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, 2003 Cluster File Systems, Inc.
   5  * Author: Phil Schwan <phil@clusterfs.com>
   6  *         Peter Braam <braam@clusterfs.com>
   7  *         Mike Shaver <shaver@clusterfs.com>
   8  *
   9  *   This file is part of Lustre, http://www.lustre.org.
  10  *
  11  *   Lustre is free software; you can redistribute it and/or
  12  *   modify it under the terms of version 2 of the GNU General Public
  13  *   License as published by the Free Software Foundation.
  14  *
  15  *   Lustre is distributed in the hope that it will be useful,
  16  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  *   GNU General Public License for more details.
  19  *
  20  *   You should have received a copy of the GNU General Public License
  21  *   along with Lustre; if not, write to the Free Software
  22  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  23  */
  24
  25 #ifndef EXPORT_SYMTAB
  26 # define EXPORT_SYMTAB
  27 #endif
  28 #define DEBUG_SUBSYSTEM S_LOV
  29 #ifdef __KERNEL__
  30 #include <linux/slab.h>
  31 #include <linux/module.h>
  32 #include <linux/init.h>
  33 #include <linux/slab.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/seq_file.h>
  36 #include <asm/div64.h>
  37 #else
  38 #include <liblustre.h>
  39 #endif
  40
  41 #include <linux/obd_support.h>
  42 #include <linux/lustre_lib.h>
  43 #include <linux/lustre_net.h>
  44 #include <linux/lustre_idl.h>
  45 #include <linux/lustre_dlm.h>
  46 #include <linux/lustre_mds.h>
  47 #include <linux/obd_class.h>
  48 #include <linux/obd_lov.h>
  49 #include <linux/obd_ost.h>
  50 #include <linux/lprocfs_status.h>
  51
  52 #include "lov_internal.h"
  53
  54 /* obd methods */
  55 #define MAX_STRING_SIZE 128
  56 static int lov_connect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt,
  57                            int activate, unsigned long connect_flags)
  58 {
  59         struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
  60         struct obd_uuid *tgt_uuid = &tgt->uuid;
  61
  62 #ifdef __KERNEL__
  63         struct proc_dir_entry *lov_proc_dir;
  64 #endif
  65         struct lov_obd *lov = &obd->u.lov;
  66         struct lustre_handle conn = {0, };
  67         struct obd_device *tgt_obd;
  68         int rc;
  69         ENTRY;
  70
  71         tgt_obd = class_find_client_obd(tgt_uuid, LUSTRE_OSC_NAME,
  72                                         &obd->obd_uuid);
  73
  74         if (!tgt_obd) {
  75                 CERROR("Target %s not attached\n", tgt_uuid->uuid);
  76                 RETURN(-EINVAL);
  77         }
  78
  79         if (!tgt_obd->obd_set_up) {
  80                 CERROR("Target %s not set up\n", tgt_uuid->uuid);
  81                 RETURN(-EINVAL);
  82         }
  83
  84         if (activate) {
  85                 tgt_obd->obd_no_recov = 0;
  86                 ptlrpc_activate_import(tgt_obd->u.cli.cl_import);
  87         }
  88
  89         if (tgt_obd->u.cli.cl_import->imp_invalid) {
  90                 CERROR("not connecting OSC %s; administratively "
  91                        "disabled\n", tgt_uuid->uuid);
  92                 rc = obd_register_observer(tgt_obd, obd);
  93                 if (rc) {
  94                         CERROR("Target %s register_observer error %d; "
  95                                "will not be able to reactivate\n",
  96                                tgt_uuid->uuid, rc);
  97                 }
  98                 RETURN(0);
  99         }
 100
 101         rc = obd_connect(&conn, tgt_obd, &lov_osc_uuid, connect_flags);
 102         if (rc) {
 103                 CERROR("Target %s connect error %d\n", tgt_uuid->uuid, rc);
 104                 RETURN(rc);
 105         }
 106         tgt->ltd_exp = class_conn2export(&conn);
 107
 108         rc = obd_register_observer(tgt_obd, obd);
 109         if (rc) {
 110                 CERROR("Target %s register_observer error %d\n",
 111                        tgt_uuid->uuid, rc);
 112                 obd_disconnect(tgt->ltd_exp, 0);
 113                 tgt->ltd_exp = NULL;
 114                 RETURN(rc);
 115         }
 116
 117         tgt->active = 1;
 118         lov->desc.ld_active_tgt_count++;
 119
 120 #ifdef __KERNEL__
 121         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 122         if (lov_proc_dir) {
 123                 struct obd_device *osc_obd = class_conn2obd(&conn);
 124                 struct proc_dir_entry *osc_symlink;
 125                 char name[MAX_STRING_SIZE + 1];
 126
 127                 LASSERT(osc_obd != NULL);
 128                 LASSERT(osc_obd->obd_type != NULL);
 129                 LASSERT(osc_obd->obd_type->typ_name != NULL);
 130                 name[MAX_STRING_SIZE] = '\0';
 131                 snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
 132                          osc_obd->obd_type->typ_name,
 133                          osc_obd->obd_name);
 134                 osc_symlink = proc_symlink(osc_obd->obd_name, lov_proc_dir,
 135                                            name);
 136                 if (osc_symlink == NULL) {
 137                         CERROR("could not register LOV target "
 138                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
 139                                obd->obd_type->typ_name, obd->obd_name,
 140                                osc_obd->obd_name);
 141                         lprocfs_remove(lov_proc_dir);
 142                         lov_proc_dir = NULL;
 143                 }
 144         }
 145 #endif
 146
 147         RETURN(0);
 148 }
 149
 150 static int lov_connect(struct lustre_handle *conn, struct obd_device *obd,
 151                        struct obd_uuid *cluuid, unsigned long flags)
 152 {
 153 #ifdef __KERNEL__
 154         struct proc_dir_entry *lov_proc_dir;
 155 #endif
 156         struct lov_obd *lov = &obd->u.lov;
 157         struct lov_tgt_desc *tgt;
 158         struct obd_export *exp;
 159         int rc, rc2, i;
 160         ENTRY;
 161
 162         rc = class_connect(conn, obd, cluuid);
 163         if (rc)
 164                 RETURN(rc);
 165
 166         exp = class_conn2export(conn);
 167
 168         /* We don't want to actually do the underlying connections more than
 169          * once, so keep track. */
 170         lov->refcount++;
 171         if (lov->refcount > 1) {
 172                 class_export_put(exp);
 173                 RETURN(0);
 174         }
 175
 176 #ifdef __KERNEL__
 177         lov_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
 178                                         NULL, NULL);
 179         if (IS_ERR(lov_proc_dir)) {
 180                 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 181                        obd->obd_type->typ_name, obd->obd_name);
 182                 lov_proc_dir = NULL;
 183         }
 184 #endif
 185
 186         /* connect_flags is the MDS number, save for use in lov_add_obd */
 187         lov->lov_connect_flags = flags;
 188         for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count; i++, tgt++) {
 189                 if (obd_uuid_empty(&tgt->uuid))
 190                         continue;
 191                 rc = lov_connect_obd(obd, tgt, 0, flags);
 192                 if (rc)
 193                         GOTO(out_disc, rc);
 194         }
 195
 196         class_export_put(exp);
 197         RETURN (0);
 198
 199  out_disc:
 200 #ifdef __KERNEL__
 201         if (lov_proc_dir)
 202                 lprocfs_remove(lov_proc_dir);
 203 #endif
 204
 205         while (i-- > 0) {
 206                 struct obd_uuid uuid;
 207                 --tgt;
 208                 --lov->desc.ld_active_tgt_count;
 209                 tgt->active = 0;
 210                 /* save for CERROR below; (we know it's terminated) */
 211                 uuid = tgt->uuid;
 212                 rc2 = obd_disconnect(tgt->ltd_exp, 0);
 213                 if (rc2)
 214                         CERROR("error: LOV target %s disconnect on OST idx %d: "
 215                                "rc = %d\n", uuid.uuid, i, rc2);
 216         }
 217         class_disconnect(exp, 0);
 218         RETURN (rc);
 219 }
 220
 221 static int lov_disconnect_obd(struct obd_device *obd,
 222                               struct lov_tgt_desc *tgt,
 223                               unsigned long flags)
 224 {
 225 #ifdef __KERNEL__
 226         struct proc_dir_entry *lov_proc_dir;
 227 #endif
 228         struct obd_device *osc_obd = class_exp2obd(tgt->ltd_exp);
 229         struct lov_obd *lov = &obd->u.lov;
 230         int rc;
 231         ENTRY;
 232
 233 #ifdef __KERNEL__
 234         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 235         if (lov_proc_dir) {
 236                 struct proc_dir_entry *osc_symlink;
 237
 238                 osc_symlink = lprocfs_srch(lov_proc_dir, osc_obd->obd_name);
 239                 if (osc_symlink) {
 240                         lprocfs_remove(osc_symlink);
 241                 } else {
 242                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 243                                obd->obd_type->typ_name, obd->obd_name,
 244                                osc_obd->obd_name);
 245                 }
 246         }
 247 #endif
 248         if (obd->obd_no_recov) {
 249                 /* Pass it on to our clients.
 250                  * XXX This should be an argument to disconnect,
 251                  * XXX not a back-door flag on the OBD.  Ah well.
 252                  */
 253                 if (osc_obd)
 254                         osc_obd->obd_no_recov = 1;
 255         }
 256
 257         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
 258         rc = obd_disconnect(tgt->ltd_exp, flags);
 259         if (rc) {
 260                 if (tgt->active) {
 261                         CERROR("Target %s disconnect error %d\n",
 262                                tgt->uuid.uuid, rc);
 263                 }
 264                 rc = 0;
 265         }
 266
 267         if (tgt->active) {
 268                 tgt->active = 0;
 269                 lov->desc.ld_active_tgt_count--;
 270         }
 271         tgt->ltd_exp = NULL;
 272         RETURN(0);
 273 }
 274
 275 static int lov_disconnect(struct obd_export *exp, unsigned long flags)
 276 {
 277         struct obd_device *obd = class_exp2obd(exp);
 278 #ifdef __KERNEL__
 279         struct proc_dir_entry *lov_proc_dir;
 280 #endif
 281         struct lov_obd *lov = &obd->u.lov;
 282         struct lov_tgt_desc *tgt;
 283         int rc, i;
 284         ENTRY;
 285
 286         if (!lov->tgts)
 287                 goto out_local;
 288
 289         /* Only disconnect the underlying layers on the final disconnect. */
 290         lov->refcount--;
 291         if (lov->refcount != 0)
 292                 goto out_local;
 293
 294         for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count; i++, tgt++) {
 295                 if (tgt->ltd_exp)
 296                         lov_disconnect_obd(obd, tgt, flags);
 297         }
 298
 299 #ifdef __KERNEL__
 300         lov_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 301         if (lov_proc_dir) {
 302                 lprocfs_remove(lov_proc_dir);
 303         } else {
 304                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing.",
 305                        obd->obd_type->typ_name, obd->obd_name);
 306         }
 307 #endif
 308
 309  out_local:
 310         rc = class_disconnect(exp, 0);
 311         RETURN(rc);
 312 }
 313
 314 /* Error codes:
 315  *
 316  *  -EINVAL  : UUID can't be found in the LOV's target list
 317  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
 318  *  -EBADF   : The UUID is found, but the OBD is the wrong type (!)
