lustre/lov/lov_obd.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 2002-2006 Cluster File Systems, Inc.
   5  * Author: Phil Schwan <phil@clusterfs.com>
   6  *         Peter Braam <braam@clusterfs.com>
   7  *         Mike Shaver <shaver@clusterfs.com>
   8  *         Nathan Rutman <nathan@clusterfs.com>
   9  *
  10  *   This file is part of the Lustre file system, http://www.lustre.org
  11  *   Lustre is a trademark of Cluster File Systems, Inc.
  12  *
  13  *   You may have signed or agreed to another license before downloading
  14  *   this software.  If so, you are bound by the terms and conditions
  15  *   of that agreement, and the following does not apply to you.  See the
  16  *   LICENSE file included with this distribution for more information.
  17  *
  18  *   If you did not agree to a different license, then this copy of Lustre
  19  *   is open source software; you can redistribute it and/or modify it
  20  *   under the terms of version 2 of the GNU General Public License as
  21  *   published by the Free Software Foundation.
  22  *
  23  *   In either case, Lustre is distributed in the hope that it will be
  24  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  25  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  26  *   license text for more details.
  27  */
  28
  29 #ifndef EXPORT_SYMTAB
  30 # define EXPORT_SYMTAB
  31 #endif
  32 #define DEBUG_SUBSYSTEM S_LOV
  33 #ifdef __KERNEL__
  34 #include <libcfs/libcfs.h>
  35 #else
  36 #include <liblustre.h>
  37 #endif
  38
  39 #include <obd_support.h>
  40 #include <lustre_lib.h>
  41 #include <lustre_net.h>
  42 #include <lustre/lustre_idl.h>
  43 #include <lustre_dlm.h>
  44 #include <lustre_mds.h>
  45 #include <lustre_debug.h>
  46 #include <obd_class.h>
  47 #include <obd_lov.h>
  48 #include <obd_ost.h>
  49 #include <lprocfs_status.h>
  50 #include <lustre_param.h>
  51
  52 #include "lov_internal.h"
  53
  54
  55 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
  56    Any function that expects lov_tgts to remain stationary must take a ref. */
  57 void lov_getref(struct obd_device *obd)
  58 {
  59         struct lov_obd *lov = &obd->u.lov;
  60
  61         /* nobody gets through here until lov_putref is done */
  62         mutex_down(&lov->lov_lock);
  63         atomic_inc(&lov->lov_refcount);
  64         mutex_up(&lov->lov_lock);
  65         return;
  66 }
  67
  68 static void __lov_del_obd(struct obd_device *obd, __u32 index);
  69
  70 void lov_putref(struct obd_device *obd)
  71 {
  72         struct lov_obd *lov = &obd->u.lov;
  73         mutex_down(&lov->lov_lock);
  74         /* ok to dec to 0 more than once -- ltd_exp's will be null */
  75         if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
  76                 int i;
  77                 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
  78                        lov->lov_death_row);
  79                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
  80                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_reap)
  81                                 continue;
  82                         /* Disconnect and delete from list */
  83                         __lov_del_obd(obd, i);
  84                         lov->lov_death_row--;
  85                 }
  86         }
  87         mutex_up(&lov->lov_lock);
  88 }
  89
  90 #define MAX_STRING_SIZE 128
  91 static int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
  92                            struct obd_connect_data *data)
  93 {
  94         struct lov_obd *lov = &obd->u.lov;
  95         struct obd_uuid tgt_uuid = lov->lov_tgts[index]->ltd_uuid;
  96         struct obd_device *tgt_obd;
  97         struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
  98         struct lustre_handle conn = {0, };
  99         struct obd_import *imp;
 100 #ifdef __KERNEL__
 101         cfs_proc_dir_entry_t *lov_proc_dir;
 102 #endif
 103         int rc;
 104         ENTRY;
 105
 106         if (!lov->lov_tgts[index])
 107                 RETURN(-EINVAL);
 108
 109         tgt_obd = class_find_client_obd(&tgt_uuid, LUSTRE_OSC_NAME,
 110                                         &obd->obd_uuid);
 111
 112         if (!tgt_obd) {
 113                 CERROR("Target %s not attached\n", obd_uuid2str(&tgt_uuid));
 114                 RETURN(-EINVAL);
 115         }
 116         if (!tgt_obd->obd_set_up) {
 117                 CERROR("Target %s not set up\n", obd_uuid2str(&tgt_uuid));
 118                 RETURN(-EINVAL);
 119         }
 120
 121         if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
 122                 data->ocd_index = index;
 123
 124         /*
 125          * Divine LOV knows that OBDs under it are OSCs.
 126          */
 127         imp = tgt_obd->u.cli.cl_import;
 128
 129         if (activate) {
 130                 tgt_obd->obd_no_recov = 0;
 131                 /* FIXME this is probably supposed to be
 132                    ptlrpc_set_import_active.  Horrible naming. */
 133                 ptlrpc_activate_import(imp);
 134         }
 135
 136         if (imp->imp_invalid) {
 137                 CERROR("not connecting OSC %s; administratively "
 138                        "disabled\n", obd_uuid2str(&tgt_uuid));
 139                 rc = obd_register_observer(tgt_obd, obd);
 140                 if (rc) {
 141                         CERROR("Target %s register_observer error %d; "
 142                                "will not be able to reactivate\n",
 143                                obd_uuid2str(&tgt_uuid), rc);
 144                 }
 145                 RETURN(0);
 146         }
 147
 148         rc = obd_connect(NULL, &conn, tgt_obd, &lov_osc_uuid, data);
 149         if (rc) {
 150                 CERROR("Target %s connect error %d\n",
 151                        obd_uuid2str(&tgt_uuid), rc);
 152                 RETURN(rc);
 153         }
 154         lov->lov_tgts[index]->ltd_exp = class_conn2export(&conn);
 155         if (!lov->lov_tgts[index]->ltd_exp) {
 156                 CERROR("Target %s: null export!\n", obd_uuid2str(&tgt_uuid));
 157                 RETURN(-ENODEV);
 158         }
 159
 160         rc = obd_register_observer(tgt_obd, obd);
 161         if (rc) {
 162                 CERROR("Target %s register_observer error %d\n",
 163                        obd_uuid2str(&tgt_uuid), rc);
 164                 obd_disconnect(lov->lov_tgts[index]->ltd_exp);
 165                 lov->lov_tgts[index]->ltd_exp = NULL;
 166                 RETURN(rc);
 167         }
 168
 169         lov->lov_tgts[index]->ltd_reap = 0;
 170         if (activate) {
 171                 lov->lov_tgts[index]->ltd_active = 1;
 172                 lov->desc.ld_active_tgt_count++;
 173         }
 174         CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
 175                obd_uuid2str(&tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
 176
 177 #ifdef __KERNEL__
 178         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 179         if (lov_proc_dir) {
 180                 struct obd_device *osc_obd = class_conn2obd(&conn);
 181                 cfs_proc_dir_entry_t *osc_symlink;
 182                 char name[MAX_STRING_SIZE];
 183
 184                 LASSERT(osc_obd != NULL);
 185                 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
 186                 LASSERT(osc_obd->obd_type->typ_name != NULL);
 187                 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
 188                          osc_obd->obd_type->typ_name,
 189                          osc_obd->obd_name);
 190                 osc_symlink = proc_symlink(osc_obd->obd_name, lov_proc_dir,
 191                                            name);
 192                 if (osc_symlink == NULL) {
 193                         CERROR("could not register LOV target "
 194                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
 195                                obd->obd_type->typ_name, obd->obd_name,
 196                                osc_obd->obd_name);
 197                         lprocfs_remove(&lov_proc_dir);
 198                 }
 199         }
 200 #endif
 201
 202         rc = qos_add_tgt(obd, index);
 203         if (rc)
 204                 CERROR("qos_add_tgt failed %d\n", rc);
 205
 206         RETURN(0);
 207 }
 208
 209 static int lov_connect(const struct lu_env *env,
 210                        struct lustre_handle *conn, struct obd_device *obd,
 211                        struct obd_uuid *cluuid, struct obd_connect_data *data)
 212 {
 213         struct lov_obd *lov = &obd->u.lov;
 214         struct lov_tgt_desc *tgt;
 215         int i, rc;
 216         ENTRY;
 217
 218         CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
 219
 220         rc = class_connect(conn, obd, cluuid);
 221         if (rc)
 222                 RETURN(rc);
 223
 224         /* Why should there ever be more than 1 connect? */
 225         lov->lov_connects++;
 226         LASSERT(lov->lov_connects == 1);
 227
 228         memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
 229         if (data)
 230                 lov->lov_ocd = *data;
 231
 232         lov_getref(obd);
 233         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 234                 tgt = lov->lov_tgts[i];
 235                 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
 236                         continue;
 237                 /* Flags will be lowest common denominator */
 238                 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
 239                 if (rc) {
 240                         CERROR("%s: lov connect tgt %d failed: %d\n",
 241                                obd->obd_name, i, rc);
 242                         continue;
 243                 }
 244         }
 245         lov_putref(obd);
 246
 247         RETURN(0);
 248 }
 249
 250 static int lov_disconnect_obd(struct obd_device *obd, __u32 index)
 251 {
 252         cfs_proc_dir_entry_t *lov_proc_dir;
 253         struct lov_obd *lov = &obd->u.lov;
 254         struct obd_device *osc_obd =
 255                 class_exp2obd(lov->lov_tgts[index]->ltd_exp);
 256         int rc;
 257         ENTRY;
 258
 259         CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
 260                obd->obd_name, osc_obd->obd_name);
 261
 262         if (lov->lov_tgts[index]->ltd_active) {
 263                 lov->lov_tgts[index]->ltd_active = 0;
 264                 lov->desc.ld_active_tgt_count--;
 265         }
 266
 267         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 268         if (lov_proc_dir) {
 269                 cfs_proc_dir_entry_t *osc_symlink;
 270
 271                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
 272                 if (osc_symlink) {
 273                         lprocfs_remove(&osc_symlink);
 274                 } else {
 275                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing.",
 276                                obd->obd_type->typ_name, obd->obd_name,
 277                                osc_obd->obd_name);
 278                 }
 279         }
 280
 281         if (osc_obd) {
 282                 /* Pass it on to our clients.
 283                  * XXX This should be an argument to disconnect,
 284                  * XXX not a back-door flag on the OBD.  Ah well.
