lustre/lov/lov_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) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, 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  * lustre/lov/lov_obd.c
  37  *
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Peter Braam <braam@clusterfs.com>
  40  * Author: Mike Shaver <shaver@clusterfs.com>
  41  * Author: Nathan Rutman <nathan@clusterfs.com>
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_LOV
  45 #ifdef __KERNEL__
  46 #include <libcfs/libcfs.h>
  47 #else
  48 #include <liblustre.h>
  49 #endif
  50
  51 #include <obd_support.h>
  52 #include <lustre_lib.h>
  53 #include <lustre_net.h>
  54 #include <lustre/lustre_idl.h>
  55 #include <lustre_dlm.h>
  56 #include <lustre_mds.h>
  57 #include <lustre_debug.h>
  58 #include <obd_class.h>
  59 #include <obd_lov.h>
  60 #include <obd_ost.h>
  61 #include <lprocfs_status.h>
  62 #include <lustre_param.h>
  63 #include <cl_object.h>
  64 #include <lclient.h> /* for cl_client_lru */
  65 #include <lustre/ll_fiemap.h>
  66 #include <lustre_log.h>
  67 #include <lustre_fid.h>
  68
  69 #include "lov_internal.h"
  70
  71 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
  72    Any function that expects lov_tgts to remain stationary must take a ref. */
  73 static void lov_getref(struct obd_device *obd)
  74 {
  75         struct lov_obd *lov = &obd->u.lov;
  76
  77         /* nobody gets through here until lov_putref is done */
  78         mutex_lock(&lov->lov_lock);
  79         cfs_atomic_inc(&lov->lov_refcount);
  80         mutex_unlock(&lov->lov_lock);
  81         return;
  82 }
  83
  84 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
  85
  86 static void lov_putref(struct obd_device *obd)
  87 {
  88         struct lov_obd *lov = &obd->u.lov;
  89
  90         mutex_lock(&lov->lov_lock);
  91         /* ok to dec to 0 more than once -- ltd_exp's will be null */
  92         if (cfs_atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
  93                 CFS_LIST_HEAD(kill);
  94                 int i;
  95                 struct lov_tgt_desc *tgt, *n;
  96                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
  97                        lov->lov_death_row);
  98                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
  99                         tgt = lov->lov_tgts[i];
 100
 101                         if (!tgt || !tgt->ltd_reap)
 102                                 continue;
 103                         cfs_list_add(&tgt->ltd_kill, &kill);
 104                         /* XXX - right now there is a dependency on ld_tgt_count
 105                          * being the maximum tgt index for computing the
 106                          * mds_max_easize. So we can't shrink it. */
 107                         lov_ost_pool_remove(&lov->lov_packed, i);
 108                         lov->lov_tgts[i] = NULL;
 109                         lov->lov_death_row--;
 110                 }
 111                 mutex_unlock(&lov->lov_lock);
 112
 113                 cfs_list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
 114                         cfs_list_del(&tgt->ltd_kill);
 115                         /* Disconnect */
 116                         __lov_del_obd(obd, tgt);
 117                 }
 118         } else {
 119                 mutex_unlock(&lov->lov_lock);
 120         }
 121 }
 122
 123 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 124                               enum obd_notify_event ev);
 125 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 126                       enum obd_notify_event ev, void *data);
 127
 128
 129 #define MAX_STRING_SIZE 128
 130 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
 131                     struct obd_connect_data *data)
 132 {
 133         struct lov_obd *lov = &obd->u.lov;
 134         struct obd_uuid *tgt_uuid;
 135         struct obd_device *tgt_obd;
 136         static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
 137         struct obd_import *imp;
 138 #ifdef __KERNEL__
 139         cfs_proc_dir_entry_t *lov_proc_dir;
 140 #endif
 141         int rc;
 142         ENTRY;
 143
 144         if (!lov->lov_tgts[index])
 145                 RETURN(-EINVAL);
 146
 147         tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
 148         tgt_obd = lov->lov_tgts[index]->ltd_obd;
 149
 150         if (!tgt_obd->obd_set_up) {
 151                 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
 152                 RETURN(-EINVAL);
 153         }
 154
 155         /* override the sp_me from lov */
 156         tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
 157
 158         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
 159                 data->ocd_index = index;
 160
 161         /*
 162          * Divine LOV knows that OBDs under it are OSCs.
 163          */
 164         imp = tgt_obd->u.cli.cl_import;
 165
 166         if (activate) {
 167                 tgt_obd->obd_no_recov = 0;
 168                 /* FIXME this is probably supposed to be
 169                    ptlrpc_set_import_active.  Horrible naming. */
 170                 ptlrpc_activate_import(imp);
 171         }
 172
 173         rc = obd_register_observer(tgt_obd, obd);
 174         if (rc) {
 175                 CERROR("Target %s register_observer error %d\n",
 176                        obd_uuid2str(tgt_uuid), rc);
 177                 RETURN(rc);
 178         }
 179
 180
 181         if (imp->imp_invalid) {
 182                 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
 183                        "disabled\n", obd_uuid2str(tgt_uuid));
 184                 RETURN(0);
 185         }
 186
 187         rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
 188                          &lov_osc_uuid, data, NULL);
 189         if (rc || !lov->lov_tgts[index]->ltd_exp) {
 190                 CERROR("Target %s connect error %d\n",
 191                        obd_uuid2str(tgt_uuid), rc);
 192                 RETURN(-ENODEV);
 193         }
 194
 195         lov->lov_tgts[index]->ltd_reap = 0;
 196
 197         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
 198                obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
 199
 200 #ifdef __KERNEL__
 201         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 202         if (lov_proc_dir) {
 203                 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
 204                 cfs_proc_dir_entry_t *osc_symlink;
 205
 206                 LASSERT(osc_obd != NULL);
 207                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
 208                 LASSERT(osc_obd->obd_type->typ_name != NULL);
 209
 210                 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
 211                                                   lov_proc_dir,
 212                                                   "../../../%s/%s",
 213                                                   osc_obd->obd_type->typ_name,
 214                                                   osc_obd->obd_name);
 215                 if (osc_symlink == NULL) {
 216                         CERROR("could not register LOV target "
 217                                 "/proc/fs/lustre/%s/%s/target_obds/%s.",
 218                                 obd->obd_type->typ_name, obd->obd_name,
 219                                 osc_obd->obd_name);
 220                         lprocfs_remove(&lov_proc_dir);
 221                 }
 222         }
 223 #endif
 224
 225         RETURN(0);
 226 }
 227
 228 static int lov_connect(const struct lu_env *env,
 229                        struct obd_export **exp, struct obd_device *obd,
 230                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 231                        void *localdata)
 232 {
 233         struct lov_obd *lov = &obd->u.lov;
 234         struct lov_tgt_desc *tgt;
 235         struct lustre_handle conn;
 236         int i, rc;
 237         ENTRY;
 238
 239         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
 240
 241         rc = class_connect(&conn, obd, cluuid);
 242         if (rc)
 243                 RETURN(rc);
 244
 245         *exp = class_conn2export(&conn);
 246
 247         /* Why should there ever be more than 1 connect? */
 248         lov->lov_connects++;
 249         LASSERT(lov->lov_connects == 1);
 250
 251         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
 252         if (data)
 253                 lov->lov_ocd = *data;
 254
 255         obd_getref(obd);
 256         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 257                 tgt = lov->lov_tgts[i];
 258                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
 259                         continue;
 260                 /* Flags will be lowest common denominator */
 261                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
 262                 if (rc) {
 263                         CERROR("%s: lov connect tgt %d failed: %d\n",
 264                                obd->obd_name, i, rc);
 265                         continue;
 266                 }
 267                 /* connect to administrative disabled ost */
 268                 if (!lov->lov_tgts[i]->ltd_exp)
 269                         continue;
 270
 271                 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
 272                                 OBD_NOTIFY_CONNECT, (void *)&i);
 273                 if (rc) {
 274                         CERROR("%s error sending notify %d\n",
 275                                obd->obd_name, rc);
 276                 }
 277         }
 278         obd_putref(obd);
 279
 280         RETURN(0);
 281 }
 282
 283 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 284 {
 285         cfs_proc_dir_entry_t *lov_proc_dir;
 286         struct lov_obd *lov = &obd->u.lov;
 287         struct obd_device *osc_obd;
 288         int rc;
 289         ENTRY;
 290
 291         osc_obd = class_exp2obd(tgt->ltd_exp);
 292         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
 293                obd->obd_name, osc_obd->obd_name);
 294
 295         if (tgt->ltd_active) {
 296                 tgt->ltd_active = 0;
 297                 lov->desc.ld_active_tgt_count--;
 298                 tgt->ltd_exp->exp_obd->obd_inactive = 1;
 299         }
 300
 301         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 302         if (lov_proc_dir) {
 303                 cfs_proc_dir_entry_t *osc_symlink;
 304
 305                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
 306                 if (osc_symlink) {
 307                         lprocfs_remove(&osc_symlink);
 308                 } else {
 309                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
 310                                obd->obd_type->typ_name, obd->obd_name,
 311                                osc_obd->obd_name);
 312                 }
 313         }
 314
 315         if (osc_obd) {
 316                 /* Pass it on to our clients.
 317                  * XXX This should be an argument to disconnect,
 318                  * XXX not a back-door flag on the OBD.  Ah well.
 319                  */
 320                 osc_obd->obd_force = obd->obd_force;
 321                 osc_obd->obd_fail = obd->obd_fail;
 322                 osc_obd->obd_no_recov = obd->obd_no_recov;
 323         }
 324
 325         obd_register_observer(osc_obd, NULL);
 326
 327         rc = obd_disconnect(tgt->ltd_exp);
 328         if (rc) {
 329                 CERROR("Target %s disconnect error %d\n",
 330                        tgt->ltd_uuid.uuid, rc);
 331                 rc = 0;
 332         }
 333
 334         tgt->ltd_exp = NULL;
 335         RETURN(0);
 336 }
 337
 338 static int lov_disconnect(struct obd_export *exp)
 339 {
 340         struct obd_device *obd = class_exp2obd(exp);
 341         struct lov_obd *lov = &obd->u.lov;
 342         int i, rc;
 343         ENTRY;
 344
 345         if (!lov->lov_tgts)
 346                 goto out;
 347
 348         /* Only disconnect the underlying layers on the final disconnect. */
 349         lov->lov_connects--;
 350         if (lov->lov_connects != 0) {
 351                 /* why should there be more than 1 connect? */
 352                 CERROR("disconnect #%d\n", lov->lov_connects);
 353                 goto out;
 354         }
 355
 356         /* Let's hold another reference so lov_del_obd doesn't spin through
 357            putref every time */
 358         obd_getref(obd);
 359
 360         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 361                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
 362                         /* Disconnection is the last we know about an obd */
 363                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
 364                 }
 365         }
 366         obd_putref(obd);
 367
 368 out:
 369         rc = class_disconnect(exp); /* bz 9811 */
 370         RETURN(rc);
 371 }
 372
 373 /* Error codes:
 374  *
 375  *  -EINVAL  : UUID can't be found in the LOV's target list
 376  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 377  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
 378  *  any >= 0 : is log target index
 379  */
 380 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 381                               enum obd_notify_event ev)
 382 {
 383         struct lov_obd *lov = &obd->u.lov;
 384         struct lov_tgt_desc *tgt;
 385         int index, activate, active;
 386         ENTRY;
 387
 388         CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
 389                lov, uuid->uuid, ev);
 390
 391         obd_getref(obd);
 392         for (index = 0; index < lov->desc.ld_tgt_count; index++) {
 393                 tgt = lov->lov_tgts[index];
 394                 if (!tgt || !tgt->ltd_exp)
 395                         continue;
 396
 397                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
 398                        index, obd_uuid2str(&tgt->ltd_uuid),
 399                        tgt->ltd_exp->exp_handle.h_cookie);
 400                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 401                         break;
 402         }
 403
 404         if (index == lov->desc.ld_tgt_count)
 405                 GOTO(out, index = -EINVAL);
 406
 407         if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
 408                 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
 409
 410                 if (lov->lov_tgts[index]->ltd_activate == activate) {
 411                         CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
 412                                uuid->uuid, activate ? "" : "de");
 413                 } else {
 414                         lov->lov_tgts[index]->ltd_activate = activate;
 415                         CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
 416                                activate ? "" : "de", obd_uuid2str(uuid));
 417                 }
 418
 419         } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
 420                 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
 421
 422                 if (lov->lov_tgts[index]->ltd_active == active) {
 423                         CDEBUG(D_INFO, "OSC %s already %sactive!\n",
 424                                uuid->uuid, active ? "" : "in");
 425                         GOTO(out, index);
 426                 } else {
 427                         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
 428                                obd_uuid2str(uuid), active ? "" : "in");
 429                 }
 430
 431                 lov->lov_tgts[index]->ltd_active = active;
 432                 if (active) {
 433                         lov->desc.ld_active_tgt_count++;
 434                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
 435                 } else {
 436                         lov->desc.ld_active_tgt_count--;
 437                         lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
 438                 }
 439         } else {
 440                 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
 441         }
 442
 443  out:
 444         obd_putref(obd);
 445         RETURN(index);
 446 }
 447
 448 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 449                       enum obd_notify_event ev, void *data)
 450 {
 451         int rc = 0;
 452         ENTRY;
 453
 454         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
 455             ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
 456                 struct obd_uuid *uuid;
 457
 458                 LASSERT(watched);
 459
 460                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 461                         CERROR("unexpected notification of %s %s!\n",
 462                                watched->obd_type->typ_name,
 463                                watched->obd_name);
 464                         RETURN(-EINVAL);
 465                 }
 466                 uuid = &watched->u.cli.cl_target_uuid;
 467
 468                 /* Set OSC as active before notifying the observer, so the
 469                  * observer can use the OSC normally.