 319  */
 320 static int lov_set_osc_active(struct lov_obd *lov, struct obd_uuid *uuid,
 321                               int activate)
 322 {
 323         struct lov_tgt_desc *tgt;
 324         int i, rc = 0;
 325         ENTRY;
 326
 327         CDEBUG(D_INFO, "Searching in lov %p for uuid %s (activate=%d)\n",
 328                lov, uuid->uuid, activate);
 329
 330         spin_lock(&lov->lov_lock);
 331         for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count; i++, tgt++) {
 332                 if (tgt->ltd_exp == NULL)
 333                         continue;
 334
 335                 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
 336                        i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
 337                 if (strncmp(uuid->uuid, tgt->uuid.uuid, sizeof uuid->uuid) == 0)
 338                         break;
 339         }
 340
 341         if (i == lov->desc.ld_tgt_count)
 342                 GOTO(out, rc = -EINVAL);
 343
 344
 345         if (tgt->active == activate) {
 346                 CDEBUG(D_INFO, "OSC %s already %sactive!\n", uuid->uuid,
 347                         activate ? "" : "in");
 348                 GOTO(out, rc);
 349         }
 350
 351         CDEBUG(D_INFO, "Marking OSC %s %sactive\n", uuid->uuid,
 352                activate ? "" : "in");
 353
 354         tgt->active = activate;
 355         if (activate)
 356                 lov->desc.ld_active_tgt_count++;
 357         else
 358                 lov->desc.ld_active_tgt_count--;
 359
 360         EXIT;
 361  out:
 362         spin_unlock(&lov->lov_lock);
 363         return rc;
 364 }
 365
 366 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
 367                       int active, void *data)
 368 {
 369         struct obd_uuid *uuid;
 370         int rc;
 371         ENTRY;
 372
 373         if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
 374                 CERROR("unexpected notification of %s %s!\n",
 375                        watched->obd_type->typ_name,
 376                        watched->obd_name);
 377                 return -EINVAL;
 378         }
 379         uuid = &watched->u.cli.cl_import->imp_target_uuid;
 380
 381         /* Set OSC as active before notifying the observer, so the
 382          * observer can use the OSC normally.
 383          */
 384         rc = lov_set_osc_active(&obd->u.lov, uuid, active);
 385         if (rc) {
 386                 CERROR("%sactivation of %s failed: %d\n",
 387                        active ? "" : "de", uuid->uuid, rc);
 388                 RETURN(rc);
 389         }
 390
 391         if (obd->obd_observer)
 392                 /* Pass the notification up the chain. */
 393                 rc = obd_notify(obd->obd_observer, watched, active, data);
 394
 395         RETURN(rc);
 396 }
 397
 398 int lov_attach(struct obd_device *dev, obd_count len, void *data)
 399 {
 400         struct lprocfs_static_vars lvars;
 401         int rc;
 402
 403         lprocfs_init_vars(lov, &lvars);
 404         rc = lprocfs_obd_attach(dev, lvars.obd_vars);
 405         if (rc == 0) {
 406 #ifdef __KERNEL__
 407                 struct proc_dir_entry *entry;
 408
 409                 entry = create_proc_entry("target_obd_status", 0444,
 410                                           dev->obd_proc_entry);
 411                 if (entry == NULL) {
 412                         rc = -ENOMEM;
 413                 } else {
 414                         entry->proc_fops = &lov_proc_target_fops;
 415                         entry->data = dev;
 416                 }
 417 #endif
 418         }
 419         return rc;
 420 }
 421
 422 int lov_detach(struct obd_device *dev)
 423 {
 424         return lprocfs_obd_detach(dev);
 425 }
 426
 427 static int lov_setup(struct obd_device *obd, obd_count len, void *buf)
 428 {
 429         struct lov_obd *lov = &obd->u.lov;
 430         struct lustre_cfg *lcfg = buf;
 431         struct lov_desc *desc;
 432         int count;
 433         ENTRY;
 434
 435         if (lcfg->lcfg_inllen1 < 1) {
 436                 CERROR("LOV setup requires a descriptor\n");
 437                 RETURN(-EINVAL);
 438         }
 439
 440         desc = (struct lov_desc *)lcfg->lcfg_inlbuf1;
 441         if (sizeof(*desc) > lcfg->lcfg_inllen1) {
 442                 CERROR("descriptor size wrong: %d > %d\n",
 443                        (int)sizeof(*desc), lcfg->lcfg_inllen1);
 444                 RETURN(-EINVAL);
 445         }
 446
 447         /* Because of 64-bit divide/mod operations only work with a 32-bit
 448          * divisor in a 32-bit kernel, we cannot support a stripe width
 449          * of 4GB or larger on 32-bit CPUs.
 450          */
 451
 452         count = desc->ld_default_stripe_count;
 453         if (count && (count * desc->ld_default_stripe_size) > ~0UL) {
 454                 CERROR("LOV: stripe width "LPU64"x%u > %lu on 32-bit system\n",
 455                        desc->ld_default_stripe_size, count, ~0UL);
 456                 RETURN(-EINVAL);
 457         }
 458         if (desc->ld_tgt_count > 0) {
 459                 lov->bufsize= sizeof(struct lov_tgt_desc) * desc->ld_tgt_count;
 460         } else {
 461                 lov->bufsize = sizeof(struct lov_tgt_desc) * LOV_MAX_TGT_COUNT;
 462         }
 463         OBD_ALLOC(lov->tgts, lov->bufsize);
 464         if (lov->tgts == NULL) {
 465                 lov->bufsize = 0;
 466                 CERROR("couldn't allocate %d bytes for target table.\n",
 467                        lov->bufsize);
 468                 RETURN(-EINVAL);
 469         }
 470
 471         desc->ld_tgt_count = 0;
 472         desc->ld_active_tgt_count = 0;
 473         lov->desc = *desc;
 474         spin_lock_init(&lov->lov_lock);
 475         sema_init(&lov->lov_llog_sem, 1);
 476
 477         RETURN(0);
 478 }
 479
 480 static int lov_cleanup(struct obd_device *obd, int flags)
 481 {
 482         struct lov_obd *lov = &obd->u.lov;
 483
 484         OBD_FREE(lov->tgts, lov->bufsize);
 485         RETURN(0);
 486 }
 487
 488 static int
 489 lov_add_obd(struct obd_device *obd, struct obd_uuid *uuidp, int index, int gen)
 490 {
 491         struct lov_obd *lov = &obd->u.lov;
 492         struct lov_tgt_desc *tgt;
 493         int rc;
 494         ENTRY;
 495
 496         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d\n",
 497                uuidp->uuid, index, gen);
 498
 499         if ((index < 0) || (index >= LOV_MAX_TGT_COUNT)) {
 500                 CERROR("request to add OBD %s at invalid index: %d\n",
 501                        uuidp->uuid, index);
 502                 RETURN(-EINVAL);
 503         }
 504
 505         if (gen <= 0) {
 506                 CERROR("request to add OBD %s with invalid generation: %d\n",
 507                        uuidp->uuid, gen);
 508                 RETURN(-EINVAL);
 509         }
 510
 511         tgt = lov->tgts + index;
 512         if (!obd_uuid_empty(&tgt->uuid)) {
 513                 CERROR("OBD already assigned at LOV target index %d\n",
 514                        index);
 515                 RETURN(-EEXIST);
 516         }
 517
 518         tgt->uuid = *uuidp;
 519         /* XXX - add a sanity check on the generation number. */
 520         tgt->ltd_gen = gen;
 521
 522         if (index >= lov->desc.ld_tgt_count)
 523                 lov->desc.ld_tgt_count = index + 1;
 524
 525         CDEBUG(D_CONFIG, "idx: %d ltd_gen: %d ld_tgt_count: %d\n",
 526                 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
 527
 528         if (lov->refcount == 0)
 529                 RETURN(0);
 530
 531         if (tgt->ltd_exp) {
 532                 struct obd_device *osc_obd;
 533
 534                 osc_obd = class_exp2obd(tgt->ltd_exp);
 535                 if (osc_obd)
 536                         osc_obd->obd_no_recov = 0;
 537         }
 538
 539         rc = lov_connect_obd(obd, tgt, 1, lov->lov_connect_flags);
 540         if (rc)
 541                 GOTO(out, rc);
 542
 543         if (obd->obd_observer) {
 544                 /* tell the mds_lov about the new target */
 545                 rc = obd_notify(obd->obd_observer, tgt->ltd_exp->exp_obd, 1,
 546                                 (void *)index);
 547         }
 548
 549         GOTO(out, rc);
 550  out:
 551         if (rc && tgt->ltd_exp != NULL)
 552                 lov_disconnect_obd(obd, tgt, 0);
 553         return rc;
 554 }
 555
 556 static int
 557 lov_del_obd(struct obd_device *obd, struct obd_uuid *uuidp, int index, int gen)
 558 {
 559         struct lov_obd *lov = &obd->u.lov;
 560         struct lov_tgt_desc *tgt;
 561         int count = lov->desc.ld_tgt_count;
 562         int rc = 0;
 563         ENTRY;
 564
 565         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d\n",
 566                uuidp->uuid, index, gen);
 567
 568         if (index >= count) {
 569                 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
 570                        index, count);
 571                 RETURN(-EINVAL);
 572         }
 573
 574         tgt = lov->tgts + index;
 575
 576         if (obd_uuid_empty(&tgt->uuid)) {
 577                 CERROR("LOV target at index %d is not setup.\n", index);
 578                 RETURN(-EINVAL);
 579         }
 580
 581         if (strncmp(uuidp->uuid, tgt->uuid.uuid, sizeof uuidp->uuid) != 0) {
 582                 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
 583                        tgt->uuid.uuid, index, uuidp->uuid);
 584                 RETURN(-EINVAL);
 585         }
 586
 587         if (tgt->ltd_exp) {
 588                 struct obd_device *osc_obd;
 589
 590                 osc_obd = class_exp2obd(tgt->ltd_exp);
 591                 if (osc_obd) {
 592                         osc_obd->obd_no_recov = 1;
 593                         rc = obd_llog_finish(osc_obd, &osc_obd->obd_llogs, 1);
 594                         if (rc)
 595                                 CERROR("osc_llog_finish error: %d\n", rc);
 596                 }
 597                 lov_disconnect_obd(obd, tgt, 0);
 598         }
 599
 600         /* XXX - right now there is a dependency on ld_tgt_count being the
 601          * maximum tgt index for computing the mds_max_easize. So we can't
 602          * shrink it. */
 603
 604         /* lt_gen = 0 will mean it will not match the gen of any valid loi */
 605         memset(tgt, 0, sizeof(*tgt));
 606
 607         CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
 608                tgt->uuid.uuid, index, tgt->ltd_gen, tgt->ltd_exp, tgt->active);
 609
 610         RETURN(rc);
 611 }
 612
 613 static int lov_process_config(struct obd_device *obd, obd_count len, void *buf)
 614 {
 615         struct lustre_cfg *lcfg = buf;
 616         struct obd_uuid obd_uuid;
 617         int cmd;
 618         int index;
 619         int gen;
 620         int rc = 0;
 621         ENTRY;
 622
 623         switch(cmd = lcfg->lcfg_command) {
 624         case LCFG_LOV_ADD_OBD:
 625         case LCFG_LOV_DEL_OBD: {
 626                 if (lcfg->lcfg_inllen1 > sizeof(obd_uuid.uuid))
 627                         GOTO(out, rc = -EINVAL);
 628
 629                 obd_str2uuid(&obd_uuid, lcfg->lcfg_inlbuf1);
 630
 631                 if (sscanf(lcfg->lcfg_inlbuf2, "%d", &index) != 1)
 632                         GOTO(out, rc = -EINVAL);
 633                 if (sscanf(lcfg->lcfg_inlbuf3, "%d", &gen) != 1)
 634                         GOTO(out, rc = -EINVAL);
 635                 if (cmd == LCFG_LOV_ADD_OBD)
 636                         rc = lov_add_obd(obd, &obd_uuid, index, gen);
 637                 else
 638                         rc = lov_del_obd(obd, &obd_uuid, index, gen);
 639                 GOTO(out, rc);
 640         }
 641         default: {
 642                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
 643                 GOTO(out, rc = -EINVAL);
 644