 285                  */
 286                 osc_obd->obd_force = obd->obd_force;
 287                 osc_obd->obd_fail = obd->obd_fail;
 288                 osc_obd->obd_no_recov = obd->obd_no_recov;
 289         }
 290
 291         obd_register_observer(osc_obd, NULL);
 292
 293         rc = obd_disconnect(lov->lov_tgts[index]->ltd_exp);
 294         if (rc) {
 295                 CERROR("Target %s disconnect error %d\n",
 296                        lov_uuid2str(lov, index), rc);
 297                 rc = 0;
 298         }
 299
 300         qos_del_tgt(obd, index);
 301
 302         lov->lov_tgts[index]->ltd_exp = NULL;
 303         RETURN(0);
 304 }
 305
 306 static int lov_del_target(struct obd_device *obd, __u32 index,
 307                           struct obd_uuid *uuidp, int gen);
 308
 309 static int lov_disconnect(struct obd_export *exp)
 310 {
 311         struct obd_device *obd = class_exp2obd(exp);
 312         struct lov_obd *lov = &obd->u.lov;
 313         int i, rc;
 314         ENTRY;
 315
 316         if (!lov->lov_tgts)
 317                 goto out;
 318
 319         /* Only disconnect the underlying layers on the final disconnect. */
 320         lov->lov_connects--;
 321         if (lov->lov_connects != 0) {
 322                 /* why should there be more than 1 connect? */
 323                 CERROR("disconnect #%d\n", lov->lov_connects);
 324                 goto out;
 325         }
 326
 327         /* Let's hold another reference so lov_del_obd doesn't spin through
 328            putref every time */
 329         lov_getref(obd);
 330         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 331                 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
 332                         /* Disconnection is the last we know about an obd */
 333                         lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
 334                 }
 335         }
 336         lov_putref(obd);
 337
 338 out:
 339         rc = class_disconnect(exp); /* bz 9811 */
 340         RETURN(rc);
 341 }
 342
 343 /* Error codes:
 344  *
 345  *  -EINVAL  : UUID can't be found in the LOV's target list
 346  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 347  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
 348  */
 349 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
 350                               int activate)
 351 {
 352         struct lov_obd *lov = &obd->u.lov;
 353         struct lov_tgt_desc *tgt;
 354         int i, rc = 0;
 355         ENTRY;
 356
 357         CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
 358                lov, uuid->uuid, activate);
 359
 360         lov_getref(obd);
 361         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 362                 tgt = lov->lov_tgts[i];
 363                 if (!tgt || !tgt->ltd_exp)
 364                         continue;
 365
 366                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
 367                        i, obd_uuid2str(&tgt->ltd_uuid),
 368                        tgt->ltd_exp->exp_handle.h_cookie);
 369                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
 370                         break;
 371         }
 372
 373         if (i == lov->desc.ld_tgt_count)
 374                 GOTO(out, rc = -EINVAL);
 375
 376         if (lov->lov_tgts[i]->ltd_active == activate) {
 377                 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
 378                        activate ? "" : "in");
 379                 GOTO(out, rc);
 380         }
 381
 382         CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n", obd_uuid2str(uuid),
 383                activate ? "" : "in");
 384
 385         lov->lov_tgts[i]->ltd_active = activate;
 386         if (activate)
 387                 lov->desc.ld_active_tgt_count++;
 388         else
 389                 lov->desc.ld_active_tgt_count--;
 390         /* remove any old qos penalty */
 391         lov->lov_tgts[i]->ltd_qos.ltq_penalty = 0;
 392
 393  out:
 394         lov_putref(obd);
 395         RETURN(rc);
 396 }
 397
 398 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 399                       enum obd_notify_event ev, void *data)
 400 {
 401         int rc = 0;
 402         ENTRY;
 403
 404         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
 405                 struct obd_uuid *uuid;
 406
 407                 LASSERT(watched);
 408
 409                 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 410                         CERROR("unexpected notification of %s %s!\n",
 411                                watched->obd_type->typ_name,
 412                                watched->obd_name);
 413                         RETURN(-EINVAL);
 414                 }
 415                 uuid = &watched->u.cli.cl_target_uuid;
 416
 417                 /* Set OSC as active before notifying the observer, so the
 418                  * observer can use the OSC normally.
 419                  */
 420                 rc = lov_set_osc_active(obd, uuid, ev == OBD_NOTIFY_ACTIVE);
 421                 if (rc) {
 422                         CERROR("%sactivation of %s failed: %d\n",
 423                                (ev == OBD_NOTIFY_ACTIVE) ? "" : "de",
 424                                obd_uuid2str(uuid), rc);
 425                         RETURN(rc);
 426                 }
 427         }
 428
 429         /* Pass the notification up the chain. */
 430         if (watched) {
 431                 rc = obd_notify_observer(obd, watched, ev, data);
 432         } else {
 433                 /* NULL watched means all osc's in the lov (only for syncs) */
 434                 struct lov_obd *lov = &obd->u.lov;
 435                 struct obd_device *tgt_obd;
 436                 int i;
 437                 lov_getref(obd);
 438                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 439                         if (!lov->lov_tgts[i])
 440                                 continue;
 441                         tgt_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
 442                         rc = obd_notify_observer(obd, tgt_obd, ev, data);
 443                         if (rc) {
 444                                 CERROR("%s: notify %s of %s failed %d\n",
 445                                        obd->obd_name,
 446                                        obd->obd_observer->obd_name,
 447                                        tgt_obd->obd_name, rc);
 448                                 break;
 449                         }
 450                 }
 451                 lov_putref(obd);
 452         }
 453
 454         RETURN(rc);
 455 }
 456
 457 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
 458                           __u32 index, int gen, int active)
 459 {
 460         struct lov_obd *lov = &obd->u.lov;
 461         struct lov_tgt_desc *tgt;
 462         int rc;
 463         ENTRY;
 464
 465         CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
 466                uuidp->uuid, index, gen, active);
 467
 468         if (gen <= 0) {
 469                 CERROR("request to add OBD %s with invalid generation: %d\n",
 470                        uuidp->uuid, gen);
 471                 RETURN(-EINVAL);
 472         }
 473
 474         mutex_down(&lov->lov_lock);
 475
 476         if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
 477                 tgt = lov->lov_tgts[index];
 478                 CERROR("UUID %s already assigned at LOV target index %d\n",
 479                        obd_uuid2str(&tgt->ltd_uuid), index);
 480                 mutex_up(&lov->lov_lock);
 481                 RETURN(-EEXIST);
 482         }
 483
 484         if (index >= lov->lov_tgt_size) {
 485                 /* We need to reallocate the lov target array. */
 486                 struct lov_tgt_desc **newtgts, **old = NULL;
 487                 __u32 newsize, oldsize = 0;
 488
 489                 newsize = max(lov->lov_tgt_size, (__u32)2);
 490                 while (newsize < index + 1)
 491                         newsize = newsize << 1;
 492                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
 493                 if (newtgts == NULL) {
 494                         mutex_up(&lov->lov_lock);
 495                         RETURN(-ENOMEM);
 496                 }
 497
 498                 if (lov->lov_tgt_size) {
 499                         memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
 500                                lov->lov_tgt_size);
 501                         old = lov->lov_tgts;
 502                         oldsize = lov->lov_tgt_size;
 503                 }
 504
 505                 lov->lov_tgts = newtgts;
 506                 lov->lov_tgt_size = newsize;
 507 #ifdef __KERNEL__
 508                 smp_rmb();
 509 #endif
 510                 if (old)
 511                         OBD_FREE(old, sizeof(*old) * oldsize);
 512
 513                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
 514                        lov->lov_tgts, lov->lov_tgt_size);
 515         }
 516
 517
 518         OBD_ALLOC_PTR(tgt);
 519         if (!tgt) {
 520                 mutex_up(&lov->lov_lock);
 521                 RETURN(-ENOMEM);
 522         }
 523
 524         memset(tgt, 0, sizeof(*tgt));
 525         tgt->ltd_uuid = *uuidp;
 526         /* XXX - add a sanity check on the generation number. */
 527         tgt->ltd_gen = gen;
 528         tgt->ltd_index = index;
 529         tgt->ltd_activate = active;
 530         lov->lov_tgts[index] = tgt;
 531         if (index >= lov->desc.ld_tgt_count)
 532                 lov->desc.ld_tgt_count = index + 1;
 533         mutex_up(&lov->lov_lock);
 534
 535         CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
 536                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 537
 538         if (lov->lov_connects == 0) {
 539                 /* lov_connect hasn't been called yet. We'll do the
 540                    lov_connect_obd on this target when that fn first runs,
 541                    because we don't know the connect flags yet. */
 542                 RETURN(0);
 543         }
 544
 545         lov_getref(obd);
 546
 547         rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
 548         if (rc)
 549                 GOTO(out, rc);
 550
 551         rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
 552                         active ? OBD_NOTIFY_ACTIVE : OBD_NOTIFY_INACTIVE,
 553                         (void *)&index);
 554
 555 out:
 556         if (rc) {
 557                 CERROR("add failed (%d), deleting %s\n", rc,
 558                        obd_uuid2str(&tgt->ltd_uuid));
 559                 lov_del_target(obd, index, 0, 0);
 560         }
 561         lov_putref(obd);
 562         RETURN(rc);
 563 }
 564
 565 /* Schedule a target for deletion */
 566 static int lov_del_target(struct obd_device *obd, __u32 index,
 567                           struct obd_uuid *uuidp, int gen)
 568 {
 569         struct lov_obd *lov = &obd->u.lov;
 570         int count = lov->desc.ld_tgt_count;
 571         int rc = 0;
 572         ENTRY;
 573
 574         if (index >= count) {
 575                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 576                        index, count);
 577                 RETURN(-EINVAL);
 578         }
 579
 580         lov_getref(obd);
 581
 582         if (!lov->lov_tgts[index]) {
 583                 CERROR("LOV target at index %d is not setup.\n", index);
 584                 GOTO(out, rc = -EINVAL);
 585         }
 586
 587         if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
 588                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 589                        lov_uuid2str(lov, index), index,
 590                        obd_uuid2str(uuidp));
 591                 GOTO(out, rc = -EINVAL);
 592         }
 593
 594         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 595                lov_uuid2str(lov, index), index,
 596                lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
 597                lov->lov_tgts[index]->ltd_active);
 598
 599         lov->lov_tgts[index]->ltd_reap = 1;
 600         lov->lov_death_row++;
 601         /* we really delete it from lov_putref */
 602 out:
 603         lov_putref(obd);
 604
 605         RETURN(rc);
 606 }
 607
 608 /* We are holding lov_lock */
 609 static void __lov_del_obd(struct obd_device *obd, __u32 index)
 610 {
 611         struct lov_obd *lov = &obd->u.lov;
 612         struct obd_device *osc_obd;
 613         struct lov_tgt_desc *tgt = lov->lov_tgts[index];
 614
 615         LASSERT(tgt);
 616         LASSERT(tgt->ltd_reap);
 617
 618         osc_obd = class_exp2obd(tgt->ltd_exp);
 619
 620         CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
 621                lov_uuid2str(lov, index),
 622                osc_obd ? osc_obd->obd_name : "<no obd>");
 623
 624         if (tgt->ltd_exp)
 625                 lov_disconnect_obd(obd, index);
 626
 627         /* XXX - right now there is a dependency on ld_tgt_count being the
 628          * maximum tgt index for computing the mds_max_easize. So we can't
 629          * shrink it. */
 630
 631         lov->lov_tgts[index] = NULL;
 632         OBD_FREE_PTR(tgt);
 633
 634         /* Manual cleanup - no cleanup logs to clean up the osc's.  We must
 635            do it ourselves. And we can't do it from lov_cleanup,
 636            because we just lost our only reference to it. */
 637         if (osc_obd)
 638                 class_manual_cleanup(osc_obd);
 639 }
 640
 641 void lov_fix_desc(struct lov_desc *desc)
 642 {
 643         if (desc->ld_default_stripe_size < PTLRPC_MAX_BRW_SIZE) {
 644                 LCONSOLE_WARN("Increasing default stripe size to min %u\n",
 645                               PTLRPC_MAX_BRW_SIZE);
 646                 desc->ld_default_stripe_size = PTLRPC_MAX_BRW_SIZE;
 647         } else if (desc->ld_default_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
 648                 desc->ld_default_stripe_size &= ~(LOV_MIN_STRIPE_SIZE - 1);
 649                 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
 650                               "multiple of %u)\n",
 651                               desc->ld_default_stripe_size,LOV_MIN_STRIPE_SIZE);
 652         }
 653
 654         if (desc->ld_default_stripe_count == 0)
 655                 desc->ld_default_stripe_count = 1;
 656
 657         /* from lov_setstripe */
 658         if ((desc->ld_pattern != 0) &&
 659             (desc->ld_pattern != LOV_PATTERN_RAID0)) {
 660                 LCONSOLE_WARN("Unknown stripe pattern: %#x\n",desc->ld_pattern);
 661                 desc->ld_pattern = 0;
 662         }
 663
 664         if (desc->ld_qos_maxage == 0)
 665                 desc->ld_qos_maxage = QOS_DEFAULT_MAXAGE;
 666 }
 667
 668 static int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
 669 {
 670         struct lprocfs_static_vars lvars;
 671         struct lov_desc *desc;
 672         struct lov_obd *lov = &obd->u.lov;
 673         int count;
 674         ENTRY;
 675
 676         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
 677                 CERROR("LOV setup requires a descriptor\n");
 678                 RETURN(-EINVAL);
 679         }
 680
 681         desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
 682
 683         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
 684                 CERROR("descriptor size wrong: %d > %d\n",
 685                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
 686                 RETURN(-EINVAL);
 687         }
 688
 689         if (desc->ld_magic != LOV_DESC_MAGIC) {
 690                 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
 691                             CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
 692                                    obd->obd_name, desc);
 693                             lustre_swab_lov_desc(desc);
 694                 } else {
 695                         CERROR("%s: Bad lov desc magic: %#x\n",
 696                                obd->obd_name, desc->ld_magic);
 697                         RETURN(-EINVAL);
 698                 }
 699         }
 700
 701         lov_fix_desc(desc);
 702
 703         /* Because of 64-bit divide/mod operations only work with a 32-bit
 704          * divisor in a 32-bit kernel, we cannot support a stripe width
 705          * of 4GB or larger on 32-bit CPUs. */
 706         count = desc->ld_default_stripe_count;
 707         if ((count > 0 ? count : desc->ld_tgt_count) *
 708             desc->ld_default_stripe_size > 0xffffffff) {
 709                 CERROR("LOV: stripe width "LPU64"x%u > 4294967295 bytes\n",
 710                        desc->ld_default_stripe_size, count);
 711                 RETURN(-EINVAL);
 712         }
 713
 714         desc->ld_active_tgt_count = 0;
 715         lov->desc = *desc;
 716         lov->lov_tgt_size = 0;
 717         sema_init(&lov->lov_lock, 1);
 718         atomic_set(&lov->lov_refcount, 0);
 719         INIT_LIST_HEAD(&lov->lov_qos.lq_oss_list);
 720         init_rwsem(&lov->lov_qos.lq_rw_sem);
 721         lov->lov_qos.lq_dirty = 1;
 722         lov->lov_qos.lq_dirty_rr = 1;
 723         lov->lov_qos.lq_reset = 1;
 724         /* Default priority is toward free space balance */
 725         lov->lov_qos.lq_prio_free = 232;
 726
 727         lprocfs_init_vars(lov, &lvars);
 728         lprocfs_obd_setup(obd, lvars.obd_vars);
 729 #ifdef LPROCFS
 730         {
 731                 cfs_proc_dir_entry_t *entry;
 732
 733                 entry = create_proc_entry("target_obd", 0444,
 734                                           obd->obd_proc_entry);
 735                 if (entry != NULL) {
 736                         entry->proc_fops = &lov_proc_target_fops;
 737                         entry->data = obd;
 738                 }
 739         }
 740 #endif
 741
 742         RETURN(0);
 743 }
 744
 745 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
 746 {
 747         int rc = 0;
 748         ENTRY;
 749
 750         switch (stage) {
 751         case OBD_CLEANUP_EARLY: {
 752                 struct lov_obd *lov = &obd->u.lov;
 753                 int i;
 754                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 755                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
 756                                 continue;
 757                         obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
 758                                        OBD_CLEANUP_EARLY);
 759                 }
 760                 break;
 761         }
 762         case OBD_CLEANUP_EXPORTS:
 763                 break;
 764         case OBD_CLEANUP_SELF_EXP:
 765                 rc = obd_llog_finish(obd, 0);
 766                 if (rc != 0)
 767                         CERROR("failed to cleanup llogging subsystems\n");
 768                 break;
 769         case OBD_CLEANUP_OBD:
 770                 break;
 771         }
 772         RETURN(rc);
 773 }
 774
 775 static int lov_cleanup(struct obd_device *obd)
 776 {
 777         struct lov_obd *lov = &obd->u.lov;
 778
 779         lprocfs_obd_cleanup(obd);
 780         if (lov->lov_tgts) {
 781                 int i;
 782                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 783                         if (lov->lov_tgts[i]) {
 784                                 /* Inactive targets may never have connected */
 785                                 if (lov->lov_tgts[i]->ltd_active ||
 786                                     atomic_read(&lov->lov_refcount))