 470                  */
 471                 rc = lov_set_osc_active(obd, uuid, ev);
 472                 if (rc < 0) {
 473                         CERROR("event(%d) of %s failed: %d\n", ev,
 474                                obd_uuid2str(uuid), rc);
 475                         RETURN(rc);
 476                 }
 477                 /* active event should be pass lov target index as data */
 478                 data = &rc;
 479         }
 480
 481         /* Pass the notification up the chain. */
 482         if (watched) {
 483                 rc = obd_notify_observer(obd, watched, ev, data);
 484         } else {
 485                 /* NULL watched means all osc's in the lov (only for syncs) */
 486                 /* sync event should be send lov idx as data */
 487                 struct lov_obd *lov = &obd->u.lov;
 488                 int i, is_sync;
 489
 490                 data = &i;
 491                 is_sync = (ev == OBD_NOTIFY_SYNC) ||
 492                           (ev == OBD_NOTIFY_SYNC_NONBLOCK);
 493
 494                 obd_getref(obd);
 495                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 496                         if (!lov->lov_tgts[i])
 497                                 continue;
 498
 499                         /* don't send sync event if target not
 500                          * connected/activated */
 501                         if (is_sync &&  !lov->lov_tgts[i]->ltd_active)
 502                                 continue;
 503
 504                         rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
 505                                                  ev, data);
 506                         if (rc) {
 507                                 CERROR("%s: notify %s of %s failed %d\n",
 508                                        obd->obd_name,
 509                                        obd->obd_observer->obd_name,
 510                                        lov->lov_tgts[i]->ltd_obd->obd_name,
 511                                        rc);
 512                         }
 513                 }
 514                 obd_putref(obd);
 515         }
 516
 517         RETURN(rc);
 518 }
 519
 520 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 521                           __u32 index, int gen, int active)
 522 {
 523         struct lov_obd *lov = &obd->u.lov;
 524         struct lov_tgt_desc *tgt;
 525         struct obd_device *tgt_obd;
 526         int rc;
 527         ENTRY;
 528
 529         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
 530                uuidp->uuid, index, gen, active);
 531
 532         if (gen <= 0) {
 533                 CERROR("request to add OBD %s with invalid generation: %d\n",
 534                        uuidp->uuid, gen);
 535                 RETURN(-EINVAL);
 536         }
 537
 538         tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
 539                                         &obd->obd_uuid);
 540         if (tgt_obd == NULL)
 541                 RETURN(-EINVAL);
 542
 543         mutex_lock(&lov->lov_lock);
 544
 545         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
 546                 tgt = lov->lov_tgts[index];
 547                 CERROR("UUID %s already assigned at LOV target index %d\n",
 548                        obd_uuid2str(&tgt->ltd_uuid), index);
 549                 mutex_unlock(&lov->lov_lock);
 550                 RETURN(-EEXIST);
 551         }
 552
 553         if (index >= lov->lov_tgt_size) {
 554                 /* We need to reallocate the lov target array. */
 555                 struct lov_tgt_desc **newtgts, **old = NULL;
 556                 __u32 newsize, oldsize = 0;
 557
 558                 newsize = max(lov->lov_tgt_size, (__u32)2);
 559                 while (newsize < index + 1)
 560                         newsize = newsize << 1;
 561                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 562                 if (newtgts == NULL) {
 563                         mutex_unlock(&lov->lov_lock);
 564                         RETURN(-ENOMEM);
 565                 }
 566
 567                 if (lov->lov_tgt_size) {
 568                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
 569                                lov->lov_tgt_size);
 570                         old = lov->lov_tgts;
 571                         oldsize = lov->lov_tgt_size;
 572                 }
 573
 574                 lov->lov_tgts = newtgts;
 575                 lov->lov_tgt_size = newsize;
 576                 smp_rmb();
 577                 if (old)
 578                         OBD_FREE(old, sizeof(*old) * oldsize);
 579
 580                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
 581                        lov->lov_tgts, lov->lov_tgt_size);
 582         }
 583
 584         OBD_ALLOC_PTR(tgt);
 585         if (!tgt) {
 586                 mutex_unlock(&lov->lov_lock);
 587                 RETURN(-ENOMEM);
 588         }
 589
 590         rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
 591         if (rc) {
 592                 mutex_unlock(&lov->lov_lock);
 593                 OBD_FREE_PTR(tgt);
 594                 RETURN(rc);
 595         }
 596
 597         tgt->ltd_uuid = *uuidp;
 598         tgt->ltd_obd = tgt_obd;
 599         /* XXX - add a sanity check on the generation number. */
 600         tgt->ltd_gen = gen;
 601         tgt->ltd_index = index;
 602         tgt->ltd_activate = active;
 603         lov->lov_tgts[index] = tgt;
 604         if (index >= lov->desc.ld_tgt_count)
 605                 lov->desc.ld_tgt_count = index + 1;
 606
 607         mutex_unlock(&lov->lov_lock);
 608
 609         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
 610                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 611
 612         rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
 613
 614         if (lov->lov_connects == 0) {
 615                 /* lov_connect hasn't been called yet. We'll do the
 616                    lov_connect_obd on this target when that fn first runs,
 617                    because we don't know the connect flags yet. */
 618                 RETURN(0);
 619         }
 620
 621         obd_getref(obd);
 622
 623         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
 624         if (rc)
 625                 GOTO(out, rc);
 626
 627         /* connect to administrative disabled ost */
 628         if (!tgt->ltd_exp)
 629                 GOTO(out, rc = 0);
 630
 631         if (lov->lov_cache != NULL) {
 632                 rc = obd_set_info_async(NULL, tgt->ltd_exp,
 633                                 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
 634                                 sizeof(struct cl_client_cache), lov->lov_cache,
 635                                 NULL);
 636                 if (rc < 0)
 637                         GOTO(out, rc);
 638         }
 639
 640         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
 641                         active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
 642                         (void *)&index);
 643
 644 out:
 645         if (rc) {
 646                 CERROR("add failed (%d), deleting %s\n", rc,
 647                        obd_uuid2str(&tgt->ltd_uuid));
 648                 lov_del_target(obd, index, 0, 0);
 649         }
 650         obd_putref(obd);
 651         RETURN(rc);
 652 }
 653
 654 /* Schedule a target for deletion */
 655 int lov_del_target(struct obd_device *obd, __u32 index,
 656                    struct obd_uuid *uuidp, int gen)
 657 {
 658         struct lov_obd *lov = &obd->u.lov;
 659         int count = lov->desc.ld_tgt_count;
 660         int rc = 0;
 661         ENTRY;
 662
 663         if (index >= count) {
 664                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 665                        index, count);
 666                 RETURN(-EINVAL);
 667         }
 668
 669         obd_getref(obd);
 670
 671         if (!lov->lov_tgts[index]) {
 672                 CERROR("LOV target at index %d is not setup.\n", index);
 673                 GOTO(out, rc = -EINVAL);
 674         }
 675
 676         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
 677                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 678                        lov_uuid2str(lov, index), index,
 679                        obd_uuid2str(uuidp));
 680                 GOTO(out, rc = -EINVAL);
 681         }
 682
 683         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 684                lov_uuid2str(lov, index), index,
 685                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
 686                lov->lov_tgts[index]->ltd_active);
 687
 688         lov->lov_tgts[index]->ltd_reap = 1;
 689         lov->lov_death_row++;
 690         /* we really delete it from obd_putref */
 691 out:
 692         obd_putref(obd);
 693
 694         RETURN(rc);
 695 }
 696
 697 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
 698 {
 699         struct obd_device *osc_obd;
 700
 701         LASSERT(tgt);
 702         LASSERT(tgt->ltd_reap);
 703
 704         osc_obd = class_exp2obd(tgt->ltd_exp);
 705
 706         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
 707                tgt->ltd_uuid.uuid,
 708                osc_obd ? osc_obd->obd_name : "<no obd>");
 709
 710         if (tgt->ltd_exp)
 711                 lov_disconnect_obd(obd, tgt);
 712
 713         OBD_FREE_PTR(tgt);
 714
 715         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
 716            do it ourselves. And we can't do it from lov_cleanup,
 717            because we just lost our only reference to it. */
 718         if (osc_obd)
 719                 class_manual_cleanup(osc_obd);
 720 }
 721
 722 void lov_fix_desc_stripe_size(__u64 *val)
 723 {
 724         if (*val < LOV_DEFAULT_STRIPE_SIZE) {
 725                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 726                               LOV_DEFAULT_STRIPE_SIZE);
 727                 *val = LOV_DEFAULT_STRIPE_SIZE;
 728         } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
 729                 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
 730                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 731                               "multiple of %u)\n",
 732                               *val, LOV_MIN_STRIPE_SIZE);
 733         }
 734 }
 735
 736 void lov_fix_desc_stripe_count(__u32 *val)
 737 {
 738         if (*val == 0)
 739                 *val = 1;
 740 }
 741
 742 void lov_fix_desc_pattern(__u32 *val)
 743 {
 744         /* from lov_setstripe */
 745         if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
 746                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
 747                 *val = 0;
 748         }
 749 }
 750
 751 void lov_fix_desc_qos_maxage(__u32 *val)
 752 {
 753         /* fix qos_maxage */
 754         if (*val == 0)
 755                 *val = QOS_DEFAULT_MAXAGE;
 756 }
 757
 758 void lov_fix_desc(struct lov_desc *desc)
 759 {
 760         lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
 761         lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
 762         lov_fix_desc_pattern(&desc->ld_pattern);
 763         lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
 764 }
 765
 766 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 767 {
 768         struct lprocfs_static_vars lvars = { 0 };
 769         struct lov_desc *desc;
 770         struct lov_obd *lov = &obd->u.lov;
 771         int rc;
 772         ENTRY;
 773
 774         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 775                 CERROR("LOV setup requires a descriptor\n");
 776                 RETURN(-EINVAL);
 777         }
 778
 779         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 780
 781         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 782                 CERROR("descriptor size wrong: %d > %d\n",
 783                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 784                 RETURN(-EINVAL);
 785         }
 786
 787         if (desc->ld_magic != LOV_DESC_MAGIC) {
 788                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 789                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 790                                    obd->obd_name, desc);
 791                             lustre_swab_lov_desc(desc);
 792                 } else {
 793                         CERROR("%s: Bad lov desc magic: %#x\n",
 794                                obd->obd_name, desc->ld_magic);
 795                         RETURN(-EINVAL);
 796                 }
 797         }
 798
 799         lov_fix_desc(desc);
 800
 801         desc->ld_active_tgt_count = 0;
 802         lov->desc = *desc;
 803         lov->lov_tgt_size = 0;
 804
 805         mutex_init(&lov->lov_lock);
 806         cfs_atomic_set(&lov->lov_refcount, 0);
 807         lov->lov_sp_me = LUSTRE_SP_CLI;
 808
 809         lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
 810                                                    HASH_POOLS_MAX_BITS,
 811                                                    HASH_POOLS_BKT_BITS, 0,
 812                                                    CFS_HASH_MIN_THETA,
 813                                                    CFS_HASH_MAX_THETA,
 814                                                    &pool_hash_operations,
 815                                                    CFS_HASH_DEFAULT);
 816         CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
 817         lov->lov_pool_count = 0;
 818         rc = lov_ost_pool_init(&lov->lov_packed, 0);
 819         if (rc)
 820                 GOTO(out, rc);
 821
 822         lprocfs_lov_init_vars(&lvars);
 823         lprocfs_obd_setup(obd, lvars.obd_vars);
 824 #ifdef LPROCFS
 825         {
 826                 int rc;
 827
