lustre/obdecho/echo_client.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  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 <lustre_debug.h>
  48 #include <lprocfs_status.h>
  49 #include <cl_object.h>
  50 #include <lustre_fid.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_net.h>
  53 #include <obd_lov.h>
  54
  55 #include "echo_internal.h"
  56
  57 /** \defgroup echo_client Echo Client
  58  * @{
  59  */
  60
  61 struct echo_device {
  62         struct cl_device        ed_cl;
  63         struct echo_client_obd *ed_ec;
  64
  65         struct cl_site          ed_site_myself;
  66         struct cl_site         *ed_site;
  67         struct lu_device       *ed_next;
  68         int                     ed_next_islov;
  69         int                     ed_next_ismd;
  70         struct lu_client_seq   *ed_cl_seq;
  71 };
  72
  73 struct echo_object {
  74         struct cl_object        eo_cl;
  75         struct cl_object_header eo_hdr;
  76
  77         struct echo_device     *eo_dev;
  78         cfs_list_t              eo_obj_chain;
  79         struct lov_stripe_md   *eo_lsm;
  80         cfs_atomic_t            eo_npages;
  81         int                     eo_deleted;
  82 };
  83
  84 struct echo_object_conf {
  85         struct cl_object_conf  eoc_cl;
  86         struct lov_stripe_md **eoc_md;
  87 };
  88
  89 struct echo_page {
  90         struct cl_page_slice   ep_cl;
  91         struct mutex            ep_lock;
  92         cfs_page_t            *ep_vmpage;
  93 };
  94
  95 struct echo_lock {
  96         struct cl_lock_slice   el_cl;
  97         cfs_list_t             el_chain;
  98         struct echo_object    *el_object;
  99         __u64                  el_cookie;
 100         cfs_atomic_t           el_refcount;
 101 };
 102
 103 struct echo_io {
 104         struct cl_io_slice     ei_cl;
 105 };
 106
 107 #if 0
 108 struct echo_req {
 109         struct cl_req_slice er_cl;
 110 };
 111 #endif
 112
 113 static int echo_client_setup(const struct lu_env *env,
 114                              struct obd_device *obddev,
 115                              struct lustre_cfg *lcfg);
 116 static int echo_client_cleanup(struct obd_device *obddev);
 117
 118
 119 /** \defgroup echo_helpers Helper functions
 120  * @{
 121  */
 122 static inline struct echo_device *cl2echo_dev(const struct cl_device *dev)
 123 {
 124         return container_of0(dev, struct echo_device, ed_cl);
 125 }
 126
 127 static inline struct cl_device *echo_dev2cl(struct echo_device *d)
 128 {
 129         return &d->ed_cl;
 130 }
 131
 132 static inline struct echo_device *obd2echo_dev(const struct obd_device *obd)
 133 {
 134         return cl2echo_dev(lu2cl_dev(obd->obd_lu_dev));
 135 }
 136
 137 static inline struct cl_object *echo_obj2cl(struct echo_object *eco)
 138 {
 139         return &eco->eo_cl;
 140 }
 141
 142 static inline struct echo_object *cl2echo_obj(const struct cl_object *o)
 143 {
 144         return container_of(o, struct echo_object, eo_cl);
 145 }
 146
 147 static inline struct echo_page *cl2echo_page(const struct cl_page_slice *s)
 148 {
 149         return container_of(s, struct echo_page, ep_cl);
 150 }
 151
 152 static inline struct echo_lock *cl2echo_lock(const struct cl_lock_slice *s)
 153 {
 154         return container_of(s, struct echo_lock, el_cl);
 155 }
 156
 157 static inline struct cl_lock *echo_lock2cl(const struct echo_lock *ecl)
 158 {
 159         return ecl->el_cl.cls_lock;
 160 }
 161
 162 static struct lu_context_key echo_thread_key;
 163 static inline struct echo_thread_info *echo_env_info(const struct lu_env *env)
