->lk_wfd);
 983         if (filp == NULL)
 984                 RETURN(-EBADF);
 985
 986         OBD_ALLOC_PTR(kcd);
 987         if (kcd == NULL) {
 988                 fput(filp);
 989                 RETURN(-ENOMEM);
 990         }
 991         kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
 992         kcd->kcd_uuid = lmv->cluuid;
 993         kcd->kcd_archive = lk->lk_data;
 994
 995         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
 996         if (rc != 0) {
 997                 if (filp != NULL)
 998                         fput(filp);
 999                 OBD_FREE_PTR(kcd);
1000         }
1001
1002         RETURN(rc);
1003 }
1004
1005
1006
1007
1008 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
1009                          int len, void *karg, void *uarg)
1010 {
1011         struct obd_device       *obddev = class_exp2obd(exp);
1012         struct lmv_obd          *lmv = &obddev->u.lmv;
1013         struct lmv_tgt_desc     *tgt = NULL;
1014         __u32                    i = 0;
1015         int                      rc = 0;
1016         int                      set = 0;
1017         __u32                    count = lmv->desc.ld_tgt_count;
1018         ENTRY;
1019
1020         if (count == 0)
1021                 RETURN(-ENOTTY);
1022
1023         switch (cmd) {
1024         case IOC_OBD_STATFS: {
1025                 struct obd_ioctl_data *data = karg;
1026                 struct obd_device *mdc_obd;
1027                 struct obd_statfs stat_buf = {0};
1028                 __u32 index;
1029
1030                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1031                 if ((index >= count))
1032                         RETURN(-ENODEV);
1033
1034                 tgt = lmv->tgts[index];
1035                 if (tgt == NULL || !tgt->ltd_active)
1036                         RETURN(-ENODATA);
1037
1038                 mdc_obd = class_exp2obd(tgt->ltd_exp);
1039                 if (!mdc_obd)
1040                         RETURN(-EINVAL);
1041
1042                 /* copy UUID */
1043                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
1044                                  min((int) data->ioc_plen2,
1045                                      (int) sizeof(struct obd_uuid))))
1046                         RETURN(-EFAULT);
1047
1048                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
1049                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1050                                 0);
1051                 if (rc)
1052                         RETURN(rc);
1053                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1054                                  min((int) data->ioc_plen1,
1055                                      (int) sizeof(stat_buf))))
1056                         RETURN(-EFAULT);
1057                 break;
1058         }
1059         case OBD_IOC_QUOTACTL: {
1060                 struct if_quotactl *qctl = karg;
1061                 struct obd_quotactl *oqctl;
1062
1063                 if (qctl->qc_valid == QC_MDTIDX) {
1064                         if (count <= qctl->qc_idx)
1065                                 RETURN(-EINVAL);
1066
1067                         tgt = lmv->tgts[qctl->qc_idx];
1068                         if (tgt == NULL || tgt->ltd_exp == NULL)
1069                                 RETURN(-EINVAL);
1070                 } else if (qctl->qc_valid == QC_UUID) {
1071                         for (i = 0; i < count; i++) {
1072                                 tgt = lmv->tgts[i];
1073                                 if (tgt == NULL)
1074                                         continue;
1075                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
1076                                                      &qctl->obd_uuid))
1077                                         continue;
1078
1079                                 if (tgt->ltd_exp == NULL)
1080                                         RETURN(-EINVAL);
1081
1082                                 break;
1083                         }
1084                 } else {
1085                         RETURN(-EINVAL);
1086                 }
1087
1088                 if (i >= count)
1089                         RETURN(-EAGAIN);
1090
1091                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1092                 OBD_ALLOC_PTR(oqctl);
1093                 if (!oqctl)
1094                         RETURN(-ENOMEM);
1095
1096                 QCTL_COPY(oqctl, qctl);
1097                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1098                 if (rc == 0) {
1099                         QCTL_COPY(qctl, oqctl);
1100                         qctl->qc_valid = QC_MDTIDX;
1101                         qctl->obd_uuid = tgt->ltd_uuid;
1102                 }
1103                 OBD_FREE_PTR(oqctl);
1104                 break;
1105         }
1106         case OBD_IOC_CHANGELOG_SEND:
1107         case OBD_IOC_CHANGELOG_CLEAR: {
1108                 struct ioc_changelog *icc = karg;
1109
1110                 if (icc->icc_mdtindex >= count)
1111                         RETURN(-ENODEV);
1112
1113                 tgt = lmv->tgts[icc->icc_mdtindex];
1114                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1115                         RETURN(-ENODEV);
1116                 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1117                 break;
1118         }
1119         case LL_IOC_GET_CONNECT_FLAGS: {
1120                 tgt = lmv->tgts[0];
1121                 if (tgt == NULL || tgt->ltd_exp == NULL)
1122                         RETURN(-ENODATA);
1123                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1124                 break;
1125         }
1126         case OBD_IOC_FID2PATH: {
1127                 rc = lmv_fid2path(exp, len, karg, uarg);
1128                 break;
1129         }
1130         case LL_IOC_HSM_STATE_GET:
1131         case LL_IOC_HSM_STATE_SET:
1132         case LL_IOC_HSM_ACTION: {
1133                 struct md_op_data       *op_data = karg;
1134
1135                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1136                 if (IS_ERR(tgt))
1137                         RETURN(PTR_ERR(tgt));
1138
1139                 if (tgt->ltd_exp == NULL)
1140                         RETURN(-EINVAL);
1141
1142                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1143                 break;
1144         }
1145         case LL_IOC_HSM_PROGRESS: {
1146                 const struct hsm_progress_kernel *hpk = karg;
1147
1148                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1149                 if (IS_ERR(tgt))
1150                         RETURN(PTR_ERR(tgt));
1151                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1152                 break;
1153         }
1154         case LL_IOC_HSM_REQUEST: {
1155                 struct hsm_user_request *hur = karg;
1156                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1157
1158                 if (reqcount == 0)
1159                         RETURN(0);
1160
1161                 /* if the request is about a single fid
1162                  * or if there is a single MDS, no need to split
1163                  * the request. */
1164                 if (reqcount == 1 || count == 1) {
1165                         tgt = lmv_find_target(lmv,
1166                                               &hur->hur_user_item[0].hui_fid);
1167                         if (IS_ERR(tgt))
1168                                 RETURN(PTR_ERR(tgt));
1169                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1170                 } else {
1171                         /* split fid list to their respective MDS */
1172                         for (i = 0; i < count; i++) {
1173                                 unsigned int            nr, reqlen;
1174                                 int                     rc1;
1175                                 struct hsm_user_request *req;
1176
1177                                 tgt = lmv->tgts[i];
1178                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1179                                         continue;
1180
1181                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1182                                 if (nr == 0) /* nothing for this MDS */
1183                                         continue;
1184
1185                                 /* build a request with fids for this MDS */
1186                                 reqlen = offsetof(typeof(*hur),
1187                                                   hur_user_item[nr])
1188                                                 + hur->hur_request.hr_data_len;
1189                                 OBD_ALLOC_LARGE(req, reqlen);
1190                                 if (req == NULL)
1191                                         RETURN(-ENOMEM);
1192
1193                                 lmv_hsm_req_build(lmv, hur, tgt, req);
1194
1195                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1196                                                     req, uarg);
1197                                 if (rc1 != 0 && rc == 0)
1198                                         rc = rc1;
1199                                 OBD_FREE_LARGE(req, reqlen);
1200                         }
1201                 }
1202                 break;
1203         }
1204         case LL_IOC_LOV_SWAP_LAYOUTS: {
1205                 struct md_op_data       *op_data = karg;
1206                 struct lmv_tgt_desc     *tgt1, *tgt2;
1207
1208                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1209                 if (IS_ERR(tgt1))
1210                         RETURN(PTR_ERR(tgt1));
1211
1212                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1213                 if (IS_ERR(tgt2))
1214                         RETURN(PTR_ERR(tgt2));
1215
1216                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1217                         RETURN(-EINVAL);
1218
1219                 /* only files on same MDT can have their layouts swapped */
1220                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1221                         RETURN(-EPERM);
1222
1223                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1224                 break;
1225         }
1226         case LL_IOC_HSM_CT_START: {
1227                 struct lustre_kernelcomm *lk = karg;
1228                 if (lk->lk_flags & LK_FLG_STOP)
1229                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1230                 else
1231                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1232                 break;
1233         }
1234         default:
1235                 for (i = 0; i < count; i++) {
1236                         struct obd_device *mdc_obd;
1237                         int err;
1238
1239                         tgt = lmv->tgts[i];
1240                         if (tgt == NULL || tgt->ltd_exp == NULL)
1241                                 continue;
1242                         /* ll_umount_begin() sets force flag but for lmv, not
1243                          * mdc. Let's pass it through */
1244                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1245                         mdc_obd->obd_force = obddev->obd_force;
1246                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1247                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1248                                 RETURN(err);
1249                         } else if (err) {
1250                                 if (tgt->ltd_active) {
1251                                         CERROR("error: iocontrol MDC %s on MDT"
1252                                                " idx %d cmd %x: err = %d\n",
1253                                                tgt->ltd_uuid.uuid, i, cmd, err);
1254                                         if (!rc)
1255                                                 rc = err;
1256                                 }
1257                         } else
1258                                 set = 1;
1259                 }
1260                 if (!set && !rc)
1261                         rc = -EIO;
1262         }
1263         RETURN(rc);
1264 }
1265
1266 #if 0
1267 static int lmv_all_chars_policy(int count, const char *name,
1268                                 int len)
1269 {
1270         unsigned int c = 0;
1271
1272         while (len > 0)
1273                 c += name[--len];
1274         c = c % count;
1275         return c;
1276 }
1277
1278 static int lmv_nid_policy(struct lmv_obd *lmv)
1279 {
1280         struct obd_import *imp;
1281         __u32              id;
1282
1283         /*
1284          * XXX: To get nid we assume that underlying obd device is mdc.
