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