lustre/lmv/lmv_obd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LMV
  38 #include <linux/slab.h>
  39 #include <linux/module.h>
  40 #include <linux/init.h>
  41 #include <linux/user_namespace.h>
  42 #ifdef HAVE_UIDGID_HEADER
  43 # include <linux/uidgid.h>
  44 #endif
  45 #include <linux/slab.h>
  46 #include <linux/pagemap.h>
  47 #include <linux/mm.h>
  48 #include <linux/math64.h>
  49 #include <linux/seq_file.h>
  50 #include <linux/namei.h>
  51
  52 #include <lustre/lustre_idl.h>
  53 #include <obd_support.h>
  54 #include <lustre_lib.h>
  55 #include <lustre_net.h>
  56 #include <obd_class.h>
  57 #include <lustre_lmv.h>
  58 #include <lprocfs_status.h>
  59 #include <cl_object.h>
  60 #include <lustre_fid.h>
  61 #include <lustre_ioctl.h>
  62 #include <lustre_kernelcomm.h>
  63 #include "lmv_internal.h"
  64
  65 static void lmv_activate_target(struct lmv_obd *lmv,
  66                                 struct lmv_tgt_desc *tgt,
  67                                 int activate)
  68 {
  69         if (tgt->ltd_active == activate)
  70                 return;
  71
  72         tgt->ltd_active = activate;
  73         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  74
  75         tgt->ltd_exp->exp_obd->obd_inactive = !activate;
  76 }
  77
  78 /**
  79  * Error codes:
  80  *
  81  *  -EINVAL  : UUID can't be found in the LMV's target list
  82  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  83  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  84  */
  85 static int lmv_set_mdc_active(struct lmv_obd *lmv,
  86                               const struct obd_uuid *uuid,
  87                               int activate)
  88 {
  89         struct lmv_tgt_desc     *tgt = NULL;
  90         struct obd_device       *obd;
  91         __u32                    i;
  92         int                      rc = 0;
  93         ENTRY;
  94
  95         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  96                         lmv, uuid->uuid, activate);
  97
  98         spin_lock(&lmv->lmv_lock);
  99         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 100                 tgt = lmv->tgts[i];
 101                 if (tgt == NULL || tgt->ltd_exp == NULL)
 102                         continue;
 103
 104                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
 105                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 106
 107                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 108                         break;
 109         }
 110
 111         if (i == lmv->desc.ld_tgt_count)
 112                 GOTO(out_lmv_lock, rc = -EINVAL);
 113
 114         obd = class_exp2obd(tgt->ltd_exp);
 115         if (obd == NULL)
 116                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 117
 118         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 119                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 120                obd->obd_type->typ_name, i);
 121         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 122
 123         if (tgt->ltd_active == activate) {
 124                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 125                        activate ? "" : "in");
 126                 GOTO(out_lmv_lock, rc);
 127         }
 128
 129         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
 130                activate ? "" : "in");
 131         lmv_activate_target(lmv, tgt, activate);
 132         EXIT;
 133
 134  out_lmv_lock:
 135         spin_unlock(&lmv->lmv_lock);
 136         return rc;
 137 }
 138
 139 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
 140 {
 141         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
 142         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
 143
 144         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
 145 }
 146
 147 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 148                       enum obd_notify_event ev, void *data)
 149 {
 150         struct obd_connect_data *conn_data;
 151         struct lmv_obd          *lmv = &obd->u.lmv;
 152         struct obd_uuid         *uuid;
 153         int                      rc = 0;
 154         ENTRY;
 155
 156         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 157                 CERROR("unexpected notification of %s %s!\n",
 158                        watched->obd_type->typ_name,
 159                        watched->obd_name);
 160                 RETURN(-EINVAL);
 161         }
 162
 163         uuid = &watched->u.cli.cl_target_uuid;
 164         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 165                 /*
 166                  * Set MDC as active before notifying the observer, so the
 167                  * observer can use the MDC normally.
 168                  */
 169                 rc = lmv_set_mdc_active(lmv, uuid,
 170                                         ev == OBD_NOTIFY_ACTIVE);
 171                 if (rc) {
 172                         CERROR("%sactivation of %s failed: %d\n",
 173                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
 174                                uuid->uuid, rc);
 175                         RETURN(rc);
 176                 }
 177         } else if (ev == OBD_NOTIFY_OCD) {
 178                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
 179                 /*
 180                  * XXX: Make sure that ocd_connect_flags from all targets are
 181                  * the same. Otherwise one of MDTs runs wrong version or
 182                  * something like this.  --umka
 183                  */
 184                 obd->obd_self_export->exp_connect_data = *conn_data;
 185         }
 186 #if 0
 187         else if (ev == OBD_NOTIFY_DISCON) {
 188                 /*
 189                  * For disconnect event, flush fld cache for failout MDS case.
 190                  */
 191                 fld_client_flush(&lmv->lmv_fld);
 192         }
 193 #endif
 194         /*
 195          * Pass the notification up the chain.
 196          */
 197         if (obd->obd_observer)
 198                 rc = obd_notify(obd->obd_observer, watched, ev, data);
 199
 200         RETURN(rc);
 201 }
 202
 203 /**
 204  * This is fake connect function. Its purpose is to initialize lmv and say
 205  * caller that everything is okay. Real connection will be performed later.
 206  */
 207 static int lmv_connect(const struct lu_env *env,
 208                        struct obd_export **exp, struct obd_device *obd,
 209                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 210                        void *localdata)
 211 {
 212         struct lmv_obd        *lmv = &obd->u.lmv;
 213         struct lustre_handle  conn = { 0 };
 214         int                    rc = 0;
 215         ENTRY;
 216
 217         /*
 218          * We don't want to actually do the underlying connections more than
 219          * once, so keep track.
 220          */
 221         lmv->refcount++;
 222         if (lmv->refcount > 1) {
 223                 *exp = NULL;
 224                 RETURN(0);
 225         }
 226
 227         rc = class_connect(&conn, obd, cluuid);
 228         if (rc) {
 229                 CERROR("class_connection() returned %d\n", rc);
 230                 RETURN(rc);
 231         }
 232
 233         *exp = class_conn2export(&conn);
 234         class_export_get(*exp);
 235
 236         lmv->exp = *exp;
 237         lmv->connected = 0;
 238         lmv->cluuid = *cluuid;
 239
 240         if (data)
 241                 lmv->conn_data = *data;
 242
 243         if (lmv->targets_proc_entry == NULL) {
 244                 lmv->targets_proc_entry = lprocfs_register("target_obds",
 245                                                            obd->obd_proc_entry,
 246                                                            NULL, NULL);
 247                 if (IS_ERR(lmv->targets_proc_entry)) {
 248                         CERROR("%s: cannot register "
 249                                "/proc/fs/lustre/%s/%s/target_obds\n",
 250                                obd->obd_name, obd->obd_type->typ_name,
 251                                obd->obd_name);
 252                         lmv->targets_proc_entry = NULL;
 253                 }
 254         }
 255
 256         /*
 257          * All real clients should perform actual connection right away, because
 258          * it is possible, that LMV will not have opportunity to connect targets
 259          * and MDC stuff will be called directly, for instance while reading
 260          * ../mdc/../kbytesfree procfs file, etc.
 261          */
 262         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
 263                 rc = lmv_check_connect(obd);
 264
 265         if (rc && lmv->targets_proc_entry != NULL)
 266                 lprocfs_remove(&lmv->targets_proc_entry);
 267         RETURN(rc);
 268 }
 269
 270 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
 271                             __u32 def_easize)
 272 {
 273         struct obd_device       *obd = exp->exp_obd;
 274         struct lmv_obd          *lmv = &obd->u.lmv;
 275         __u32                    i;
 276         int                      rc = 0;
 277         int                      change = 0;
 278         ENTRY;
 279
 280         if (lmv->max_easize < easize) {
 281                 lmv->max_easize = easize;
 282                 change = 1;
 283         }
 284         if (lmv->max_def_easize < def_easize) {
 285                 lmv->max_def_easize = def_easize;
 286                 change = 1;
 287         }
 288
 289         if (change == 0)
 290                 RETURN(0);
 291
 292         if (lmv->connected == 0)
 293                 RETURN(0);
 294
 295         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 296                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 297
 298                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
 299                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 300                         continue;
 301                 }
 302
 303                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
 304                 if (rc) {
 305                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
 306                                " rc = %d\n", obd->obd_name, i, rc);
 307                         break;
 308                 }
 309         }
 310         RETURN(rc);
 311 }
 312
 313 #define MAX_STRING_SIZE 128
 314
 315 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 316 {
 317         struct lmv_obd          *lmv = &obd->u.lmv;
 318         struct obd_uuid         *cluuid = &lmv->cluuid;
 319         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
 320         struct obd_device       *mdc_obd;
 321         struct obd_export       *mdc_exp;
 322         struct lu_fld_target     target;
 323         int                      rc;
 324         ENTRY;
 325
 326         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
 327                                         &obd->obd_uuid);
 328         if (!mdc_obd) {
 329                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
 330                 RETURN(-EINVAL);
 331         }
 332
 333         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
 334                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 335                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
 336                 cluuid->uuid);
 337
 338         if (!mdc_obd->obd_set_up) {
 339                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
 340                 RETURN(-EINVAL);
 341         }
 342
 343         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
 344                          &lmv->conn_data, NULL);
 345         if (rc) {
 346                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
 347                 RETURN(rc);
 348         }
 349
 350         /*
 351          * Init fid sequence client for this mdc and add new fld target.
 352          */
 353         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
 354         if (rc)
 355                 RETURN(rc);
 356
 357         target.ft_srv = NULL;
 358         target.ft_exp = mdc_exp;
 359         target.ft_idx = tgt->ltd_idx;
 360
 361         fld_client_add_target(&lmv->lmv_fld, &target);
 362
 363         rc = obd_register_observer(mdc_obd, obd);
 364         if (rc) {
 365                 obd_disconnect(mdc_exp);
 366                 CERROR("target %s register_observer error %d\n",
 367                        tgt->ltd_uuid.uuid, rc);
 368                 RETURN(rc);
 369         }
 370
 371         if (obd->obd_observer) {
 372                 /*
 373                  * Tell the observer about the new target.