1285          */
1286         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1287         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1288         return id % lmv->desc.ld_tgt_count;
1289 }
1290
1291 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1292                           placement_policy_t placement)
1293 {
1294         switch (placement) {
1295         case PLACEMENT_CHAR_POLICY:
1296                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1297                                             op_data->op_name,
1298                                             op_data->op_namelen);
1299         case PLACEMENT_NID_POLICY:
1300                 return lmv_nid_policy(lmv);
1301
1302         default:
1303                 break;
1304         }
1305
1306         CERROR("Unsupported placement policy %x\n", placement);
1307         return -EINVAL;
1308 }
1309 #endif
1310
1311 /**
1312  * This is _inode_ placement policy function (not name).
1313  */
1314 static int lmv_placement_policy(struct obd_device *obd,
1315                                 struct md_op_data *op_data,
1316                                 mdsno_t *mds)
1317 {
1318         struct lmv_obd          *lmv = &obd->u.lmv;
1319         ENTRY;
1320
1321         LASSERT(mds != NULL);
1322
1323         if (lmv->desc.ld_tgt_count == 1) {
1324                 *mds = 0;
1325                 RETURN(0);
1326         }
1327
1328         /**
1329          * If stripe_offset is provided during setdirstripe
1330          * (setdirstripe -i xx), xx MDS will be choosen.
1331          */
1332         if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1333                 struct lmv_user_md *lum;
1334
1335                 lum = op_data->op_data;
1336
1337                 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1338                         *mds = le32_to_cpu(lum->lum_stripe_offset);
1339                 } else {
1340                         /* -1 means default, which will be in the same MDT with
1341                          * the stripe */
1342                         *mds = op_data->op_mds;
1343                         lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1344                 }
1345         } else {
1346                 /* Allocate new fid on target according to operation type and
1347                  * parent home mds. */
1348                 *mds = op_data->op_mds;
1349         }
1350
1351         RETURN(0);
1352 }
1353
1354 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
1355                     mdsno_t mds)
1356 {
1357         struct lmv_tgt_desc     *tgt;
1358         int                      rc;
1359         ENTRY;
1360
1361         tgt = lmv_get_target(lmv, mds);
1362         if (IS_ERR(tgt))
1363                 RETURN(PTR_ERR(tgt));
1364
1365         /*
1366          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1367          * on server that seq in new allocated fid is not yet known.
1368          */
1369         mutex_lock(&tgt->ltd_fid_mutex);
1370
1371         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1372                 GOTO(out, rc = -ENODEV);
1373
1374         /*
1375          * Asking underlaying tgt layer to allocate new fid.
1376          */
1377         rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
1378         if (rc > 0) {
1379                 LASSERT(fid_is_sane(fid));
1380                 rc = 0;
1381         }
1382
1383         EXIT;
1384 out:
1385         mutex_unlock(&tgt->ltd_fid_mutex);
1386         return rc;
1387 }
1388
1389 int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1390                   struct md_op_data *op_data)
1391 {
1392         struct obd_device     *obd = class_exp2obd(exp);
1393         struct lmv_obd        *lmv = &obd->u.lmv;
1394         mdsno_t                mds = 0;
1395         int                    rc;
1396         ENTRY;
1397
1398         LASSERT(op_data != NULL);
1399         LASSERT(fid != NULL);
1400
1401         rc = lmv_placement_policy(obd, op_data, &mds);
1402         if (rc) {
1403                 CERROR("Can't get target for allocating fid, "
1404                        "rc %d\n", rc);
1405                 RETURN(rc);
1406         }
1407
1408         rc = __lmv_fid_alloc(lmv, fid, mds);
1409         if (rc) {
1410                 CERROR("Can't alloc new fid, rc %d\n", rc);
1411                 RETURN(rc);
1412         }
1413
1414         RETURN(rc);
1415 }
1416
1417 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1418 {
1419         struct lmv_obd  *lmv = &obd->u.lmv;
1420         struct lmv_desc *desc;
1421         int             rc;
1422         ENTRY;
1423
1424         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1425                 CERROR("LMV setup requires a descriptor\n");
1426                 RETURN(-EINVAL);
1427         }
1428
1429         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1430         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1431                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1432                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1433                 RETURN(-EINVAL);
1434         }
1435
1436         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1437         if (lmv->tgts == NULL)
1438                 RETURN(-ENOMEM);
1439         lmv->tgts_size = 32;
1440
1441         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1442         lmv->desc.ld_tgt_count = 0;
1443         lmv->desc.ld_active_tgt_count = 0;
1444         lmv->max_cookiesize = 0;
1445         lmv->max_def_easize = 0;
1446         lmv->max_easize = 0;
1447         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1448
1449         spin_lock_init(&lmv->lmv_lock);
1450         mutex_init(&lmv->init_mutex);
1451
1452 #ifdef LPROCFS
1453         obd->obd_vars = lprocfs_lmv_obd_vars;
1454         lprocfs_seq_obd_setup(obd);
1455         lprocfs_alloc_md_stats(obd, 0);
1456         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1457                                 0444, &lmv_proc_target_fops, obd);
1458         if (rc)
1459                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1460                       obd->obd_name, rc);
1461 #endif
1462         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1463                              LUSTRE_CLI_FLD_HASH_DHT);
1464         if (rc) {
1465                 CERROR("Can't init FLD, err %d\n", rc);
1466                 GOTO(out, rc);
1467         }
1468
1469         RETURN(0);
1470
1471 out:
1472         return rc;
1473 }
1474
1475 static int lmv_cleanup(struct obd_device *obd)
1476 {
1477         struct lmv_obd   *lmv = &obd->u.lmv;
1478         ENTRY;
1479
1480         fld_client_fini(&lmv->lmv_fld);
1481         if (lmv->tgts != NULL) {
1482                 int i;
1483                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1484                         if (lmv->tgts[i] == NULL)
1485                                 continue;
1486                         lmv_del_target(lmv, i);
1487                 }
1488                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1489                 lmv->tgts_size = 0;
1490         }
1491         RETURN(0);
1492 }
1493
1494 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1495 {
1496         struct lustre_cfg       *lcfg = buf;
1497         struct obd_uuid         obd_uuid;
1498         int                     gen;
1499         __u32                   index;
1500         int                     rc;
1501         ENTRY;
1502
1503         switch (lcfg->lcfg_command) {
1504         case LCFG_ADD_MDC:
1505                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1506                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1507                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1508                         GOTO(out, rc = -EINVAL);
1509
1510                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1511
1512                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1513                         GOTO(out, rc = -EINVAL);
1514                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1515                         GOTO(out, rc = -EINVAL);
1516                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1517                 GOTO(out, rc);
1518         default:
1519                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1520                 GOTO(out, rc = -EINVAL);
1521         }
1522 out:
1523         RETURN(rc);
1524 }
1525
1526 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1527                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1528 {
1529         struct obd_device       *obd = class_exp2obd(exp);
1530         struct lmv_obd          *lmv = &obd->u.lmv;
1531         struct obd_statfs       *temp;
1532         int                      rc = 0;
1533         __u32                    i;
1534         ENTRY;
1535
1536         rc = lmv_check_connect(obd);
1537         if (rc)
1538                 RETURN(rc);
1539
1540         OBD_ALLOC(temp, sizeof(*temp));
1541         if (temp == NULL)
1542                 RETURN(-ENOMEM);
1543
1544         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1545                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1546                         continue;
1547
1548                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1549                                 max_age, flags);
1550                 if (rc) {
1551                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1552                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1553                                rc);
1554                         GOTO(out_free_temp, rc);
1555                 }
1556
1557                 if (i == 0) {
1558                         *osfs = *temp;
1559                         /* If the statfs is from mount, it will needs
1560                          * retrieve necessary information from MDT0.
1561                          * i.e. mount does not need the merged osfs
1562                          * from all of MDT.