 645         }
 646         }
 647 out:
 648         RETURN(rc);
 649 }
 650
 651 #ifndef log2
 652 #define log2(n) ffz(~(n))
 653 #endif
 654
 655 static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
 656                              struct lov_stripe_md **ea,
 657                              struct obd_trans_info *oti)
 658 {
 659         struct lov_obd *lov;
 660         struct obdo *tmp_oa;
 661         struct obd_uuid *ost_uuid = NULL;
 662         int rc = 0, i;
 663         ENTRY;
 664
 665         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
 666                 src_oa->o_flags == OBD_FL_DELORPHAN);
 667
 668         lov = &export->exp_obd->u.lov;
 669
 670         tmp_oa = obdo_alloc();
 671         if (tmp_oa == NULL)
 672                 RETURN(-ENOMEM);
 673
 674         if (src_oa->o_valid & OBD_MD_FLINLINE) {
 675                 ost_uuid = (struct obd_uuid *)src_oa->o_inline;
 676                 CDEBUG(D_HA, "clearing orphans only for %s\n",
 677                        ost_uuid->uuid);
 678         }
 679
 680         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 681                 struct lov_stripe_md obj_md;
 682                 struct lov_stripe_md *obj_mdp = &obj_md;
 683                 int err;
 684
 685                 /* if called for a specific target, we don't
 686                    care if it is not active. */
 687                 if (lov->tgts[i].active == 0 && ost_uuid == NULL) {
 688                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
 689                         continue;
 690                 }
 691
 692                 if (ost_uuid && !obd_uuid_equals(ost_uuid, &lov->tgts[i].uuid))
 693                         continue;
 694
 695                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
 696
 697                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
 698                 err = obd_create(lov->tgts[i].ltd_exp, tmp_oa, &obj_mdp, oti);
 699                 if (err)
 700                         /* This export will be disabled until it is recovered,
 701                            and then orphan recovery will be completed. */
 702                         CERROR("error in orphan recovery on OST idx %d/%d: "
 703                                "rc = %d\n", i, lov->desc.ld_tgt_count, err);
 704
 705                 if (ost_uuid)
 706                         break;
 707         }
 708         obdo_free(tmp_oa);
 709         RETURN(rc);
 710 }
 711
 712 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
 713                         struct lov_stripe_md **ea, struct obd_trans_info *oti)
 714 {
 715         struct lov_stripe_md *obj_mdp, *lsm;
 716         struct lov_obd *lov = &exp->exp_obd->u.lov;
 717         unsigned ost_idx;
 718         int rc, i;
 719         ENTRY;
 720
 721         LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
 722                 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
 723
 724         OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
 725         if (obj_mdp == NULL)
 726                 RETURN(-ENOMEM);
 727
 728         ost_idx = src_oa->o_nlink;
 729         lsm = *ea;
 730         if (lsm == NULL)
 731                 GOTO(out, rc = -EINVAL);
 732         if (ost_idx >= lov->desc.ld_tgt_count)
 733                 GOTO(out, rc = -EINVAL);
 734
 735         for (i = 0; i < lsm->lsm_stripe_count; i++) {
 736                 if (lsm->lsm_oinfo[i].loi_ost_idx == ost_idx) {
 737                         if (lsm->lsm_oinfo[i].loi_id != src_oa->o_id)
 738                                 GOTO(out, rc = -EINVAL);
 739                         break;
 740                 }
 741         }
 742         if (i == lsm->lsm_stripe_count)
 743                 GOTO(out, rc = -EINVAL);
 744
 745         rc = obd_create(lov->tgts[ost_idx].ltd_exp, src_oa, &obj_mdp, oti);
 746 out:
 747         OBD_FREE(obj_mdp, sizeof(*obj_mdp));
 748         RETURN(rc);
 749 }
 750
 751 /* the LOV expects oa->o_id to be set to the LOV object id */
 752 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
 753                       struct lov_stripe_md **ea, struct obd_trans_info *oti)
 754 {
 755         struct lov_request_set *set = NULL;
 756         struct list_head *pos;
 757         struct lov_obd *lov;
 758         int rc = 0;
 759         ENTRY;
 760
 761         LASSERT(ea != NULL);
 762         if (exp == NULL)
 763                 RETURN(-EINVAL);
 764
 765         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
 766             src_oa->o_flags == OBD_FL_DELORPHAN) {
 767                 rc = lov_clear_orphans(exp, src_oa, ea, oti);
 768                 RETURN(rc);
 769         }
 770
 771         lov = &exp->exp_obd->u.lov;
 772         if (!lov->desc.ld_active_tgt_count)
 773                 RETURN(-EIO);
 774
 775         /* Recreate a specific object id at the given OST index */
 776         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
 777             (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
 778                  rc = lov_recreate(exp, src_oa, ea, oti);
 779                  RETURN(rc);
 780         }
 781
 782         rc = lov_prep_create_set(exp, ea, src_oa, oti, &set);
 783         if (rc)
 784                 RETURN(rc);
 785
 786         list_for_each (pos, &set->set_list) {
 787                 struct lov_request *req =
 788                         list_entry(pos, struct lov_request, rq_link);
 789
 790                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
 791                 rc = obd_create(lov->tgts[req->rq_idx].ltd_exp,
 792                                 req->rq_oa, &req->rq_md, oti);
 793                 lov_update_create_set(set, req, rc);
 794         }
 795         rc = lov_fini_create_set(set, ea);
 796         RETURN(rc);
 797 }
 798
 799 #define lsm_bad_magic(LSMP)                                     \
 800 ({                                                              \
 801         struct lov_stripe_md *_lsm__ = (LSMP);                  \
 802         int _ret__ = 0;                                         \
 803         if (!_lsm__) {                                          \
 804                 CERROR("LOV requires striping ea\n");           \
 805                 _ret__ = 1;                                     \
 806         } else if (_lsm__->lsm_magic != LOV_MAGIC) {            \
 807                 CERROR("LOV striping magic bad %#x != %#x\n",   \
 808                        _lsm__->lsm_magic, LOV_MAGIC);           \
 809                 _ret__ = 1;                                     \
 810         }                                                       \
 811         _ret__;                                                 \
 812 })
 813
 814 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
 815                        struct lov_stripe_md *lsm, struct obd_trans_info *oti)
 816 {
 817         struct lov_request_set *set;
 818         struct lov_request *req;
 819         struct list_head *pos;
 820         struct lov_obd *lov;
 821         int rc = 0;
 822         ENTRY;
 823
 824         if (lsm_bad_magic(lsm))
 825                 RETURN(-EINVAL);
 826
 827         if (!exp || !exp->exp_obd)
 828                 RETURN(-ENODEV);
 829
 830         lov = &exp->exp_obd->u.lov;
 831         rc = lov_prep_destroy_set(exp, oa, lsm, oti, &set);
 832         if (rc)
 833                 RETURN(rc);
 834
 835         list_for_each (pos, &set->set_list) {
 836                 int err;
 837                 req = list_entry(pos, struct lov_request, rq_link);
 838
 839                 /* XXX update the cookie position */
 840                 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
 841                 rc = obd_destroy(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa,
 842                                  NULL, oti);
 843                 err = lov_update_common_set(set, req, rc);
 844                 if (rc) {
 845                         CERROR("error: destroying objid "LPX64" subobj "
 846                                LPX64" on OST idx %d: rc = %d\n",
 847                                set->set_oa->o_id, req->rq_oa->o_id,
 848                                req->rq_idx, rc);
 849                         if (!rc)
 850                                 rc = err;
 851                 }
 852         }
 853         lov_fini_destroy_set(set);
 854         RETURN(rc);
 855 }
 856
 857 static int lov_getattr(struct obd_export *exp, struct obdo *oa,
 858                        struct lov_stripe_md *lsm)
 859 {
 860         struct lov_request_set *set;
 861         struct lov_request *req;
 862         struct list_head *pos;
 863         struct lov_obd *lov;
 864         int err = 0, rc = 0;
 865         ENTRY;
 866
 867         if (lsm_bad_magic(lsm))
 868                 RETURN(-EINVAL);
 869
 870         if (!exp || !exp->exp_obd)
 871                 RETURN(-ENODEV);
 872
 873         lov = &exp->exp_obd->u.lov;
 874
 875         rc = lov_prep_getattr_set(exp, oa, lsm, &set);
 876         if (rc)
 877                 RETURN(rc);
 878
 879         list_for_each (pos, &set->set_list) {
 880                 req = list_entry(pos, struct lov_request, rq_link);
 881
 882                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
 883                        "%u\n", oa->o_id, req->rq_stripe, req->rq_oa->o_id,
 884                        req->rq_idx);
 885
 886                 rc = obd_getattr(lov->tgts[req->rq_idx].ltd_exp,
 887                                  req->rq_oa, NULL);
 888                 err = lov_update_common_set(set, req, rc);
 889                 if (err) {
 890                         CERROR("error: getattr objid "LPX64" subobj "
 891                                LPX64" on OST idx %d: rc = %d\n",
 892                                set->set_oa->o_id, req->rq_oa->o_id,
 893                                req->rq_idx, err);
 894                         break;
 895                 }
 896         }
 897
 898         rc = lov_fini_getattr_set(set);
 899         if (err)
 900                 rc = err;
 901         RETURN(rc);
 902 }
 903
 904 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset, void *data,
 905                                  int rc)
 906 {
 907         struct lov_request_set *lovset = (struct lov_request_set *)data;
 908         ENTRY;
 909
 910         /* don't do attribute merge if this aysnc op failed */
 911         if (rc) {
 912                 lovset->set_completes = 0;
 913                 lov_fini_getattr_set(lovset);
 914         } else {
 915                 rc = lov_fini_getattr_set(lovset);
 916         }
 917         RETURN (rc);
 918 }
 919
 920 static int lov_getattr_async(struct obd_export *exp, struct obdo *oa,
 921                               struct lov_stripe_md *lsm,
 922                               struct ptlrpc_request_set *rqset)
 923 {
 924         struct lov_request_set *lovset;
 925         struct lov_obd *lov;
 926         struct list_head *pos;
 927         struct lov_request *req;
 928         int rc = 0;
 929         ENTRY;
 930
 931         if (lsm_bad_magic(lsm))
 932                 RETURN(-EINVAL);
 933
 934         if (!exp || !exp->exp_obd)
 935                 RETURN(-ENODEV);
 936
 937         lov = &exp->exp_obd->u.lov;
 938
 939         rc = lov_prep_getattr_set(exp, oa, lsm, &lovset);
 940         if (rc)
 941                 RETURN(rc);
 942
 943         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
 944                lsm->lsm_object_id, lsm->lsm_stripe_count, lsm->lsm_stripe_size);
 945
 946         list_for_each (pos, &lovset->set_list) {
 947                 req = list_entry(pos, struct lov_request, rq_link);
 948
 949                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
 950                        "%u\n", oa->o_id, req->rq_stripe, req->rq_oa->o_id,
 951                        req->rq_idx);