 828                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
 829                                         0444, &lov_proc_target_fops, obd);
 830                 if (rc)
 831                         CWARN("Error adding the target_obd file\n");
 832         }
 833 #endif
 834         lov->lov_pool_proc_entry = lprocfs_register("pools",
 835                                                     obd->obd_proc_entry,
 836                                                     NULL, NULL);
 837
 838         RETURN(0);
 839
 840 out:
 841         return rc;
 842 }
 843
 844 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 845 {
 846         int rc = 0;
 847         struct lov_obd *lov = &obd->u.lov;
 848
 849         ENTRY;
 850
 851         switch (stage) {
 852         case OBD_CLEANUP_EARLY: {
 853                 int i;
 854                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 855                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
 856                                 continue;
 857                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
 858                                        OBD_CLEANUP_EARLY);
 859                 }
 860                 break;
 861         }
 862         case OBD_CLEANUP_EXPORTS:
 863                 rc = obd_llog_finish(obd, 0);
 864                 if (rc != 0)
 865                         CERROR("failed to cleanup llogging subsystems\n");
 866                 break;
 867         }
 868         RETURN(rc);
 869 }
 870
 871 static int lov_cleanup(struct obd_device *obd)
 872 {
 873         struct lov_obd *lov = &obd->u.lov;
 874         cfs_list_t *pos, *tmp;
 875         struct pool_desc *pool;
 876         ENTRY;
 877
 878         cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
 879                 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
 880                 /* free pool structs */
 881                 CDEBUG(D_INFO, "delete pool %p\n", pool);
 882                 /* In the function below, .hs_keycmp resolves to
 883                  * pool_hashkey_keycmp() */
 884                 /* coverity[overrun-buffer-val] */
 885                 lov_pool_del(obd, pool->pool_name);
 886         }
 887         cfs_hash_putref(lov->lov_pools_hash_body);
 888         lov_ost_pool_free(&lov->lov_packed);
 889
 890         lprocfs_obd_cleanup(obd);
 891         if (lov->lov_tgts) {
 892                 int i;
 893                 obd_getref(obd);
 894                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 895                         if (!lov->lov_tgts[i])
 896                                 continue;
 897
 898                         /* Inactive targets may never have connected */
 899                         if (lov->lov_tgts[i]->ltd_active ||
 900                             cfs_atomic_read(&lov->lov_refcount))
 901                             /* We should never get here - these
 902                                should have been removed in the
 903                              disconnect. */
 904                                 CERROR("lov tgt %d not cleaned!"
 905                                        " deathrow=%d, lovrc=%d\n",
 906                                        i, lov->lov_death_row,
 907                                        cfs_atomic_read(&lov->lov_refcount));
 908                         lov_del_target(obd, i, 0, 0);
 909                 }
 910                 obd_putref(obd);
 911                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
 912                          lov->lov_tgt_size);
 913                 lov->lov_tgt_size = 0;
 914         }
 915         RETURN(0);
 916 }
 917
 918 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
 919                             __u32 *indexp, int *genp)
 920 {
 921         struct obd_uuid obd_uuid;
 922         int cmd;
 923         int rc = 0;
 924         ENTRY;
 925
 926         switch(cmd = lcfg->lcfg_command) {
 927         case LCFG_LOV_ADD_OBD:
 928         case LCFG_LOV_ADD_INA:
 929         case LCFG_LOV_DEL_OBD: {
 930                 __u32 index;
 931                 int gen;
 932                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
 933                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
 934                         GOTO(out, rc = -EINVAL);
 935
 936                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
 937
 938                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
 939                         GOTO(out, rc = -EINVAL);
 940                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
 941                         GOTO(out, rc = -EINVAL);
 942                 index = *indexp;
 943                 gen = *genp;
 944                 if (cmd == LCFG_LOV_ADD_OBD)
 945                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
 946                 else if (cmd == LCFG_LOV_ADD_INA)
 947                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
 948                 else
 949                         rc = lov_del_target(obd, index, &obd_uuid, gen);
 950                 GOTO(out, rc);
 951         }
 952         case LCFG_PARAM: {
 953                 struct lprocfs_static_vars lvars = { 0 };
 954                 struct lov_desc *desc = &(obd->u.lov.desc);
 955
 956                 if (!desc)
 957                         GOTO(out, rc = -EINVAL);
 958
 959                 lprocfs_lov_init_vars(&lvars);
 960
 961                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
 962                                               lcfg, obd);
 963                 if (rc > 0)
 964                         rc = 0;
 965                 GOTO(out, rc);
 966         }
 967         case LCFG_POOL_NEW:
 968         case LCFG_POOL_ADD:
 969         case LCFG_POOL_DEL:
 970         case LCFG_POOL_REM:
 971                 GOTO(out, rc);
 972
 973         default: {
 974                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 975                 GOTO(out, rc = -EINVAL);
 976
 977         }
 978         }
 979 out:
 980         RETURN(rc);
 981 }
 982
 983 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
 984                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
 985 {
 986         struct lov_stripe_md *obj_mdp, *lsm;
 987         struct lov_obd *lov = &exp->exp_obd->u.lov;
 988         unsigned ost_idx;
 989         int rc, i;
 990         ENTRY;
 991
 992         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
 993                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
 994
 995         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
 996         if (obj_mdp == NULL)
 997                 RETURN(-ENOMEM);
 998
 999         ost_idx = src_oa->o_nlink;
1000         lsm = *ea;
1001         if (lsm == NULL)
1002                 GOTO(out, rc = -EINVAL);
1003         if (ost_idx >= lov->desc.ld_tgt_count ||
1004             !lov->lov_tgts[ost_idx])
1005                 GOTO(out, rc = -EINVAL);
1006
1007         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1008                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1009                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
1010                                 GOTO(out, rc = -EINVAL);
1011                         break;
1012                 }
1013         }
1014         if (i == lsm->lsm_stripe_count)
1015                 GOTO(out, rc = -EINVAL);
1016
1017         rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1018                         src_oa, &obj_mdp, oti);
1019 out:
1020         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1021         RETURN(rc);
1022 }
1023
1024 /* the LOV expects oa->o_id to be set to the LOV object id */
1025 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1026                       struct obdo *src_oa, struct lov_stripe_md **ea,
1027                       struct obd_trans_info *oti)
1028 {
1029         struct lov_obd *lov;
1030         int rc = 0;
1031         ENTRY;
1032
1033         LASSERT(ea != NULL);
1034         if (exp == NULL)
1035                 RETURN(-EINVAL);
1036
1037         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1038             src_oa->o_flags == OBD_FL_DELORPHAN) {
1039                 /* should be used with LOV anymore */
1040                 LBUG();
1041         }
1042
1043         lov = &exp->exp_obd->u.lov;
1044         if (!lov->desc.ld_active_tgt_count)
1045                 RETURN(-EIO);
1046
1047         obd_getref(exp->exp_obd);
1048         /* Recreate a specific object id at the given OST index */
1049         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1050             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1051                  rc = lov_recreate(exp, src_oa, ea, oti);
1052         }
1053
1054         obd_putref(exp->exp_obd);
1055         RETURN(rc);
1056 }
1057
1058 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1059 do {                                                                            \
1060         LASSERT((lsmp) != NULL);                                                \
1061         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 ||                          \
1062                  (lsmp)->lsm_magic == LOV_MAGIC_V3),                            \
1063                  "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic);              \
1064 } while (0)
1065
1066 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1067                        struct obdo *oa, struct lov_stripe_md *lsm,
1068                        struct obd_trans_info *oti, struct obd_export *md_exp,
1069                        void *capa)
1070 {
1071         struct lov_request_set *set;
1072         struct obd_info oinfo;
1073         struct lov_request *req;
1074         cfs_list_t *pos;
1075         struct lov_obd *lov;
1076         int rc = 0, err = 0;
1077         ENTRY;
1078
1079         ASSERT_LSM_MAGIC(lsm);
1080
1081         if (!exp || !exp->exp_obd)
1082                 RETURN(-ENODEV);
1083
1084         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1085                 LASSERT(oti);
1086                 LASSERT(oti->oti_logcookies);
1087         }
1088
1089         lov = &exp->exp_obd->u.lov;
1090         obd_getref(exp->exp_obd);
1091         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1092         if (rc)
1093                 GOTO(out, rc);
1094
1095         cfs_list_for_each (pos, &set->set_list) {
1096                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1097
1098                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1099                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1100
1101                 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1102                                   req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1103                 err = lov_update_common_set(set, req, err);
1104                 if (err) {
1105                         CERROR("error: destroying objid "LPX64" subobj "
1106                                LPX64" on OST idx %d: rc = %d\n",
1107                                oa->o_id, req->rq_oi.oi_oa->o_id,
1108                                req->rq_idx, err);
1109                         if (!rc)
1110                                 rc = err;
1111                 }
1112         }
1113
1114         if (rc == 0) {
1115                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1116                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1117         }
1118         err = lov_fini_destroy_set(set);
1119 out:
1120         obd_putref(exp->exp_obd);
1121         RETURN(rc ? rc : err);
1122 }
1123
1124 static int lov_getattr(const struct lu_env *env, struct obd_export *exp,
1125                        struct obd_info *oinfo)
1126 {
1127         struct lov_request_set *set;
1128         struct lov_request *req;
1129         cfs_list_t *pos;
1130         struct lov_obd *lov;
1131         int err = 0, rc = 0;
1132         ENTRY;
1133
1134         LASSERT(oinfo);
1135         ASSERT_LSM_MAGIC(oinfo->oi_md);
1136
1137         if (!exp || !exp->exp_obd)
1138                 RETURN(-ENODEV);
1139
1140         lov = &exp->exp_obd->u.lov;
1141
1142         rc = lov_prep_getattr_set(exp, oinfo, &set);
1143         if (rc)
1144                 RETURN(rc);
1145
1146         cfs_list_for_each (pos, &set->set_list) {
1147                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1148
1149                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1150                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1151                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1152
1153                 rc = obd_getattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1154                                  &req->rq_oi);
1155                 err = lov_update_common_set(set, req, rc);
1156                 if (err) {
1157                         CERROR("error: getattr objid "LPX64" subobj "
1158                                LPX64" on OST idx %d: rc = %d\n",
1159                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1160                                req->rq_idx, err);
1161                         break;
1162                 }
1163         }
1164
1165         rc = lov_fini_getattr_set(set);
1166         if (err)
1167                 rc = err;
1168         RETURN(rc);
1169 }
1170
1171 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1172                                  void *data, int rc)
1173 {
1174         struct lov_request_set *lovset = (struct lov_request_set *)data;
1175         int err;
1176         ENTRY;
1177
1178         /* don't do attribute merge if this aysnc op failed */
1179         if (rc)
1180                 cfs_atomic_set(&lovset->set_completes, 0);
1181         err = lov_fini_getattr_set(lovset);
1182         RETURN(rc ? rc : err);
1183 }
1184