1563                          * And also clients can be mounted as long as
1564                          * MDT0 is in service*/
1565                         if (flags & OBD_STATFS_FOR_MDT0)
1566                                 GOTO(out_free_temp, rc);
1567                 } else {
1568                         osfs->os_bavail += temp->os_bavail;
1569                         osfs->os_blocks += temp->os_blocks;
1570                         osfs->os_ffree += temp->os_ffree;
1571                         osfs->os_files += temp->os_files;
1572                 }
1573         }
1574
1575         EXIT;
1576 out_free_temp:
1577         OBD_FREE(temp, sizeof(*temp));
1578         return rc;
1579 }
1580
1581 static int lmv_getstatus(struct obd_export *exp,
1582                          struct lu_fid *fid,
1583                          struct obd_capa **pc)
1584 {
1585         struct obd_device    *obd = exp->exp_obd;
1586         struct lmv_obd       *lmv = &obd->u.lmv;
1587         int                   rc;
1588         ENTRY;
1589
1590         rc = lmv_check_connect(obd);
1591         if (rc)
1592                 RETURN(rc);
1593
1594         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1595         RETURN(rc);
1596 }
1597
1598 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1599                         struct obd_capa *oc, obd_valid valid, const char *name,
1600                         const char *input, int input_size, int output_size,
1601                         int flags, struct ptlrpc_request **request)
1602 {
1603         struct obd_device      *obd = exp->exp_obd;
1604         struct lmv_obd         *lmv = &obd->u.lmv;
1605         struct lmv_tgt_desc    *tgt;
1606         int                     rc;
1607         ENTRY;
1608
1609         rc = lmv_check_connect(obd);
1610         if (rc)
1611                 RETURN(rc);
1612
1613         tgt = lmv_find_target(lmv, fid);
1614         if (IS_ERR(tgt))
1615                 RETURN(PTR_ERR(tgt));
1616
1617         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1618                          input_size, output_size, flags, request);
1619
1620         RETURN(rc);
1621 }
1622
1623 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1624                         struct obd_capa *oc, obd_valid valid, const char *name,
1625                         const char *input, int input_size, int output_size,
1626                         int flags, __u32 suppgid,
1627                         struct ptlrpc_request **request)
1628 {
1629         struct obd_device      *obd = exp->exp_obd;
1630         struct lmv_obd         *lmv = &obd->u.lmv;
1631         struct lmv_tgt_desc    *tgt;
1632         int                     rc;
1633         ENTRY;
1634
1635         rc = lmv_check_connect(obd);
1636         if (rc)
1637                 RETURN(rc);
1638
1639         tgt = lmv_find_target(lmv, fid);
1640         if (IS_ERR(tgt))
1641                 RETURN(PTR_ERR(tgt));
1642
1643         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1644                          input_size, output_size, flags, suppgid,
1645                          request);
1646
1647         RETURN(rc);
1648 }
1649
1650 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1651                        struct ptlrpc_request **request)
1652 {
1653         struct obd_device       *obd = exp->exp_obd;
1654         struct lmv_obd          *lmv = &obd->u.lmv;
1655         struct lmv_tgt_desc     *tgt;
1656         int                      rc;
1657         ENTRY;
1658
1659         rc = lmv_check_connect(obd);
1660         if (rc)
1661                 RETURN(rc);
1662
1663         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1664         if (IS_ERR(tgt))
1665                 RETURN(PTR_ERR(tgt));
1666
1667         if (op_data->op_flags & MF_GET_MDT_IDX) {
1668                 op_data->op_mds = tgt->ltd_idx;
1669                 RETURN(0);
1670         }
1671
1672         rc = md_getattr(tgt->ltd_exp, op_data, request);
1673
1674         RETURN(rc);
1675 }
1676
1677 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1678 {
1679         struct obd_device   *obd = exp->exp_obd;
1680         struct lmv_obd      *lmv = &obd->u.lmv;
1681         __u32                i;
1682         int                  rc;
1683         ENTRY;
1684
1685         rc = lmv_check_connect(obd);
1686         if (rc)
1687                 RETURN(rc);
1688
1689         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1690
1691         /*
1692          * With DNE every object can have two locks in different namespaces:
1693          * lookup lock in space of MDT storing direntry and update/open lock in
1694          * space of MDT storing inode.
1695          */
1696         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1697                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1698                         continue;
1699                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1700         }
1701
1702         RETURN(0);
1703 }
1704
1705 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1706                            ldlm_iterator_t it, void *data)
1707 {
1708         struct obd_device   *obd = exp->exp_obd;
1709         struct lmv_obd      *lmv = &obd->u.lmv;
1710         __u32                i;
1711         int                  rc;
1712         ENTRY;
1713
1714         rc = lmv_check_connect(obd);
1715         if (rc)
1716                 RETURN(rc);
1717
1718         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1719
1720         /*
1721          * With DNE every object can have two locks in different namespaces:
1722          * lookup lock in space of MDT storing direntry and update/open lock in
1723          * space of MDT storing inode.
1724          */
1725         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1726                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1727                         continue;
1728                 rc = md_find_cbdata(lmv->tgts[i]->ltd_exp, fid, it, data);
1729                 if (rc)
1730                         RETURN(rc);
1731         }
1732
1733         RETURN(rc);
1734 }
1735
1736
1737 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1738                      struct md_open_data *mod, struct ptlrpc_request **request)
1739 {
1740         struct obd_device     *obd = exp->exp_obd;
1741         struct lmv_obd        *lmv = &obd->u.lmv;
1742         struct lmv_tgt_desc   *tgt;
1743         int                    rc;
1744         ENTRY;
1745
1746         rc = lmv_check_connect(obd);
1747         if (rc)
1748                 RETURN(rc);
1749
1750         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1751         if (IS_ERR(tgt))
1752                 RETURN(PTR_ERR(tgt));
1753
1754         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1755         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1756         RETURN(rc);
1757 }
1758
1759 /**
1760  * Choosing the MDT by name or FID in @op_data.
1761  * For non-striped directory, it will locate MDT by fid.
1762  * For striped-directory, it will locate MDT by name. And also
1763  * it will reset op_fid1 with the FID of the choosen stripe.
1764  **/
1765 struct lmv_tgt_desc *
1766 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1767                            const char *name, int namelen, struct lu_fid *fid,
1768                            mdsno_t *mds)
1769 {
1770         struct lmv_tgt_desc     *tgt;
1771         const struct lmv_oinfo  *oinfo;
1772
1773         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1774         if (IS_ERR(oinfo))
1775                 RETURN((void *)oinfo);
1776         *fid = oinfo->lmo_fid;
1777         *mds = oinfo->lmo_mds;
1778         tgt = lmv_get_target(lmv, *mds);
1779
1780         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid));
1781         return tgt;
1782 }
1783
1784 struct lmv_tgt_desc
1785 *lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1786                 struct lu_fid *fid)
1787 {
1788         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1789         struct lmv_tgt_desc     *tgt;
1790
1791         if (lsm == NULL || lsm->lsm_md_stripe_count <= 1 ||
1792             op_data->op_namelen == 0 ||
1793             lsm->lsm_md_magic == LMV_MAGIC_MIGRATE) {
1794                 tgt = lmv_find_target(lmv, fid);
1795                 if (IS_ERR(tgt))
1796                         return tgt;
1797
1798                 op_data->op_mds = tgt->ltd_idx;
1799                 return tgt;
1800         }
1801
1802         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1803                                           op_data->op_namelen, fid,
1804                                           &op_data->op_mds);
1805 }
1806
1807 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1808                const void *data, int datalen, int mode, __u32 uid,
1809                __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1810                struct ptlrpc_request **request)
1811 {
1812         struct obd_device       *obd = exp->exp_obd;
1813         struct lmv_obd          *lmv = &obd->u.lmv;
1814         struct lmv_tgt_desc     *tgt;
1815         int                      rc;
1816         ENTRY;
1817
1818         rc = lmv_check_connect(obd);
1819         if (rc)
1820                 RETURN(rc);
1821
1822         if (!lmv->desc.ld_active_tgt_count)
1823                 RETURN(-EIO);
1824
1825         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1826         if (IS_ERR(tgt))
1827                 RETURN(PTR_ERR(tgt));
1828
1829         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1830                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1831                op_data->op_mds);
1832
1833         rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
1834         if (rc)
1835                 RETURN(rc);
1836
1837         /* Send the create request to the MDT where the object
1838          * will be located */
1839         tgt = lmv_find_target(lmv, &op_data->op_fid2);
1840         if (IS_ERR(tgt))
1841                 RETURN(PTR_ERR(tgt));
1842
1843         op_data->op_mds = tgt->ltd_idx;
1844
1845         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1846                PFID(&op_data->op_fid2), op_data->op_mds);
1847
1848         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1849         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1850                        cap_effective, rdev, request);
1851         if (rc == 0) {
1852                 if (*request == NULL)
1853                         RETURN(rc);
1854                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1855         }
1856         RETURN(rc);
1857 }
1858
1859 static int lmv_done_writing(struct obd_export *exp,
1860                             struct md_op_data *op_data,
1861                             struct md_open_data *mod)
1862 {
1863         struct obd_device     *obd = exp->exp_obd;
1864         struct lmv_obd        *lmv = &obd->u.lmv;
1865         struct lmv_tgt_desc   *tgt;
1866         int                    rc;
1867         ENTRY;
1868
1869         rc = lmv_check_connect(obd);
1870         if (rc)
1871                 RETURN(rc);
1872
1873         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1874         if (IS_ERR(tgt))
1875                 RETURN(PTR_ERR(tgt));
1876
1877         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1878         RETURN(rc);
1879 }
1880
1881 static int
1882 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1883             const union ldlm_policy_data *policy,
1884             struct lookup_intent *it, struct md_op_data *op_data,
1885             struct lustre_handle *lockh, __u64 extra_lock_flags)
1886 {
1887         struct obd_device        *obd = exp->exp_obd;
1888         struct lmv_obd           *lmv = &obd->u.lmv;
1889         struct lmv_tgt_desc      *tgt;
1890         int                       rc;
1891         ENTRY;
1892
1893         rc = lmv_check_connect(obd);
1894         if (rc)
1895                 RETURN(rc);
1896
1897         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1898                LL_IT2STR(it), PFID(&op_data->op_fid1));
1899
1900         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1901         if (IS_ERR(tgt))
1902                 RETURN(PTR_ERR(tgt));
1903
1904         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1905                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1906
1907         rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1908                         extra_lock_flags);
1909
1910         RETURN(rc);
1911 }
1912
1913 static int
1914 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1915                  struct ptlrpc_request **preq)
1916 {
1917         struct ptlrpc_request   *req = NULL;