 374                  */
 375                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
 376                                 OBD_NOTIFY_ACTIVE,
 377                                 (void *)(tgt - lmv->tgts[0]));
 378                 if (rc) {
 379                         obd_disconnect(mdc_exp);
 380                         RETURN(rc);
 381                 }
 382         }
 383
 384         tgt->ltd_active = 1;
 385         tgt->ltd_exp = mdc_exp;
 386         lmv->desc.ld_active_tgt_count++;
 387
 388         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
 389
 390         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
 391                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
 392                 atomic_read(&obd->obd_refcount));
 393
 394         if (lmv->targets_proc_entry != NULL) {
 395                 struct proc_dir_entry *mdc_symlink;
 396
 397                 LASSERT(mdc_obd->obd_type != NULL);
 398                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
 399                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
 400                                                   lmv->targets_proc_entry,
 401                                                   "../../../%s/%s",
 402                                                   mdc_obd->obd_type->typ_name,
 403                                                   mdc_obd->obd_name);
 404                 if (mdc_symlink == NULL) {
 405                         CERROR("cannot register LMV target "
 406                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
 407                                obd->obd_type->typ_name, obd->obd_name,
 408                                mdc_obd->obd_name);
 409                 }
 410         }
 411         RETURN(0);
 412 }
 413
 414 static void lmv_del_target(struct lmv_obd *lmv, int index)
 415 {
 416         if (lmv->tgts[index] == NULL)
 417                 return;
 418
 419         OBD_FREE_PTR(lmv->tgts[index]);
 420         lmv->tgts[index] = NULL;
 421         return;
 422 }
 423
 424 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 425                            __u32 index, int gen)
 426 {
 427         struct obd_device *mdc_obd;
 428         struct lmv_obd      *lmv = &obd->u.lmv;
 429         struct lmv_tgt_desc *tgt;
 430         int                  orig_tgt_count = 0;
 431         int                  rc = 0;
 432         ENTRY;
 433
 434         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
 435         mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
 436                                         &obd->obd_uuid);
 437         if (!mdc_obd) {
 438                 CERROR("%s: Target %s not attached: rc = %d\n",
 439                        obd->obd_name, uuidp->uuid, -EINVAL);
 440                 RETURN(-EINVAL);
 441         }
 442
 443         mutex_lock(&lmv->lmv_init_mutex);
 444         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
 445                 tgt = lmv->tgts[index];
 446                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
 447                        " rc = %d\n", obd->obd_name,
 448                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
 449                 mutex_unlock(&lmv->lmv_init_mutex);
 450                 RETURN(-EEXIST);
 451         }
 452
 453         if (index >= lmv->tgts_size) {
 454                 /* We need to reallocate the lmv target array. */
 455                 struct lmv_tgt_desc **newtgts, **old = NULL;
 456                 __u32 newsize = 1;
 457                 __u32 oldsize = 0;
 458
 459                 while (newsize < index + 1)
 460                         newsize = newsize << 1;
 461                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 462                 if (newtgts == NULL) {
 463                         mutex_unlock(&lmv->lmv_init_mutex);
 464                         RETURN(-ENOMEM);
 465                 }
 466
 467                 if (lmv->tgts_size) {
 468                         memcpy(newtgts, lmv->tgts,
 469                                sizeof(*newtgts) * lmv->tgts_size);
 470                         old = lmv->tgts;
 471                         oldsize = lmv->tgts_size;
 472                 }
 473
 474                 lmv->tgts = newtgts;
 475                 lmv->tgts_size = newsize;
 476                 smp_rmb();
 477                 if (old)
 478                         OBD_FREE(old, sizeof(*old) * oldsize);
 479
 480                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
 481                        lmv->tgts_size);
 482         }
 483
 484         OBD_ALLOC_PTR(tgt);
 485         if (!tgt) {
 486                 mutex_unlock(&lmv->lmv_init_mutex);
 487                 RETURN(-ENOMEM);
 488         }
 489
 490         mutex_init(&tgt->ltd_fid_mutex);
 491         tgt->ltd_idx = index;
 492         tgt->ltd_uuid = *uuidp;
 493         tgt->ltd_active = 0;
 494         lmv->tgts[index] = tgt;
 495         if (index >= lmv->desc.ld_tgt_count) {
 496                 orig_tgt_count = lmv->desc.ld_tgt_count;
 497                 lmv->desc.ld_tgt_count = index + 1;
 498         }
 499
 500         if (lmv->connected == 0) {
 501                 /* lmv_check_connect() will connect this target. */
 502                 mutex_unlock(&lmv->lmv_init_mutex);
 503                 RETURN(0);
 504         }
 505
 506         /* Otherwise let's connect it ourselves */
 507         mutex_unlock(&lmv->lmv_init_mutex);
 508         rc = lmv_connect_mdc(obd, tgt);
 509         if (rc != 0) {
 510                 spin_lock(&lmv->lmv_lock);
 511                 if (lmv->desc.ld_tgt_count == index + 1)
 512                         lmv->desc.ld_tgt_count = orig_tgt_count;
 513                 memset(tgt, 0, sizeof(*tgt));
 514                 spin_unlock(&lmv->lmv_lock);
 515         } else {
 516                 int easize = sizeof(struct lmv_stripe_md) +
 517                         lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
 518                 lmv_init_ea_size(obd->obd_self_export, easize, 0);
 519         }
 520
 521         RETURN(rc);
 522 }
 523
 524 int lmv_check_connect(struct obd_device *obd)
 525 {
 526         struct lmv_obd          *lmv = &obd->u.lmv;
 527         struct lmv_tgt_desc     *tgt;
 528         __u32                    i;
 529         int                      rc;
 530         int                      easize;
 531         ENTRY;
 532
 533         if (lmv->connected)
 534                 RETURN(0);
 535
 536         mutex_lock(&lmv->lmv_init_mutex);
 537         if (lmv->connected) {
 538                 mutex_unlock(&lmv->lmv_init_mutex);
 539                 RETURN(0);
 540         }
 541
 542         if (lmv->desc.ld_tgt_count == 0) {
 543                 mutex_unlock(&lmv->lmv_init_mutex);
 544                 CERROR("%s: no targets configured.\n", obd->obd_name);
 545                 RETURN(-EINVAL);
 546         }
 547
 548         LASSERT(lmv->tgts != NULL);
 549
 550         if (lmv->tgts[0] == NULL) {
 551                 mutex_unlock(&lmv->lmv_init_mutex);
 552                 CERROR("%s: no target configured for index 0.\n",
 553                        obd->obd_name);
 554                 RETURN(-EINVAL);
 555         }
 556
 557         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
 558                lmv->cluuid.uuid, obd->obd_name);
 559
 560         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 561                 tgt = lmv->tgts[i];
 562                 if (tgt == NULL)
 563                         continue;
 564                 rc = lmv_connect_mdc(obd, tgt);
 565                 if (rc)
 566                         GOTO(out_disc, rc);
 567         }
 568
 569         class_export_put(lmv->exp);
 570         lmv->connected = 1;
 571         easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
 572         lmv_init_ea_size(obd->obd_self_export, easize, 0);
 573         mutex_unlock(&lmv->lmv_init_mutex);
 574         RETURN(0);
 575
 576  out_disc:
 577         while (i-- > 0) {
 578                 int rc2;
 579                 tgt = lmv->tgts[i];
 580                 if (tgt == NULL)
 581                         continue;
 582                 tgt->ltd_active = 0;
 583                 if (tgt->ltd_exp) {
 584                         --lmv->desc.ld_active_tgt_count;
 585                         rc2 = obd_disconnect(tgt->ltd_exp);
 586                         if (rc2) {
 587                                 CERROR("LMV target %s disconnect on "
 588                                        "MDC idx %d: error %d\n",
 589                                        tgt->ltd_uuid.uuid, i, rc2);
 590                         }
 591                 }
 592         }
 593         class_disconnect(lmv->exp);
 594         mutex_unlock(&lmv->lmv_init_mutex);
 595         RETURN(rc);
 596 }
 597
 598 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
 599 {
 600         struct lmv_obd         *lmv = &obd->u.lmv;
 601         struct obd_device      *mdc_obd;
 602         int                     rc;
 603         ENTRY;
 604
 605         LASSERT(tgt != NULL);
 606         LASSERT(obd != NULL);
 607
 608         mdc_obd = class_exp2obd(tgt->ltd_exp);
 609
 610         if (mdc_obd) {
 611                 mdc_obd->obd_force = obd->obd_force;
 612                 mdc_obd->obd_fail = obd->obd_fail;
 613                 mdc_obd->obd_no_recov = obd->obd_no_recov;
 614
 615                 if (lmv->targets_proc_entry != NULL)
 616                         lprocfs_remove_proc_entry(mdc_obd->obd_name,
 617                                                   lmv->targets_proc_entry);
 618         }
 619
 620         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
 621         if (rc)
 622                 CERROR("Can't finanize fids factory\n");
 623
 624         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
 625                tgt->ltd_exp->exp_obd->obd_name,
 626                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
 627
 628         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 629         rc = obd_disconnect(tgt->ltd_exp);
 630         if (rc) {
 631                 if (tgt->ltd_active) {
 632                         CERROR("Target %s disconnect error %d\n",
 633                                tgt->ltd_uuid.uuid, rc);
 634                 }
 635         }
 636
 637         lmv_activate_target(lmv, tgt, 0);
 638         tgt->ltd_exp = NULL;
 639         RETURN(0);
 640 }
 641
 642 static int lmv_disconnect(struct obd_export *exp)
 643 {
 644         struct obd_device       *obd = class_exp2obd(exp);
 645         struct lmv_obd          *lmv = &obd->u.lmv;
 646         int                      rc;
 647         __u32                    i;
 648         ENTRY;
 649
 650         if (!lmv->tgts)
 651                 goto out_local;
 652
 653         /*
 654          * Only disconnect the underlying layers on the final disconnect.
 655          */
 656         lmv->refcount--;
 657         if (lmv->refcount != 0)
 658                 goto out_local;
 659
 660         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 661                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
 662                         continue;
 663
 664                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
 665         }
 666
 667         if (lmv->targets_proc_entry != NULL)
 668                 lprocfs_remove(&lmv->targets_proc_entry);
 669         else
 670                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 671                        obd->obd_type->typ_name, obd->obd_name);
 672
 673 out_local:
 674         /*
 675          * This is the case when no real connection is established by
 676          * lmv_check_connect().
 677          */
 678         if (!lmv->connected)
 679                 class_export_put(exp);
 680         rc = class_disconnect(exp);
 681         if (lmv->refcount == 0)
 682                 lmv->connected = 0;
 683         RETURN(rc);
 684 }
 685
 686 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
 687                         void __user *uarg)
 688 {
 689         struct obd_device       *obddev = class_exp2obd(exp);
 690         struct lmv_obd          *lmv = &obddev->u.lmv;
 691         struct getinfo_fid2path *gf;
 692         struct lmv_tgt_desc     *tgt;
 693         struct getinfo_fid2path *remote_gf = NULL;
 694         struct lu_fid           root_fid;
 695         int                     remote_gf_size = 0;
 696         int                     rc;
 697
 698         gf = karg;
 699         tgt = lmv_find_target(lmv, &gf->gf_fid);
 700         if (IS_ERR(tgt))
 701                 RETURN(PTR_ERR(tgt));
 702
 703         root_fid = *gf->gf_u.gf_root_fid;
 704         LASSERT(fid_is_sane(&root_fid));
 705
 706 repeat_fid2path:
 707         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
 708         if (rc != 0 && rc != -EREMOTE)
 709                 GOTO(out_fid2path, rc);
 710
 711         /* If remote_gf != NULL, it means just building the
 712          * path on the remote MDT, copy this path segement to gf */
 713         if (remote_gf != NULL) {
 714                 struct getinfo_fid2path *ori_gf;
 715                 char *ptr;
 716
 717                 ori_gf = (struct getinfo_fid2path *)karg;
 718                 if (strlen(ori_gf->gf_u.gf_path) +
 719                     strlen(gf->gf_u.gf_path) > ori_gf->gf_pathlen)
 720                         GOTO(out_fid2path, rc = -EOVERFLOW);
 721
 722                 ptr = ori_gf->gf_u.gf_path;
 723
 724                 memmove(ptr + strlen(gf->gf_u.gf_path) + 1, ptr,
 725                         strlen(ori_gf->gf_u.gf_path));
 726
 727                 strncpy(ptr, gf->gf_u.gf_path,
 728                         strlen(gf->gf_u.gf_path));
 729                 ptr += strlen(gf->gf_u.gf_path);
 730                 *ptr = '/';
 731         }
 732
 733         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
 734                tgt->ltd_exp->exp_obd->obd_name,
 735                gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
 736                gf->gf_linkno);
 737
 738         if (rc == 0)
 739                 GOTO(out_fid2path, rc);
 740
 741         /* sigh, has to go to another MDT to do path building further */
 742         if (remote_gf == NULL) {
 743                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
 744                 OBD_ALLOC(remote_gf, remote_gf_size);
 745                 if (remote_gf == NULL)
 746                         GOTO(out_fid2path, rc = -ENOMEM);
 747                 remote_gf->gf_pathlen = PATH_MAX;
 748         }
 749
 750         if (!fid_is_sane(&gf->gf_fid)) {
 751                 CERROR("%s: invalid FID "DFID": rc = %d\n",
 752                        tgt->ltd_exp->exp_obd->obd_name,
 753                        PFID(&gf->gf_fid), -EINVAL);
 754                 GOTO(out_fid2path, rc = -EINVAL);
 755         }
 756
 757         tgt = lmv_find_target(lmv, &gf->gf_fid);
 758         if (IS_ERR(tgt))
 759                 GOTO(out_fid2path, rc = -EINVAL);
 760
 761         remote_gf->gf_fid = gf->gf_fid;
 762         remote_gf->gf_recno = -1;
 763         remote_gf->gf_linkno = -1;
 764         memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
 765         *remote_gf->gf_u.gf_root_fid = root_fid;
 766         gf = remote_gf;
 767         goto repeat_fid2path;
 768
 769 out_fid2path:
 770         if (remote_gf != NULL)
 771                 OBD_FREE(remote_gf, remote_gf_size);
 772         RETURN(rc);
 773 }
 774
 775 static int lmv_hsm_req_count(struct lmv_obd *lmv,
 776                              const struct hsm_user_request *hur,
 777                              const struct lmv_tgt_desc *tgt_mds)
 778 {
 779         __u32                    i;
 780         int                      nr = 0;
 781         struct lmv_tgt_desc     *curr_tgt;
 782
 783         /* count how many requests must be sent to the given target */
 784         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
 785                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
 786                 if (IS_ERR(curr_tgt))
 787                         RETURN(PTR_ERR(curr_tgt));
 788                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
 789                         nr++;
 790         }
 791         return nr;
 792 }
 793
 794 static int lmv_hsm_req_build(struct lmv_obd *lmv,
 795                               struct hsm_user_request *hur_in,
 796                               const struct lmv_tgt_desc *tgt_mds,
 797                               struct hsm_user_request *hur_out)
 798 {
 799         __u32                    i, nr_out;
 800         struct lmv_tgt_desc     *curr_tgt;
 801
 802         /* build the hsm_user_request for the given target */
 803         hur_out->hur_request = hur_in->hur_request;
 804         nr_out = 0;
 805         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
 806                 curr_tgt = lmv_find_target(lmv,
 807                                            &hur_in->hur_user_item[i].hui_fid);
 808                 if (IS_ERR(curr_tgt))
 809                         RETURN(PTR_ERR(curr_tgt));
 810                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
 811                         hur_out->hur_user_item[nr_out] =
 812                                                 hur_in->hur_user_item[i];
 813                         nr_out++;
 814                 }
 815         }
 816         hur_out->hur_request.hr_itemcount = nr_out;
 817         memcpy(hur_data(hur_out), hur_data(hur_in),
 818                hur_in->hur_request.hr_data_len);
 819
 820         RETURN(0);
 821 }
 822
 823 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
 824                                  struct lustre_kernelcomm *lk,
 825                                  void __user *uarg)
 826 {
 827         __u32   i;
 828         int     rc;
 829         ENTRY;
 830
 831         /* unregister request (call from llapi_hsm_copytool_fini) */
 832         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 833                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 834
 835                 if (tgt == NULL || tgt->ltd_exp == NULL)
 836                         continue;
 837                 /* best effort: try to clean as much as possible
 838                  * (continue on error) */
 839                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 840         }
 841
 842         /* Whatever the result, remove copytool from kuc groups.
 843          * Unreached coordinators will get EPIPE on next requests
 844          * and will unregister automatically.
