lustre/obdecho/echo_client.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_ECHO
  38 #ifdef __KERNEL__
  39 #include <libcfs/libcfs.h>
  40 #else
  41 #include <liblustre.h>
  42 #endif
  43
  44 #include <obd.h>
  45 #include <obd_support.h>
  46 #include <obd_class.h>
  47 #include <obd_echo.h>
  48 #include <lustre_debug.h>
  49 #include <lprocfs_status.h>
  50
  51 static obd_id last_object_id;
  52
  53 #if 0
  54 static void
  55 echo_printk_object (char *msg, struct ec_object *eco)
  56 {
  57         struct lov_stripe_md *lsm = eco->eco_lsm;
  58         int                   i;
  59
  60         CDEBUG(D_INFO, "%s: object %p: "LPX64", refs %d%s: "LPX64"=%u!%u\n",
  61                msg, eco, eco->eco_id, eco->eco_refcount,
  62                eco->eco_deleted ? "(deleted) " : "",
  63                lsm->lsm_object_id, lsm->lsm_stripe_size,
  64                lsm->lsm_stripe_count);
  65
  66         for (i = 0; i < lsm->lsm_stripe_count; i++)
  67                 CDEBUG(D_INFO, "@%2u:"LPX64"\n",
  68                        lsm->lsm_oinfo[i].loi_ost_idx,
  69                        lsm->lsm_oinfo[i].loi_id);
  70 }
  71 #endif
  72
  73 static struct ec_object *
  74 echo_find_object_locked (struct obd_device *obd, obd_id id)
  75 {
  76         struct echo_client_obd *ec = &obd->u.echo_client;
  77         struct ec_object       *eco = NULL;
  78         struct list_head       *el;
  79
  80         list_for_each (el, &ec->ec_objects) {
  81                 eco = list_entry (el, struct ec_object, eco_obj_chain);
  82
  83                 if (eco->eco_id == id)
  84                         return (eco);
  85         }
  86         return (NULL);
  87 }
  88
  89 static int
  90 echo_copyout_lsm (struct lov_stripe_md *lsm, void *_ulsm, int ulsm_nob)
  91 {
  92         struct lov_stripe_md *ulsm = _ulsm;
  93         int nob, i;
  94
  95         nob = offsetof (struct lov_stripe_md, lsm_oinfo[lsm->lsm_stripe_count]);
  96         if (nob > ulsm_nob)
  97                 return (-EINVAL);
  98
  99         if (copy_to_user (ulsm, lsm, sizeof(ulsm)))
 100                 return (-EFAULT);
 101
 102         for (i = 0; i < lsm->lsm_stripe_count; i++) {
 103                 if (copy_to_user (ulsm->lsm_oinfo[i], lsm->lsm_oinfo[i],
 104                                   sizeof(lsm->lsm_oinfo[0])))
 105                         return (-EFAULT);
 106         }
 107         return 0;
 108 }
 109
 110 static int
 111 echo_copyin_lsm (struct obd_device *obd, struct lov_stripe_md *lsm,
 112                  void *ulsm, int ulsm_nob)
 113 {
 114         struct echo_client_obd *ec = &obd->u.echo_client;
 115         int                     i;
 116
 117         if (ulsm_nob < sizeof (*lsm))
 118                 return (-EINVAL);
 119
 120         if (copy_from_user (lsm, ulsm, sizeof (*lsm)))
 121                 return (-EFAULT);
 122
 123         if (lsm->lsm_stripe_count > ec->ec_nstripes ||
 124             lsm->lsm_magic != LOV_MAGIC ||
 125             (lsm->lsm_stripe_size & (~CFS_PAGE_MASK)) != 0 ||
 126             ((__u64)lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL))
 127                 return (-EINVAL);
 128
 129
 130         for (i = 0; i < lsm->lsm_stripe_count; i++) {
 131                 if (copy_from_user(lsm->lsm_oinfo[i],
 132                                    ((struct lov_stripe_md *)ulsm)->lsm_oinfo[i],
 133                                    sizeof(lsm->lsm_oinfo[0])))
 134                         return (-EFAULT);
 135         }
 136         return (0);
 137 }
 138
 139 static struct ec_object *
 140 echo_allocate_object (struct obd_device *obd)
 141 {
 142         struct echo_client_obd *ec = &obd->u.echo_client;
 143         struct ec_object       *eco;
 144         int rc;
 145
 146         OBD_ALLOC(eco, sizeof (*eco));
 147         if (eco == NULL)
 148                 return NULL;
 149
 150         rc = obd_alloc_memmd(ec->ec_exp, &eco->eco_lsm);
 151         if (rc < 0) {
 152                 OBD_FREE(eco, sizeof (*eco));
 153                 return NULL;
 154         }
 155
 156         eco->eco_device = obd;
 157         eco->eco_deleted = 0;
 158         eco->eco_refcount = 0;
 159         eco->eco_lsm->lsm_magic = LOV_MAGIC;
 160         /* leave stripe count 0 by default */
 161
 162         return (eco);
 163 }
 164
 165 static void
 166 echo_free_object (struct ec_object *eco)
 167 {
 168         struct obd_device      *obd = eco->eco_device;
 169         struct echo_client_obd *ec = &obd->u.echo_client;
 170
 171         LASSERT (eco->eco_refcount == 0);
 172         if (!eco->eco_lsm)
 173                 CERROR("No object %s\n", obd->obd_name);
 174         else
 175                 obd_free_memmd(ec->ec_exp, &eco->eco_lsm);
 176         OBD_FREE (eco, sizeof (*eco));
 177 }
 178
 179 static int echo_create_object(struct obd_device *obd, int on_target,
 180                               struct obdo *oa, void *ulsm, int ulsm_nob,
 181                               struct obd_trans_info *oti)
 182 {
 183         struct echo_client_obd *ec = &obd->u.echo_client;
 184         struct ec_object       *eco2;
 185         struct ec_object       *eco;
 186         struct lov_stripe_md   *lsm;
 187         int                     rc;
 188         int                     i, idx;
 189
 190         if ((oa->o_valid & OBD_MD_FLID) == 0 && /* no obj id */
 191             (on_target ||                       /* set_stripe */
 192              ec->ec_nstripes != 0)) {           /* LOV */
 193                 CERROR ("No valid oid\n");
 194                 return (-EINVAL);
 195         }
 196
 197         if (ulsm != NULL) {
 198                 eco = echo_allocate_object (obd);
 199                 if (eco == NULL)
 200                         return (-ENOMEM);
 201
 202                 lsm = eco->eco_lsm;
 203
 204                 rc = echo_copyin_lsm (obd, lsm, ulsm, ulsm_nob);
 205                 if (rc != 0)
 206                         goto failed;
 207
 208                 /* setup object ID here for !on_target and LOV hint */
 209                 if ((oa->o_valid & OBD_MD_FLID) != 0)
 210                         eco->eco_id = lsm->lsm_object_id = oa->o_id;
 211
 212                 if (lsm->lsm_stripe_count == 0)
 213                         lsm->lsm_stripe_count = ec->ec_nstripes;
 214
 215                 if (lsm->lsm_stripe_size == 0)
 216                         lsm->lsm_stripe_size = CFS_PAGE_SIZE;
 217
 218                 idx = ll_rand();
 219
