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(__u32 lmv_hash_type, unsigned int stripe_count,
  94                              const char *name, int namelen)
  95 {
  96         int     idx;
  97         __u32   hash_type = lmv_hash_type & LMV_HASH_TYPE_MASK;
  98
  99         LASSERT(namelen > 0);
 100         if (stripe_count <= 1)
 101                 return 0;
 102
 103         /* for migrating object, always start from 0 stripe */
 104         if (lmv_hash_type & LMV_HASH_FLAG_MIGRATION)
 105                 return 0;
 106
 107         switch (hash_type) {
 108         case LMV_HASH_TYPE_ALL_CHARS:
 109                 idx = lmv_hash_all_chars(stripe_count, name, namelen);
 110                 break;
 111         case LMV_HASH_TYPE_FNV_1A_64:
 112                 idx = lmv_hash_fnv1a(stripe_count, name, namelen);
 113                 break;
 114         default:
 115                 CERROR("Unknown hash type 0x%x\n", hash_type);
 116                 return -EINVAL;
 117         }
 118
 119         CDEBUG(D_INFO, "name %.*s hash_type %d idx %d\n", namelen, name,
 120                hash_type, idx);
 121
 122         return idx;
 123 }
 124
 125 static void lmv_activate_target(struct lmv_obd *lmv,
 126                                 struct lmv_tgt_desc *tgt,
 127                                 int activate)
 128 {
 129         if (tgt->ltd_active == activate)
 130                 return;
 131
 132         tgt->ltd_active = activate;
 133         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
 134 }
 135
 136 /**
 137  * Error codes:
 138  *
 139  *  -EINVAL  : UUID can't be found in the LMV's target list
 140  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 141  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
 142  */
 143 static int lmv_set_mdc_active(struct lmv_obd *lmv,
 144                               const struct obd_uuid *uuid,
 145                               int activate)
 146 {
 147         struct lmv_tgt_desc     *tgt = NULL;
 148         struct obd_device       *obd;
 149         __u32                    i;
 150         int                      rc = 0;
 151         ENTRY;
 152
 153         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
 154                         lmv, uuid->uuid, activate);
 155
 156         spin_lock(&lmv->lmv_lock);
 157         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 158                 tgt = lmv->tgts[i];
 159                 if (tgt == NULL || tgt->ltd_exp == NULL)
 160                         continue;
 161
 162                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
 163                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 164
 165                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 166                         break;
 167         }
 168
 169         if (i == lmv->desc.ld_tgt_count)
 170                 GOTO(out_lmv_lock, rc = -EINVAL);
 171
 172         obd = class_exp2obd(tgt->ltd_exp);
 173         if (obd == NULL)
 174                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 175
 176         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 177                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 178                obd->obd_type->typ_name, i);
 179         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 180
 181         if (tgt->ltd_active == activate) {
 182                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 183                        activate ? "" : "in");
 184                 GOTO(out_lmv_lock, rc);
 185         }
 186
 187         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 188                activate ? "" : "in");
 189         lmv_activate_target(lmv, tgt, activate);
 190         EXIT;
 191
 192  out_lmv_lock:
 193         spin_unlock(&lmv->lmv_lock);
 194         return rc;
 195 }
 196
 197 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 198 {
 199         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
 200         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
 201
 202         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
 203 }
 204
 205 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 206                       enum obd_notify_event ev, void *data)
 207 {
 208         struct obd_connect_data *conn_data;
 209         struct lmv_obd          *lmv = &obd->u.lmv;
 210         struct obd_uuid         *uuid;
 211         int                      rc = 0;
 212         ENTRY;
 213
 214         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 215                 CERROR("unexpected notification of %s %s!\n",
 216                        watched->obd_type->typ_name,
 217                        watched->obd_name);
 218                 RETURN(-EINVAL);
 219         }
 220
 221         uuid = &watched->u.cli.cl_target_uuid;
 222         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 223                 /*
 224                  * Set MDC as active before notifying the observer, so the
 225                  * observer can use the MDC normally.
 226                  */
 227                 rc = lmv_set_mdc_active(lmv, uuid,
 228                                         ev == OBD_NOTIFY_ACTIVE);
 229                 if (rc) {
 230                         CERROR("%sactivation of %s failed: %d\n",
 231                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 232                                uuid->uuid, rc);
 233                         RETURN(rc);
 234                 }
 235         } else if (ev == OBD_NOTIFY_OCD) {
 236                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 237                 /*
 238                  * XXX: Make sure that ocd_connect_flags from all targets are
 239                  * the same. Otherwise one of MDTs runs wrong version or
 240                  * something like this.  --umka
 241                  */
 242                 obd->obd_self_export->exp_connect_data = *conn_data;
 243         }
 244 #if 0
 245         else if (ev == OBD_NOTIFY_DISCON) {
 246                 /*
 247                  * For disconnect event, flush fld cache for failout MDS case.
 248                  */
 249                 fld_client_flush(&lmv->lmv_fld);
 250         }
 251 #endif
 252         /*
 253          * Pass the notification up the chain.
 254          */
 255         if (obd->obd_observer)
 256                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 257
 258         RETURN(rc);
 259 }
 260
 261 /**
 262  * This is fake connect function. Its purpose is to initialize lmv and say
 263  * caller that everything is okay. Real connection will be performed later.
 264  */
 265 static int lmv_connect(const struct lu_env *env,
 266                        struct obd_export **exp, struct obd_device *obd,
 267                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 268                        void *localdata)
 269 {
 270         struct lmv_obd        *lmv = &obd->u.lmv;
 271         struct lustre_handle  conn = { 0 };
 272         int                    rc = 0;
 273         ENTRY;
 274
 275         /*
 276          * We don't want to actually do the underlying connections more than
 277          * once, so keep track.
 278          */
 279         lmv->refcount++;
 280         if (lmv->refcount > 1) {
 281                 *exp = NULL;
 282                 RETURN(0);
 283         }
 284
 285         rc = class_connect(&conn, obd, cluuid);
 286         if (rc) {
 287                 CERROR("class_connection() returned %d\n", rc);
 288                 RETURN(rc);
 289         }
 290
 291         *exp = class_conn2export(&conn);
 292         class_export_get(*exp);
 293
 294         lmv->exp = *exp;
 295         lmv->connected = 0;
 296         lmv->cluuid = *cluuid;
 297
 298         if (data)
 299                 lmv->conn_data = *data;
 300
 301         if (lmv->targets_proc_entry == NULL) {
 302                 lmv->targets_proc_entry = lprocfs_seq_register("target_obds",
 303                                                         obd->obd_proc_entry,
 304                                                         NULL, NULL);
 305                 if (IS_ERR(lmv->targets_proc_entry)) {
 306                         CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 307                                obd->obd_type->typ_name, obd->obd_name);
 308                         lmv->targets_proc_entry = NULL;
 309                 }
 310         }
 311
 312         /*
 313          * All real clients should perform actual connection right away, because
 314          * it is possible, that LMV will not have opportunity to connect targets
 315          * and MDC stuff will be called directly, for instance while reading
 316          * ../mdc/../kbytesfree procfs file, etc.
 317          */
 318         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
 319                 rc = lmv_check_connect(obd);
 320
 321         if (rc && lmv->targets_proc_entry != NULL)
 322                 lprocfs_remove(&lmv->targets_proc_entry);
 323         RETURN(rc);
 324 }
 325
 326 static void lmv_set_timeouts(struct obd_device *obd)
 327 {
 328         struct lmv_obd          *lmv;
 329         __u32                    i;
 330
 331         lmv = &obd->u.lmv;
 332         if (lmv->server_timeout == 0)
 333                 return;
 334
 335         if (lmv->connected == 0)
 336                 return;
 337
 338         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 339                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 340
 341                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
 342                         continue;
 343
 344                 obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
 345                                    KEY_INTERMDS, 0, NULL, NULL);
 346         }
 347 }
 348
 349 static int lmv_init_ea_size(struct obd_export *exp, int easize,
 350                             int def_easize, int cookiesize, int def_cookiesize)
 351 {
 352         struct obd_device       *obd = exp->exp_obd;
 353         struct lmv_obd          *lmv = &obd->u.lmv;
 354         __u32                    i;
 355         int                      rc = 0;
 356         int                      change = 0;
 357         ENTRY;
 358
 359         if (lmv->max_easize < easize) {
 360                 lmv->max_easize = easize;
 361                 change = 1;
 362         }
 363         if (lmv->max_def_easize < def_easize) {
 364                 lmv->max_def_easize = def_easize;
 365                 change = 1;
 366         }
 367         if (lmv->max_cookiesize < cookiesize) {
 368                 lmv->max_cookiesize = cookiesize;
 369                 change = 1;
 370         }
 371         if (lmv->max_def_cookiesize < def_cookiesize) {
 372                 lmv->max_def_cookiesize = def_cookiesize;
 373                 change = 1;
 374         }
 375         if (change == 0)
 376                 RETURN(0);
 377
 378         if (lmv->connected == 0)
 379                 RETURN(0);
 380
 381         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 382                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 383
 384                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
 385                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 386                         continue;
 387                 }
 388
 389                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
 390                                      cookiesize, def_cookiesize);
 391                 if (rc) {
 392                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 393                                " rc = %d.\n", obd->obd_name, i, rc);
 394                         break;
 395                 }
 396         }
 397         RETURN(rc);
 398 }
 399
 400 #define MAX_STRING_SIZE 128
 401
 402 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 403 {
 404         struct lmv_obd          *lmv = &obd->u.lmv;
 405         struct obd_uuid         *cluuid = &lmv->cluuid;
 406         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 407         struct obd_device       *mdc_obd;
 408         struct obd_export       *mdc_exp;
 409         struct lu_fld_target     target;
 410         int                      rc;
 411         ENTRY;
 412
 413         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 414                                         &obd->obd_uuid);
 415         if (!mdc_obd) {
 416                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 417                 RETURN(-EINVAL);
 418         }
 419
 420         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 421                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 422                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 423                 cluuid->uuid);
 424
 425         if (!mdc_obd->obd_set_up) {
 426                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 427                 RETURN(-EINVAL);
 428         }
 429
 430         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 431                          &lmv->conn_data, NULL);
 432         if (rc) {
 433                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 434                 RETURN(rc);
 435         }
 436
 437         /*
 438          * Init fid sequence client for this mdc and add new fld target.
 439          */
 440         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 441         if (rc)
 442                 RETURN(rc);
 443
 444         target.ft_srv = NULL;
 445         target.ft_exp = mdc_exp;
 446         target.ft_idx = tgt->ltd_idx;
 447
 448         fld_client_add_target(&lmv->lmv_fld, &target);
 449
 450         rc = obd_register_observer(mdc_obd, obd);
 451         if (rc) {
 452                 obd_disconnect(mdc_exp);
 453                 CERROR("target %s register_observer error %d\n",
 454                        tgt->ltd_uuid.uuid, rc);
 455                 RETURN(rc);
 456         }
 457
 458         if (obd->obd_observer) {
 459                 /*
 460                  * Tell the observer about the new target.