 787                                         /* We should never get here - these
 788                                            should have been removed in the
 789                                            disconnect. */
 790                                         CERROR("lov tgt %d not cleaned!"
 791                                                " deathrow=%d, lovrc=%d\n",
 792                                                i, lov->lov_death_row,
 793                                                atomic_read(&lov->lov_refcount));
 794                                 lov_del_target(obd, i, 0, 0);
 795                         }
 796                 }
 797                 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
 798                          lov->lov_tgt_size);
 799                 lov->lov_tgt_size = 0;
 800         }
 801
 802         if (lov->lov_qos.lq_rr_size)
 803                 OBD_FREE(lov->lov_qos.lq_rr_array, lov->lov_qos.lq_rr_size);
 804
 805         RETURN(0);
 806 }
 807
 808 static int lov_process_config(struct obd_device *obd, obd_count len, void *buf)
 809 {
 810         struct lustre_cfg *lcfg = buf;
 811         struct obd_uuid obd_uuid;
 812         int cmd;
 813         int rc = 0;
 814         ENTRY;
 815
 816         switch(cmd = lcfg->lcfg_command) {
 817         case LCFG_LOV_ADD_OBD:
 818         case LCFG_LOV_ADD_INA:
 819         case LCFG_LOV_DEL_OBD: {
 820                 __u32 index;
 821                 int gen;
 822                 /* lov_modify_tgts add  0:lov_mdsA  1:ost1_UUID  2:0  3:1 */
 823                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
 824                         GOTO(out, rc = -EINVAL);
 825
 826                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
 827
 828                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
 829                         GOTO(out, rc = -EINVAL);
 830                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
 831                         GOTO(out, rc = -EINVAL);
 832                 if (cmd == LCFG_LOV_ADD_OBD)
 833                         rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
 834                 else if (cmd == LCFG_LOV_ADD_INA)
 835                         rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
 836                 else
 837                         rc = lov_del_target(obd, index, &obd_uuid, gen);
 838                 GOTO(out, rc);
 839         }
 840         case LCFG_PARAM: {
 841                 struct lprocfs_static_vars lvars;
 842                 struct lov_desc *desc = &(obd->u.lov.desc);
 843
 844                 if (!desc)
 845                         GOTO(out, rc = -EINVAL);
 846
 847                 lprocfs_init_vars(lov, &lvars);
 848
 849                 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
 850                                               lcfg, obd);
 851                 GOTO(out, rc);
 852         }
 853         default: {
 854                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 855                 GOTO(out, rc = -EINVAL);
 856
 857         }
 858         }
 859 out:
 860         RETURN(rc);
 861 }
 862
 863 #ifndef log2
 864 #define log2(n) ffz(~(n))
 865 #endif
 866
 867 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
 868                              struct lov_stripe_md **ea,
 869                              struct obd_trans_info *oti)
 870 {
 871         struct lov_obd *lov;
 872         struct obdo *tmp_oa;
 873         struct obd_uuid *ost_uuid = NULL;
 874         int rc = 0, i;
 875         ENTRY;
 876
 877         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
 878                 src_oa->o_flags == OBD_FL_DELORPHAN);
 879
 880         lov = &export->exp_obd->u.lov;
 881
 882         OBDO_ALLOC(tmp_oa);
 883         if (tmp_oa == NULL)
 884                 RETURN(-ENOMEM);
 885
 886         if (src_oa->o_valid & OBD_MD_FLINLINE) {
 887                 ost_uuid = (struct obd_uuid *)src_oa->o_inline;
 888                 CDEBUG(D_HA, "clearing orphans only for %s\n",
 889                        ost_uuid->uuid);
 890         }
 891
 892         lov_getref(export->exp_obd);
 893         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 894                 struct lov_stripe_md obj_md;
 895                 struct lov_stripe_md *obj_mdp = &obj_md;
 896                 struct lov_tgt_desc *tgt;
 897                 int err;
 898
 899                 tgt = lov->lov_tgts[i];
 900                 if (!tgt)
 901                         continue;
 902
 903                 /* if called for a specific target, we don't
 904                    care if it is not active. */
 905                 if (!lov->lov_tgts[i]->ltd_active && ost_uuid == NULL) {
 906                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
 907                         continue;
 908                 }
 909
 910                 if (ost_uuid && !obd_uuid_equals(ost_uuid, &tgt->ltd_uuid))
 911                         continue;
 912
 913                 CDEBUG(D_CONFIG,"Clear orphans for %d:%s\n", i,
 914                        obd_uuid2str(ost_uuid));
 915
 916                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
 917
 918                 LASSERT(lov->lov_tgts[i]->ltd_exp);
 919                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
 920                 err = obd_create(lov->lov_tgts[i]->ltd_exp,
 921                                  tmp_oa, &obj_mdp, oti);
 922                 if (err)
 923                         /* This export will be disabled until it is recovered,
 924                            and then orphan recovery will be completed. */
 925                         CERROR("error in orphan recovery on OST idx %d/%d: "
 926                                "rc = %d\n", i, lov->desc.ld_tgt_count, err);
 927
 928                 if (ost_uuid)
 929                         break;
 930         }
 931         lov_putref(export->exp_obd);
 932
 933         OBDO_FREE(tmp_oa);
 934         RETURN(rc);
 935 }
 936
 937 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
 938                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
 939 {
 940         struct lov_stripe_md *obj_mdp, *lsm;
 941         struct lov_obd *lov = &exp->exp_obd->u.lov;
 942         unsigned ost_idx;
 943         int rc, i;
 944         ENTRY;
 945
 946         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
 947                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
 948
 949         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
 950         if (obj_mdp == NULL)
 951                 RETURN(-ENOMEM);
 952
 953         ost_idx = src_oa->o_nlink;
 954         lsm = *ea;
 955         if (lsm == NULL)
 956                 GOTO(out, rc = -EINVAL);
 957         if (ost_idx >= lov->desc.ld_tgt_count ||
 958             !lov->lov_tgts[ost_idx])
 959                 GOTO(out, rc = -EINVAL);
 960
 961         for (i = 0; i < lsm->lsm_stripe_count; i++) {
 962                 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
 963                         if (lsm->lsm_oinfo[i]->loi_id != src_oa->o_id)
 964                                 GOTO(out, rc = -EINVAL);
 965                         break;
 966                 }
 967         }
 968         if (i == lsm->lsm_stripe_count)
 969                 GOTO(out, rc = -EINVAL);
 970
 971         rc = obd_create(lov->lov_tgts[ost_idx]->ltd_exp, src_oa, &obj_mdp, oti);
 972 out:
 973         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
 974         RETURN(rc);
 975 }
 976
 977 /* the LOV expects oa->o_id to be set to the LOV object id */
 978 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
 979                       struct lov_stripe_md **ea, struct obd_trans_info *oti)
 980 {
 981         struct lov_obd *lov;
 982         struct obd_info oinfo;
 983         struct lov_request_set *set = NULL;
 984         struct lov_request *req;
 985         struct obd_statfs osfs;
 986         __u64 maxage;
 987         int rc = 0;
 988         ENTRY;
 989
 990         LASSERT(ea != NULL);
 991         if (exp == NULL)
 992                 RETURN(-EINVAL);
 993
 994         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
 995             src_oa->o_flags == OBD_FL_DELORPHAN) {
 996                 rc = lov_clear_orphans(exp, src_oa, ea, oti);
 997                 RETURN(rc);
 998         }
 999
1000         lov = &exp->exp_obd->u.lov;
1001         if (!lov->desc.ld_active_tgt_count)
1002                 RETURN(-EIO);
1003
1004         /* Recreate a specific object id at the given OST index */
1005         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1006             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1007                  rc = lov_recreate(exp, src_oa, ea, oti);
1008                  RETURN(rc);
1009         }
1010
1011         maxage = cfs_time_shift_64(-lov->desc.ld_qos_maxage);
1012         obd_statfs_rqset(exp->exp_obd, &osfs, maxage);
1013
1014         rc = lov_prep_create_set(exp, &oinfo, ea, src_oa, oti, &set);
1015         if (rc)
1016                 RETURN(rc);
1017
1018         list_for_each_entry(req, &set->set_list, rq_link) {
1019                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1020                 rc = obd_create(lov->lov_tgts[req->rq_idx]->ltd_exp,
1021                                 req->rq_oi.oi_oa, &req->rq_oi.oi_md, oti);
1022                 lov_update_create_set(set, req, rc);
1023         }
1024         rc = lov_fini_create_set(set, ea);
1025         RETURN(rc);
1026 }
1027
1028 #define ASSERT_LSM_MAGIC(lsmp)                                                  \
1029 do {                                                                            \
1030         LASSERT((lsmp) != NULL);                                                \
1031         LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC ||                             \
1032                  (lsmp)->lsm_magic == LOV_MAGIC_JOIN), "%p->lsm_magic=%x\n",    \
1033                  (lsmp), (lsmp)->lsm_magic);                                    \
1034 } while (0)
1035
1036 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
1037                        struct lov_stripe_md *lsm, struct obd_trans_info *oti,
1038                        struct obd_export *md_exp)
1039 {
1040         struct lov_request_set *set;
1041         struct obd_info oinfo;
1042         struct lov_request *req;
1043         struct list_head *pos;
1044         struct lov_obd *lov;
1045         int rc = 0, err;
1046         ENTRY;
1047
1048         ASSERT_LSM_MAGIC(lsm);
1049
1050         if (!exp || !exp->exp_obd)
1051                 RETURN(-ENODEV);
1052
1053         if (oa->o_valid & OBD_MD_FLCOOKIE) {
1054                 LASSERT(oti);
1055                 LASSERT(oti->oti_logcookies);
1056         }
1057
1058         lov = &exp->exp_obd->u.lov;
1059         rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1060         if (rc)
1061                 RETURN(rc);
1062
1063         list_for_each (pos, &set->set_list) {
1064                 int err;
1065                 req = list_entry(pos, struct lov_request, rq_link);
1066
1067                 if (oa->o_valid & OBD_MD_FLCOOKIE)
1068                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1069
1070                 err = obd_destroy(lov->lov_tgts[req->rq_idx]->ltd_exp,
1071                                   req->rq_oi.oi_oa, NULL, oti, NULL);