1185 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1186                               struct ptlrpc_request_set *rqset)
1187 {
1188         struct lov_request_set *lovset;
1189         struct lov_obd *lov;
1190         cfs_list_t *pos;
1191         struct lov_request *req;
1192         int rc = 0, err;
1193         ENTRY;
1194
1195         LASSERT(oinfo);
1196         ASSERT_LSM_MAGIC(oinfo->oi_md);
1197
1198         if (!exp || !exp->exp_obd)
1199                 RETURN(-ENODEV);
1200
1201         lov = &exp->exp_obd->u.lov;
1202
1203         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1204         if (rc)
1205                 RETURN(rc);
1206
1207         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1208                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1209                oinfo->oi_md->lsm_stripe_size);
1210
1211         cfs_list_for_each (pos, &lovset->set_list) {
1212                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1213
1214                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1215                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1216                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1217                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1218                                        &req->rq_oi, rqset);
1219                 if (rc) {
1220                         CERROR("error: getattr objid "LPX64" subobj "
1221                                LPX64" on OST idx %d: rc = %d\n",
1222                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1223                                req->rq_idx, rc);
1224                         GOTO(out, rc);
1225                 }
1226         }
1227
1228         if (!cfs_list_empty(&rqset->set_requests)) {
1229                 LASSERT(rc == 0);
1230                 LASSERT (rqset->set_interpret == NULL);
1231                 rqset->set_interpret = lov_getattr_interpret;
1232                 rqset->set_arg = (void *)lovset;
1233                 RETURN(rc);
1234         }
1235 out:
1236         if (rc)
1237                 cfs_atomic_set(&lovset->set_completes, 0);
1238         err = lov_fini_getattr_set(lovset);
1239         RETURN(rc ? rc : err);
1240 }
1241
1242 static int lov_setattr(const struct lu_env *env, struct obd_export *exp,
1243                        struct obd_info *oinfo, struct obd_trans_info *oti)
1244 {
1245         struct lov_request_set *set;
1246         struct lov_obd *lov;
1247         cfs_list_t *pos;
1248         struct lov_request *req;
1249         int err = 0, rc = 0;
1250         ENTRY;
1251
1252         LASSERT(oinfo);
1253         ASSERT_LSM_MAGIC(oinfo->oi_md);
1254
1255         if (!exp || !exp->exp_obd)
1256                 RETURN(-ENODEV);
1257
1258         /* for now, we only expect the following updates here */
1259         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1260                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1261                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1262                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1263                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1264                                             OBD_MD_FLGID | OBD_MD_FLFID |
1265                                             OBD_MD_FLGENER)));
1266         lov = &exp->exp_obd->u.lov;
1267         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1268         if (rc)
1269                 RETURN(rc);
1270
1271         cfs_list_for_each (pos, &set->set_list) {
1272                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1273
1274                 rc = obd_setattr(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1275                                  &req->rq_oi, NULL);
1276                 err = lov_update_setattr_set(set, req, rc);
1277                 if (err) {
1278                         CERROR("error: setattr objid "LPX64" subobj "
1279                                LPX64" on OST idx %d: rc = %d\n",
1280                                set->set_oi->oi_oa->o_id,
1281                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1282                         if (!rc)
1283                                 rc = err;
1284                 }
1285         }
1286         err = lov_fini_setattr_set(set);
1287         if (!rc)
1288                 rc = err;
1289         RETURN(rc);
1290 }
1291
1292 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1293                                  void *data, int rc)
1294 {
1295         struct lov_request_set *lovset = (struct lov_request_set *)data;
1296         int err;
1297         ENTRY;
1298
1299         if (rc)
1300                 cfs_atomic_set(&lovset->set_completes, 0);
1301         err = lov_fini_setattr_set(lovset);
1302         RETURN(rc ? rc : err);
1303 }
1304
1305 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1306    needed. Otherwise, a client is waiting for responses. */
1307 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1308                              struct obd_trans_info *oti,
1309                              struct ptlrpc_request_set *rqset)
1310 {
1311         struct lov_request_set *set;
1312         struct lov_request *req;
1313         cfs_list_t *pos;
1314         struct lov_obd *lov;
1315         int rc = 0;
1316         ENTRY;
1317
1318         LASSERT(oinfo);
1319         ASSERT_LSM_MAGIC(oinfo->oi_md);
1320         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1321                 LASSERT(oti);
1322                 LASSERT(oti->oti_logcookies);
1323         }
1324
1325         if (!exp || !exp->exp_obd)
1326                 RETURN(-ENODEV);
1327
1328         lov = &exp->exp_obd->u.lov;
1329         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1330         if (rc)
1331                 RETURN(rc);
1332
1333         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1334                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1335                oinfo->oi_md->lsm_stripe_size);
1336
1337         cfs_list_for_each (pos, &set->set_list) {
1338                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1339
1340                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1341                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1342
1343                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1344                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1345                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1346
1347                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1348                                        &req->rq_oi, oti, rqset);
1349                 if (rc) {
1350                         CERROR("error: setattr objid "LPX64" subobj "
1351                                LPX64" on OST idx %d: rc = %d\n",
1352                                set->set_oi->oi_oa->o_id,
1353                                req->rq_oi.oi_oa->o_id,
1354                                req->rq_idx, rc);
1355                         break;
1356                 }
1357         }
1358
1359         /* If we are not waiting for responses on async requests, return. */
1360         if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1361                 int err;
1362                 if (rc)
1363                         cfs_atomic_set(&set->set_completes, 0);
1364                 err = lov_fini_setattr_set(set);
1365                 RETURN(rc ? rc : err);
1366         }
1367
1368         LASSERT(rqset->set_interpret == NULL);
1369         rqset->set_interpret = lov_setattr_interpret;
1370         rqset->set_arg = (void *)set;
1371
1372         RETURN(0);
1373 }
1374
1375 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1376                                void *data, int rc)
1377 {
1378         struct lov_request_set *lovset = (struct lov_request_set *)data;
1379         int err;
1380         ENTRY;
1381
1382         if (rc)
1383                 cfs_atomic_set(&lovset->set_completes, 0);
1384         err = lov_fini_punch_set(lovset);
1385         RETURN(rc ? rc : err);
1386 }
1387
1388 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1389  * we can send this 'punch' to just the authoritative node and the nodes
1390  * that the punch will affect. */
1391 static int lov_punch(const struct lu_env *env, struct obd_export *exp,
1392                      struct obd_info *oinfo, struct obd_trans_info *oti,
1393                      struct ptlrpc_request_set *rqset)
1394 {
1395         struct lov_request_set *set;
1396         struct lov_obd *lov;
1397         cfs_list_t *pos;
1398         struct lov_request *req;
1399         int rc = 0;
1400         ENTRY;
1401
1402         LASSERT(oinfo);
1403         ASSERT_LSM_MAGIC(oinfo->oi_md);
1404
1405         if (!exp || !exp->exp_obd)
1406                 RETURN(-ENODEV);
1407
1408         lov = &exp->exp_obd->u.lov;
1409         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1410         if (rc)
1411                 RETURN(rc);
1412
1413         cfs_list_for_each (pos, &set->set_list) {
1414                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1415
1416                 rc = obd_punch(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1417                                &req->rq_oi, NULL, rqset);
1418                 if (rc) {
1419                         CERROR("error: punch objid "LPX64" subobj "LPX64
1420                                " on OST idx %d: rc = %d\n",
1421                                set->set_oi->oi_oa->o_id,
1422                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1423                         break;
1424                 }
1425         }
1426
1427         if (rc || cfs_list_empty(&rqset->set_requests)) {
1428                 int err;
1429                 err = lov_fini_punch_set(set);
1430                 RETURN(rc ? rc : err);
1431         }
1432
1433         LASSERT(rqset->set_interpret == NULL);
1434         rqset->set_interpret = lov_punch_interpret;
1435         rqset->set_arg = (void *)set;
1436
1437         RETURN(0);
1438 }
1439
1440 static int lov_sync_interpret(struct ptlrpc_request_set *rqset,
1441                               void *data, int rc)
1442 {
1443         struct lov_request_set *lovset = data;
1444         int err;
1445         ENTRY;
1446
1447         if (rc)
1448                 cfs_atomic_set(&lovset->set_completes, 0);
1449         err = lov_fini_sync_set(lovset);
1450         RETURN(rc ?: err);
1451 }
1452
1453 static int lov_sync(const struct lu_env *env, struct obd_export *exp,
1454                     struct obd_info *oinfo, obd_off start, obd_off end,
1455                     struct ptlrpc_request_set *rqset)
1456 {
1457         struct lov_request_set *set = NULL;
1458         struct lov_obd *lov;
1459         cfs_list_t *pos;
1460         struct lov_request *req;
1461         int rc = 0;
1462         ENTRY;
1463
1464         ASSERT_LSM_MAGIC(oinfo->oi_md);
1465         LASSERT(rqset != NULL);
1466
1467         if (!exp->exp_obd)
1468                 RETURN(-ENODEV);
1469
1470         lov = &exp->exp_obd->u.lov;
1471         rc = lov_prep_sync_set(exp, oinfo, start, end, &set);
1472         if (rc)
1473                 RETURN(rc);
1474
1475         CDEBUG(D_INFO, "fsync objid "LPX64" ["LPX64", "LPX64"]\n",
1476                set->set_oi->oi_oa->o_id, start, end);
1477
1478         cfs_list_for_each (pos, &set->set_list) {
1479                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1480
1481                 rc = obd_sync(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1482                               &req->rq_oi, req->rq_oi.oi_policy.l_extent.start,
1483                               req->rq_oi.oi_policy.l_extent.end, rqset);
1484                 if (rc) {
1485                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1486                                " on OST idx %d: rc = %d\n",
1487                                set->set_oi->oi_oa->o_id,
1488                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1489                         break;
1490                 }
1491         }
1492
1493         /* If we are not waiting for responses on async requests, return. */
1494         if (rc || cfs_list_empty(&rqset->set_requests)) {
1495                 int err = lov_fini_sync_set(set);
1496
1497                 RETURN(rc ?: err);
1498         }
1499
1500         LASSERT(rqset->set_interpret == NULL);
1501         rqset->set_interpret = lov_sync_interpret;
1502         rqset->set_arg = (void *)set;
1503
1504         RETURN(0);
1505 }
1506
1507 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1508                          obd_count oa_bufs, struct brw_page *pga)
1509 {
1510         struct obd_info oinfo = { { { 0 } } };
1511         int i, rc = 0;
1512
1513         oinfo.oi_oa = lov_oinfo->oi_oa;
1514
1515         /* The caller just wants to know if there's a chance that this
1516          * I/O can succeed */
1517         for (i = 0; i < oa_bufs; i++) {
1518                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1519                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1520                 obd_off start, end;
1521
1522                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1523                                            pga[i].off + pga[i].count - 1,
1524                                            &start, &end))
1525                         continue;
1526
1527                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1528                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1529                         return -EIO;
1530                 }
1531
1532                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1533                              1, &pga[i], NULL);
1534                 if (rc)
1535                         break;