 220                 /* setup stripes: indices + default ids if required */
 221                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
 222                         if (lsm->lsm_oinfo[i]->loi_id == 0)
 223                                 lsm->lsm_oinfo[i]->loi_id = lsm->lsm_object_id;
 224
 225                         lsm->lsm_oinfo[i]->loi_ost_idx =
 226                                 (idx + i) % ec->ec_nstripes;
 227                 }
 228         } else {
 229                 OBD_ALLOC(eco, sizeof(*eco));
 230                 if (!eco)
 231                         return (-ENOMEM);
 232                 eco->eco_device = obd;
 233                 lsm = NULL;
 234         }
 235
 236         if (oa->o_id == 0)
 237                 oa->o_id = ++last_object_id;
 238
 239         if (on_target) {
 240                 oa->o_gr = FILTER_GROUP_ECHO;
 241                 oa->o_valid |= OBD_MD_FLGROUP;
 242
 243                 rc = obd_create(ec->ec_exp, oa, &lsm, oti);
 244                 if (rc != 0)
 245                         goto failed;
 246
 247                 /* See what object ID we were given */
 248                 eco->eco_id = oa->o_id = lsm->lsm_object_id;
 249                 oa->o_valid |= OBD_MD_FLID;
 250
 251                 LASSERT(eco->eco_lsm == NULL || eco->eco_lsm == lsm);
 252                 eco->eco_lsm = lsm;
 253         }
 254
 255         spin_lock (&ec->ec_lock);
 256
 257         eco2 = echo_find_object_locked (obd, oa->o_id);
 258         if (eco2 != NULL) {                     /* conflict */
 259                 spin_unlock (&ec->ec_lock);
 260
 261                 CERROR ("Can't create object id "LPX64": id already exists%s\n",
 262                         oa->o_id, on_target ? " (undoing create)" : "");
 263
 264                 if (on_target)
 265                         obd_destroy(ec->ec_exp, oa, lsm, oti, NULL);
 266
 267                 rc = -EEXIST;
 268                 goto failed;
 269         }
 270
 271         list_add (&eco->eco_obj_chain, &ec->ec_objects);
 272         spin_unlock (&ec->ec_lock);
 273         CDEBUG (D_INFO,
 274                 "created %p: "LPX64"=%u#%u@%u refs %d del %d\n",
 275                 eco, eco->eco_id,
 276                 eco->eco_lsm->lsm_stripe_size,
 277                 eco->eco_lsm->lsm_stripe_count,
 278                 eco->eco_lsm->lsm_oinfo[0]->loi_ost_idx,
 279                 eco->eco_refcount, eco->eco_deleted);
 280         return (0);
 281
 282  failed:
 283         echo_free_object (eco);
 284         if (rc)
 285                 CERROR("%s: err %d on create\n", obd->obd_name, rc);
 286         return (rc);
 287 }
 288
 289 static int
 290 echo_get_object (struct ec_object **ecop, struct obd_device *obd,
 291                  struct obdo *oa)
 292 {
 293         struct echo_client_obd *ec = &obd->u.echo_client;
 294         struct ec_object       *eco;
 295         struct ec_object       *eco2;
 296         int                     rc;
 297
 298         if ((oa->o_valid & OBD_MD_FLID) == 0 ||
 299             oa->o_id == 0)                      /* disallow use of object id 0 */
 300         {
 301                 CERROR ("No valid oid\n");
 302                 return (-EINVAL);
 303         }
 304
 305         spin_lock (&ec->ec_lock);
 306         eco = echo_find_object_locked (obd, oa->o_id);
 307         if (eco != NULL) {
 308                 if (eco->eco_deleted) {           /* being deleted */
 309                         spin_unlock(&ec->ec_lock);/* (see comment in cleanup) */
 310                         return (-EAGAIN);
 311                 }
 312
 313                 eco->eco_refcount++;
 314                 spin_unlock (&ec->ec_lock);
 315                 *ecop = eco;
 316                 CDEBUG (D_INFO,
 317                         "found %p: "LPX64"=%u#%u@%u refs %d del %d\n",
 318                         eco, eco->eco_id,
 319                         eco->eco_lsm->lsm_stripe_size,
 320                         eco->eco_lsm->lsm_stripe_count,
 321                         eco->eco_lsm->lsm_oinfo[0]->loi_ost_idx,
 322                         eco->eco_refcount, eco->eco_deleted);
 323                 return (0);
 324         }
 325         spin_unlock (&ec->ec_lock);
 326
 327         if (ec->ec_nstripes != 0)               /* striping required */
 328                 return (-ENOENT);
 329
 330         eco = echo_allocate_object (obd);
 331         if (eco == NULL)
 332                 return (-ENOMEM);
 333
 334         eco->eco_id = eco->eco_lsm->lsm_object_id = oa->o_id;
 335
 336         spin_lock (&ec->ec_lock);
 337
 338         eco2 = echo_find_object_locked (obd, oa->o_id);
 339         if (eco2 == NULL) {                     /* didn't race */
 340                 list_add (&eco->eco_obj_chain, &ec->ec_objects);
 341                 spin_unlock (&ec->ec_lock);
 342                 eco->eco_refcount = 1;
 343                 *ecop = eco;
 344                 CDEBUG (D_INFO,
 345                         "created %p: "LPX64"=%u#%u@%d refs %d del %d\n",
 346                         eco, eco->eco_id,
 347                         eco->eco_lsm->lsm_stripe_size,
 348                         eco->eco_lsm->lsm_stripe_count,
 349                         eco->eco_lsm->lsm_oinfo[0]->loi_ost_idx,
 350                         eco->eco_refcount, eco->eco_deleted);
 351                 return (0);
 352         }
 353
 354         if (eco2->eco_deleted)
 355                 rc = -EAGAIN;                   /* lose race */
 356         else {
 357                 eco2->eco_refcount++;           /* take existing */
 358                 *ecop = eco2;
 359                 rc = 0;
 360                 LASSERT (eco2->eco_id == eco2->eco_lsm->lsm_object_id);
 361                 CDEBUG (D_INFO,
 362                         "found(2) %p: "LPX64"=%u#%u@%d refs %d del %d\n",
 363                         eco2, eco2->eco_id,
 364                         eco2->eco_lsm->lsm_stripe_size,
 365                         eco2->eco_lsm->lsm_stripe_count,
 366                         eco2->eco_lsm->lsm_oinfo[0]->loi_ost_idx,
 367                         eco2->eco_refcount, eco2->eco_deleted);
 368         }
 369
 370         spin_unlock (&ec->ec_lock);
 371
 372         echo_free_object (eco);
 373         return (rc);
 374 }
 375
 376 static void
 377 echo_put_object (struct ec_object *eco)
 378 {
 379         struct obd_device      *obd = eco->eco_device;
 380         struct echo_client_obd *ec = &obd->u.echo_client;
 381
 382         /* Release caller's ref on the object.