 845          */
 846         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group);
 847
 848         RETURN(rc);
 849 }
 850
 851 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
 852                                struct lustre_kernelcomm *lk, __user void *uarg)
 853 {
 854         struct file             *filp;
 855         __u32                    i, j;
 856         int                      err, rc;
 857         bool                     any_set = false;
 858         struct kkuc_ct_data      kcd = { 0 };
 859         ENTRY;
 860
 861         /* All or nothing: try to register to all MDS.
 862          * In case of failure, unregister from previous MDS,
 863          * except if it because of inactive target. */
 864         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 865                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
 866
 867                 if (tgt == NULL || tgt->ltd_exp == NULL)
 868                         continue;
 869                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
 870                 if (err) {
 871                         if (tgt->ltd_active) {
 872                                 /* permanent error */
 873                                 CERROR("%s: iocontrol MDC %s on MDT"
 874                                        " idx %d cmd %x: err = %d\n",
 875                                        class_exp2obd(lmv->exp)->obd_name,
 876                                        tgt->ltd_uuid.uuid, i, cmd, err);
 877                                 rc = err;
 878                                 lk->lk_flags |= LK_FLG_STOP;
 879                                 /* unregister from previous MDS */
 880                                 for (j = 0; j < i; j++) {
 881                                         tgt = lmv->tgts[j];
 882                                         if (tgt == NULL || tgt->ltd_exp == NULL)
 883                                                 continue;
 884                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
 885                                                       lk, uarg);
 886                                 }
 887                                 RETURN(rc);
 888                         }
 889                         /* else: transient error.
 890                          * kuc will register to the missing MDT
 891                          * when it is back */
 892                 } else {
 893                         any_set = true;
 894                 }
 895         }
 896
 897         if (!any_set)
 898                 /* no registration done: return error */
 899                 RETURN(-ENOTCONN);
 900
 901         /* at least one registration done, with no failure */
 902         filp = fget(lk->lk_wfd);
 903         if (filp == NULL)
 904                 RETURN(-EBADF);
 905
 906         kcd.kcd_magic = KKUC_CT_DATA_MAGIC;
 907         kcd.kcd_uuid = lmv->cluuid;
 908         kcd.kcd_archive = lk->lk_data;
 909
 910         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group,
 911                                    &kcd, sizeof(kcd));
 912         if (rc != 0)
 913                 fput(filp);
 914
 915         RETURN(rc);
 916 }
 917
 918
 919
 920
 921 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 922                          int len, void *karg, void __user *uarg)
 923 {
 924         struct obd_device       *obddev = class_exp2obd(exp);
 925         struct lmv_obd          *lmv = &obddev->u.lmv;
 926         struct lmv_tgt_desc     *tgt = NULL;
 927         __u32                    i = 0;
 928         int                      rc = 0;
 929         int                      set = 0;
 930         __u32                    count = lmv->desc.ld_tgt_count;
 931         ENTRY;
 932
 933         if (count == 0)
 934                 RETURN(-ENOTTY);
 935
 936         switch (cmd) {
 937         case IOC_OBD_STATFS: {
 938                 struct obd_ioctl_data *data = karg;
 939                 struct obd_device *mdc_obd;
 940                 struct obd_statfs stat_buf = {0};
 941                 __u32 index;
 942
 943                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
 944                 if ((index >= count))
 945                         RETURN(-ENODEV);
 946
 947                 tgt = lmv->tgts[index];
 948                 if (tgt == NULL || !tgt->ltd_active)
 949                         RETURN(-ENODATA);
 950
 951                 mdc_obd = class_exp2obd(tgt->ltd_exp);
 952                 if (!mdc_obd)
 953                         RETURN(-EINVAL);
 954
 955                 /* copy UUID */
 956                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
 957                                  min((int) data->ioc_plen2,
 958                                      (int) sizeof(struct obd_uuid))))
 959                         RETURN(-EFAULT);
 960
 961                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
 962                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 963                                 0);
 964                 if (rc)
 965                         RETURN(rc);
 966                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
 967                                  min((int) data->ioc_plen1,
 968                                      (int) sizeof(stat_buf))))
 969                         RETURN(-EFAULT);
 970                 break;
 971         }
 972         case OBD_IOC_QUOTACTL: {
 973                 struct if_quotactl *qctl = karg;
 974                 struct obd_quotactl *oqctl;
 975
 976                 if (qctl->qc_valid == QC_MDTIDX) {
 977                         if (count <= qctl->qc_idx)
 978                                 RETURN(-EINVAL);
 979
 980                         tgt = lmv->tgts[qctl->qc_idx];
 981                         if (tgt == NULL || tgt->ltd_exp == NULL)
 982                                 RETURN(-EINVAL);
 983                 } else if (qctl->qc_valid == QC_UUID) {
 984                         for (i = 0; i < count; i++) {
 985                                 tgt = lmv->tgts[i];
 986                                 if (tgt == NULL)
 987                                         continue;
 988                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
 989                                                      &qctl->obd_uuid))
 990                                         continue;
 991
 992                                 if (tgt->ltd_exp == NULL)
 993                                         RETURN(-EINVAL);
 994
 995                                 break;
 996                         }
 997                 } else {
 998                         RETURN(-EINVAL);
 999                 }
1000
1001                 if (i >= count)
1002                         RETURN(-EAGAIN);
1003
1004                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1005                 OBD_ALLOC_PTR(oqctl);
1006                 if (!oqctl)
1007                         RETURN(-ENOMEM);
1008
1009                 QCTL_COPY(oqctl, qctl);
1010                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1011                 if (rc == 0) {
1012                         QCTL_COPY(qctl, oqctl);
1013                         qctl->qc_valid = QC_MDTIDX;
1014                         qctl->obd_uuid = tgt->ltd_uuid;
1015                 }
1016                 OBD_FREE_PTR(oqctl);
1017                 break;
1018         }
1019         case OBD_IOC_CHANGELOG_SEND:
1020         case OBD_IOC_CHANGELOG_CLEAR: {
1021                 struct ioc_changelog *icc = karg;
1022
1023                 if (icc->icc_mdtindex >= count)
1024                         RETURN(-ENODEV);
1025
1026                 tgt = lmv->tgts[icc->icc_mdtindex];
1027                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1028                         RETURN(-ENODEV);
1029                 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1030                 break;
1031         }
1032         case LL_IOC_GET_CONNECT_FLAGS: {
1033                 tgt = lmv->tgts[0];
1034                 if (tgt == NULL || tgt->ltd_exp == NULL)
1035                         RETURN(-ENODATA);
1036                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1037                 break;
1038         }
1039         case LL_IOC_FID2MDTIDX: {
1040                 struct lu_fid *fid = karg;
1041                 int             mdt_index;
1042
1043                 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1044                 if (rc != 0)
1045                         RETURN(rc);
1046
1047                 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1048                  * point to user space memory for FID2MDTIDX. */
1049                 *(__u32 *)uarg = mdt_index;
1050                 break;
1051         }
1052         case OBD_IOC_FID2PATH: {
1053                 rc = lmv_fid2path(exp, len, karg, uarg);
1054                 break;
1055         }
1056         case LL_IOC_HSM_STATE_GET:
1057         case LL_IOC_HSM_STATE_SET:
1058         case LL_IOC_HSM_ACTION: {
1059                 struct md_op_data       *op_data = karg;
1060
1061                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1062                 if (IS_ERR(tgt))
1063                         RETURN(PTR_ERR(tgt));
1064
1065                 if (tgt->ltd_exp == NULL)
1066                         RETURN(-EINVAL);
1067
1068                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1069                 break;
1070         }
1071         case LL_IOC_HSM_PROGRESS: {
1072                 const struct hsm_progress_kernel *hpk = karg;
1073
1074                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1075                 if (IS_ERR(tgt))
1076                         RETURN(PTR_ERR(tgt));
1077                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1078                 break;
1079         }
1080         case LL_IOC_HSM_REQUEST: {
1081                 struct hsm_user_request *hur = karg;
1082                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1083
1084                 if (reqcount == 0)
1085                         RETURN(0);
1086
1087                 /* if the request is about a single fid
1088                  * or if there is a single MDS, no need to split
1089                  * the request. */
1090                 if (reqcount == 1 || count == 1) {
1091                         tgt = lmv_find_target(lmv,
1092                                               &hur->hur_user_item[0].hui_fid);
1093                         if (IS_ERR(tgt))
1094                                 RETURN(PTR_ERR(tgt));
1095                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1096                 } else {
1097                         /* split fid list to their respective MDS */
1098                         for (i = 0; i < count; i++) {
1099                                 int nr, rc1;
1100                                 size_t reqlen;
1101                                 struct hsm_user_request *req;
1102
1103                                 tgt = lmv->tgts[i];
1104                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1105                                         continue;
1106
1107                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1108                                 if (nr < 0)
1109                                         RETURN(nr);
1110                                 if (nr == 0) /* nothing for this MDS */
1111                                         continue;
1112
1113                                 /* build a request with fids for this MDS */
1114                                 reqlen = offsetof(typeof(*hur),
1115                                                   hur_user_item[nr])
1116                                                 + hur->hur_request.hr_data_len;
1117                                 OBD_ALLOC_LARGE(req, reqlen);
1118                                 if (req == NULL)
1119                                         RETURN(-ENOMEM);
1120                                 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
1121                                 if (rc1 < 0)
1122                                         GOTO(hsm_req_err, rc1);
1123                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1124                                                     req, uarg);
1125 hsm_req_err:
1126                                 if (rc1 != 0 && rc == 0)
1127                                         rc = rc1;
1128                                 OBD_FREE_LARGE(req, reqlen);
1129                         }
1130                 }
1131                 break;
1132         }
1133         case LL_IOC_LOV_SWAP_LAYOUTS: {
1134                 struct md_op_data       *op_data = karg;
1135                 struct lmv_tgt_desc     *tgt1, *tgt2;
1136
1137                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1138                 if (IS_ERR(tgt1))
1139                         RETURN(PTR_ERR(tgt1));
1140
1141                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1142                 if (IS_ERR(tgt2))
1143                         RETURN(PTR_ERR(tgt2));
1144
1145                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1146                         RETURN(-EINVAL);
1147
1148                 /* only files on same MDT can have their layouts swapped */
1149                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1150                         RETURN(-EPERM);
1151
1152                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1153                 break;
1154         }
1155         case LL_IOC_HSM_CT_START: {
1156                 struct lustre_kernelcomm *lk = karg;
1157                 if (lk->lk_flags & LK_FLG_STOP)
1158                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1159                 else
1160                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1161                 break;
1162         }
1163         default:
1164                 for (i = 0; i < count; i++) {
1165                         struct obd_device *mdc_obd;
1166                         int err;
1167
1168                         tgt = lmv->tgts[i];
1169                         if (tgt == NULL || tgt->ltd_exp == NULL)
1170                                 continue;
1171                         /* ll_umount_begin() sets force flag but for lmv, not
1172                          * mdc. Let's pass it through */
1173                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1174                         mdc_obd->obd_force = obddev->obd_force;
1175                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1176                         if (err) {
1177                                 if (tgt->ltd_active) {
1178                                         CERROR("error: iocontrol MDC %s on MDT"
1179                                                " idx %d cmd %x: err = %d\n",
1180                                                tgt->ltd_uuid.uuid, i, cmd, err);
1181                                         if (!rc)
1182                                                 rc = err;
1183                                 }
1184                         } else
1185                                 set = 1;
1186                 }
1187                 if (!set && !rc)
1188                         rc = -EIO;
1189         }
1190         RETURN(rc);
1191 }
1192
1193 #if 0
1194 static int lmv_all_chars_policy(int count, const char *name,
1195                                 int len)
1196 {
1197         unsigned int c = 0;
1198
1199         while (len > 0)
1200                 c += name[--len];
1201         c = c % count;
1202         return c;
1203 }
1204
1205 static int lmv_nid_policy(struct lmv_obd *lmv)
1206 {
1207         struct obd_import *imp;
1208         __u32              id;
1209
1210         /*
1211          * XXX: To get nid we assume that underlying obd device is mdc.
1212          */
1213         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1214         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1215         return id % lmv->desc.ld_tgt_count;
1216 }
1217
1218 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1219                           placement_policy_t placement)
1220 {
1221         switch (placement) {
1222         case PLACEMENT_CHAR_POLICY:
1223                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1224                                             op_data->op_name,
1225                                             op_data->op_namelen);
1226         case PLACEMENT_NID_POLICY:
1227                 return lmv_nid_policy(lmv);
1228
1229         default:
1230                 break;
1231         }
1232
1233         CERROR("Unsupported placement policy %x\n", placement);
1234         return -EINVAL;
1235 }
1236 #endif
1237
1238 /**
1239  * This is _inode_ placement policy function (not name).
1240  */
1241 static int lmv_placement_policy(struct obd_device *obd,
1242                                 struct md_op_data *op_data, u32 *mds)
1243 {
1244         struct lmv_obd          *lmv = &obd->u.lmv;
1245         ENTRY;
1246
1247         LASSERT(mds != NULL);
1248
1249         if (lmv->desc.ld_tgt_count == 1) {
1250                 *mds = 0;
1251                 RETURN(0);
1252         }
1253
1254         if (op_data->op_default_stripe_offset != -1) {
1255                 *mds = op_data->op_default_stripe_offset;
1256                 RETURN(0);
1257         }
1258
1259         /**
1260          * If stripe_offset is provided during setdirstripe
1261          * (setdirstripe -i xx), xx MDS will be choosen.