1072                 err = lov_update_common_set(set, req, err);
1073                 if (err) {
1074                         CERROR("error: destroying objid "LPX64" subobj "
1075                                LPX64" on OST idx %d: rc = %d\n",
1076                                oa->o_id, req->rq_oi.oi_oa->o_id,
1077                                req->rq_idx, err);
1078                         if (!rc)
1079                                 rc = err;
1080                 }
1081         }
1082
1083         if (rc == 0) {
1084                 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1085                 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1086         }
1087         err = lov_fini_destroy_set(set);
1088         RETURN(rc ? rc : err);
1089 }
1090
1091 static int lov_getattr(struct obd_export *exp, struct obd_info *oinfo)
1092 {
1093         struct lov_request_set *set;
1094         struct lov_request *req;
1095         struct list_head *pos;
1096         struct lov_obd *lov;
1097         int err = 0, rc = 0;
1098         ENTRY;
1099
1100         LASSERT(oinfo);
1101         ASSERT_LSM_MAGIC(oinfo->oi_md);
1102
1103         if (!exp || !exp->exp_obd)
1104                 RETURN(-ENODEV);
1105
1106         lov = &exp->exp_obd->u.lov;
1107
1108         rc = lov_prep_getattr_set(exp, oinfo, &set);
1109         if (rc)
1110                 RETURN(rc);
1111
1112         list_for_each (pos, &set->set_list) {
1113                 req = list_entry(pos, struct lov_request, rq_link);
1114
1115                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1116                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1117                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1118
1119                 rc = obd_getattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1120                                  &req->rq_oi);
1121                 err = lov_update_common_set(set, req, rc);
1122                 if (err) {
1123                         CERROR("error: getattr objid "LPX64" subobj "
1124                                LPX64" on OST idx %d: rc = %d\n",
1125                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1126                                req->rq_idx, err);
1127                         break;
1128                 }
1129         }
1130
1131         rc = lov_fini_getattr_set(set);
1132         if (err)
1133                 rc = err;
1134         RETURN(rc);
1135 }
1136
1137 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1138                                  void *data, int rc)
1139 {
1140         struct lov_request_set *lovset = (struct lov_request_set *)data;
1141         int err;
1142         ENTRY;
1143
1144         /* don't do attribute merge if this aysnc op failed */
1145         if (rc)
1146                 lovset->set_completes = 0;
1147         err = lov_fini_getattr_set(lovset);
1148         RETURN(rc ? rc : err);
1149 }
1150
1151 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1152                               struct ptlrpc_request_set *rqset)
1153 {
1154         struct lov_request_set *lovset;
1155         struct lov_obd *lov;
1156         struct list_head *pos;
1157         struct lov_request *req;
1158         int rc = 0, err;
1159         ENTRY;
1160
1161         LASSERT(oinfo);
1162         ASSERT_LSM_MAGIC(oinfo->oi_md);
1163
1164         if (!exp || !exp->exp_obd)
1165                 RETURN(-ENODEV);
1166
1167         lov = &exp->exp_obd->u.lov;
1168
1169         rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1170         if (rc)
1171                 RETURN(rc);
1172
1173         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1174                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1175                oinfo->oi_md->lsm_stripe_size);
1176
1177         list_for_each (pos, &lovset->set_list) {
1178                 req = list_entry(pos, struct lov_request, rq_link);
1179
1180                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1181                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1182                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1183                 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1184                                        &req->rq_oi, rqset);
1185                 if (rc) {
1186                         CERROR("error: getattr objid "LPX64" subobj "
1187                                LPX64" on OST idx %d: rc = %d\n",
1188                                oinfo->oi_oa->o_id, req->rq_oi.oi_oa->o_id,
1189                                req->rq_idx, rc);
1190                         GOTO(out, rc);
1191                 }
1192         }
1193
1194         if (!list_empty(&rqset->set_requests)) {
1195                 LASSERT(rc == 0);
1196                 LASSERT (rqset->set_interpret == NULL);
1197                 rqset->set_interpret = lov_getattr_interpret;
1198                 rqset->set_arg = (void *)lovset;
1199                 RETURN(rc);
1200         }
1201 out:
1202         if (rc)
1203                 lovset->set_completes = 0;
1204         err = lov_fini_getattr_set(lovset);
1205         RETURN(rc ? rc : err);
1206 }
1207
1208 static int lov_setattr(struct obd_export *exp, struct obd_info *oinfo,
1209                        struct obd_trans_info *oti)
1210 {
1211         struct lov_request_set *set;
1212         struct lov_obd *lov;
1213         struct list_head *pos;
1214         struct lov_request *req;
1215         int err = 0, rc = 0;
1216         ENTRY;
1217
1218         LASSERT(oinfo);
1219         ASSERT_LSM_MAGIC(oinfo->oi_md);
1220
1221         if (!exp || !exp->exp_obd)
1222                 RETURN(-ENODEV);
1223
1224         /* for now, we only expect the following updates here */
1225         LASSERT(!(oinfo->oi_oa->o_valid & ~(OBD_MD_FLID | OBD_MD_FLTYPE |
1226                                             OBD_MD_FLMODE | OBD_MD_FLATIME |
1227                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1228                                             OBD_MD_FLFLAGS | OBD_MD_FLSIZE |
1229                                             OBD_MD_FLGROUP | OBD_MD_FLUID |
1230                                             OBD_MD_FLGID | OBD_MD_FLINLINE |
1231                                             OBD_MD_FLFID | OBD_MD_FLGENER)));
1232         lov = &exp->exp_obd->u.lov;
1233         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1234         if (rc)
1235                 RETURN(rc);
1236
1237         list_for_each (pos, &set->set_list) {
1238                 req = list_entry(pos, struct lov_request, rq_link);
1239
1240                 rc = obd_setattr(lov->lov_tgts[req->rq_idx]->ltd_exp,
1241                                  &req->rq_oi, NULL);
1242                 err = lov_update_setattr_set(set, req, rc);
1243                 if (err) {
1244                         CERROR("error: setattr objid "LPX64" subobj "
1245                                LPX64" on OST idx %d: rc = %d\n",
1246                                set->set_oi->oi_oa->o_id,
1247                                req->rq_oi.oi_oa->o_id, req->rq_idx, err);
1248                         if (!rc)
1249                                 rc = err;
1250                 }
1251         }
1252         err = lov_fini_setattr_set(set);
1253         if (!rc)
1254                 rc = err;
1255         RETURN(rc);
1256 }
1257
1258 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1259                                  void *data, int rc)
1260 {
1261         struct lov_request_set *lovset = (struct lov_request_set *)data;
1262         int err;
1263         ENTRY;
1264
1265         if (rc)
1266                 lovset->set_completes = 0;
1267         err = lov_fini_setattr_set(lovset);
1268         RETURN(rc ? rc : err);
1269 }
1270
1271 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1272    needed. Otherwise, a client is waiting for responses. */
1273 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1274                              struct obd_trans_info *oti,
1275                              struct ptlrpc_request_set *rqset)
1276 {
1277         struct lov_request_set *set;
1278         struct lov_request *req;
1279         struct list_head *pos;
1280         struct lov_obd *lov;
1281         int rc = 0;
1282         ENTRY;
1283
1284         LASSERT(oinfo);
1285         ASSERT_LSM_MAGIC(oinfo->oi_md);
1286         if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1287                 LASSERT(oti);
1288                 LASSERT(oti->oti_logcookies);
1289         }
1290
1291         if (!exp || !exp->exp_obd)
1292                 RETURN(-ENODEV);
1293
1294         lov = &exp->exp_obd->u.lov;
1295         rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1296         if (rc)
1297                 RETURN(rc);
1298
1299         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
1300                oinfo->oi_md->lsm_object_id, oinfo->oi_md->lsm_stripe_count,
1301                oinfo->oi_md->lsm_stripe_size);
1302
1303         list_for_each (pos, &set->set_list) {
1304                 req = list_entry(pos, struct lov_request, rq_link);
1305
1306                 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1307                         oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1308
1309                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
1310                        "%u\n", oinfo->oi_oa->o_id, req->rq_stripe,
1311                        req->rq_oi.oi_oa->o_id, req->rq_idx);
1312
1313                 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1314                                        &req->rq_oi, oti, rqset);
1315                 if (rc) {
1316                         CERROR("error: setattr objid "LPX64" subobj "
1317                                LPX64" on OST idx %d: rc = %d\n",
1318                                set->set_oi->oi_oa->o_id,
1319                                req->rq_oi.oi_oa->o_id,
1320                                req->rq_idx, rc);
1321                         break;
1322                 }
1323         }
1324
1325         /* If we are not waiting for responses on async requests, return. */
1326         if (rc || !rqset || list_empty(&rqset->set_requests)) {
1327                 int err;
1328                 if (rc)
1329                         set->set_completes = 0;
1330                 err = lov_fini_setattr_set(set);
1331                 RETURN(rc ? rc : err);
1332         }
1333
1334         LASSERT(rqset->set_interpret == NULL);
1335         rqset->set_interpret = lov_setattr_interpret;
1336         rqset->set_arg = (void *)set;
1337
1338         RETURN(0);
1339 }
1340
1341 static int lov_punch_interpret(struct ptlrpc_request_set *rqset,
1342                                void *data, int rc)
1343 {
1344         struct lov_request_set *lovset = (struct lov_request_set *)data;
1345         int err;
1346         ENTRY;
1347
1348         if (rc)
1349                 lovset->set_completes = 0;
1350         err = lov_fini_punch_set(lovset);
1351         RETURN(rc ? rc : err);
1352 }
1353
1354 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1355  * we can send this 'punch' to just the authoritative node and the nodes
1356  * that the punch will affect. */
1357 static int lov_punch(struct obd_export *exp, struct obd_info *oinfo,
1358                      struct obd_trans_info *oti,
1359                      struct ptlrpc_request_set *rqset)
1360 {
1361         struct lov_request_set *set;