 383          * delete => mark for deletion when last ref goes
 384          */
 385
 386         spin_lock (&ec->ec_lock);
 387
 388         eco->eco_refcount--;
 389         LASSERT (eco->eco_refcount >= 0);
 390
 391         CDEBUG(D_INFO, "put %p: "LPX64"=%u#%u@%d refs %d del %d\n",
 392                eco, eco->eco_id,
 393                eco->eco_lsm->lsm_stripe_size,
 394                eco->eco_lsm->lsm_stripe_count,
 395                eco->eco_lsm->lsm_oinfo[0]->loi_ost_idx,
 396                eco->eco_refcount, eco->eco_deleted);
 397
 398         if (eco->eco_refcount != 0 || !eco->eco_deleted) {
 399                 spin_unlock (&ec->ec_lock);
 400                 return;
 401         }
 402
 403         spin_unlock (&ec->ec_lock);
 404
 405         /* NB leave obj in the object list.  We must prevent anyone from
 406          * attempting to enqueue on this object number until we can be
 407          * sure there will be no more lock callbacks.
 408          */
 409         obd_cancel_unused(ec->ec_exp, eco->eco_lsm, 0, NULL);
 410
 411         /* now we can let it go */
 412         spin_lock (&ec->ec_lock);
 413         list_del (&eco->eco_obj_chain);
 414         spin_unlock (&ec->ec_lock);
 415
 416         LASSERT (eco->eco_refcount == 0);
 417
 418         echo_free_object (eco);
 419 }
 420
 421 static void
 422 echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp)
 423 {
 424         unsigned long stripe_count;
 425         unsigned long stripe_size;
 426         unsigned long width;
 427         unsigned long woffset;
 428         int           stripe_index;
 429         obd_off       offset;
 430
 431         if (lsm->lsm_stripe_count <= 1)
 432                 return;
 433
 434         offset       = *offp;
 435         stripe_size  = lsm->lsm_stripe_size;
 436         stripe_count = lsm->lsm_stripe_count;
 437
 438         /* width = # bytes in all stripes */
 439         width = stripe_size * stripe_count;
 440
 441         /* woffset = offset within a width; offset = whole number of widths */
 442         woffset = do_div (offset, width);
 443
 444         stripe_index = woffset / stripe_size;
 445
 446         *idp = lsm->lsm_oinfo[stripe_index]->loi_id;
 447         *offp = offset * stripe_size + woffset % stripe_size;
 448 }
 449
 450 static void
 451 echo_client_page_debug_setup(struct lov_stripe_md *lsm,
 452                              cfs_page_t *page, int rw, obd_id id,
 453                              obd_off offset, obd_off count)
 454 {
 455         char    *addr;
 456         obd_off  stripe_off;
 457         obd_id   stripe_id;
 458         int      delta;
 459
 460         /* no partial pages on the client */
 461         LASSERT(count == CFS_PAGE_SIZE);
 462
 463         addr = cfs_kmap(page);
 464
 465         for (delta = 0; delta < CFS_PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
 466                 if (rw == OBD_BRW_WRITE) {
 467                         stripe_off = offset + delta;
 468                         stripe_id = id;
 469                         echo_get_stripe_off_id(lsm, &stripe_off, &stripe_id);
 470                 } else {
 471                         stripe_off = 0xdeadbeef00c0ffeeULL;
 472                         stripe_id = 0xdeadbeef00c0ffeeULL;
 473                 }
 474                 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
 475                                   stripe_off, stripe_id);
 476         }
 477
 478         cfs_kunmap(page);
 479 }
 480
 481 static int echo_client_page_debug_check(struct lov_stripe_md *lsm,
 482                                         cfs_page_t *page, obd_id id,
 483                                         obd_off offset, obd_off count)
 484 {
 485         obd_off stripe_off;
 486         obd_id  stripe_id;
 487         char   *addr;
 488         int     delta;
 489         int     rc;
 490         int     rc2;
 491
 492         /* no partial pages on the client */
 493         LASSERT(count == CFS_PAGE_SIZE);
 494
 495         addr = cfs_kmap(page);
 496
 497         for (rc = delta = 0; delta < CFS_PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
 498                 stripe_off = offset + delta;
 499                 stripe_id = id;
 500                 echo_get_stripe_off_id (lsm, &stripe_off, &stripe_id);
 501
 502                 rc2 = block_debug_check("test_brw",
 503                                         addr + delta, OBD_ECHO_BLOCK_SIZE,
 504                                         stripe_off, stripe_id);
 505                 if (rc2 != 0) {
 506                         CERROR ("Error in echo object "LPX64"\n", id);
 507                         rc = rc2;
 508                 }
 509         }
 510
 511         cfs_kunmap(page);
 512         return rc;
 513 }
 514
 515 static int echo_client_kbrw(struct obd_device *obd, int rw, struct obdo *oa,
 516                             struct lov_stripe_md *lsm, obd_off offset,
 517                             obd_size count, struct obd_trans_info *oti)
 518 {
 519         struct echo_client_obd *ec = &obd->u.echo_client;
 520         struct obd_info         oinfo = { { { 0 } } };
 521         obd_count               npages;
 522         struct brw_page        *pga;
 523         struct brw_page        *pgp;
 524         obd_off                 off;
 525         int                     i;
 526         int                     rc;
 527         int                     verify;
 528         int                     gfp_mask;
 529
 530         verify = ((oa->o_id) != ECHO_PERSISTENT_OBJID &&
 531                   (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
 532                   (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
 533
 534         gfp_mask = ((oa->o_id & 2) == 0) ? CFS_ALLOC_STD : CFS_ALLOC_HIGHUSER;
 535
 536         LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
 537         LASSERT(lsm != NULL);
 538         LASSERT(lsm->lsm_object_id == oa->o_id);
 539
 540         if (count <= 0 ||
 541             (count & (~CFS_PAGE_MASK)) != 0)
 542                 return (-EINVAL);
 543
 544         /* XXX think again with misaligned I/O */
 545         npages = count >> CFS_PAGE_SHIFT;
 546
 547         OBD_ALLOC(pga, npages * sizeof(*pga));
 548         if (pga == NULL)
 549                 return (-ENOMEM);
 550
 551         for (i = 0, pgp = pga, off = offset;
 552              i < npages;
 553              i++, pgp++, off += CFS_PAGE_SIZE) {
 554
 555                 LASSERT (pgp->pg == NULL);      /* for cleanup */
 556
 557                 rc = -ENOMEM;
 558                 OBD_PAGE_ALLOC(pgp->pg, gfp_mask);
 559                 if (pgp->pg == NULL)
 560                         goto out;
 561
 562                 pgp->count = CFS_PAGE_SIZE;
 563                 pgp->off = off;