1918         struct obd_device       *obd = exp->exp_obd;
1919         struct lmv_obd          *lmv = &obd->u.lmv;
1920         struct lmv_tgt_desc     *tgt;
1921         struct mdt_body         *body;
1922         int                      rc;
1923         ENTRY;
1924
1925         rc = lmv_check_connect(obd);
1926         if (rc)
1927                 RETURN(rc);
1928
1929         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1930         if (IS_ERR(tgt))
1931                 RETURN(PTR_ERR(tgt));
1932
1933         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1934                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1935                tgt->ltd_idx);
1936
1937         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1938         if (rc != 0)
1939                 RETURN(rc);
1940
1941         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1942         LASSERT(body != NULL);
1943
1944         if (body->valid & OBD_MD_MDS) {
1945                 struct lu_fid rid = body->fid1;
1946                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1947                        PFID(&rid));
1948
1949                 tgt = lmv_find_target(lmv, &rid);
1950                 if (IS_ERR(tgt)) {
1951                         ptlrpc_req_finished(*preq);
1952                         preq = NULL;
1953                         RETURN(PTR_ERR(tgt));
1954                 }
1955
1956                 op_data->op_fid1 = rid;
1957                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1958                 op_data->op_namelen = 0;
1959                 op_data->op_name = NULL;
1960                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1961                 ptlrpc_req_finished(*preq);
1962                 *preq = req;
1963         }
1964
1965         RETURN(rc);
1966 }
1967
1968 #define md_op_data_fid(op_data, fl)                     \
1969         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1970          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1971          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1972          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1973          NULL)
1974
1975 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1976                             struct md_op_data *op_data,
1977                             int op_tgt, ldlm_mode_t mode, int bits, int flag)
1978 {
1979         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1980         struct obd_device      *obd = exp->exp_obd;
1981         struct lmv_obd         *lmv = &obd->u.lmv;
1982         ldlm_policy_data_t      policy = {{ 0 }};
1983         int                     rc = 0;
1984         ENTRY;
1985
1986         if (!fid_is_sane(fid))
1987                 RETURN(0);
1988
1989         if (tgt == NULL) {
1990                 tgt = lmv_find_target(lmv, fid);
1991                 if (IS_ERR(tgt))
1992                         RETURN(PTR_ERR(tgt));
1993         }
1994
1995         if (tgt->ltd_idx != op_tgt) {
1996                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1997                 policy.l_inodebits.bits = bits;
1998                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1999                                       mode, LCF_ASYNC, NULL);
2000         } else {
2001                 CDEBUG(D_INODE,
2002                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
2003                        op_tgt, PFID(fid));
2004                 op_data->op_flags |= flag;
2005                 rc = 0;
2006         }
2007
2008         RETURN(rc);
2009 }
2010
2011 /*
2012  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2013  * op_data->op_fid2
2014  */
2015 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2016                     struct ptlrpc_request **request)
2017 {
2018         struct obd_device       *obd = exp->exp_obd;
2019         struct lmv_obd          *lmv = &obd->u.lmv;
2020         struct lmv_tgt_desc     *tgt;
2021         int                      rc;
2022         ENTRY;
2023
2024         rc = lmv_check_connect(obd);
2025         if (rc)
2026                 RETURN(rc);
2027
2028         LASSERT(op_data->op_namelen != 0);
2029
2030         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2031                PFID(&op_data->op_fid2), op_data->op_namelen,
2032                op_data->op_name, PFID(&op_data->op_fid1));
2033
2034         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2035         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2036         op_data->op_cap = cfs_curproc_cap_pack();
2037         if (op_data->op_mea2 != NULL) {
2038                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
2039                 const struct lmv_oinfo  *oinfo;
2040
2041                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2042                                                 op_data->op_namelen);
2043                 if (IS_ERR(oinfo))
2044                         RETURN(PTR_ERR(oinfo));
2045
2046                 op_data->op_fid2 = oinfo->lmo_fid;
2047         }
2048
2049         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2050         if (IS_ERR(tgt))
2051                 RETURN(PTR_ERR(tgt));
2052
2053         /*
2054          * Cancel UPDATE lock on child (fid1).
2055          */
2056         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2057         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2058                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2059         if (rc != 0)
2060                 RETURN(rc);
2061
2062         rc = md_link(tgt->ltd_exp, op_data, request);
2063
2064         RETURN(rc);
2065 }
2066
2067 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2068                       const char *old, int oldlen, const char *new, int newlen,
2069                       struct ptlrpc_request **request)
2070 {
2071         struct obd_device       *obd = exp->exp_obd;
2072         struct lmv_obd          *lmv = &obd->u.lmv;
2073         struct lmv_tgt_desc     *src_tgt;
2074         int                     rc;
2075         ENTRY;
2076
2077         LASSERT(oldlen != 0);
2078
2079         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2080                oldlen, old, PFID(&op_data->op_fid1),
2081                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2082                newlen, new, PFID(&op_data->op_fid2),
2083                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2084
2085         rc = lmv_check_connect(obd);
2086         if (rc)
2087                 RETURN(rc);
2088
2089         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2090         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2091         op_data->op_cap = cfs_curproc_cap_pack();
2092         if (op_data->op_cli_flags & CLI_MIGRATE) {
2093                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2094                          PFID(&op_data->op_fid3));
2095                 rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
2096                 if (rc)
2097                         RETURN(rc);
2098                 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3);
2099         } else {
2100                 if (op_data->op_mea1 != NULL) {
2101                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2102
2103                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2104                                                              oldlen,
2105                                                              &op_data->op_fid1,
2106                                                              &op_data->op_mds);
2107                         if (IS_ERR(src_tgt))
2108                                 RETURN(PTR_ERR(src_tgt));
2109                 } else {
2110                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2111                         if (IS_ERR(src_tgt))
2112                                 RETURN(PTR_ERR(src_tgt));
2113
2114                         op_data->op_mds = src_tgt->ltd_idx;
2115                 }
2116
2117                 if (op_data->op_mea2) {
2118                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
2119                         const struct lmv_oinfo  *oinfo;
2120
2121                         oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2122                         if (IS_ERR(oinfo))
2123                                 RETURN(PTR_ERR(oinfo));
2124
2125                         op_data->op_fid2 = oinfo->lmo_fid;
2126                 }
2127         }
2128         if (IS_ERR(src_tgt))
2129                 RETURN(PTR_ERR(src_tgt));
2130
2131         /*
2132          * LOOKUP lock on src child (fid3) should also be cancelled for
2133          * src_tgt in mdc_rename.
2134          */
2135         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2136
2137         /*
2138          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2139          * own target.
2140          */
2141         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2142                               LCK_EX, MDS_INODELOCK_UPDATE,
2143                               MF_MDC_CANCEL_FID2);
2144
2145         if (rc != 0)
2146                 RETURN(rc);
2147         /*
2148          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2149          */
2150         if (fid_is_sane(&op_data->op_fid3)) {
2151                 struct lmv_tgt_desc *tgt;
2152
2153                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2154                 if (IS_ERR(tgt))
2155                         RETURN(PTR_ERR(tgt));
2156
2157                 /* Cancel LOOKUP lock on its parent */
2158                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2159                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2160                                       MF_MDC_CANCEL_FID3);
2161                 if (rc != 0)
2162                         RETURN(rc);
2163
2164                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2165                                       LCK_EX, MDS_INODELOCK_FULL,
2166                                       MF_MDC_CANCEL_FID3);
2167                 if (rc != 0)
2168                         RETURN(rc);
2169         }
2170
2171         /*
2172          * Cancel all the locks on tgt child (fid4).
2173          */
2174         if (fid_is_sane(&op_data->op_fid4))
2175                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2176                                       LCK_EX, MDS_INODELOCK_FULL,
2177                                       MF_MDC_CANCEL_FID4);
2178
2179         CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2180                op_data->op_mds, PFID(&op_data->op_fid2));
2181
2182         rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2183                        request);
2184
2185         RETURN(rc);
2186 }
2187
2188 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2189                        void *ea, int ealen, void *ea2, int ea2len,
2190                        struct ptlrpc_request **request,
2191                        struct md_open_data **mod)
2192 {
2193         struct obd_device       *obd = exp->exp_obd;
2194         struct lmv_obd          *lmv = &obd->u.lmv;
2195         struct lmv_tgt_desc     *tgt;
2196         int                      rc = 0;
2197         ENTRY;
2198
2199         rc = lmv_check_connect(obd);
2200         if (rc)
2201                 RETURN(rc);
2202
2203         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2204                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2205
2206         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2207         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2208         if (IS_ERR(tgt))
2209                 RETURN(PTR_ERR(tgt));
2210
2211         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2212                         ea2len, request, mod);
2213
2214         RETURN(rc);
2215 }
2216
2217 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2218                      struct obd_capa *oc, struct ptlrpc_request **request)
2219 {
2220         struct obd_device       *obd = exp->exp_obd;
2221         struct lmv_obd          *lmv = &obd->u.lmv;
2222         struct lmv_tgt_desc     *tgt;
2223         int                      rc;
2224         ENTRY;
2225
2226         rc = lmv_check_connect(obd);
2227         if (rc != 0)
2228                 RETURN(rc);
2229
2230         tgt = lmv_find_target(lmv, fid);
2231         if (IS_ERR(tgt))
2232                 RETURN(PTR_ERR(tgt));
2233
2234         rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2235         RETURN(rc);
2236 }
2237
2238 /*
2239  * Adjust a set of pages, each page containing an array of lu_dirpages,
2240  * so that each page can be used as a single logical lu_dirpage.