 164 {
 165         struct echo_thread_info *info;
 166         info = lu_context_key_get(&env->le_ctx, &echo_thread_key);
 167         LASSERT(info != NULL);
 168         return info;
 169 }
 170
 171 static inline
 172 struct echo_object_conf *cl2echo_conf(const struct cl_object_conf *c)
 173 {
 174         return container_of(c, struct echo_object_conf, eoc_cl);
 175 }
 176
 177 /** @} echo_helpers */
 178
 179 static struct echo_object *cl_echo_object_find(struct echo_device *d,
 180                                                struct lov_stripe_md **lsm);
 181 static int cl_echo_object_put(struct echo_object *eco);
 182 static int cl_echo_enqueue   (struct echo_object *eco, obd_off start,
 183                               obd_off end, int mode, __u64 *cookie);
 184 static int cl_echo_cancel    (struct echo_device *d, __u64 cookie);
 185 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
 186                               cfs_page_t **pages, int npages, int async);
 187
 188 static struct echo_thread_info *echo_env_info(const struct lu_env *env);
 189
 190 struct echo_thread_info {
 191         struct echo_object_conf eti_conf;
 192         struct lustre_md        eti_md;
 193
 194         struct cl_2queue        eti_queue;
 195         struct cl_io            eti_io;
 196         struct cl_lock_descr    eti_descr;
 197         struct lu_fid           eti_fid;
 198         struct lu_fid           eti_fid2;
 199         struct md_op_spec       eti_spec;
 200         struct lov_mds_md_v3    eti_lmm;
 201         struct lov_user_md_v3   eti_lum;
 202         struct md_attr          eti_ma;
 203         struct lu_name          eti_lname;
 204         /* per-thread values, can be re-used */
 205         void                    *eti_big_lmm;
 206         int                     eti_big_lmmsize;
 207         char                    eti_name[20];
 208         struct lu_buf           eti_buf;
 209         char                    eti_xattr_buf[LUSTRE_POSIX_ACL_MAX_SIZE];
 210 };
 211
 212 /* No session used right now */
 213 struct echo_session_info {
 214         unsigned long dummy;
 215 };
 216
 217 static cfs_mem_cache_t *echo_lock_kmem;
 218 static cfs_mem_cache_t *echo_object_kmem;
 219 static cfs_mem_cache_t *echo_thread_kmem;
 220 static cfs_mem_cache_t *echo_session_kmem;
 221 //static cfs_mem_cache_t *echo_req_kmem;
 222
 223 static struct lu_kmem_descr echo_caches[] = {
 224         {
 225                 .ckd_cache = &echo_lock_kmem,
 226                 .ckd_name  = "echo_lock_kmem",
 227                 .ckd_size  = sizeof (struct echo_lock)
 228         },
 229         {
 230                 .ckd_cache = &echo_object_kmem,
 231                 .ckd_name  = "echo_object_kmem",
 232                 .ckd_size  = sizeof (struct echo_object)
 233         },
 234         {
 235                 .ckd_cache = &echo_thread_kmem,
 236                 .ckd_name  = "echo_thread_kmem",
 237                 .ckd_size  = sizeof (struct echo_thread_info)
 238         },
 239         {
 240                 .ckd_cache = &echo_session_kmem,
 241                 .ckd_name  = "echo_session_kmem",
 242                 .ckd_size  = sizeof (struct echo_session_info)
 243         },
 244 #if 0
 245         {
 246                 .ckd_cache = &echo_req_kmem,
 247                 .ckd_name  = "echo_req_kmem",
 248                 .ckd_size  = sizeof (struct echo_req)
 249         },
 250 #endif
 251         {
 252                 .ckd_cache = NULL
 253         }
 254 };
 255
 256 /** \defgroup echo_page Page operations
 257  *
 258  * Echo page operations.
