lustre/lmv/lmv_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  *
   6  *   This file is part of Lustre, http://www.lustre.org.
   7  *
   8  *   Lustre is free software; you can redistribute it and/or
   9  *   modify it under the terms of version 2 of the GNU General Public
  10  *   License as published by the Free Software Foundation.
  11  *
  12  *   Lustre is distributed in the hope that it will be useful,
  13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *   GNU General Public License for more details.
  16  *
  17  *   You should have received a copy of the GNU General Public License
  18  *   along with Lustre; if not, write to the Free Software
  19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  */
  21
  22 #ifndef EXPORT_SYMTAB
  23 # define EXPORT_SYMTAB
  24 #endif
  25 #define DEBUG_SUBSYSTEM S_LMV
  26 #ifdef __KERNEL__
  27 #include <linux/slab.h>
  28 #include <linux/module.h>
  29 #include <linux/init.h>
  30 #include <linux/slab.h>
  31 #include <linux/pagemap.h>
  32 #include <asm/div64.h>
  33 #include <linux/seq_file.h>
  34 #include <linux/namei.h>
  35 #else
  36 #include <liblustre.h>
  37 #endif
  38 #include <linux/ext2_fs.h>
  39
  40 #include <linux/obd_support.h>
  41 #include <linux/lustre_lib.h>
  42 #include <linux/lustre_net.h>
  43 #include <linux/lustre_idl.h>
  44 #include <linux/lustre_dlm.h>
  45 #include <linux/lustre_mds.h>
  46 #include <linux/obd_class.h>
  47 #include <linux/obd_ost.h>
  48 #include <linux/lprocfs_status.h>
  49 #include <linux/lustre_fsfilt.h>
  50 #include <linux/obd_lmv.h>
  51 #include <linux/lustre_lite.h>
  52 #include "lmv_internal.h"
  53
  54 /* not defined for liblustre building */
  55 #if !defined(ATOMIC_INIT)
  56 #define ATOMIC_INIT(val) { (val) }
  57 #endif
  58
  59 /* object cache. */
  60 kmem_cache_t *obj_cache;
  61 atomic_t obj_cache_count = ATOMIC_INIT(0);
  62
  63 static void lmv_activate_target(struct lmv_obd *lmv,
  64                                 struct lmv_tgt_desc *tgt,
  65                                 int activate)
  66 {
  67         if (tgt->active == activate)
  68                 return;
  69
  70         tgt->active = activate;
  71         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
  72 }
  73
  74 /* Error codes:
  75  *
  76  *  -EINVAL  : UUID can't be found in the LMV's target list
  77  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
  78  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
  79  */
  80 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
  81                               int activate)
  82 {
  83         struct lmv_tgt_desc *tgt;
  84         struct obd_device *obd;
  85         int i, rc = 0;
  86         ENTRY;
  87
  88         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
  89                lmv, uuid->uuid, activate);
  90
  91         spin_lock(&lmv->lmv_lock);
  92         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
  93                 if (tgt->ltd_exp == NULL)
  94                         continue;
  95
  96                 CDEBUG(D_INFO, "lmv idx %d is %s conn "LPX64"\n",
  97                        i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
  98
  99                 if (strncmp(uuid->uuid, tgt->uuid.uuid, sizeof uuid->uuid) == 0)
 100                         break;
 101         }
 102
 103         if (i == lmv->desc.ld_tgt_count)
 104                 GOTO(out_lmv_lock, rc = -EINVAL);
 105
 106         obd = class_exp2obd(tgt->ltd_exp);
 107         if (obd == NULL)
 108                 GOTO(out_lmv_lock, rc = -ENOTCONN);
 109
 110         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
 111                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
 112                obd->obd_type->typ_name, i);
 113         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
 114
 115         if (tgt->active == activate) {
 116                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
 117                        activate ? "" : "in");
 118                 GOTO(out_lmv_lock, rc);
 119         }
 120
 121         CDEBUG(D_INFO, "Marking OBD %p %sactive\n",
 122                obd, activate ? "" : "in");
 123
 124         lmv_activate_target(lmv, tgt, activate);
 125
 126         EXIT;
 127
 128  out_lmv_lock:
 129         spin_unlock(&lmv->lmv_lock);
 130         return rc;
 131 }
 132
 133 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
 134                       int active, void *data)
 135 {
 136         struct obd_uuid *uuid;
 137         int rc;
 138         ENTRY;
 139
 140         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
 141                 CERROR("unexpected notification of %s %s!\n",
 142                        watched->obd_type->typ_name,
 143                        watched->obd_name);
 144                 RETURN(-EINVAL);
 145         }
 146         uuid = &watched->u.cli.cl_import->imp_target_uuid;
 147
 148         /* Set MDC as active before notifying the observer, so the observer can
 149          * use the MDC normally.
 150          */
 151         rc = lmv_set_mdc_active(&obd->u.lmv, uuid, active);
 152         if (rc) {
 153                 CERROR("%sactivation of %s failed: %d\n",
 154                        active ? "" : "de", uuid->uuid, rc);
 155                 RETURN(rc);
 156         }
 157
 158         if (obd->obd_observer)
 159                 /* Pass the notification up the chain. */
 160                 rc = obd_notify(obd->obd_observer, watched, active, data);
 161
 162         RETURN(rc);
 163 }
 164
 165 int lmv_attach(struct obd_device *dev, obd_count len, void *data)
 166 {
 167         struct lprocfs_static_vars lvars;
 168         int rc;
 169         ENTRY;
 170
 171         lprocfs_init_vars(lmv, &lvars);
 172         rc = lprocfs_obd_attach(dev, lvars.obd_vars);
 173 #ifdef __KERNEL__
 174         if (rc == 0) {
 175                 struct proc_dir_entry *entry;
 176
 177                 entry = create_proc_entry("target_obd_status", 0444,
 178                                            dev->obd_proc_entry);
 179                 if (entry == NULL)
 180                         RETURN(-ENOMEM);
 181                 entry->proc_fops = &lmv_proc_target_fops;
 182                 entry->data = dev;
 183        }
 184 #endif
 185         RETURN (rc);
 186 }
 187
 188 int lmv_detach(struct obd_device *dev)
 189 {
 190         return lprocfs_obd_detach(dev);
 191 }
 192
 193 /* this is fake connect function. Its purpose is to initialize lmv and say
 194  * caller that everything is okay. Real connection will be performed later. */
 195 static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd,
 196                        struct obd_uuid *cluuid, struct obd_connect_data *data,
 197                        unsigned long flags)
 198 {
 199 #ifdef __KERNEL__
 200         struct proc_dir_entry *lmv_proc_dir;
 201 #endif
 202         struct lmv_obd *lmv = &obd->u.lmv;
 203         struct obd_export *exp;
 204         int rc = 0;
 205         ENTRY;
 206
 207         rc = class_connect(conn, obd, cluuid);
 208         if (rc) {
 209                 CERROR("class_connection() returned %d\n", rc);
 210                 RETURN(rc);
 211         }
 212
 213         exp = class_conn2export(conn);
 214
 215         /* we don't want to actually do the underlying connections more than
 216          * once, so keep track. */
 217         lmv->refcount++;
 218         if (lmv->refcount > 1) {
 219                 class_export_put(exp);
 220                 RETURN(0);
 221         }
 222
 223         lmv->exp = exp;
 224         lmv->connected = 0;
 225         lmv->cluuid = *cluuid;
 226         lmv->connect_flags = flags;
 227         sema_init(&lmv->init_sem, 1);
 228         if (data)
 229                 memcpy(&lmv->conn_data, data, sizeof(*data));
 230
 231 #ifdef __KERNEL__
 232         lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
 233                                         NULL, NULL);
 234         if (IS_ERR(lmv_proc_dir)) {
 235                 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
 236                        obd->obd_type->typ_name, obd->obd_name);
 237                 lmv_proc_dir = NULL;
 238         }
 239 #endif
 240
 241         /*
 242          * all real clients shouls perform actual connection rightaway, because
 243          * it is possible, that LMV will not have opportunity to connect
 244          * targets, as MDC stuff will bit called directly, for instance while
 245          * reading ../mdc/../kbytesfree procfs file, etc.