 461                  */
 462                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 463                                 OBD_NOTIFY_ACTIVE,
 464                                 (void *)(tgt - lmv->tgts[0]));
 465                 if (rc) {
 466                         obd_disconnect(mdc_exp);
 467                         RETURN(rc);
 468                 }
 469         }
 470
 471         tgt->ltd_active = 1;
 472         tgt->ltd_exp = mdc_exp;
 473         lmv->desc.ld_active_tgt_count++;
 474
 475         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
 476                         lmv->max_cookiesize, lmv->max_def_cookiesize);
 477
 478         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 479                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 480                 atomic_read(&obd->obd_refcount));
 481
 482         if (lmv->targets_proc_entry != NULL) {
 483                 struct proc_dir_entry *mdc_symlink;
 484
 485                 LASSERT(mdc_obd->obd_type != NULL);
 486                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 487                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 488                                                   lmv->targets_proc_entry,
 489                                                   "../../../%s/%s",
 490                                                   mdc_obd->obd_type->typ_name,
 491                                                   mdc_obd->obd_name);
 492                 if (mdc_symlink == NULL) {
 493                         CERROR("Could not register LMV target "
 494                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 495                                obd->obd_type->typ_name, obd->obd_name,
 496                                mdc_obd->obd_name);
 497                 }
 498         }
 499         RETURN(0);
 500 }
 501
 502 static void lmv_del_target(struct lmv_obd *lmv, int index)
 503 {
 504         if (lmv->tgts[index] == NULL)
 505                 return;
 506
 507         OBD_FREE_PTR(lmv->tgts[index]);
 508         lmv->tgts[index] = NULL;
 509         return;
 510 }
 511
 512 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 513                            __u32 index, int gen)
 514 {
 515         struct lmv_obd      *lmv = &obd->u.lmv;
 516         struct lmv_tgt_desc *tgt;
 517         int                  orig_tgt_count = 0;
 518         int                  rc = 0;
 519         ENTRY;
 520
 521         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 522
 523         lmv_init_lock(lmv);
 524
 525         if (lmv->desc.ld_tgt_count == 0) {
 526                 struct obd_device *mdc_obd;
 527
 528                 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 529                                                 &obd->obd_uuid);
 530                 if (!mdc_obd) {
 531                         lmv_init_unlock(lmv);
 532                         CERROR("%s: Target %s not attached: rc = %d\n",
 533                                obd->obd_name, uuidp->uuid, -EINVAL);
 534                         RETURN(-EINVAL);
 535                 }
 536         }
 537
 538         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
 539                 tgt = lmv->tgts[index];
 540                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
 541                        " rc = %d\n", obd->obd_name,
 542                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
 543                 lmv_init_unlock(lmv);
 544                 RETURN(-EEXIST);
 545         }
 546
 547         if (index >= lmv->tgts_size) {
 548                 /* We need to reallocate the lmv target array. */
 549                 struct lmv_tgt_desc **newtgts, **old = NULL;
 550                 __u32 newsize = 1;
 551                 __u32 oldsize = 0;
 552
 553                 while (newsize < index + 1)
 554                         newsize = newsize << 1;
 555                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 556                 if (newtgts == NULL) {
 557                         lmv_init_unlock(lmv);
 558                         RETURN(-ENOMEM);
 559                 }
 560
 561                 if (lmv->tgts_size) {
 562                         memcpy(newtgts, lmv->tgts,
 563                                sizeof(*newtgts) * lmv->tgts_size);
 564                         old = lmv->tgts;
 565                         oldsize = lmv->tgts_size;
 566                 }
 567
 568                 lmv->tgts = newtgts;
 569                 lmv->tgts_size = newsize;
 570                 smp_rmb();
 571                 if (old)
 572                         OBD_FREE(old, sizeof(*old) * oldsize);
 573
 574                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
 575                        lmv->tgts_size);
 576         }
 577
 578         OBD_ALLOC_PTR(tgt);
 579         if (!tgt) {
 580                 lmv_init_unlock(lmv);
 581                 RETURN(-ENOMEM);
 582         }
 583
 584         mutex_init(&tgt->ltd_fid_mutex);
 585         tgt->ltd_idx = index;
 586         tgt->ltd_uuid = *uuidp;
 587         tgt->ltd_active = 0;
 588         lmv->tgts[index] = tgt;
 589         if (index >= lmv->desc.ld_tgt_count) {
 590                 orig_tgt_count = lmv->desc.ld_tgt_count;
 591                 lmv->desc.ld_tgt_count = index + 1;
 592         }
 593
 594         if (lmv->connected) {
 595                 rc = lmv_connect_mdc(obd, tgt);
 596                 if (rc != 0) {
 597                         spin_lock(&lmv->lmv_lock);
 598                         if (lmv->desc.ld_tgt_count == index + 1)
 599                                 lmv->desc.ld_tgt_count = orig_tgt_count;
 600                         memset(tgt, 0, sizeof(*tgt));
 601                         spin_unlock(&lmv->lmv_lock);
 602                 } else {
 603                         int easize = sizeof(struct lmv_stripe_md) +
 604                                 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
 605                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
 606                 }
 607         }
 608
 609         lmv_init_unlock(lmv);
 610         RETURN(rc);
 611 }
 612
 613 int lmv_check_connect(struct obd_device *obd)
 614 {
 615         struct lmv_obd          *lmv = &obd->u.lmv;
 616         struct lmv_tgt_desc     *tgt;
 617         __u32                    i;
 618         int                      rc;
 619         int                      easize;
 620         ENTRY;
 621
 622         if (lmv->connected)
 623                 RETURN(0);
 624
 625         lmv_init_lock(lmv);
 626         if (lmv->connected) {
 627                 lmv_init_unlock(lmv);
 628                 RETURN(0);
 629         }
 630
 631         if (lmv->desc.ld_tgt_count == 0) {
 632                 lmv_init_unlock(lmv);
 633                 CERROR("%s: no targets configured.\n", obd->obd_name);
 634                 RETURN(-EINVAL);
 635         }
 636
 637         LASSERT(lmv->tgts != NULL);
 638
 639         if (lmv->tgts[0] == NULL) {
 640                 lmv_init_unlock(lmv);
 641                 CERROR("%s: no target configured for index 0.\n",
 642                        obd->obd_name);
 643                 RETURN(-EINVAL);
 644         }
 645
 646         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 647                lmv->cluuid.uuid, obd->obd_name);
 648
 649         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 650                 tgt = lmv->tgts[i];
 651                 if (tgt == NULL)
 652                         continue;
 653                 rc = lmv_connect_mdc(obd, tgt);
 654                 if (rc)
 655                         GOTO(out_disc, rc);
 656         }
 657
 658         lmv_set_timeouts(obd);
 659         class_export_put(lmv->exp);
 660         lmv->connected = 1;
 661         easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
 662         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
 663         lmv_init_unlock(lmv);
 664         RETURN(0);
 665
 666  out_disc:
 667         while (i-- > 0) {
 668                 int rc2;
 669                 tgt = lmv->tgts[i];
 670                 if (tgt == NULL)
 671                         continue;
 672                 tgt->ltd_active = 0;
 673                 if (tgt->ltd_exp) {
 674                         --lmv->desc.ld_active_tgt_count;
 675                         rc2 = obd_disconnect(tgt->ltd_exp);
 676                         if (rc2) {
 677                                 CERROR("LMV target %s disconnect on "
 678                                        "MDC idx %d: error %d\n",
 679                                        tgt->ltd_uuid.uuid, i, rc2);
 680                         }
 681                 }
 682         }
 683         class_disconnect(lmv->exp);
 684         lmv_init_unlock(lmv);
 685         RETURN(rc);
 686 }
 687
 688 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 689 {
 690         struct lmv_obd         *lmv = &obd->u.lmv;
 691         struct obd_device      *mdc_obd;
 692         int                     rc;
 693         ENTRY;
 694
 695         LASSERT(tgt != NULL);
 696         LASSERT(obd != NULL);
 697
 698         mdc_obd = class_exp2obd(tgt->ltd_exp);
 699
 700         if (mdc_obd) {
 701                 mdc_obd->obd_force = obd->obd_force;
 702                 mdc_obd->obd_fail = obd->obd_fail;
 703                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 704         }
 705
 706         if (lmv->targets_proc_entry != NULL)
 707                 lprocfs_remove_proc_entry(mdc_obd->obd_name,
 708                                           lmv->targets_proc_entry);
 709
 710         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 711         if (rc)
 712                 CERROR("Can't finanize fids factory\n");
 713
 714         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 715                tgt->ltd_exp->exp_obd->obd_name,
 716                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 717
 718         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 719         rc = obd_disconnect(tgt->ltd_exp);
 720         if (rc) {
 721                 if (tgt->ltd_active) {
 722                         CERROR("Target %s disconnect error %d\n",
 723                                tgt->ltd_uuid.uuid, rc);
 724                 }
 725         }
 726
 727         lmv_activate_target(lmv, tgt, 0);
 728         tgt->ltd_exp = NULL;
 729         RETURN(0);
 730 }
 731
 732 static int lmv_disconnect(struct obd_export *exp)
 733 {
 734         struct obd_device       *obd = class_exp2obd(exp);
 735         struct lmv_obd          *lmv = &obd->u.lmv;
 736         int                      rc;
 737         __u32                    i;
 738         ENTRY;
 739
 740         if (!lmv->tgts)
 741                 goto out_local;
 742
 743         /*
 744          * Only disconnect the underlying layers on the final disconnect.
 745          */
 746         lmv->refcount--;
 747         if (lmv->refcount != 0)
 748                 goto out_local;
 749
 750         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 751                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
 752                         continue;
 753
 754                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
 755         }
 756
 757         if (lmv->targets_proc_entry != NULL)
 758                 lprocfs_remove(&lmv->targets_proc_entry);
 759         else
 760                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 761                        obd->obd_type->typ_name, obd->obd_name);
 762
 763 out_local:
 764         /*
 765          * This is the case when no real connection is established by
 766          * lmv_check_connect().
 767          */
 768         if (!lmv->connected)
 769                 class_export_put(exp);
 770         rc = class_disconnect(exp);
 771         if (lmv->refcount == 0)
 772                 lmv->connected = 0;
 773         RETURN(rc);
 774 }
 775
 776 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
 777 {
 778         struct obd_device       *obddev = class_exp2obd(exp);
 779         struct lmv_obd          *lmv = &obddev->u.lmv;
 780         struct getinfo_fid2path *gf;
 781         struct lmv_tgt_desc     *tgt;
 782         struct getinfo_fid2path *remote_gf = NULL;
 783         int                     remote_gf_size = 0;
 784         int                     rc;
 785
 786         gf = (struct getinfo_fid2path *)karg;
 787         tgt = lmv_find_target(lmv, &gf->gf_fid);
 788         if (IS_ERR(tgt))
 789                 RETURN(PTR_ERR(tgt));
 790
 791 repeat_fid2path:
 792         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 793         if (rc != 0 && rc != -EREMOTE)
 794                 GOTO(out_fid2path, rc);
 795
 796         /* If remote_gf != NULL, it means just building the
 797          * path on the remote MDT, copy this path segement to gf */
 798         if (remote_gf != NULL) {
 799                 struct getinfo_fid2path *ori_gf;
 800                 char *ptr;
 801
 802                 ori_gf = (struct getinfo_fid2path *)karg;
 803                 if (strlen(ori_gf->gf_path) +
 804                     strlen(gf->gf_path) > ori_gf->gf_pathlen)
 805                         GOTO(out_fid2path, rc = -EOVERFLOW);
 806
 807                 ptr = ori_gf->gf_path;
 808
 809                 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
 810                         strlen(ori_gf->gf_path));
 811
 812                 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
 813                 ptr += strlen(gf->gf_path);
 814                 *ptr = '/';
 815         }
 816
 817         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
 818                tgt->ltd_exp->exp_obd->obd_name,
 819                gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 820                gf->gf_linkno);
 821
 822         if (rc == 0)
 823                 GOTO(out_fid2path, rc);
 824
 825         /* sigh, has to go to another MDT to do path building further */
 826         if (remote_gf == NULL) {
 827                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 828                 OBD_ALLOC(remote_gf, remote_gf_size);
 829                 if (remote_gf == NULL)
 830                         GOTO(out_fid2path, rc = -ENOMEM);
 831                 remote_gf->gf_pathlen = PATH_MAX;
 832         }
 833
 834         if (!fid_is_sane(&gf->gf_fid)) {
 835                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 836                        tgt->ltd_exp->exp_obd->obd_name,
 837                        PFID(&gf->gf_fid), -EINVAL);
 838                 GOTO(out_fid2path, rc = -EINVAL);
 839         }
 840
 841         tgt = lmv_find_target(lmv, &gf->gf_fid);
 842         if (IS_ERR(tgt))
 843                 GOTO(out_fid2path, rc = -EINVAL);
 844
 845         remote_gf->gf_fid = gf->gf_fid;
 846         remote_gf->gf_recno = -1;
 847         remote_gf->gf_linkno = -1;
 848         memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
 849         gf = remote_gf;
 850         goto repeat_fid2path;
 851
 852 out_fid2path:
 853         if (remote_gf != NULL)
 854                 OBD_FREE(remote_gf, remote_gf_size);
 855         RETURN(rc);
 856 }
 857
 858 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 859                              const struct hsm_user_request *hur,
 860                              const struct lmv_tgt_desc *tgt_mds)
 861 {
 862         __u32                    i;
 863         int                      nr = 0;
 864         struct lmv_tgt_desc     *curr_tgt;
 865
 866         /* count how many requests must be sent to the given target */
 867         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 868                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
 869                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 870                         nr++;
 871         }
 872         return nr;
 873 }
 874
 875 static void lmv_hsm_req_build(struct lmv_obd *lmv,
 876                               struct hsm_user_request *hur_in,
 877                               const struct lmv_tgt_desc *tgt_mds,
 878                               struct hsm_user_request *hur_out)
 879 {
 880         __u32                    i, nr_out;
 881         struct lmv_tgt_desc     *curr_tgt;
 882
 883         /* build the hsm_user_request for the given target */
 884         hur_out->hur_request = hur_in->hur_request;
 885         nr_out = 0;
 886         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 887                 curr_tgt = lmv_find_target(lmv,
 888                                            &hur_in->hur_user_item[i].hui_fid);
 889                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 890                         hur_out->hur_user_item[nr_out] =
 891                                                 hur_in->hur_user_item[i];
 892                         nr_out++;
 893                 }
 894         }
 895         hur_out->hur_request.hr_itemcount = nr_out;
 896         memcpy(hur_data(hur_out), hur_data(hur_in),
 897                hur_in->hur_request.hr_data_len);
 898 }
 899
 900 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
 901                                  struct lustre_kernelcomm *lk, void *uarg)
 902 {
 903         __u32                    i;
 904         int                      rc;
 905         struct kkuc_ct_data     *kcd = NULL;
 906         ENTRY;
 907
 908         /* unregister request (call from llapi_hsm_copytool_fini) */
 909         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 910                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 911
 912                 if (tgt == NULL || tgt->ltd_exp == NULL)
 913                         continue;
 914                 /* best effort: try to clean as much as possible
 915                  * (continue on error) */
 916                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 917         }
 918
 919         /* Whatever the result, remove copytool from kuc groups.
 920          * Unreached coordinators will get EPIPE on next requests
 921          * and will unregister automatically.
 922          */
 923         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
 924         if (kcd != NULL)
 925                 OBD_FREE_PTR(kcd);
 926
 927         RETURN(rc);
 928 }
 929
 930 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
 931                                struct lustre_kernelcomm *lk, void *uarg)
 932 {
 933         struct file             *filp;
 934         __u32                    i, j;
 935         int                      err, rc;
 936         bool                     any_set = false;
 937         struct kkuc_ct_data     *kcd;
 938         ENTRY;
 939
 940         /* All or nothing: try to register to all MDS.