 259  *
 260  * @{
 261  */
 262 static cfs_page_t *echo_page_vmpage(const struct lu_env *env,
 263                                     const struct cl_page_slice *slice)
 264 {
 265         return cl2echo_page(slice)->ep_vmpage;
 266 }
 267
 268 static int echo_page_own(const struct lu_env *env,
 269                          const struct cl_page_slice *slice,
 270                          struct cl_io *io, int nonblock)
 271 {
 272         struct echo_page *ep = cl2echo_page(slice);
 273
 274         if (!nonblock)
 275                 mutex_lock(&ep->ep_lock);
 276         else if (!mutex_trylock(&ep->ep_lock))
 277                 return -EAGAIN;
 278         return 0;
 279 }
 280
 281 static void echo_page_disown(const struct lu_env *env,
 282                              const struct cl_page_slice *slice,
 283                              struct cl_io *io)
 284 {
 285         struct echo_page *ep = cl2echo_page(slice);
 286
 287         LASSERT(mutex_is_locked(&ep->ep_lock));
 288         mutex_unlock(&ep->ep_lock);
 289 }
 290
 291 static void echo_page_discard(const struct lu_env *env,
 292                               const struct cl_page_slice *slice,
 293                               struct cl_io *unused)
 294 {
 295         cl_page_delete(env, slice->cpl_page);
 296 }
 297
 298 static int echo_page_is_vmlocked(const struct lu_env *env,
 299                                  const struct cl_page_slice *slice)
 300 {
 301         if (mutex_is_locked(&cl2echo_page(slice)->ep_lock))
 302                 return -EBUSY;
 303         return -ENODATA;
 304 }
 305
 306 static void echo_page_completion(const struct lu_env *env,
 307                                  const struct cl_page_slice *slice,
 308                                  int ioret)
 309 {
 310         LASSERT(slice->cpl_page->cp_sync_io != NULL);
 311 }
 312
 313 static void echo_page_fini(const struct lu_env *env,
 314                            struct cl_page_slice *slice)
 315 {
 316         struct echo_page *ep    = cl2echo_page(slice);
 317         struct echo_object *eco = cl2echo_obj(slice->cpl_obj);
 318         cfs_page_t *vmpage      = ep->ep_vmpage;
 319         ENTRY;
 320
 321         cfs_atomic_dec(&eco->eo_npages);
 322         page_cache_release(vmpage);
 323         EXIT;
 324 }
 325
 326 static int echo_page_prep(const struct lu_env *env,
 327                           const struct cl_page_slice *slice,
 328                           struct cl_io *unused)
 329 {
 330         return 0;
 331 }
 332
 333 static int echo_page_print(const struct lu_env *env,
 334                            const struct cl_page_slice *slice,
 335                            void *cookie, lu_printer_t printer)
 336 {
 337         struct echo_page *ep = cl2echo_page(slice);
 338
 339         (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p %d vm@%p\n",
 340                    ep, mutex_is_locked(&ep->ep_lock), ep->ep_vmpage);
 341         return 0;
 342 }
 343
 344 static const struct cl_page_operations echo_page_ops = {
 345         .cpo_own           = echo_page_own,
 346         .cpo_disown        = echo_page_disown,
 347         .cpo_discard       = echo_page_discard,
 348         .cpo_vmpage        = echo_page_vmpage,
 349         .cpo_fini          = echo_page_fini,
 350         .cpo_print         = echo_page_print,
 351         .cpo_is_vmlocked   = echo_page_is_vmlocked,
 352         .io = {
 353                 [CRT_READ] = {
 354                         .cpo_prep        = echo_page_prep,
 355                         .cpo_completion  = echo_page_completion,