1262          */
1263         if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1264                 struct lmv_user_md *lum;
1265
1266                 lum = op_data->op_data;
1267
1268                 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1269                         *mds = le32_to_cpu(lum->lum_stripe_offset);
1270                 } else {
1271                         /* -1 means default, which will be in the same MDT with
1272                          * the stripe */
1273                         *mds = op_data->op_mds;
1274                         lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1275                 }
1276         } else {
1277                 /* Allocate new fid on target according to operation type and
1278                  * parent home mds. */
1279                 *mds = op_data->op_mds;
1280         }
1281
1282         RETURN(0);
1283 }
1284
1285 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1286 {
1287         struct lmv_tgt_desc     *tgt;
1288         int                      rc;
1289         ENTRY;
1290
1291         tgt = lmv_get_target(lmv, mds, NULL);
1292         if (IS_ERR(tgt))
1293                 RETURN(PTR_ERR(tgt));
1294
1295         /*
1296          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1297          * on server that seq in new allocated fid is not yet known.
1298          */
1299         mutex_lock(&tgt->ltd_fid_mutex);
1300
1301         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1302                 GOTO(out, rc = -ENODEV);
1303
1304         /*
1305          * Asking underlying tgt layer to allocate new fid.
1306          */
1307         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1308         if (rc > 0) {
1309                 LASSERT(fid_is_sane(fid));
1310                 rc = 0;
1311         }
1312
1313         EXIT;
1314 out:
1315         mutex_unlock(&tgt->ltd_fid_mutex);
1316         return rc;
1317 }
1318
1319 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1320                   struct lu_fid *fid, struct md_op_data *op_data)
1321 {
1322         struct obd_device     *obd = class_exp2obd(exp);
1323         struct lmv_obd        *lmv = &obd->u.lmv;
1324         u32                    mds = 0;
1325         int                    rc;
1326         ENTRY;
1327
1328         LASSERT(op_data != NULL);
1329         LASSERT(fid != NULL);
1330
1331         rc = lmv_placement_policy(obd, op_data, &mds);
1332         if (rc) {
1333                 CERROR("Can't get target for allocating fid, "
1334                        "rc %d\n", rc);
1335                 RETURN(rc);
1336         }
1337
1338         rc = __lmv_fid_alloc(lmv, fid, mds);
1339         if (rc) {
1340                 CERROR("Can't alloc new fid, rc %d\n", rc);
1341                 RETURN(rc);
1342         }
1343
1344         RETURN(rc);
1345 }
1346
1347 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1348 {
1349         struct lmv_obd  *lmv = &obd->u.lmv;
1350         struct lmv_desc *desc;
1351         int             rc;
1352         ENTRY;
1353
1354         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1355                 CERROR("LMV setup requires a descriptor\n");
1356                 RETURN(-EINVAL);
1357         }
1358
1359         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1360         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1361                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1362                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1363                 RETURN(-EINVAL);
1364         }
1365
1366         lmv->tgts_size = 32U;
1367         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1368         if (lmv->tgts == NULL)
1369                 RETURN(-ENOMEM);
1370
1371         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1372         lmv->desc.ld_tgt_count = 0;
1373         lmv->desc.ld_active_tgt_count = 0;
1374         lmv->max_def_easize = 0;
1375         lmv->max_easize = 0;
1376         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1377
1378         spin_lock_init(&lmv->lmv_lock);
1379         mutex_init(&lmv->lmv_init_mutex);
1380
1381 #ifdef CONFIG_PROC_FS
1382         obd->obd_vars = lprocfs_lmv_obd_vars;
1383         lprocfs_obd_setup(obd);
1384         lprocfs_alloc_md_stats(obd, 0);
1385         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1386                                 0444, &lmv_proc_target_fops, obd);
1387         if (rc)
1388                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1389                       obd->obd_name, rc);
1390 #endif
1391         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1392                              LUSTRE_CLI_FLD_HASH_DHT);
1393         if (rc) {
1394                 CERROR("Can't init FLD, err %d\n", rc);
1395                 GOTO(out, rc);
1396         }
1397
1398         RETURN(0);
1399
1400 out:
1401         return rc;
1402 }
1403
1404 static int lmv_cleanup(struct obd_device *obd)
1405 {
1406         struct lmv_obd   *lmv = &obd->u.lmv;
1407         ENTRY;
1408
1409         fld_client_fini(&lmv->lmv_fld);
1410         if (lmv->tgts != NULL) {
1411                 int i;
1412                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1413                         if (lmv->tgts[i] == NULL)
1414                                 continue;
1415                         lmv_del_target(lmv, i);
1416                 }
1417                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1418                 lmv->tgts_size = 0;
1419         }
1420         RETURN(0);
1421 }
1422
1423 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1424 {
1425         struct lustre_cfg       *lcfg = buf;
1426         struct obd_uuid         obd_uuid;
1427         int                     gen;
1428         __u32                   index;
1429         int                     rc;
1430         ENTRY;
1431
1432         switch (lcfg->lcfg_command) {
1433         case LCFG_ADD_MDC:
1434                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1435                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1436                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1437                         GOTO(out, rc = -EINVAL);
1438
1439                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1440
1441                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1442                         GOTO(out, rc = -EINVAL);
1443                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1444                         GOTO(out, rc = -EINVAL);
1445                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1446                 GOTO(out, rc);
1447         default:
1448                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1449                 GOTO(out, rc = -EINVAL);
1450         }
1451 out:
1452         RETURN(rc);
1453 }
1454
1455 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1456                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1457 {
1458         struct obd_device       *obd = class_exp2obd(exp);
1459         struct lmv_obd          *lmv = &obd->u.lmv;
1460         struct obd_statfs       *temp;
1461         int                      rc = 0;
1462         __u32                    i;
1463         ENTRY;
1464
1465         rc = lmv_check_connect(obd);
1466         if (rc)
1467                 RETURN(rc);
1468
1469         OBD_ALLOC(temp, sizeof(*temp));
1470         if (temp == NULL)
1471                 RETURN(-ENOMEM);
1472
1473         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1474                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1475                         continue;
1476
1477                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1478                                 max_age, flags);
1479                 if (rc) {
1480                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1481                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1482                                rc);
1483                         GOTO(out_free_temp, rc);
1484                 }
1485
1486                 if (i == 0) {
1487                         *osfs = *temp;
1488                         /* If the statfs is from mount, it will needs
1489                          * retrieve necessary information from MDT0.
1490                          * i.e. mount does not need the merged osfs
1491                          * from all of MDT.
1492                          * And also clients can be mounted as long as
1493                          * MDT0 is in service*/
1494                         if (flags & OBD_STATFS_FOR_MDT0)
1495                                 GOTO(out_free_temp, rc);
1496                 } else {
1497                         osfs->os_bavail += temp->os_bavail;
1498                         osfs->os_blocks += temp->os_blocks;
1499                         osfs->os_ffree += temp->os_ffree;
1500                         osfs->os_files += temp->os_files;
1501                 }
1502         }
1503
1504         EXIT;
1505 out_free_temp:
1506         OBD_FREE(temp, sizeof(*temp));
1507         return rc;
1508 }
1509
1510 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1511                         struct lu_fid *fid)
1512 {
1513         struct obd_device    *obd = exp->exp_obd;
1514         struct lmv_obd       *lmv = &obd->u.lmv;
1515         int                   rc;
1516         ENTRY;
1517
1518         rc = lmv_check_connect(obd);
1519         if (rc)
1520                 RETURN(rc);
1521
1522         rc = md_get_root(lmv->tgts[0]->ltd_exp, fileset, fid);
1523         RETURN(rc);
1524 }
1525
1526 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1527                         u64 valid, const char *name,
1528                         const char *input, int input_size, int output_size,
1529                         int flags, struct ptlrpc_request **request)
1530 {
1531         struct obd_device      *obd = exp->exp_obd;
1532         struct lmv_obd         *lmv = &obd->u.lmv;
1533         struct lmv_tgt_desc    *tgt;
1534         int                     rc;
1535         ENTRY;
1536
1537         rc = lmv_check_connect(obd);
1538         if (rc)
1539                 RETURN(rc);
1540
1541         tgt = lmv_find_target(lmv, fid);
1542         if (IS_ERR(tgt))
1543                 RETURN(PTR_ERR(tgt));
1544
1545         rc = md_getxattr(tgt->ltd_exp, fid, valid, name, input,
1546                          input_size, output_size, flags, request);
1547
1548         RETURN(rc);
1549 }
1550
1551 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1552                         u64 valid, const char *name,
1553                         const char *input, int input_size, int output_size,
1554                         int flags, __u32 suppgid,
1555                         struct ptlrpc_request **request)
1556 {
1557         struct obd_device      *obd = exp->exp_obd;
1558         struct lmv_obd         *lmv = &obd->u.lmv;
1559         struct lmv_tgt_desc    *tgt;
1560         int                     rc;
1561         ENTRY;
1562
1563         rc = lmv_check_connect(obd);
1564         if (rc)
1565                 RETURN(rc);
1566
1567         tgt = lmv_find_target(lmv, fid);
1568         if (IS_ERR(tgt))
1569                 RETURN(PTR_ERR(tgt));
1570
1571         rc = md_setxattr(tgt->ltd_exp, fid, valid, name, input,
1572                          input_size, output_size, flags, suppgid,
1573                          request);
1574
1575         RETURN(rc);
1576 }
1577
1578 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1579                        struct ptlrpc_request **request)
1580 {
1581         struct obd_device       *obd = exp->exp_obd;
1582         struct lmv_obd          *lmv = &obd->u.lmv;
1583         struct lmv_tgt_desc     *tgt;
1584         int                      rc;
1585         ENTRY;
1586
1587         rc = lmv_check_connect(obd);
1588         if (rc)
1589                 RETURN(rc);
1590
1591         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1592         if (IS_ERR(tgt))
1593                 RETURN(PTR_ERR(tgt));
1594
1595         if (op_data->op_flags & MF_GET_MDT_IDX) {
1596                 op_data->op_mds = tgt->ltd_idx;
1597                 RETURN(0);
1598         }
1599
1600         rc = md_getattr(tgt->ltd_exp, op_data, request);
1601
1602         RETURN(rc);
1603 }
1604
1605 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1606 {
1607         struct obd_device   *obd = exp->exp_obd;
1608         struct lmv_obd      *lmv = &obd->u.lmv;
1609         __u32                i;
1610         int                  rc;
1611         ENTRY;
1612
1613         rc = lmv_check_connect(obd);
1614         if (rc)
1615                 RETURN(rc);
1616
1617         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1618
1619         /*
1620          * With DNE every object can have two locks in different namespaces:
1621          * lookup lock in space of MDT storing direntry and update/open lock in
1622          * space of MDT storing inode.
1623          */
1624         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1625                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1626                         continue;
1627                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1628         }
1629
1630         RETURN(0);
1631 }
1632
1633 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1634                      struct md_open_data *mod, struct ptlrpc_request **request)
1635 {
1636         struct obd_device     *obd = exp->exp_obd;
1637         struct lmv_obd        *lmv = &obd->u.lmv;
1638         struct lmv_tgt_desc   *tgt;
1639         int                    rc;
1640         ENTRY;
1641
1642         rc = lmv_check_connect(obd);
1643         if (rc)
1644                 RETURN(rc);
1645
1646         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1647         if (IS_ERR(tgt))
1648                 RETURN(PTR_ERR(tgt));
1649
1650         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1651         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1652         RETURN(rc);
1653 }
1654
1655 /**
1656  * Choosing the MDT by name or FID in @op_data.
1657  * For non-striped directory, it will locate MDT by fid.
1658  * For striped-directory, it will locate MDT by name. And also
1659  * it will reset op_fid1 with the FID of the choosen stripe.
1660  **/
1661 struct lmv_tgt_desc *
1662 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1663                            const char *name, int namelen, struct lu_fid *fid,
1664                            u32 *mds)
1665 {
1666         struct lmv_tgt_desc     *tgt;
1667         const struct lmv_oinfo  *oinfo;
1668
1669         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1670                 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1671                         RETURN(ERR_PTR(-EBADF));
1672                 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1673         } else {
1674                 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1675                 if (IS_ERR(oinfo))
1676                         RETURN(ERR_CAST(oinfo));
1677         }
1678
1679         if (fid != NULL)
1680                 *fid = oinfo->lmo_fid;
1681         if (mds != NULL)
1682                 *mds = oinfo->lmo_mds;
1683
1684         tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1685
1686         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", oinfo->lmo_mds,
1687                PFID(&oinfo->lmo_fid));
1688         return tgt;
1689 }
1690
1691 /**
1692  * Locate mds by fid or name
1693  *
1694  * For striped directory (lsm != NULL), it will locate the stripe
1695  * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1696  * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1697  * walk through all of stripes to locate the entry.
1698  *
1699  * For normal direcotry, it will locate MDS by FID directly.
1700  * \param[in] lmv       LMV device
1701  * \param[in] op_data   client MD stack parameters, name, namelen
1702  *                      mds_num etc.
1703  * \param[in] fid       object FID used to locate MDS.
1704  *
1705  * retval               pointer to the lmv_tgt_desc if succeed.
1706  *                      ERR_PTR(errno) if failed.