2241  *
2242  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
2243  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
2244  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
2245  * value is used as a cookie to request the next lu_dirpage in a
2246  * directory listing that spans multiple pages (two in this example):
2247  *   ________
2248  *  |        |
2249  * .|--------v-------   -----.
2250  * |s|e|f|p|ent|ent| ... |ent|
2251  * '--|--------------   -----'   Each CFS_PAGE contains a single
2252  *    '------.                   lu_dirpage.
2253  * .---------v-------   -----.
2254  * |s|e|f|p|ent| 0 | ... | 0 |
2255  * '-----------------   -----'
2256  *
2257  * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
2258  * larger than LU_PAGE_SIZE, a single host page may contain multiple
2259  * lu_dirpages. After reading the lu_dirpages from the MDS, the
2260  * ldp_hash_end of the first lu_dirpage refers to the one immediately
2261  * after it in the same CFS_PAGE (arrows simplified for brevity, but
2262  * in general e0==s1, e1==s2, etc.):
2263  *
2264  * .--------------------   -----.
2265  * |s0|e0|f0|p|ent|ent| ... |ent|
2266  * |---v----------------   -----|
2267  * |s1|e1|f1|p|ent|ent| ... |ent|
2268  * |---v----------------   -----|  Here, each CFS_PAGE contains
2269  *             ...                 multiple lu_dirpages.
2270  * |---v----------------   -----|
2271  * |s'|e'|f'|p|ent|ent| ... |ent|
2272  * '---|----------------   -----'
2273  *     v
2274  * .----------------------------.
2275  * |        next CFS_PAGE       |
2276  *
2277  * This structure is transformed into a single logical lu_dirpage as follows:
2278  *
2279  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
2280  *   labeled 'next CFS_PAGE'.
2281  *
2282  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
2283  *   a hash collision with the next page exists.
2284  *
2285  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
2286  *   to the first entry of the next lu_dirpage.
2287  */
2288 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
2289 static void lmv_adjust_dirpages(struct page **pages, int ncfspgs, int nlupgs)
2290 {
2291         int i;
2292
2293         for (i = 0; i < ncfspgs; i++) {
2294                 struct lu_dirpage       *dp = kmap(pages[i]);
2295                 struct lu_dirpage       *first = dp;
2296                 struct lu_dirent        *end_dirent = NULL;
2297                 struct lu_dirent        *ent;
2298                 __u64                   hash_end = dp->ldp_hash_end;
2299                 __u32                   flags = dp->ldp_flags;
2300
2301                 while (--nlupgs > 0) {
2302                         ent = lu_dirent_start(dp);
2303                         for (end_dirent = ent; ent != NULL;
2304                              end_dirent = ent, ent = lu_dirent_next(ent));
2305
2306                         /* Advance dp to next lu_dirpage. */
2307                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
2308
2309                         /* Check if we've reached the end of the CFS_PAGE. */
2310                         if (!((unsigned long)dp & ~CFS_PAGE_MASK))
2311                                 break;
2312
2313                         /* Save the hash and flags of this lu_dirpage. */
2314                         hash_end = dp->ldp_hash_end;
2315                         flags = dp->ldp_flags;
2316
2317                         /* Check if lu_dirpage contains no entries. */
2318                         if (!end_dirent)
2319                                 break;
2320
2321                         /* Enlarge the end entry lde_reclen from 0 to
2322                          * first entry of next lu_dirpage. */
2323                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
2324                         end_dirent->lde_reclen =
2325                                 cpu_to_le16((char *)(dp->ldp_entries) -
2326                                             (char *)end_dirent);
2327                 }
2328
2329                 first->ldp_hash_end = hash_end;
2330                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
2331                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
2332
2333                 kunmap(pages[i]);
2334         }
2335         LASSERTF(nlupgs == 0, "left = %d", nlupgs);
2336 }
2337 #else
2338 #define lmv_adjust_dirpages(pages, ncfspgs, nlupgs) do {} while (0)
2339 #endif  /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
2340
2341 #define NORMAL_MAX_STRIPES 4
2342 int lmv_read_entry(struct obd_export *exp, struct md_op_data *op_data,
2343                    struct md_callback *cb_op, struct lu_dirent **ldp,
2344                    struct page **ppage)
2345 {
2346         struct obd_device       *obd = exp->exp_obd;
2347         struct lmv_obd          *lmv = &obd->u.lmv;
2348         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2349         struct lu_dirent        *tmp_ents[NORMAL_MAX_STRIPES];
2350         struct lu_dirent        **ents = NULL;
2351         int                     stripe_count;
2352         __u64                   min_hash;
2353         int                     min_idx = 0;
2354         struct page             *min_page = NULL;
2355         int                     i;
2356         int                     rc;
2357         ENTRY;
2358
2359         rc = lmv_check_connect(obd);
2360         if (rc)
2361                 RETURN(rc);
2362
2363         if (lsm == NULL)
2364                 stripe_count = 1;
2365         else
2366                 stripe_count = lsm->lsm_md_stripe_count;
2367
2368         if (stripe_count > NORMAL_MAX_STRIPES) {
2369                 OBD_ALLOC(ents, sizeof(ents[0]) * stripe_count);
2370                 if (ents == NULL)
2371                         GOTO(out, rc = -ENOMEM);
2372         } else {
2373                 ents = tmp_ents;
2374                 memset(ents, 0, sizeof(ents[0]) * stripe_count);
2375         }
2376
2377         min_hash = MDS_DIR_END_OFF;
2378         for (i = 0; i < stripe_count; i++) {
2379                 struct lmv_tgt_desc *tgt;
2380                 struct page *page = NULL;
2381
2382                 if (likely(lsm == NULL)) {
2383                         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2384                         if (IS_ERR(tgt))
2385                                 GOTO(out, rc = PTR_ERR(tgt));
2386                         LASSERT(op_data->op_data != NULL);
2387                 } else {
2388                         tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds);
2389                         if (IS_ERR(tgt))
2390                                 GOTO(out, rc = PTR_ERR(tgt));
2391                         op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2392                         op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2393                         op_data->op_stripe_offset = i;
2394                 }
2395
2396                 rc = md_read_entry(tgt->ltd_exp, op_data, cb_op, &ents[i],
2397                                    &page);
2398                 if (rc != 0)
2399                         GOTO(out, rc);
2400
2401                 if (ents[i] != NULL &&
2402                     le64_to_cpu(ents[i]->lde_hash) <= min_hash) {
2403                         if (min_page != NULL)
2404                                 page_cache_release(min_page);
2405                         min_page = page;
2406                         min_hash = le64_to_cpu(ents[i]->lde_hash);
2407                         min_idx = i;
2408                 }
2409         }
2410
2411         if (min_hash != MDS_DIR_END_OFF)
2412                 *ldp = ents[min_idx];
2413         else
2414                 *ldp = NULL;
2415 out:
2416         if (stripe_count > NORMAL_MAX_STRIPES && ents != NULL)
2417                 OBD_FREE(ents, sizeof(ents[0]) * stripe_count);
2418
2419         if (rc != 0 && min_page != NULL) {
2420                 kunmap(min_page);
2421                 page_cache_release(min_page);
2422         } else {
2423                 *ppage = min_page;
2424         }
2425
2426         RETURN(rc);
2427 }
2428
2429 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2430                       struct ptlrpc_request **request)
2431 {
2432         struct obd_device       *obd = exp->exp_obd;
2433         struct lmv_obd          *lmv = &obd->u.lmv;
2434         struct lmv_tgt_desc     *tgt = NULL;
2435         struct lmv_tgt_desc     *parent_tgt = NULL;
2436         struct mdt_body         *body;
2437         int                     rc;
2438         ENTRY;
2439
2440         rc = lmv_check_connect(obd);
2441         if (rc)
2442                 RETURN(rc);
2443 retry:
2444         /* Send unlink requests to the MDT where the child is located */
2445         if (likely(!fid_is_zero(&op_data->op_fid2))) {
2446                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2447                 if (IS_ERR(tgt))
2448                         RETURN(PTR_ERR(tgt));
2449
2450                 /* For striped dir, we need to locate the parent as well */
2451                 if (op_data->op_mea1 != NULL &&
2452                     op_data->op_mea1->lsm_md_stripe_count > 1) {
2453                         struct lmv_tgt_desc *tmp;
2454
2455                         LASSERT(op_data->op_name != NULL &&
2456                                 op_data->op_namelen != 0);
2457                         tmp = lmv_locate_target_for_name(lmv,
2458                                                    op_data->op_mea1,
2459                                                    op_data->op_name,
2460                                                    op_data->op_namelen,
2461                                                    &op_data->op_fid1,
2462                                                    &op_data->op_mds);
2463                         if (IS_ERR(tmp))
2464                                 RETURN(PTR_ERR(tmp));
2465                 }
2466         } else {
2467                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2468                 if (IS_ERR(tgt))
2469                         RETURN(PTR_ERR(tgt));
2470         }
2471
2472         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2473         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2474         op_data->op_cap = cfs_curproc_cap_pack();
2475
2476         /*
2477          * If child's fid is given, cancel unused locks for it if it is from
2478          * another export than parent.