 941          * In case of failure, unregister from previous MDS,
 942          * except if it because of inactive target. */
 943         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 944                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 945
 946                 if (tgt == NULL || tgt->ltd_exp == NULL)
 947                         continue;
 948                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 949                 if (err) {
 950                         if (tgt->ltd_active) {
 951                                 /* permanent error */
 952                                 CERROR("%s: iocontrol MDC %s on MDT"
 953                                        " idx %d cmd %x: err = %d\n",
 954                                        class_exp2obd(lmv->exp)->obd_name,
 955                                        tgt->ltd_uuid.uuid, i, cmd, err);
 956                                 rc = err;
 957                                 lk->lk_flags |= LK_FLG_STOP;
 958                                 /* unregister from previous MDS */
 959                                 for (j = 0; j < i; j++) {
 960                                         tgt = lmv->tgts[j];
 961                                         if (tgt == NULL || tgt->ltd_exp == NULL)
 962                                                 continue;
 963                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
 964                                                       lk, uarg);
 965                                 }
 966                                 RETURN(rc);
 967                         }
 968                         /* else: transient error.
 969                          * kuc will register to the missing MDT
 970                          * when it is back */
 971                 } else {
 972                         any_set = true;
 973                 }
 974         }
 975
 976         if (!any_set)
 977                 /* no registration done: return error */
 978                 RETURN(-ENOTCONN);
 979
 980         /* at least one registration done, with no failure */
 981         filp = fget(lk->lk_wfd);
 982         if (filp == NULL)
 983                 RETURN(-EBADF);
 984
 985         OBD_ALLOC_PTR(kcd);
 986         if (kcd == NULL) {
 987                 fput(filp);
 988                 RETURN(-ENOMEM);
 989         }
 990         kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
 991         kcd->kcd_uuid = lmv->cluuid;
 992         kcd->kcd_archive = lk->lk_data;
 993
 994         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
 995         if (rc != 0) {
 996                 if (filp != NULL)
 997                         fput(filp);
 998                 OBD_FREE_PTR(kcd);
 999         }
1000
1001         RETURN(rc);
1002 }
1003
1004
1005
1006
1007 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
1008                          int len, void *karg, void *uarg)
1009 {
1010         struct obd_device       *obddev = class_exp2obd(exp);
1011         struct lmv_obd          *lmv = &obddev->u.lmv;
1012         struct lmv_tgt_desc     *tgt = NULL;
1013         __u32                    i = 0;
1014         int                      rc = 0;
1015         int                      set = 0;
1016         __u32                    count = lmv->desc.ld_tgt_count;
1017         ENTRY;
1018
1019         if (count == 0)
1020                 RETURN(-ENOTTY);
1021
1022         switch (cmd) {
1023         case IOC_OBD_STATFS: {
1024                 struct obd_ioctl_data *data = karg;
1025                 struct obd_device *mdc_obd;
1026                 struct obd_statfs stat_buf = {0};
1027                 __u32 index;
1028
1029                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1030                 if ((index >= count))
1031                         RETURN(-ENODEV);
1032
1033                 tgt = lmv->tgts[index];
1034                 if (tgt == NULL || !tgt->ltd_active)
1035                         RETURN(-ENODATA);
1036
1037                 mdc_obd = class_exp2obd(tgt->ltd_exp);
1038                 if (!mdc_obd)
1039                         RETURN(-EINVAL);
1040
1041                 /* copy UUID */
1042                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
1043                                  min((int) data->ioc_plen2,
1044                                      (int) sizeof(struct obd_uuid))))
1045                         RETURN(-EFAULT);
1046
1047                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
1048                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1049                                 0);
1050                 if (rc)
1051                         RETURN(rc);
1052                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1053                                  min((int) data->ioc_plen1,
1054                                      (int) sizeof(stat_buf))))
1055                         RETURN(-EFAULT);
1056                 break;
1057         }
1058         case OBD_IOC_QUOTACTL: {
1059                 struct if_quotactl *qctl = karg;
1060                 struct obd_quotactl *oqctl;
1061
1062                 if (qctl->qc_valid == QC_MDTIDX) {
1063                         if (count <= qctl->qc_idx)
1064                                 RETURN(-EINVAL);
1065
1066                         tgt = lmv->tgts[qctl->qc_idx];
1067                         if (tgt == NULL || tgt->ltd_exp == NULL)
1068                                 RETURN(-EINVAL);
1069                 } else if (qctl->qc_valid == QC_UUID) {
1070                         for (i = 0; i < count; i++) {
1071                                 tgt = lmv->tgts[i];
1072                                 if (tgt == NULL)
1073                                         continue;
1074                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
1075                                                      &qctl->obd_uuid))
1076                                         continue;
1077
1078                                 if (tgt->ltd_exp == NULL)
1079                                         RETURN(-EINVAL);
1080
1081                                 break;
1082                         }
1083                 } else {
1084                         RETURN(-EINVAL);
1085                 }
1086
1087                 if (i >= count)
1088                         RETURN(-EAGAIN);
1089
1090                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1091                 OBD_ALLOC_PTR(oqctl);
1092                 if (!oqctl)
1093                         RETURN(-ENOMEM);
1094
1095                 QCTL_COPY(oqctl, qctl);
1096                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1097                 if (rc == 0) {
1098                         QCTL_COPY(qctl, oqctl);
1099                         qctl->qc_valid = QC_MDTIDX;
1100                         qctl->obd_uuid = tgt->ltd_uuid;
1101                 }
1102                 OBD_FREE_PTR(oqctl);
1103                 break;
1104         }
1105         case OBD_IOC_CHANGELOG_SEND:
1106         case OBD_IOC_CHANGELOG_CLEAR: {
1107                 struct ioc_changelog *icc = karg;
1108
1109                 if (icc->icc_mdtindex >= count)
1110                         RETURN(-ENODEV);
1111
1112                 tgt = lmv->tgts[icc->icc_mdtindex];
1113                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1114                         RETURN(-ENODEV);
1115                 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1116                 break;
1117         }
1118         case LL_IOC_GET_CONNECT_FLAGS: {
1119                 tgt = lmv->tgts[0];
1120                 if (tgt == NULL || tgt->ltd_exp == NULL)
1121                         RETURN(-ENODATA);
1122                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1123                 break;
1124         }
1125         case OBD_IOC_FID2PATH: {
1126                 rc = lmv_fid2path(exp, len, karg, uarg);
1127                 break;
1128         }
1129         case LL_IOC_HSM_STATE_GET:
1130         case LL_IOC_HSM_STATE_SET:
1131         case LL_IOC_HSM_ACTION: {
1132                 struct md_op_data       *op_data = karg;
1133
1134                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1135                 if (IS_ERR(tgt))
1136                         RETURN(PTR_ERR(tgt));
1137
1138                 if (tgt->ltd_exp == NULL)
1139                         RETURN(-EINVAL);
1140
1141                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1142                 break;
1143         }
1144         case LL_IOC_HSM_PROGRESS: {
1145                 const struct hsm_progress_kernel *hpk = karg;
1146
1147                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1148                 if (IS_ERR(tgt))
1149                         RETURN(PTR_ERR(tgt));
1150                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1151                 break;
1152         }
1153         case LL_IOC_HSM_REQUEST: {
1154                 struct hsm_user_request *hur = karg;
1155                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1156
1157                 if (reqcount == 0)
1158                         RETURN(0);
1159
1160                 /* if the request is about a single fid
1161                  * or if there is a single MDS, no need to split
1162                  * the request. */
1163                 if (reqcount == 1 || count == 1) {
1164                         tgt = lmv_find_target(lmv,
1165                                               &hur->hur_user_item[0].hui_fid);
1166                         if (IS_ERR(tgt))
1167                                 RETURN(PTR_ERR(tgt));
1168                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1169                 } else {
1170                         /* split fid list to their respective MDS */
1171                         for (i = 0; i < count; i++) {
1172                                 unsigned int            nr, reqlen;
1173                                 int                     rc1;
1174                                 struct hsm_user_request *req;
1175
1176                                 tgt = lmv->tgts[i];
1177                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1178                                         continue;
1179
1180                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1181                                 if (nr == 0) /* nothing for this MDS */
1182                                         continue;
1183
1184                                 /* build a request with fids for this MDS */
1185                                 reqlen = offsetof(typeof(*hur),
1186                                                   hur_user_item[nr])
1187                                                 + hur->hur_request.hr_data_len;
1188                                 OBD_ALLOC_LARGE(req, reqlen);
1189                                 if (req == NULL)
1190                                         RETURN(-ENOMEM);
1191
1192                                 lmv_hsm_req_build(lmv, hur, tgt, req);
1193
1194                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1195                                                     req, uarg);
1196                                 if (rc1 != 0 && rc == 0)
1197                                         rc = rc1;
1198                                 OBD_FREE_LARGE(req, reqlen);
1199                         }
1200                 }
1201                 break;
1202         }
1203         case LL_IOC_LOV_SWAP_LAYOUTS: {
1204                 struct md_op_data       *op_data = karg;
1205                 struct lmv_tgt_desc     *tgt1, *tgt2;
1206
1207                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1208                 if (IS_ERR(tgt1))
1209                         RETURN(PTR_ERR(tgt1));
1210
1211                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1212                 if (IS_ERR(tgt2))
1213                         RETURN(PTR_ERR(tgt2));
1214
1215                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1216                         RETURN(-EINVAL);
1217
1218                 /* only files on same MDT can have their layouts swapped */
1219                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1220                         RETURN(-EPERM);
1221
1222                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1223                 break;
1224         }
1225         case LL_IOC_HSM_CT_START: {
1226                 struct lustre_kernelcomm *lk = karg;
1227                 if (lk->lk_flags & LK_FLG_STOP)
1228                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1229                 else
1230                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1231                 break;
1232         }
1233         default:
1234                 for (i = 0; i < count; i++) {
1235                         struct obd_device *mdc_obd;
1236                         int err;
1237
1238                         tgt = lmv->tgts[i];
1239                         if (tgt == NULL || tgt->ltd_exp == NULL)
1240                                 continue;
1241                         /* ll_umount_begin() sets force flag but for lmv, not
1242                          * mdc. Let's pass it through */
1243                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1244                         mdc_obd->obd_force = obddev->obd_force;
1245                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1246                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1247                                 RETURN(err);
1248                         } else if (err) {
1249                                 if (tgt->ltd_active) {
1250                                         CERROR("error: iocontrol MDC %s on MDT"
1251                                                " idx %d cmd %x: err = %d\n",
1252                                                tgt->ltd_uuid.uuid, i, cmd, err);
1253                                         if (!rc)
1254                                                 rc = err;
1255                                 }
1256                         } else
1257                                 set = 1;
1258                 }
1259                 if (!set && !rc)
1260                         rc = -EIO;
1261         }
1262         RETURN(rc);
1263 }
1264
1265 #if 0
1266 static int lmv_all_chars_policy(int count, const char *name,
1267                                 int len)
1268 {
1269         unsigned int c = 0;
1270
1271         while (len > 0)
1272                 c += name[--len];
1273         c = c % count;
1274         return c;
1275 }
1276
1277 static int lmv_nid_policy(struct lmv_obd *lmv)
1278 {
1279         struct obd_import *imp;
1280         __u32              id;
1281
1282         /*
1283          * XXX: To get nid we assume that underlying obd device is mdc.
1284          */
1285         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1286         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1287         return id % lmv->desc.ld_tgt_count;
1288 }
1289
1290 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1291                           placement_policy_t placement)
1292 {
1293         switch (placement) {
1294         case PLACEMENT_CHAR_POLICY:
1295                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1296                                             op_data->op_name,
1297                                             op_data->op_namelen);
1298         case PLACEMENT_NID_POLICY:
1299                 return lmv_nid_policy(lmv);
1300
1301         default:
1302                 break;
1303         }
1304
1305         CERROR("Unsupported placement policy %x\n", placement);
1306         return -EINVAL;
1307 }
1308 #endif
1309
1310 /**
1311  * This is _inode_ placement policy function (not name).
1312  */
1313 static int lmv_placement_policy(struct obd_device *obd,
1314                                 struct md_op_data *op_data,
1315                                 mdsno_t *mds)
1316 {
1317         struct lmv_obd          *lmv = &obd->u.lmv;
1318         ENTRY;
1319
1320         LASSERT(mds != NULL);
1321
1322         if (lmv->desc.ld_tgt_count == 1) {
1323                 *mds = 0;
1324                 RETURN(0);
1325         }
1326
1327         /**
1328          * If stripe_offset is provided during setdirstripe
1329          * (setdirstripe -i xx), xx MDS will be choosen.
1330          */
1331         if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1332                 struct lmv_user_md *lum;
1333
1334                 lum = op_data->op_data;
1335
1336                 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1337                         *mds = le32_to_cpu(lum->lum_stripe_offset);
1338                 } else {
1339                         /* -1 means default, which will be in the same MDT with
1340                          * the stripe */
1341                         *mds = op_data->op_mds;
1342                         lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1343                 }
1344         } else {
1345                 /* Allocate new fid on target according to operation type and
1346                  * parent home mds. */
1347                 *mds = op_data->op_mds;
1348         }
1349
1350         RETURN(0);
1351 }
1352
1353 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
1354                     mdsno_t mds)
1355 {
1356         struct lmv_tgt_desc     *tgt;
1357         int                      rc;
1358         ENTRY;
1359
1360         tgt = lmv_get_target(lmv, mds, NULL);
1361         if (IS_ERR(tgt))
1362                 RETURN(PTR_ERR(tgt));
1363
1364         /*
1365          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1366          * on server that seq in new allocated fid is not yet known.