 246          */
 247         if (flags & OBD_OPT_REAL_CLIENT)
 248                 rc = lmv_check_connect(obd);
 249
 250 #ifdef __KERNEL__
 251         if (rc) {
 252                 if (lmv_proc_dir)
 253                         lprocfs_remove(lmv_proc_dir);
 254         }
 255 #endif
 256
 257         RETURN(rc);
 258 }
 259
 260 void lmv_set_timeouts(struct obd_device *obd)
 261 {
 262         struct lmv_tgt_desc *tgts;
 263         struct lmv_obd *lmv;
 264         int i;
 265
 266         lmv = &obd->u.lmv;
 267         if (lmv->server_timeout == 0)
 268                 return;
 269
 270         if (lmv->connected == 0)
 271                 return;
 272
 273         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
 274                 if (tgts->ltd_exp == NULL)
 275                         continue;
 276
 277                 obd_set_info(tgts->ltd_exp, strlen("inter_mds"),
 278                              "inter_mds", 0, NULL);
 279         }
 280 }
 281
 282 #define MAX_STRING_SIZE 128
 283
 284 /* performs a check if passed obd is connected. If no - connect it. */
 285 int lmv_check_connect(struct obd_device *obd)
 286 {
 287 #ifdef __KERNEL__
 288         struct proc_dir_entry *lmv_proc_dir;
 289 #endif
 290         struct lmv_obd *lmv = &obd->u.lmv;
 291         struct lmv_tgt_desc *tgts;
 292         struct obd_uuid *cluuid;
 293         struct obd_export *exp;
 294         int rc, rc2, i;
 295         ENTRY;
 296
 297         if (lmv->connected)
 298                 RETURN(0);
 299
 300         down(&lmv->init_sem);
 301         if (lmv->connected) {
 302                 up(&lmv->init_sem);
 303                 RETURN(0);
 304         }
 305
 306         cluuid = &lmv->cluuid;
 307         exp = lmv->exp;
 308
 309         CDEBUG(D_OTHER, "time to connect %s to %s\n",
 310                cluuid->uuid, obd->obd_name);
 311
 312         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
 313                 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
 314                 struct lustre_handle conn = {0, };
 315                 struct obd_device *tgt_obd;
 316
 317                 LASSERT(tgts != NULL);
 318
 319                 tgt_obd = class_find_client_obd(&tgts->uuid, LUSTRE_MDC_NAME,
 320                                                 &obd->obd_uuid);
 321                 if (!tgt_obd) {
 322                         CERROR("target %s not attached\n", tgts->uuid.uuid);
 323                         GOTO(out_disc, rc = -EINVAL);
 324                 }
 325
 326                 /* for MDS: don't connect to yourself */
 327                 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
 328                         CDEBUG(D_OTHER, "don't connect back to %s\n",
 329                                cluuid->uuid);
 330                         tgts->ltd_exp = NULL;
 331                         continue;
 332                 }
 333
 334                 CDEBUG(D_OTHER, "connect to %s(%s) - %s, %s FOR %s\n",
 335                         tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
 336                         tgts->uuid.uuid, obd->obd_uuid.uuid,
 337                         cluuid->uuid);
 338
 339                 if (!tgt_obd->obd_set_up) {
 340                         CERROR("target %s not set up\n", tgts->uuid.uuid);
 341                         GOTO(out_disc, rc = -EINVAL);
 342                 }
 343
 344                 rc = obd_connect(&conn, tgt_obd, &lmv_mdc_uuid, &lmv->conn_data,
 345                                  lmv->connect_flags);
 346                 if (rc) {
 347                         CERROR("target %s connect error %d\n",
 348                                 tgts->uuid.uuid, rc);
 349                         GOTO(out_disc, rc);
 350                 }
 351                 tgts->ltd_exp = class_conn2export(&conn);
 352
 353                 obd_init_ea_size(tgts->ltd_exp, lmv->max_easize,
 354                                  lmv->max_cookiesize);
 355
 356                 rc = obd_register_observer(tgt_obd, obd);
 357                 if (rc) {
 358                         CERROR("target %s register_observer error %d\n",
 359                                tgts->uuid.uuid, rc);
 360                         obd_disconnect(tgts->ltd_exp, 0);
 361                         GOTO(out_disc, rc);
 362                 }
 363
 364                 lmv->desc.ld_active_tgt_count++;
 365                 tgts->active = 1;
 366
 367                 CDEBUG(D_OTHER, "connected to %s(%s) successfully (%d)\n",
 368                         tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
 369                         atomic_read(&obd->obd_refcount));
 370
 371 #ifdef __KERNEL__
 372                 lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 373                 if (lmv_proc_dir) {
 374                         struct obd_device *mdc_obd = class_conn2obd(&conn);
 375                         struct proc_dir_entry *mdc_symlink;
 376                         char name[MAX_STRING_SIZE + 1];
 377
 378                         LASSERT(mdc_obd != NULL);
 379                         LASSERT(mdc_obd->obd_type != NULL);
 380                         LASSERT(mdc_obd->obd_type->typ_name != NULL);
 381                         name[MAX_STRING_SIZE] = '\0';
 382                         snprintf(name, MAX_STRING_SIZE, "../../../%s/%s",
 383                                  mdc_obd->obd_type->typ_name,
 384                                  mdc_obd->obd_name);
 385                         mdc_symlink = proc_symlink(mdc_obd->obd_name,
 386                                                    lmv_proc_dir, name);
 387                         if (mdc_symlink == NULL) {
 388                                 CERROR("could not register LMV target "
 389                                        "/proc/fs/lustre/%s/%s/target_obds/%s.",
 390                                        obd->obd_type->typ_name, obd->obd_name,
 391                                        mdc_obd->obd_name);
 392                                 lprocfs_remove(lmv_proc_dir);
 393                                 lmv_proc_dir = NULL;
 394                         }
 395                 }
 396 #endif
 397         }
 398
 399         lmv_set_timeouts(obd);
 400         class_export_put(exp);
 401         lmv->connected = 1;
 402         up(&lmv->init_sem);
 403         RETURN(0);
 404
 405  out_disc:
 406         while (i-- > 0) {
 407                 struct obd_uuid uuid;
 408                 --tgts;
 409                 --lmv->desc.ld_active_tgt_count;
 410                 tgts->active = 0;
 411                 /* save for CERROR below; (we know it's terminated) */
 412                 uuid = tgts->uuid;
 413                 rc2 = obd_disconnect(tgts->ltd_exp, 0);
 414                 if (rc2)
 415                         CERROR("error: LMV target %s disconnect on MDC idx %d: "
 416                                "error %d\n", uuid.uuid, i, rc2);
 417         }
 418         class_disconnect(exp, 0);
 419         up(&lmv->init_sem);
 420         return rc;
 421 }
 422
 423 static int lmv_disconnect(struct obd_export *exp, unsigned long flags)
 424 {
 425         struct obd_device *obd = class_exp2obd(exp);
 426         struct lmv_obd *lmv = &obd->u.lmv;
 427
 428 #ifdef __KERNEL__
 429         struct proc_dir_entry *lmv_proc_dir;
 430 #endif
 431         int rc, i;
 432         ENTRY;
 433
 434         if (!lmv->tgts)
 435                 goto out_local;
 436
 437         /* Only disconnect the underlying layers on the final disconnect. */
 438         lmv->refcount--;
 439         if (lmv->refcount != 0)
 440                 goto out_local;
 441
 442 #ifdef __KERNEL__
 443         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
 444 #endif
 445
 446         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 447                 struct obd_device *mdc_obd;
 448
 449                 if (lmv->tgts[i].ltd_exp == NULL)
 450                         continue;
 451
 452                 mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
 453
 454                 if (mdc_obd)
 455                         mdc_obd->obd_no_recov = obd->obd_no_recov;
 456
 457 #ifdef __KERNEL__
 458                 if (lmv_proc_dir) {
 459                         struct proc_dir_entry *mdc_symlink;
 460
 461                         mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
 462                         if (mdc_symlink) {
 463                                 lprocfs_remove(mdc_symlink);
 464                         } else {
 465                                 CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
 466                                        obd->obd_type->typ_name, obd->obd_name,
 467                                        mdc_obd->obd_name);
 468                         }
 469                 }
 470 #endif
 471                 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
 472                         lmv->tgts[i].ltd_exp->exp_obd->obd_name,
 473                         lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
 474
 475                 obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL);
 476                 rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags);
 477                 if (rc) {
 478                         if (lmv->tgts[i].active) {
 479                                 CERROR("Target %s disconnect error %d\n",
 480                                        lmv->tgts[i].uuid.uuid, rc);
 481                         }
 482                         rc = 0;
 483                 }
 484
 485                 lmv_activate_target(lmv, &lmv->tgts[i], 0);
 486                 lmv->tgts[i].ltd_exp = NULL;
 487         }
 488
 489 #ifdef __KERNEL__
 490         if (lmv_proc_dir) {
 491                 lprocfs_remove(lmv_proc_dir);
 492         } else {
 493                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
 494                        obd->obd_type->typ_name, obd->obd_name);
 495         }
 496 #endif
 497
 498 out_local:
 499         /* this is the case when no real connection is established by
 500          * lmv_check_connect(). */
 501         if (!lmv->connected)
 502                 class_export_put(exp);
 503         rc = class_disconnect(exp, 0);
 504         if (lmv->refcount == 0)
 505                 lmv->connected = 0;
 506         RETURN(rc);
 507 }
 508
 509 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
 510                          int len, void *karg, void *uarg)
 511 {
 512         struct obd_device *obddev = class_exp2obd(exp);
 513         struct lmv_obd *lmv = &obddev->u.lmv;
 514         int i, rc = 0, set = 0;
 515         ENTRY;
 516
 517         if (lmv->desc.ld_tgt_count == 0)
 518                 RETURN(-ENOTTY);
 519
 520         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 521                 int err;
 522
 523                 if (lmv->tgts[i].ltd_exp == NULL)
 524                         continue;
 525
 526                 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
 527                 if (err) {