 564                 pgp->flag = 0;
 565
 566                 if (verify)
 567                         echo_client_page_debug_setup(lsm, pgp->pg, rw,
 568                                                      oa->o_id, off, pgp->count);
 569         }
 570
 571         oinfo.oi_oa = oa;
 572         oinfo.oi_md = lsm;
 573         rc = obd_brw(rw, ec->ec_exp, &oinfo, npages, pga, oti);
 574
 575  out:
 576         if (rc != 0 || rw != OBD_BRW_READ)
 577                 verify = 0;
 578
 579         for (i = 0, pgp = pga; i < npages; i++, pgp++) {
 580                 if (pgp->pg == NULL)
 581                         continue;
 582
 583                 if (verify) {
 584                         int vrc;
 585                         vrc = echo_client_page_debug_check(lsm, pgp->pg, oa->o_id,
 586                                                            pgp->off, pgp->count);
 587                         if (vrc != 0 && rc == 0)
 588                                 rc = vrc;
 589                 }
 590                 OBD_PAGE_FREE(pgp->pg);
 591         }
 592         OBD_FREE(pga, npages * sizeof(*pga));
 593         return (rc);
 594 }
 595
 596 struct echo_async_state;
 597
 598 #define EAP_MAGIC 79277927
 599 struct echo_async_page {
 600         int                     eap_magic;
 601         cfs_page_t             *eap_page;
 602         void                    *eap_cookie;
 603         obd_off                 eap_off;
 604         struct echo_async_state *eap_eas;
 605         struct list_head        eap_item;
 606 };
 607
 608 static inline struct echo_async_page *eap_from_cookie(void *ptr)
 609 {
 610         struct echo_async_page *ap = ptr;
 611         LASSERT(ap->eap_magic == EAP_MAGIC);
 612         return ap;
 613 }
 614
 615 struct echo_async_state {
 616         spinlock_t              eas_lock;
 617         obd_off                 eas_next_offset;
 618         obd_off                 eas_end_offset;
 619         int                     eas_in_flight;
 620         int                     eas_rc;
 621         cfs_waitq_t             eas_waitq;
 622         struct list_head        eas_avail;
 623         struct obdo             eas_oa;
 624         struct lov_stripe_md    *eas_lsm;
 625 };
 626
 627 static int eas_should_wake(struct echo_async_state *eas)
 628 {
 629         int rc = 0;
 630
 631         spin_lock(&eas->eas_lock);
 632         if (eas->eas_rc == 0 && !list_empty(&eas->eas_avail))
 633             rc = 1;
 634         spin_unlock(&eas->eas_lock);
 635         return rc;
 636 };
 637
 638 static int ec_ap_make_ready(void *data, int cmd)
 639 {
 640         /* our pages are issued ready */
 641         LBUG();
 642         return 0;
 643 }
 644 static int ec_ap_refresh_count(void *data, int cmd)
 645 {
 646         /* our pages are issued with a stable count */
 647         LBUG();
 648         return CFS_PAGE_SIZE;
 649 }
 650 static void ec_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
 651 {
 652         struct echo_async_page *eap = eap_from_cookie(data);
 653
 654         memcpy(oa, &eap->eap_eas->eas_oa, sizeof(*oa));
 655 }
 656
 657 static int ec_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
 658 {
 659         struct echo_async_page *eap = eap_from_cookie(data);
 660         struct echo_async_state *eas;
 661
 662         eas = eap->eap_eas;
 663
 664         if (cmd == OBD_BRW_READ &&
 665             eas->eas_oa.o_id != ECHO_PERSISTENT_OBJID &&
 666             (eas->eas_oa.o_valid & OBD_MD_FLFLAGS) != 0 &&
 667             (eas->eas_oa.o_flags & OBD_FL_DEBUG_CHECK) != 0)
 668                 echo_client_page_debug_check(eas->eas_lsm, eap->eap_page,
 669                                              eas->eas_oa.o_id, eap->eap_off,
 670                                              CFS_PAGE_SIZE);
 671
 672         spin_lock(&eas->eas_lock);
 673         if (rc && !eas->eas_rc)
 674                 eas->eas_rc = rc;
 675         eas->eas_in_flight--;
 676         list_add(&eap->eap_item, &eas->eas_avail);
 677         cfs_waitq_signal(&eas->eas_waitq);
 678         spin_unlock(&eas->eas_lock);
 679         return 0;
 680 }
 681
 682 static struct obd_async_page_ops ec_async_page_ops = {
 683         .ap_make_ready =        ec_ap_make_ready,
 684         .ap_refresh_count =     ec_ap_refresh_count,
 685         .ap_fill_obdo =         ec_ap_fill_obdo,
 686         .ap_completion =        ec_ap_completion,
 687 };
 688
 689 static int echo_client_async_page(struct obd_export *exp, int rw,
 690                                    struct obdo *oa, struct lov_stripe_md *lsm,
 691                                    obd_off offset, obd_size count,
 692                                    obd_size batching)
 693 {
 694         obd_count npages, i;
 695         struct echo_async_page *eap;
 696         struct echo_async_state eas;
 697         int rc = 0;
 698         struct echo_async_page **aps = NULL;
 699
 700         ENTRY;
 701 #if 0
 702         int                     verify;
 703         int                     gfp_mask;
 704
 705         verify = ((oa->o_id) != ECHO_PERSISTENT_OBJID &&
 706                   (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
 707                   (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
 708
 709         gfp_mask = ((oa->o_id & 2) == 0) ? GFP_KERNEL : GFP_HIGHUSER;
 710 #endif
 711
 712         LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
 713
 714         if (count <= 0 ||
 715             (count & (~CFS_PAGE_MASK)) != 0 ||
 716             (lsm != NULL &&
 717              lsm->lsm_object_id != oa->o_id))
 718                 return (-EINVAL);
 719
 720         /* XXX think again with misaligned I/O */
 721         npages = batching >> CFS_PAGE_SHIFT;
 722
 723         memcpy(&eas.eas_oa, oa, sizeof(*oa));
 724         eas.eas_next_offset = offset;
 725         eas.eas_end_offset = offset + count;
 726         spin_lock_init(&eas.eas_lock);
 727         cfs_waitq_init(&eas.eas_waitq);
 728         eas.eas_in_flight = 0;
 729         eas.eas_rc = 0;
 730         eas.eas_lsm = lsm;
 731         CFS_INIT_LIST_HEAD(&eas.eas_avail);
 732
 733         OBD_ALLOC(aps, npages * sizeof aps[0]);
 734         if (aps == NULL)
 735                 return (-ENOMEM);
 736
 737         /* prepare the group of pages that we're going to be keeping
 738          * in flight */
 739         for (i = 0; i < npages; i++) {
 740                 cfs_page_t *page;
 741                 OBD_PAGE_ALLOC(page, CFS_ALLOC_STD);
 742                 if (page == NULL)
 743                         GOTO(out, rc = -ENOMEM);
 744
 745                 OBD_ALLOC(eap, sizeof(*eap));
 746                 if (eap == NULL) {
 747                         OBD_PAGE_FREE(page);
 748                         GOTO(out, rc = -ENOMEM);
 749                 }