1707  */
1708 struct lmv_tgt_desc*
1709 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1710                struct lu_fid *fid)
1711 {
1712         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1713         struct lmv_tgt_desc     *tgt;
1714
1715         /* During creating VOLATILE file, it should honor the mdt
1716          * index if the file under striped dir is being restored, see
1717          * ct_restore(). */
1718         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1719             (int)op_data->op_mds != -1) {
1720                 int i;
1721                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1722                 if (IS_ERR(tgt))
1723                         return tgt;
1724
1725                 if (lsm != NULL) {
1726                         /* refill the right parent fid */
1727                         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1728                                 struct lmv_oinfo *oinfo;
1729
1730                                 oinfo = &lsm->lsm_md_oinfo[i];
1731                                 if (oinfo->lmo_mds == op_data->op_mds) {
1732                                         *fid = oinfo->lmo_fid;
1733                                         break;
1734                                 }
1735                         }
1736
1737                         if (i == lsm->lsm_md_stripe_count)
1738                                 *fid = lsm->lsm_md_oinfo[0].lmo_fid;
1739                 }
1740
1741                 return tgt;
1742         }
1743
1744         if (lsm == NULL || op_data->op_namelen == 0) {
1745                 tgt = lmv_find_target(lmv, fid);
1746                 if (IS_ERR(tgt))
1747                         return tgt;
1748
1749                 op_data->op_mds = tgt->ltd_idx;
1750                 return tgt;
1751         }
1752
1753         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1754                                           op_data->op_namelen, fid,
1755                                           &op_data->op_mds);
1756 }
1757
1758 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1759                 const void *data, size_t datalen, umode_t mode, uid_t uid,
1760                 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1761                 struct ptlrpc_request **request)
1762 {
1763         struct obd_device       *obd = exp->exp_obd;
1764         struct lmv_obd          *lmv = &obd->u.lmv;
1765         struct lmv_tgt_desc     *tgt;
1766         int                      rc;
1767         ENTRY;
1768
1769         rc = lmv_check_connect(obd);
1770         if (rc)
1771                 RETURN(rc);
1772
1773         if (!lmv->desc.ld_active_tgt_count)
1774                 RETURN(-EIO);
1775
1776         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1777         if (IS_ERR(tgt))
1778                 RETURN(PTR_ERR(tgt));
1779
1780         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1781                 (int)op_data->op_namelen, op_data->op_name,
1782                 PFID(&op_data->op_fid1), op_data->op_mds);
1783
1784         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1785         if (rc)
1786                 RETURN(rc);
1787         if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1788                 /* Send the create request to the MDT where the object
1789                  * will be located */
1790                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1791                 if (IS_ERR(tgt))
1792                         RETURN(PTR_ERR(tgt));
1793
1794                 op_data->op_mds = tgt->ltd_idx;
1795         } else {
1796                 CDEBUG(D_CONFIG, "Server doesn't support striped dirs\n");
1797         }
1798
1799         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1800                PFID(&op_data->op_fid2), op_data->op_mds);
1801
1802         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1803         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1804                        cap_effective, rdev, request);
1805         if (rc == 0) {
1806                 if (*request == NULL)
1807                         RETURN(rc);
1808                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1809         }
1810         RETURN(rc);
1811 }
1812
1813 static int
1814 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1815             const union ldlm_policy_data *policy,
1816             struct lookup_intent *it, struct md_op_data *op_data,
1817             struct lustre_handle *lockh, __u64 extra_lock_flags)
1818 {
1819         struct obd_device        *obd = exp->exp_obd;
1820         struct lmv_obd           *lmv = &obd->u.lmv;
1821         struct lmv_tgt_desc      *tgt;
1822         int                       rc;
1823         ENTRY;
1824
1825         rc = lmv_check_connect(obd);
1826         if (rc)
1827                 RETURN(rc);
1828
1829         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1830                LL_IT2STR(it), PFID(&op_data->op_fid1));
1831
1832         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1833         if (IS_ERR(tgt))
1834                 RETURN(PTR_ERR(tgt));
1835
1836         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1837                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1838
1839         rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1840                         extra_lock_flags);
1841
1842         RETURN(rc);
1843 }
1844
1845 static int
1846 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1847                  struct ptlrpc_request **preq)
1848 {
1849         struct ptlrpc_request   *req = NULL;
1850         struct obd_device       *obd = exp->exp_obd;
1851         struct lmv_obd          *lmv = &obd->u.lmv;
1852         struct lmv_tgt_desc     *tgt;
1853         struct mdt_body         *body;
1854         int                      rc;
1855         ENTRY;
1856
1857         rc = lmv_check_connect(obd);
1858         if (rc)
1859                 RETURN(rc);
1860
1861         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1862         if (IS_ERR(tgt))
1863                 RETURN(PTR_ERR(tgt));
1864
1865         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1866                 (int)op_data->op_namelen, op_data->op_name,
1867                 PFID(&op_data->op_fid1), tgt->ltd_idx);
1868
1869         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1870         if (rc != 0)
1871                 RETURN(rc);
1872
1873         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1874         LASSERT(body != NULL);
1875
1876         if (body->mbo_valid & OBD_MD_MDS) {
1877                 struct lu_fid rid = body->mbo_fid1;
1878                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1879                        PFID(&rid));
1880
1881                 tgt = lmv_find_target(lmv, &rid);
1882                 if (IS_ERR(tgt)) {
1883                         ptlrpc_req_finished(*preq);
1884                         preq = NULL;
1885                         RETURN(PTR_ERR(tgt));
1886                 }
1887
1888                 op_data->op_fid1 = rid;
1889                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1890                 op_data->op_namelen = 0;
1891                 op_data->op_name = NULL;
1892                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1893                 ptlrpc_req_finished(*preq);
1894                 *preq = req;
1895         }
1896
1897         RETURN(rc);
1898 }
1899
1900 #define md_op_data_fid(op_data, fl)                     \
1901         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1902          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1903          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1904          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1905          NULL)
1906
1907 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1908                             struct md_op_data *op_data, __u32 op_tgt,
1909                             enum ldlm_mode mode, int bits, int flag)
1910 {
1911         struct lu_fid *fid = md_op_data_fid(op_data, flag);
1912         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
1913         union ldlm_policy_data policy = { { 0 } };
1914         int rc = 0;
1915         ENTRY;
1916
1917         if (!fid_is_sane(fid))
1918                 RETURN(0);
1919
1920         if (tgt == NULL) {
1921                 tgt = lmv_find_target(lmv, fid);
1922                 if (IS_ERR(tgt))
1923                         RETURN(PTR_ERR(tgt));
1924         }
1925
1926         if (tgt->ltd_idx != op_tgt) {
1927                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1928                 policy.l_inodebits.bits = bits;
1929                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1930                                       mode, LCF_ASYNC, NULL);
1931         } else {
1932                 CDEBUG(D_INODE,
1933                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1934                        op_tgt, PFID(fid));
1935                 op_data->op_flags |= flag;
1936                 rc = 0;
1937         }
1938
1939         RETURN(rc);
1940 }
1941
1942 /*
1943  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1944  * op_data->op_fid2
1945  */
1946 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1947                     struct ptlrpc_request **request)
1948 {
1949         struct obd_device       *obd = exp->exp_obd;
1950         struct lmv_obd          *lmv = &obd->u.lmv;
1951         struct lmv_tgt_desc     *tgt;
1952         int                      rc;
1953         ENTRY;
1954
1955         rc = lmv_check_connect(obd);
1956         if (rc)
1957                 RETURN(rc);
1958
1959         LASSERT(op_data->op_namelen != 0);
1960
1961         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1962                PFID(&op_data->op_fid2), (int)op_data->op_namelen,
1963                op_data->op_name, PFID(&op_data->op_fid1));
1964
1965         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
1966         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
1967         op_data->op_cap = cfs_curproc_cap_pack();
1968         if (op_data->op_mea2 != NULL) {
1969                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
1970                 const struct lmv_oinfo  *oinfo;
1971
1972                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
1973                                                 op_data->op_namelen);
1974                 if (IS_ERR(oinfo))
1975                         RETURN(PTR_ERR(oinfo));
1976
1977                 op_data->op_fid2 = oinfo->lmo_fid;
1978         }
1979
1980         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1981         if (IS_ERR(tgt))
1982                 RETURN(PTR_ERR(tgt));
1983
1984         /*
1985          * Cancel UPDATE lock on child (fid1).
1986          */
1987         op_data->op_flags |= MF_MDC_CANCEL_FID2;
1988         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
1989                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
1990         if (rc != 0)
1991                 RETURN(rc);
1992
1993         rc = md_link(tgt->ltd_exp, op_data, request);
1994
1995         RETURN(rc);
1996 }
1997
1998 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
1999                       const char *old, size_t oldlen,
2000                       const char *new, size_t newlen,
2001                       struct ptlrpc_request **request)
2002 {
2003         struct obd_device       *obd = exp->exp_obd;
2004         struct lmv_obd          *lmv = &obd->u.lmv;
2005         struct lmv_tgt_desc     *src_tgt;
2006         struct lmv_tgt_desc     *tgt_tgt;
2007         struct obd_export       *target_exp;
2008         struct mdt_body         *body;
2009         int                     rc;
2010         ENTRY;
2011
2012         LASSERT(oldlen != 0);
2013
2014         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2015                (int)oldlen, old, PFID(&op_data->op_fid1),
2016                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2017                (int)newlen, new, PFID(&op_data->op_fid2),
2018                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2019
2020         rc = lmv_check_connect(obd);
2021         if (rc)
2022                 RETURN(rc);
2023
2024         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2025         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2026         op_data->op_cap = cfs_curproc_cap_pack();
2027         if (op_data->op_cli_flags & CLI_MIGRATE) {
2028                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2029                          PFID(&op_data->op_fid3));
2030
2031                 if (op_data->op_mea1 != NULL) {
2032                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2033                         struct lmv_tgt_desc     *tmp;
2034
2035                         /* Fix the parent fid for striped dir */
2036                         tmp = lmv_locate_target_for_name(lmv, lsm, old,
2037                                                          oldlen,
2038                                                          &op_data->op_fid1,
2039                                                          NULL);
2040                         if (IS_ERR(tmp))
2041                                 RETURN(PTR_ERR(tmp));
2042                 }
2043
2044                 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2045                 if (rc != 0)
2046                         RETURN(rc);
2047
2048                 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2049                 if (IS_ERR(src_tgt))
2050                         RETURN(PTR_ERR(src_tgt));
2051
2052                 target_exp = src_tgt->ltd_exp;
2053         } else {
2054                 if (op_data->op_mea1 != NULL) {
2055                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2056
2057                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2058                                                              oldlen,
2059                                                              &op_data->op_fid1,
2060                                                              &op_data->op_mds);
2061                 } else {
2062                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2063                 }
2064                 if (IS_ERR(src_tgt))
2065                         RETURN(PTR_ERR(src_tgt));
2066
2067
2068                 if (op_data->op_mea2 != NULL) {
2069                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
2070
2071                         tgt_tgt = lmv_locate_target_for_name(lmv, lsm, new,
2072                                                              newlen,
2073                                                              &op_data->op_fid2,
2074                                                              &op_data->op_mds);
2075                 } else {
2076                         tgt_tgt = lmv_find_target(lmv, &op_data->op_fid2);
2077
2078                 }
2079                 if (IS_ERR(tgt_tgt))
2080                         RETURN(PTR_ERR(tgt_tgt));
2081
2082                 target_exp = tgt_tgt->ltd_exp;
2083         }
2084
2085         /*
2086          * LOOKUP lock on src child (fid3) should also be cancelled for
2087          * src_tgt in mdc_rename.
2088          */
2089         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2090
2091         /*
2092          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2093          * own target.
2094          */
2095         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2096                               LCK_EX, MDS_INODELOCK_UPDATE,
2097                               MF_MDC_CANCEL_FID2);
2098
2099         if (rc != 0)
2100                 RETURN(rc);
2101         /*
2102          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2103          */
2104         if (fid_is_sane(&op_data->op_fid3)) {
2105                 struct lmv_tgt_desc *tgt;
2106
2107                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2108                 if (IS_ERR(tgt))
2109                         RETURN(PTR_ERR(tgt));
2110
2111                 /* Cancel LOOKUP lock on its parent */
2112                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2113                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2114                                       MF_MDC_CANCEL_FID3);
2115                 if (rc != 0)
2116                         RETURN(rc);
2117
2118                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2119                                       LCK_EX, MDS_INODELOCK_FULL,
2120                                       MF_MDC_CANCEL_FID3);
2121                 if (rc != 0)
2122                         RETURN(rc);
2123         }
2124
2125 retry_rename:
2126         /*
2127          * Cancel all the locks on tgt child (fid4).