1367          */
1368         mutex_lock(&tgt->ltd_fid_mutex);
1369
1370         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1371                 GOTO(out, rc = -ENODEV);
1372
1373         /*
1374          * Asking underlying tgt layer to allocate new fid.
1375          */
1376         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1377         if (rc > 0) {
1378                 LASSERT(fid_is_sane(fid));
1379                 rc = 0;
1380         }
1381
1382         EXIT;
1383 out:
1384         mutex_unlock(&tgt->ltd_fid_mutex);
1385         return rc;
1386 }
1387
1388 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1389                   struct lu_fid *fid, struct md_op_data *op_data)
1390 {
1391         struct obd_device     *obd = class_exp2obd(exp);
1392         struct lmv_obd        *lmv = &obd->u.lmv;
1393         mdsno_t                mds = 0;
1394         int                    rc;
1395         ENTRY;
1396
1397         LASSERT(op_data != NULL);
1398         LASSERT(fid != NULL);
1399
1400         rc = lmv_placement_policy(obd, op_data, &mds);
1401         if (rc) {
1402                 CERROR("Can't get target for allocating fid, "
1403                        "rc %d\n", rc);
1404                 RETURN(rc);
1405         }
1406
1407         rc = __lmv_fid_alloc(lmv, fid, mds);
1408         if (rc) {
1409                 CERROR("Can't alloc new fid, rc %d\n", rc);
1410                 RETURN(rc);
1411         }
1412
1413         RETURN(rc);
1414 }
1415
1416 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1417 {
1418         struct lmv_obd  *lmv = &obd->u.lmv;
1419         struct lmv_desc *desc;
1420         int             rc;
1421         ENTRY;
1422
1423         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1424                 CERROR("LMV setup requires a descriptor\n");
1425                 RETURN(-EINVAL);
1426         }
1427
1428         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1429         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1430                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1431                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1432                 RETURN(-EINVAL);
1433         }
1434
1435         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1436         if (lmv->tgts == NULL)
1437                 RETURN(-ENOMEM);
1438         lmv->tgts_size = 32;
1439
1440         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1441         lmv->desc.ld_tgt_count = 0;
1442         lmv->desc.ld_active_tgt_count = 0;
1443         lmv->max_cookiesize = 0;
1444         lmv->max_def_easize = 0;
1445         lmv->max_easize = 0;
1446         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1447
1448         spin_lock_init(&lmv->lmv_lock);
1449         mutex_init(&lmv->init_mutex);
1450
1451 #ifdef LPROCFS
1452         obd->obd_vars = lprocfs_lmv_obd_vars;
1453         lprocfs_seq_obd_setup(obd);
1454         lprocfs_alloc_md_stats(obd, 0);
1455         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1456                                 0444, &lmv_proc_target_fops, obd);
1457         if (rc)
1458                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1459                       obd->obd_name, rc);
1460 #endif
1461         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1462                              LUSTRE_CLI_FLD_HASH_DHT);
1463         if (rc) {
1464                 CERROR("Can't init FLD, err %d\n", rc);
1465                 GOTO(out, rc);
1466         }
1467
1468         RETURN(0);
1469
1470 out:
1471         return rc;
1472 }
1473
1474 static int lmv_cleanup(struct obd_device *obd)
1475 {
1476         struct lmv_obd   *lmv = &obd->u.lmv;
1477         ENTRY;
1478
1479         fld_client_fini(&lmv->lmv_fld);
1480         if (lmv->tgts != NULL) {
1481                 int i;
1482                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1483                         if (lmv->tgts[i] == NULL)
1484                                 continue;
1485                         lmv_del_target(lmv, i);
1486                 }
1487                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1488                 lmv->tgts_size = 0;
1489         }
1490         RETURN(0);
1491 }
1492
1493 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1494 {
1495         struct lustre_cfg       *lcfg = buf;
1496         struct obd_uuid         obd_uuid;
1497         int                     gen;
1498         __u32                   index;
1499         int                     rc;
1500         ENTRY;
1501
1502         switch (lcfg->lcfg_command) {
1503         case LCFG_ADD_MDC:
1504                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1505                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1506                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1507                         GOTO(out, rc = -EINVAL);
1508
1509                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1510
1511                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1512                         GOTO(out, rc = -EINVAL);
1513                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1514                         GOTO(out, rc = -EINVAL);
1515                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1516                 GOTO(out, rc);
1517         default:
1518                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1519                 GOTO(out, rc = -EINVAL);
1520         }
1521 out:
1522         RETURN(rc);
1523 }
1524
1525 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1526                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1527 {
1528         struct obd_device       *obd = class_exp2obd(exp);
1529         struct lmv_obd          *lmv = &obd->u.lmv;
1530         struct obd_statfs       *temp;
1531         int                      rc = 0;
1532         __u32                    i;
1533         ENTRY;
1534
1535         rc = lmv_check_connect(obd);
1536         if (rc)
1537                 RETURN(rc);
1538
1539         OBD_ALLOC(temp, sizeof(*temp));
1540         if (temp == NULL)
1541                 RETURN(-ENOMEM);
1542
1543         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1544                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1545                         continue;
1546
1547                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1548                                 max_age, flags);
1549                 if (rc) {
1550                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1551                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1552                                rc);
1553                         GOTO(out_free_temp, rc);
1554                 }
1555
1556                 if (i == 0) {
1557                         *osfs = *temp;
1558                         /* If the statfs is from mount, it will needs
1559                          * retrieve necessary information from MDT0.
1560                          * i.e. mount does not need the merged osfs
1561                          * from all of MDT.
1562                          * And also clients can be mounted as long as
1563                          * MDT0 is in service*/
1564                         if (flags & OBD_STATFS_FOR_MDT0)
1565                                 GOTO(out_free_temp, rc);
1566                 } else {
1567                         osfs->os_bavail += temp->os_bavail;
1568                         osfs->os_blocks += temp->os_blocks;
1569                         osfs->os_ffree += temp->os_ffree;
1570                         osfs->os_files += temp->os_files;
1571                 }
1572         }
1573
1574         EXIT;
1575 out_free_temp:
1576         OBD_FREE(temp, sizeof(*temp));
1577         return rc;
1578 }
1579
1580 static int lmv_getstatus(struct obd_export *exp,
1581                          struct lu_fid *fid,
1582                          struct obd_capa **pc)
1583 {
1584         struct obd_device    *obd = exp->exp_obd;
1585         struct lmv_obd       *lmv = &obd->u.lmv;
1586         int                   rc;
1587         ENTRY;
1588
1589         rc = lmv_check_connect(obd);
1590         if (rc)
1591                 RETURN(rc);
1592
1593         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1594         RETURN(rc);
1595 }
1596
1597 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1598                         struct obd_capa *oc, obd_valid valid, const char *name,
1599                         const char *input, int input_size, int output_size,
1600                         int flags, struct ptlrpc_request **request)
1601 {
1602         struct obd_device      *obd = exp->exp_obd;
1603         struct lmv_obd         *lmv = &obd->u.lmv;
1604         struct lmv_tgt_desc    *tgt;
1605         int                     rc;
1606         ENTRY;
1607
1608         rc = lmv_check_connect(obd);
1609         if (rc)
1610                 RETURN(rc);
1611
1612         tgt = lmv_find_target(lmv, fid);
1613         if (IS_ERR(tgt))
1614                 RETURN(PTR_ERR(tgt));
1615
1616         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1617                          input_size, output_size, flags, request);
1618
1619         RETURN(rc);
1620 }
1621
1622 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1623                         struct obd_capa *oc, obd_valid valid, const char *name,
1624                         const char *input, int input_size, int output_size,
1625                         int flags, __u32 suppgid,
1626                         struct ptlrpc_request **request)
1627 {
1628         struct obd_device      *obd = exp->exp_obd;
1629         struct lmv_obd         *lmv = &obd->u.lmv;
1630         struct lmv_tgt_desc    *tgt;
1631         int                     rc;
1632         ENTRY;
1633
1634         rc = lmv_check_connect(obd);
1635         if (rc)
1636                 RETURN(rc);
1637
1638         tgt = lmv_find_target(lmv, fid);
1639         if (IS_ERR(tgt))
1640                 RETURN(PTR_ERR(tgt));
1641
1642         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1643                          input_size, output_size, flags, suppgid,
1644                          request);
1645
1646         RETURN(rc);
1647 }
1648
1649 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1650                        struct ptlrpc_request **request)
1651 {
1652         struct obd_device       *obd = exp->exp_obd;
1653         struct lmv_obd          *lmv = &obd->u.lmv;
1654         struct lmv_tgt_desc     *tgt;
1655         int                      rc;
1656         ENTRY;
1657
1658         rc = lmv_check_connect(obd);
1659         if (rc)
1660                 RETURN(rc);
1661
1662         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1663         if (IS_ERR(tgt))
1664                 RETURN(PTR_ERR(tgt));
1665
1666         if (op_data->op_flags & MF_GET_MDT_IDX) {
1667                 op_data->op_mds = tgt->ltd_idx;
1668                 RETURN(0);
1669         }
1670
1671         rc = md_getattr(tgt->ltd_exp, op_data, request);
1672
1673         RETURN(rc);
1674 }
1675
1676 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1677 {
1678         struct obd_device   *obd = exp->exp_obd;
1679         struct lmv_obd      *lmv = &obd->u.lmv;
1680         __u32                i;
1681         int                  rc;
1682         ENTRY;
1683
1684         rc = lmv_check_connect(obd);
1685         if (rc)
1686                 RETURN(rc);
1687
1688         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1689
1690         /*
1691          * With DNE every object can have two locks in different namespaces:
1692          * lookup lock in space of MDT storing direntry and update/open lock in
1693          * space of MDT storing inode.
1694          */
1695         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1696                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1697                         continue;
1698                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1699         }
1700
1701         RETURN(0);
1702 }
1703
1704 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1705                            ldlm_iterator_t it, void *data)
1706 {
1707         struct obd_device       *obd = exp->exp_obd;
1708         struct lmv_obd          *lmv = &obd->u.lmv;
1709         int                     i;
1710         int                     tgt;
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.  Try the MDT that the FID maps to first,
1724          * since this can be easily found, and only try others if that fails.
1725          */
1726         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1727              i < lmv->desc.ld_tgt_count;
1728              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1729                 if (tgt < 0) {
1730                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1731                                obd->obd_name, PFID(fid), tgt);
1732                         tgt = 0;
1733                 }
1734
1735                 if (lmv->tgts[tgt] == NULL ||
1736                     lmv->tgts[tgt]->ltd_exp == NULL)
1737                         continue;
1738
1739                 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1740                 if (rc)
1741                         RETURN(rc);
1742         }
1743
1744         RETURN(rc);
1745 }
1746
1747
1748 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1749                      struct md_open_data *mod, struct ptlrpc_request **request)
1750 {
1751         struct obd_device     *obd = exp->exp_obd;
1752         struct lmv_obd        *lmv = &obd->u.lmv;
1753         struct lmv_tgt_desc   *tgt;
1754         int                    rc;
1755         ENTRY;
1756
1757         rc = lmv_check_connect(obd);
1758         if (rc)
1759                 RETURN(rc);
1760
1761         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1762         if (IS_ERR(tgt))
1763                 RETURN(PTR_ERR(tgt));
1764
1765         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1766         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1767         RETURN(rc);
1768 }
1769
1770 /**
1771  * Choosing the MDT by name or FID in @op_data.
1772  * For non-striped directory, it will locate MDT by fid.
1773  * For striped-directory, it will locate MDT by name. And also
1774  * it will reset op_fid1 with the FID of the choosen stripe.