 528                         if (lmv->tgts[i].active) {
 529                                 CERROR("error: iocontrol MDC %s on MDT"
 530                                        "idx %d: err = %d\n",
 531                                        lmv->tgts[i].uuid.uuid, i, err);
 532                                 if (!rc)
 533                                         rc = err;
 534                         }
 535                 } else
 536                         set = 1;
 537         }
 538         if (!set && !rc)
 539                 rc = -EIO;
 540
 541         RETURN(rc);
 542 }
 543
 544 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
 545 {
 546         int i, rc = 0;
 547         struct lmv_desc *desc;
 548         struct obd_uuid *uuids;
 549         struct lmv_tgt_desc *tgts;
 550         struct obd_device *tgt_obd;
 551         struct lustre_cfg *lcfg = buf;
 552         struct lmv_obd *lmv = &obd->u.lmv;
 553         ENTRY;
 554
 555         if (lcfg->lcfg_inllen1 < 1) {
 556                 CERROR("LMV setup requires a descriptor\n");
 557                 RETURN(-EINVAL);
 558         }
 559
 560         if (lcfg->lcfg_inllen2 < 1) {
 561                 CERROR("LMV setup requires an MDT UUID list\n");
 562                 RETURN(-EINVAL);
 563         }
 564
 565         desc = (struct lmv_desc *)lcfg->lcfg_inlbuf1;
 566         if (sizeof(*desc) > lcfg->lcfg_inllen1) {
 567                 CERROR("descriptor size wrong: %d > %d\n",
 568                        (int)sizeof(*desc), lcfg->lcfg_inllen1);
 569                 RETURN(-EINVAL);
 570         }
 571
 572         uuids = (struct obd_uuid *)lcfg->lcfg_inlbuf2;
 573         if (sizeof(*uuids) * desc->ld_tgt_count != lcfg->lcfg_inllen2) {
 574                 CERROR("UUID array size wrong: %u * %u != %u\n",
 575                        sizeof(*uuids), desc->ld_tgt_count, lcfg->lcfg_inllen2);
 576                 RETURN(-EINVAL);
 577         }
 578
 579         lmv->tgts_size = sizeof(struct lmv_tgt_desc) * desc->ld_tgt_count;
 580         OBD_ALLOC(lmv->tgts, lmv->tgts_size);
 581         if (lmv->tgts == NULL) {
 582                 CERROR("Out of memory\n");
 583                 RETURN(-ENOMEM);
 584         }
 585
 586         lmv->desc = *desc;
 587         spin_lock_init(&lmv->lmv_lock);
 588
 589         for (i = 0, tgts = lmv->tgts; i < desc->ld_tgt_count; i++, tgts++)
 590                 tgts->uuid = uuids[i];
 591
 592         lmv->max_cookiesize = 0;
 593
 594         lmv->max_easize = sizeof(struct lustre_id) *
 595                 desc->ld_tgt_count + sizeof(struct mea);
 596
 597         rc = lmv_setup_mgr(obd);
 598         if (rc) {
 599                 CERROR("Can't setup LMV object manager, "
 600                        "error %d.\n", rc);
 601                 OBD_FREE(lmv->tgts, lmv->tgts_size);
 602         }
 603
 604         tgt_obd = class_find_client_obd(&lmv->tgts->uuid, LUSTRE_MDC_NAME,
 605                                         &obd->obd_uuid);
 606         if (!tgt_obd) {
 607                 CERROR("Target %s not attached\n", lmv->tgts->uuid.uuid);
 608                 RETURN(-EINVAL);
 609         }
 610
 611         RETURN(rc);
 612 }
 613
 614 static int lmv_cleanup(struct obd_device *obd, int flags)
 615 {
 616         struct lmv_obd *lmv = &obd->u.lmv;
 617         ENTRY;
 618
 619         lmv_cleanup_mgr(obd);
 620         OBD_FREE(lmv->tgts, lmv->tgts_size);
 621
 622         RETURN(0);
 623 }
 624
 625 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
 626                       unsigned long max_age)
 627 {
 628         struct lmv_obd *lmv = &obd->u.lmv;
 629         struct obd_statfs *temp;
 630         int rc = 0, i;
 631         ENTRY;
 632
 633         rc = lmv_check_connect(obd);
 634         if (rc)
 635                 RETURN(rc);
 636
 637         OBD_ALLOC(temp, sizeof(*temp));
 638         if (temp == NULL)
 639                 RETURN(-ENOMEM);
 640
 641         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
 642                 if (lmv->tgts[i].ltd_exp == NULL) {
 643                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
 644                         continue;
 645                 }
 646
 647                 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp, max_age);
 648                 if (rc) {
 649                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
 650                                lmv->tgts[i].ltd_exp->exp_obd->obd_name,
 651                                rc);
 652                         GOTO(out_free_temp, rc);
 653                 }
 654                 if (i == 0) {
 655                         memcpy(osfs, temp, sizeof(*temp));
 656                 } else {
 657                         osfs->os_bavail += temp->os_bavail;
 658                         osfs->os_blocks += temp->os_blocks;
 659                         osfs->os_ffree += temp->os_ffree;
 660                         osfs->os_files += temp->os_files;
 661                 }
 662         }
 663
 664         EXIT;
 665 out_free_temp:
 666         OBD_FREE(temp, sizeof(*temp));
 667         return rc;
 668 }
 669
 670 static int lmv_getstatus(struct obd_export *exp, struct lustre_id *id)
 671 {
 672         struct obd_device *obd = exp->exp_obd;
 673         struct lmv_obd *lmv = &obd->u.lmv;
 674         int rc;
 675         ENTRY;
 676
 677         rc = lmv_check_connect(obd);
 678         if (rc)
 679                 RETURN(rc);
 680
 681         rc = md_getstatus(lmv->tgts[0].ltd_exp, id);
 682         id_group(id) = 0;
 683
 684         RETURN(rc);
 685 }
 686
 687 static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
 688                        __u64 valid, const char *ea_name, int ea_namelen,
 689                        unsigned int ea_size, struct ptlrpc_request **request)
 690 {
 691         struct obd_device *obd = exp->exp_obd;
 692         struct lmv_obd *lmv = &obd->u.lmv;
 693         int rc, i = id_group(id);
 694         struct lmv_obj *obj;
 695         ENTRY;
 696
 697         rc = lmv_check_connect(obd);
 698         if (rc)
 699                 RETURN(rc);
 700
 701         LASSERT(i < lmv->desc.ld_tgt_count);
 702
 703
 704         rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid,
 705                         ea_name, ea_namelen, ea_size, request);
 706         if (rc)
 707                 RETURN(rc);
 708
 709         obj = lmv_grab_obj(obd, id);
 710
 711         CDEBUG(D_OTHER, "GETATTR for "DLID4" %s\n",
 712                OLID4(id), obj ? "(splitted)" : "");
 713
 714         /*
 715          * if object is splitted, then we loop over all the slaves and gather
 716          * size attribute. In ideal world we would have to gather also mds field
 717          * from all slaves, as object is spread over the cluster and this is
 718          * definitely interesting information and it is not good to loss it,
 719          * but...
 720          */
 721         if (obj) {
 722                 struct mds_body *body;
 723
 724                 if (*request == NULL) {
 725                         lmv_put_obj(obj);
 726                         RETURN(rc);
 727                 }
 728
 729                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
 730                                       sizeof(*body));
 731                 LASSERT(body != NULL);
 732
 733                 lmv_lock_obj(obj);
 734
 735                 for (i = 0; i < obj->objcount; i++) {
 736
 737                         if (lmv->tgts[i].ltd_exp == NULL) {
 738                                 CWARN("%s: NULL export for %d\n",
 739                                       obd->obd_name, i);
 740                                 continue;
 741                         }
 742
 743                         /* skip master obj. */
 744                         if (id_equal_fid(&obj->id, &obj->objs[i].id))
 745                                 continue;
 746
 747                         body->size += obj->objs[i].size;
 748                 }
 749
 750                 lmv_unlock_obj(obj);
 751                 lmv_put_obj(obj);
 752         }
 753
 754         RETURN(rc);
 755 }
 756
 757 static int lmv_change_cbdata(struct obd_export *exp,
 758                              struct lustre_id *id,
 759                              ldlm_iterator_t it,
 760                              void *data)
 761 {
 762         struct obd_device *obd = exp->exp_obd;
 763         struct lmv_obd *lmv = &obd->u.lmv;
 764         int rc = 0;
 765         ENTRY;
 766
 767         rc = lmv_check_connect(obd);
 768         if (rc)
 769                 RETURN(rc);
 770
 771         CDEBUG(D_OTHER, "CBDATA for "DLID4"\n", OLID4(id));
 772         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
 773
 774         rc = md_change_cbdata(lmv->tgts[id_group(id)].ltd_exp,
 775                               id, it, data);
 776
 777         RETURN(rc);
 778 }
 779
 780 static int lmv_change_cbdata_name(struct obd_export *exp,
 781                                   struct lustre_id *pid,
 782                                   char *name, int len,
 783                                   struct lustre_id *cid,
 784                                   ldlm_iterator_t it,
 785                                   void *data)
 786 {
 787         struct obd_device *obd = exp->exp_obd;
 788         struct lmv_obd *lmv = &obd->u.lmv;
 789         struct lustre_id rcid = *cid;
 790         struct lmv_obj *obj;
 791         int rc = 0, mds;
 792         ENTRY;
 793
 794         rc = lmv_check_connect(obd);
 795         if (rc)
 796                 RETURN(rc);
 797
 798         LASSERT(id_group(pid) < lmv->desc.ld_tgt_count);
 799         LASSERT(id_group(cid) < lmv->desc.ld_tgt_count);
 800
 801         CDEBUG(D_OTHER, "CBDATA for "DLID4":%*s -> "DLID4"\n",
 802                OLID4(pid), len, name, OLID4(cid));
 803
 804         /* this is default mds for directory name belongs to. */
 805         mds = id_group(pid);
 806         obj = lmv_grab_obj(obd, pid);
 807         if (obj) {
 808                 /* directory is splitted. look for right mds for this name. */
 809                 mds = raw_name2idx(obj->hashtype, obj->objcount, name, len);
 810                 rcid = obj->objs[mds].id;
 811                 mds = id_group(&rcid);
 812                 lmv_put_obj(obj);
 813         }
 814         rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, &rcid, it, data);
 815         RETURN(rc);
 816 }
 817
 818 static int lmv_valid_attrs(struct obd_export *exp, struct lustre_id *id)
 819 {
 820         struct obd_device *obd = exp->exp_obd;
 821         struct lmv_obd *lmv = &obd->u.lmv;
 822         int rc = 0;
 823         ENTRY;
 824
 825         rc = lmv_check_connect(obd);
 826         if (rc)
 827                 RETURN(rc);
 828
 829         CDEBUG(D_OTHER, "validate "DLID4"\n", OLID4(id));
 830         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
 831         rc = md_valid_attrs(lmv->tgts[id_group(id)].ltd_exp, id);
 832         RETURN(rc);
 833 }
 834
 835 int lmv_close(struct obd_export *exp, struct obdo *obdo,