2479          *
2480          * LOOKUP lock for child (fid3) should also be cancelled on parent
2481          * tgt_tgt in mdc_unlink().
2482          */
2483         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2484
2485         /*
2486          * Cancel FULL locks on child (fid3).
2487          */
2488         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2489         if (IS_ERR(parent_tgt))
2490                 RETURN(PTR_ERR(parent_tgt));
2491
2492         if (parent_tgt != tgt) {
2493                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2494                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2495                                       MF_MDC_CANCEL_FID3);
2496         }
2497
2498         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2499                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2500         if (rc != 0)
2501                 RETURN(rc);
2502
2503         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2504                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2505
2506         rc = md_unlink(tgt->ltd_exp, op_data, request);
2507         if (rc != 0 && rc != -EREMOTE)
2508                 RETURN(rc);
2509
2510         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2511         if (body == NULL)
2512                 RETURN(-EPROTO);
2513
2514         /* Not cross-ref case, just get out of here. */
2515         if (likely(!(body->valid & OBD_MD_MDS)))
2516                 RETURN(0);
2517
2518         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2519                exp->exp_obd->obd_name, PFID(&body->fid1));
2520
2521         /* This is a remote object, try remote MDT, Note: it may
2522          * try more than 1 time here, Considering following case
2523          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2524          * 1. Initially A does not know where remote1 is, it send
2525          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2526          *    resend unlink RPC to MDT1 (retry 1st time).
2527          *
2528          * 2. During the unlink RPC in flight,
2529          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2530          *    and create new remote1, but on MDT0
2531          *
2532          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2533          *    /mnt/lustre, then lookup get fid of remote1, and find
2534          *    it is remote dir again, and replay -EREMOTE again.
2535          *
2536          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2537          *
2538          * In theory, it might try unlimited time here, but it should
2539          * be very rare case.  */
2540         op_data->op_fid2 = body->fid1;
2541         ptlrpc_req_finished(*request);
2542         *request = NULL;
2543
2544         goto retry;
2545 }
2546
2547 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2548 {
2549         struct lmv_obd *lmv = &obd->u.lmv;
2550         int rc = 0;
2551
2552         switch (stage) {
2553         case OBD_CLEANUP_EARLY:
2554                 /* XXX: here should be calling obd_precleanup() down to
2555                  * stack. */
2556                 break;
2557         case OBD_CLEANUP_EXPORTS:
2558                 fld_client_proc_fini(&lmv->lmv_fld);
2559                 lprocfs_obd_cleanup(obd);
2560                 lprocfs_free_md_stats(obd);
2561                 break;
2562         default:
2563                 break;
2564         }
2565         RETURN(rc);
2566 }
2567
2568 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2569                         __u32 keylen, void *key, __u32 *vallen, void *val,
2570                         struct lov_stripe_md *lsm)
2571 {
2572         struct obd_device       *obd;
2573         struct lmv_obd          *lmv;
2574         int                      rc = 0;
2575         ENTRY;
2576
2577         obd = class_exp2obd(exp);
2578         if (obd == NULL) {
2579                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2580                        exp->exp_handle.h_cookie);
2581                 RETURN(-EINVAL);
2582         }
2583
2584         lmv = &obd->u.lmv;
2585         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2586                 int i;
2587
2588                 rc = lmv_check_connect(obd);
2589                 if (rc)
2590                         RETURN(rc);
2591
2592                 LASSERT(*vallen == sizeof(__u32));
2593                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2594                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2595                         /*
2596                          * All tgts should be connected when this gets called.
2597                          */
2598                         if (tgt == NULL || tgt->ltd_exp == NULL)
2599                                 continue;
2600
2601                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2602                                           vallen, val, NULL))
2603                                 RETURN(0);
2604                 }
2605                 RETURN(-EINVAL);
2606         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2607                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2608                    KEY_IS(KEY_MAX_COOKIESIZE) ||
2609                    KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2610                    KEY_IS(KEY_CONN_DATA)) {
2611                 rc = lmv_check_connect(obd);
2612                 if (rc)
2613                         RETURN(rc);
2614
2615                 /*
2616                  * Forwarding this request to first MDS, it should know LOV
2617                  * desc.
2618                  */
2619                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2620                                   vallen, val, NULL);
2621                 if (!rc && KEY_IS(KEY_CONN_DATA))
2622                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2623                 RETURN(rc);
2624         } else if (KEY_IS(KEY_TGT_COUNT)) {
2625                 *((int *)val) = lmv->desc.ld_tgt_count;
2626                 RETURN(0);
2627         }
2628
2629         CDEBUG(D_IOCTL, "Invalid key\n");
2630         RETURN(-EINVAL);
2631 }
2632
2633 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2634                        obd_count keylen, void *key, obd_count vallen,
2635                        void *val, struct ptlrpc_request_set *set)
2636 {
2637         struct lmv_tgt_desc    *tgt = NULL;
2638         struct obd_device      *obd;
2639         struct lmv_obd         *lmv;
2640         int rc = 0;
2641         ENTRY;
2642
2643         obd = class_exp2obd(exp);
2644         if (obd == NULL) {
2645                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2646                        exp->exp_handle.h_cookie);
2647                 RETURN(-EINVAL);
2648         }
2649         lmv = &obd->u.lmv;
2650
2651         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2652                 int i, err = 0;
2653
2654                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2655                         tgt = lmv->tgts[i];
2656
2657                         if (tgt == NULL || tgt->ltd_exp == NULL)
2658                                 continue;
2659
2660                         err = obd_set_info_async(env, tgt->ltd_exp,
2661                                                  keylen, key, vallen, val, set);
2662                         if (err && rc == 0)
2663                                 rc = err;
2664                 }
2665
2666                 RETURN(rc);
2667         }
2668
2669         RETURN(-EINVAL);
2670 }
2671
2672 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2673                           struct lmv_mds_md_v1 *lmm1)
2674 {
2675         int     cplen;
2676         int     i;
2677
2678         lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2679         lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2680         lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2681         lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2682         cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2683                         sizeof(lmm1->lmv_pool_name));
2684         if (cplen >= sizeof(lmm1->lmv_pool_name))
2685                 return -E2BIG;
2686
2687         for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2688                 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2689                               &lsm->lsm_md_oinfo[i].lmo_fid);
2690         return 0;
2691 }
2692
2693 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2694                 int stripe_count)
2695 {
2696         int     lmm_size = 0;
2697         bool    allocated = false;
2698         int     rc = 0;
2699         ENTRY;
2700
2701         LASSERT(lmmp != NULL);
2702         /* Free lmm */
2703         if (*lmmp != NULL && lsm == NULL) {
2704                 int stripe_count;
2705
2706                 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2707                 lmm_size = lmv_mds_md_size(stripe_count,
2708                                            le32_to_cpu((*lmmp)->lmv_magic));
2709                 if (lmm_size == 0)
2710                         RETURN(-EINVAL);
2711                 OBD_FREE(*lmmp, lmm_size);
2712                 *lmmp = NULL;
2713                 RETURN(0);
2714         }
2715
2716         /* Alloc lmm */
2717         if (*lmmp == NULL && lsm == NULL) {
2718                 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2719                 LASSERT(lmm_size > 0);
2720                 OBD_ALLOC(*lmmp, lmm_size);
2721                 if (*lmmp == NULL)
2722                         RETURN(-ENOMEM);
2723                 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2724                 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2725                 RETURN(lmm_size);
2726         }
2727
2728         /* pack lmm */
2729         LASSERT(lsm != NULL);
2730         lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2731         if (*lmmp == NULL) {
2732                 OBD_ALLOC(*lmmp, lmm_size);
2733                 if (*lmmp == NULL)
2734                         RETURN(-ENOMEM);
2735                 allocated = true;
2736         }
2737
2738         switch (lsm->lsm_md_magic) {
2739         case LMV_MAGIC_V1:
2740                 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2741                 break;
2742         default:
2743                 rc = -EINVAL;
2744                 break;
2745         }
2746
2747         if (rc != 0 && allocated) {
2748                 OBD_FREE(*lmmp, lmm_size);
2749                 *lmmp = NULL;
2750         }
2751
2752         RETURN(lmm_size);
2753 }
2754 EXPORT_SYMBOL(lmv_pack_md);
2755
2756 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2757                             const struct lmv_mds_md_v1 *lmm1)
2758 {
2759         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
2760         int             stripe_count;
2761         int             cplen;
2762         int             i;
2763         int             rc = 0;
2764         ENTRY;
2765
2766         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2767         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2768         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2769         lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2770         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2771         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2772                         sizeof(lsm->lsm_md_pool_name));
2773
2774         if (cplen >= sizeof(lsm->lsm_md_pool_name))
2775                 RETURN(-E2BIG);
2776
2777         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2778                "layout_version %d\n", lsm->lsm_md_stripe_count,
2779                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2780                lsm->lsm_md_layout_version);
2781
2782         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2783         for (i = 0; i < le32_to_cpu(stripe_count); i++) {
2784                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2785                               &lmm1->lmv_stripe_fids[i]);
2786                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2787                                     &lsm->lsm_md_oinfo[i].lmo_mds);
2788                 if (rc != 0)
2789                         RETURN(rc);
2790                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2791                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2792         }
2793
2794         RETURN(rc);
2795 }
2796
2797 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2798                   const union lmv_mds_md *lmm, int stripe_count)
2799 {
2800         struct lmv_stripe_md     *lsm;
2801         int                      lsm_size;
2802         int                      rc;
2803         bool                     allocated = false;
2804         ENTRY;
2805
2806         LASSERT(lsmp != NULL);
2807
2808         lsm = *lsmp;
2809         /* Free memmd */
2810         if (lsm != NULL && lmm == NULL) {
2811 #ifdef __KERNEL__
2812                 int i;
2813                 for (i = 1; i < lsm->lsm_md_stripe_count; i++) {
2814                         if (lsm->lsm_md_oinfo[i].lmo_root != NULL)
2815                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
2816                 }
2817 #endif
2818                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2819                 OBD_FREE(lsm, lsm_size);
2820                 *lsmp = NULL;
2821                 RETURN(0);
2822         }
2823
2824         /* Alloc memmd */
2825         if (lsm == NULL && lmm == NULL) {
2826                 lsm_size = lmv_stripe_md_size(stripe_count);
2827                 OBD_ALLOC(lsm, lsm_size);
2828                 if (lsm == NULL)
2829                         RETURN(-ENOMEM);
2830                 lsm->lsm_md_stripe_count = stripe_count;
2831                 *lsmp = lsm;
2832                 RETURN(0);
2833         }
2834
2835         /* Unpack memmd */
2836         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
2837             le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_MIGRATE &&
2838             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
2839                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
2840                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
2841                        -EIO);
2842                 RETURN(-EIO);
2843         }
2844
2845         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1 ||
2846             le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_MIGRATE)
2847                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2848         else
2849                 /**
2850                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
2851                  * stripecount should be 0 then.