1775  **/
1776 struct lmv_tgt_desc *
1777 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1778                            const char *name, int namelen, struct lu_fid *fid,
1779                            mdsno_t *mds)
1780 {
1781         struct lmv_tgt_desc     *tgt;
1782         const struct lmv_oinfo  *oinfo;
1783
1784         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1785         if (IS_ERR(oinfo))
1786                 RETURN((void *)oinfo);
1787         *fid = oinfo->lmo_fid;
1788         *mds = oinfo->lmo_mds;
1789         tgt = lmv_get_target(lmv, *mds, NULL);
1790
1791         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid));
1792         return tgt;
1793 }
1794
1795 struct lmv_tgt_desc
1796 *lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1797                 struct lu_fid *fid)
1798 {
1799         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1800         struct lmv_tgt_desc     *tgt;
1801
1802         if (lsm == NULL || op_data->op_namelen == 0) {
1803                 tgt = lmv_find_target(lmv, fid);
1804                 if (IS_ERR(tgt))
1805                         return tgt;
1806
1807                 op_data->op_mds = tgt->ltd_idx;
1808                 return tgt;
1809         }
1810
1811         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1812                                           op_data->op_namelen, fid,
1813                                           &op_data->op_mds);
1814 }
1815
1816 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1817                const void *data, int datalen, int mode, __u32 uid,
1818                __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1819                struct ptlrpc_request **request)
1820 {
1821         struct obd_device       *obd = exp->exp_obd;
1822         struct lmv_obd          *lmv = &obd->u.lmv;
1823         struct lmv_tgt_desc     *tgt;
1824         int                      rc;
1825         ENTRY;
1826
1827         rc = lmv_check_connect(obd);
1828         if (rc)
1829                 RETURN(rc);
1830
1831         if (!lmv->desc.ld_active_tgt_count)
1832                 RETURN(-EIO);
1833
1834         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1835         if (IS_ERR(tgt))
1836                 RETURN(PTR_ERR(tgt));
1837
1838         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1839                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1840                op_data->op_mds);
1841
1842         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1843         if (rc)
1844                 RETURN(rc);
1845
1846         /* Send the create request to the MDT where the object
1847          * will be located */
1848         tgt = lmv_find_target(lmv, &op_data->op_fid2);
1849         if (IS_ERR(tgt))
1850                 RETURN(PTR_ERR(tgt));
1851
1852         op_data->op_mds = tgt->ltd_idx;
1853
1854         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1855                PFID(&op_data->op_fid2), op_data->op_mds);
1856
1857         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1858         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1859                        cap_effective, rdev, request);
1860         if (rc == 0) {
1861                 if (*request == NULL)
1862                         RETURN(rc);
1863                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1864         }
1865         RETURN(rc);
1866 }
1867
1868 static int lmv_done_writing(struct obd_export *exp,
1869                             struct md_op_data *op_data,
1870                             struct md_open_data *mod)
1871 {
1872         struct obd_device     *obd = exp->exp_obd;
1873         struct lmv_obd        *lmv = &obd->u.lmv;
1874         struct lmv_tgt_desc   *tgt;
1875         int                    rc;
1876         ENTRY;
1877
1878         rc = lmv_check_connect(obd);
1879         if (rc)
1880                 RETURN(rc);
1881
1882         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1883         if (IS_ERR(tgt))
1884                 RETURN(PTR_ERR(tgt));
1885
1886         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1887         RETURN(rc);
1888 }
1889
1890 static int
1891 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1892             const union ldlm_policy_data *policy,
1893             struct lookup_intent *it, struct md_op_data *op_data,
1894             struct lustre_handle *lockh, __u64 extra_lock_flags)
1895 {
1896         struct obd_device        *obd = exp->exp_obd;
1897         struct lmv_obd           *lmv = &obd->u.lmv;
1898         struct lmv_tgt_desc      *tgt;
1899         int                       rc;
1900         ENTRY;
1901
1902         rc = lmv_check_connect(obd);
1903         if (rc)
1904                 RETURN(rc);
1905
1906         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1907                LL_IT2STR(it), PFID(&op_data->op_fid1));
1908
1909         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1910         if (IS_ERR(tgt))
1911                 RETURN(PTR_ERR(tgt));
1912
1913         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1914                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1915
1916         rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1917                         extra_lock_flags);
1918
1919         RETURN(rc);
1920 }
1921
1922 static int
1923 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1924                  struct ptlrpc_request **preq)
1925 {
1926         struct ptlrpc_request   *req = NULL;
1927         struct obd_device       *obd = exp->exp_obd;
1928         struct lmv_obd          *lmv = &obd->u.lmv;
1929         struct lmv_tgt_desc     *tgt;
1930         struct mdt_body         *body;
1931         int                      rc;
1932         ENTRY;
1933
1934         rc = lmv_check_connect(obd);
1935         if (rc)
1936                 RETURN(rc);
1937
1938         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1939         if (IS_ERR(tgt))
1940                 RETURN(PTR_ERR(tgt));
1941
1942         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1943                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1944                tgt->ltd_idx);
1945
1946         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1947         if (rc != 0)
1948                 RETURN(rc);
1949
1950         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1951         LASSERT(body != NULL);
1952
1953         if (body->mbo_valid & OBD_MD_MDS) {
1954                 struct lu_fid rid = body->mbo_fid1;
1955                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1956                        PFID(&rid));
1957
1958                 tgt = lmv_find_target(lmv, &rid);
1959                 if (IS_ERR(tgt)) {
1960                         ptlrpc_req_finished(*preq);
1961                         preq = NULL;
1962                         RETURN(PTR_ERR(tgt));
1963                 }
1964
1965                 op_data->op_fid1 = rid;
1966                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1967                 op_data->op_namelen = 0;
1968                 op_data->op_name = NULL;
1969                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1970                 ptlrpc_req_finished(*preq);
1971                 *preq = req;
1972         }
1973
1974         RETURN(rc);
1975 }
1976
1977 #define md_op_data_fid(op_data, fl)                     \
1978         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1979          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1980          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1981          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1982          NULL)
1983
1984 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1985                             struct md_op_data *op_data,
1986                             int op_tgt, ldlm_mode_t mode, int bits, int flag)
1987 {
1988         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1989         struct obd_device      *obd = exp->exp_obd;
1990         struct lmv_obd         *lmv = &obd->u.lmv;
1991         ldlm_policy_data_t      policy = {{ 0 }};
1992         int                     rc = 0;
1993         ENTRY;
1994
1995         if (!fid_is_sane(fid))
1996                 RETURN(0);
1997
1998         if (tgt == NULL) {
1999                 tgt = lmv_find_target(lmv, fid);
2000                 if (IS_ERR(tgt))
2001                         RETURN(PTR_ERR(tgt));
2002         }
2003
2004         if (tgt->ltd_idx != op_tgt) {
2005                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2006                 policy.l_inodebits.bits = bits;
2007                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2008                                       mode, LCF_ASYNC, NULL);
2009         } else {
2010                 CDEBUG(D_INODE,
2011                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
2012                        op_tgt, PFID(fid));
2013                 op_data->op_flags |= flag;
2014                 rc = 0;
2015         }
2016
2017         RETURN(rc);
2018 }
2019
2020 /*
2021  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2022  * op_data->op_fid2
2023  */
2024 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2025                     struct ptlrpc_request **request)
2026 {
2027         struct obd_device       *obd = exp->exp_obd;
2028         struct lmv_obd          *lmv = &obd->u.lmv;
2029         struct lmv_tgt_desc     *tgt;
2030         int                      rc;
2031         ENTRY;
2032
2033         rc = lmv_check_connect(obd);
2034         if (rc)
2035                 RETURN(rc);
2036
2037         LASSERT(op_data->op_namelen != 0);
2038
2039         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2040                PFID(&op_data->op_fid2), op_data->op_namelen,
2041                op_data->op_name, PFID(&op_data->op_fid1));
2042
2043         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2044         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2045         op_data->op_cap = cfs_curproc_cap_pack();
2046         if (op_data->op_mea2 != NULL) {
2047                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
2048                 const struct lmv_oinfo  *oinfo;
2049
2050                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2051                                                 op_data->op_namelen);
2052                 if (IS_ERR(oinfo))
2053                         RETURN(PTR_ERR(oinfo));
2054
2055                 op_data->op_fid2 = oinfo->lmo_fid;
2056         }
2057
2058         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2059         if (IS_ERR(tgt))
2060                 RETURN(PTR_ERR(tgt));
2061
2062         /*
2063          * Cancel UPDATE lock on child (fid1).
2064          */
2065         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2066         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2067                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2068         if (rc != 0)
2069                 RETURN(rc);
2070
2071         rc = md_link(tgt->ltd_exp, op_data, request);
2072
2073         RETURN(rc);
2074 }
2075
2076 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2077                       const char *old, int oldlen, const char *new, int newlen,
2078                       struct ptlrpc_request **request)
2079 {
2080         struct obd_device       *obd = exp->exp_obd;
2081         struct lmv_obd          *lmv = &obd->u.lmv;
2082         struct lmv_tgt_desc     *src_tgt;
2083         int                     rc;
2084         ENTRY;
2085
2086         LASSERT(oldlen != 0);
2087
2088         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2089                oldlen, old, PFID(&op_data->op_fid1),
2090                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2091                newlen, new, PFID(&op_data->op_fid2),
2092                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2093
2094         rc = lmv_check_connect(obd);
2095         if (rc)
2096                 RETURN(rc);
2097
2098         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2099         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2100         op_data->op_cap = cfs_curproc_cap_pack();
2101         if (op_data->op_cli_flags & CLI_MIGRATE) {
2102                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2103                          PFID(&op_data->op_fid3));
2104                 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2105                 if (rc)
2106                         RETURN(rc);
2107                 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3);
2108         } else {
2109                 if (op_data->op_mea1 != NULL) {
2110                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2111
2112                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2113                                                              oldlen,
2114                                                              &op_data->op_fid1,
2115                                                              &op_data->op_mds);
2116                         if (IS_ERR(src_tgt))
2117                                 RETURN(PTR_ERR(src_tgt));
2118                 } else {
2119                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2120                         if (IS_ERR(src_tgt))
2121                                 RETURN(PTR_ERR(src_tgt));
2122
2123                         op_data->op_mds = src_tgt->ltd_idx;
2124                 }
2125
2126                 if (op_data->op_mea2) {
2127                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
2128                         const struct lmv_oinfo  *oinfo;
2129
2130                         oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2131                         if (IS_ERR(oinfo))
2132                                 RETURN(PTR_ERR(oinfo));
2133
2134                         op_data->op_fid2 = oinfo->lmo_fid;
2135                 }
2136         }
2137         if (IS_ERR(src_tgt))
2138                 RETURN(PTR_ERR(src_tgt));
2139
2140         /*
2141          * LOOKUP lock on src child (fid3) should also be cancelled for
2142          * src_tgt in mdc_rename.
2143          */
2144         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2145
2146         /*
2147          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2148          * own target.
2149          */
2150         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2151                               LCK_EX, MDS_INODELOCK_UPDATE,
2152                               MF_MDC_CANCEL_FID2);
2153
2154         if (rc != 0)
2155                 RETURN(rc);
2156         /*
2157          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2158          */
2159         if (fid_is_sane(&op_data->op_fid3)) {
2160                 struct lmv_tgt_desc *tgt;
2161
2162                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2163                 if (IS_ERR(tgt))
2164                         RETURN(PTR_ERR(tgt));
2165
2166                 /* Cancel LOOKUP lock on its parent */
2167                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2168                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2169                                       MF_MDC_CANCEL_FID3);
2170                 if (rc != 0)
2171                         RETURN(rc);
2172
2173                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2174                                       LCK_EX, MDS_INODELOCK_FULL,
2175                                       MF_MDC_CANCEL_FID3);
2176                 if (rc != 0)
2177                         RETURN(rc);
2178         }
2179
2180         /*
2181          * Cancel all the locks on tgt child (fid4).
2182          */
2183         if (fid_is_sane(&op_data->op_fid4))
2184                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2185                                       LCK_EX, MDS_INODELOCK_FULL,
2186                                       MF_MDC_CANCEL_FID4);
2187
2188         CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2189                op_data->op_mds, PFID(&op_data->op_fid2));
2190
2191         rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2192                        request);
2193
2194         RETURN(rc);
2195 }
2196
2197 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2198                        void *ea, int ealen, void *ea2, int ea2len,
2199                        struct ptlrpc_request **request,
2200                        struct md_open_data **mod)
2201 {
2202         struct obd_device       *obd = exp->exp_obd;
2203         struct lmv_obd          *lmv = &obd->u.lmv;
2204         struct lmv_tgt_desc     *tgt;
2205         int                      rc = 0;
2206         ENTRY;
2207
2208         rc = lmv_check_connect(obd);
2209         if (rc)
2210                 RETURN(rc);
2211
2212         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2213                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2214
2215         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2216         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2217         if (IS_ERR(tgt))
2218                 RETURN(PTR_ERR(tgt));
2219
2220         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2221                         ea2len, request, mod);
2222
2223         RETURN(rc);
2224 }
2225
2226 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2227                      struct obd_capa *oc, struct ptlrpc_request **request)
2228 {
2229         struct obd_device       *obd = exp->exp_obd;
2230         struct lmv_obd          *lmv = &obd->u.lmv;
2231         struct lmv_tgt_desc     *tgt;
2232         int                      rc;
2233         ENTRY;
2234
2235         rc = lmv_check_connect(obd);
2236         if (rc != 0)
2237                 RETURN(rc);
2238
2239         tgt = lmv_find_target(lmv, fid);
2240         if (IS_ERR(tgt))
2241                 RETURN(PTR_ERR(tgt));
2242
2243         rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2244         RETURN(rc);
2245 }
2246
2247 /*
2248  * Adjust a set of pages, each page containing an array of lu_dirpages,
2249  * so that each page can be used as a single logical lu_dirpage.
2250  *
2251  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
2252  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
2253  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
2254  * value is used as a cookie to request the next lu_dirpage in a
2255  * directory listing that spans multiple pages (two in this example):
2256  *   ________
2257  *  |        |
2258  * .|--------v-------   -----.
2259  * |s|e|f|p|ent|ent| ... |ent|
2260  * '--|--------------   -----'   Each CFS_PAGE contains a single
2261  *    '------.                   lu_dirpage.
2262  * .---------v-------   -----.
2263  * |s|e|f|p|ent| 0 | ... | 0 |
2264  * '-----------------   -----'
2265  *
2266  * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
2267  * larger than LU_PAGE_SIZE, a single host page may contain multiple
2268  * lu_dirpages. After reading the lu_dirpages from the MDS, the
2269  * ldp_hash_end of the first lu_dirpage refers to the one immediately
2270  * after it in the same CFS_PAGE (arrows simplified for brevity, but
2271  * in general e0==s1, e1==s2, etc.):
2272  *
2273  * .--------------------   -----.
2274  * |s0|e0|f0|p|ent|ent| ... |ent|
2275  * |---v----------------   -----|
2276  * |s1|e1|f1|p|ent|ent| ... |ent|
2277  * |---v----------------   -----|  Here, each CFS_PAGE contains
2278  *             ...                 multiple lu_dirpages.
2279  * |---v----------------   -----|
2280  * |s'|e'|f'|p|ent|ent| ... |ent|
2281  * '---|----------------   -----'
2282  *     v
2283  * .----------------------------.
2284  * |        next CFS_PAGE       |
2285  *
2286  * This structure is transformed into a single logical lu_dirpage as follows:
2287  *
2288  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
2289  *   labeled 'next CFS_PAGE'.