 836               struct obd_client_handle *och,
 837               struct ptlrpc_request **request)
 838 {
 839         struct obd_device *obd = exp->exp_obd;
 840         struct lmv_obd *lmv = &obd->u.lmv;
 841         int rc, i = obdo->o_mds;
 842         ENTRY;
 843
 844         rc = lmv_check_connect(obd);
 845         if (rc)
 846                 RETURN(rc);
 847
 848         LASSERT(i < lmv->desc.ld_tgt_count);
 849         CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long)obdo->o_mds,
 850                (unsigned long)obdo->o_id, (unsigned long)obdo->o_generation);
 851         rc = md_close(lmv->tgts[i].ltd_exp, obdo, och, request);
 852         RETURN(rc);
 853 }
 854
 855 int lmv_get_mea_and_update_object(struct obd_export *exp,
 856                                   struct lustre_id *id)
 857 {
 858         struct obd_device *obd = exp->exp_obd;
 859         struct lmv_obd *lmv = &obd->u.lmv;
 860         struct ptlrpc_request *req = NULL;
 861         struct lmv_obj *obj;
 862         struct lustre_md md;
 863         int mealen, rc;
 864         __u64 valid;
 865         ENTRY;
 866
 867         md.mea = NULL;
 868         mealen = MEA_SIZE_LMV(lmv);
 869
 870         valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA;
 871
 872         /* time to update mea of parent id */
 873         rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp,
 874                         id, valid, NULL, 0, mealen, &req);
 875         if (rc) {
 876                 CERROR("md_getattr() failed, error %d\n", rc);
 877                 GOTO(cleanup, rc);
 878         }
 879
 880         rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
 881         if (rc) {
 882                 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
 883                 GOTO(cleanup, rc);
 884         }
 885
 886         if (md.mea == NULL)
 887                 GOTO(cleanup, rc = -ENODATA);
 888
 889         obj = lmv_create_obj(exp, id, md.mea);
 890         if (IS_ERR(obj))
 891                 rc = PTR_ERR(obj);
 892
 893         lmv_put_obj(obj);
 894         obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
 895
 896         EXIT;
 897 cleanup:
 898         if (req)
 899                 ptlrpc_req_finished(req);
 900         return rc;
 901 }
 902
 903 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
 904                const void *data, int datalen, int mode, __u32 uid,
 905                __u32 gid, __u64 rdev, struct ptlrpc_request **request)
 906 {
 907         struct obd_device *obd = exp->exp_obd;
 908         struct lmv_obd *lmv = &obd->u.lmv;
 909         struct mds_body *body;
 910         struct lmv_obj *obj;
 911         int rc, mds, loop = 0;
 912         ENTRY;
 913
 914         rc = lmv_check_connect(obd);
 915         if (rc)
 916                 RETURN(rc);
 917
 918         if (!lmv->desc.ld_active_tgt_count)
 919                 RETURN(-EIO);
 920 repeat:
 921         LASSERT(++loop <= 2);
 922         obj = lmv_grab_obj(obd, &op_data->id1);
 923         if (obj) {
 924                 mds = raw_name2idx(obj->hashtype, obj->objcount,
 925                                    op_data->name, op_data->namelen);
 926                 op_data->id1 = obj->objs[mds].id;
 927                 lmv_put_obj(obj);
 928         }
 929
 930         CDEBUG(D_OTHER, "CREATE '%*s' on "DLID4"\n", op_data->namelen,
 931                op_data->name, OLID4(&op_data->id1));
 932
 933         rc = md_create(lmv->tgts[id_group(&op_data->id1)].ltd_exp,
 934                        op_data, data, datalen, mode, uid, gid, rdev,
 935                        request);
 936         if (rc == 0) {
 937                 if (*request == NULL)
 938                         RETURN(rc);
 939
 940                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
 941                                       sizeof(*body));
 942                 if (body == NULL)
 943                         RETURN(-ENOMEM);
 944
 945                 CDEBUG(D_OTHER, "created. "DLID4"\n", OLID4(&op_data->id1));
 946         } else if (rc == -ERESTART) {
 947                 /*
 948                  * directory got splitted. time to update local object and
 949                  * repeat the request with proper MDS.
 950                  */
 951                 rc = lmv_get_mea_and_update_object(exp, &op_data->id1);
 952                 if (rc == 0) {
 953                         ptlrpc_req_finished(*request);
 954                         goto repeat;
 955                 }
 956         }
 957         RETURN(rc);
 958 }
 959
 960 int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
 961 {
 962         struct obd_device *obd = exp->exp_obd;
 963         struct lmv_obd *lmv = &obd->u.lmv;
 964         int rc;
 965         ENTRY;
 966
 967         rc = lmv_check_connect(obd);
 968         if (rc)
 969                 RETURN(rc);
 970
 971         /* FIXME: choose right MDC here */
 972         CWARN("this method isn't implemented yet\n");
 973         rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
 974         RETURN(rc);
 975 }
 976
 977 int lmv_enqueue_slaves(struct obd_export *exp, int locktype,
 978                        struct lookup_intent *it, int lockmode,
 979                        struct mdc_op_data *data, struct lustre_handle *lockh,
 980                        void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
 981                        ldlm_blocking_callback cb_blocking, void *cb_data)
 982 {
 983         struct obd_device *obd = exp->exp_obd;
 984         struct lmv_obd *lmv = &obd->u.lmv;
 985         struct mea *mea = data->mea1;
 986         struct mdc_op_data *data2;
 987         int i, rc, mds;
 988         ENTRY;
 989
 990         OBD_ALLOC(data2, sizeof(*data2));
 991         if (data2 == NULL)
 992                 RETURN(-ENOMEM);
 993
 994         LASSERT(mea != NULL);
 995         for (i = 0; i < mea->mea_count; i++) {
 996                 memset(data2, 0, sizeof(*data2));
 997                 data2->id1 = mea->mea_ids[i];
 998                 mds = id_group(&data2->id1);
 999
1000                 if (lmv->tgts[mds].ltd_exp == NULL)
1001                         continue;
1002
1003                 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it,
1004                                 lockmode, data2, lockh + i, lmm, lmmsize,
1005                                 cb_compl, cb_blocking, cb_data);
1006
1007                 CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n",
1008                        OLID4(&mea->mea_ids[i]), rc, LUSTRE_IT(it)->it_status);
1009                 if (rc)
1010                         GOTO(cleanup, rc);
1011                 if (LUSTRE_IT(it)->it_data) {
1012                         struct ptlrpc_request *req;
1013                         req = (struct ptlrpc_request *) LUSTRE_IT(it)->it_data;
1014                         ptlrpc_req_finished(req);
1015                 }
1016
1017                 if (LUSTRE_IT(it)->it_status)
1018                         GOTO(cleanup, rc = LUSTRE_IT(it)->it_status);
1019         }
1020
1021         OBD_FREE(data2, sizeof(*data2));
1022         RETURN(0);
1023 cleanup:
1024         OBD_FREE(data2, sizeof(*data2));
1025
1026         /* drop all taken locks */
1027         while (--i >= 0) {
1028                 if (lockh[i].cookie)
1029                         ldlm_lock_decref(lockh + i, lockmode);
1030                 lockh[i].cookie = 0;
1031         }
1032         return rc;
1033 }
1034
1035 int lmv_enqueue(struct obd_export *exp, int lock_type,
1036                 struct lookup_intent *it, int lock_mode,
1037                 struct mdc_op_data *data, struct lustre_handle *lockh,
1038                 void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
1039                 ldlm_blocking_callback cb_blocking, void *cb_data)
1040 {
1041         struct obd_device *obd = exp->exp_obd;
1042         struct lmv_obd *lmv = &obd->u.lmv;
1043         struct lmv_obj *obj;
1044         int rc, mds;
1045         ENTRY;
1046
1047         rc = lmv_check_connect(obd);
1048         if (rc)
1049                 RETURN(rc);
1050
1051         if (data->mea1 && it->it_op == IT_UNLINK) {
1052                 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
1053                                         data, lockh, lmm, lmmsize,
1054                                         cb_compl, cb_blocking, cb_data);
1055                 RETURN(rc);
1056         }
1057
1058         if (data->namelen) {
1059                 obj = lmv_grab_obj(obd, &data->id1);
1060                 if (obj) {
1061                         /* directory is splitted. look for right mds for this
1062                          * name */
1063                         mds = raw_name2idx(obj->hashtype, obj->objcount,
1064                                            (char *)data->name, data->namelen);
1065                         data->id1 = obj->objs[mds].id;
1066                         lmv_put_obj(obj);
1067                 }
1068         }
1069         CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4"\n", LL_IT2STR(it),
1070                OLID4(&data->id1));
1071
1072         rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp,
1073                         lock_type, it, lock_mode, data, lockh, lmm,
1074                         lmmsize, cb_compl, cb_blocking, cb_data);
1075         RETURN(rc);
1076 }
1077
1078 int lmv_getattr_lock(struct obd_export *exp, struct lustre_id *id,
1079                      char *filename, int namelen, __u64 valid,
1080                      unsigned int ea_size, struct ptlrpc_request **request)
1081 {
1082         int rc, mds = id_group(id), loop = 0;
1083         struct obd_device *obd = exp->exp_obd;
1084         struct lmv_obd *lmv = &obd->u.lmv;
1085         struct lustre_id rid = *id;
1086         struct mds_body *body;
1087         struct lmv_obj *obj;
1088         ENTRY;
1089
1090         rc = lmv_check_connect(obd);
1091         if (rc)
1092                 RETURN(rc);
1093 repeat:
1094         LASSERT(++loop <= 2);
1095         obj = lmv_grab_obj(obd, id);
1096         if (obj) {
1097                 /* directory is splitted. look for right mds for this name */
1098                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1099                                    filename, namelen - 1);
1100                 rid = obj->objs[mds].id;
1101                 lmv_put_obj(obj);
1102         }
1103
1104         CDEBUG(D_OTHER, "getattr_lock for %*s on "DLID4" -> "DLID4"\n",
1105                namelen, filename, OLID4(id), OLID4(&rid));
1106
1107         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp, &rid,
1108                              filename, namelen, (valid | OBD_MD_FID),
1109                              ea_size, request);
1110         if (rc == 0) {
1111                 /*
1112                  * this could be cross-node reference. in this case all we have
1113                  * right now is lustre_id triple. we'd like to find other
1114                  * attributes.