 952                 rc = obd_getattr_async(lov->tgts[req->rq_idx].ltd_exp,
 953                                        req->rq_oa, NULL, rqset);
 954                 if (rc) {
 955                         CERROR("error: getattr objid "LPX64" subobj "
 956                                LPX64" on OST idx %d: rc = %d\n",
 957                                lovset->set_oa->o_id, req->rq_oa->o_id,
 958                                req->rq_idx, rc);
 959                         GOTO(out, rc);
 960                 }
 961                 lov_update_common_set(lovset, req, rc);
 962         }
 963
 964         LASSERT(rc == 0);
 965         LASSERT (rqset->set_interpret == NULL);
 966         rqset->set_interpret = lov_getattr_interpret;
 967         rqset->set_arg = (void *)lovset;
 968         RETURN(rc);
 969 out:
 970         LASSERT(rc);
 971         lov_fini_getattr_set(lovset);
 972         RETURN(rc);
 973 }
 974
 975 static int lov_setattr(struct obd_export *exp, struct obdo *src_oa,
 976                        struct lov_stripe_md *lsm, struct obd_trans_info *oti)
 977 {
 978         struct lov_request_set *set;
 979         struct lov_obd *lov;
 980         struct list_head *pos;
 981         struct lov_request *req;
 982         int err = 0, rc = 0;
 983         ENTRY;
 984
 985         if (lsm_bad_magic(lsm))
 986                 RETURN(-EINVAL);
 987
 988         if (!exp || !exp->exp_obd)
 989                 RETURN(-ENODEV);
 990
 991         /* for now, we only expect time updates here */
 992         LASSERT(!(src_oa->o_valid & ~(OBD_MD_FLID|OBD_MD_FLTYPE | OBD_MD_FLMODE|
 993                                       OBD_MD_FLATIME | OBD_MD_FLMTIME |
 994                                       OBD_MD_FLCTIME | OBD_MD_FLFLAGS |
 995                                       OBD_MD_FLSIZE | OBD_MD_FLGROUP)));
 996
 997         LASSERT(!(src_oa->o_valid & OBD_MD_FLGROUP) || src_oa->o_gr > 0);
 998
 999         lov = &exp->exp_obd->u.lov;
1000         rc = lov_prep_setattr_set(exp, src_oa, lsm, NULL, &set);
1001         if (rc)
1002                 RETURN(rc);
1003
1004         list_for_each (pos, &set->set_list) {
1005                 req = list_entry(pos, struct lov_request, rq_link);
1006
1007                 rc = obd_setattr(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa,
1008                                  NULL, NULL);
1009                 err = lov_update_common_set(set, req, rc);
1010                 if (err) {
1011                         CERROR("error: setattr objid "LPX64" subobj "
1012                                LPX64" on OST idx %d: rc = %d\n",
1013                                set->set_oa->o_id, req->rq_oa->o_id,
1014                                req->rq_idx, err);
1015                         if (!rc)
1016                                 rc = err;
1017                 }
1018         }
1019         err = lov_fini_setattr_set(set);
1020         if (!rc)
1021                 rc = err;
1022         RETURN(rc);
1023 }
1024
1025 static int lov_revalidate_policy(struct lov_obd *lov, struct lov_stripe_md *lsm)
1026 {
1027         static int next_idx = 0;
1028         struct lov_tgt_desc *tgt;
1029         int i, count;
1030
1031         /* XXX - we should do something clever and take lsm
1032          * into account but just do round robin for now. */
1033
1034         /* last_idx must always be less that count because
1035          * ld_tgt_count currently cannot shrink. */
1036         count = lov->desc.ld_tgt_count;
1037
1038         for (i = next_idx, tgt = lov->tgts + i; i < count; i++, tgt++) {
1039                 if (tgt->active) {
1040                         next_idx = (i + 1) % count;
1041                         RETURN(i);
1042                 }
1043         }
1044
1045         for (i = 0, tgt = lov->tgts; i < next_idx; i++, tgt++) {
1046                 if (tgt->active) {
1047                         next_idx = (i + 1) % count;
1048                         RETURN(i);
1049                 }
1050         }
1051
1052         RETURN(-EIO);
1053 }
1054
1055 static int lov_revalidate_md(struct obd_export *exp, struct obdo *src_oa,
1056                              struct lov_stripe_md *ea,
1057                              struct obd_trans_info *oti)
1058 {
1059         struct obd_export *osc_exp;
1060         struct lov_obd *lov = &exp->exp_obd->u.lov;
1061         struct lov_stripe_md *lsm = ea;
1062         struct lov_stripe_md obj_md;
1063         struct lov_stripe_md *obj_mdp = &obj_md;
1064         struct lov_oinfo *loi;
1065         struct obdo *tmp_oa;
1066         int ost_idx, updates = 0, i;
1067         ENTRY;
1068
1069         tmp_oa = obdo_alloc();
1070         if (tmp_oa == NULL)
1071                 RETURN(-ENOMEM);
1072
1073         loi = lsm->lsm_oinfo;
1074         for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
1075                 int rc;
1076                 if (!obd_uuid_empty(&lov->tgts[loi->loi_ost_idx].uuid))
1077                         continue;
1078
1079                 ost_idx = lov_revalidate_policy(lov, lsm);
1080                 if (ost_idx < 0) {
1081                         /* FIXME: punt for now. */
1082                         CERROR("lov_revalidate_policy failed; no active "
1083                                "OSCs?\n");
1084                         continue;
1085                 }
1086
1087                 /* create a new object */
1088                 memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
1089                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
1090                 osc_exp = lov->tgts[ost_idx].ltd_exp;
1091                 rc = obd_create(osc_exp, tmp_oa, &obj_mdp, oti);
1092                 if (rc) {
1093                         CERROR("error creating new subobj at idx %d; "
1094                                "rc = %d\n", ost_idx, rc);
1095                         continue;
1096                 }
1097                 if (oti->oti_objid)
1098                         oti->oti_objid[ost_idx] = tmp_oa->o_id;
1099                 loi->loi_id = tmp_oa->o_id;
1100                 loi->loi_gr = tmp_oa->o_gr;
1101                 loi->loi_ost_idx = ost_idx;
1102                 loi->loi_ost_gen = lov->tgts[ost_idx].ltd_gen;
1103                 CDEBUG(D_INODE, "replacing objid "LPX64" subobj "LPX64
1104                        " with idx %d gen %d.\n", lsm->lsm_object_id,
1105                        loi->loi_id, ost_idx, loi->loi_ost_gen);
1106                 updates = 1;
1107         }
1108
1109         /* If we got an error revalidating an entry there's no need to
1110          * cleanup up objects we allocated here because the bad entry
1111          * still points to a deleted OST. */
1112
1113         obdo_free(tmp_oa);
1114         RETURN(updates);
1115 }
1116
1117 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
1118  * we can send this 'punch' to just the authoritative node and the nodes
1119  * that the punch will affect. */
1120 static int lov_punch(struct obd_export *exp, struct obdo *oa,
1121                      struct lov_stripe_md *lsm,
1122                      obd_off start, obd_off end, struct obd_trans_info *oti)
1123 {
1124         struct lov_request_set *set;
1125         struct lov_obd *lov;
1126         struct list_head *pos;
1127         struct lov_request *req;
1128         int err = 0, rc = 0;
1129         ENTRY;
1130
1131         if (lsm_bad_magic(lsm))
1132                 RETURN(-EINVAL);
1133
1134         if (!exp || !exp->exp_obd)
1135                 RETURN(-ENODEV);
1136
1137         lov = &exp->exp_obd->u.lov;
1138         rc = lov_prep_punch_set(exp, oa, lsm, start, end, oti, &set);
1139         if (rc)
1140                 RETURN(rc);
1141
1142         list_for_each (pos, &set->set_list) {
1143                 req = list_entry(pos, struct lov_request, rq_link);
1144
1145                 rc = obd_punch(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa,
1146                                NULL, req->rq_extent.start,
1147                                req->rq_extent.end, NULL);
1148                 err = lov_update_punch_set(set, req, rc);
1149                 if (err) {
1150                         CERROR("error: punch objid "LPX64" subobj "LPX64
1151                                " on OST idx %d: rc = %d\n", set->set_oa->o_id,
1152                                req->rq_oa->o_id, req->rq_idx, rc);
1153                         if (!rc)
1154                                 rc = err;
1155                 }
1156         }
1157         err = lov_fini_punch_set(set);
1158         if (!rc)
1159                 rc = err;
1160         RETURN(rc);
1161 }
1162
1163 static int lov_sync(struct obd_export *exp, struct obdo *oa,
1164                     struct lov_stripe_md *lsm, obd_off start, obd_off end)
1165 {
1166         struct lov_request_set *set;
1167         struct lov_obd *lov;
1168         struct list_head *pos;
1169         struct lov_request *req;
1170         int err = 0, rc = 0;
1171         ENTRY;
1172
1173         if (lsm_bad_magic(lsm))
1174                 RETURN(-EINVAL);
1175
1176         if (!exp->exp_obd)
1177                 RETURN(-ENODEV);
1178
1179         lov = &exp->exp_obd->u.lov;
1180         rc = lov_prep_sync_set(exp, oa, lsm, start, end, &set);
1181         if (rc)
1182                 RETURN(rc);
1183
1184         list_for_each (pos, &set->set_list) {
1185                 req = list_entry(pos, struct lov_request, rq_link);
1186
1187                 rc = obd_sync(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa,
1188                               NULL, req->rq_extent.start, req->rq_extent.end);
1189                 err = lov_update_common_set(set, req, rc);
1190                 if (err) {
1191                         CERROR("error: fsync objid "LPX64" subobj "LPX64
1192                                " on OST idx %d: rc = %d\n", set->set_oa->o_id,
1193                                req->rq_oa->o_id, req->rq_idx, rc);
1194                         if (!rc)
1195                                 rc = err;
1196                 }
1197         }
1198         err = lov_fini_sync_set(set);
1199         if (!rc)
1200                 rc = err;
1201         RETURN(rc);
1202 }
1203
1204 static int lov_brw_check(struct lov_obd *lov, struct obdo *oa,
1205                          struct lov_stripe_md *lsm,
1206                          obd_count oa_bufs, struct brw_page *pga)
1207 {
1208         int i, rc = 0;
1209         ENTRY;
1210
1211         /* The caller just wants to know if there's a chance that this
1212          * I/O can succeed */
1213         for (i = 0; i < oa_bufs; i++) {
1214                 int stripe = lov_stripe_number(lsm, pga[i].disk_offset);
1215                 int ost = lsm->lsm_oinfo[stripe].loi_ost_idx;
1216                 obd_off start, end;
1217
1218                 if (!lov_stripe_intersects(lsm, i, pga[i].disk_offset,
1219                                            pga[i].disk_offset + pga[i].count,
1220                                            &start, &end))
1221                         continue;
1222
1223                 if (lov->tgts[ost].active == 0) {
1224                         CDEBUG(D_HA, "lov idx %d inactive\n", ost);
1225                         RETURN(-EIO);
1226                 }
1227                 rc = obd_brw(OBD_BRW_CHECK, lov->tgts[ost].ltd_exp, oa,
1228                              NULL, 1, &pga[i], NULL);
1229                 if (rc)
1230                         break;
1231         }
1232         RETURN(rc);
1233 }
1234
1235 static int lov_brw(int cmd, struct obd_export *exp, struct obdo *src_oa,
1236                    struct lov_stripe_md *lsm, obd_count oa_bufs,
1237                    struct brw_page *pga, struct obd_trans_info *oti)
1238 {
1239         struct lov_request_set *set;
1240         struct lov_request *req;
1241         struct list_head *pos;
1242         struct lov_obd *lov = &exp->exp_obd->u.lov;