2290  *
2291  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
2292  *   a hash collision with the next page exists.
2293  *
2294  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
2295  *   to the first entry of the next lu_dirpage.
2296  */
2297 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
2298 static void lmv_adjust_dirpages(struct page **pages, int ncfspgs, int nlupgs)
2299 {
2300         int i;
2301
2302         for (i = 0; i < ncfspgs; i++) {
2303                 struct lu_dirpage       *dp = kmap(pages[i]);
2304                 struct lu_dirpage       *first = dp;
2305                 struct lu_dirent        *end_dirent = NULL;
2306                 struct lu_dirent        *ent;
2307                 __u64                   hash_end = dp->ldp_hash_end;
2308                 __u32                   flags = dp->ldp_flags;
2309
2310                 while (--nlupgs > 0) {
2311                         ent = lu_dirent_start(dp);
2312                         for (end_dirent = ent; ent != NULL;
2313                              end_dirent = ent, ent = lu_dirent_next(ent));
2314
2315                         /* Advance dp to next lu_dirpage. */
2316                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
2317
2318                         /* Check if we've reached the end of the CFS_PAGE. */
2319                         if (!((unsigned long)dp & ~CFS_PAGE_MASK))
2320                                 break;
2321
2322                         /* Save the hash and flags of this lu_dirpage. */
2323                         hash_end = dp->ldp_hash_end;
2324                         flags = dp->ldp_flags;
2325
2326                         /* Check if lu_dirpage contains no entries. */
2327                         if (!end_dirent)
2328                                 break;
2329
2330                         /* Enlarge the end entry lde_reclen from 0 to
2331                          * first entry of next lu_dirpage. */
2332                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
2333                         end_dirent->lde_reclen =
2334                                 cpu_to_le16((char *)(dp->ldp_entries) -
2335                                             (char *)end_dirent);
2336                 }
2337
2338                 first->ldp_hash_end = hash_end;
2339                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
2340                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
2341
2342                 kunmap(pages[i]);
2343         }
2344         LASSERTF(nlupgs == 0, "left = %d", nlupgs);
2345 }
2346 #else
2347 #define lmv_adjust_dirpages(pages, ncfspgs, nlupgs) do {} while (0)
2348 #endif  /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
2349
2350 /**
2351  * This function will read entry from a striped directory, bascially, it will
2352  * read entries from all of stripes, and choose one closest to the required
2353  * offset(&op_data->op_hash_offset). A few notes
2354  * 1. skip . and .. for non-zero stripes, because there can only have one .
2355  * and .. in a directory.
2356  * 2. op_data will be shared by all of stripes, instead of allocating new
2357  * one, so need to restore before reusing.
2358  * 3. release the entry page if that is not being chosen.
2359  *
2360  * param[in]exp         obd export refer to LMV
2361  * param[in]op_data     hold those MD parameters of read_entry.
2362  * param[in]cb_op       ldlm callback being used in enqueue in mdc_read_entry
2363  * param[out]ldp        the entry being read.
2364  * param[out]ppage      the page holding the entry, note: because the entry
2365  *                      will be accessed in upper layer, so we need hold the
2366  *                      page until the usages of entry is finished, see
2367  *                      ll_dir_entry_next.
2368  *
2369  * retval               =0 if get entry successfully
2370  *                      <0 can not get entry.
2371  */
2372 #define NORMAL_MAX_STRIPES 4
2373 static int lmv_read_striped_entry(struct obd_export *exp,
2374                                   struct md_op_data *op_data,
2375                                   struct md_callback *cb_op,
2376                                   struct lu_dirent **ldp,
2377                                   struct page **ppage)
2378 {
2379         struct obd_device       *obd = exp->exp_obd;
2380         struct lmv_obd          *lmv = &obd->u.lmv;
2381         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2382         struct lmv_tgt_desc     *tgt;
2383         struct lu_dirent        *tmp_ents[NORMAL_MAX_STRIPES];
2384         struct lu_dirent        **ents = NULL;
2385         struct lu_fid           master_fid = op_data->op_fid1;
2386         void                    *master_data = op_data->op_data;
2387         __u64                   last_idx = op_data->op_stripe_offset;
2388         __u64                   hash_offset = op_data->op_hash_offset;
2389         __u32                   same_hash_offset = op_data->op_same_hash_offset;
2390         __u32                   cli_flags = op_data->op_cli_flags;
2391         int                     stripe_count;
2392         __u64                   min_hash;
2393         int                     min_same_hash_offset = 0;
2394         int                     min_idx = 0;
2395         struct page             *min_page = NULL;
2396         int                     i;
2397         int                     rc;
2398         ENTRY;
2399
2400         LASSERT(lsm != NULL);
2401
2402         rc = lmv_check_connect(obd);
2403         if (rc)
2404                 RETURN(rc);
2405
2406         /* . and .. will be stored on the master object, so we need iterate
2407          * the master object as well */
2408         stripe_count = lsm->lsm_md_stripe_count;
2409         if (stripe_count > NORMAL_MAX_STRIPES) {
2410                 OBD_ALLOC(ents, sizeof(ents[0]) * stripe_count);
2411                 if (ents == NULL)
2412                         GOTO(out, rc = -ENOMEM);
2413         } else {
2414                 ents = tmp_ents;
2415                 memset(ents, 0, sizeof(ents[0]) * stripe_count);
2416         }
2417
2418         min_hash = MDS_DIR_END_OFF;
2419         for (i = 0; i < stripe_count; i++) {
2420                 struct page *page = NULL;
2421
2422                 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2423                 if (IS_ERR(tgt))
2424                         GOTO(out, rc = PTR_ERR(tgt));
2425
2426                 if (last_idx != i)
2427                         op_data->op_same_hash_offset = 0;
2428                 else
2429                         op_data->op_same_hash_offset = same_hash_offset;
2430
2431                 /* op_data will be shared by each stripe, so we need
2432                  * reset these value for each stripe */
2433                 op_data->op_stripe_offset = i;
2434                 op_data->op_hash_offset = hash_offset;
2435                 op_data->op_cli_flags = cli_flags;
2436                 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2437                 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2438                 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2439
2440 next:
2441                 rc = md_read_entry(tgt->ltd_exp, op_data, cb_op, &ents[i],
2442                                    &page);
2443                 if (rc != 0)
2444                         GOTO(out, rc);
2445
2446                 if (ents[i] != NULL &&
2447                     (strncmp(ents[i]->lde_name, ".",
2448                              le16_to_cpu(ents[i]->lde_namelen)) == 0 ||
2449                     strncmp(ents[i]->lde_name, "..",
2450                             le16_to_cpu(ents[i]->lde_namelen)) == 0)) {
2451                         if (i == 0) {
2452                                 /* replace . with master FID */
2453                                 if (le16_to_cpu(ents[i]->lde_namelen) == 1)
2454                                         fid_cpu_to_le(&ents[i]->lde_fid,
2455                                                       &master_fid);
2456                                 else
2457                                         fid_cpu_to_le(&ents[i]->lde_fid,
2458                                                       &op_data->op_fid3);
2459                         } else {
2460                                 /* skip . and .. for other stripes */
2461                                 op_data->op_cli_flags |= CLI_NEXT_ENTRY;
2462                                 op_data->op_hash_offset =
2463                                         le64_to_cpu(ents[i]->lde_hash);
2464                                 kunmap(page);
2465                                 page_cache_release(page);
2466                                 goto next;
2467                         }
2468                 }
2469
2470                 if (ents[i] != NULL) {
2471                         /* If the hash value of read_entry is equal to the
2472                          * current min_hash, which is very rare and only
2473                          * happens if two entries have the same hash value
2474                          * but on different stripes, in this case, we need
2475                          * make sure these entries are being reading forward,
2476                          * not backward, i.e. only reset the min_entry, if
2477                          * current stripe is ahead of last entry. Note: if
2478                          * there are hash conflict inside the entry, MDC
2479                          * (see mdc_read_entry) will resolve them. */
2480                         if (le64_to_cpu(ents[i]->lde_hash) < min_hash ||
2481                             (le64_to_cpu(ents[i]->lde_hash) == min_hash &&
2482                              i >= last_idx)) {
2483                                 if (min_page != NULL) {
2484                                         kunmap(min_page);
2485                                         page_cache_release(min_page);
2486                                 }
2487                                 min_page = page;
2488                                 min_hash = le64_to_cpu(ents[i]->lde_hash);
2489                                 min_same_hash_offset =
2490                                            op_data->op_same_hash_offset;
2491                                 min_idx = i;
2492                         } else {
2493                                 kunmap(page);
2494                                 page_cache_release(page);
2495                         }
2496                 }
2497         }
2498
2499         if (min_hash != MDS_DIR_END_OFF) {
2500                 *ldp = ents[min_idx];
2501                 op_data->op_stripe_offset = min_idx;
2502                 op_data->op_same_hash_offset = min_same_hash_offset;
2503                 *ppage = min_page;
2504         } else {
2505                 *ldp = NULL;
2506                 *ppage = NULL;
2507         }
2508 out:
2509         /* We do not want to allocate md_op_data during each
2510          * dir entry reading, so op_data will be shared by every stripe,
2511          * then we need to restore it back to original value before
2512          * return to the upper layer */
2513         op_data->op_hash_offset = hash_offset;
2514         op_data->op_fid1 = master_fid;
2515         op_data->op_fid2 = master_fid;
2516         op_data->op_data = master_data;
2517         op_data->op_cli_flags = cli_flags;
2518         if (stripe_count > NORMAL_MAX_STRIPES && ents != NULL)
2519                 OBD_FREE(ents, sizeof(ents[0]) * stripe_count);
2520
2521         if (rc != 0 && min_page != NULL) {
2522                 kunmap(min_page);
2523                 page_cache_release(min_page);
2524         }
2525
2526         RETURN(rc);
2527 }
2528
2529 int lmv_read_entry(struct obd_export *exp, struct md_op_data *op_data,
2530                    struct md_callback *cb_op, struct lu_dirent **ldp,
2531                    struct page **ppage)
2532 {
2533         struct obd_device       *obd = exp->exp_obd;
2534         struct lmv_obd          *lmv = &obd->u.lmv;
2535         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2536         struct lmv_tgt_desc     *tgt;
2537         int                     rc;
2538         ENTRY;
2539
2540         rc = lmv_check_connect(obd);
2541         if (rc != 0)
2542                 RETURN(rc);
2543
2544         if (unlikely(lsm != NULL)) {
2545                 rc = lmv_read_striped_entry(exp, op_data, cb_op,
2546                                             ldp, ppage);
2547                 RETURN(rc);
2548         }
2549
2550         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2551         if (IS_ERR(tgt))
2552                 RETURN(PTR_ERR(tgt));
2553
2554         rc = md_read_entry(tgt->ltd_exp, op_data, cb_op, ldp,
2555                            ppage);
2556         RETURN(rc);
2557 }
2558
2559 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2560                       struct ptlrpc_request **request)
2561 {
2562         struct obd_device       *obd = exp->exp_obd;
2563         struct lmv_obd          *lmv = &obd->u.lmv;
2564         struct lmv_tgt_desc     *tgt = NULL;
2565         struct lmv_tgt_desc     *parent_tgt = NULL;
2566         struct mdt_body         *body;
2567         int                     rc;
2568         ENTRY;
2569
2570         rc = lmv_check_connect(obd);
2571         if (rc)
2572                 RETURN(rc);
2573 retry:
2574         /* Send unlink requests to the MDT where the child is located */
2575         if (likely(!fid_is_zero(&op_data->op_fid2))) {
2576                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2577                 if (IS_ERR(tgt))
2578                         RETURN(PTR_ERR(tgt));
2579
2580                 /* For striped dir, we need to locate the parent as well */
2581                 if (op_data->op_mea1 != NULL) {
2582                         struct lmv_tgt_desc *tmp;
2583
2584                         LASSERT(op_data->op_name != NULL &&
2585                                 op_data->op_namelen != 0);
2586                         tmp = lmv_locate_target_for_name(lmv,
2587                                                    op_data->op_mea1,
2588                                                    op_data->op_name,
2589                                                    op_data->op_namelen,
2590                                                    &op_data->op_fid1,
2591                                                    &op_data->op_mds);
2592                         if (IS_ERR(tmp))
2593                                 RETURN(PTR_ERR(tmp));
2594                 }
2595         } else {
2596                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2597                 if (IS_ERR(tgt))
2598                         RETURN(PTR_ERR(tgt));
2599         }
2600
2601         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2602         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2603         op_data->op_cap = cfs_curproc_cap_pack();
2604
2605         /*
2606          * If child's fid is given, cancel unused locks for it if it is from
2607          * another export than parent.
2608          *
2609          * LOOKUP lock for child (fid3) should also be cancelled on parent
2610          * tgt_tgt in mdc_unlink().
2611          */
2612         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2613
2614         /*
2615          * Cancel FULL locks on child (fid3).
2616          */
2617         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2618         if (IS_ERR(parent_tgt))
2619                 RETURN(PTR_ERR(parent_tgt));
2620
2621         if (parent_tgt != tgt) {
2622                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2623                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2624                                       MF_MDC_CANCEL_FID3);
2625         }
2626
2627         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2628                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2629         if (rc != 0)
2630                 RETURN(rc);
2631
2632         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2633                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2634
2635         rc = md_unlink(tgt->ltd_exp, op_data, request);
2636         if (rc != 0 && rc != -EREMOTE)
2637                 RETURN(rc);
2638
2639         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2640         if (body == NULL)
2641                 RETURN(-EPROTO);
2642
2643         /* Not cross-ref case, just get out of here. */
2644         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2645                 RETURN(0);
2646
2647         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2648                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2649
2650         /* This is a remote object, try remote MDT, Note: it may
2651          * try more than 1 time here, Considering following case
2652          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2653          * 1. Initially A does not know where remote1 is, it send
2654          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2655          *    resend unlink RPC to MDT1 (retry 1st time).