1115                  */
1116                 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
1117                 LASSERT(body != NULL);
1118                 LASSERT((body->valid & OBD_MD_FID) != 0);
1119
1120                 if (body->valid & OBD_MD_MDS) {
1121                         struct ptlrpc_request *req = NULL;
1122
1123                         rid = body->id1;
1124                         CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid));
1125
1126                         rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
1127                                              &rid, NULL, 1, valid, ea_size, &req);
1128                         ptlrpc_req_finished(*request);
1129                         *request = req;
1130                 }
1131         } else if (rc == -ERESTART) {
1132                 /* directory got splitted. time to update local object and
1133                  * repeat the request with proper MDS */
1134                 rc = lmv_get_mea_and_update_object(exp, &rid);
1135                 if (rc == 0) {
1136                         ptlrpc_req_finished(*request);
1137                         goto repeat;
1138                 }
1139         }
1140         RETURN(rc);
1141 }
1142
1143 /*
1144  * llite passes id of an target inode in data->id1 and id of directory in
1145  * data->id2
1146  */
1147 int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
1148              struct ptlrpc_request **request)
1149 {
1150         struct obd_device *obd = exp->exp_obd;
1151         struct lmv_obd *lmv = &obd->u.lmv;
1152         struct lmv_obj *obj;
1153         int rc;
1154         ENTRY;
1155
1156         rc = lmv_check_connect(obd);
1157         if (rc)
1158                 RETURN(rc);
1159
1160         if (data->namelen != 0) {
1161                 /* usual link request */
1162                 obj = lmv_grab_obj(obd, &data->id1);
1163                 if (obj) {
1164                         rc = raw_name2idx(obj->hashtype, obj->objcount,
1165                                           data->name, data->namelen);
1166                         data->id1 = obj->objs[rc].id;
1167                         lmv_put_obj(obj);
1168                 }
1169
1170                 CDEBUG(D_OTHER,"link "DLID4":%*s to "DLID4"\n",
1171                        OLID4(&data->id2), data->namelen, data->name,
1172                        OLID4(&data->id1));
1173         } else {
1174                 /* request from MDS to acquire i_links for inode by id1 */
1175                 CDEBUG(D_OTHER, "inc i_nlinks for "DLID4"\n",
1176                        OLID4(&data->id1));
1177         }
1178
1179         rc = md_link(lmv->tgts[id_group(&data->id1)].ltd_exp,
1180                      data, request);
1181         RETURN(rc);
1182 }
1183
1184 int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1185                const char *old, int oldlen, const char *new, int newlen,
1186                struct ptlrpc_request **request)
1187 {
1188         struct obd_device *obd = exp->exp_obd;
1189         struct lmv_obd *lmv = &obd->u.lmv;
1190         struct lmv_obj *obj;
1191         int rc, mds;
1192         ENTRY;
1193
1194         CDEBUG(D_OTHER, "rename %*s in "DLID4" to %*s in "DLID4"\n",
1195                oldlen, old, OLID4(&data->id1), newlen, new,
1196                OLID4(&data->id2));
1197
1198         rc = lmv_check_connect(obd);
1199         if (rc)
1200                 RETURN(rc);
1201
1202         if (oldlen == 0) {
1203                 /*
1204                  * MDS with old dir entry is asking another MDS to create name
1205                  * there.
1206                  */
1207                 CDEBUG(D_OTHER,
1208                        "create %*s(%d/%d) in "DLID4" pointing "
1209                        "to "DLID4"\n", newlen, new, oldlen, newlen,
1210                        OLID4(&data->id2), OLID4(&data->id1));
1211
1212                 mds = id_group(&data->id2);
1213
1214                 /*
1215                  * target directory can be splitted, sowe should forward request
1216                  * to the right MDS.
1217                  */
1218                 obj = lmv_grab_obj(obd, &data->id2);
1219                 if (obj) {
1220                         mds = raw_name2idx(obj->hashtype, obj->objcount,
1221                                            (char *)new, newlen);
1222                         data->id2 = obj->objs[mds].id;
1223                         CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1224                                OLID4(&data->id2));
1225                         lmv_put_obj(obj);
1226                 }
1227                 goto request;
1228         }
1229
1230         obj = lmv_grab_obj(obd, &data->id1);
1231         if (obj) {
1232                 /*
1233                  * directory is already splitted, so we have to forward request
1234                  * to the right MDS.
1235                  */
1236                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1237                                    (char *)old, oldlen);
1238                 data->id1 = obj->objs[mds].id;
1239                 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1240                        OLID4(&data->id1));
1241                 lmv_put_obj(obj);
1242         }
1243
1244         obj = lmv_grab_obj(obd, &data->id2);
1245         if (obj) {
1246                 /*
1247                  * directory is already splitted, so we have to forward request
1248                  * to the right MDS.
1249                  */
1250                 mds = raw_name2idx(obj->hashtype, obj->objcount,
1251                                    (char *)new, newlen);
1252
1253                 data->id2 = obj->objs[mds].id;
1254                 CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds,
1255                        OLID4(&data->id2));
1256                 lmv_put_obj(obj);
1257         }
1258
1259         mds = id_group(&data->id1);
1260
1261 request:
1262         if (id_group(&data->id1) != id_group(&data->id2)) {
1263                 CDEBUG(D_OTHER,"cross-node rename "DLID4"/%*s to "DLID4"/%*s\n",
1264                        OLID4(&data->id1), oldlen, old, OLID4(&data->id2),
1265                        newlen, new);
1266         }
1267
1268         rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1269                        new, newlen, request);
1270         RETURN(rc);
1271 }
1272
1273 int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1274                 struct iattr *iattr, void *ea, int ealen, void *ea2,
1275                 int ea2len, struct ptlrpc_request **request)
1276 {
1277         struct obd_device *obd = exp->exp_obd;
1278         struct lmv_obd *lmv = &obd->u.lmv;
1279         struct ptlrpc_request *req;
1280         struct mds_body *body;
1281         struct lmv_obj *obj;
1282         int rc = 0, i;
1283         ENTRY;
1284
1285         rc = lmv_check_connect(obd);
1286         if (rc)
1287                 RETURN(rc);
1288
1289         obj = lmv_grab_obj(obd, &data->id1);
1290
1291         CDEBUG(D_OTHER, "SETATTR for "DLID4", valid 0x%x%s\n",
1292                OLID4(&data->id1), iattr->ia_valid, obj ? ", splitted" : "");
1293
1294         if (obj) {
1295                 for (i = 0; i < obj->objcount; i++) {
1296                         data->id1 = obj->objs[i].id;
1297
1298                         rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1299                                         data, iattr, ea, ealen, ea2, ea2len, &req);
1300
1301                         if (id_equal_fid(&obj->id, &obj->objs[i].id)) {
1302                                 /*
1303                                  * this is master object and this request should
1304                                  * be returned back to llite.