 750
 751                 eap->eap_magic = EAP_MAGIC;
 752                 eap->eap_page = page;
 753                 eap->eap_eas = &eas;
 754                 list_add_tail(&eap->eap_item, &eas.eas_avail);
 755                 aps[i] = eap;
 756         }
 757
 758         /* first we spin queueing io and being woken by its completion */
 759         spin_lock(&eas.eas_lock);
 760         for(;;) {
 761                 int rc;
 762
 763                 /* sleep until we have a page to send */
 764                 spin_unlock(&eas.eas_lock);
 765                 rc = wait_event_interruptible(eas.eas_waitq,
 766                                               eas_should_wake(&eas));
 767                 spin_lock(&eas.eas_lock);
 768                 if (rc && !eas.eas_rc)
 769                         eas.eas_rc = rc;
 770                 if (eas.eas_rc)
 771                         break;
 772                 if (list_empty(&eas.eas_avail))
 773                         continue;
 774                 eap = list_entry(eas.eas_avail.next, struct echo_async_page,
 775                                  eap_item);
 776                 list_del(&eap->eap_item);
 777                 spin_unlock(&eas.eas_lock);
 778
 779                 /* unbind the eap from its old page offset */
 780                 if (eap->eap_cookie != NULL) {
 781                         obd_teardown_async_page(exp, lsm, NULL,
 782                                                 eap->eap_cookie);
 783                         eap->eap_cookie = NULL;
 784                 }
 785
 786                 eas.eas_next_offset += CFS_PAGE_SIZE;
 787                 eap->eap_off = eas.eas_next_offset;
 788
 789                 rc = obd_prep_async_page(exp, lsm, NULL, eap->eap_page,
 790                                          eap->eap_off, &ec_async_page_ops,
 791                                          eap, &eap->eap_cookie, 1, NULL);
 792                 if (rc) {
 793                         spin_lock(&eas.eas_lock);
 794                         eas.eas_rc = rc;
 795                         break;
 796                 }
 797
 798                 if (oa->o_id != ECHO_PERSISTENT_OBJID &&
 799                     (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
 800                     (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0)
 801                         echo_client_page_debug_setup(lsm, eap->eap_page, rw,
 802                                                      oa->o_id,
 803                                                      eap->eap_off, CFS_PAGE_SIZE);
 804
 805                 /* always asserts urgent, which isn't quite right */
 806                 rc = obd_queue_async_io(exp, lsm, NULL, eap->eap_cookie,
 807                                         rw, 0, CFS_PAGE_SIZE, 0,
 808                                         ASYNC_READY | ASYNC_URGENT |
 809                                         ASYNC_COUNT_STABLE);
 810                 spin_lock(&eas.eas_lock);
 811                 if (rc && !eas.eas_rc) {
 812                         eas.eas_rc = rc;
 813                         break;
 814                 }
 815                 eas.eas_in_flight++;
 816                 if (eas.eas_next_offset == eas.eas_end_offset)
 817                         break;
 818         }
 819
 820         /* still hold the eas_lock here.. */
 821
 822         /* now we just spin waiting for all the rpcs to complete */
 823         while(eas.eas_in_flight) {
 824                 spin_unlock(&eas.eas_lock);
 825                 wait_event_interruptible(eas.eas_waitq,
 826                                          eas.eas_in_flight == 0);
 827                 spin_lock(&eas.eas_lock);
 828         }
 829         spin_unlock(&eas.eas_lock);
 830
 831 out:
 832         if (aps != NULL) {
 833                 for (i = 0; i < npages; ++ i) {
 834                         cfs_page_t *page;
 835
 836                         eap = aps[i];
 837                         page = eap->eap_page;
 838                         if (eap->eap_cookie != NULL)
 839                                 obd_teardown_async_page(exp, lsm, NULL,
 840                                                         eap->eap_cookie);
 841                         OBD_FREE(eap, sizeof(*eap));
 842                         OBD_PAGE_FREE(page);
 843                 }
 844                 OBD_FREE(aps, npages * sizeof aps[0]);
 845         }
 846
 847         RETURN(rc);
 848 }
 849
 850 static int echo_client_prep_commit(struct obd_export *exp, int rw,
 851                                    struct obdo *oa, struct lov_stripe_md *lsm,
 852                                    obd_off offset, obd_size count,
 853                                    obd_size batch, struct obd_trans_info *oti)
 854 {
 855         struct obd_ioobj ioo;
 856         struct niobuf_local *lnb;
 857         struct niobuf_remote *rnb;
 858         obd_off off;
 859         obd_size npages, tot_pages;
 860         int i, ret = 0;
 861         ENTRY;
 862
 863         if (count <= 0 || (count & (~CFS_PAGE_MASK)) != 0 ||
 864             (lsm != NULL && lsm->lsm_object_id != oa->o_id))
 865                 RETURN(-EINVAL);
 866
 867         npages = batch >> CFS_PAGE_SHIFT;
 868         tot_pages = count >> CFS_PAGE_SHIFT;
 869
 870         OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
 871         OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote));
 872
 873         if (lnb == NULL || rnb == NULL)
 874                 GOTO(out, ret = -ENOMEM);
 875
 876         obdo_to_ioobj(oa, &ioo);
 877
 878         off = offset;
 879
 880         for(; tot_pages; tot_pages -= npages) {
 881                 int lpages;
 882
 883                 if (tot_pages < npages)
 884                         npages = tot_pages;
 885
 886                 for (i = 0; i < npages; i++, off += CFS_PAGE_SIZE) {
 887                         rnb[i].offset = off;
 888                         rnb[i].len = CFS_PAGE_SIZE;
 889                 }
 890
 891                 ioo.ioo_bufcnt = npages;
 892                 oti->oti_transno = 0;
 893
 894                 lpages = npages;
 895                 ret = obd_preprw(rw, exp, oa, 1, &ioo, rnb, &lpages, lnb, oti,
 896                                  NULL);
 897                 if (ret != 0)
 898                         GOTO(out, ret);
 899                 LASSERT(lpages == npages);
 900
 901                 for (i = 0; i < lpages; i++) {
 902                         cfs_page_t *page = lnb[i].page;
 903
 904                         /* read past eof? */
 905                         if (page == NULL && lnb[i].rc == 0)
 906                                 continue;
 907
 908                         if (oa->o_id == ECHO_PERSISTENT_OBJID ||
 909                             (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