1243         int err, rc = 0;
1244         ENTRY;
1245
1246         if (lsm_bad_magic(lsm))
1247                 RETURN(-EINVAL);
1248
1249         if (cmd == OBD_BRW_CHECK) {
1250                 rc = lov_brw_check(lov, src_oa, lsm, oa_bufs, pga);
1251                 RETURN(rc);
1252         }
1253
1254         rc = lov_prep_brw_set(exp, src_oa, lsm, oa_bufs, pga, oti, &set);
1255         if (rc)
1256                 RETURN(rc);
1257
1258         list_for_each (pos, &set->set_list) {
1259                 struct obd_export *sub_exp;
1260                 struct brw_page *sub_pga;
1261                 req = list_entry(pos, struct lov_request, rq_link);
1262
1263                 sub_exp = lov->tgts[req->rq_idx].ltd_exp;
1264                 sub_pga = set->set_pga + req->rq_pgaidx;
1265                 rc = obd_brw(cmd, sub_exp, req->rq_oa, req->rq_md,
1266                              req->rq_oabufs, sub_pga, oti);
1267                 if (rc)
1268                         break;
1269                 lov_update_common_set(set, req, rc);
1270         }
1271
1272         err = lov_fini_brw_set(set);
1273         if (!rc)
1274                 rc = err;
1275         RETURN(rc);
1276 }
1277
1278 static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
1279                              int rc)
1280 {
1281         struct lov_request_set *lovset = (struct lov_request_set *)data;
1282         ENTRY;
1283
1284         if (rc) {
1285                 lovset->set_completes = 0;
1286                 lov_fini_brw_set(lovset);
1287         } else {
1288                 rc = lov_fini_brw_set(lovset);
1289         }
1290
1291         RETURN(rc);
1292 }
1293
1294 static int lov_brw_async(int cmd, struct obd_export *exp, struct obdo *oa,
1295                          struct lov_stripe_md *lsm, obd_count oa_bufs,
1296                          struct brw_page *pga, struct ptlrpc_request_set *set,
1297                          struct obd_trans_info *oti)
1298 {
1299         struct lov_request_set *lovset;
1300         struct lov_request *req;
1301         struct list_head *pos;
1302         struct lov_obd *lov = &exp->exp_obd->u.lov;
1303         int rc = 0;
1304         ENTRY;
1305
1306         if (lsm_bad_magic(lsm))
1307                 RETURN(-EINVAL);
1308
1309         if (cmd == OBD_BRW_CHECK) {
1310                 rc = lov_brw_check(lov, oa, lsm, oa_bufs, pga);
1311                 RETURN(rc);
1312         }
1313
1314         rc = lov_prep_brw_set(exp, oa, lsm, oa_bufs, pga, oti, &lovset);
1315         if (rc)
1316                 RETURN(rc);
1317
1318         list_for_each (pos, &lovset->set_list) {
1319                 struct obd_export *sub_exp;
1320                 struct brw_page *sub_pga;
1321                 req = list_entry(pos, struct lov_request, rq_link);
1322
1323                 sub_exp = lov->tgts[req->rq_idx].ltd_exp;
1324                 sub_pga = lovset->set_pga + req->rq_pgaidx;
1325                 rc = obd_brw_async(cmd, sub_exp, req->rq_oa, req->rq_md,
1326                                    req->rq_oabufs, sub_pga, set, oti);
1327                 if (rc)
1328                         GOTO(out, rc);
1329                 lov_update_common_set(lovset, req, rc);
1330         }
1331         LASSERT(rc == 0);
1332         LASSERT(set->set_interpret == NULL);
1333         set->set_interpret = (set_interpreter_func)lov_brw_interpret;
1334         set->set_arg = (void *)lovset;
1335
1336         RETURN(rc);
1337 out:
1338         lov_fini_brw_set(lovset);
1339         RETURN(rc);
1340 }
1341
1342 static int lov_ap_make_ready(void *data, int cmd)
1343 {
1344         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1345
1346         return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd);
1347 }
1348 static int lov_ap_refresh_count(void *data, int cmd)
1349 {
1350         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1351
1352         return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data,
1353                                                      cmd);
1354 }
1355 static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
1356 {
1357         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1358
1359         lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa);
1360         /* XXX woah, shouldn't we be altering more here?  size? */
1361         oa->o_id = lap->lap_loi_id;
1362 }
1363
1364 static void lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
1365 {
1366         struct lov_async_page *lap = LAP_FROM_COOKIE(data);
1367
1368         /* in a raid1 regime this would down a count of many ios
1369          * in flight, onl calling the caller_ops completion when all
1370          * the raid1 ios are complete */
1371         lap->lap_caller_ops->ap_completion(lap->lap_caller_data, cmd, oa, rc);
1372 }
1373
1374 static struct obd_async_page_ops lov_async_page_ops = {
1375         .ap_make_ready =        lov_ap_make_ready,
1376         .ap_refresh_count =     lov_ap_refresh_count,
1377         .ap_fill_obdo =         lov_ap_fill_obdo,
1378         .ap_completion =        lov_ap_completion,
1379 };
1380
1381 static int lov_prep_async_page(struct obd_export *exp,
1382                                struct lov_stripe_md *lsm,
1383                                struct lov_oinfo *loi, struct page *page,
1384                                obd_off offset, struct obd_async_page_ops *ops,
1385                                void *data, void **res)
1386 {
1387         struct lov_obd *lov = &exp->exp_obd->u.lov;
1388         struct lov_async_page *lap;
1389         int rc, stripe;
1390         ENTRY;
1391
1392         if (lsm_bad_magic(lsm))
1393                 RETURN(-EINVAL);
1394         LASSERT(loi == NULL);
1395
1396         stripe = lov_stripe_number(lsm, offset);
1397         loi = &lsm->lsm_oinfo[stripe];
1398
1399         if (obd_uuid_empty(&lov->tgts[loi->loi_ost_idx].uuid))
1400                 RETURN(-EIO);
1401         if (lov->tgts[loi->loi_ost_idx].active == 0)
1402                 RETURN(-EIO);
1403         if (lov->tgts[loi->loi_ost_idx].ltd_exp == NULL) {
1404                 CERROR("ltd_exp == NULL, but OST idx %d doesn't appear to be "
1405                        "deleted or inactive.\n", loi->loi_ost_idx);
1406                 RETURN(-EIO);
1407         }
1408
1409         OBD_ALLOC(lap, sizeof(*lap));
1410         if (lap == NULL)
1411                 RETURN(-ENOMEM);
1412
1413         lap->lap_magic = LAP_MAGIC;
1414         lap->lap_caller_ops = ops;
1415         lap->lap_caller_data = data;
1416
1417         /* FIXME handle multiple oscs after landing b_raid1 */
1418         lap->lap_stripe = stripe;
1419         switch (lsm->lsm_pattern) {
1420                 case LOV_PATTERN_RAID0:
1421                         lov_stripe_offset(lsm, offset, lap->lap_stripe,
1422                                           &lap->lap_sub_offset);
1423                         break;
1424                 case LOV_PATTERN_CMOBD:
1425                         lap->lap_sub_offset = offset;
1426                         break;
1427                 default:
1428                         LBUG();
1429         }
1430
1431         /* so the callback doesn't need the lsm */
1432         lap->lap_loi_id = loi->loi_id;
1433
1434         rc = obd_prep_async_page(lov->tgts[loi->loi_ost_idx].ltd_exp,
1435                                  lsm, loi, page, lap->lap_sub_offset,
1436                                  &lov_async_page_ops, lap,
1437                                  &lap->lap_sub_cookie);
1438         if (rc) {
1439                 OBD_FREE(lap, sizeof(*lap));
1440                 RETURN(rc);
1441         }
1442         CDEBUG(D_CACHE, "lap %p page %p cookie %p off "LPU64"\n", lap, page,
1443                lap->lap_sub_cookie, offset);
1444         *res = lap;
1445         RETURN(0);
1446 }
1447
1448 static int lov_queue_async_io(struct obd_export *exp,
1449                               struct lov_stripe_md *lsm,
1450                               struct lov_oinfo *loi, void *cookie,
1451                               int cmd, obd_off off, int count,
1452                               obd_flags brw_flags, obd_flags async_flags)
1453 {
1454         struct lov_obd *lov = &exp->exp_obd->u.lov;
1455         struct lov_async_page *lap;
1456         int rc;
1457
1458         LASSERT(loi == NULL);
1459
1460         if (lsm_bad_magic(lsm))
1461                 RETURN(-EINVAL);
1462
1463         lap = LAP_FROM_COOKIE(cookie);
1464
1465         loi = &lsm->lsm_oinfo[lap->lap_stripe];
1466
1467         rc = obd_queue_async_io(lov->tgts[loi->loi_ost_idx].ltd_exp, lsm,
1468                                 loi, lap->lap_sub_cookie, cmd, off, count,
1469                                 brw_flags, async_flags);
1470         RETURN(rc);
1471 }
1472
1473 static int lov_set_async_flags(struct obd_export *exp,
1474                                struct lov_stripe_md *lsm,
1475                                struct lov_oinfo *loi, void *cookie,
1476                                obd_flags async_flags)
1477 {
1478         struct lov_obd *lov = &exp->exp_obd->u.lov;
1479         struct lov_async_page *lap;
1480         int rc;
1481
1482         LASSERT(loi == NULL);
1483
1484         if (lsm_bad_magic(lsm))
1485                 RETURN(-EINVAL);
1486
1487         lap = LAP_FROM_COOKIE(cookie);
1488
1489         loi = &lsm->lsm_oinfo[lap->lap_stripe];
1490
1491         rc = obd_set_async_flags(lov->tgts[loi->loi_ost_idx].ltd_exp,
1492                                  lsm, loi, lap->lap_sub_cookie, async_flags);
1493         RETURN(rc);
1494 }
1495
1496 static int lov_queue_group_io(struct obd_export *exp,
1497                               struct lov_stripe_md *lsm,
1498                               struct lov_oinfo *loi,
1499                               struct obd_io_group *oig, void *cookie,
1500                               int cmd, obd_off off, int count,
1501                               obd_flags brw_flags, obd_flags async_flags)
1502 {
1503         struct lov_obd *lov = &exp->exp_obd->u.lov;
1504         struct lov_async_page *lap;
1505         int rc;
1506
1507         LASSERT(loi == NULL);
1508
1509         if (lsm_bad_magic(lsm))
1510                 RETURN(-EINVAL);
1511
1512         lap = LAP_FROM_COOKIE(cookie);
1513
1514         loi = &lsm->lsm_oinfo[lap->lap_stripe];
1515
1516         rc = obd_queue_group_io(lov->tgts[loi->loi_ost_idx].ltd_exp, lsm, loi,
1517                                 oig, lap->lap_sub_cookie, cmd, off, count,
1518                                 brw_flags, async_flags);
1519         RETURN(rc);
1520 }
1521
1522 /* this isn't exactly optimal.  we may have queued sync io in oscs on
1523  * all stripes, but we don't record that fact at queue time.  so we
1524  * trigger sync io on all stripes. */
1525 static int lov_trigger_group_io(struct obd_export *exp,
1526                                 struct lov_stripe_md *lsm,
1527                                 struct lov_oinfo *loi,
1528                                 struct obd_io_group *oig)
1529 {
1530         struct lov_obd *lov = &exp->exp_obd->u.lov;
1531         int rc = 0, i, err;
1532
1533         LASSERT(loi == NULL);
1534
1535         if (lsm_bad_magic(lsm))
1536                 RETURN(-EINVAL);
1537
1538         loi = lsm->lsm_oinfo;
1539         for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
1540                 if (lov->tgts[loi->loi_ost_idx].active == 0) {
1541                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1542                         continue;
1543                 }
1544
1545                 err = obd_trigger_group_io(lov->tgts[loi->loi_ost_idx].ltd_exp,
1546                                            lsm, loi, oig);
1547                 if (rc == 0 && err != 0)
1548                         rc = err;
1549         };
1550         RETURN(rc);
1551 }
1552