1305                                  */
1306                                 *request = req;
1307                         } else {
1308                                 ptlrpc_req_finished(req);
1309                         }
1310
1311                         if (rc)
1312                                 break;
1313                 }
1314                 lmv_put_obj(obj);
1315         } else {
1316                 LASSERT(id_group(&data->id1) < lmv->desc.ld_tgt_count);
1317                 rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp,
1318                                 data, iattr, ea, ealen, ea2, ea2len, request);
1319                 if (rc == 0) {
1320                         body = lustre_msg_buf((*request)->rq_repmsg, 0,
1321                                               sizeof(*body));
1322                         LASSERT(body != NULL);
1323                         LASSERT((body->valid & OBD_MD_FID) != 0);
1324                         LASSERT(id_group(&body->id1) == id_group(&data->id1));
1325                 }
1326         }
1327         RETURN(rc);
1328 }
1329
1330 int lmv_sync(struct obd_export *exp, struct lustre_id *id,
1331              struct ptlrpc_request **request)
1332 {
1333         struct obd_device *obd = exp->exp_obd;
1334         struct lmv_obd *lmv = &obd->u.lmv;
1335         int rc;
1336         ENTRY;
1337
1338         rc = lmv_check_connect(obd);
1339         if (rc)
1340                 RETURN(rc);
1341
1342         rc = md_sync(lmv->tgts[id_group(id)].ltd_exp,
1343                      id, request);
1344         RETURN(rc);
1345 }
1346
1347 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock,
1348                             struct ldlm_lock_desc *desc,
1349                             void *data, int flag)
1350 {
1351         struct lustre_handle lockh;
1352         struct lmv_obj *obj;
1353         int rc;
1354         ENTRY;
1355
1356         switch (flag) {
1357         case LDLM_CB_BLOCKING:
1358                 ldlm_lock2handle(lock, &lockh);
1359                 rc = ldlm_cli_cancel(&lockh);
1360                 if (rc < 0) {
1361                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1362                         RETURN(rc);
1363                 }
1364                 break;
1365         case LDLM_CB_CANCELING:
1366                 /* time to drop cached attrs for dirobj */
1367                 obj = lock->l_ast_data;
1368                 if (obj) {
1369                         CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64
1370                                ", master "DLID4"\n",
1371                                lock->l_resource->lr_name.name[3] == 1 ?
1372                                "LOOKUP" : "UPDATE",
1373                                lock->l_resource->lr_name.name[0],
1374                                lock->l_resource->lr_name.name[1],
1375                                OLID4(&obj->id));
1376                         lmv_put_obj(obj);
1377                 }
1378                 break;
1379         default:
1380                 LBUG();
1381         }
1382         RETURN(0);
1383 }
1384
1385 void lmv_remove_dots(struct page *page)
1386 {
1387         unsigned limit = PAGE_CACHE_SIZE;
1388         char *kaddr = page_address(page);
1389         struct ext2_dir_entry_2 *p;
1390         unsigned offs, rec_len;
1391
1392         for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1393                 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1394                 rec_len = le16_to_cpu(p->rec_len);
1395
1396                 if ((p->name_len == 1 && p->name[0] == '.') ||
1397                     (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1398                         p->inode = 0;
1399         }
1400 }
1401
1402 int lmv_readpage(struct obd_export *exp, struct lustre_id *id,
1403                  __u64 offset, struct page *page,
1404                  struct ptlrpc_request **request)
1405 {
1406         struct obd_device *obd = exp->exp_obd;
1407         struct lmv_obd *lmv = &obd->u.lmv;
1408         struct lustre_id rid = *id;
1409         struct lmv_obj *obj;
1410         int rc, i;
1411         ENTRY;
1412
1413 #warning "we need well-desgined readdir() implementation"
1414         rc = lmv_check_connect(obd);
1415         if (rc)
1416                 RETURN(rc);
1417
1418         LASSERT(id_group(id) < lmv->desc.ld_tgt_count);
1419         CDEBUG(D_OTHER, "READPAGE at %llu from "DLID4"\n",
1420                offset, OLID4(&rid));
1421
1422         obj = lmv_grab_obj(obd, id);
1423         if (obj) {
1424                 lmv_lock_obj(obj);
1425
1426                 /* find dirobj containing page with requested offset. */
1427                 for (i = 0; i < obj->objcount; i++) {
1428                         if (offset < obj->objs[i].size)
1429                                 break;
1430                         offset -= obj->objs[i].size;
1431                 }
1432                 rid = obj->objs[i].id;
1433
1434                 lmv_unlock_obj(obj);
1435                 lmv_put_obj(obj);
1436
1437                 CDEBUG(D_OTHER, "forward to "DLID4" with offset %lu\n",
1438                        OLID4(&rid), (unsigned long)offset);
1439         }
1440         rc = md_readpage(lmv->tgts[id_group(&rid)].ltd_exp, &rid,
1441                          offset, page, request);
1442
1443         if (rc == 0 && !id_equal_fid(&rid, id))
1444                 /* this page isn't from master object. To avoid "." and ".."
1445                  * duplication in directory, we have to remove them from all
1446                  * slave objects */
1447                 lmv_remove_dots(page);
1448
1449         RETURN(rc);
1450 }
1451
1452 int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1453                       struct ptlrpc_request **req)
1454 {
1455         struct obd_device *obd = exp->exp_obd;
1456         struct lmv_obd *lmv = &obd->u.lmv;
1457         struct mea *mea = data->mea1;
1458         struct mdc_op_data *data2;
1459         int i, rc = 0;
1460         ENTRY;
1461
1462         OBD_ALLOC(data2, sizeof(*data2));
1463         if (data2 == NULL)
1464                 RETURN(-ENOMEM);
1465
1466         LASSERT(mea != NULL);
1467         for (i = 0; i < mea->mea_count; i++) {
1468                 memset(data2, 0, sizeof(*data2));
1469                 data2->id1 = mea->mea_ids[i];
1470                 data2->create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1471
1472                 if (lmv->tgts[id_group(&data2->id1)].ltd_exp == NULL)
1473                         continue;
1474
1475                 rc = md_unlink(lmv->tgts[id_group(&data2->id1)].ltd_exp,
1476                                data2, req);
1477
1478                 CDEBUG(D_OTHER, "unlink slave "DLID4" -> %d\n",
1479                        OLID4(&mea->mea_ids[i]), rc);
1480
1481                 if (*req) {
1482                         ptlrpc_req_finished(*req);
1483                         *req = NULL;
1484                 }
1485                 if (rc)
1486                         RETURN(rc);
1487         }
1488         OBD_FREE(data2, sizeof(*data2));
1489         RETURN(rc);
1490 }
1491
1492 int lmv_delete_inode(struct obd_export *exp, struct lustre_id *id)
1493 {
1494         ENTRY;
1495
1496         LASSERT(exp && id);
1497         if (lmv_delete_obj(exp, id)) {
1498                 CDEBUG(D_OTHER, "lmv object "DLID4" is destroyed.\n",
1499                        OLID4(id));
1500         }
1501         RETURN(0);
1502 }
1503
1504 int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1505                struct ptlrpc_request **request)
1506 {
1507         struct obd_device *obd = exp->exp_obd;
1508         struct lmv_obd *lmv = &obd->u.lmv;
1509         int rc, i = 0;
1510         ENTRY;
1511
1512         rc = lmv_check_connect(obd);
1513         if (rc)
1514                 RETURN(rc);
1515
1516         if (data->namelen == 0 && data->mea1 != NULL) {
1517                 /* mds asks to remove slave objects */
1518                 rc = lmv_unlink_slaves(exp, data, request);
1519                 RETURN(rc);
1520         }
1521
1522         if (data->namelen != 0) {
1523                 struct lmv_obj *obj;
1524
1525                 obj = lmv_grab_obj(obd, &data->id1);
1526                 if (obj) {
1527                         i = raw_name2idx(obj->hashtype, obj->objcount,
1528                                          data->name, data->namelen);
1529                         data->id1 = obj->objs[i].id;
1530                         lmv_put_obj(obj);
1531                 }
1532                 CDEBUG(D_OTHER, "unlink '%*s' in "DLID4" -> %u\n",
1533                        data->namelen, data->name, OLID4(&data->id1),
1534                        i);
1535         } else {
1536                 CDEBUG(D_OTHER, "drop i_nlink on "DLID4"\n",
1537                        OLID4(&data->id1));
1538         }
1539         rc = md_unlink(lmv->tgts[id_group(&data->id1)].ltd_exp,
1540                        data, request);
1541         RETURN(rc);
1542 }
1543
1544 struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1545                                     struct lustre_id *id)
1546 {
1547         struct obd_device *obd = exp->exp_obd;
1548         struct lmv_obd *lmv = &obd->u.lmv;
1549         int rc;
1550         ENTRY;
1551
1552         rc = lmv_check_connect(obd);
1553         if (rc)
1554                 RETURN(ERR_PTR(rc));
1555         obd = lmv->tgts[id_group(id)].ltd_exp->exp_obd;
1556         EXIT;
1557
1558         return obd;
1559 }
1560
1561 int lmv_init_ea_size(struct obd_export *exp, int easize,
1562                      int cookiesize)
1563 {
1564         struct obd_device *obd = exp->exp_obd;
1565         struct lmv_obd *lmv = &obd->u.lmv;
1566         int i, rc = 0, change = 0;
1567         ENTRY;
1568
1569         if (lmv->max_easize < easize) {
1570                 lmv->max_easize = easize;
1571                 change = 1;
1572         }
1573         if (lmv->max_cookiesize < cookiesize) {
1574                 lmv->max_cookiesize = cookiesize;
1575                 change = 1;
1576         }
1577         if (change == 0)
1578                 RETURN(0);
1579
1580         if (lmv->connected == 0)
1581                 RETURN(0);
1582
1583         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1584                 if (lmv->tgts[i].ltd_exp == NULL) {
1585                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
1586                         continue;
1587                 }
1588
1589                 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
1590                 if (rc) {
1591                         CERROR("obd_init_ea_size() failed on MDT target %d, "
1592                                "error %d.\n", i, rc);
1593                         break;
1594                 }
1595         }
1596         RETURN(rc);
1597 }
1598
1599 int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1600                           struct lov_stripe_md **ea, struct obd_trans_info *oti)
1601 {
1602         struct obd_device *obd = exp->exp_obd;
1603         struct lmv_obd *lmv = &obd->u.lmv;
1604         struct lov_stripe_md obj_md;
1605         struct lov_stripe_md *obj_mdp = &obj_md;
1606         int rc = 0;
1607         ENTRY;
1608
1609         LASSERT(ea == NULL);
1610         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1611
1612         rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp,
1613                         oa, &obj_mdp, oti);
1614
1615         RETURN(rc);
1616 }
1617
1618 int lmv_getready(struct obd_export *exp)
1619 {
1620         struct obd_device *obd = exp->exp_obd;
1621         int rc = 0;
1622
1623         ENTRY;
1624         rc = lmv_check_connect(obd);
1625         RETURN(rc);
1626 }
1627
1628 /*
1629  * to be called from MDS only. @oa should have correct store cookie and o_fid
1630  * values for "master" object, as it will be used.