 910                             (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
 911                                 continue;
 912
 913                         if (rw == OBD_BRW_WRITE)
 914                                 echo_client_page_debug_setup(lsm, page, rw,
 915                                                              oa->o_id,
 916                                                              rnb[i].offset,
 917                                                              rnb[i].len);
 918                         else
 919                                 echo_client_page_debug_check(lsm, page,
 920                                                              oa->o_id,
 921                                                              rnb[i].offset,
 922                                                              rnb[i].len);
 923                 }
 924
 925                 ret = obd_commitrw(rw, exp, oa, 1,&ioo,rnb,npages,lnb,oti,ret);
 926                 if (ret != 0)
 927                         GOTO(out, ret);
 928         }
 929
 930 out:
 931         if (lnb)
 932                 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
 933         if (rnb)
 934                 OBD_FREE(rnb, npages * sizeof(struct niobuf_remote));
 935         RETURN(ret);
 936 }
 937
 938 int echo_client_brw_ioctl(int rw, struct obd_export *exp,
 939                           struct obd_ioctl_data *data)
 940 {
 941         struct obd_device *obd = class_exp2obd(exp);
 942         struct echo_client_obd *ec = &obd->u.echo_client;
 943         struct obd_trans_info dummy_oti = { .oti_thread_id = -1 };
 944         struct ec_object *eco;
 945         int rc;
 946         ENTRY;
 947
 948         rc = echo_get_object(&eco, obd, &data->ioc_obdo1);
 949         if (rc)
 950                 RETURN(rc);
 951
 952         data->ioc_obdo1.o_valid &= ~OBD_MD_FLHANDLE;
 953         data->ioc_obdo1.o_valid |= OBD_MD_FLGROUP;
 954         data->ioc_obdo1.o_gr = FILTER_GROUP_ECHO;
 955
 956         switch((long)data->ioc_pbuf1) {
 957         case 1:
 958                 rc = echo_client_kbrw(obd, rw, &data->ioc_obdo1,
 959                                               eco->eco_lsm, data->ioc_offset,
 960                                               data->ioc_count, &dummy_oti);
 961                 break;
 962         case 2:
 963                 rc = echo_client_async_page(ec->ec_exp, rw, &data->ioc_obdo1,
 964                                            eco->eco_lsm, data->ioc_offset,
 965                                            data->ioc_count, data->ioc_plen1);
 966                 break;
 967         case 3:
 968                 rc = echo_client_prep_commit(ec->ec_exp, rw, &data->ioc_obdo1,
 969                                             eco->eco_lsm, data->ioc_offset,
 970                                             data->ioc_count, data->ioc_plen1,
 971                                             &dummy_oti);
 972                 break;
 973         default:
 974                 rc = -EINVAL;
 975         }
 976         echo_put_object(eco);
 977         RETURN(rc);
 978 }
 979
 980 static int
 981 echo_ldlm_callback (struct ldlm_lock *lock, struct ldlm_lock_desc *new,
 982                     void *data, int flag)
 983 {
 984         struct ec_object       *eco = (struct ec_object *)data;
 985         struct echo_client_obd *ec = &(eco->eco_device->u.echo_client);
 986         struct lustre_handle    lockh;
 987         struct list_head       *el;
 988         int                     found = 0;
 989         int                     rc;
 990
 991         ldlm_lock2handle (lock, &lockh);
 992
 993         /* #ifdef this out if we're not feeling paranoid */
 994         spin_lock (&ec->ec_lock);
 995         list_for_each (el, &ec->ec_objects) {
 996                 found = (eco == list_entry(el, struct ec_object,
 997                                            eco_obj_chain));
 998                 if (found)
 999                         break;
1000         }
1001         spin_unlock (&ec->ec_lock);
1002         LASSERT (found);
1003
1004         switch (flag) {
1005         case LDLM_CB_BLOCKING:
1006                 CDEBUG(D_INFO, "blocking callback on "LPX64", handle "LPX64"\n",
1007                        eco->eco_id, lockh.cookie);
1008                 rc = ldlm_cli_cancel (&lockh);
1009                 if (rc != ELDLM_OK)
1010                         CERROR ("ldlm_cli_cancel failed: %d\n", rc);
1011                 break;
1012
1013         case LDLM_CB_CANCELING:
1014                 CDEBUG(D_INFO, "cancel callback on "LPX64", handle "LPX64"\n",
1015                        eco->eco_id, lockh.cookie);
1016                 break;
1017
1018         default:
1019                 LBUG ();
1020         }
1021
1022         return (0);
1023 }
1024
1025 static int
1026 echo_client_enqueue(struct obd_export *exp, struct obdo *oa,
1027                     int mode, obd_off offset, obd_size nob)
1028 {
1029         struct obd_device      *obd = exp->exp_obd;
1030         struct echo_client_obd *ec = &obd->u.echo_client;
1031         struct lustre_handle   *ulh = &oa->o_handle;
1032         struct ldlm_enqueue_info einfo = { 0 };
1033         struct obd_info oinfo = { { { 0 } } };
1034         struct ec_object       *eco;
1035         struct ec_lock         *ecl;
1036         int                     rc;
1037
1038         if (!(mode == LCK_PR || mode == LCK_PW))
1039                 return -EINVAL;
1040
1041         if ((offset & (~CFS_PAGE_MASK)) != 0 ||
1042             (nob & (~CFS_PAGE_MASK)) != 0)
1043                 return -EINVAL;
1044
1045         rc = echo_get_object (&eco, obd, oa);
1046         if (rc != 0)
1047                 return rc;
1048
1049         rc = -ENOMEM;
1050         OBD_ALLOC (ecl, sizeof (*ecl));
1051         if (ecl == NULL)
1052                 goto failed_0;
1053
1054         ecl->ecl_mode = mode;
1055         ecl->ecl_object = eco;
1056         ecl->ecl_policy.l_extent.start = offset;
1057         ecl->ecl_policy.l_extent.end =
1058                 (nob == 0) ? ((obd_off) -1) : (offset + nob - 1);
1059
1060         einfo.ei_type = LDLM_EXTENT;
1061         einfo.ei_mode = mode;
1062         einfo.ei_cb_bl = echo_ldlm_callback;
1063         einfo.ei_cb_cp = ldlm_completion_ast;
1064         einfo.ei_cb_gl = NULL;
1065         einfo.ei_cbdata = eco;
1066
1067         oinfo.oi_policy = ecl->ecl_policy;
1068         oinfo.oi_lockh = &ecl->ecl_lock_handle;
1069         oinfo.oi_md = eco->eco_lsm;
1070         rc = obd_enqueue(ec->ec_exp, &oinfo, &einfo, NULL);
1071         if (rc != 0)
1072                 goto failed_1;
1073
1074         CDEBUG(D_INFO, "enqueue handle "LPX64"\n", ecl->ecl_lock_handle.cookie);
1075
1076         /* NB ecl takes object ref from echo_get_object() above */