2128          */
2129         if (fid_is_sane(&op_data->op_fid4)) {
2130                 struct lmv_tgt_desc *tgt;
2131
2132                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2133                                       LCK_EX, MDS_INODELOCK_FULL,
2134                                       MF_MDC_CANCEL_FID4);
2135                 if (rc != 0)
2136                         RETURN(rc);
2137
2138                 tgt = lmv_find_target(lmv, &op_data->op_fid4);
2139                 if (IS_ERR(tgt))
2140                         RETURN(PTR_ERR(tgt));
2141
2142                 /* Since the target child might be destroyed, and it might
2143                  * become orphan, and we can only check orphan on the local
2144                  * MDT right now, so we send rename request to the MDT where
2145                  * target child is located. If target child does not exist,
2146                  * then it will send the request to the target parent */
2147                 target_exp = tgt->ltd_exp;
2148         }
2149
2150         rc = md_rename(target_exp, op_data, old, oldlen, new, newlen,
2151                        request);
2152
2153         if (rc != 0 && rc != -EXDEV)
2154                 RETURN(rc);
2155
2156         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2157         if (body == NULL)
2158                 RETURN(-EPROTO);
2159
2160         /* Not cross-ref case, just get out of here. */
2161         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2162                 RETURN(rc);
2163
2164         CDEBUG(D_INODE, "%s: try rename to another MDT for "DFID"\n",
2165                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2166
2167         op_data->op_fid4 = body->mbo_fid1;
2168         ptlrpc_req_finished(*request);
2169         *request = NULL;
2170         goto retry_rename;
2171 }
2172
2173 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2174                        void *ea, size_t ealen, struct ptlrpc_request **request)
2175 {
2176         struct obd_device       *obd = exp->exp_obd;
2177         struct lmv_obd          *lmv = &obd->u.lmv;
2178         struct lmv_tgt_desc     *tgt;
2179         int                      rc = 0;
2180         ENTRY;
2181
2182         rc = lmv_check_connect(obd);
2183         if (rc)
2184                 RETURN(rc);
2185
2186         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2187                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2188
2189         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2190         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2191         if (IS_ERR(tgt))
2192                 RETURN(PTR_ERR(tgt));
2193
2194         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2195
2196         RETURN(rc);
2197 }
2198
2199 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2200                      struct ptlrpc_request **request)
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;
2206         ENTRY;
2207
2208         rc = lmv_check_connect(obd);
2209         if (rc != 0)
2210                 RETURN(rc);
2211
2212         tgt = lmv_find_target(lmv, fid);
2213         if (IS_ERR(tgt))
2214                 RETURN(PTR_ERR(tgt));
2215
2216         rc = md_fsync(tgt->ltd_exp, fid, request);
2217         RETURN(rc);
2218 }
2219
2220 /**
2221  * Get current minimum entry from striped directory
2222  *
2223  * This function will search the dir entry, whose hash value is the
2224  * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2225  * only being called for striped directory.
2226  *
2227  * \param[in] exp               export of LMV
2228  * \param[in] op_data           parameters transferred beween client MD stack
2229  *                              stripe_information will be included in this
2230  *                              parameter
2231  * \param[in] cb_op             ldlm callback being used in enqueue in
2232  *                              mdc_read_page
2233  * \param[in] hash_offset       the hash value, which is used to locate
2234  *                              minum(closet) dir entry
2235  * \param[in|out] stripe_offset the caller use this to indicate the stripe
2236  *                              index of last entry, so to avoid hash conflict
2237  *                              between stripes. It will also be used to
2238  *                              return the stripe index of current dir entry.
2239  * \param[in|out] entp          the minum entry and it also is being used
2240  *                              to input the last dir entry to resolve the
2241  *                              hash conflict
2242  *
2243  * \param[out] ppage            the page which holds the minum entry
2244  *
2245  * \retval                      = 0 get the entry successfully
2246  *                              negative errno (< 0) does not get the entry
2247  */
2248 static int lmv_get_min_striped_entry(struct obd_export *exp,
2249                                      struct md_op_data *op_data,
2250                                      struct md_callback *cb_op,
2251                                      __u64 hash_offset, int *stripe_offset,
2252                                      struct lu_dirent **entp,
2253                                      struct page **ppage)
2254 {
2255         struct obd_device       *obd = exp->exp_obd;
2256         struct lmv_obd          *lmv = &obd->u.lmv;
2257         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2258         struct lmv_tgt_desc     *tgt;
2259         int                     stripe_count;
2260         struct lu_dirent        *min_ent = NULL;
2261         struct page             *min_page = NULL;
2262         int                     min_idx = 0;
2263         int                     i;
2264         int                     rc = 0;
2265         ENTRY;
2266
2267         stripe_count = lsm->lsm_md_stripe_count;
2268         for (i = 0; i < stripe_count; i++) {
2269                 struct lu_dirent        *ent = NULL;
2270                 struct page             *page = NULL;
2271                 struct lu_dirpage       *dp;
2272                 __u64                   stripe_hash = hash_offset;
2273
2274                 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2275                 if (IS_ERR(tgt))
2276                         GOTO(out, rc = PTR_ERR(tgt));
2277
2278                 /* op_data will be shared by each stripe, so we need
2279                  * reset these value for each stripe */
2280                 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2281                 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2282                 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2283 next:
2284                 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2285                                   &page);
2286                 if (rc != 0)
2287                         GOTO(out, rc);
2288
2289                 dp = page_address(page);
2290                 for (ent = lu_dirent_start(dp); ent != NULL;
2291                      ent = lu_dirent_next(ent)) {
2292                         /* Skip dummy entry */
2293                         if (le16_to_cpu(ent->lde_namelen) == 0)
2294                                 continue;
2295
2296                         if (le64_to_cpu(ent->lde_hash) < hash_offset)
2297                                 continue;
2298
2299                         if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2300                             (*entp == ent || i < *stripe_offset))
2301                                 continue;
2302
2303                         /* skip . and .. for other stripes */
2304                         if (i != 0 &&
2305                             (strncmp(ent->lde_name, ".",
2306                                      le16_to_cpu(ent->lde_namelen)) == 0 ||
2307                              strncmp(ent->lde_name, "..",
2308                                      le16_to_cpu(ent->lde_namelen)) == 0))
2309                                 continue;
2310                         break;
2311                 }
2312
2313                 if (ent == NULL) {
2314                         stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2315
2316                         kunmap(page);
2317                         page_cache_release(page);
2318                         page = NULL;
2319
2320                         /* reach the end of current stripe, go to next stripe */
2321                         if (stripe_hash == MDS_DIR_END_OFF)
2322                                 continue;
2323                         else
2324                                 goto next;
2325                 }
2326
2327                 if (min_ent != NULL) {
2328                         if (le64_to_cpu(min_ent->lde_hash) >
2329                             le64_to_cpu(ent->lde_hash)) {
2330                                 min_ent = ent;
2331                                 kunmap(min_page);
2332                                 page_cache_release(min_page);
2333                                 min_idx = i;
2334                                 min_page = page;
2335                         } else {
2336                                 kunmap(page);
2337                                 page_cache_release(page);
2338                                 page = NULL;
2339                         }
2340                 } else {
2341                         min_ent = ent;
2342                         min_page = page;
2343                         min_idx = i;
2344                 }
2345         }
2346
2347 out:
2348         if (*ppage != NULL) {
2349                 kunmap(*ppage);
2350                 page_cache_release(*ppage);
2351         }
2352         *stripe_offset = min_idx;
2353         *entp = min_ent;
2354         *ppage = min_page;
2355         RETURN(rc);
2356 }
2357
2358 /**
2359  * Build dir entry page from a striped directory
2360  *
2361  * This function gets one entry by @offset from a striped directory. It will
2362  * read entries from all of stripes, and choose one closest to the required
2363  * offset(&offset). A few notes
2364  * 1. skip . and .. for non-zero stripes, because there can only have one .
2365  * and .. in a directory.
2366  * 2. op_data will be shared by all of stripes, instead of allocating new
2367  * one, so need to restore before reusing.
2368  * 3. release the entry page if that is not being chosen.
2369  *
2370  * \param[in] exp       obd export refer to LMV
2371  * \param[in] op_data   hold those MD parameters of read_entry
2372  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2373  * \param[out] ldp      the entry being read
2374  * \param[out] ppage    the page holding the entry. Note: because the entry
2375  *                      will be accessed in upper layer, so we need hold the
2376  *                      page until the usages of entry is finished, see
2377  *                      ll_dir_entry_next.
2378  *
2379  * retval               =0 if get entry successfully
2380  *                      <0 cannot get entry
2381  */
2382 static int lmv_read_striped_page(struct obd_export *exp,
2383                                  struct md_op_data *op_data,
2384                                  struct md_callback *cb_op,
2385                                  __u64 offset, struct page **ppage)
2386 {
2387         struct obd_device       *obd = exp->exp_obd;
2388         struct lu_fid           master_fid = op_data->op_fid1;
2389         struct inode            *master_inode = op_data->op_data;
2390         __u64                   hash_offset = offset;
2391         struct lu_dirpage       *dp;
2392         struct page             *min_ent_page = NULL;
2393         struct page             *ent_page = NULL;
2394         struct lu_dirent        *ent;
2395         void                    *area;
2396         int                     ent_idx = 0;
2397         struct lu_dirent        *min_ent = NULL;
2398         struct lu_dirent        *last_ent;
2399         size_t                  left_bytes;
2400         int                     rc;
2401         ENTRY;
2402
2403         rc = lmv_check_connect(obd);
2404         if (rc)
2405                 RETURN(rc);
2406
2407         /* Allocate a page and read entries from all of stripes and fill
2408          * the page by hash order */
2409         ent_page = alloc_page(GFP_KERNEL);
2410         if (ent_page == NULL)
2411                 RETURN(-ENOMEM);
2412
2413         /* Initialize the entry page */
2414         dp = kmap(ent_page);
2415         memset(dp, 0, sizeof(*dp));
2416         dp->ldp_hash_start = cpu_to_le64(offset);
2417         dp->ldp_flags |= LDF_COLLIDE;
2418
2419         area = dp + 1;
2420         left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2421         ent = area;
2422         last_ent = ent;
2423         do {
2424                 __u16   ent_size;
2425
2426                 /* Find the minum entry from all sub-stripes */
2427                 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2428                                                &ent_idx, &min_ent,
2429                                                &min_ent_page);
2430                 if (rc != 0)
2431                         GOTO(out, rc);
2432
2433                 /* If it can not get minum entry, it means it already reaches
2434                  * the end of this directory */
2435                 if (min_ent == NULL) {
2436                         last_ent->lde_reclen = 0;
2437                         hash_offset = MDS_DIR_END_OFF;
2438                         GOTO(out, rc);
2439                 }
2440
2441                 ent_size = le16_to_cpu(min_ent->lde_reclen);
2442
2443                 /* the last entry lde_reclen is 0, but it might not
2444                  * the end of this entry of this temporay entry */
2445                 if (ent_size == 0)
2446                         ent_size = lu_dirent_calc_size(
2447                                         le16_to_cpu(min_ent->lde_namelen),
2448                                         le32_to_cpu(min_ent->lde_attrs));
2449                 if (ent_size > left_bytes) {
2450                         last_ent->lde_reclen = cpu_to_le16(0);
2451                         hash_offset = le64_to_cpu(min_ent->lde_hash);
2452                         GOTO(out, rc);
2453                 }
2454
2455                 memcpy(ent, min_ent, ent_size);
2456
2457                 /* Replace . with master FID and Replace .. with the parent FID
2458                  * of master object */
2459                 if (strncmp(ent->lde_name, ".",
2460                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2461                     le16_to_cpu(ent->lde_namelen) == 1)
2462                         fid_cpu_to_le(&ent->lde_fid, &master_fid);
2463                 else if (strncmp(ent->lde_name, "..",
2464                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2465                            le16_to_cpu(ent->lde_namelen) == 2)
2466                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2467
2468                 left_bytes -= ent_size;
2469                 ent->lde_reclen = cpu_to_le16(ent_size);
2470                 last_ent = ent;
2471                 ent = (void *)ent + ent_size;
2472                 hash_offset = le64_to_cpu(min_ent->lde_hash);
2473                 if (hash_offset == MDS_DIR_END_OFF) {
2474                         last_ent->lde_reclen = 0;
2475                         break;
2476                 }
2477         } while (1);
2478 out:
2479         if (min_ent_page != NULL) {
2480                 kunmap(min_ent_page);
2481                 page_cache_release(min_ent_page);
2482         }
2483
2484         if (unlikely(rc != 0)) {
2485                 __free_page(ent_page);
2486                 ent_page = NULL;
2487         } else {
2488                 if (ent == area)
2489                         dp->ldp_flags |= LDF_EMPTY;
2490                 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2491                 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2492         }
2493
2494         /* We do not want to allocate md_op_data during each
2495          * dir entry reading, so op_data will be shared by every stripe,
2496          * then we need to restore it back to original value before
2497          * return to the upper layer */
2498         op_data->op_fid1 = master_fid;
2499         op_data->op_fid2 = master_fid;
2500         op_data->op_data = master_inode;
2501
2502         *ppage = ent_page;
2503
2504         RETURN(rc);
2505 }
2506
2507 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2508                   struct md_callback *cb_op, __u64 offset,
2509                   struct page **ppage)
2510 {
2511         struct obd_device       *obd = exp->exp_obd;
2512         struct lmv_obd          *lmv = &obd->u.lmv;
2513         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2514         struct lmv_tgt_desc     *tgt;
2515         int                     rc;
2516         ENTRY;
2517
2518         rc = lmv_check_connect(obd);
2519         if (rc != 0)
2520                 RETURN(rc);
2521
2522         if (unlikely(lsm != NULL)) {
2523                 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2524                 RETURN(rc);
2525         }
2526
2527         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2528         if (IS_ERR(tgt))
2529                 RETURN(PTR_ERR(tgt));
2530
2531         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2532
2533         RETURN(rc);
2534 }
2535
2536 /**
2537  * Unlink a file/directory
2538  *
2539  * Unlink a file or directory under the parent dir. The unlink request
2540  * usually will be sent to the MDT where the child is located, but if
2541  * the client does not have the child FID then request will be sent to the
2542  * MDT where the parent is located.
2543  *
2544  * If the parent is a striped directory then it also needs to locate which
2545  * stripe the name of the child is located, and replace the parent FID
2546  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2547  * it will walk through all of sub-stripes until the child is being
2548  * unlinked finally.