1631  */
1632 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
1633                    struct lov_stripe_md **ea, struct obd_trans_info *oti)
1634 {
1635         struct obd_device *obd = exp->exp_obd;
1636         struct lmv_obd *lmv = &obd->u.lmv;
1637         struct lustre_id mid;
1638         int i, c, rc = 0;
1639         struct mea *mea;
1640         ENTRY;
1641
1642         rc = lmv_check_connect(obd);
1643         if (rc)
1644                 RETURN(rc);
1645
1646         LASSERT(oa != NULL);
1647
1648         if (ea == NULL) {
1649                 rc = lmv_obd_create_single(exp, oa, NULL, oti);
1650                 if (rc)
1651                         CERROR("Can't create object, rc = %d\n", rc);
1652                 RETURN(rc);
1653         }
1654
1655         if (*ea == NULL) {
1656                 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1657                 if (rc < 0) {
1658                         CERROR("obd_alloc_diskmd() failed, error %d\n",
1659                                rc);
1660                         RETURN(rc);
1661                 } else
1662                         rc = 0;
1663
1664                 if (*ea == NULL)
1665                         RETURN(-ENOMEM);
1666         }
1667
1668         /*
1669          * here we should take care about splitted dir, so store cookie and fid
1670          * for "master" object should already be allocated and passed in @oa.
1671          */
1672         LASSERT(oa->o_id != 0);
1673         LASSERT(oa->o_fid != 0);
1674
1675         /* save "master" object id */
1676         obdo2id(&mid, oa);
1677
1678         mea = (struct mea *)*ea;
1679         mea->mea_master = -1;
1680         mea->mea_magic = MEA_MAGIC_ALL_CHARS;
1681
1682         if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1683                 mea->mea_count = lmv->desc.ld_tgt_count;
1684
1685         for (i = 0, c = 0; c < mea->mea_count && i < lmv->desc.ld_tgt_count; i++) {
1686                 struct lov_stripe_md obj_md;
1687                 struct lov_stripe_md *obj_mdp = &obj_md;
1688
1689                 if (lmv->tgts[i].ltd_exp == NULL) {
1690                         /* this is "master" MDS */
1691                         mea->mea_master = i;
1692                         mea->mea_ids[c] = mid;
1693                         c++;
1694                         continue;
1695                 }
1696
1697                 /*
1698                  * "master" MDS should always be part of stripped dir,
1699                  * so scan for it.
1700                  */
1701                 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1702                         continue;
1703
1704                 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE |
1705                         OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1706
1707                 rc = obd_create(lmv->tgts[c].ltd_exp, oa, &obj_mdp, oti);
1708                 if (rc) {
1709                         CERROR("obd_create() failed on MDT target %d, "
1710                                "error %d\n", c, rc);
1711                         RETURN(rc);
1712                 }
1713
1714                 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1715                        i, oa->o_id, oa->o_generation);
1716
1717
1718                 /*
1719                  * here, when object is created (or it is master and was passed
1720                  * from caller) on desired MDS we save its fid to local mea_ids.
1721                  */
1722                 LASSERT(oa->o_fid);
1723
1724                 /*
1725                  * store cookie should be defined here for both cases (master
1726                  * object and not master), because master is already created.
1727                  */
1728                 LASSERT(oa->o_id);
1729
1730                 /* fill mea by store cookie and fid */
1731                 obdo2id(&mea->mea_ids[c], oa);
1732                 c++;
1733         }
1734         LASSERT(c == mea->mea_count);
1735
1736         CDEBUG(D_OTHER, "%d dirobjects created\n",
1737                (int)mea->mea_count);
1738
1739         RETURN(rc);
1740 }
1741
1742 static int lmv_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1743                          struct obd_device *tgt, int count,
1744                          struct llog_catid *logid)
1745 {
1746         struct llog_ctxt *ctxt;
1747         int rc;
1748         ENTRY;
1749
1750         rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1751                             &llog_client_ops);
1752         if (rc == 0) {
1753                 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1754                 ctxt->loc_imp = tgt->u.cli.cl_import;
1755         }
1756
1757         RETURN(rc);
1758 }
1759
1760 static int lmv_llog_finish(struct obd_device *obd,
1761                            struct obd_llogs *llogs, int count)
1762 {
1763         int rc;
1764         ENTRY;
1765
1766         rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1767         RETURN(rc);
1768 }
1769
1770 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1771                         void *key, __u32 *vallen, void *val)
1772 {
1773         struct obd_device *obd;
1774         struct lmv_obd *lmv;
1775         int rc = 0;
1776         ENTRY;
1777
1778         obd = class_exp2obd(exp);
1779         if (obd == NULL) {
1780                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1781                        exp->exp_handle.h_cookie);
1782                 RETURN(-EINVAL);
1783         }
1784
1785         lmv = &obd->u.lmv;
1786         if (keylen == 6 && memcmp(key, "mdsize", 6) == 0) {
1787                 __u32 *mdsize = val;
1788                 *vallen = sizeof(__u32);
1789                 *mdsize = sizeof(struct lustre_id) * lmv->desc.ld_tgt_count
1790                         + sizeof(struct mea);
1791                 RETURN(0);
1792         } else if (keylen == 6 && memcmp(key, "mdsnum", 6) == 0) {
1793                 struct obd_uuid *cluuid = &lmv->cluuid;
1794                 struct lmv_tgt_desc *tgts;
1795                 __u32 *mdsnum = val;
1796                 int i;
1797
1798                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1799                         if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1800                                 *vallen = sizeof(__u32);
1801                                 *mdsnum = i;
1802                                 RETURN(0);
1803                         }
1804                 }
1805                 LASSERT(0);
1806         } else if (keylen == 6 && memcmp(key, "rootid", 6) == 0) {
1807                 /* getting rootid from first MDS. */
1808                 rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key,
1809                                   vallen, val);
1810                 RETURN(rc);
1811         } else if (keylen >= strlen("lmvdesc") && strcmp(key, "lmvdesc") == 0) {
1812                 struct lmv_desc *desc_ret = val;
1813                 *desc_ret = lmv->desc;
1814                 RETURN(0);
1815         } else if (keylen == strlen("remote_flag") &&
1816                    !strcmp(key, "remote_flag")) {
1817                 struct lmv_tgt_desc *tgts;
1818                 int i;
1819
1820                 LASSERT(*vallen == sizeof(__u32));
1821                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
1822                      i++, tgts++) {
1823
1824                         /* all tgts should be connected when this get called. */
1825                         if (!tgts || !tgts->ltd_exp) {
1826                                 CERROR("target not setup?\n");
1827                                 continue;
1828                         }
1829
1830                         if (!obd_get_info(tgts->ltd_exp, keylen, key,
1831                                           vallen, val))
1832                                 RETURN(0);
1833                 }
1834                 RETURN(-EINVAL);
1835         }
1836
1837         CDEBUG(D_IOCTL, "invalid key\n");
1838         RETURN(-EINVAL);
1839 }
1840
1841 int lmv_set_info(struct obd_export *exp, obd_count keylen,
1842                  void *key, obd_count vallen, void *val)
1843 {
1844         struct obd_device *obd;
1845         struct lmv_obd *lmv;
1846         ENTRY;
1847
1848         obd = class_exp2obd(exp);
1849         if (obd == NULL) {
1850                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1851                        exp->exp_handle.h_cookie);
1852                 RETURN(-EINVAL);
1853         }
1854         lmv = &obd->u.lmv;
1855
1856         if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
1857                 lmv->server_timeout = 1;
1858                 lmv_set_timeouts(obd);
1859                 RETURN(0);
1860         }
1861
1862         /* maybe this could be default */
1863         if ((keylen == strlen("sec") && strcmp(key, "sec") == 0) ||
1864             (keylen == strlen("nllu") && strcmp(key, "nllu") == 0)) {
1865                 struct lmv_tgt_desc *tgt;
1866                 struct obd_export *exp;
1867                 int rc = 0, err, i;
1868
1869                 spin_lock(&lmv->lmv_lock);
1870                 for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count;
1871                      i++, tgt++) {
1872                         exp = tgt->ltd_exp;
1873                         /* during setup time the connections to mdc might
1874                          * haven't been established.