1553 static int lov_teardown_async_page(struct obd_export *exp,
1554                                    struct lov_stripe_md *lsm,
1555                                    struct lov_oinfo *loi, void *cookie)
1556 {
1557         struct lov_obd *lov = &exp->exp_obd->u.lov;
1558         struct lov_async_page *lap;
1559         int rc;
1560
1561         LASSERT(loi == NULL);
1562
1563         if (lsm_bad_magic(lsm))
1564                 RETURN(-EINVAL);
1565
1566         lap = LAP_FROM_COOKIE(cookie);
1567
1568         loi = &lsm->lsm_oinfo[lap->lap_stripe];
1569
1570         rc = obd_teardown_async_page(lov->tgts[loi->loi_ost_idx].ltd_exp,
1571                                      lsm, loi, lap->lap_sub_cookie);
1572         if (rc) {
1573                 CERROR("unable to teardown sub cookie %p: %d\n",
1574                        lap->lap_sub_cookie, rc);
1575                 RETURN(rc);
1576         }
1577         OBD_FREE(lap, sizeof(*lap));
1578         RETURN(rc);
1579 }
1580
1581 static int lov_enqueue(struct obd_export *exp, struct lov_stripe_md *lsm,
1582                        __u32 type, ldlm_policy_data_t *policy, __u32 mode,
1583                        int *flags, void *bl_cb, void *cp_cb, void *gl_cb,
1584                        void *data,__u32 lvb_len, void *lvb_swabber,
1585                        struct lustre_handle *lockh)
1586 {
1587         struct lov_request_set *set;
1588         struct lov_request *req;
1589         struct list_head *pos;
1590         struct lustre_handle *lov_lockhp;
1591         struct lov_obd *lov;
1592         ldlm_error_t rc;
1593         int save_flags = *flags;
1594         ENTRY;
1595
1596         if (lsm_bad_magic(lsm))
1597                 RETURN(-EINVAL);
1598
1599         /* we should never be asked to replay a lock this way. */
1600         LASSERT((*flags & LDLM_FL_REPLAY) == 0);
1601
1602         if (!exp || !exp->exp_obd)
1603                 RETURN(-ENODEV);
1604
1605         lov = &exp->exp_obd->u.lov;
1606         rc = lov_prep_enqueue_set(exp, lsm, policy, mode, lockh, &set);
1607         if (rc)
1608                 RETURN(rc);
1609
1610         list_for_each (pos, &set->set_list) {
1611                 ldlm_policy_data_t sub_policy;
1612                 req = list_entry(pos, struct lov_request, rq_link);
1613                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1614                 LASSERT(lov_lockhp);
1615
1616                 *flags = save_flags;
1617                 sub_policy.l_extent.start = req->rq_extent.start;
1618                 sub_policy.l_extent.end = req->rq_extent.end;
1619
1620                 rc = obd_enqueue(lov->tgts[req->rq_idx].ltd_exp, req->rq_md,
1621                                  type, &sub_policy, mode, flags, bl_cb,
1622                                  cp_cb, gl_cb, data, lvb_len, lvb_swabber,
1623                                  lov_lockhp);
1624                 rc = lov_update_enqueue_set(set, req, rc, save_flags);
1625                 if (rc != ELDLM_OK)
1626                         break;
1627         }
1628
1629         lov_fini_enqueue_set(set, mode);
1630         RETURN(rc);
1631 }
1632
1633 static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
1634                      __u32 type, ldlm_policy_data_t *policy, __u32 mode,
1635                      int *flags, void *data, struct lustre_handle *lockh)
1636 {
1637         struct lov_request_set *set;
1638         struct lov_request *req;
1639         struct list_head *pos;
1640         struct lov_obd *lov = &exp->exp_obd->u.lov;
1641         struct lustre_handle *lov_lockhp;
1642         int lov_flags, rc = 0;
1643         ENTRY;
1644
1645         if (lsm_bad_magic(lsm))
1646                 RETURN(-EINVAL);
1647
1648         if (!exp || !exp->exp_obd)
1649                 RETURN(-ENODEV);
1650
1651         lov = &exp->exp_obd->u.lov;
1652         rc = lov_prep_match_set(exp, lsm, policy, mode, lockh, &set);
1653         if (rc)
1654                 RETURN(rc);
1655
1656         list_for_each (pos, &set->set_list) {
1657                 ldlm_policy_data_t sub_policy;
1658                 req = list_entry(pos, struct lov_request, rq_link);
1659                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1660                 LASSERT(lov_lockhp);
1661
1662                 sub_policy.l_extent.start = req->rq_extent.start;
1663                 sub_policy.l_extent.end = req->rq_extent.end;
1664                 lov_flags = *flags;
1665
1666                 rc = obd_match(lov->tgts[req->rq_idx].ltd_exp, req->rq_md,
1667                                type, &sub_policy, mode, &lov_flags, data,
1668                                lov_lockhp);
1669                 rc = lov_update_match_set(set, req, rc);
1670                 if (rc != 1)
1671                         break;
1672         }
1673         lov_fini_match_set(set, mode, *flags);
1674         RETURN(rc);
1675 }
1676
1677 static int lov_change_cbdata(struct obd_export *exp,
1678                              struct lov_stripe_md *lsm, ldlm_iterator_t it,
1679                              void *data)
1680 {
1681         struct lov_obd *lov;
1682         struct lov_oinfo *loi;
1683         int rc = 0, i;
1684         ENTRY;
1685
1686         if (lsm_bad_magic(lsm))
1687                 RETURN(-EINVAL);
1688
1689         if (!exp || !exp->exp_obd)
1690                 RETURN(-ENODEV);
1691
1692         LASSERT(lsm->lsm_object_gr > 0);
1693
1694         lov = &exp->exp_obd->u.lov;
1695         for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
1696                 struct lov_stripe_md submd;
1697                 if (lov->tgts[loi->loi_ost_idx].active == 0) {
1698                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1699                         continue;
1700                 }
1701
1702                 submd.lsm_object_id = loi->loi_id;
1703                 submd.lsm_object_gr = lsm->lsm_object_gr;
1704                 submd.lsm_stripe_count = 0;
1705                 rc = obd_change_cbdata(lov->tgts[loi->loi_ost_idx].ltd_exp,
1706                                        &submd, it, data);
1707         }
1708         RETURN(rc);
1709 }
1710
1711 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
1712                       __u32 mode, struct lustre_handle *lockh)
1713 {
1714         struct lov_request_set *set;
1715         struct lov_request *req;
1716         struct list_head *pos;
1717         struct lov_obd *lov = &exp->exp_obd->u.lov;
1718         struct lustre_handle *lov_lockhp;
1719         int err = 0, rc = 0;
1720         ENTRY;
1721
1722         if (lsm_bad_magic(lsm))
1723                 RETURN(-EINVAL);
1724
1725         if (!exp || !exp->exp_obd)
1726                 RETURN(-ENODEV);
1727
1728         LASSERT(lsm->lsm_object_gr > 0);
1729
1730         LASSERT(lockh);
1731         lov = &exp->exp_obd->u.lov;
1732         rc = lov_prep_cancel_set(exp, lsm, mode, lockh, &set);
1733         if (rc)
1734                 RETURN(rc);
1735
1736         list_for_each (pos, &set->set_list) {
1737                 req = list_entry(pos, struct lov_request, rq_link);
1738                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
1739
1740                 rc = obd_cancel(lov->tgts[req->rq_idx].ltd_exp, req->rq_md,
1741                                 mode, lov_lockhp);
1742                 rc = lov_update_common_set(set, req, rc);
1743                 if (rc) {
1744                         CERROR("error: cancel objid "LPX64" subobj "
1745                                LPX64" on OST idx %d: rc = %d\n",
1746                                lsm->lsm_object_id,
1747                                req->rq_md->lsm_object_id, req->rq_idx, rc);
1748                         err = rc;
1749                 }
1750
1751         }
1752         lov_fini_cancel_set(set);
1753         RETURN(err);
1754 }
1755
1756 static int lov_cancel_unused(struct obd_export *exp,
1757                              struct lov_stripe_md *lsm,
1758                              int flags, void *opaque)
1759 {
1760         struct lov_obd *lov;
1761         struct lov_oinfo *loi;
1762         int rc = 0, i;
1763         ENTRY;
1764
1765         lov = &exp->exp_obd->u.lov;
1766         if (lsm == NULL) {
1767                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1768                         int err = obd_cancel_unused(lov->tgts[i].ltd_exp, NULL,
1769                                                     flags, opaque);
1770                         if (!rc)
1771                                 rc = err;
1772                 }
1773                 RETURN(rc);
1774         }
1775
1776         if (lsm_bad_magic(lsm))
1777                 RETURN(-EINVAL);
1778
1779         if (!exp || !exp->exp_obd)
1780                 RETURN(-ENODEV);
1781
1782         LASSERT(lsm->lsm_object_gr > 0);
1783
1784         for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
1785                 struct lov_stripe_md submd;
1786                 int err;
1787
1788                 if (lov->tgts[loi->loi_ost_idx].active == 0)
1789                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1790
1791                 submd.lsm_object_id = loi->loi_id;
1792                 submd.lsm_object_gr = lsm->lsm_object_gr;
1793                 submd.lsm_stripe_count = 0;
1794                 err = obd_cancel_unused(lov->tgts[loi->loi_ost_idx].ltd_exp,
1795                                         &submd, flags, opaque);
1796                 if (err && lov->tgts[loi->loi_ost_idx].active) {
1797                         CERROR("error: cancel unused objid "LPX64" subobj "LPX64
1798                                " on OST idx %d: rc = %d\n", lsm->lsm_object_id,
1799                                loi->loi_id, loi->loi_ost_idx, err);
1800                         if (!rc)
1801                                 rc = err;
1802                 }
1803         }
1804         RETURN(rc);
1805 }
1806
1807 #define LOV_U64_MAX ((__u64)~0ULL)
1808 #define LOV_SUM_MAX(tot, add)                                           \
1809         do {                                                            \
1810                 if ((tot) + (add) < (tot))                              \
1811                         (tot) = LOV_U64_MAX;                            \
1812                 else                                                    \
1813                         (tot) += (add);                                 \
1814         } while(0)
1815
1816 static int lov_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1817                       unsigned long max_age)
1818 {
1819         struct lov_obd *lov = &obd->u.lov;
1820         struct obd_statfs lov_sfs;
1821         int set = 0;
1822         int rc = 0;
1823         int i;
1824         ENTRY;
1825
1826
1827         /* We only get block data from the OBD */
1828         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1829                 int err;
1830                 if (!lov->tgts[i].active) {
1831                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
1832                         continue;
1833                 }
1834
1835                 err = obd_statfs(class_exp2obd(lov->tgts[i].ltd_exp), &lov_sfs,
1836                                  max_age);
1837                 if (err) {
1838                         if (lov->tgts[i].active && !rc)
1839                                 rc = err;
1840                         continue;
1841                 }
1842
1843                 if (!set) {
1844                         memcpy(osfs, &lov_sfs, sizeof(lov_sfs));
1845                         set = 1;
1846                 } else {
1847                         osfs->os_bfree += lov_sfs.os_bfree;
1848                         osfs->os_bavail += lov_sfs.os_bavail;
1849                         osfs->os_blocks += lov_sfs.os_blocks;
1850                         /* XXX not sure about this one - depends on policy.