1362         struct lov_obd *lov;
1363         struct list_head *pos;
1364         struct lov_request *req;
1365         int rc = 0;
1366         ENTRY;
1367
1368         LASSERT(oinfo);
1369         ASSERT_LSM_MAGIC(oinfo->oi_md);
1370
1371         if (!exp || !exp->exp_obd)
1372                 RETURN(-ENODEV);
1373
1374         lov = &exp->exp_obd->u.lov;
1375         rc = lov_prep_punch_set(exp, oinfo, oti, &set);
1376         if (rc)
1377                 RETURN(rc);
1378
1379         list_for_each (pos, &set->set_list) {
1380                 req = list_entry(pos, struct lov_request, rq_link);
1381
1382                 rc = obd_punch(lov->lov_tgts[req->rq_idx]->ltd_exp,
1383                                &req->rq_oi, NULL, rqset);
1384                 if (rc) {
1385                         CERROR("error: punch objid "LPX64" subobj "LPX64
1386                                " on OST idx %d: rc = %d\n",
1387                                set->set_oi->oi_oa->o_id,
1388                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1389                         break;
1390                 }
1391         }
1392
1393         if (rc || list_empty(&rqset->set_requests)) {
1394                 int err;
1395                 err = lov_fini_punch_set(set);
1396                 RETURN(rc ? rc : err);
1397         }
1398
1399         LASSERT(rqset->set_interpret == NULL);
1400         rqset->set_interpret = lov_punch_interpret;
1401         rqset->set_arg = (void *)set;
1402
1403         RETURN(0);
1404 }
1405
1406 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1407                     struct lov_stripe_md *lsm, obd_off start, obd_off end,
1408                     void *capa)
1409 {
1410         struct lov_request_set *set;
1411         struct obd_info oinfo;
1412         struct lov_obd *lov;
1413         struct list_head *pos;
1414         struct lov_request *req;
1415         int err = 0, rc = 0;
1416         ENTRY;
1417
1418         ASSERT_LSM_MAGIC(lsm);
1419
1420         if (!exp->exp_obd)
1421                 RETURN(-ENODEV);
1422
1423         lov = &exp->exp_obd->u.lov;
1424         rc = lov_prep_sync_set(exp, &oinfo, oa, lsm, start, end, &set);
1425         if (rc)
1426                 RETURN(rc);
1427
1428         list_for_each (pos, &set->set_list) {
1429                 req = list_entry(pos, struct lov_request, rq_link);
1430
1431                 rc = obd_sync(lov->lov_tgts[req->rq_idx]->ltd_exp,
1432                               req->rq_oi.oi_oa, NULL,
1433                               req->rq_oi.oi_policy.l_extent.start,
1434                               req->rq_oi.oi_policy.l_extent.end, capa);
1435                 err = lov_update_common_set(set, req, rc);
1436                 if (err) {
1437                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1438                                " on OST idx %d: rc = %d\n",
1439                                set->set_oi->oi_oa->o_id,
1440                                req->rq_oi.oi_oa->o_id, req->rq_idx, rc);
1441                         if (!rc)
1442                                 rc = err;
1443                 }
1444         }
1445         err = lov_fini_sync_set(set);
1446         if (!rc)
1447                 rc = err;
1448         RETURN(rc);
1449 }
1450
1451 static int lov_brw_check(struct lov_obd *lov, struct obd_info *lov_oinfo,
1452                          obd_count oa_bufs, struct brw_page *pga)
1453 {
1454         struct obd_info oinfo = { { { 0 } } };
1455         int i, rc = 0;
1456
1457         oinfo.oi_oa = lov_oinfo->oi_oa;
1458
1459         /* The caller just wants to know if there's a chance that this
1460          * I/O can succeed */
1461         for (i = 0; i < oa_bufs; i++) {
1462                 int stripe = lov_stripe_number(lov_oinfo->oi_md, pga[i].off);
1463                 int ost = lov_oinfo->oi_md->lsm_oinfo[stripe]->loi_ost_idx;
1464                 obd_off start, end;
1465
1466                 if (!lov_stripe_intersects(lov_oinfo->oi_md, i, pga[i].off,
1467                                            pga[i].off + pga[i].count,
1468                                            &start, &end))
1469                         continue;
1470
1471                 if (!lov->lov_tgts[ost] || !lov->lov_tgts[ost]->ltd_active) {
1472                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1473                         return -EIO;
1474                 }
1475
1476                 rc = obd_brw(OBD_BRW_CHECK, lov->lov_tgts[ost]->ltd_exp, &oinfo,
1477                              1, &pga[i], NULL);
1478                 if (rc)
1479                         break;
1480         }
1481         return rc;
1482 }
1483
1484 static int lov_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
1485                    obd_count oa_bufs, struct brw_page *pga,
1486                    struct obd_trans_info *oti)
1487 {
1488         struct lov_request_set *set;
1489         struct lov_request *req;
1490         struct list_head *pos;
1491         struct lov_obd *lov = &exp->exp_obd->u.lov;
1492         int err, rc = 0;
1493         ENTRY;
1494
1495         ASSERT_LSM_MAGIC(oinfo->oi_md);
1496
1497         if (cmd == OBD_BRW_CHECK) {
1498                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1499                 RETURN(rc);
1500         }
1501
1502         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &set);
1503         if (rc)
1504                 RETURN(rc);
1505
1506         list_for_each (pos, &set->set_list) {
1507                 struct obd_export *sub_exp;
1508                 struct brw_page *sub_pga;
1509                 req = list_entry(pos, struct lov_request, rq_link);
1510
1511                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1512                 sub_pga = set->set_pga + req->rq_pgaidx;
1513                 rc = obd_brw(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1514                              sub_pga, oti);
1515                 if (rc)
1516                         break;
1517                 lov_update_common_set(set, req, rc);
1518         }
1519
1520         err = lov_fini_brw_set(set);
1521         if (!rc)
1522                 rc = err;
1523         RETURN(rc);
1524 }
1525
1526 static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
1527                              int rc)
1528 {
1529         struct lov_request_set *lovset = (struct lov_request_set *)data;
1530         ENTRY;
1531
1532         if (rc) {
1533                 lovset->set_completes = 0;
1534                 lov_fini_brw_set(lovset);
1535         } else {
1536                 rc = lov_fini_brw_set(lovset);
1537         }
1538
1539         RETURN(rc);
1540 }
1541
1542 static int lov_brw_async(int cmd, struct obd_export *exp,
1543                          struct obd_info *oinfo, obd_count oa_bufs,
1544                          struct brw_page *pga, struct obd_trans_info *oti,
1545                          struct ptlrpc_request_set *set)
1546 {
1547         struct lov_request_set *lovset;
1548         struct lov_request *req;
1549         struct list_head *pos;
1550         struct lov_obd *lov = &exp->exp_obd->u.lov;
1551         int rc = 0;
1552         ENTRY;
1553
1554         LASSERT(oinfo);
1555         ASSERT_LSM_MAGIC(oinfo->oi_md);
1556
1557         if (cmd == OBD_BRW_CHECK) {
1558                 rc = lov_brw_check(lov, oinfo, oa_bufs, pga);
1559                 RETURN(rc);
1560         }
1561
1562         rc = lov_prep_brw_set(exp, oinfo, oa_bufs, pga, oti, &lovset);
1563         if (rc)
1564                 RETURN(rc);
1565
1566         list_for_each (pos, &lovset->set_list) {
1567                 struct obd_export *sub_exp;
1568                 struct brw_page *sub_pga;
1569                 req = list_entry(pos, struct lov_request, rq_link);
1570
1571                 sub_exp = lov->lov_tgts[req->rq_idx]->ltd_exp;
1572                 sub_pga = lovset->set_pga + req->rq_pgaidx;
1573                 rc = obd_brw_async(cmd, sub_exp, &req->rq_oi, req->rq_oabufs,
1574                                    sub_pga, oti, set);
1575                 if (rc)
1576                         GOTO(out, rc);
1577                 lov_update_common_set(lovset, req, rc);
1578         }
1579         LASSERT(rc == 0);
1580         LASSERT(set->set_interpret == NULL);
1581         set->set_interpret = (set_interpreter_func)lov_brw_interpret;
1582         set->set_arg = (void *)lovset;
1583
1584         RETURN(rc);
1585 out:
1586         lov_fini_brw_set(lovset);
1587         RETURN(rc);
1588 }
1589
1590 static int lov_ap_make_ready(void *data, int cmd)
1591 {
1592         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1593
1594         return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd);
1595 }
1596
1597 static int lov_ap_refresh_count(void *data, int cmd)
1598 {
1599         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1600
1601         return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data,
1602                                                      cmd);
1603 }
1604
1605 static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
1606 {
1607         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1608
1609         lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa);
1610         /* XXX woah, shouldn't we be altering more here?  size? */
1611         oa->o_id = lap->lap_loi_id;
1612         oa->o_gr = lap->lap_loi_gr;
1613         oa->o_valid |= OBD_MD_FLGROUP;
1614         oa->o_stripe_idx = lap->lap_stripe;
1615 }
1616
1617 static void lov_ap_update_obdo(void *data, int cmd, struct obdo *oa,
1618                                obd_valid valid)
1619 {
1620         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1621
1622         lap->lap_caller_ops->ap_update_obdo(lap->lap_caller_data, cmd,oa,valid);
1623 }
1624
1625 static int lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
1626 {
1627         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1628
1629         /* in a raid1 regime this would down a count of many ios
1630          * in flight, onl calling the caller_ops completion when all
1631          * the raid1 ios are complete */
1632         rc = lap->lap_caller_ops->ap_completion(lap->lap_caller_data,cmd,oa,rc);
1633         return rc;
1634 }
1635
1636 static struct obd_capa *lov_ap_lookup_capa(void *data, int cmd)
1637 {
1638         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1639         return lap->lap_caller_ops->ap_lookup_capa(lap->lap_caller_data, cmd);
1640 }
1641
1642 static struct obd_async_page_ops lov_async_page_ops = {
1643         .ap_make_ready =        lov_ap_make_ready,
1644         .ap_refresh_count =     lov_ap_refresh_count,
1645         .ap_fill_obdo =         lov_ap_fill_obdo,
1646         .ap_update_obdo =       lov_ap_update_obdo,
1647         .ap_completion =        lov_ap_completion,
1648         .ap_lookup_capa =       lov_ap_lookup_capa,
1649 };
1650
1651 int lov_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm,
1652                            struct lov_oinfo *loi, cfs_page_t *page,
1653                            obd_off offset, struct obd_async_page_ops *ops,
1654                            void *data, void **res)
1655 {
1656         struct lov_obd *lov = &exp->exp_obd->u.lov;
1657         struct lov_async_page *lap;
1658         int rc = 0;
1659         ENTRY;
1660
1661         if (!page) {
1662                 int i = 0;
1663                 /* Find an existing osc so we can get it's stupid sizeof(*oap).