1536         }
1537         return rc;
1538 }
1539
1540 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1541                    obd_count oa_bufs, struct brw_page *pga,
1542                    struct obd_trans_info *oti)
1543 {
1544         struct lov_request_set *set;
1545         struct lov_request *req;
1546         cfs_list_t *pos;
1547         struct lov_obd *lov = &exp->exp_obd->u.lov;
1548         int err, rc = 0;
1549         ENTRY;
1550
1551         ASSERT_LSM_MAGIC(oinfo->oi_md);
1552
1553         if (cmd == OBD_BRW_CHECK) {
1554                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1555                 RETURN(rc);
1556         }
1557
1558         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1559         if (rc)
1560                 RETURN(rc);
1561
1562         cfs_list_for_each (pos, &set->set_list) {
1563                 struct obd_export *sub_exp;
1564                 struct brw_page *sub_pga;
1565                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1566
1567                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1568                 sub_pga = set->set_pga + req->rq_pgaidx;
1569                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1570                              sub_pga, oti);
1571                 if (rc)
1572                         break;
1573                 lov_update_common_set(set, req, rc);
1574         }
1575
1576         err = lov_fini_brw_set(set);
1577         if (!rc)
1578                 rc = err;
1579         RETURN(rc);
1580 }
1581
1582 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1583                                  void *data, int rc)
1584 {
1585         struct lov_request_set *lovset = (struct lov_request_set *)data;
1586         ENTRY;
1587         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1588         RETURN(rc);
1589 }
1590
1591 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1592                        struct ldlm_enqueue_info *einfo,
1593                        struct ptlrpc_request_set *rqset)
1594 {
1595         ldlm_mode_t mode = einfo->ei_mode;
1596         struct lov_request_set *set;
1597         struct lov_request *req;
1598         cfs_list_t *pos;
1599         struct lov_obd *lov;
1600         ldlm_error_t rc;
1601         ENTRY;
1602
1603         LASSERT(oinfo);
1604         ASSERT_LSM_MAGIC(oinfo->oi_md);
1605         LASSERT(mode == (mode & -mode));
1606
1607         /* we should never be asked to replay a lock this way. */
1608         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1609
1610         if (!exp || !exp->exp_obd)
1611                 RETURN(-ENODEV);
1612
1613         lov = &exp->exp_obd->u.lov;
1614         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1615         if (rc)
1616                 RETURN(rc);
1617
1618         cfs_list_for_each (pos, &set->set_list) {
1619                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1620
1621                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1622                                  &req->rq_oi, einfo, rqset);
1623                 if (rc != ELDLM_OK)
1624                         GOTO(out, rc);
1625         }
1626
1627         if (rqset && !cfs_list_empty(&rqset->set_requests)) {
1628                 LASSERT(rc == 0);
1629                 LASSERT(rqset->set_interpret == NULL);
1630                 rqset->set_interpret = lov_enqueue_interpret;
1631                 rqset->set_arg = (void *)set;
1632                 RETURN(rc);
1633         }
1634 out:
1635         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1636         RETURN(rc);
1637 }
1638
1639 static int lov_change_cbdata(struct obd_export *exp,
1640                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1641                              void *data)
1642 {
1643         struct lov_obd *lov;
1644         int rc = 0, i;
1645         ENTRY;
1646
1647         ASSERT_LSM_MAGIC(lsm);
1648
1649         if (!exp || !exp->exp_obd)
1650                 RETURN(-ENODEV);
1651
1652         lov = &exp->exp_obd->u.lov;
1653         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1654                 struct lov_stripe_md submd;
1655                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1656
1657                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1658                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1659                         continue;
1660                 }
1661
1662                 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1663                 submd.lsm_object_id = loi->loi_id;
1664                 submd.lsm_object_seq = loi->loi_seq;
1665                 submd.lsm_stripe_count = 0;
1666                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1667                                        &submd, it, data);
1668         }
1669         RETURN(rc);
1670 }
1671
1672 /* find any ldlm lock of the inode in lov
1673  * return 0    not find
1674  *        1    find one
1675  *      < 0    error */
1676 static int lov_find_cbdata(struct obd_export *exp,
1677                            struct lov_stripe_md *lsm, ldlm_iterator_t it,
1678                            void *data)
1679 {
1680         struct lov_obd *lov;
1681         int rc = 0, i;
1682         ENTRY;
1683
1684         ASSERT_LSM_MAGIC(lsm);
1685
1686         if (!exp || !exp->exp_obd)
1687                 RETURN(-ENODEV);
1688
1689         lov = &exp->exp_obd->u.lov;
1690         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1691                 struct lov_stripe_md submd;
1692                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1693
1694                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1695                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1696                         continue;
1697                 }
1698
1699                 LASSERT_SEQ_IS_MDT(loi->loi_seq);
1700                 submd.lsm_object_id = loi->loi_id;
1701                 submd.lsm_object_seq = loi->loi_seq;
1702                 submd.lsm_stripe_count = 0;
1703                 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1704                                      &submd, it, data);
1705                 if (rc != 0)
1706                         RETURN(rc);
1707         }
1708         RETURN(rc);
1709 }
1710
1711 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1712                       __u32 mode, struct lustre_handle *lockh)
1713 {
1714         struct lov_request_set *set;
1715         struct obd_info oinfo;
1716         struct lov_request *req;
1717         cfs_list_t *pos;
1718         struct lov_obd *lov;
1719         struct lustre_handle *lov_lockhp;
1720         int err = 0, rc = 0;
1721         ENTRY;
1722
1723         ASSERT_LSM_MAGIC(lsm);
1724
1725         if (!exp || !exp->exp_obd)
1726                 RETURN(-ENODEV);
1727
1728         LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1729         LASSERT(lockh);
1730         lov = &exp->exp_obd->u.lov;
1731         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1732         if (rc)
1733                 RETURN(rc);
1734
1735         cfs_list_for_each (pos, &set->set_list) {
1736                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1737                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1738
1739                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
1740                                 req->rq_oi.oi_md, mode, lov_lockhp);
1741                 rc = lov_update_common_set(set, req, rc);
1742                 if (rc) {
1743                         CERROR("error: cancel objid "LPX64" subobj "
1744                                LPX64" on OST idx %d: rc = %d\n",
1745                                lsm->lsm_object_id,
1746                                req->rq_oi.oi_md->lsm_object_id,
1747                                req->rq_idx, rc);
1748                         err = rc;
1749                 }
1750
1751         }
1752         lov_fini_cancel_set(set);
1753         RETURN(err);
1754 }
1755
1756 static int lov_cancel_unused(struct obd_export *exp,
1757                              struct lov_stripe_md *lsm,
1758                              ldlm_cancel_flags_t flags, void *opaque)
1759 {
1760         struct lov_obd *lov;
1761         int rc = 0, i;
1762         ENTRY;
1763
1764         if (!exp || !exp->exp_obd)
1765                 RETURN(-ENODEV);
1766
1767         lov = &exp->exp_obd->u.lov;
1768         if (lsm == NULL) {
1769                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1770                         int err;
1771                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1772                                 continue;
1773
1774                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
1775                                                 flags, opaque);
1776                         if (!rc)
1777                                 rc = err;
1778                 }
1779                 RETURN(rc);
1780         }
1781
1782         ASSERT_LSM_MAGIC(lsm);
1783
1784         LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
1785         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1786                 struct lov_stripe_md submd;
1787                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1788                 int err;
1789
1790                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1791                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1792                         continue;
1793                 }
1794
1795                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
1796                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1797
1798                 submd.lsm_object_id = loi->loi_id;
1799                 submd.lsm_object_seq = loi->loi_seq;
1800                 submd.lsm_stripe_count = 0;
1801                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1802                                         &submd, flags, opaque);
1803                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1804                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1805                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1806                                loi->loi_id, loi->loi_ost_idx, err);
1807                         if (!rc)
1808                                 rc = err;
1809                 }
1810         }
1811         RETURN(rc);
1812 }
1813
1814 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1815 {
1816         struct lov_request_set *lovset = (struct lov_request_set *)data;
1817         int err;
1818         ENTRY;
1819
1820         if (rc)
1821                 cfs_atomic_set(&lovset->set_completes, 0);
1822
1823         err = lov_fini_statfs_set(lovset);
1824         RETURN(rc ? rc : err);
1825 }
1826
1827 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1828                             __u64 max_age, struct ptlrpc_request_set *rqset)
1829 {
1830         struct obd_device      *obd = class_exp2obd(exp);
1831         struct lov_request_set *set;
1832         struct lov_request *req;
1833         cfs_list_t *pos;
1834         struct lov_obd *lov;
1835         int rc = 0;
1836         ENTRY;
1837
1838         LASSERT(oinfo != NULL);
1839         LASSERT(oinfo->oi_osfs != NULL);
1840
1841         lov = &obd->u.lov;
1842         rc = lov_prep_statfs_set(obd, oinfo, &set);
1843         if (rc)
1844                 RETURN(rc);
1845
1846         cfs_list_for_each (pos, &set->set_list) {
1847                 req = cfs_list_entry(pos, struct lov_request, rq_link);
1848                 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1849                                       &req->rq_oi, max_age, rqset);
1850                 if (rc)
1851                         break;
1852         }
1853
1854         if (rc || cfs_list_empty(&rqset->set_requests)) {
1855                 int err;
1856                 if (rc)
1857                         cfs_atomic_set(&set->set_completes, 0);
1858                 err = lov_fini_statfs_set(set);
1859                 RETURN(rc ? rc : err);
1860         }
1861
1862         LASSERT(rqset->set_interpret == NULL);
1863         rqset->set_interpret = lov_statfs_interpret;
1864         rqset->set_arg = (void *)set;
1865         RETURN(0);
1866 }
1867
1868 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1869                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1870 {
1871         struct ptlrpc_request_set *set = NULL;
1872         struct obd_info oinfo = { { { 0 } } };
1873         int rc = 0;
1874         ENTRY;
1875
1876
1877         /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1878          * statfs requests */
1879         set = ptlrpc_prep_set();
1880         if (set == NULL)
1881                 RETURN(-ENOMEM);
1882
1883         oinfo.oi_osfs = osfs;
1884         oinfo.oi_flags = flags;
1885         rc = lov_statfs_async(exp, &oinfo, max_age, set);
1886         if (rc == 0)
1887                 rc = ptlrpc_set_wait(set);
1888         ptlrpc_set_destroy(set);
1889
1890         RETURN(rc);
1891 }
1892
1893 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1894                          void *karg, void *uarg)
1895 {
1896         struct obd_device *obddev = class_exp2obd(exp);
1897         struct lov_obd *lov = &obddev->u.lov;
1898         int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1899         struct obd_uuid *uuidp;
1900         ENTRY;
1901
1902         switch (cmd) {
1903         case IOC_OBD_STATFS: {
1904                 struct obd_ioctl_data *data = karg;
1905                 struct obd_device *osc_obd;
1906                 struct obd_statfs stat_buf = {0};
1907                 __u32 index;
1908                 __u32 flags;
1909
1910                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1911                 if ((index >= count))