 356                 },
 357                 [CRT_WRITE] = {
 358                         .cpo_prep        = echo_page_prep,
 359                         .cpo_completion  = echo_page_completion,
 360                 }
 361         }
 362 };
 363 /** @} echo_page */
 364
 365 /** \defgroup echo_lock Locking
 366  *
 367  * echo lock operations
 368  *
 369  * @{
 370  */
 371 static void echo_lock_fini(const struct lu_env *env,
 372                            struct cl_lock_slice *slice)
 373 {
 374         struct echo_lock *ecl = cl2echo_lock(slice);
 375
 376         LASSERT(cfs_list_empty(&ecl->el_chain));
 377         OBD_SLAB_FREE_PTR(ecl, echo_lock_kmem);
 378 }
 379
 380 static void echo_lock_delete(const struct lu_env *env,
 381                              const struct cl_lock_slice *slice)
 382 {
 383         struct echo_lock *ecl      = cl2echo_lock(slice);
 384
 385         LASSERT(cfs_list_empty(&ecl->el_chain));
 386 }
 387
 388 static int echo_lock_fits_into(const struct lu_env *env,
 389                                const struct cl_lock_slice *slice,
 390                                const struct cl_lock_descr *need,
 391                                const struct cl_io *unused)
 392 {
 393         return 1;
 394 }
 395
 396 static struct cl_lock_operations echo_lock_ops = {
 397         .clo_fini      = echo_lock_fini,
 398         .clo_delete    = echo_lock_delete,
 399         .clo_fits_into = echo_lock_fits_into
 400 };
 401
 402 /** @} echo_lock */
 403
 404 /** \defgroup echo_cl_ops cl_object operations
 405  *
 406  * operations for cl_object
 407  *
 408  * @{
 409  */
 410 static int echo_page_init(const struct lu_env *env, struct cl_object *obj,
 411                         struct cl_page *page, cfs_page_t *vmpage)
 412 {
 413         struct echo_page *ep = cl_object_page_slice(obj, page);
 414         struct echo_object *eco = cl2echo_obj(obj);
 415         ENTRY;
 416
 417         ep->ep_vmpage = vmpage;
 418         page_cache_get(vmpage);
 419         mutex_init(&ep->ep_lock);
 420         cl_page_slice_add(page, &ep->ep_cl, obj, &echo_page_ops);
 421         cfs_atomic_inc(&eco->eo_npages);
 422         RETURN(0);
 423 }
 424
 425 static int echo_io_init(const struct lu_env *env, struct cl_object *obj,
 426                         struct cl_io *io)
 427 {
 428         return 0;
 429 }
 430
 431 static int echo_lock_init(const struct lu_env *env,
 432                           struct cl_object *obj, struct cl_lock *lock,
 433                           const struct cl_io *unused)
 434 {
 435         struct echo_lock *el;
 436         ENTRY;
 437
 438         OBD_SLAB_ALLOC_PTR_GFP(el, echo_lock_kmem, CFS_ALLOC_IO);
 439         if (el != NULL) {
 440                 cl_lock_slice_add(lock, &el->el_cl, obj, &echo_lock_ops);
 441                 el->el_object = cl2echo_obj(obj);
 442                 CFS_INIT_LIST_HEAD(&el->el_chain);
 443                 cfs_atomic_set(&el->el_refcount, 0);
 444         }
 445         RETURN(el == NULL ? -ENOMEM : 0);
 446 }
 447
 448 static int echo_conf_set(const struct lu_env *env, struct cl_object *obj,
 449                          const struct cl_object_conf *conf)
 450 {
 451         return 0;
 452 }
 453
 454 static const struct cl_object_operations echo_cl_obj_ops = {
 455         .coo_page_init = echo_page_init,
 456         .coo_lock_init = echo_lock_init,
 457         .coo_io_init   = echo_io_init,
 458         .coo_conf_set  = echo_conf_set
 459 };
 460 /** @} echo_cl_ops */
 461
 462 /** \defgroup echo_lu_ops lu_object operations
 463  *
 464  * operations for echo lu object.