2852                  */
2853                 lsm_size = lmv_stripe_md_size(0);
2854
2855         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2856         if (lsm == NULL) {
2857                 OBD_ALLOC(lsm, lsm_size);
2858                 if (lsm == NULL)
2859                         RETURN(-ENOMEM);
2860                 allocated = true;
2861                 *lsmp = lsm;
2862         }
2863
2864         switch (le32_to_cpu(lmm->lmv_magic)) {
2865         case LMV_MAGIC_V1:
2866         case LMV_MAGIC_MIGRATE:
2867                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
2868                 break;
2869         default:
2870                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
2871                        le32_to_cpu(lmm->lmv_magic));
2872                 rc = -EINVAL;
2873                 break;
2874         }
2875
2876         if (rc != 0 && allocated) {
2877                 OBD_FREE(lsm, lsm_size);
2878                 *lsmp = NULL;
2879                 lsm_size = rc;
2880         }
2881         RETURN(lsm_size);
2882 }
2883
2884 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
2885 {
2886         return lmv_unpack_md(NULL, lsmp, NULL, stripes);
2887 }
2888 EXPORT_SYMBOL(lmv_alloc_memmd);
2889
2890 void lmv_free_memmd(struct lmv_stripe_md *lsm)
2891 {
2892         lmv_unpack_md(NULL, &lsm, NULL, 0);
2893 }
2894 EXPORT_SYMBOL(lmv_free_memmd);
2895
2896 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2897                  struct lov_mds_md *lmm, int disk_len)
2898 {
2899         return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
2900                              (union lmv_mds_md *)lmm, disk_len);
2901 }
2902
2903 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2904                struct lov_stripe_md *lsm)
2905 {
2906         struct obd_device               *obd = exp->exp_obd;
2907         struct lmv_obd                  *lmv_obd = &obd->u.lmv;
2908         const struct lmv_stripe_md      *lmv = (struct lmv_stripe_md *)lsm;
2909         int                             stripe_count;
2910
2911         if (lmmp == NULL) {
2912                 if (lsm != NULL)
2913                         stripe_count = lmv->lsm_md_stripe_count;
2914                 else
2915                         stripe_count = lmv_obd->desc.ld_tgt_count;
2916
2917                 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
2918         }
2919
2920         return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
2921 }
2922
2923 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2924                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
2925                              ldlm_cancel_flags_t flags, void *opaque)
2926 {
2927         struct obd_device       *obd = exp->exp_obd;
2928         struct lmv_obd          *lmv = &obd->u.lmv;
2929         int                      rc = 0;
2930         int                      err;
2931         __u32                    i;
2932         ENTRY;
2933
2934         LASSERT(fid != NULL);
2935
2936         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2937                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2938
2939                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
2940                         continue;
2941
2942                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
2943                                        opaque);
2944                 if (!rc)
2945                         rc = err;
2946         }
2947         RETURN(rc);
2948 }
2949
2950 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
2951                       __u64 *bits)
2952 {
2953         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2954         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
2955         int                      rc;
2956         ENTRY;
2957
2958         if (tgt == NULL || tgt->ltd_exp == NULL)
2959                 RETURN(-EINVAL);
2960         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
2961         RETURN(rc);
2962 }
2963
2964 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
2965                            const struct lu_fid *fid, ldlm_type_t type,
2966                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
2967                            struct lustre_handle *lockh)
2968 {
2969         struct obd_device       *obd = exp->exp_obd;
2970         struct lmv_obd          *lmv = &obd->u.lmv;
2971         ldlm_mode_t              rc;
2972         __u32                    i;
2973         ENTRY;
2974
2975         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2976
2977         /*
2978          * With CMD every object can have two locks in different namespaces:
2979          * lookup lock in space of mds storing direntry and update/open lock in
2980          * space of mds storing inode. Thus we check all targets, not only that
2981          * one fid was created in.
2982          */
2983         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2984                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2985
2986                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
2987                         continue;
2988
2989                 rc = md_lock_match(tgt->ltd_exp, flags, fid, type, policy, mode,
2990                                    lockh);
2991                 if (rc)
2992                         RETURN(rc);
2993         }
2994
2995         RETURN(0);
2996 }
2997
2998 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2999                       struct obd_export *dt_exp, struct obd_export *md_exp,
3000                       struct lustre_md *md)
3001 {
3002         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3003         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3004
3005         if (tgt == NULL || tgt->ltd_exp == NULL)
3006                 RETURN(-EINVAL);
3007
3008         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3009 }
3010
3011 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3012 {
3013         struct obd_device       *obd = exp->exp_obd;
3014         struct lmv_obd          *lmv = &obd->u.lmv;
3015         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3016         ENTRY;
3017
3018         if (md->lmv != NULL) {
3019                 lmv_free_memmd(md->lmv);
3020                 md->lmv = NULL;
3021         }
3022         if (tgt == NULL || tgt->ltd_exp == NULL)
3023                 RETURN(-EINVAL);
3024         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3025 }
3026
3027 int lmv_set_open_replay_data(struct obd_export *exp,
3028                              struct obd_client_handle *och,
3029                              struct lookup_intent *it)
3030 {
3031         struct obd_device       *obd = exp->exp_obd;
3032         struct lmv_obd          *lmv = &obd->u.lmv;
3033         struct lmv_tgt_desc     *tgt;
3034         ENTRY;
3035
3036         tgt = lmv_find_target(lmv, &och->och_fid);
3037         if (IS_ERR(tgt))
3038                 RETURN(PTR_ERR(tgt));
3039
3040         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3041 }
3042
3043 int lmv_clear_open_replay_data(struct obd_export *exp,
3044                                struct obd_client_handle *och)
3045 {
3046         struct obd_device       *obd = exp->exp_obd;
3047         struct lmv_obd          *lmv = &obd->u.lmv;
3048         struct lmv_tgt_desc     *tgt;
3049         ENTRY;
3050
3051         tgt = lmv_find_target(lmv, &och->och_fid);
3052         if (IS_ERR(tgt))
3053                 RETURN(PTR_ERR(tgt));
3054
3055         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3056 }
3057
3058 static int lmv_get_remote_perm(struct obd_export *exp,
3059                                const struct lu_fid *fid,
3060                                struct obd_capa *oc, __u32 suppgid,
3061                                struct ptlrpc_request **request)
3062 {
3063         struct obd_device       *obd = exp->exp_obd;
3064         struct lmv_obd          *lmv = &obd->u.lmv;
3065         struct lmv_tgt_desc     *tgt;
3066         int                      rc;
3067         ENTRY;
3068
3069         rc = lmv_check_connect(obd);
3070         if (rc)
3071                 RETURN(rc);
3072
3073         tgt = lmv_find_target(lmv, fid);
3074         if (IS_ERR(tgt))
3075                 RETURN(PTR_ERR(tgt));
3076
3077         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3078         RETURN(rc);
3079 }
3080
3081 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3082                           renew_capa_cb_t cb)
3083 {
3084         struct obd_device       *obd = exp->exp_obd;
3085         struct lmv_obd          *lmv = &obd->u.lmv;
3086         struct lmv_tgt_desc     *tgt;
3087         int                      rc;
3088         ENTRY;
3089
3090         rc = lmv_check_connect(obd);
3091         if (rc)
3092                 RETURN(rc);
3093
3094         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3095         if (IS_ERR(tgt))
3096                 RETURN(PTR_ERR(tgt));
3097
3098         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3099         RETURN(rc);
3100 }
3101
3102 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3103                     const struct req_msg_field *field, struct obd_capa **oc)
3104 {
3105         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3106         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3107
3108         if (tgt == NULL || tgt->ltd_exp == NULL)
3109                 RETURN(-EINVAL);
3110         return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3111 }
3112
3113 int lmv_intent_getattr_async(struct obd_export *exp,
3114                              struct md_enqueue_info *minfo,
3115                              struct ldlm_enqueue_info *einfo)
3116 {
3117         struct md_op_data       *op_data = &minfo->mi_data;
3118         struct obd_device       *obd = exp->exp_obd;
3119         struct lmv_obd          *lmv = &obd->u.lmv;
3120         struct lmv_tgt_desc     *tgt = NULL;
3121         int                      rc;
3122         ENTRY;
3123
3124         rc = lmv_check_connect(obd);
3125         if (rc)
3126                 RETURN(rc);
3127
3128         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3129         if (IS_ERR(tgt))
3130                 RETURN(PTR_ERR(tgt));
3131
3132         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3133         RETURN(rc);
3134 }
3135
3136 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3137                         struct lu_fid *fid, __u64 *bits)
3138 {
3139         struct obd_device       *obd = exp->exp_obd;
3140         struct lmv_obd          *lmv = &obd->u.lmv;
3141         struct lmv_tgt_desc     *tgt;
3142         int                      rc;
3143         ENTRY;
3144
3145         rc = lmv_check_connect(obd);
3146         if (rc)
3147                 RETURN(rc);
3148
3149         tgt = lmv_find_target(lmv, fid);
3150         if (IS_ERR(tgt))
3151                 RETURN(PTR_ERR(tgt));
3152
3153         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3154         RETURN(rc);
3155 }
3156
3157 /**
3158  * For lmv, only need to send request to master MDT, and the master MDT will
3159  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3160  * we directly fetch data from the slave MDTs.