2549  *
2550  * \param[in] exp       export refer to LMV
2551  * \param[in] op_data   different parameters transferred beween client
2552  *                      MD stacks, name, namelen, FIDs etc.
2553  *                      op_fid1 is the parent FID, op_fid2 is the child
2554  *                      FID.
2555  * \param[out] request  point to the request of unlink.
2556  *
2557  * retval               0 if succeed
2558  *                      negative errno if failed.
2559  */
2560 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2561                       struct ptlrpc_request **request)
2562 {
2563         struct obd_device       *obd = exp->exp_obd;
2564         struct lmv_obd          *lmv = &obd->u.lmv;
2565         struct lmv_tgt_desc     *tgt = NULL;
2566         struct lmv_tgt_desc     *parent_tgt = NULL;
2567         struct mdt_body         *body;
2568         int                     rc;
2569         int                     stripe_index = 0;
2570         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2571         ENTRY;
2572
2573         rc = lmv_check_connect(obd);
2574         if (rc)
2575                 RETURN(rc);
2576 retry_unlink:
2577         /* For striped dir, we need to locate the parent as well */
2578         if (lsm != NULL) {
2579                 struct lmv_tgt_desc *tmp;
2580
2581                 LASSERT(op_data->op_name != NULL &&
2582                         op_data->op_namelen != 0);
2583
2584                 tmp = lmv_locate_target_for_name(lmv, lsm,
2585                                                  op_data->op_name,
2586                                                  op_data->op_namelen,
2587                                                  &op_data->op_fid1,
2588                                                  &op_data->op_mds);
2589
2590                 /* return -EBADFD means unknown hash type, might
2591                  * need try all sub-stripe here */
2592                 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2593                         RETURN(PTR_ERR(tmp));
2594
2595                 /* Note: both migrating dir and unknown hash dir need to
2596                  * try all of sub-stripes, so we need start search the
2597                  * name from stripe 0, but migrating dir is already handled
2598                  * inside lmv_locate_target_for_name(), so we only check
2599                  * unknown hash type directory here */
2600                 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2601                         struct lmv_oinfo *oinfo;
2602
2603                         oinfo = &lsm->lsm_md_oinfo[stripe_index];
2604
2605                         op_data->op_fid1 = oinfo->lmo_fid;
2606                         op_data->op_mds = oinfo->lmo_mds;
2607                 }
2608         }
2609
2610 try_next_stripe:
2611         /* Send unlink requests to the MDT where the child is located */
2612         if (likely(!fid_is_zero(&op_data->op_fid2)))
2613                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2614         else if (lsm != NULL)
2615                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2616         else
2617                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2618
2619         if (IS_ERR(tgt))
2620                 RETURN(PTR_ERR(tgt));
2621
2622         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2623         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2624         op_data->op_cap = cfs_curproc_cap_pack();
2625
2626         /*
2627          * If child's fid is given, cancel unused locks for it if it is from
2628          * another export than parent.
2629          *
2630          * LOOKUP lock for child (fid3) should also be cancelled on parent
2631          * tgt_tgt in mdc_unlink().
2632          */
2633         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2634
2635         /*
2636          * Cancel FULL locks on child (fid3).
2637          */
2638         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2639         if (IS_ERR(parent_tgt))
2640                 RETURN(PTR_ERR(parent_tgt));
2641
2642         if (parent_tgt != tgt) {
2643                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2644                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2645                                       MF_MDC_CANCEL_FID3);
2646         }
2647
2648         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2649                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2650         if (rc != 0)
2651                 RETURN(rc);
2652
2653         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2654                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2655
2656         rc = md_unlink(tgt->ltd_exp, op_data, request);
2657         if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2658                 RETURN(rc);
2659
2660         /* Try next stripe if it is needed. */
2661         if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2662                 struct lmv_oinfo *oinfo;
2663
2664                 stripe_index++;
2665                 if (stripe_index >= lsm->lsm_md_stripe_count)
2666                         RETURN(rc);
2667
2668                 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2669
2670                 op_data->op_fid1 = oinfo->lmo_fid;
2671                 op_data->op_mds = oinfo->lmo_mds;
2672
2673                 ptlrpc_req_finished(*request);
2674                 *request = NULL;
2675
2676                 goto try_next_stripe;
2677         }
2678
2679         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2680         if (body == NULL)
2681                 RETURN(-EPROTO);
2682
2683         /* Not cross-ref case, just get out of here. */
2684         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2685                 RETURN(rc);
2686
2687         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2688                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2689
2690         /* This is a remote object, try remote MDT, Note: it may
2691          * try more than 1 time here, Considering following case
2692          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2693          * 1. Initially A does not know where remote1 is, it send
2694          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2695          *    resend unlink RPC to MDT1 (retry 1st time).
2696          *
2697          * 2. During the unlink RPC in flight,
2698          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2699          *    and create new remote1, but on MDT0
2700          *
2701          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2702          *    /mnt/lustre, then lookup get fid of remote1, and find
2703          *    it is remote dir again, and replay -EREMOTE again.
2704          *
2705          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2706          *
2707          * In theory, it might try unlimited time here, but it should
2708          * be very rare case.  */
2709         op_data->op_fid2 = body->mbo_fid1;
2710         ptlrpc_req_finished(*request);
2711         *request = NULL;
2712
2713         goto retry_unlink;
2714 }
2715
2716 static int lmv_precleanup(struct obd_device *obd)
2717 {
2718         ENTRY;
2719         fld_client_proc_fini(&obd->u.lmv.lmv_fld);
2720         lprocfs_obd_cleanup(obd);
2721         lprocfs_free_md_stats(obd);
2722         RETURN(0);
2723 }
2724
2725 /**
2726  * Get by key a value associated with a LMV device.
2727  *
2728  * Dispatch request to lower-layer devices as needed.
2729  *
2730  * \param[in] env               execution environment for this thread
2731  * \param[in] exp               export for the LMV device
2732  * \param[in] keylen            length of key identifier
2733  * \param[in] key               identifier of key to get value for
2734  * \param[in] vallen            size of \a val
2735  * \param[out] val              pointer to storage location for value
2736  * \param[in] lsm               optional striping metadata of object
2737  *
2738  * \retval 0            on success
2739  * \retval negative     negated errno on failure
2740  */
2741 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2742                         __u32 keylen, void *key, __u32 *vallen, void *val)
2743 {
2744         struct obd_device       *obd;
2745         struct lmv_obd          *lmv;
2746         int                      rc = 0;
2747         ENTRY;
2748
2749         obd = class_exp2obd(exp);
2750         if (obd == NULL) {
2751                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2752                        exp->exp_handle.h_cookie);
2753                 RETURN(-EINVAL);
2754         }
2755
2756         lmv = &obd->u.lmv;
2757         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2758                 int i;
2759
2760                 rc = lmv_check_connect(obd);
2761                 if (rc)
2762                         RETURN(rc);
2763
2764                 LASSERT(*vallen == sizeof(__u32));
2765                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2766                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2767                         /*
2768                          * All tgts should be connected when this gets called.
2769                          */
2770                         if (tgt == NULL || tgt->ltd_exp == NULL)
2771                                 continue;
2772
2773                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2774                                           vallen, val))
2775                                 RETURN(0);
2776                 }
2777                 RETURN(-EINVAL);
2778         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2779                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2780                    KEY_IS(KEY_CONN_DATA)) {
2781                 rc = lmv_check_connect(obd);
2782                 if (rc)
2783                         RETURN(rc);
2784
2785                 /*
2786                  * Forwarding this request to first MDS, it should know LOV
2787                  * desc.
2788                  */
2789                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2790                                   vallen, val);
2791                 if (!rc && KEY_IS(KEY_CONN_DATA))
2792                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2793                 RETURN(rc);
2794         } else if (KEY_IS(KEY_TGT_COUNT)) {
2795                 *((int *)val) = lmv->desc.ld_tgt_count;
2796                 RETURN(0);
2797         }
2798
2799         CDEBUG(D_IOCTL, "Invalid key\n");
2800         RETURN(-EINVAL);
2801 }
2802
2803 /**
2804  * Asynchronously set by key a value associated with a LMV device.
2805  *
2806  * Dispatch request to lower-layer devices as needed.
2807  *
2808  * \param[in] env       execution environment for this thread
2809  * \param[in] exp       export for the LMV device
2810  * \param[in] keylen    length of key identifier
2811  * \param[in] key       identifier of key to store value for
2812  * \param[in] vallen    size of value to store
2813  * \param[in] val       pointer to data to be stored
2814  * \param[in] set       optional list of related ptlrpc requests
2815  *
2816  * \retval 0            on success
2817  * \retval negative     negated errno on failure
2818  */
2819 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2820                         __u32 keylen, void *key, __u32 vallen, void *val,
2821                         struct ptlrpc_request_set *set)
2822 {
2823         struct lmv_tgt_desc     *tgt = NULL;
2824         struct obd_device       *obd;
2825         struct lmv_obd          *lmv;
2826         int rc = 0;
2827         ENTRY;
2828
2829         obd = class_exp2obd(exp);
2830         if (obd == NULL) {
2831                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2832                        exp->exp_handle.h_cookie);
2833                 RETURN(-EINVAL);
2834         }
2835         lmv = &obd->u.lmv;
2836
2837         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2838             KEY_IS(KEY_DEFAULT_EASIZE)) {
2839                 int i, err = 0;
2840
2841                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2842                         tgt = lmv->tgts[i];
2843
2844                         if (tgt == NULL || tgt->ltd_exp == NULL)
2845                                 continue;
2846
2847                         err = obd_set_info_async(env, tgt->ltd_exp,
2848                                                  keylen, key, vallen, val, set);
2849                         if (err && rc == 0)
2850                                 rc = err;
2851                 }
2852
2853                 RETURN(rc);
2854         }
2855
2856         RETURN(-EINVAL);
2857 }
2858
2859 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2860                             const struct lmv_mds_md_v1 *lmm1)
2861 {
2862         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
2863         int             stripe_count;
2864         int             cplen;
2865         int             i;
2866         int             rc = 0;
2867         ENTRY;
2868
2869         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2870         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2871         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2872         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2873                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2874         else
2875                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2876         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2877         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2878                         sizeof(lsm->lsm_md_pool_name));
2879
2880         if (cplen >= sizeof(lsm->lsm_md_pool_name))
2881                 RETURN(-E2BIG);
2882
2883         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2884                "layout_version %d\n", lsm->lsm_md_stripe_count,
2885                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2886                lsm->lsm_md_layout_version);
2887
2888         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2889         for (i = 0; i < stripe_count; i++) {
2890                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2891                               &lmm1->lmv_stripe_fids[i]);
2892                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2893                                     &lsm->lsm_md_oinfo[i].lmo_mds);
2894                 if (rc != 0)
2895                         RETURN(rc);
2896                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2897                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2898         }
2899
2900         RETURN(rc);
2901 }
2902
2903 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2904                         const union lmv_mds_md *lmm, size_t lmm_size)
2905 {
2906         struct lmv_stripe_md     *lsm;
2907         int                      lsm_size;
2908         int                      rc;
2909         bool                     allocated = false;
2910         ENTRY;
2911
2912         LASSERT(lsmp != NULL);
2913
2914         lsm = *lsmp;
2915         /* Free memmd */
2916         if (lsm != NULL && lmm == NULL) {
2917                 int i;
2918                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2919                         /* For migrating inode, the master stripe and master
2920                          * object will be the same, so do not need iput, see
2921                          * ll_update_lsm_md */
2922                         if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2923                               i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2924                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
2925                 }
2926                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2927                 OBD_FREE(lsm, lsm_size);
2928                 *lsmp = NULL;
2929                 RETURN(0);
2930         }
2931
2932         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
2933                 RETURN(-EPERM);
2934
2935         /* Unpack memmd */
2936         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
2937             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
2938                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
2939                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
2940                        -EIO);
2941                 RETURN(-EIO);
2942         }
2943
2944         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
2945                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2946         else
2947                 /**
2948                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
2949                  * stripecount should be 0 then.