1077         spin_lock(&ec->ec_lock);
1078
1079         list_add(&ecl->ecl_exp_chain, &exp->exp_ec_data.eced_locks);
1080         ulh->cookie = ecl->ecl_cookie = ec->ec_unique++;
1081
1082         spin_unlock(&ec->ec_lock);
1083
1084         oa->o_valid |= OBD_MD_FLHANDLE;
1085         return 0;
1086
1087  failed_1:
1088         OBD_FREE (ecl, sizeof (*ecl));
1089  failed_0:
1090         echo_put_object (eco);
1091         return (rc);
1092 }
1093
1094 static int
1095 echo_client_cancel(struct obd_export *exp, struct obdo *oa)
1096 {
1097         struct obd_device      *obd = exp->exp_obd;
1098         struct echo_client_obd *ec = &obd->u.echo_client;
1099         struct lustre_handle   *ulh = &oa->o_handle;
1100         struct ec_lock         *ecl = NULL;
1101         int                     found = 0;
1102         struct list_head       *el;
1103         int                     rc;
1104
1105         if ((oa->o_valid & OBD_MD_FLHANDLE) == 0)
1106                 return -EINVAL;
1107
1108         spin_lock (&ec->ec_lock);
1109
1110         list_for_each (el, &exp->exp_ec_data.eced_locks) {
1111                 ecl = list_entry (el, struct ec_lock, ecl_exp_chain);
1112                 found = (ecl->ecl_cookie == ulh->cookie);
1113                 if (found) {
1114                         list_del (&ecl->ecl_exp_chain);
1115                         break;
1116                 }
1117         }
1118
1119         spin_unlock (&ec->ec_lock);
1120
1121         if (!found)
1122                 return (-ENOENT);
1123
1124         rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm, ecl->ecl_mode,
1125                         &ecl->ecl_lock_handle);
1126
1127         echo_put_object (ecl->ecl_object);
1128         OBD_FREE (ecl, sizeof (*ecl));
1129
1130         return rc;
1131 }
1132
1133 static int
1134 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp,
1135                       int len, void *karg, void *uarg)
1136 {
1137         struct obd_device      *obd;
1138         struct echo_client_obd *ec;
1139         struct ec_object       *eco;
1140         struct obd_ioctl_data  *data = karg;
1141         struct obd_trans_info   dummy_oti;
1142         struct oti_req_ack_lock *ack_lock;
1143         struct obdo            *oa;
1144         int                     rw = OBD_BRW_READ;
1145         int                     rc = 0;
1146         int                     i;
1147         ENTRY;
1148
1149         unlock_kernel();
1150
1151         memset(&dummy_oti, 0, sizeof(dummy_oti));
1152
1153         obd = exp->exp_obd;
1154         ec = &obd->u.echo_client;
1155
1156         switch (cmd) {
1157         case OBD_IOC_CREATE:                    /* may create echo object */
1158                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1159                         GOTO (out, rc = -EPERM);
1160
1161                 rc = echo_create_object (obd, 1, &data->ioc_obdo1,
1162                                          data->ioc_pbuf1, data->ioc_plen1,
1163                                          &dummy_oti);
1164                 GOTO(out, rc);
1165
1166         case OBD_IOC_DESTROY:
1167                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1168                         GOTO (out, rc = -EPERM);
1169
1170                 rc = echo_get_object (&eco, obd, &data->ioc_obdo1);
1171                 if (rc == 0) {
1172                         oa = &data->ioc_obdo1;
1173                         oa->o_gr = FILTER_GROUP_ECHO;
1174                         oa->o_valid |= OBD_MD_FLGROUP;
1175                         rc = obd_destroy(ec->ec_exp, oa, eco->eco_lsm,
1176                                          &dummy_oti, NULL);
1177                         if (rc == 0)
1178                                 eco->eco_deleted = 1;
1179                         echo_put_object(eco);
1180                 }
1181                 GOTO(out, rc);
1182
1183         case OBD_IOC_GETATTR:
1184                 rc = echo_get_object (&eco, obd, &data->ioc_obdo1);
1185                 if (rc == 0) {
1186                         struct obd_info oinfo = { { { 0 } } };
1187                         oinfo.oi_md = eco->eco_lsm;
1188                         oinfo.oi_oa = &data->ioc_obdo1;
1189                         rc = obd_getattr(ec->ec_exp, &oinfo);
1190                         echo_put_object(eco);
1191                 }
1192                 GOTO(out, rc);
1193
1194         case OBD_IOC_SETATTR:
1195                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1196                         GOTO (out, rc = -EPERM);
1197
1198                 rc = echo_get_object (&eco, obd, &data->ioc_obdo1);
1199                 if (rc == 0) {
1200                         struct obd_info oinfo = { { { 0 } } };
1201                         oinfo.oi_oa = &data->ioc_obdo1;
1202                         oinfo.oi_md = eco->eco_lsm;
1203
1204                         rc = obd_setattr(ec->ec_exp, &oinfo, NULL);
1205                         echo_put_object(eco);
1206                 }
1207                 GOTO(out, rc);
1208
1209         case OBD_IOC_BRW_WRITE:
1210                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1211                         GOTO (out, rc = -EPERM);
1212
1213                 rw = OBD_BRW_WRITE;
1214                 /* fall through */
1215         case OBD_IOC_BRW_READ:
1216                 rc = echo_client_brw_ioctl(rw, exp, data);
1217                 GOTO(out, rc);
1218
1219         case ECHO_IOC_GET_STRIPE:
1220                 rc = echo_get_object(&eco, obd, &data->ioc_obdo1);
1221                 if (rc == 0) {
1222                         rc = echo_copyout_lsm(eco->eco_lsm, data->ioc_pbuf1,
1223                                               data->ioc_plen1);
1224                         echo_put_object(eco);
1225                 }
1226                 GOTO(out, rc);
1227
1228         case ECHO_IOC_SET_STRIPE:
1229                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1230                         GOTO (out, rc = -EPERM);
1231
1232                 if (data->ioc_pbuf1 == NULL) {  /* unset */
1233                         rc = echo_get_object(&eco, obd, &data->ioc_obdo1);
1234                         if (rc == 0) {
1235                                 eco->eco_deleted = 1;
1236                                 echo_put_object(eco);
1237                         }
1238                 } else {
1239                         rc = echo_create_object(obd, 0, &data->ioc_obdo1,
1240                                                 data->ioc_pbuf1,
1241                                                 data->ioc_plen1, &dummy_oti);
1242                 }
1243                 GOTO (out, rc);
1244
1245         case ECHO_IOC_ENQUEUE:
1246                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1247                         GOTO (out, rc = -EPERM);