1664                    Only because of this layering limitation will a client
1665                    mount with no osts fail */
1666                 while (!lov->lov_tgts || !lov->lov_tgts[i] ||
1667                        !lov->lov_tgts[i]->ltd_exp) {
1668                         i++;
1669                         if (i >= lov->desc.ld_tgt_count)
1670                                 RETURN(-ENOMEDIUM);
1671                 }
1672                 rc = size_round(sizeof(*lap)) +
1673                         obd_prep_async_page(lov->lov_tgts[i]->ltd_exp, NULL,
1674                                             NULL, NULL, 0, NULL, NULL, NULL);
1675                 RETURN(rc);
1676         }
1677         ASSERT_LSM_MAGIC(lsm);
1678         LASSERT(loi == NULL);
1679
1680         lap = *res;
1681         lap->lap_magic = LOV_AP_MAGIC;
1682         lap->lap_caller_ops = ops;
1683         lap->lap_caller_data = data;
1684
1685         /* for now only raid 0 which passes through */
1686         lap->lap_stripe = lov_stripe_number(lsm, offset);
1687         lov_stripe_offset(lsm, offset, lap->lap_stripe, &lap->lap_sub_offset);
1688         loi = lsm->lsm_oinfo[lap->lap_stripe];
1689
1690         /* so the callback doesn't need the lsm */
1691         lap->lap_loi_id = loi->loi_id;
1692         lap->lap_loi_gr = lsm->lsm_object_gr;
1693         LASSERT(lsm->lsm_object_gr > 0);
1694
1695         lap->lap_sub_cookie = (void *)lap + size_round(sizeof(*lap));
1696
1697         rc = obd_prep_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1698                                  lsm, loi, page, lap->lap_sub_offset,
1699                                  &lov_async_page_ops, lap,
1700                                  &lap->lap_sub_cookie);
1701         if (rc)
1702                 RETURN(rc);
1703         CDEBUG(D_CACHE, "lap %p page %p cookie %p off "LPU64"\n", lap, page,
1704                lap->lap_sub_cookie, offset);
1705         RETURN(0);
1706 }
1707
1708 static int lov_queue_async_io(struct obd_export *exp,
1709                               struct lov_stripe_md *lsm,
1710                               struct lov_oinfo *loi, void *cookie,
1711                               int cmd, obd_off off, int count,
1712                               obd_flag brw_flags, obd_flag async_flags)
1713 {
1714         struct lov_obd *lov = &exp->exp_obd->u.lov;
1715         struct lov_async_page *lap;
1716         int rc;
1717
1718         LASSERT(loi == NULL);
1719
1720         ASSERT_LSM_MAGIC(lsm);
1721
1722         lap = LAP_FROM_COOKIE(cookie);
1723
1724         loi = lsm->lsm_oinfo[lap->lap_stripe];
1725
1726         rc = obd_queue_async_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1727                                 loi, lap->lap_sub_cookie, cmd, off, count,
1728                                 brw_flags, async_flags);
1729         RETURN(rc);
1730 }
1731
1732 static int lov_set_async_flags(struct obd_export *exp,
1733                                struct lov_stripe_md *lsm,
1734                                struct lov_oinfo *loi, void *cookie,
1735                                obd_flag async_flags)
1736 {
1737         struct lov_obd *lov = &exp->exp_obd->u.lov;
1738         struct lov_async_page *lap;
1739         int rc;
1740
1741         LASSERT(loi == NULL);
1742
1743         ASSERT_LSM_MAGIC(lsm);
1744
1745         lap = LAP_FROM_COOKIE(cookie);
1746
1747         loi = lsm->lsm_oinfo[lap->lap_stripe];
1748
1749         rc = obd_set_async_flags(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1750                                  lsm, loi, lap->lap_sub_cookie, async_flags);
1751         RETURN(rc);
1752 }
1753
1754 static int lov_queue_group_io(struct obd_export *exp,
1755                               struct lov_stripe_md *lsm,
1756                               struct lov_oinfo *loi,
1757                               struct obd_io_group *oig, void *cookie,
1758                               int cmd, obd_off off, int count,
1759                               obd_flag brw_flags, obd_flag async_flags)
1760 {
1761         struct lov_obd *lov = &exp->exp_obd->u.lov;
1762         struct lov_async_page *lap;
1763         int rc;
1764
1765         LASSERT(loi == NULL);
1766
1767         ASSERT_LSM_MAGIC(lsm);
1768
1769         lap = LAP_FROM_COOKIE(cookie);
1770
1771         loi = lsm->lsm_oinfo[lap->lap_stripe];
1772
1773         rc = obd_queue_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp, lsm,
1774                                 loi, oig, lap->lap_sub_cookie, cmd, off, count,
1775                                 brw_flags, async_flags);
1776         RETURN(rc);
1777 }
1778
1779 /* this isn't exactly optimal.  we may have queued sync io in oscs on
1780  * all stripes, but we don't record that fact at queue time.  so we
1781  * trigger sync io on all stripes. */
1782 static int lov_trigger_group_io(struct obd_export *exp,
1783                                 struct lov_stripe_md *lsm,
1784                                 struct lov_oinfo *loi,
1785                                 struct obd_io_group *oig)
1786 {
1787         struct lov_obd *lov = &exp->exp_obd->u.lov;
1788         int rc = 0, i, err;
1789
1790         LASSERT(loi == NULL);
1791
1792         ASSERT_LSM_MAGIC(lsm);
1793
1794         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1795                 loi = lsm->lsm_oinfo[i];
1796                 if (!lov->lov_tgts[loi->loi_ost_idx] ||
1797                     !lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
1798                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1799                         continue;
1800                 }
1801
1802                 err = obd_trigger_group_io(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1803                                            lsm, loi, oig);
1804                 if (rc == 0 && err != 0)
1805                         rc = err;
1806         };
1807         RETURN(rc);
1808 }
1809
1810 static int lov_teardown_async_page(struct obd_export *exp,
1811                                    struct lov_stripe_md *lsm,
1812                                    struct lov_oinfo *loi, void *cookie)
1813 {
1814         struct lov_obd *lov = &exp->exp_obd->u.lov;
1815         struct lov_async_page *lap;
1816         int rc;
1817
1818         LASSERT(loi == NULL);
1819
1820         ASSERT_LSM_MAGIC(lsm);
1821
1822         lap = LAP_FROM_COOKIE(cookie);
1823
1824         loi = lsm->lsm_oinfo[lap->lap_stripe];
1825
1826         rc = obd_teardown_async_page(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1827                                      lsm, loi, lap->lap_sub_cookie);
1828         if (rc) {
1829                 CERROR("unable to teardown sub cookie %p: %d\n",
1830                        lap->lap_sub_cookie, rc);
1831                 RETURN(rc);
1832         }
1833         RETURN(rc);
1834 }
1835
1836 static int lov_enqueue_interpret(struct ptlrpc_request_set *rqset,
1837                                  void *data, int rc)
1838 {
1839         struct lov_request_set *lovset = (struct lov_request_set *)data;
1840         ENTRY;
1841         rc = lov_fini_enqueue_set(lovset, lovset->set_ei->ei_mode, rc, rqset);
1842         RETURN(rc);
1843 }
1844
1845 static int lov_enqueue(struct obd_export *exp, struct obd_info *oinfo,
1846                        struct ldlm_enqueue_info *einfo,
1847                        struct ptlrpc_request_set *rqset)
1848 {
1849         ldlm_mode_t mode = einfo->ei_mode;
1850         struct lov_request_set *set;
1851         struct lov_request *req;
1852         struct list_head *pos;
1853         struct lov_obd *lov;
1854         ldlm_error_t rc;
1855         ENTRY;
1856
1857         LASSERT(oinfo);
1858         ASSERT_LSM_MAGIC(oinfo->oi_md);
1859         LASSERT(mode == (mode & -mode));
1860
1861         /* we should never be asked to replay a lock this way. */
1862         LASSERT((oinfo->oi_flags & LDLM_FL_REPLAY) == 0);
1863
1864         if (!exp || !exp->exp_obd)
1865                 RETURN(-ENODEV);
1866
1867         lov = &exp->exp_obd->u.lov;
1868         rc = lov_prep_enqueue_set(exp, oinfo, einfo, &set);
1869         if (rc)
1870                 RETURN(rc);
1871
1872         list_for_each (pos, &set->set_list) {
1873                 req = list_entry(pos, struct lov_request, rq_link);
1874
1875                 rc = obd_enqueue(lov->lov_tgts[req->rq_idx]->ltd_exp,
1876                                  &req->rq_oi, einfo, rqset);
1877                 if (rc != ELDLM_OK)
1878                         GOTO(out, rc);
1879         }
1880
1881         if (rqset && !list_empty(&rqset->set_requests)) {
1882                 LASSERT(rc == 0);
1883                 LASSERT(rqset->set_interpret == NULL);
1884                 rqset->set_interpret = lov_enqueue_interpret;
1885                 rqset->set_arg = (void *)set;
1886                 RETURN(rc);
1887         }
1888 out:
1889         rc = lov_fini_enqueue_set(set, mode, rc, rqset);
1890         RETURN(rc);
1891 }
1892
1893 static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
1894                      __u32 type, ldlm_policy_data_t *policy, __u32 mode,
1895                      int *flags, void *data, struct lustre_handle *lockh)
1896 {
1897         struct lov_request_set *set;
1898         struct obd_info oinfo;
1899         struct lov_request *req;
1900         struct list_head *pos;
1901         struct lov_obd *lov = &exp->exp_obd->u.lov;
1902         struct lustre_handle *lov_lockhp;
1903         int lov_flags, rc = 0;
1904         ENTRY;
1905
1906         ASSERT_LSM_MAGIC(lsm);
1907         LASSERT((*flags & LDLM_FL_TEST_LOCK) || mode == (mode & -mode));
1908
1909         if (!exp || !exp->exp_obd)
1910                 RETURN(-ENODEV);
1911
1912         lov = &exp->exp_obd->u.lov;
1913         rc = lov_prep_match_set(exp, &oinfo, lsm, policy, mode, lockh, &set);
1914         if (rc)
1915                 RETURN(rc);
1916
1917         list_for_each (pos, &set->set_list) {
1918                 ldlm_policy_data_t sub_policy;
1919                 req = list_entry(pos, struct lov_request, rq_link);
1920                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1921                 LASSERT(lov_lockhp);
1922
1923                 lov_flags = *flags;
1924                 sub_policy.l_extent = req->rq_oi.oi_policy.l_extent;
1925
1926                 rc = obd_match(lov->lov_tgts[req->rq_idx]->ltd_exp,
1927                                req->rq_oi.oi_md, type, &sub_policy,
1928                                mode, &lov_flags, data, lov_lockhp);
1929                 rc = lov_update_match_set(set, req, rc);
1930                 if (rc <= 0)
1931                         break;
1932         }
1933         lov_fini_match_set(set, mode, *flags);
1934         RETURN(rc);
1935 }
1936
1937 static int lov_change_cbdata(struct obd_export *exp,
1938                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1939                              void *data)
1940 {
1941         struct lov_obd *lov;
1942         int rc = 0, i;
1943         ENTRY;
1944
1945         ASSERT_LSM_MAGIC(lsm);
1946
1947         if (!exp || !exp->exp_obd)
1948                 RETURN(-ENODEV);
1949
1950         LASSERT(lsm->lsm_object_gr > 0);
1951
1952         lov = &exp->exp_obd->u.lov;
1953         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1954                 struct lov_stripe_md submd;
1955                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1956
1957                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1958                         CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1959                         continue;
1960                 }
1961
1962                 submd.lsm_object_id = loi->loi_id;
1963                 submd.lsm_object_gr = lsm->lsm_object_gr;
1964                 submd.lsm_stripe_count = 0;
1965                 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1966                                        &submd, it, data);
1967         }
1968         RETURN(rc);
1969 }
1970
1971 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1972                       __u32 mode, struct lustre_handle *lockh)
1973 {
1974         struct lov_request_set *set;
1975         struct obd_info oinfo;
1976         struct lov_request *req;
1977         struct list_head *pos;
1978         struct lov_obd *lov = &exp->exp_obd->u.lov;
1979         struct lustre_handle *lov_lockhp;
1980         int err = 0, rc = 0;
1981         ENTRY;
1982
1983         ASSERT_LSM_MAGIC(lsm);
1984
1985         if (!exp || !exp->exp_obd)
1986                 RETURN(-ENODEV);
1987
1988         LASSERT(lsm->lsm_object_gr > 0);
1989         LASSERT(lockh);
1990         lov = &exp->exp_obd->u.lov;
1991         rc = lov_prep_cancel_set(exp, &oinfo, lsm, mode, lockh, &set);
1992         if (rc)
1993                 RETURN(rc);
1994
1995         list_for_each (pos, &set->set_list) {
1996                 req = list_entry(pos, struct lov_request, rq_link);
1997                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1998
1999                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
2000                                 req->rq_oi.oi_md, mode, lov_lockhp);
2001                 rc = lov_update_common_set(set, req, rc);
2002                 if (rc) {
2003                         CERROR("error: cancel objid "LPX64" subobj "
2004                                LPX64" on OST idx %d: rc = %d\n",
2005                                lsm->lsm_object_id,
2006                                req->rq_oi.oi_md->lsm_object_id,
2007                                req->rq_idx, rc);
2008                         err = rc;
2009                 }
2010
2011         }
2012         lov_fini_cancel_set(set);
2013         RETURN(err);
2014 }
2015
2016 static int lov_cancel_unused(struct obd_export *exp,
2017                              struct lov_stripe_md *lsm,
2018                              int flags, void *opaque)
2019 {
2020         struct lov_obd *lov;