3161  */
3162 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3163                  struct obd_quotactl *oqctl)
3164 {
3165         struct obd_device   *obd = class_exp2obd(exp);
3166         struct lmv_obd      *lmv = &obd->u.lmv;
3167         struct lmv_tgt_desc *tgt = lmv->tgts[0];
3168         int                  rc = 0;
3169         __u32                i;
3170         __u64                curspace, curinodes;
3171         ENTRY;
3172
3173         if (tgt == NULL ||
3174             tgt->ltd_exp == NULL ||
3175             !tgt->ltd_active ||
3176             lmv->desc.ld_tgt_count == 0) {
3177                 CERROR("master lmv inactive\n");
3178                 RETURN(-EIO);
3179         }
3180
3181         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3182                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3183                 RETURN(rc);
3184         }
3185
3186         curspace = curinodes = 0;
3187         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3188                 int err;
3189                 tgt = lmv->tgts[i];
3190
3191                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3192                         continue;
3193
3194                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3195                 if (err) {
3196                         CERROR("getquota on mdt %d failed. %d\n", i, err);
3197                         if (!rc)
3198                                 rc = err;
3199                 } else {
3200                         curspace += oqctl->qc_dqblk.dqb_curspace;
3201                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3202                 }
3203         }
3204         oqctl->qc_dqblk.dqb_curspace = curspace;
3205         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3206
3207         RETURN(rc);
3208 }
3209
3210 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3211                    struct obd_quotactl *oqctl)
3212 {
3213         struct obd_device       *obd = class_exp2obd(exp);
3214         struct lmv_obd          *lmv = &obd->u.lmv;
3215         struct lmv_tgt_desc     *tgt;
3216         __u32                    i;
3217         int                      rc = 0;
3218         ENTRY;
3219
3220         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3221                 int err;
3222                 tgt = lmv->tgts[i];
3223                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3224                         CERROR("lmv idx %d inactive\n", i);
3225                         RETURN(-EIO);
3226                 }
3227
3228                 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3229                 if (err && !rc)
3230                         rc = err;
3231         }
3232
3233         RETURN(rc);
3234 }
3235
3236 int lmv_update_lsm_md(struct obd_export *exp, struct lmv_stripe_md *lsm,
3237                       struct mdt_body *body, ldlm_blocking_callback cb_blocking)
3238 {
3239         if (lsm->lsm_md_stripe_count <= 1)
3240                 return 0;
3241
3242         return lmv_revalidate_slaves(exp, body, lsm, cb_blocking, 0);
3243 }
3244
3245 int lmv_merge_attr(struct obd_export *exp, const struct lmv_stripe_md *lsm,
3246                    struct cl_attr *attr)
3247 {
3248 #ifdef __KERNEL__
3249         int i;
3250
3251         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3252                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3253
3254                 CDEBUG(D_INFO, ""DFID" size %llu, nlink %u, atime %lu ctime"
3255                        "%lu, mtime %lu.\n", PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3256                        i_size_read(inode), inode->i_nlink,
3257                        LTIME_S(inode->i_atime), LTIME_S(inode->i_ctime),
3258                        LTIME_S(inode->i_mtime));
3259
3260                 /* for slave stripe, it needs to subtract nlink for . and .. */
3261                 if (i != 0)
3262                         attr->cat_nlink += inode->i_nlink - 2;
3263                 else
3264                         attr->cat_nlink = inode->i_nlink;
3265
3266                 attr->cat_size += i_size_read(inode);
3267
3268                 if (attr->cat_atime < LTIME_S(inode->i_atime))
3269                         attr->cat_atime = LTIME_S(inode->i_atime);
3270
3271                 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3272                         attr->cat_ctime = LTIME_S(inode->i_ctime);
3273
3274                 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3275                         attr->cat_mtime = LTIME_S(inode->i_mtime);
3276         }
3277 #endif
3278         return 0;
3279 }
3280
3281 struct obd_ops lmv_obd_ops = {
3282         .o_owner                = THIS_MODULE,
3283         .o_setup                = lmv_setup,
3284         .o_cleanup              = lmv_cleanup,
3285         .o_precleanup           = lmv_precleanup,
3286         .o_process_config       = lmv_process_config,
3287         .o_connect              = lmv_connect,
3288         .o_disconnect           = lmv_disconnect,
3289         .o_statfs               = lmv_statfs,
3290         .o_get_info             = lmv_get_info,
3291         .o_set_info_async       = lmv_set_info_async,
3292         .o_packmd               = lmv_packmd,
3293         .o_unpackmd             = lmv_unpackmd,
3294         .o_notify               = lmv_notify,
3295         .o_get_uuid             = lmv_get_uuid,
3296         .o_iocontrol            = lmv_iocontrol,
3297         .o_quotacheck           = lmv_quotacheck,
3298         .o_quotactl             = lmv_quotactl
3299 };
3300
3301 struct md_ops lmv_md_ops = {
3302         .m_getstatus            = lmv_getstatus,
3303         .m_null_inode           = lmv_null_inode,
3304         .m_find_cbdata          = lmv_find_cbdata,
3305         .m_close                = lmv_close,
3306         .m_create               = lmv_create,
3307         .m_done_writing         = lmv_done_writing,
3308         .m_enqueue              = lmv_enqueue,
3309         .m_getattr              = lmv_getattr,
3310         .m_getxattr             = lmv_getxattr,
3311         .m_getattr_name         = lmv_getattr_name,
3312         .m_intent_lock          = lmv_intent_lock,
3313         .m_link                 = lmv_link,
3314         .m_rename               = lmv_rename,
3315         .m_setattr              = lmv_setattr,
3316         .m_setxattr             = lmv_setxattr,
3317         .m_fsync                = lmv_fsync,
3318         .m_read_entry           = lmv_read_entry,
3319         .m_unlink               = lmv_unlink,
3320         .m_init_ea_size         = lmv_init_ea_size,
3321         .m_cancel_unused        = lmv_cancel_unused,
3322         .m_set_lock_data        = lmv_set_lock_data,
3323         .m_lock_match           = lmv_lock_match,
3324         .m_get_lustre_md        = lmv_get_lustre_md,
3325         .m_free_lustre_md       = lmv_free_lustre_md,
3326         .m_update_lsm_md        = lmv_update_lsm_md,
3327         .m_merge_attr           = lmv_merge_attr,
3328         .m_set_open_replay_data = lmv_set_open_replay_data,
3329         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3330         .m_renew_capa           = lmv_renew_capa,
3331         .m_unpack_capa          = lmv_unpack_capa,
3332         .m_get_remote_perm      = lmv_get_remote_perm,
3333         .m_intent_getattr_async = lmv_intent_getattr_async,
3334         .m_revalidate_lock      = lmv_revalidate_lock
3335 };
3336
3337 int __init lmv_init(void)
3338 {
3339         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3340 #ifndef HAVE_ONLY_PROCFS_SEQ
3341                                    NULL,
3342 #endif
3343                                    LUSTRE_LMV_NAME, NULL);
3344 }
3345
3346 #ifdef __KERNEL__
3347 static void lmv_exit(void)
3348 {
3349         class_unregister_type(LUSTRE_LMV_NAME);
3350 }
3351
3352 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3353 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3354 MODULE_LICENSE("GPL");
3355
3356 module_init(lmv_init);
3357 module_exit(lmv_exit);
3358 #endif