2656          *
2657          * 2. During the unlink RPC in flight,
2658          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2659          *    and create new remote1, but on MDT0
2660          *
2661          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2662          *    /mnt/lustre, then lookup get fid of remote1, and find
2663          *    it is remote dir again, and replay -EREMOTE again.
2664          *
2665          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2666          *
2667          * In theory, it might try unlimited time here, but it should
2668          * be very rare case.  */
2669         op_data->op_fid2 = body->mbo_fid1;
2670         ptlrpc_req_finished(*request);
2671         *request = NULL;
2672
2673         goto retry;
2674 }
2675
2676 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2677 {
2678         struct lmv_obd *lmv = &obd->u.lmv;
2679         int rc = 0;
2680
2681         switch (stage) {
2682         case OBD_CLEANUP_EARLY:
2683                 /* XXX: here should be calling obd_precleanup() down to
2684                  * stack. */
2685                 break;
2686         case OBD_CLEANUP_EXPORTS:
2687                 fld_client_proc_fini(&lmv->lmv_fld);
2688                 lprocfs_obd_cleanup(obd);
2689                 lprocfs_free_md_stats(obd);
2690                 break;
2691         default:
2692                 break;
2693         }
2694         RETURN(rc);
2695 }
2696
2697 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2698                         __u32 keylen, void *key, __u32 *vallen, void *val,
2699                         struct lov_stripe_md *lsm)
2700 {
2701         struct obd_device       *obd;
2702         struct lmv_obd          *lmv;
2703         int                      rc = 0;
2704         ENTRY;
2705
2706         obd = class_exp2obd(exp);
2707         if (obd == NULL) {
2708                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2709                        exp->exp_handle.h_cookie);
2710                 RETURN(-EINVAL);
2711         }
2712
2713         lmv = &obd->u.lmv;
2714         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2715                 int i;
2716
2717                 rc = lmv_check_connect(obd);
2718                 if (rc)
2719                         RETURN(rc);
2720
2721                 LASSERT(*vallen == sizeof(__u32));
2722                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2723                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2724                         /*
2725                          * All tgts should be connected when this gets called.
2726                          */
2727                         if (tgt == NULL || tgt->ltd_exp == NULL)
2728                                 continue;
2729
2730                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2731                                           vallen, val, NULL))
2732                                 RETURN(0);
2733                 }
2734                 RETURN(-EINVAL);
2735         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2736                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2737                    KEY_IS(KEY_MAX_COOKIESIZE) ||
2738                    KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2739                    KEY_IS(KEY_CONN_DATA)) {
2740                 rc = lmv_check_connect(obd);
2741                 if (rc)
2742                         RETURN(rc);
2743
2744                 /*
2745                  * Forwarding this request to first MDS, it should know LOV
2746                  * desc.
2747                  */
2748                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2749                                   vallen, val, NULL);
2750                 if (!rc && KEY_IS(KEY_CONN_DATA))
2751                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2752                 RETURN(rc);
2753         } else if (KEY_IS(KEY_TGT_COUNT)) {
2754                 *((int *)val) = lmv->desc.ld_tgt_count;
2755                 RETURN(0);
2756         }
2757
2758         CDEBUG(D_IOCTL, "Invalid key\n");
2759         RETURN(-EINVAL);
2760 }
2761
2762 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2763                        obd_count keylen, void *key, obd_count vallen,
2764                        void *val, struct ptlrpc_request_set *set)
2765 {
2766         struct lmv_tgt_desc    *tgt = NULL;
2767         struct obd_device      *obd;
2768         struct lmv_obd         *lmv;
2769         int rc = 0;
2770         ENTRY;
2771
2772         obd = class_exp2obd(exp);
2773         if (obd == NULL) {
2774                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2775                        exp->exp_handle.h_cookie);
2776                 RETURN(-EINVAL);
2777         }
2778         lmv = &obd->u.lmv;
2779
2780         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2781                 int i, err = 0;
2782
2783                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2784                         tgt = lmv->tgts[i];
2785
2786                         if (tgt == NULL || tgt->ltd_exp == NULL)
2787                                 continue;
2788
2789                         err = obd_set_info_async(env, tgt->ltd_exp,
2790                                                  keylen, key, vallen, val, set);
2791                         if (err && rc == 0)
2792                                 rc = err;
2793                 }
2794
2795                 RETURN(rc);
2796         }
2797
2798         RETURN(-EINVAL);
2799 }
2800
2801 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2802                           struct lmv_mds_md_v1 *lmm1)
2803 {
2804         int     cplen;
2805         int     i;
2806
2807         lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2808         lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2809         lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2810         lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2811         cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2812                         sizeof(lmm1->lmv_pool_name));
2813         if (cplen >= sizeof(lmm1->lmv_pool_name))
2814                 return -E2BIG;
2815
2816         for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2817                 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2818                               &lsm->lsm_md_oinfo[i].lmo_fid);
2819         return 0;
2820 }
2821
2822 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2823                 int stripe_count)
2824 {
2825         int     lmm_size = 0;
2826         bool    allocated = false;
2827         int     rc = 0;
2828         ENTRY;
2829
2830         LASSERT(lmmp != NULL);
2831         /* Free lmm */
2832         if (*lmmp != NULL && lsm == NULL) {
2833                 int stripe_count;
2834
2835                 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2836                 lmm_size = lmv_mds_md_size(stripe_count,
2837                                            le32_to_cpu((*lmmp)->lmv_magic));
2838                 if (lmm_size == 0)
2839                         RETURN(-EINVAL);
2840                 OBD_FREE(*lmmp, lmm_size);
2841                 *lmmp = NULL;
2842                 RETURN(0);
2843         }
2844
2845         /* Alloc lmm */
2846         if (*lmmp == NULL && lsm == NULL) {
2847                 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2848                 LASSERT(lmm_size > 0);
2849                 OBD_ALLOC(*lmmp, lmm_size);
2850                 if (*lmmp == NULL)
2851                         RETURN(-ENOMEM);
2852                 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2853                 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2854                 RETURN(lmm_size);
2855         }
2856
2857         /* pack lmm */
2858         LASSERT(lsm != NULL);
2859         lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2860         if (*lmmp == NULL) {
2861                 OBD_ALLOC(*lmmp, lmm_size);
2862                 if (*lmmp == NULL)
2863                         RETURN(-ENOMEM);
2864                 allocated = true;
2865         }
2866
2867         switch (lsm->lsm_md_magic) {
2868         case LMV_MAGIC_V1:
2869                 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2870                 break;
2871         default:
2872                 rc = -EINVAL;
2873                 break;
2874         }
2875
2876         if (rc != 0 && allocated) {
2877                 OBD_FREE(*lmmp, lmm_size);
2878                 *lmmp = NULL;
2879         }
2880
2881         RETURN(lmm_size);
2882 }
2883 EXPORT_SYMBOL(lmv_pack_md);
2884
2885 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2886                             const struct lmv_mds_md_v1 *lmm1)
2887 {
2888         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
2889         int             stripe_count;
2890         int             cplen;
2891         int             i;
2892         int             rc = 0;
2893         ENTRY;
2894
2895         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2896         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2897         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2898         lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2899         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2900         fid_le_to_cpu(&lsm->lsm_md_master_fid, &lmm1->lmv_master_fid);
2901         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2902                         sizeof(lsm->lsm_md_pool_name));
2903
2904         if (!fid_is_sane(&lsm->lsm_md_master_fid))
2905                 RETURN(-EPROTO);
2906
2907         if (cplen >= sizeof(lsm->lsm_md_pool_name))
2908                 RETURN(-E2BIG);
2909
2910         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2911                "layout_version %d\n", lsm->lsm_md_stripe_count,
2912                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2913                lsm->lsm_md_layout_version);
2914
2915         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2916         for (i = 0; i < le32_to_cpu(stripe_count); i++) {
2917                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2918                               &lmm1->lmv_stripe_fids[i]);
2919                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2920                                     &lsm->lsm_md_oinfo[i].lmo_mds);
2921                 if (rc != 0)
2922                         RETURN(rc);
2923                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2924                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2925         }
2926
2927         RETURN(rc);
2928 }
2929
2930 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2931                   const union lmv_mds_md *lmm, int stripe_count)
2932 {
2933         struct lmv_stripe_md     *lsm;
2934         int                      lsm_size;
2935         int                      rc;
2936         bool                     allocated = false;
2937         ENTRY;
2938
2939         LASSERT(lsmp != NULL);
2940
2941         lsm = *lsmp;
2942         /* Free memmd */
2943         if (lsm != NULL && lmm == NULL) {
2944 #ifdef __KERNEL__
2945                 int i;
2946                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2947                         /* For migrating inode, the master stripe and master
2948                          * object will be the same, so do not need iput, see
2949                          * ll_update_lsm_md */
2950                         if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2951                               i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2952                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
2953                 }
2954 #endif
2955                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2956                 OBD_FREE(lsm, lsm_size);
2957                 *lsmp = NULL;
2958                 RETURN(0);
2959         }
2960
2961         /* Alloc memmd */
2962         if (lsm == NULL && lmm == NULL) {
2963                 lsm_size = lmv_stripe_md_size(stripe_count);
2964                 OBD_ALLOC(lsm, lsm_size);
2965                 if (lsm == NULL)
2966                         RETURN(-ENOMEM);
2967                 lsm->lsm_md_stripe_count = stripe_count;
2968                 *lsmp = lsm;
2969                 RETURN(0);
2970         }
2971
2972         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
2973                 RETURN(-EPERM);
2974
2975         /* Unpack memmd */
2976         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
2977             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
2978                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
2979                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
2980                        -EIO);
2981                 RETURN(-EIO);
2982         }
2983
2984         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
2985                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2986         else
2987                 /**
2988                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
2989                  * stripecount should be 0 then.
2990                  */
2991                 lsm_size = lmv_stripe_md_size(0);
2992
2993         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2994         if (lsm == NULL) {
2995                 OBD_ALLOC(lsm, lsm_size);
2996                 if (lsm == NULL)
2997                         RETURN(-ENOMEM);
2998                 allocated = true;
2999                 *lsmp = lsm;
3000         }
3001
3002         switch (le32_to_cpu(lmm->lmv_magic)) {
3003         case LMV_MAGIC_V1:
3004                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3005                 break;
3006         default:
3007                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3008                        le32_to_cpu(lmm->lmv_magic));
3009                 rc = -EINVAL;
3010                 break;
3011         }
3012
3013         if (rc != 0 && allocated) {
3014                 OBD_FREE(lsm, lsm_size);
3015                 *lsmp = NULL;
3016                 lsm_size = rc;
3017         }
3018         RETURN(lsm_size);
3019 }
3020
3021 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3022 {
3023         return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3024 }
3025 EXPORT_SYMBOL(lmv_alloc_memmd);
3026
3027 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3028 {
3029         lmv_unpack_md(NULL, &lsm, NULL, 0);
3030 }
3031 EXPORT_SYMBOL(lmv_free_memmd);
3032
3033 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3034                  struct lov_mds_md *lmm, int disk_len)
3035 {
3036         return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3037                              (union lmv_mds_md *)lmm, disk_len);
3038 }
3039
3040 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3041                struct lov_stripe_md *lsm)
3042 {
3043         struct obd_device               *obd = exp->exp_obd;
3044         struct lmv_obd                  *lmv_obd = &obd->u.lmv;
3045         const struct lmv_stripe_md      *lmv = (struct lmv_stripe_md *)lsm;
3046         int                             stripe_count;
3047
3048         if (lmmp == NULL) {
3049                 if (lsm != NULL)
3050                         stripe_count = lmv->lsm_md_stripe_count;
3051                 else
3052                         stripe_count = lmv_obd->desc.ld_tgt_count;
3053
3054                 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3055         }
3056
3057         return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3058 }
3059
3060 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3061                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
3062                              ldlm_cancel_flags_t flags, void *opaque)
3063 {
3064         struct obd_device       *obd = exp->exp_obd;
3065         struct lmv_obd          *lmv = &obd->u.lmv;
3066         int                      rc = 0;
3067         int                      err;
3068         __u32                    i;
3069         ENTRY;
3070
3071         LASSERT(fid != NULL);
3072
3073         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3074                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3075
3076                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3077                         continue;
3078
3079                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3080                                        opaque);
3081                 if (!rc)
3082                         rc = err;
3083         }
3084         RETURN(rc);
3085 }
3086
3087 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3088                       __u64 *bits)
3089 {
3090         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3091         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3092         int                      rc;
3093         ENTRY;
3094
3095         if (tgt == NULL || tgt->ltd_exp == NULL)
3096                 RETURN(-EINVAL);
3097         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3098         RETURN(rc);
3099 }
3100
3101 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3102                            const struct lu_fid *fid, ldlm_type_t type,
3103                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
3104                            struct lustre_handle *lockh)
3105 {
3106         struct obd_device       *obd = exp->exp_obd;
3107         struct lmv_obd          *lmv = &obd->u.lmv;
3108         ldlm_mode_t             rc;
3109         int                     tgt;
3110         int                     i;
3111         ENTRY;
3112
3113         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3114
3115         /*
3116          * With DNE every object can have two locks in different namespaces:
3117          * lookup lock in space of MDT storing direntry and update/open lock in
3118          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3119          * since this can be easily found, and only try others if that fails.