1248
1249                 rc = echo_client_enqueue(exp, &data->ioc_obdo1,
1250                                          data->ioc_conn1, /* lock mode */
1251                                    data->ioc_offset, data->ioc_count);/*extent*/
1252                 GOTO (out, rc);
1253
1254         case ECHO_IOC_CANCEL:
1255                 rc = echo_client_cancel(exp, &data->ioc_obdo1);
1256                 GOTO (out, rc);
1257
1258         default:
1259                 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
1260                 GOTO (out, rc = -ENOTTY);
1261         }
1262
1263         EXIT;
1264  out:
1265
1266         /* XXX this should be in a helper also called by target_send_reply */
1267         for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4;
1268              i++, ack_lock++) {
1269                 if (!ack_lock->mode)
1270                         break;
1271                 ldlm_lock_decref(&ack_lock->lock, ack_lock->mode);
1272         }
1273
1274         lock_kernel();
1275
1276         return rc;
1277 }
1278
1279 static int echo_client_setup(struct obd_device *obddev, struct lustre_cfg *lcfg)
1280 {
1281         struct echo_client_obd *ec = &obddev->u.echo_client;
1282         struct obd_device *tgt;
1283         struct lustre_handle conn = {0, };
1284         struct obd_uuid echo_uuid = { "ECHO_UUID" };
1285         struct obd_connect_data *ocd = NULL;
1286         int rc;
1287         ENTRY;
1288
1289         if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1290                 CERROR("requires a TARGET OBD name\n");
1291                 RETURN(-EINVAL);
1292         }
1293
1294         tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
1295         if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
1296                 CERROR("device not attached or not set up (%s)\n",
1297                        lustre_cfg_string(lcfg, 1));
1298                 RETURN(-EINVAL);
1299         }
1300
1301         spin_lock_init (&ec->ec_lock);
1302         CFS_INIT_LIST_HEAD (&ec->ec_objects);
1303         ec->ec_unique = 0;
1304
1305         OBD_ALLOC(ocd, sizeof(*ocd));
1306         if (ocd == NULL) {
1307                 CERROR("Can't alloc ocd connecting to %s\n",
1308                        lustre_cfg_string(lcfg, 1));
1309                 return -ENOMEM;
1310         }
1311
1312         ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL;
1313         ocd->ocd_version = LUSTRE_VERSION_CODE;
1314         ocd->ocd_group = FILTER_GROUP_ECHO;
1315
1316         rc = obd_connect(NULL, &conn, tgt, &echo_uuid, ocd, NULL);
1317
1318         OBD_FREE(ocd, sizeof(*ocd));
1319
1320         if (rc != 0) {
1321                 CERROR("fail to connect to device %s\n",
1322                        lustre_cfg_string(lcfg, 1));
1323                 return (rc);
1324         }
1325         ec->ec_exp = class_conn2export(&conn);
1326
1327         RETURN(rc);
1328 }
1329
1330 static int echo_client_cleanup(struct obd_device *obddev)
1331 {
1332         struct list_head       *el;
1333         struct ec_object       *eco;
1334         struct echo_client_obd *ec = &obddev->u.echo_client;
1335         int rc;
1336         ENTRY;
1337
1338         if (!list_empty(&obddev->obd_exports)) {
1339                 CERROR("still has clients!\n");
1340                 RETURN(-EBUSY);
1341         }
1342
1343         /* XXX assuming sole access */
1344         while (!list_empty(&ec->ec_objects)) {
1345                 el = ec->ec_objects.next;
1346                 eco = list_entry(el, struct ec_object, eco_obj_chain);
1347
1348                 LASSERT(eco->eco_refcount == 0);
1349                 eco->eco_refcount = 1;
1350                 eco->eco_deleted = 1;
1351                 echo_put_object(eco);
1352         }
1353
1354         rc = obd_disconnect(ec->ec_exp);
1355         if (rc != 0)
1356                 CERROR("fail to disconnect device: %d\n", rc);
1357
1358         RETURN(rc);
1359 }
1360
1361 static int echo_client_connect(const struct lu_env *env,
1362                                struct lustre_handle *conn,
1363                                struct obd_device *src, struct obd_uuid *cluuid,
1364                                struct obd_connect_data *data, void *localdata)
1365 {
1366         struct obd_export *exp;
1367         int                rc;
1368
1369         ENTRY;
1370         rc = class_connect(conn, src, cluuid);
1371         if (rc == 0) {
1372                 exp = class_conn2export(conn);
1373                 CFS_INIT_LIST_HEAD(&exp->exp_ec_data.eced_locks);
1374                 class_export_put(exp);
1375         }
1376
1377         RETURN (rc);
1378 }
1379
1380 static int echo_client_disconnect(struct obd_export *exp)
1381 {
1382         struct obd_device      *obd;
1383         struct echo_client_obd *ec;
1384         struct ec_lock         *ecl;
1385         int                     rc;
1386         ENTRY;
1387
1388         if (exp == NULL)
1389                 GOTO(out, rc = -EINVAL);
1390
1391         obd = exp->exp_obd;
1392         ec = &obd->u.echo_client;
1393
1394         /* no more contention on export's lock list */
1395         while (!list_empty (&exp->exp_ec_data.eced_locks)) {
1396                 ecl = list_entry (exp->exp_ec_data.eced_locks.next,
1397                                   struct ec_lock, ecl_exp_chain);
1398                 list_del (&ecl->ecl_exp_chain);
1399
1400                 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm,
1401                                  ecl->ecl_mode, &ecl->ecl_lock_handle);
1402
1403                 CDEBUG (D_INFO, "Cancel lock on object "LPX64" on disconnect "
1404                         "(%d)\n", ecl->ecl_object->eco_id, rc);
1405
1406                 echo_put_object (ecl->ecl_object);
1407                 OBD_FREE (ecl, sizeof (*ecl));
1408         }
1409
1410         rc = class_disconnect(exp);
1411         GOTO(out, rc);
1412  out:
1413         return rc;
1414 }
1415
1416 static struct obd_ops echo_obd_ops = {
1417         .o_owner       = THIS_MODULE,
1418         .o_setup       = echo_client_setup,
1419         .o_cleanup     = echo_client_cleanup,
1420         .o_iocontrol   = echo_client_iocontrol,
1421         .o_connect     = echo_client_connect,
1422         .o_disconnect  = echo_client_disconnect
1423 };
1424
1425 int echo_client_init(void)
1426 {
1427         struct lprocfs_static_vars lvars = { 0 };
1428
1429         lprocfs_echo_init_vars(&lvars);
1430         return class_register_type(&echo_obd_ops, NULL, lvars.module_vars,
1431                                    LUSTRE_ECHO_CLIENT_NAME, NULL);
1432 }
1433
1434 void echo_client_exit(void)
1435 {
1436         class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
1437 }