1851                          *   - could be minimum if we always stripe on all OBDs
1852                          *     (but that would be wrong for any other policy,
1853                          *     if one of the OBDs has no more objects left)
1854                          *   - could be sum if we stripe whole objects
1855                          *   - could be average, just to give a nice number
1856                          *
1857                          * To give a "reasonable" (if not wholly accurate)
1858                          * number, we divide the total number of free objects
1859                          * by expected stripe count (watch out for overflow).
1860                          */
1861                         LOV_SUM_MAX(osfs->os_files, lov_sfs.os_files);
1862                         LOV_SUM_MAX(osfs->os_ffree, lov_sfs.os_ffree);
1863                 }
1864         }
1865
1866         if (set) {
1867                 __u32 expected_stripes = lov->desc.ld_default_stripe_count ?
1868                                          lov->desc.ld_default_stripe_count :
1869                                          lov->desc.ld_active_tgt_count;
1870
1871                 if (osfs->os_files != LOV_U64_MAX)
1872                         do_div(osfs->os_files, expected_stripes);
1873                 if (osfs->os_ffree != LOV_U64_MAX)
1874                         do_div(osfs->os_ffree, expected_stripes);
1875         } else if (!rc)
1876                 rc = -EIO;
1877
1878         RETURN(rc);
1879 }
1880
1881 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1882                          void *karg, void *uarg)
1883 {
1884         struct obd_device *obddev = class_exp2obd(exp);
1885         struct lov_obd *lov = &obddev->u.lov;
1886         int i, rc, count = lov->desc.ld_tgt_count;
1887         struct obd_uuid *uuidp;
1888         ENTRY;
1889
1890         switch (cmd) {
1891         case OBD_IOC_LOV_GET_CONFIG: {
1892                 struct obd_ioctl_data *data = karg;
1893                 struct lov_tgt_desc *tgtdesc;
1894                 struct lov_desc *desc;
1895                 char *buf = NULL;
1896                 __u32 *genp;
1897
1898                 buf = NULL;
1899                 len = 0;
1900                 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1901                         RETURN(-EINVAL);
1902
1903                 data = (struct obd_ioctl_data *)buf;
1904
1905                 if (sizeof(*desc) > data->ioc_inllen1) {
1906                         obd_ioctl_freedata(buf, len);
1907                         RETURN(-EINVAL);
1908                 }
1909
1910                 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1911                         obd_ioctl_freedata(buf, len);
1912                         RETURN(-EINVAL);
1913                 }
1914
1915                 if (sizeof(__u32) * count > data->ioc_inllen3) {
1916                         obd_ioctl_freedata(buf, len);
1917                         RETURN(-EINVAL);
1918                 }
1919
1920                 desc = (struct lov_desc *)data->ioc_inlbuf1;
1921                 memcpy(desc, &(lov->desc), sizeof(*desc));
1922
1923                 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1924                 genp = (__u32 *)data->ioc_inlbuf3;
1925                 tgtdesc = lov->tgts;
1926                 /* the uuid will be empty for deleted OSTs */
1927                 for (i = 0; i < count; i++, uuidp++, genp++, tgtdesc++) {
1928                         obd_str2uuid(uuidp, tgtdesc->uuid.uuid);
1929                         *genp = tgtdesc->ltd_gen;
1930                 }
1931
1932                 rc = copy_to_user((void *)uarg, buf, len);
1933                 if (rc)
1934                         rc = -EFAULT;
1935                 obd_ioctl_freedata(buf, len);
1936                 break;
1937         }
1938         case LL_IOC_LOV_SETSTRIPE:
1939                 rc = lov_setstripe(exp, karg, uarg);
1940                 break;
1941         case LL_IOC_LOV_GETSTRIPE:
1942                 rc = lov_getstripe(exp, karg, uarg);
1943                 break;
1944         case LL_IOC_LOV_SETEA:
1945                 rc = lov_setea(exp, karg, uarg);
1946                 break;
1947         default: {
1948                 int set = 0;
1949                 if (count == 0)
1950                         RETURN(-ENOTTY);
1951                 rc = 0;
1952                 for (i = 0; i < count; i++) {
1953                         int err;
1954
1955                         /* OST was deleted */
1956                         if (obd_uuid_empty(&lov->tgts[i].uuid))
1957                                 continue;
1958
1959                         err = obd_iocontrol(cmd, lov->tgts[i].ltd_exp,
1960                                             len, karg, uarg);
1961                         if (err) {
1962                                 if (lov->tgts[i].active) {
1963                                         CERROR("error: iocontrol OSC %s on OST "
1964                                                "idx %d cmd %x: err = %d\n",
1965                                                lov->tgts[i].uuid.uuid, i,
1966                                                cmd, err);
1967                                         if (!rc)
1968                                                 rc = err;
1969                                 }
1970                         } else
1971                                 set = 1;
1972                 }
1973                 if (!set && !rc)
1974                         rc = -EIO;
1975         }
1976         }
1977
1978         RETURN(rc);
1979 }
1980
1981 static int lov_get_info(struct obd_export *exp, __u32 keylen,
1982                         void *key, __u32 *vallen, void *val)
1983 {
1984         struct obd_device *obddev = class_exp2obd(exp);
1985         struct lov_obd *lov = &obddev->u.lov;
1986         int i;
1987         ENTRY;
1988
1989         if (!vallen || !val)
1990                 RETURN(-EFAULT);
1991
1992         if (keylen > strlen("lock_to_stripe") &&
1993             strcmp(key, "lock_to_stripe") == 0) {
1994                 struct {
1995                         char name[16];
1996                         struct ldlm_lock *lock;
1997                         struct lov_stripe_md *lsm;
1998                 } *data = key;
1999                 struct lov_oinfo *loi;
2000                 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2001                 __u32 *stripe = val;
2002
2003                 if (*vallen < sizeof(*stripe))
2004                         RETURN(-EFAULT);
2005                 *vallen = sizeof(*stripe);
2006
2007                 /* XXX This is another one of those bits that will need to
2008                  * change if we ever actually support nested LOVs.  It uses
2009                  * the lock's export to find out which stripe it is. */
2010                 /* XXX - it's assumed all the locks for deleted OSTs have
2011                  * been cancelled. Also, the export for deleted OSTs will
2012                  * be NULL and won't match the lock's export. */
2013                 for (i = 0, loi = data->lsm->lsm_oinfo;
2014                      i < data->lsm->lsm_stripe_count;
2015                      i++, loi++) {
2016                         if (lov->tgts[loi->loi_ost_idx].ltd_exp ==
2017                                         data->lock->l_conn_export &&
2018                             loi->loi_id == res_id->name[0] &&
2019                             loi->loi_gr == res_id->name[2]) {
2020                                 *stripe = i;
2021                                 RETURN(0);
2022                         }
2023                 }
2024                 LDLM_ERROR(data->lock, "lock on inode without such object\n");
2025                 dump_lsm(D_ERROR, data->lsm);
2026                 RETURN(-ENXIO);
2027         } else if (keylen >= strlen("size_to_stripe") &&
2028                    strcmp(key, "size_to_stripe") == 0) {
2029                 struct {
2030                         int stripe_number;
2031                         __u64 size;
2032                         struct lov_stripe_md *lsm;
2033                 } *data = val;
2034
2035                 if (*vallen < sizeof(*data))
2036                         RETURN(-EFAULT);
2037
2038                 data->size = lov_size_to_stripe(data->lsm, data->size,
2039                                                 data->stripe_number);
2040                 RETURN(0);
2041         } else if (keylen >= strlen("last_id") && strcmp(key, "last_id") == 0) {
2042                 obd_id *ids = val;
2043                 int rc, size = sizeof(obd_id);
2044                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2045                         if (!lov->tgts[i].active)
2046                                 continue;
2047                         rc = obd_get_info(lov->tgts[i].ltd_exp,
2048                                           keylen, key, &size, &(ids[i]));
2049                         if (rc != 0)
2050                                 RETURN(rc);
2051                 }
2052                 RETURN(0);
2053         } else if (keylen >= strlen("lovdesc") && strcmp(key, "lovdesc") == 0) {
2054                 struct lov_desc *desc_ret = val;
2055                 *desc_ret = lov->desc;
2056
2057                 RETURN(0);
2058         }
2059
2060         RETURN(-EINVAL);
2061 }
2062
2063 static int lov_set_info(struct obd_export *exp, obd_count keylen,
2064                         void *key, obd_count vallen, void *val)
2065 {
2066         struct obd_device *obddev = class_exp2obd(exp);
2067         struct lov_obd *lov = &obddev->u.lov;
2068         int i, rc = 0, err;
2069         ENTRY;
2070
2071 #define KEY_IS(str) \
2072         (keylen == strlen(str) && memcmp(key, str, keylen) == 0)
2073
2074         if (KEY_IS("next_id")) {
2075                 if (vallen != lov->desc.ld_tgt_count)
2076                         RETURN(-EINVAL);
2077                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2078                         /* initialize all OSCs, even inactive ones */
2079
2080                         err = obd_set_info(lov->tgts[i].ltd_exp,
2081                                           keylen, key, sizeof(obd_id),
2082                                           ((obd_id*)val) + i);
2083                         if (!rc)
2084                                 rc = err;
2085                 }
2086                 RETURN(rc);
2087         }
2088         if (KEY_IS("async")) {