2021         int rc = 0, i;
2022         ENTRY;
2023
2024         if (!exp || !exp->exp_obd)
2025                 RETURN(-ENODEV);
2026
2027         lov = &exp->exp_obd->u.lov;
2028         if (lsm == NULL) {
2029                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2030                         int err;
2031                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2032                                 continue;
2033
2034                         err = obd_cancel_unused(lov->lov_tgts[i]->ltd_exp, NULL,
2035                                                 flags, opaque);
2036                         if (!rc)
2037                                 rc = err;
2038                 }
2039                 RETURN(rc);
2040         }
2041
2042         ASSERT_LSM_MAGIC(lsm);
2043
2044         LASSERT(lsm->lsm_object_gr > 0);
2045         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2046                 struct lov_stripe_md submd;
2047                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2048                 int err;
2049
2050                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2051                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2052                         continue;
2053                 }
2054
2055                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2056                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2057
2058                 submd.lsm_object_id = loi->loi_id;
2059                 submd.lsm_object_gr = lsm->lsm_object_gr;
2060                 submd.lsm_stripe_count = 0;
2061                 err = obd_cancel_unused(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2062                                         &submd, flags, opaque);
2063                 if (err && lov->lov_tgts[loi->loi_ost_idx]->ltd_active) {
2064                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
2065                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
2066                                loi->loi_id, loi->loi_ost_idx, err);
2067                         if (!rc)
2068                                 rc = err;
2069                 }
2070         }
2071         RETURN(rc);
2072 }
2073
2074 static int lov_join_lru(struct obd_export *exp,
2075                         struct lov_stripe_md *lsm, int join)
2076 {
2077         struct lov_obd *lov;
2078         int i, count = 0;
2079         ENTRY;
2080
2081         ASSERT_LSM_MAGIC(lsm);
2082         if (!exp || !exp->exp_obd)
2083                 RETURN(-ENODEV);
2084
2085         lov = &exp->exp_obd->u.lov;
2086         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2087                 struct lov_stripe_md submd;
2088                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2089                 int rc = 0;
2090
2091                 if (!lov->lov_tgts[loi->loi_ost_idx]) {
2092                         CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
2093                         continue;
2094                 }
2095
2096                 if (!lov->lov_tgts[loi->loi_ost_idx]->ltd_active)
2097                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
2098
2099                 submd.lsm_object_id = loi->loi_id;
2100                 submd.lsm_object_gr = lsm->lsm_object_gr;
2101                 submd.lsm_stripe_count = 0;
2102                 rc = obd_join_lru(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
2103                                   &submd, join);
2104                 if (rc < 0) {
2105                         CERROR("join lru failed. objid: "LPX64" subobj: "LPX64
2106                                " ostidx: %d rc: %d\n", lsm->lsm_object_id,
2107                                loi->loi_id, loi->loi_ost_idx, rc);
2108                         return rc;
2109                 } else {
2110                         count += rc;
2111                 }
2112         }
2113         RETURN(count);
2114 }
2115
2116 static int lov_statfs_interpret(struct ptlrpc_request_set *rqset,
2117                                 void *data, int rc)
2118 {
2119         struct lov_request_set *lovset = (struct lov_request_set *)data;
2120         int err;
2121         ENTRY;
2122
2123         if (rc)
2124                 lovset->set_completes = 0;
2125
2126         err = lov_fini_statfs_set(lovset);
2127         RETURN(rc ? rc : err);
2128 }
2129
2130 static int lov_statfs_async(struct obd_device *obd, struct obd_info *oinfo,
2131                             __u64 max_age, struct ptlrpc_request_set *rqset)
2132 {
2133         struct lov_request_set *set;
2134         struct lov_request *req;
2135         struct list_head *pos;
2136         struct lov_obd *lov;
2137         int rc = 0;
2138         ENTRY;
2139
2140         LASSERT(oinfo != NULL);
2141         LASSERT(oinfo->oi_osfs != NULL);
2142
2143         lov = &obd->u.lov;
2144         rc = lov_prep_statfs_set(obd, oinfo, &set);
2145         if (rc)
2146                 RETURN(rc);
2147
2148         list_for_each (pos, &set->set_list) {
2149                 struct obd_device *osc_obd;
2150
2151                 req = list_entry(pos, struct lov_request, rq_link);
2152
2153                 osc_obd = class_exp2obd(lov->lov_tgts[req->rq_idx]->ltd_exp);
2154                 rc = obd_statfs_async(osc_obd, &req->rq_oi, max_age, rqset);
2155                 if (rc)
2156                         break;
2157         }
2158
2159         if (rc || list_empty(&rqset->set_requests)) {
2160                 int err;
2161                 if (rc)
2162                         set->set_completes = 0;
2163                 err = lov_fini_statfs_set(set);
2164                 RETURN(rc ? rc : err);
2165         }
2166
2167         LASSERT(rqset->set_interpret == NULL);
2168         rqset->set_interpret = lov_statfs_interpret;
2169         rqset->set_arg = (void *)set;
2170         RETURN(0);
2171 }
2172
2173 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
2174                       __u64 max_age)
2175 {
2176         struct lov_obd *lov = &obd->u.lov;
2177         struct obd_statfs lov_sfs;
2178         int set = 0;
2179         int rc = 0, err;
2180         int i;
2181         ENTRY;
2182
2183         /* We only get block data from the OBD */
2184         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2185                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active) {
2186                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
2187                         continue;
2188                 }
2189
2190                 err = obd_statfs(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
2191                                  &lov_sfs, max_age);
2192                 if (err) {
2193                         if (lov->lov_tgts[i]->ltd_active && !rc)
2194                                 rc = err;
2195                         continue;
2196                 }
2197                 lov_update_statfs(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
2198                                   osfs, &lov_sfs, set);
2199                 set++;
2200         }
2201
2202         err = lov_fini_statfs(obd, osfs, set);
2203         qos_update(lov);
2204
2205         RETURN(rc ? rc : err);
2206 }
2207
2208 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2209                          void *karg, void *uarg)
2210 {
2211         struct obd_device *obddev = class_exp2obd(exp);
2212         struct lov_obd *lov = &obddev->u.lov;
2213         int i, rc, count = lov->desc.ld_tgt_count;
2214         struct obd_uuid *uuidp;
2215         ENTRY;
2216
2217         switch (cmd) {
2218         case OBD_IOC_LOV_GET_CONFIG: {
2219                 struct obd_ioctl_data *data;
2220                 struct lov_desc *desc;
2221                 char *buf = NULL;
2222                 __u32 *genp;
2223
2224                 len = 0;
2225                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
2226                         RETURN(-EINVAL);
2227
2228                 data = (struct obd_ioctl_data *)buf;
2229
2230                 if (sizeof(*desc) > data->ioc_inllen1) {
2231                         obd_ioctl_freedata(buf, len);
2232                         RETURN(-EINVAL);
2233                 }
2234
2235                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
2236                         obd_ioctl_freedata(buf, len);
2237                         RETURN(-EINVAL);
2238                 }
2239
2240                 if (sizeof(__u32) * count > data->ioc_inllen3) {
2241                         obd_ioctl_freedata(buf, len);
2242                         RETURN(-EINVAL);
2243                 }
2244
2245                 desc = (struct lov_desc *)data->ioc_inlbuf1;
2246                 memcpy(desc, &(lov->desc), sizeof(*desc));
2247
2248                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
2249                 genp = (__u32 *)data->ioc_inlbuf3;
2250                 /* the uuid will be empty for deleted OSTs */
2251                 for (i = 0; i < count; i++, uuidp++, genp++) {
2252                         if (!lov->lov_tgts[i])
2253                                 continue;
2254                         *uuidp = lov->lov_tgts[i]->ltd_uuid;
2255                         *genp = lov->lov_tgts[i]->ltd_gen;
2256                 }
2257
2258                 rc = copy_to_user((void *)uarg, buf, len);
2259                 if (rc)
2260                         rc = -EFAULT;
2261                 obd_ioctl_freedata(buf, len);
2262                 break;
2263         }
2264         case LL_IOC_LOV_SETSTRIPE:
2265                 rc = lov_setstripe(exp, karg, uarg);
2266                 break;
2267         case LL_IOC_LOV_GETSTRIPE:
2268                 rc = lov_getstripe(exp, karg, uarg);
2269                 break;
2270         case LL_IOC_LOV_SETEA:
2271                 rc = lov_setea(exp, karg, uarg);
2272                 break;
2273         default: {
2274                 int set = 0;
2275
2276                 if (count == 0)
2277                         RETURN(-ENOTTY);
2278
2279                 rc = 0;
2280                 for (i = 0; i < count; i++) {
2281                         int err;
2282
2283                         /* OST was disconnected */
2284                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2285                                 continue;
2286
2287                         err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
2288                                             len, karg, uarg);
2289                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
2290                                 RETURN(err);
2291                         } else if (err) {
2292                                 if (lov->lov_tgts[i]->ltd_active) {
2293                                         CDEBUG(err == -ENOTTY ?
2294                                                D_IOCTL : D_WARNING,
2295                                                "iocontrol OSC %s on OST "
2296                                                "idx %d cmd %x: err = %d\n",
2297                                                lov_uuid2str(lov, i),
2298                                                i, cmd, err);
2299                                         if (!rc)
2300                                                 rc = err;
2301                                 }
2302                         } else {
2303                                 set = 1;
2304                         }
2305                 }
2306                 if (!set && !rc)
2307                         rc = -EIO;
2308         }
2309         }
2310
2311         RETURN(rc);
2312 }
2313
2314 static int lov_get_info(struct obd_export *exp, __u32 keylen,
2315                         void *key, __u32 *vallen, void *val)
2316 {
2317         struct obd_device *obddev = class_exp2obd(exp);
2318         struct lov_obd *lov = &obddev->u.lov;
2319         int i, rc;
2320         ENTRY;
2321
2322         if (!vallen || !val)
2323                 RETURN(-EFAULT);
2324
2325         lov_getref(obddev);
2326
2327         if (KEY_IS("lock_to_stripe")) {
2328                 struct {
2329                         char name[16];
2330                         struct ldlm_lock *lock;
2331                         struct lov_stripe_md *lsm;
2332                 } *data = key;
2333                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2334                 struct lov_oinfo *loi;
2335                 __u32 *stripe = val;
2336
2337                 if (*vallen < sizeof(*stripe))
2338                         GOTO(out, rc = -EFAULT);
2339                 *vallen = sizeof(*stripe);
2340
2341                 /* XXX This is another one of those bits that will need to
2342                  * change if we ever actually support nested LOVs.  It uses
2343                  * the lock's export to find out which stripe it is. */
2344                 /* XXX - it's assumed all the locks for deleted OSTs have
2345                  * been cancelled. Also, the export for deleted OSTs will
2346                  * be NULL and won't match the lock's export. */
2347                 for (i = 0; i < data->lsm->lsm_stripe_count; i++) {
2348                         loi = data->lsm->lsm_oinfo[i];
2349                         if (!lov->lov_tgts[loi->loi_ost_idx])
2350                                 continue;
2351                         if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2352                             data->lock->l_conn_export &&
2353                             loi->loi_id == res_id->name[0] &&
2354                             loi->loi_gr == res_id->name[2]) {
2355                                 *stripe = i;
2356                                 GOTO(out, rc = 0);
2357                         }
2358                 }
2359                 LDLM_ERROR(data->lock, "lock on inode without such object");
2360                 dump_lsm(D_ERROR, data->lsm);
2361                 GOTO(out, rc = -ENXIO);
2362         } else if (KEY_IS("last_id")) {
2363                 obd_id *ids = val;