1912                         RETURN(-ENODEV);
1913
1914                 if (!lov->lov_tgts[index])
1915                         /* Try again with the next index */
1916                         RETURN(-EAGAIN);
1917                 if (!lov->lov_tgts[index]->ltd_active)
1918                         RETURN(-ENODATA);
1919
1920                 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1921                 if (!osc_obd)
1922                         RETURN(-EINVAL);
1923
1924                 /* copy UUID */
1925                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1926                                      min((int) data->ioc_plen2,
1927                                          (int) sizeof(struct obd_uuid))))
1928                         RETURN(-EFAULT);
1929
1930                 flags = uarg ? *(__u32*)uarg : 0;
1931                 /* got statfs data */
1932                 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1933                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1934                                 flags);
1935                 if (rc)
1936                         RETURN(rc);
1937                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
1938                                      min((int) data->ioc_plen1,
1939                                          (int) sizeof(stat_buf))))
1940                         RETURN(-EFAULT);
1941                 break;
1942         }
1943         case OBD_IOC_LOV_GET_CONFIG: {
1944                 struct obd_ioctl_data *data;
1945                 struct lov_desc *desc;
1946                 char *buf = NULL;
1947                 __u32 *genp;
1948
1949                 len = 0;
1950                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1951                         RETURN(-EINVAL);
1952
1953                 data = (struct obd_ioctl_data *)buf;
1954
1955                 if (sizeof(*desc) > data->ioc_inllen1) {
1956                         obd_ioctl_freedata(buf, len);
1957                         RETURN(-EINVAL);
1958                 }
1959
1960                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1961                         obd_ioctl_freedata(buf, len);
1962                         RETURN(-EINVAL);
1963                 }
1964
1965                 if (sizeof(__u32) * count > data->ioc_inllen3) {
1966                         obd_ioctl_freedata(buf, len);
1967                         RETURN(-EINVAL);
1968                 }
1969
1970                 desc = (struct lov_desc *)data->ioc_inlbuf1;
1971                 memcpy(desc, &(lov->desc), sizeof(*desc));
1972
1973                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1974                 genp = (__u32 *)data->ioc_inlbuf3;
1975                 /* the uuid will be empty for deleted OSTs */
1976                 for (i = 0; i < count; i++, uuidp++, genp++) {
1977                         if (!lov->lov_tgts[i])
1978                                 continue;
1979                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
1980                         *genp = lov->lov_tgts[i]->ltd_gen;
1981                 }
1982
1983                 if (cfs_copy_to_user((void *)uarg, buf, len))
1984                         rc = -EFAULT;
1985                 obd_ioctl_freedata(buf, len);
1986                 break;
1987         }
1988         case LL_IOC_LOV_SETSTRIPE:
1989                 rc = lov_setstripe(exp, len, karg, uarg);
1990                 break;
1991         case LL_IOC_LOV_GETSTRIPE:
1992                 rc = lov_getstripe(exp, karg, uarg);
1993                 break;
1994         case LL_IOC_LOV_SETEA:
1995                 rc = lov_setea(exp, karg, uarg);
1996                 break;
1997         case OBD_IOC_QUOTACTL: {
1998                 struct if_quotactl *qctl = karg;
1999                 struct lov_tgt_desc *tgt = NULL;
2000                 struct obd_quotactl *oqctl;
2001
2002                 if (qctl->qc_valid == QC_OSTIDX) {
2003                         if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
2004                                 RETURN(-EINVAL);
2005
2006                         tgt = lov->lov_tgts[qctl->qc_idx];
2007                         if (!tgt || !tgt->ltd_exp)
2008                                 RETURN(-EINVAL);
2009                 } else if (qctl->qc_valid == QC_UUID) {
2010                         for (i = 0; i < count; i++) {
2011                                 tgt = lov->lov_tgts[i];
2012                                 if (!tgt ||
2013                                     !obd_uuid_equals(&tgt->ltd_uuid,
2014                                                      &qctl->obd_uuid))
2015                                         continue;
2016
2017                                 if (tgt->ltd_exp == NULL)
2018                                         RETURN(-EINVAL);
2019
2020                                 break;
2021                         }
2022                 } else {
2023                         RETURN(-EINVAL);
2024                 }
2025
2026                 if (i >= count)
2027                         RETURN(-EAGAIN);
2028
2029                 LASSERT(tgt && tgt->ltd_exp);
2030                 OBD_ALLOC_PTR(oqctl);
2031                 if (!oqctl)
2032                         RETURN(-ENOMEM);
2033
2034                 QCTL_COPY(oqctl, qctl);
2035                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2036                 if (rc == 0) {
2037                         QCTL_COPY(qctl, oqctl);
2038                         qctl->qc_valid = QC_OSTIDX;
2039                         qctl->obd_uuid = tgt->ltd_uuid;
2040                 }
2041                 OBD_FREE_PTR(oqctl);
2042                 break;
2043         }
2044         default: {
2045                 int set = 0;
2046
2047                 if (count == 0)
2048                         RETURN(-ENOTTY);
2049
2050                 for (i = 0; i < count; i++) {
2051                         int err;
2052                         struct obd_device *osc_obd;
2053
2054                         /* OST was disconnected */
2055                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2056                                 continue;
2057
2058                         /* ll_umount_begin() sets force flag but for lov, not
2059                          * osc. Let's pass it through */
2060                         osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
2061                         osc_obd->obd_force = obddev->obd_force;
2062                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2063                                             len, karg, uarg);
2064                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2065                                 RETURN(err);
2066                         } else if (err) {
2067                                 if (lov->lov_tgts[i]->ltd_active) {
2068                                         CDEBUG(err == -ENOTTY ?
2069                                                D_IOCTL : D_WARNING,
2070                                                "iocontrol OSC %s on OST "
2071                                                "idx %d cmd %x: err = %d\n",
2072                                                lov_uuid2str(lov, i),
2073                                                i, cmd, err);
2074                                         if (!rc)
2075                                                 rc = err;
2076                                 }
2077                         } else {
2078                                 set = 1;
2079                         }
2080                 }
2081                 if (!set && !rc)
2082                         rc = -EIO;
2083         }
2084         }
2085
2086         RETURN(rc);
2087 }
2088
2089 #define FIEMAP_BUFFER_SIZE 4096
2090
2091 /**
2092  * Non-zero fe_logical indicates that this is a continuation FIEMAP
2093  * call. The local end offset and the device are sent in the first
2094  * fm_extent. This function calculates the stripe number from the index.
2095  * This function returns a stripe_no on which mapping is to be restarted.
2096  *
2097  * This function returns fm_end_offset which is the in-OST offset at which
2098  * mapping should be restarted. If fm_end_offset=0 is returned then caller
2099  * will re-calculate proper offset in next stripe.
2100  * Note that the first extent is passed to lov_get_info via the value field.
2101  *
2102  * \param fiemap fiemap request header
2103  * \param lsm striping information for the file
2104  * \param fm_start logical start of mapping
2105  * \param fm_end logical end of mapping
2106  * \param start_stripe starting stripe will be returned in this
2107  */
2108 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
2109                                    struct lov_stripe_md *lsm, obd_size fm_start,
2110                                    obd_size fm_end, int *start_stripe)
2111 {
2112         obd_size local_end = fiemap->fm_extents[0].fe_logical;
2113         obd_off lun_start, lun_end;
2114         obd_size fm_end_offset;
2115         int stripe_no = -1, i;
2116
2117         if (fiemap->fm_extent_count == 0 ||
2118             fiemap->fm_extents[0].fe_logical == 0)
2119                 return 0;
2120
2121         /* Find out stripe_no from ost_index saved in the fe_device */
2122         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2123                 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
2124                                         fiemap->fm_extents[0].fe_device) {
2125                         stripe_no = i;
2126                         break;
2127                 }
2128         }
2129         if (stripe_no == -1)
2130                 return -EINVAL;
2131
2132         /* If we have finished mapping on previous device, shift logical
2133          * offset to start of next device */
2134         if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
2135                                    &lun_start, &lun_end)) != 0 &&
2136                                    local_end < lun_end) {
2137                 fm_end_offset = local_end;
2138                 *start_stripe = stripe_no;
2139         } else {
2140                 /* This is a special value to indicate that caller should
2141                  * calculate offset in next stripe. */
2142                 fm_end_offset = 0;
2143                 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
2144         }
2145
2146         return fm_end_offset;
2147 }
2148
2149 /**
2150  * We calculate on which OST the mapping will end. If the length of mapping
2151  * is greater than (stripe_size * stripe_count) then the last_stripe will
2152  * will be one just before start_stripe. Else we check if the mapping
2153  * intersects each OST and find last_stripe.
2154  * This function returns the last_stripe and also sets the stripe_count
2155  * over which the mapping is spread
2156  *
2157  * \param lsm striping information for the file
2158  * \param fm_start logical start of mapping
2159  * \param fm_end logical end of mapping
2160  * \param start_stripe starting stripe of the mapping
2161  * \param stripe_count the number of stripes across which to map is returned
2162  *
2163  * \retval last_stripe return the last stripe of the mapping
2164  */
2165 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
2166                             obd_size fm_end, int start_stripe,
2167                             int *stripe_count)
2168 {
2169         int last_stripe;
2170         obd_off obd_start, obd_end;
2171         int i, j;
2172
2173         if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
2174                 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
2175                                                               start_stripe - 1);
2176                 *stripe_count = lsm->lsm_stripe_count;
2177         } else {
2178                 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
2179                      i = (i + 1) % lsm->lsm_stripe_count, j++) {
2180                         if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
2181                                                    &obd_start, &obd_end)) == 0)
2182                                 break;
2183                 }
2184                 *stripe_count = j;
2185                 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
2186         }
2187
2188         return last_stripe;
2189 }
2190
2191 /**
2192  * Set fe_device and copy extents from local buffer into main return buffer.
2193  *
2194  * \param fiemap fiemap request header
2195  * \param lcl_fm_ext array of local fiemap extents to be copied
2196  * \param ost_index OST index to be written into the fm_device field for each
2197                     extent
2198  * \param ext_count number of extents to be copied
2199  * \param current_extent where to start copying in main extent array
2200  */
2201 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
2202                                   struct ll_fiemap_extent *lcl_fm_ext,
2203                                   int ost_index, unsigned int ext_count,
2204                                   int current_extent)
2205 {
2206         char *to;
2207         int ext;
2208
2209         for (ext = 0; ext < ext_count; ext++) {
2210                 lcl_fm_ext[ext].fe_device = ost_index;
2211                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
2212         }
2213
2214         /* Copy fm_extent's from fm_local to return buffer */
2215         to = (char *)fiemap + fiemap_count_to_size(current_extent);
2216         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
2217 }
2218
2219 /**
2220  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
2221  * This also handles the restarting of FIEMAP calls in case mapping overflows
2222  * the available number of extents in single call.