1875                          */
1876                         if (exp == NULL) {
1877                                 struct obd_device *tgt_obd;
1878
1879                                 tgt_obd = class_find_client_obd(&tgt->uuid,
1880                                                                 LUSTRE_MDC_NAME,
1881                                                                 &obd->obd_uuid);
1882                                 if (!tgt_obd) {
1883                                         CERROR("can't set info %s, "
1884                                                "device %s not attached?\n",
1885                                                 (char *) key, tgt->uuid.uuid);
1886                                         rc = -EINVAL;
1887                                         continue;
1888                                 }
1889                                 exp = tgt_obd->obd_self_export;
1890                         }
1891
1892                         err = obd_set_info(exp, keylen, key, vallen, val);
1893                         if (!rc)
1894                                 rc = err;
1895                 }
1896                 spin_unlock(&lmv->lmv_lock);
1897
1898                 RETURN(rc);
1899         }
1900
1901         RETURN(-EINVAL);
1902 }
1903
1904 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
1905                struct lov_stripe_md *lsm)
1906 {
1907         struct obd_device *obd = class_exp2obd(exp);
1908         struct lmv_obd *lmv = &obd->u.lmv;
1909         struct mea *meap, *lsmp;
1910         int mea_size, i;
1911         ENTRY;
1912
1913         mea_size = (sizeof(struct lustre_id) *
1914                     lmv->desc.ld_tgt_count) + sizeof(struct mea);
1915         if (!lmmp)
1916                 RETURN(mea_size);
1917
1918         if (*lmmp && !lsm) {
1919                 OBD_FREE(*lmmp, mea_size);
1920                 *lmmp = NULL;
1921                 RETURN(0);
1922         }
1923
1924         if (*lmmp == NULL) {
1925                 OBD_ALLOC(*lmmp, mea_size);
1926                 if (*lmmp == NULL)
1927                         RETURN(-ENOMEM);
1928         }
1929
1930         if (!lsm)
1931                 RETURN(mea_size);
1932
1933         lsmp = (struct mea *)lsm;
1934         meap = (struct mea *)*lmmp;
1935
1936         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
1937             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
1938                 RETURN(-EINVAL);
1939
1940         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
1941         meap->mea_count = cpu_to_le32(lsmp->mea_count);
1942         meap->mea_master = cpu_to_le32(lsmp->mea_master);
1943
1944         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1945                 meap->mea_ids[i] = meap->mea_ids[i];
1946                 id_cpu_to_le(&meap->mea_ids[i]);
1947         }
1948
1949         RETURN(mea_size);
1950 }
1951
1952 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
1953                  struct lov_mds_md *lmm, int lmm_size)
1954 {
1955         struct obd_device *obd = class_exp2obd(exp);
1956         struct mea **tmea = (struct mea **)lsmp;
1957         struct mea *mea = (struct mea *)lmm;
1958         struct lmv_obd *lmv = &obd->u.lmv;
1959         int mea_size, i, rc = 0;
1960         __u32 magic;
1961         ENTRY;
1962
1963         mea_size = sizeof(struct lustre_id) *
1964                 lmv->desc.ld_tgt_count + sizeof(struct mea);
1965
1966         if (lsmp == NULL)
1967                 return mea_size;
1968
1969         if (*lsmp != NULL && lmm == NULL) {
1970                 OBD_FREE(*tmea, mea_size);
1971                 RETURN(0);
1972         }
1973
1974         LASSERT(mea_size == lmm_size);
1975
1976         OBD_ALLOC(*tmea, mea_size);
1977         if (*tmea == NULL)
1978                 RETURN(-ENOMEM);
1979
1980         if (!lmm)
1981                 RETURN(mea_size);
1982
1983         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
1984             mea->mea_magic == MEA_MAGIC_ALL_CHARS)
1985         {
1986                 magic = le32_to_cpu(mea->mea_magic);
1987         } else {
1988                 struct mea_old *old = (struct mea_old *)lmm;
1989
1990                 mea_size = sizeof(struct lustre_id) * old->mea_count +
1991                         sizeof(struct mea_old);
1992
1993                 if (old->mea_count > 256 || old->mea_master > 256 ||
1994                     lmm_size < mea_size || old->mea_master > old->mea_count) {
1995                         CWARN("bad MEA: count %u, master %u, size %u\n",
1996                               old->mea_count, old->mea_master, mea_size);
1997                         GOTO(out_free_mea, rc = -EINVAL);
1998                 }
1999                 magic = MEA_MAGIC_LAST_CHAR;
2000         }
2001
2002         (*tmea)->mea_magic = magic;
2003         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2004         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2005
2006         for (i = 0; i < (*tmea)->mea_count; i++) {
2007                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2008                 id_le_to_cpu(&(*tmea)->mea_ids[i]);
2009         }
2010         RETURN(mea_size);
2011
2012 out_free_mea:
2013         OBD_FREE(*tmea, mea_size);
2014         return rc;
2015 }
2016
2017 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
2018             struct lov_stripe_md *ea, obd_count oa_bufs,
2019             struct brw_page *pgarr, struct obd_trans_info *oti)
2020 {
2021         struct obd_device *obd = exp->exp_obd;
2022         struct lmv_obd *lmv = &obd->u.lmv;
2023         struct mea *mea = (struct mea *) ea;
2024         int err;
2025
2026         LASSERT(oa != NULL);
2027         LASSERT(ea != NULL);
2028         LASSERT(pgarr != NULL);
2029         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
2030
2031         oa->o_gr = id_gen(&mea->mea_ids[oa->o_mds]);
2032         oa->o_id = id_ino(&mea->mea_ids[oa->o_mds]);
2033         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2034
2035         err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp,
2036                       oa, NULL, oa_bufs, pgarr, oti);
2037         RETURN(err);
2038 }
2039
2040 struct obd_ops lmv_obd_ops = {
2041         .o_owner                = THIS_MODULE,
2042         .o_attach               = lmv_attach,
2043         .o_detach               = lmv_detach,
2044         .o_setup                = lmv_setup,
2045         .o_cleanup              = lmv_cleanup,
2046         .o_connect              = lmv_connect,
2047         .o_disconnect           = lmv_disconnect,
2048         .o_statfs               = lmv_statfs,
2049         .o_llog_init            = lmv_llog_init,
2050         .o_llog_finish          = lmv_llog_finish,
2051         .o_get_info             = lmv_get_info,
2052         .o_set_info             = lmv_set_info,
2053         .o_create               = lmv_obd_create,
2054         .o_packmd               = lmv_packmd,
2055         .o_unpackmd             = lmv_unpackmd,
2056         .o_brw                  = lmv_brw,
2057         .o_init_ea_size         = lmv_init_ea_size,
2058         .o_notify               = lmv_notify,
2059         .o_iocontrol            = lmv_iocontrol,
2060         .o_getready             = lmv_getready,
2061 };
2062
2063 struct md_ops lmv_md_ops = {
2064         .m_getstatus           = lmv_getstatus,
2065         .m_getattr             = lmv_getattr,
2066         .m_change_cbdata       = lmv_change_cbdata,
2067         .m_change_cbdata_name  = lmv_change_cbdata_name,
2068         .m_close               = lmv_close,
2069         .m_create              = lmv_create,
2070         .m_done_writing        = lmv_done_writing,
2071         .m_enqueue             = lmv_enqueue,
2072         .m_getattr_lock        = lmv_getattr_lock,
2073         .m_intent_lock         = lmv_intent_lock,
2074         .m_link                = lmv_link,
2075         .m_rename              = lmv_rename,
2076         .m_setattr             = lmv_setattr,
2077         .m_sync                = lmv_sync,
2078         .m_readpage            = lmv_readpage,
2079         .m_unlink              = lmv_unlink,
2080         .m_get_real_obd        = lmv_get_real_obd,
2081         .m_valid_attrs         = lmv_valid_attrs,
2082         .m_delete_inode        = lmv_delete_inode,
2083 };
2084
2085 int __init lmv_init(void)
2086 {
2087         struct lprocfs_static_vars lvars;
2088         int rc;
2089
2090         obj_cache = kmem_cache_create("lmv_objects",
2091                                       sizeof(struct lmv_obj),
2092                                       0, 0, NULL, NULL);
2093         if (!obj_cache) {
2094                 CERROR("error allocating lmv objects cache\n");
2095                 return -ENOMEM;
2096         }
2097
2098         lprocfs_init_vars(lmv, &lvars);
2099         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2100                                  lvars.module_vars,
2101                                  OBD_LMV_DEVICENAME);
2102         if (rc)
2103                 kmem_cache_destroy(obj_cache);
2104
2105         return rc;
2106 }
2107
2108 #ifdef __KERNEL__
2109 static void lmv_exit(void)
2110 {
2111         class_unregister_type(OBD_LMV_DEVICENAME);
2112
2113         LASSERTF(kmem_cache_destroy(obj_cache) == 0,
2114                  "can't free lmv objects cache, %d object(s)"
2115                  "still in use\n", atomic_read(&obj_cache_count));
2116 }
2117
2118 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2119 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2120 MODULE_LICENSE("GPL");
2121
2122 module_init(lmv_init);
2123 module_exit(lmv_exit);
2124 #endif