2364                 int size = sizeof(obd_id);
2365                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2366                         if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
2367                                 continue;
2368                         rc = obd_get_info(lov->lov_tgts[i]->ltd_exp,
2369                                           keylen, key, &size, ids + i);
2370                         if (rc != 0)
2371                                 GOTO(out, rc);
2372                 }
2373                 GOTO(out, rc = 0);
2374         } else if (KEY_IS(KEY_LOVDESC)) {
2375                 struct lov_desc *desc_ret = val;
2376                 *desc_ret = lov->desc;
2377
2378                 GOTO(out, rc = 0);
2379         }
2380
2381         rc = -EINVAL;
2382 out:
2383         lov_putref(obddev);
2384         RETURN(rc);
2385 }
2386
2387 static int lov_set_info_async(struct obd_export *exp, obd_count keylen,
2388                               void *key, obd_count vallen, void *val,
2389                               struct ptlrpc_request_set *set)
2390 {
2391         struct obd_device *obddev = class_exp2obd(exp);
2392         struct lov_obd *lov = &obddev->u.lov;
2393         obd_count count;
2394         int i, rc = 0, err;
2395         int no_set = !set;
2396         int incr = 0, check_uuid = 0, do_inactive = 0;
2397         ENTRY;
2398
2399         if (no_set) {
2400                 set = ptlrpc_prep_set();
2401                 if (!set)
2402                         RETURN(-ENOMEM);
2403         }
2404
2405         lov_getref(obddev);
2406         count = lov->desc.ld_tgt_count;
2407
2408         if (KEY_IS(KEY_NEXT_ID)) {
2409                 /* We must use mds's idea of # osts for indexing into
2410                    mds->mds_lov_objids */
2411                 count = vallen / sizeof(obd_id);
2412                 LASSERT(count <= lov->desc.ld_tgt_count);
2413                 vallen = sizeof(obd_id);
2414                 incr = sizeof(obd_id);
2415                 do_inactive = 1;
2416         } else if (KEY_IS("checksum")) {
2417                 do_inactive = 1;
2418         } else if (KEY_IS("unlinked")) {
2419                 check_uuid = val ? 1 : 0;
2420         } else if (KEY_IS("evict_by_nid")) {
2421                 /* use defaults:  do_inactive = incr = 0; */
2422         }
2423
2424         for (i = 0; i < count; i++, val = (char *)val + incr) {
2425                 /* OST was disconnected */
2426                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
2427                         continue;
2428
2429                 /* OST is inactive and we don't want inactive OSCs */
2430                 if (!lov->lov_tgts[i]->ltd_active && !do_inactive)
2431                         continue;
2432
2433                 if (KEY_IS(KEY_MDS_CONN)) {
2434                         struct mds_group_info *mgi;
2435
2436                         LASSERT(vallen == sizeof(*mgi));
2437                         mgi = (struct mds_group_info *)val;
2438
2439                         /* Only want a specific OSC */
2440                         if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2441                                                 &lov->lov_tgts[i]->ltd_uuid))
2442                                 continue;
2443
2444                         err = obd_set_info_async(lov->lov_tgts[i]->ltd_exp,
2445                                          keylen, key, sizeof(int),
2446                                          &mgi->group, set);
2447                 } else {
2448                         /* Only want a specific OSC */
2449                         if (check_uuid &&
2450                             !obd_uuid_equals(val, &lov->lov_tgts[i]->ltd_uuid))
2451                                 continue;
2452
2453                         err = obd_set_info_async(lov->lov_tgts[i]->ltd_exp,
2454                                          keylen, key, vallen, val, set);
2455                 }
2456
2457                 if (!rc)
2458                         rc = err;
2459         }
2460
2461         lov_putref(obddev);
2462         if (no_set) {
2463                 err = ptlrpc_set_wait(set);
2464                 if (!rc)
2465                         rc = err;
2466                 ptlrpc_set_destroy(set);
2467         }
2468         RETURN(rc);
2469 }
2470
2471 static int lov_checkmd(struct obd_export *exp, struct obd_export *md_exp,
2472                        struct lov_stripe_md *lsm)
2473 {
2474         int rc;
2475         ENTRY;
2476
2477         if (!lsm)
2478                 RETURN(0);
2479         LASSERT(md_exp);
2480         LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
2481         rc = lsm_op_find(lsm->lsm_magic)->lsm_revalidate(lsm, md_exp->exp_obd);
2482
2483         RETURN(rc);
2484 }
2485
2486 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm)
2487 {
2488         int i, rc = 0;
2489         ENTRY;
2490
2491         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2492                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
2493                 if (loi->loi_ar.ar_rc && !rc)
2494                         rc = loi->loi_ar.ar_rc;
2495                 loi->loi_ar.ar_rc = 0;
2496         }
2497         RETURN(rc);
2498 }
2499 EXPORT_SYMBOL(lov_test_and_clear_async_rc);
2500
2501
2502 static int lov_extent_calc(struct obd_export *exp, struct lov_stripe_md *lsm,
2503                            int cmd, __u64 *offset)
2504 {
2505         __u32 ssize = lsm->lsm_stripe_size;
2506         __u64 start;
2507
2508         start = *offset;
2509         do_div(start, ssize);
2510         start = start * ssize;
2511
2512         CDEBUG(D_DLMTRACE, "offset "LPU64", stripe %u, start "LPU64
2513                            ", end "LPU64"\n", *offset, ssize, start,
2514                            start + ssize - 1);
2515         if (cmd == OBD_CALC_STRIPE_END) {
2516                 *offset = start + ssize - 1;
2517         } else if (cmd == OBD_CALC_STRIPE_START) {
2518                 *offset = start;
2519         } else {
2520                 LBUG();
2521         }
2522
2523         RETURN(0);
2524 }
2525
2526
2527 #if 0
2528 struct lov_multi_wait {
2529         struct ldlm_lock *lock;
2530         wait_queue_t      wait;
2531         int               completed;
2532         int               generation;
2533 };
2534
2535 int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
2536                       struct lustre_handle *lockh)
2537 {
2538         struct lov_lock_handles *lov_lockh = NULL;
2539         struct lustre_handle *lov_lockhp;
2540         struct lov_obd *lov;
2541         struct lov_oinfo *loi;
2542         struct lov_multi_wait *queues;
2543         int rc = 0, i;
2544         ENTRY;
2545
2546         ASSERT_LSM_MAGIC(lsm);
2547
2548         if (!exp || !exp->exp_obd)
2549                 RETURN(-ENODEV);
2550
2551         LASSERT(lockh != NULL);
2552         if (lsm->lsm_stripe_count > 1) {
2553                 lov_lockh = lov_handle2llh(lockh);
2554                 if (lov_lockh == NULL) {
2555                         CERROR("LOV: invalid lov lock handle %p\n", lockh);
2556                         RETURN(-EINVAL);
2557                 }
2558
2559                 lov_lockhp = lov_lockh->llh_handles;
2560         } else {
2561                 lov_lockhp = lockh;
2562         }
2563
2564         OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues));
2565         if (queues == NULL)
2566                 GOTO(out, rc = -ENOMEM);
2567
2568         lov = &exp->exp_obd->u.lov;
2569         for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
2570              i++, loi++, lov_lockhp++) {
2571                 struct ldlm_lock *lock;
2572                 struct obd_device *obd;
2573
2574                 lock = ldlm_handle2lock(lov_lockhp);
2575                 if (lock == NULL) {
2576                         CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
2577                                loi->loi_ost_idx, loi->loi_id);
2578                         queues[i].completed = 1;
2579                         continue;
2580                 }
2581
2582                 queues[i].lock = lock;
2583                 init_waitqueue_entry(&(queues[i].wait), current);
2584                 add_wait_queue(lock->l_waitq, &(queues[i].wait));
2585
2586                 obd = class_exp2obd(lock->l_conn_export);
2587                 if (obd != NULL)
2588                         imp = obd->u.cli.cl_import;
2589                 if (imp != NULL) {
2590                         spin_lock(&imp->imp_lock);
2591                         queues[i].generation = imp->imp_generation;
2592                         spin_unlock(&imp->imp_lock);
2593                 }
2594         }
2595
2596         lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait,
2597                                interrupted_completion_wait, &lwd);
2598         rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi);
2599
2600         for (i = 0; i < lsm->lsm_stripe_count; i++)
2601                 remove_wait_queue(lock->l_waitq, &(queues[i].wait));
2602
2603         if (rc == -EINTR || rc == -ETIMEDOUT) {
2604
2605
2606         }
2607
2608  out:
2609         if (lov_lockh != NULL)
2610                 lov_llh_put(lov_lockh);
2611         RETURN(rc);
2612 }
2613 #endif
2614
2615 void lov_stripe_lock(struct lov_stripe_md *md)
2616 {
2617         LASSERT(md->lsm_lock_owner != cfs_current());
2618         spin_lock(&md->lsm_lock);
2619         LASSERT(md->lsm_lock_owner == NULL);
2620         md->lsm_lock_owner = cfs_current();
2621 }
2622 EXPORT_SYMBOL(lov_stripe_lock);
2623
2624 void lov_stripe_unlock(struct lov_stripe_md *md)
2625 {
2626         LASSERT(md->lsm_lock_owner == cfs_current());
2627         md->lsm_lock_owner = NULL;
2628         spin_unlock(&md->lsm_lock);
2629 }
2630 EXPORT_SYMBOL(lov_stripe_unlock);
2631
2632 struct obd_ops lov_obd_ops = {
2633         .o_owner               = THIS_MODULE,
2634         .o_setup               = lov_setup,
2635         .o_precleanup          = lov_precleanup,
2636         .o_cleanup             = lov_cleanup,
2637         .o_process_config      = lov_process_config,
2638         .o_connect             = lov_connect,
2639         .o_disconnect          = lov_disconnect,
2640         .o_statfs              = lov_statfs,
2641         .o_statfs_async        = lov_statfs_async,
2642         .o_packmd              = lov_packmd,
2643         .o_unpackmd            = lov_unpackmd,
2644         .o_checkmd             = lov_checkmd,
2645         .o_create              = lov_create,
2646         .o_destroy             = lov_destroy,
2647         .o_getattr             = lov_getattr,
2648         .o_getattr_async       = lov_getattr_async,
2649         .o_setattr             = lov_setattr,
2650         .o_setattr_async       = lov_setattr_async,
2651         .o_brw                 = lov_brw,
2652         .o_brw_async           = lov_brw_async,
2653         .o_prep_async_page     = lov_prep_async_page,
2654         .o_queue_async_io      = lov_queue_async_io,
2655         .o_set_async_flags     = lov_set_async_flags,
2656         .o_queue_group_io      = lov_queue_group_io,
2657         .o_trigger_group_io    = lov_trigger_group_io,
2658         .o_teardown_async_page = lov_teardown_async_page,
2659         .o_merge_lvb           = lov_merge_lvb,
2660         .o_adjust_kms          = lov_adjust_kms,
2661         .o_punch               = lov_punch,
2662         .o_sync                = lov_sync,
2663         .o_enqueue             = lov_enqueue,
2664         .o_match               = lov_match,
2665         .o_change_cbdata       = lov_change_cbdata,
2666         .o_cancel              = lov_cancel,
2667         .o_cancel_unused       = lov_cancel_unused,
2668         .o_join_lru            = lov_join_lru,
2669         .o_iocontrol           = lov_iocontrol,
2670         .o_get_info            = lov_get_info,
2671         .o_set_info_async      = lov_set_info_async,
2672         .o_extent_calc         = lov_extent_calc,
2673         .o_llog_init           = lov_llog_init,
2674         .o_llog_finish         = lov_llog_finish,
2675         .o_notify              = lov_notify,
2676 };
2677
2678 static quota_interface_t *quota_interface;
2679 extern quota_interface_t lov_quota_interface;
2680
2681 cfs_mem_cache_t *lov_oinfo_slab;
2682
2683 int __init lov_init(void)
2684 {
2685         struct lprocfs_static_vars lvars;
2686         int rc, rc2;
2687         ENTRY;
2688
2689         lov_oinfo_slab = cfs_mem_cache_create("lov_oinfo",
2690                                               sizeof(struct lov_oinfo),
2691                                               0, SLAB_HWCACHE_ALIGN);
2692         if (lov_oinfo_slab == NULL)
2693                 return -ENOMEM;
2694         lprocfs_init_vars(lov, &lvars);
2695
2696         request_module("lquota");
2697         quota_interface = PORTAL_SYMBOL_GET(lov_quota_interface);
2698         init_obd_quota_ops(quota_interface, &lov_obd_ops);
2699
2700         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2701                                  LUSTRE_LOV_NAME, NULL);
2702         if (rc) {
2703                 if (quota_interface)
2704                         PORTAL_SYMBOL_PUT(lov_quota_interface);
2705                 rc2 = cfs_mem_cache_destroy(lov_oinfo_slab);
2706                 LASSERT(rc2 == 0);
2707         }
2708
2709         RETURN(rc);
2710 }
2711
2712 #ifdef __KERNEL__
2713 static void /*__exit*/ lov_exit(void)
2714 {
2715         int rc;
2716
2717         if (quota_interface)
2718                 PORTAL_SYMBOL_PUT(lov_quota_interface);
2719
2720         class_unregister_type(LUSTRE_LOV_NAME);
2721         rc = cfs_mem_cache_destroy(lov_oinfo_slab);
2722         LASSERT(rc == 0);
2723 }
2724
2725 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2726 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2727 MODULE_LICENSE("GPL");
2728
2729 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);
2730 #endif