2223  */
2224 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
2225                       __u32 *vallen, void *val, struct lov_stripe_md *lsm)
2226 {
2227         struct ll_fiemap_info_key *fm_key = key;
2228         struct ll_user_fiemap *fiemap = val;
2229         struct ll_user_fiemap *fm_local = NULL;
2230         struct ll_fiemap_extent *lcl_fm_ext;
2231         int count_local;
2232         unsigned int get_num_extents = 0;
2233         int ost_index = 0, actual_start_stripe, start_stripe;
2234         obd_size fm_start, fm_end, fm_length, fm_end_offset;
2235         obd_size curr_loc;
2236         int current_extent = 0, rc = 0, i;
2237         int ost_eof = 0; /* EOF for object */
2238         int ost_done = 0; /* done with required mapping for this OST? */
2239         int last_stripe;
2240         int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
2241         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
2242
2243         if (lsm == NULL)
2244                 GOTO(out, rc = 0);
2245
2246         if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
2247                 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
2248
2249         OBD_ALLOC_LARGE(fm_local, buffer_size);
2250         if (fm_local == NULL)
2251                 GOTO(out, rc = -ENOMEM);
2252         lcl_fm_ext = &fm_local->fm_extents[0];
2253
2254         count_local = fiemap_size_to_count(buffer_size);
2255
2256         memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
2257         fm_start = fiemap->fm_start;
2258         fm_length = fiemap->fm_length;
2259         /* Calculate start stripe, last stripe and length of mapping */
2260         actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
2261         fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
2262                                                 fm_start + fm_length - 1);
2263         /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
2264         if (fm_end > fm_key->oa.o_size)
2265                 fm_end = fm_key->oa.o_size;
2266
2267         last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
2268                                             actual_start_stripe, &stripe_count);
2269
2270         fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
2271                                                   fm_end, &start_stripe);
2272         if (fm_end_offset == -EINVAL)
2273                 GOTO(out, rc = -EINVAL);
2274
2275         if (fiemap->fm_extent_count == 0) {
2276                 get_num_extents = 1;
2277                 count_local = 0;
2278         }
2279
2280         /* Check each stripe */
2281         for (cur_stripe = start_stripe, i = 0; i < stripe_count;
2282              i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
2283                 obd_size req_fm_len; /* Stores length of required mapping */
2284                 obd_size len_mapped_single_call;
2285                 obd_off lun_start, lun_end, obd_object_end;
2286                 unsigned int ext_count;
2287
2288                 cur_stripe_wrap = cur_stripe;
2289
2290                 /* Find out range of mapping on this stripe */
2291                 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
2292                                            &lun_start, &obd_object_end)) == 0)
2293                         continue;
2294
2295                 /* If this is a continuation FIEMAP call and we are on
2296                  * starting stripe then lun_start needs to be set to
2297                  * fm_end_offset */
2298                 if (fm_end_offset != 0 && cur_stripe == start_stripe)
2299                         lun_start = fm_end_offset;
2300
2301                 if (fm_length != ~0ULL) {
2302                         /* Handle fm_start + fm_length overflow */
2303                         if (fm_start + fm_length < fm_start)
2304                                 fm_length = ~0ULL - fm_start;
2305                         lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
2306                                                      cur_stripe);
2307                 } else {
2308                         lun_end = ~0ULL;
2309                 }
2310
2311                 if (lun_start == lun_end)
2312                         continue;
2313
2314                 req_fm_len = obd_object_end - lun_start;
2315                 fm_local->fm_length = 0;
2316                 len_mapped_single_call = 0;
2317
2318                 /* If the output buffer is very large and the objects have many
2319                  * extents we may need to loop on a single OST repeatedly */
2320                 ost_eof = 0;
2321                 ost_done = 0;
2322                 do {
2323                         if (get_num_extents == 0) {
2324                                 /* Don't get too many extents. */
2325                                 if (current_extent + count_local >
2326                                     fiemap->fm_extent_count)
2327                                         count_local = fiemap->fm_extent_count -
2328                                                                  current_extent;
2329                         }
2330
2331                         lun_start += len_mapped_single_call;
2332                         fm_local->fm_length = req_fm_len - len_mapped_single_call;
2333                         req_fm_len = fm_local->fm_length;
2334                         fm_local->fm_extent_count = count_local;
2335                         fm_local->fm_mapped_extents = 0;
2336                         fm_local->fm_flags = fiemap->fm_flags;
2337
2338                         fm_key->oa.o_id = lsm->lsm_oinfo[cur_stripe]->loi_id;
2339                         fm_key->oa.o_seq = lsm->lsm_oinfo[cur_stripe]->loi_seq;
2340                         ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
2341
2342                         if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
2343                                 GOTO(out, rc = -EINVAL);
2344
2345                         /* If OST is inactive, return extent with UNKNOWN flag */
2346                         if (!lov->lov_tgts[ost_index]->ltd_active) {
2347                                 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
2348                                 fm_local->fm_mapped_extents = 1;
2349
2350                                 lcl_fm_ext[0].fe_logical = lun_start;
2351                                 lcl_fm_ext[0].fe_length = obd_object_end -
2352                                                                       lun_start;
2353                                 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
2354
2355                                 goto inactive_tgt;
2356                         }
2357
2358                         fm_local->fm_start = lun_start;
2359                         fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
2360                         memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
2361                         *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
2362                         rc = obd_get_info(NULL,
2363                                           lov->lov_tgts[ost_index]->ltd_exp,
2364                                           keylen, key, vallen, fm_local, lsm);
2365                         if (rc != 0)
2366                                 GOTO(out, rc);
2367
2368 inactive_tgt:
2369                         ext_count = fm_local->fm_mapped_extents;
2370                         if (ext_count == 0) {
2371                                 ost_done = 1;
2372                                 /* If last stripe has hole at the end,
2373                                  * then we need to return */
2374                                 if (cur_stripe_wrap == last_stripe) {
2375                                         fiemap->fm_mapped_extents = 0;
2376                                         goto finish;
2377                                 }
2378                                 break;
2379                         }
2380
2381                         /* If we just need num of extents then go to next device */
2382                         if (get_num_extents) {
2383                                 current_extent += ext_count;
2384                                 break;
2385                         }
2386
2387                         len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2388                                   lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2389
2390                         /* Have we finished mapping on this device? */
2391                         if (req_fm_len <= len_mapped_single_call)
2392                                 ost_done = 1;
2393
2394                         /* Clear the EXTENT_LAST flag which can be present on
2395                          * last extent */
2396                         if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2397                                 lcl_fm_ext[ext_count - 1].fe_flags &=
2398                                                             ~FIEMAP_EXTENT_LAST;
2399
2400                         curr_loc = lov_stripe_size(lsm,
2401                                            lcl_fm_ext[ext_count - 1].fe_logical+
2402                                            lcl_fm_ext[ext_count - 1].fe_length,
2403                                            cur_stripe);
2404                         if (curr_loc >= fm_key->oa.o_size)
2405                                 ost_eof = 1;
2406
2407                         fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2408                                                      ost_index, ext_count,
2409                                                      current_extent);
2410
2411                         current_extent += ext_count;
2412
2413                         /* Ran out of available extents? */
2414                         if (current_extent >= fiemap->fm_extent_count)
2415                                 goto finish;
2416                 } while (ost_done == 0 && ost_eof == 0);
2417
2418                 if (cur_stripe_wrap == last_stripe)
2419                         goto finish;
2420         }
2421
2422 finish:
2423         /* Indicate that we are returning device offsets unless file just has
2424          * single stripe */
2425         if (lsm->lsm_stripe_count > 1)
2426                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2427
2428         if (get_num_extents)
2429                 goto skip_last_device_calc;
2430
2431         /* Check if we have reached the last stripe and whether mapping for that
2432          * stripe is done. */
2433         if (cur_stripe_wrap == last_stripe) {
2434                 if (ost_done || ost_eof)
2435                         fiemap->fm_extents[current_extent - 1].fe_flags |=
2436                                                              FIEMAP_EXTENT_LAST;
2437         }
2438
2439 skip_last_device_calc:
2440         fiemap->fm_mapped_extents = current_extent;
2441
2442 out:
2443         OBD_FREE_LARGE(fm_local, buffer_size);
2444         return rc;
2445 }
2446
2447 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2448                         __u32 keylen, void *key, __u32 *vallen, void *val,
2449                         struct lov_stripe_md *lsm)
2450 {
2451         struct obd_device *obddev = class_exp2obd(exp);
2452         struct lov_obd *lov = &obddev->u.lov;
2453         int i, rc;
2454         ENTRY;
2455
2456         if (!vallen || !val)
2457                 RETURN(-EFAULT);
2458
2459         obd_getref(obddev);
2460
2461         if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2462                 struct {
2463                         char name[16];
2464                         struct ldlm_lock *lock;
2465                 } *data = key;
2466                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2467                 struct lov_oinfo *loi;
2468                 __u32 *stripe = val;
2469
2470                 if (*vallen < sizeof(*stripe))
2471                         GOTO(out, rc = -EFAULT);
2472                 *vallen = sizeof(*stripe);
2473
2474                 /* XXX This is another one of those bits that will need to
2475                  * change if we ever actually support nested LOVs.  It uses
2476                  * the lock's export to find out which stripe it is. */
2477                 /* XXX - it's assumed all the locks for deleted OSTs have
2478                  * been cancelled. Also, the export for deleted OSTs will
2479                  * be NULL and won't match the lock's export. */
2480                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2481                         loi = lsm->lsm_oinfo[i];
2482                         if (!lov->lov_tgts[loi->loi_ost_idx])
2483                                 continue;
2484                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2485                             data->lock->l_conn_export &&
2486                             ostid_res_name_eq(&loi->loi_oi, res_id)) {
2487                                 *stripe = i;
2488                                 GOTO(out, rc = 0);
2489                         }
2490                 }
2491                 LDLM_ERROR(data->lock, "lock on inode without such object");
2492                 dump_lsm(D_ERROR, lsm);
2493                 GOTO(out, rc = -ENXIO);
2494         } else if (KEY_IS(KEY_LAST_ID)) {
2495                 struct obd_id_info *info = val;
2496                 __u32 size = sizeof(obd_id);
2497                 struct lov_tgt_desc *tgt;
2498
2499                 LASSERT(*vallen == sizeof(struct obd_id_info));
2500                 tgt = lov->lov_tgts[info->idx];
2501
2502                 if (!tgt || !tgt->ltd_active)
2503                         GOTO(out, rc = -ESRCH);
2504
2505                 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2506                                   &size, info->data, NULL);
2507                 GOTO(out, rc = 0);
2508         } else if (KEY_IS(KEY_LOVDESC)) {
2509                 struct lov_desc *desc_ret = val;
2510                 *desc_ret = lov->desc;
2511
2512                 GOTO(out, rc = 0);
2513         } else if (KEY_IS(KEY_FIEMAP)) {
2514                 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2515                 GOTO(out, rc);
2516         } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2517                 struct lov_tgt_desc *tgt;
2518                 __u64 ost_idx = *((__u64*)val);
2519
2520                 LASSERT(*vallen == sizeof(__u64));
2521                 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2522                 tgt = lov->lov_tgts[ost_idx];
2523
2524                 if (!tgt || !tgt->ltd_exp)
2525                         GOTO(out, rc = -ESRCH);