3120          */
3121         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3122              i < lmv->desc.ld_tgt_count;
3123              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3124                 if (tgt < 0) {
3125                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3126                                obd->obd_name, PFID(fid), tgt);
3127                         tgt = 0;
3128                 }
3129
3130                 if (lmv->tgts[tgt] == NULL ||
3131                     lmv->tgts[tgt]->ltd_exp == NULL ||
3132                     lmv->tgts[tgt]->ltd_active == 0)
3133                         continue;
3134
3135                 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3136                                    type, policy, mode, lockh);
3137                 if (rc)
3138                         RETURN(rc);
3139         }
3140
3141         RETURN(0);
3142 }
3143
3144 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3145                       struct obd_export *dt_exp, struct obd_export *md_exp,
3146                       struct lustre_md *md)
3147 {
3148         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3149         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3150
3151         if (tgt == NULL || tgt->ltd_exp == NULL)
3152                 RETURN(-EINVAL);
3153
3154         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3155 }
3156
3157 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3158 {
3159         struct obd_device       *obd = exp->exp_obd;
3160         struct lmv_obd          *lmv = &obd->u.lmv;
3161         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3162         ENTRY;
3163
3164         if (md->lmv != NULL) {
3165                 lmv_free_memmd(md->lmv);
3166                 md->lmv = NULL;
3167         }
3168         if (tgt == NULL || tgt->ltd_exp == NULL)
3169                 RETURN(-EINVAL);
3170         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3171 }
3172
3173 int lmv_set_open_replay_data(struct obd_export *exp,
3174                              struct obd_client_handle *och,
3175                              struct lookup_intent *it)
3176 {
3177         struct obd_device       *obd = exp->exp_obd;
3178         struct lmv_obd          *lmv = &obd->u.lmv;
3179         struct lmv_tgt_desc     *tgt;
3180         ENTRY;
3181
3182         tgt = lmv_find_target(lmv, &och->och_fid);
3183         if (IS_ERR(tgt))
3184                 RETURN(PTR_ERR(tgt));
3185
3186         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3187 }
3188
3189 int lmv_clear_open_replay_data(struct obd_export *exp,
3190                                struct obd_client_handle *och)
3191 {
3192         struct obd_device       *obd = exp->exp_obd;
3193         struct lmv_obd          *lmv = &obd->u.lmv;
3194         struct lmv_tgt_desc     *tgt;
3195         ENTRY;
3196
3197         tgt = lmv_find_target(lmv, &och->och_fid);
3198         if (IS_ERR(tgt))
3199                 RETURN(PTR_ERR(tgt));
3200
3201         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3202 }
3203
3204 static int lmv_get_remote_perm(struct obd_export *exp,
3205                                const struct lu_fid *fid,
3206                                struct obd_capa *oc, __u32 suppgid,
3207                                struct ptlrpc_request **request)
3208 {
3209         struct obd_device       *obd = exp->exp_obd;
3210         struct lmv_obd          *lmv = &obd->u.lmv;
3211         struct lmv_tgt_desc     *tgt;
3212         int                      rc;
3213         ENTRY;
3214
3215         rc = lmv_check_connect(obd);
3216         if (rc)
3217                 RETURN(rc);
3218
3219         tgt = lmv_find_target(lmv, fid);
3220         if (IS_ERR(tgt))
3221                 RETURN(PTR_ERR(tgt));
3222
3223         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3224         RETURN(rc);
3225 }
3226
3227 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3228                           renew_capa_cb_t cb)
3229 {
3230         struct obd_device       *obd = exp->exp_obd;
3231         struct lmv_obd          *lmv = &obd->u.lmv;
3232         struct lmv_tgt_desc     *tgt;
3233         int                      rc;
3234         ENTRY;
3235
3236         rc = lmv_check_connect(obd);
3237         if (rc)
3238                 RETURN(rc);
3239
3240         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3241         if (IS_ERR(tgt))
3242                 RETURN(PTR_ERR(tgt));
3243
3244         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3245         RETURN(rc);
3246 }
3247
3248 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3249                     const struct req_msg_field *field, struct obd_capa **oc)
3250 {
3251         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3252         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3253
3254         if (tgt == NULL || tgt->ltd_exp == NULL)
3255                 RETURN(-EINVAL);
3256         return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3257 }
3258
3259 int lmv_intent_getattr_async(struct obd_export *exp,
3260                              struct md_enqueue_info *minfo,
3261                              struct ldlm_enqueue_info *einfo)
3262 {
3263         struct md_op_data       *op_data = &minfo->mi_data;
3264         struct obd_device       *obd = exp->exp_obd;
3265         struct lmv_obd          *lmv = &obd->u.lmv;
3266         struct lmv_tgt_desc     *tgt = NULL;
3267         int                      rc;
3268         ENTRY;
3269
3270         rc = lmv_check_connect(obd);
3271         if (rc)
3272                 RETURN(rc);
3273
3274         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3275         if (IS_ERR(tgt))
3276                 RETURN(PTR_ERR(tgt));
3277
3278         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3279         RETURN(rc);
3280 }
3281
3282 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3283                         struct lu_fid *fid, __u64 *bits)
3284 {
3285         struct obd_device       *obd = exp->exp_obd;
3286         struct lmv_obd          *lmv = &obd->u.lmv;
3287         struct lmv_tgt_desc     *tgt;
3288         int                      rc;
3289         ENTRY;
3290
3291         rc = lmv_check_connect(obd);
3292         if (rc)
3293                 RETURN(rc);
3294
3295         tgt = lmv_find_target(lmv, fid);
3296         if (IS_ERR(tgt))
3297                 RETURN(PTR_ERR(tgt));
3298
3299         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3300         RETURN(rc);
3301 }
3302
3303 int lmv_get_fid_from_lsm(struct obd_export *exp,
3304                          const struct lmv_stripe_md *lsm,
3305                          const char *name, int namelen, struct lu_fid *fid)
3306 {
3307         const struct lmv_oinfo *oinfo;
3308
3309         LASSERT(lsm != NULL);
3310         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3311         if (IS_ERR(oinfo))
3312                 return PTR_ERR(oinfo);
3313
3314         *fid = oinfo->lmo_fid;
3315
3316         RETURN(0);
3317 }
3318
3319 /**
3320  * For lmv, only need to send request to master MDT, and the master MDT will
3321  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3322  * we directly fetch data from the slave MDTs.
3323  */
3324 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3325                  struct obd_quotactl *oqctl)
3326 {
3327         struct obd_device   *obd = class_exp2obd(exp);
3328         struct lmv_obd      *lmv = &obd->u.lmv;
3329         struct lmv_tgt_desc *tgt = lmv->tgts[0];
3330         int                  rc = 0;
3331         __u32                i;
3332         __u64                curspace, curinodes;
3333         ENTRY;
3334
3335         if (tgt == NULL ||
3336             tgt->ltd_exp == NULL ||
3337             !tgt->ltd_active ||
3338             lmv->desc.ld_tgt_count == 0) {
3339                 CERROR("master lmv inactive\n");
3340                 RETURN(-EIO);
3341         }
3342
3343         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3344                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3345                 RETURN(rc);
3346         }
3347
3348         curspace = curinodes = 0;
3349         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3350                 int err;
3351                 tgt = lmv->tgts[i];
3352
3353                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3354                         continue;
3355
3356                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3357                 if (err) {
3358                         CERROR("getquota on mdt %d failed. %d\n", i, err);
3359                         if (!rc)
3360                                 rc = err;
3361                 } else {
3362                         curspace += oqctl->qc_dqblk.dqb_curspace;
3363                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3364                 }
3365         }
3366         oqctl->qc_dqblk.dqb_curspace = curspace;
3367         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3368
3369         RETURN(rc);
3370 }
3371
3372 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3373                    struct obd_quotactl *oqctl)
3374 {
3375         struct obd_device       *obd = class_exp2obd(exp);
3376         struct lmv_obd          *lmv = &obd->u.lmv;
3377         struct lmv_tgt_desc     *tgt;
3378         __u32                    i;
3379         int                      rc = 0;
3380         ENTRY;
3381
3382         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3383                 int err;
3384                 tgt = lmv->tgts[i];
3385                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3386                         CERROR("lmv idx %d inactive\n", i);
3387                         RETURN(-EIO);
3388                 }
3389
3390                 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3391                 if (err && !rc)
3392                         rc = err;
3393         }
3394
3395         RETURN(rc);
3396 }
3397
3398 int lmv_update_lsm_md(struct obd_export *exp, struct lmv_stripe_md *lsm,
3399                       struct mdt_body *body, ldlm_blocking_callback cb_blocking)
3400 {
3401         return lmv_revalidate_slaves(exp, body, lsm, cb_blocking, 0);
3402 }
3403
3404 int lmv_merge_attr(struct obd_export *exp, const struct lmv_stripe_md *lsm,
3405                    struct cl_attr *attr)
3406 {
3407 #ifdef __KERNEL__
3408         int i;
3409
3410         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3411                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3412
3413                 CDEBUG(D_INFO, ""DFID" size %llu, nlink %u, atime %lu ctime"
3414                        "%lu, mtime %lu.\n", PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3415                        i_size_read(inode), inode->i_nlink,
3416                        LTIME_S(inode->i_atime), LTIME_S(inode->i_ctime),
3417                        LTIME_S(inode->i_mtime));
3418
3419                 /* for slave stripe, it needs to subtract nlink for . and .. */
3420                 if (i != 0)
3421                         attr->cat_nlink += inode->i_nlink - 2;
3422                 else
3423                         attr->cat_nlink = inode->i_nlink;
3424
3425                 attr->cat_size += i_size_read(inode);
3426
3427                 if (attr->cat_atime < LTIME_S(inode->i_atime))
3428                         attr->cat_atime = LTIME_S(inode->i_atime);
3429
3430                 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3431                         attr->cat_ctime = LTIME_S(inode->i_ctime);
3432
3433                 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3434                         attr->cat_mtime = LTIME_S(inode->i_mtime);
3435         }
3436 #endif
3437         return 0;
3438 }
3439
3440 struct obd_ops lmv_obd_ops = {
3441         .o_owner                = THIS_MODULE,
3442         .o_setup                = lmv_setup,
3443         .o_cleanup              = lmv_cleanup,
3444         .o_precleanup           = lmv_precleanup,
3445         .o_process_config       = lmv_process_config,
3446         .o_connect              = lmv_connect,
3447         .o_disconnect           = lmv_disconnect,
3448         .o_statfs               = lmv_statfs,
3449         .o_get_info             = lmv_get_info,
3450         .o_set_info_async       = lmv_set_info_async,
3451         .o_packmd               = lmv_packmd,
3452         .o_unpackmd             = lmv_unpackmd,
3453         .o_notify               = lmv_notify,
3454         .o_get_uuid             = lmv_get_uuid,
3455         .o_iocontrol            = lmv_iocontrol,
3456         .o_quotacheck           = lmv_quotacheck,
3457         .o_quotactl             = lmv_quotactl
3458 };
3459
3460 struct md_ops lmv_md_ops = {
3461         .m_getstatus            = lmv_getstatus,
3462         .m_null_inode           = lmv_null_inode,
3463         .m_find_cbdata          = lmv_find_cbdata,
3464         .m_close                = lmv_close,
3465         .m_create               = lmv_create,
3466         .m_done_writing         = lmv_done_writing,
3467         .m_enqueue              = lmv_enqueue,
3468         .m_getattr              = lmv_getattr,
3469         .m_getxattr             = lmv_getxattr,
3470         .m_getattr_name         = lmv_getattr_name,
3471         .m_intent_lock          = lmv_intent_lock,
3472         .m_link                 = lmv_link,
3473         .m_rename               = lmv_rename,
3474         .m_setattr              = lmv_setattr,
3475         .m_setxattr             = lmv_setxattr,
3476         .m_fsync                = lmv_fsync,
3477         .m_read_entry           = lmv_read_entry,
3478         .m_unlink               = lmv_unlink,
3479         .m_init_ea_size         = lmv_init_ea_size,
3480         .m_cancel_unused        = lmv_cancel_unused,
3481         .m_set_lock_data        = lmv_set_lock_data,
3482         .m_lock_match           = lmv_lock_match,
3483         .m_get_lustre_md        = lmv_get_lustre_md,
3484         .m_free_lustre_md       = lmv_free_lustre_md,
3485         .m_update_lsm_md        = lmv_update_lsm_md,
3486         .m_merge_attr           = lmv_merge_attr,
3487         .m_set_open_replay_data = lmv_set_open_replay_data,
3488         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3489         .m_renew_capa           = lmv_renew_capa,
3490         .m_unpack_capa          = lmv_unpack_capa,
3491         .m_get_remote_perm      = lmv_get_remote_perm,
3492         .m_intent_getattr_async = lmv_intent_getattr_async,
3493         .m_revalidate_lock      = lmv_revalidate_lock,
3494         .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
3495 };
3496
3497 int __init lmv_init(void)
3498 {
3499         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3500 #ifndef HAVE_ONLY_PROCFS_SEQ
3501                                    NULL,
3502 #endif
3503                                    LUSTRE_LMV_NAME, NULL);
3504 }
3505
3506 #ifdef __KERNEL__
3507 static void lmv_exit(void)
3508 {
3509         class_unregister_type(LUSTRE_LMV_NAME);
3510 }
3511
3512 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3513 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3514 MODULE_LICENSE("GPL");
3515
3516 module_init(lmv_init);
3517 module_exit(lmv_exit);
3518 #endif