2950                  */
2951                 lsm_size = lmv_stripe_md_size(0);
2952
2953         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2954         if (lsm == NULL) {
2955                 OBD_ALLOC(lsm, lsm_size);
2956                 if (lsm == NULL)
2957                         RETURN(-ENOMEM);
2958                 allocated = true;
2959                 *lsmp = lsm;
2960         }
2961
2962         switch (le32_to_cpu(lmm->lmv_magic)) {
2963         case LMV_MAGIC_V1:
2964                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
2965                 break;
2966         default:
2967                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
2968                        le32_to_cpu(lmm->lmv_magic));
2969                 rc = -EINVAL;
2970                 break;
2971         }
2972
2973         if (rc != 0 && allocated) {
2974                 OBD_FREE(lsm, lsm_size);
2975                 *lsmp = NULL;
2976                 lsm_size = rc;
2977         }
2978         RETURN(lsm_size);
2979 }
2980
2981 void lmv_free_memmd(struct lmv_stripe_md *lsm)
2982 {
2983         lmv_unpackmd(NULL, &lsm, NULL, 0);
2984 }
2985 EXPORT_SYMBOL(lmv_free_memmd);
2986
2987 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2988                              union ldlm_policy_data *policy,
2989                              enum ldlm_mode mode, enum ldlm_cancel_flags flags,
2990                              void *opaque)
2991 {
2992         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2993         int rc = 0;
2994         __u32 i;
2995         ENTRY;
2996
2997         LASSERT(fid != NULL);
2998
2999         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3000                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3001                 int err;
3002
3003                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3004                         continue;
3005
3006                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3007                                        opaque);
3008                 if (!rc)
3009                         rc = err;
3010         }
3011         RETURN(rc);
3012 }
3013
3014 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3015                       __u64 *bits)
3016 {
3017         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3018         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3019         int                      rc;
3020         ENTRY;
3021
3022         if (tgt == NULL || tgt->ltd_exp == NULL)
3023                 RETURN(-EINVAL);
3024         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3025         RETURN(rc);
3026 }
3027
3028 enum ldlm_mode lmv_lock_match(struct obd_export *exp, __u64 flags,
3029                               const struct lu_fid *fid, enum ldlm_type type,
3030                               union ldlm_policy_data *policy,
3031                               enum ldlm_mode mode, struct lustre_handle *lockh)
3032 {
3033         struct obd_device       *obd = exp->exp_obd;
3034         struct lmv_obd          *lmv = &obd->u.lmv;
3035         enum ldlm_mode          rc;
3036         int                     tgt;
3037         int                     i;
3038         ENTRY;
3039
3040         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3041
3042         /*
3043          * With DNE every object can have two locks in different namespaces:
3044          * lookup lock in space of MDT storing direntry and update/open lock in
3045          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3046          * since this can be easily found, and only try others if that fails.
3047          */
3048         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3049              i < lmv->desc.ld_tgt_count;
3050              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3051                 if (tgt < 0) {
3052                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3053                                obd->obd_name, PFID(fid), tgt);
3054                         tgt = 0;
3055                 }
3056
3057                 if (lmv->tgts[tgt] == NULL ||
3058                     lmv->tgts[tgt]->ltd_exp == NULL ||
3059                     lmv->tgts[tgt]->ltd_active == 0)
3060                         continue;
3061
3062                 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3063                                    type, policy, mode, lockh);
3064                 if (rc)
3065                         RETURN(rc);
3066         }
3067
3068         RETURN(0);
3069 }
3070
3071 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3072                       struct obd_export *dt_exp, struct obd_export *md_exp,
3073                       struct lustre_md *md)
3074 {
3075         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3076         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3077
3078         if (tgt == NULL || tgt->ltd_exp == NULL)
3079                 RETURN(-EINVAL);
3080
3081         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3082 }
3083
3084 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3085 {
3086         struct obd_device       *obd = exp->exp_obd;
3087         struct lmv_obd          *lmv = &obd->u.lmv;
3088         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3089         ENTRY;
3090
3091         if (md->lmv != NULL) {
3092                 lmv_free_memmd(md->lmv);
3093                 md->lmv = NULL;
3094         }
3095         if (tgt == NULL || tgt->ltd_exp == NULL)
3096                 RETURN(-EINVAL);
3097         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3098 }
3099
3100 int lmv_set_open_replay_data(struct obd_export *exp,
3101                              struct obd_client_handle *och,
3102                              struct lookup_intent *it)
3103 {
3104         struct obd_device       *obd = exp->exp_obd;
3105         struct lmv_obd          *lmv = &obd->u.lmv;
3106         struct lmv_tgt_desc     *tgt;
3107         ENTRY;
3108
3109         tgt = lmv_find_target(lmv, &och->och_fid);
3110         if (IS_ERR(tgt))
3111                 RETURN(PTR_ERR(tgt));
3112
3113         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3114 }
3115
3116 int lmv_clear_open_replay_data(struct obd_export *exp,
3117                                struct obd_client_handle *och)
3118 {
3119         struct obd_device       *obd = exp->exp_obd;
3120         struct lmv_obd          *lmv = &obd->u.lmv;
3121         struct lmv_tgt_desc     *tgt;
3122         ENTRY;
3123
3124         tgt = lmv_find_target(lmv, &och->och_fid);
3125         if (IS_ERR(tgt))
3126                 RETURN(PTR_ERR(tgt));
3127
3128         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3129 }
3130
3131 static int lmv_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
3132                                u32 suppgid, struct ptlrpc_request **request)
3133 {
3134         struct obd_device       *obd = exp->exp_obd;
3135         struct lmv_obd          *lmv = &obd->u.lmv;
3136         struct lmv_tgt_desc     *tgt;
3137         int                      rc;
3138         ENTRY;
3139
3140         rc = lmv_check_connect(obd);
3141         if (rc)
3142                 RETURN(rc);
3143
3144         tgt = lmv_find_target(lmv, fid);
3145         if (IS_ERR(tgt))
3146                 RETURN(PTR_ERR(tgt));
3147
3148         rc = md_get_remote_perm(tgt->ltd_exp, fid, suppgid, request);
3149         RETURN(rc);
3150 }
3151
3152 int lmv_intent_getattr_async(struct obd_export *exp,
3153                              struct md_enqueue_info *minfo)
3154 {
3155         struct md_op_data       *op_data = &minfo->mi_data;
3156         struct obd_device       *obd = exp->exp_obd;
3157         struct lmv_obd          *lmv = &obd->u.lmv;
3158         struct lmv_tgt_desc     *ptgt = NULL;
3159         struct lmv_tgt_desc     *ctgt = NULL;
3160         int                      rc;
3161         ENTRY;
3162
3163         if (!fid_is_sane(&op_data->op_fid2))
3164                 RETURN(-EINVAL);
3165
3166         rc = lmv_check_connect(obd);
3167         if (rc)
3168                 RETURN(rc);
3169
3170         ptgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3171         if (IS_ERR(ptgt))
3172                 RETURN(PTR_ERR(ptgt));
3173
3174         ctgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
3175         if (IS_ERR(ctgt))
3176                 RETURN(PTR_ERR(ctgt));
3177
3178         /*
3179          * if child is on remote MDT, we need 2 async RPCs to fetch both LOOKUP
3180          * lock on parent, and UPDATE lock on child MDT, which makes all
3181          * complicated. Considering remote dir is rare case, and not supporting
3182          * it in statahead won't cause any issue, drop its support for now.
3183          */
3184         if (ptgt != ctgt)
3185                 RETURN(-ENOTSUPP);
3186
3187         rc = md_intent_getattr_async(ptgt->ltd_exp, minfo);
3188         RETURN(rc);
3189 }
3190
3191 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3192                         struct lu_fid *fid, __u64 *bits)
3193 {
3194         struct obd_device       *obd = exp->exp_obd;
3195         struct lmv_obd          *lmv = &obd->u.lmv;
3196         struct lmv_tgt_desc     *tgt;
3197         int                      rc;
3198         ENTRY;
3199
3200         rc = lmv_check_connect(obd);
3201         if (rc)
3202                 RETURN(rc);
3203
3204         tgt = lmv_find_target(lmv, fid);
3205         if (IS_ERR(tgt))
3206                 RETURN(PTR_ERR(tgt));
3207
3208         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3209         RETURN(rc);
3210 }
3211
3212 int lmv_get_fid_from_lsm(struct obd_export *exp,
3213                          const struct lmv_stripe_md *lsm,
3214                          const char *name, int namelen, struct lu_fid *fid)
3215 {
3216         const struct lmv_oinfo *oinfo;
3217
3218         LASSERT(lsm != NULL);
3219         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3220         if (IS_ERR(oinfo))
3221                 return PTR_ERR(oinfo);
3222
3223         *fid = oinfo->lmo_fid;
3224
3225         RETURN(0);
3226 }
3227
3228 /**
3229  * For lmv, only need to send request to master MDT, and the master MDT will
3230  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3231  * we directly fetch data from the slave MDTs.
3232  */
3233 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3234                  struct obd_quotactl *oqctl)
3235 {
3236         struct obd_device   *obd = class_exp2obd(exp);
3237         struct lmv_obd      *lmv = &obd->u.lmv;
3238         struct lmv_tgt_desc *tgt = lmv->tgts[0];
3239         int                  rc = 0;
3240         __u32                i;
3241         __u64                curspace, curinodes;
3242         ENTRY;
3243
3244         if (tgt == NULL ||
3245             tgt->ltd_exp == NULL ||
3246             !tgt->ltd_active ||
3247             lmv->desc.ld_tgt_count == 0) {
3248                 CERROR("master lmv inactive\n");
3249                 RETURN(-EIO);
3250         }
3251
3252         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3253                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3254                 RETURN(rc);
3255         }
3256
3257         curspace = curinodes = 0;
3258         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3259                 int err;
3260                 tgt = lmv->tgts[i];
3261
3262                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3263                         continue;
3264
3265                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3266                 if (err) {
3267                         CERROR("getquota on mdt %d failed. %d\n", i, err);
3268                         if (!rc)
3269                                 rc = err;
3270                 } else {
3271                         curspace += oqctl->qc_dqblk.dqb_curspace;
3272                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3273                 }
3274         }
3275         oqctl->qc_dqblk.dqb_curspace = curspace;
3276         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3277
3278         RETURN(rc);
3279 }
3280
3281 static int lmv_merge_attr(struct obd_export *exp,
3282                           const struct lmv_stripe_md *lsm,
3283                           struct cl_attr *attr,
3284                           ldlm_blocking_callback cb_blocking)
3285 {
3286         int rc;
3287         int i;
3288
3289         rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3290         if (rc < 0)
3291                 return rc;
3292
3293         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3294                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3295
3296                 CDEBUG(D_INFO, ""DFID" size %llu, blocks %llu nlink %u,"
3297                        " atime %lu ctime %lu, mtime %lu.\n",
3298                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3299                        i_size_read(inode), (unsigned long long)inode->i_blocks,
3300                        inode->i_nlink, LTIME_S(inode->i_atime),
3301                        LTIME_S(inode->i_ctime), LTIME_S(inode->i_mtime));
3302
3303                 /* for slave stripe, it needs to subtract nlink for . and .. */
3304                 if (i != 0)
3305                         attr->cat_nlink += inode->i_nlink - 2;
3306                 else
3307                         attr->cat_nlink = inode->i_nlink;
3308
3309                 attr->cat_size += i_size_read(inode);
3310                 attr->cat_blocks += inode->i_blocks;
3311
3312                 if (attr->cat_atime < LTIME_S(inode->i_atime))
3313                         attr->cat_atime = LTIME_S(inode->i_atime);
3314
3315                 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3316                         attr->cat_ctime = LTIME_S(inode->i_ctime);
3317
3318                 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3319                         attr->cat_mtime = LTIME_S(inode->i_mtime);
3320         }
3321         return 0;
3322 }
3323
3324 struct obd_ops lmv_obd_ops = {
3325         .o_owner                = THIS_MODULE,
3326         .o_setup                = lmv_setup,
3327         .o_cleanup              = lmv_cleanup,
3328         .o_precleanup           = lmv_precleanup,
3329         .o_process_config       = lmv_process_config,
3330         .o_connect              = lmv_connect,
3331         .o_disconnect           = lmv_disconnect,
3332         .o_statfs               = lmv_statfs,
3333         .o_get_info             = lmv_get_info,
3334         .o_set_info_async       = lmv_set_info_async,
3335         .o_notify               = lmv_notify,
3336         .o_get_uuid             = lmv_get_uuid,
3337         .o_iocontrol            = lmv_iocontrol,
3338         .o_quotactl             = lmv_quotactl
3339 };
3340
3341 struct md_ops lmv_md_ops = {
3342         .m_get_root             = lmv_get_root,
3343         .m_null_inode           = lmv_null_inode,
3344         .m_close                = lmv_close,
3345         .m_create               = lmv_create,
3346         .m_enqueue              = lmv_enqueue,
3347         .m_getattr              = lmv_getattr,
3348         .m_getxattr             = lmv_getxattr,
3349         .m_getattr_name         = lmv_getattr_name,
3350         .m_intent_lock          = lmv_intent_lock,
3351         .m_link                 = lmv_link,
3352         .m_rename               = lmv_rename,
3353         .m_setattr              = lmv_setattr,
3354         .m_setxattr             = lmv_setxattr,
3355         .m_fsync                = lmv_fsync,
3356         .m_read_page            = lmv_read_page,
3357         .m_unlink               = lmv_unlink,
3358         .m_init_ea_size         = lmv_init_ea_size,
3359         .m_cancel_unused        = lmv_cancel_unused,
3360         .m_set_lock_data        = lmv_set_lock_data,
3361         .m_lock_match           = lmv_lock_match,
3362         .m_get_lustre_md        = lmv_get_lustre_md,
3363         .m_free_lustre_md       = lmv_free_lustre_md,
3364         .m_merge_attr           = lmv_merge_attr,
3365         .m_set_open_replay_data = lmv_set_open_replay_data,
3366         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3367         .m_get_remote_perm      = lmv_get_remote_perm,
3368         .m_intent_getattr_async = lmv_intent_getattr_async,
3369         .m_revalidate_lock      = lmv_revalidate_lock,
3370         .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
3371         .m_unpackmd             = lmv_unpackmd,
3372 };
3373
3374 static int __init lmv_init(void)
3375 {
3376         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3377                                    LUSTRE_LMV_NAME, NULL);
3378 }
3379
3380 static void __exit lmv_exit(void)
3381 {
3382         class_unregister_type(LUSTRE_LMV_NAME);
3383 }
3384
3385 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3386 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
3387 MODULE_VERSION(LUSTRE_VERSION_STRING);
3388 MODULE_LICENSE("GPL");
3389
3390 module_init(lmv_init);
3391 module_exit(lmv_exit);