2526
2527                 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2528                 GOTO(out, rc = 0);
2529         } else if (KEY_IS(KEY_TGT_COUNT)) {
2530                 *((int *)val) = lov->desc.ld_tgt_count;
2531                 GOTO(out, rc = 0);
2532         }
2533
2534         rc = -EINVAL;
2535
2536 out:
2537         obd_putref(obddev);
2538         RETURN(rc);
2539 }
2540
2541 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2542                               obd_count keylen, void *key, obd_count vallen,
2543                               void *val, struct ptlrpc_request_set *set)
2544 {
2545         struct obd_device *obddev = class_exp2obd(exp);
2546         struct lov_obd *lov = &obddev->u.lov;
2547         obd_count count;
2548         int i, rc = 0, err;
2549         struct lov_tgt_desc *tgt;
2550         unsigned incr, check_uuid,
2551                  do_inactive, no_set;
2552         unsigned next_id = 0,  mds_con = 0, capa = 0;
2553         ENTRY;
2554
2555         incr = check_uuid = do_inactive = no_set = 0;
2556         if (set == NULL) {
2557                 no_set = 1;
2558                 set = ptlrpc_prep_set();
2559                 if (!set)
2560                         RETURN(-ENOMEM);
2561         }
2562
2563         obd_getref(obddev);
2564         count = lov->desc.ld_tgt_count;
2565
2566         if (KEY_IS(KEY_NEXT_ID)) {
2567                 count = vallen / sizeof(struct obd_id_info);
2568                 vallen = sizeof(obd_id);
2569                 incr = sizeof(struct obd_id_info);
2570                 do_inactive = 1;
2571                 next_id = 1;
2572         } else if (KEY_IS(KEY_CHECKSUM)) {
2573                 do_inactive = 1;
2574         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2575                 /* use defaults:  do_inactive = incr = 0; */
2576         } else if (KEY_IS(KEY_MDS_CONN)) {
2577                 mds_con = 1;
2578         } else if (KEY_IS(KEY_CAPA_KEY)) {
2579                 capa = 1;
2580         } else if (KEY_IS(KEY_CACHE_SET)) {
2581                 LASSERT(lov->lov_cache == NULL);
2582                 lov->lov_cache = val;
2583                 do_inactive = 1;
2584         }
2585
2586         for (i = 0; i < count; i++, val = (char *)val + incr) {
2587                 if (next_id) {
2588                         tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2589                 } else {
2590                         tgt = lov->lov_tgts[i];
2591                 }
2592                 /* OST was disconnected */
2593                 if (!tgt || !tgt->ltd_exp)
2594                         continue;
2595
2596                 /* OST is inactive and we don't want inactive OSCs */
2597                 if (!tgt->ltd_active && !do_inactive)
2598                         continue;
2599
2600                 if (mds_con) {
2601                         struct mds_group_info *mgi;
2602
2603                         LASSERT(vallen == sizeof(*mgi));
2604                         mgi = (struct mds_group_info *)val;
2605
2606                         /* Only want a specific OSC */
2607                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2608                                                 &tgt->ltd_uuid))
2609                                 continue;
2610
2611                         err = obd_set_info_async(env, tgt->ltd_exp,
2612                                          keylen, key, sizeof(int),
2613                                          &mgi->group, set);
2614                 } else if (next_id) {
2615                         err = obd_set_info_async(env, tgt->ltd_exp,
2616                                          keylen, key, vallen,
2617                                          ((struct obd_id_info*)val)->data, set);
2618                 } else if (capa) {
2619                         struct mds_capa_info *info = (struct mds_capa_info*)val;
2620
2621                         LASSERT(vallen == sizeof(*info));
2622
2623                          /* Only want a specific OSC */
2624                         if (info->uuid &&
2625                             !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2626                                 continue;
2627
2628                         err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2629                                                  key, sizeof(*info->capa),
2630                                                  info->capa, set);
2631                 } else {
2632                         /* Only want a specific OSC */
2633                         if (check_uuid &&
2634                             !obd_uuid_equals(val, &tgt->ltd_uuid))
2635                                 continue;
2636
2637                         err = obd_set_info_async(env, tgt->ltd_exp,
2638                                          keylen, key, vallen, val, set);
2639                 }
2640
2641                 if (!rc)
2642                         rc = err;
2643         }
2644
2645         obd_putref(obddev);
2646         if (no_set) {
2647                 err = ptlrpc_set_wait(set);
2648                 if (!rc)
2649                         rc = err;
2650                 ptlrpc_set_destroy(set);
2651         }
2652         RETURN(rc);
2653 }
2654
2655 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2656                            int cmd, __u64 *offset)
2657 {
2658         __u32 ssize = lsm->lsm_stripe_size;
2659         __u64 start;
2660
2661         start = *offset;
2662         lov_do_div64(start, ssize);
2663         start = start * ssize;
2664
2665         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2666                            ", end "LPU64"\n", *offset, ssize, start,
2667                            start + ssize - 1);
2668         if (cmd == OBD_CALC_STRIPE_END) {
2669                 *offset = start + ssize - 1;
2670         } else if (cmd == OBD_CALC_STRIPE_START) {
2671                 *offset = start;
2672         } else {
2673                 LBUG();
2674         }
2675
2676         RETURN(0);
2677 }
2678
2679 void lov_stripe_lock(struct lov_stripe_md *md)
2680 {
2681         LASSERT(md->lsm_lock_owner != cfs_curproc_pid());
2682         spin_lock(&md->lsm_lock);
2683         LASSERT(md->lsm_lock_owner == 0);
2684         md->lsm_lock_owner = cfs_curproc_pid();
2685 }
2686 EXPORT_SYMBOL(lov_stripe_lock);
2687
2688 void lov_stripe_unlock(struct lov_stripe_md *md)
2689 {
2690         LASSERT(md->lsm_lock_owner == cfs_curproc_pid());
2691         md->lsm_lock_owner = 0;
2692         spin_unlock(&md->lsm_lock);
2693 }
2694 EXPORT_SYMBOL(lov_stripe_unlock);
2695
2696 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2697                         struct obd_quotactl *oqctl)
2698 {
2699         struct lov_obd      *lov = &obd->u.lov;
2700         struct lov_tgt_desc *tgt;
2701         __u64                curspace = 0;
2702         __u64                bhardlimit = 0;
2703         int                  i, rc = 0;
2704         ENTRY;
2705
2706         if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2707             oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2708             oqctl->qc_cmd != Q_GETOQUOTA &&
2709             oqctl->qc_cmd != Q_INITQUOTA &&
2710             oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2711             oqctl->qc_cmd != Q_FINVALIDATE) {
2712                 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2713                 RETURN(-EFAULT);
2714         }
2715
2716         /* for lov tgt */
2717         obd_getref(obd);
2718         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2719                 int err;
2720
2721                 tgt = lov->lov_tgts[i];
2722
2723                 if (!tgt)
2724                         continue;
2725
2726                 if (!tgt->ltd_active || tgt->ltd_reap) {
2727                         if (oqctl->qc_cmd == Q_GETOQUOTA &&
2728                             lov->lov_tgts[i]->ltd_activate) {
2729                                 rc = -EREMOTEIO;
2730                                 CERROR("ost %d is inactive\n", i);
2731                         } else {
2732                                 CDEBUG(D_HA, "ost %d is inactive\n", i);
2733                         }
2734                         continue;
2735                 }
2736
2737                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2738                 if (err) {
2739                         if (tgt->ltd_active && !rc)
2740                                 rc = err;
2741                         continue;
2742                 }
2743
2744                 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2745                         curspace += oqctl->qc_dqblk.dqb_curspace;
2746                         bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2747                 }
2748         }
2749         obd_putref(obd);
2750
2751         if (oqctl->qc_cmd == Q_GETOQUOTA) {
2752                 oqctl->qc_dqblk.dqb_curspace = curspace;
2753                 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2754         }
2755         RETURN(rc);
2756 }
2757
2758 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2759                           struct obd_quotactl *oqctl)
2760 {
2761         struct lov_obd *lov = &obd->u.lov;
2762         int             i, rc = 0;
2763         ENTRY;
2764
2765         obd_getref(obd);
2766
2767         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2768                 if (!lov->lov_tgts[i])
2769                         continue;
2770
2771                 /* Skip quota check on the administratively disabled OSTs. */
2772                 if (!lov->lov_tgts[i]->ltd_activate) {
2773                         CWARN("lov idx %d was administratively disabled, "
2774                               "skip quotacheck on it.\n", i);
2775                         continue;
2776                 }
2777
2778                 if (!lov->lov_tgts[i]->ltd_active) {
2779                         CERROR("lov idx %d inactive\n", i);
2780                         rc = -EIO;
2781                         goto out;
2782                 }
2783         }
2784
2785         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2786                 int err;
2787
2788                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2789                         continue;
2790
2791                 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2792                 if (err && !rc)
2793                         rc = err;
2794         }
2795
2796 out:
2797         obd_putref(obd);
2798
2799         RETURN(rc);
2800 }
2801
2802 struct obd_ops lov_obd_ops = {
2803         .o_owner               = THIS_MODULE,
2804         .o_setup               = lov_setup,
2805         .o_precleanup          = lov_precleanup,
2806         .o_cleanup             = lov_cleanup,
2807         //.o_process_config      = lov_process_config,
2808         .o_connect             = lov_connect,
2809         .o_disconnect          = lov_disconnect,
2810         .o_statfs              = lov_statfs,
2811         .o_statfs_async        = lov_statfs_async,
2812         .o_packmd              = lov_packmd,
2813         .o_unpackmd            = lov_unpackmd,
2814         .o_create              = lov_create,
2815         .o_destroy             = lov_destroy,
2816         .o_getattr             = lov_getattr,
2817         .o_getattr_async       = lov_getattr_async,
2818         .o_setattr             = lov_setattr,
2819         .o_setattr_async       = lov_setattr_async,
2820         .o_brw                 = lov_brw,
2821         .o_merge_lvb           = lov_merge_lvb,
2822         .o_adjust_kms          = lov_adjust_kms,
2823         .o_punch               = lov_punch,
2824         .o_sync                = lov_sync,
2825         .o_enqueue             = lov_enqueue,
2826         .o_change_cbdata       = lov_change_cbdata,
2827         .o_find_cbdata         = lov_find_cbdata,
2828         .o_cancel              = lov_cancel,
2829         .o_cancel_unused       = lov_cancel_unused,
2830         .o_iocontrol           = lov_iocontrol,
2831         .o_get_info            = lov_get_info,
2832         .o_set_info_async      = lov_set_info_async,
2833         .o_extent_calc         = lov_extent_calc,
2834         .o_llog_init           = lov_llog_init,
2835         .o_llog_finish         = lov_llog_finish,
2836         .o_notify              = lov_notify,
2837         .o_pool_new            = lov_pool_new,
2838         .o_pool_rem            = lov_pool_remove,
2839         .o_pool_add            = lov_pool_add,
2840         .o_pool_del            = lov_pool_del,
2841         .o_getref              = lov_getref,
2842         .o_putref              = lov_putref,
2843         .o_quotactl            = lov_quotactl,
2844         .o_quotacheck          = lov_quotacheck,
2845 };
2846
2847 cfs_mem_cache_t *lov_oinfo_slab;
2848
2849 extern struct lu_kmem_descr lov_caches[];
2850
2851 int __init lov_init(void)
2852 {
2853         struct lprocfs_static_vars lvars = { 0 };
2854         int rc, rc2;
2855         ENTRY;
2856
2857         /* print an address of _any_ initialized kernel symbol from this
2858          * module, to allow debugging with gdb that doesn't support data
2859          * symbols from modules.*/
2860         CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2861
2862         rc = lu_kmem_init(lov_caches);
2863         if (rc)
2864                 return rc;
2865
2866         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2867                                               sizeof(struct lov_oinfo),
2868                                               0, CFS_SLAB_HWCACHE_ALIGN);
2869         if (lov_oinfo_slab == NULL) {
2870                 lu_kmem_fini(lov_caches);
2871                 return -ENOMEM;
2872         }
2873         lprocfs_lov_init_vars(&lvars);
2874
2875         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2876                                  LUSTRE_LOV_NAME, &lov_device_type);
2877
2878         if (rc) {
2879                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2880                 LASSERT(rc2 == 0);
2881                 lu_kmem_fini(lov_caches);
2882         }
2883
2884         RETURN(rc);
2885 }
2886
2887 #ifdef __KERNEL__
2888 static void /*__exit*/ lov_exit(void)
2889 {
2890         int rc;
2891
2892         class_unregister_type(LUSTRE_LOV_NAME);
2893         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2894         LASSERT(rc == 0);
2895
2896         lu_kmem_fini(lov_caches);
2897 }
2898
2899 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2900 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2901 MODULE_LICENSE("GPL");
2902
2903 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
2904 #endif