2089                 struct lov_desc *desc = &lov->desc;
2090                 struct lov_tgt_desc *tgts = lov->tgts;
2091
2092                 if (vallen != sizeof(int))
2093                         RETURN(-EINVAL);
2094                 lov->async = *((int*) val);
2095
2096                 for (i = 0; i < desc->ld_tgt_count; i++, tgts++) {
2097                         struct obd_uuid *tgt_uuid = &tgts->uuid;
2098                         struct obd_device *tgt_obd;
2099
2100                         tgt_obd = class_find_client_obd(tgt_uuid,
2101                                                         LUSTRE_OSC_NAME,
2102                                                         &obddev->obd_uuid);
2103                         if (!tgt_obd) {
2104                                 CERROR("Target %s not attached\n",
2105                                         tgt_uuid->uuid);
2106                                 if (!rc)
2107                                         rc = -EINVAL;
2108                                 continue;
2109                         }
2110
2111                         err = obd_set_info(tgt_obd->obd_self_export,
2112                                            keylen, key, vallen, val);
2113                         if (err) {
2114                                 CERROR("Failed to set async on target %s\n",
2115                                         tgt_obd->obd_name);
2116                                 if (!rc)
2117                                         rc = err;
2118                         }
2119                 }
2120                 RETURN(rc);
2121         }
2122
2123         if (KEY_IS("growth_count")) {
2124                 if (vallen != sizeof(int))
2125                         RETURN(-EINVAL);
2126         } else if (KEY_IS("mds_conn")) {
2127                 if (vallen != sizeof(__u32))
2128                         RETURN(-EINVAL);
2129         } else if (KEY_IS("unlinked") || KEY_IS("unrecovery")) {
2130                 if (vallen != 0)
2131                         RETURN(-EINVAL);
2132         } else if (KEY_IS("sec")) {
2133                 struct lov_tgt_desc *tgt;
2134                 struct obd_export *exp;
2135                 int rc = 0, err, i;
2136
2137                 spin_lock(&lov->lov_lock);
2138                 for (i = 0, tgt = lov->tgts; i < lov->desc.ld_tgt_count;
2139                      i++, tgt++) {
2140                         exp = tgt->ltd_exp;
2141                         /* during setup time the connections to osc might
2142                          * haven't been established.
2143                          */
2144                         if (exp == NULL) {
2145                                 struct obd_device *tgt_obd;
2146
2147                                 tgt_obd = class_find_client_obd(&tgt->uuid,
2148                                                                 LUSTRE_OSC_NAME,
2149                                                                 &obddev->obd_uuid);
2150                                 if (!tgt_obd) {
2151                                         CERROR("can't set security flavor, "
2152                                                "device %s not attached?\n",
2153                                                 tgt->uuid.uuid);
2154                                         rc = -EINVAL;
2155                                         continue;
2156                                 }
2157                                 exp = tgt_obd->obd_self_export;
2158                         }
2159
2160                         err = obd_set_info(exp, keylen, key, vallen, val);
2161                         if (!rc)
2162                                 rc = err;
2163                 }
2164                 spin_unlock(&lov->lov_lock);
2165
2166                 RETURN(rc);
2167         } else {
2168                 RETURN(-EINVAL);
2169         }
2170
2171         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2172                 if (val && !obd_uuid_equals(val, &lov->tgts[i].uuid))
2173                         continue;
2174
2175                 if (!val && !lov->tgts[i].active)
2176                         continue;
2177
2178                 err = obd_set_info(lov->tgts[i].ltd_exp,
2179                                   keylen, key, vallen, val);
2180                 if (!rc)
2181                         rc = err;
2182         }
2183         RETURN(rc);
2184 #undef KEY_IS
2185 }
2186
2187 #if 0
2188 struct lov_multi_wait {
2189         struct ldlm_lock *lock;
2190         wait_queue_t      wait;
2191         int               completed;
2192         int               generation;
2193 };
2194
2195 int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
2196                       struct lustre_handle *lockh)
2197 {
2198         struct lov_lock_handles *lov_lockh = NULL;
2199         struct lustre_handle *lov_lockhp;
2200         struct lov_obd *lov;
2201         struct lov_oinfo *loi;
2202         struct lov_multi_wait *queues;
2203         int rc = 0, i;
2204         ENTRY;
2205
2206         if (lsm_bad_magic(lsm))
2207                 RETURN(-EINVAL);
2208
2209         if (!exp || !exp->exp_obd)
2210                 RETURN(-ENODEV);
2211
2212         LASSERT(lockh != NULL);
2213         if (lsm->lsm_stripe_count > 1) {
2214                 lov_lockh = lov_handle2llh(lockh);
2215                 if (lov_lockh == NULL) {
2216                         CERROR("LOV: invalid lov lock handle %p\n", lockh);
2217                         RETURN(-EINVAL);
2218                 }
2219
2220                 lov_lockhp = lov_lockh->llh_handles;
2221         } else {
2222                 lov_lockhp = lockh;
2223         }
2224
2225         OBD_ALLOC(queues, lsm->lsm_stripe_count * sizeof(*queues));
2226         if (queues == NULL)
2227                 GOTO(out, rc = -ENOMEM);
2228
2229         lov = &exp->exp_obd->u.lov;
2230         for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
2231              i++, loi++, lov_lockhp++) {
2232                 struct ldlm_lock *lock;
2233                 struct obd_device *obd;
2234                 unsigned long irqflags;
2235
2236                 lock = ldlm_handle2lock(lov_lockhp);
2237                 if (lock == NULL) {
2238                         CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
2239                                loi->loi_ost_idx, loi->loi_id);
2240                         queues[i].completed = 1;
2241                         continue;
2242                 }
2243
2244                 queues[i].lock = lock;
2245                 init_waitqueue_entry(&(queues[i].wait), current);
2246                 add_wait_queue(lock->l_waitq, &(queues[i].wait));
2247
2248                 obd = class_exp2obd(lock->l_conn_export);
2249                 if (obd != NULL)
2250                         imp = obd->u.cli.cl_import;
2251                 if (imp != NULL) {
2252                         spin_lock_irqsave(&imp->imp_lock, irqflags);
2253                         queues[i].generation = imp->imp_generation;
2254                         spin_unlock_irqrestore(&imp->imp_lock, irqflags);
2255                 }
2256         }
2257
2258         lwi = LWI_TIMEOUT_INTR(obd_timeout * HZ, ldlm_expired_completion_wait,
2259                                interrupted_completion_wait, &lwd);
2260         rc = l_wait_event_added(check_multi_complete(queues, lsm), &lwi);
2261
2262         for (i = 0; i < lsm->lsm_stripe_count; i++)
2263                 remove_wait_queue(lock->l_waitq, &(queues[i].wait));
2264
2265         if (rc == -EINTR || rc == -ETIMEDOUT) {
2266
2267
2268         }
2269
2270  out:
2271         if (lov_lockh != NULL)
2272                 lov_llh_put(lov_lockh);
2273         RETURN(rc);
2274 }
2275 #endif
2276
2277 struct obd_ops lov_obd_ops = {
2278         .o_owner               = THIS_MODULE,
2279         .o_attach              = lov_attach,
2280         .o_detach              = lov_detach,
2281         .o_setup               = lov_setup,
2282         .o_cleanup             = lov_cleanup,
2283         .o_process_config      = lov_process_config,
2284         .o_connect             = lov_connect,
2285         .o_disconnect          = lov_disconnect,
2286         .o_statfs              = lov_statfs,
2287         .o_packmd              = lov_packmd,
2288         .o_unpackmd            = lov_unpackmd,
2289         .o_revalidate_md       = lov_revalidate_md,
2290         .o_create              = lov_create,
2291         .o_destroy             = lov_destroy,
2292         .o_getattr             = lov_getattr,
2293         .o_getattr_async       = lov_getattr_async,
2294         .o_setattr             = lov_setattr,
2295         .o_brw                 = lov_brw,
2296         .o_brw_async           = lov_brw_async,
2297         .o_prep_async_page     = lov_prep_async_page,
2298         .o_queue_async_io      = lov_queue_async_io,
2299         .o_set_async_flags     = lov_set_async_flags,
2300         .o_queue_group_io      = lov_queue_group_io,
2301         .o_trigger_group_io    = lov_trigger_group_io,
2302         .o_teardown_async_page = lov_teardown_async_page,
2303         .o_adjust_kms          = lov_adjust_kms,
2304         .o_punch               = lov_punch,
2305         .o_sync                = lov_sync,
2306         .o_enqueue             = lov_enqueue,
2307         .o_match               = lov_match,
2308         .o_change_cbdata       = lov_change_cbdata,
2309         .o_cancel              = lov_cancel,
2310         .o_cancel_unused       = lov_cancel_unused,
2311         .o_iocontrol           = lov_iocontrol,
2312         .o_get_info            = lov_get_info,
2313         .o_set_info            = lov_set_info,
2314         .o_llog_init           = lov_llog_init,
2315         .o_llog_finish         = lov_llog_finish,
2316         .o_notify              = lov_notify,
2317 };
2318
2319 int __init lov_init(void)
2320 {
2321         struct lprocfs_static_vars lvars;
2322         int rc;
2323         ENTRY;
2324
2325         lprocfs_init_vars(lov, &lvars);
2326         rc = class_register_type(&lov_obd_ops, NULL, lvars.module_vars,
2327                                  OBD_LOV_DEVICENAME);
2328         RETURN(rc);
2329 }
2330
2331 #ifdef __KERNEL__
2332 static void /*__exit*/ lov_exit(void)
2333 {
2334         class_unregister_type(OBD_LOV_DEVICENAME);
2335 }
2336
2337 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2338 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2339 MODULE_LICENSE("GPL");
2340
2341 module_init(lov_init);
2342 module_exit(lov_exit);
2343 #endif