 465  *
 466  * @{
 467  */
 468 static int echo_object_init(const struct lu_env *env, struct lu_object *obj,
 469                             const struct lu_object_conf *conf)
 470 {
 471         struct echo_device *ed         = cl2echo_dev(lu2cl_dev(obj->lo_dev));
 472         struct echo_client_obd *ec     = ed->ed_ec;
 473         struct echo_object *eco        = cl2echo_obj(lu2cl(obj));
 474         ENTRY;
 475
 476         if (ed->ed_next) {
 477                 struct lu_object  *below;
 478                 struct lu_device  *under;
 479
 480                 under = ed->ed_next;
 481                 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header,
 482                                                         under);
 483                 if (below == NULL)
 484                         RETURN(-ENOMEM);
 485                 lu_object_add(obj, below);
 486         }
 487
 488         if (!ed->ed_next_ismd) {
 489                 const struct cl_object_conf *cconf = lu2cl_conf(conf);
 490                 struct echo_object_conf *econf = cl2echo_conf(cconf);
 491
 492                 LASSERT(econf->eoc_md);
 493                 eco->eo_lsm = *econf->eoc_md;
 494                 /* clear the lsm pointer so that it won't get freed. */
 495                 *econf->eoc_md = NULL;
 496         } else {
 497                 eco->eo_lsm = NULL;
 498         }
 499
 500         eco->eo_dev = ed;
 501         cfs_atomic_set(&eco->eo_npages, 0);
 502         cl_object_page_init(lu2cl(obj), sizeof(struct echo_page));
 503
 504         spin_lock(&ec->ec_lock);
 505         cfs_list_add_tail(&eco->eo_obj_chain, &ec->ec_objects);
 506         spin_unlock(&ec->ec_lock);
 507
 508         RETURN(0);
 509 }
 510
 511 /* taken from osc_unpackmd() */
 512 static int echo_alloc_memmd(struct echo_device *ed,
 513                             struct lov_stripe_md **lsmp)
 514 {
 515         int lsm_size;
 516
 517         ENTRY;
 518
 519         /* If export is lov/osc then use their obd method */
 520         if (ed->ed_next != NULL)
 521                 return obd_alloc_memmd(ed->ed_ec->ec_exp, lsmp);
 522         /* OFD has no unpackmd method, do everything here */
 523         lsm_size = lov_stripe_md_size(1);
 524
 525         LASSERT(*lsmp == NULL);
 526         OBD_ALLOC(*lsmp, lsm_size);
 527         if (*lsmp == NULL)
 528                 RETURN(-ENOMEM);
 529
 530         OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
 531         if ((*lsmp)->lsm_oinfo[0] == NULL) {
 532                 OBD_FREE(*lsmp, lsm_size);
 533                 RETURN(-ENOMEM);
 534         }
 535
 536         loi_init((*lsmp)->lsm_oinfo[0]);
 537         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
 538         ostid_set_seq_echo(&(*lsmp)->lsm_oi);
 539
 540         RETURN(lsm_size);
 541 }
 542
 543 static int echo_free_memmd(struct echo_device *ed, struct lov_stripe_md **lsmp)
 544 {
 545         int lsm_size;
 546
 547         ENTRY;
 548
 549         /* If export is lov/osc then use their obd method */
 550         if (ed->ed_next != NULL)
 551                 return obd_free_memmd(ed->ed_ec->ec_exp, lsmp);
 552         /* OFD has no unpackmd method, do everything here */
 553         lsm_size = lov_stripe_md_size(1);
 554
 555         LASSERT(*lsmp != NULL);
 556         OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
 557         OBD_FREE(*lsmp, lsm_size);
 558         *lsmp = NULL;
 559         RETURN(0);
 560 }
 561
 562 static void echo_object_free(const struct lu_env *env, struct lu_object *obj)
 563 {
 564         struct echo_object *eco    = cl2echo_obj(lu2cl(obj));
 565         struct echo_client_obd *ec = eco->eo_dev->ed_ec;
 566         ENTRY;
 567
 568         LASSERT(cfs_atomic_read(&eco->eo_npages) == 0);
 569
 570         spin_lock(&ec->ec_lock);
 571         cfs_list_del_init(&eco->eo_obj_chain);
 572         spin_unlock(&ec->ec_lock);
 573
 574         lu_object_fini(obj);
 575         lu_object_header_fini(obj->lo_header);
 576
 577         if (eco->eo_lsm)
 578                 echo_free_memmd(eco->eo_dev, &eco->eo_lsm);
 579         OBD_SLAB_FREE_PTR(eco, echo_object_kmem);
 580         EXIT;
 581 }
 582
 583 static int echo_object_print(const struct lu_env *env, void *cookie,
 584                             lu_printer_t p, const struct lu_object *o)
 585 {
 586         struct echo_object *obj = cl2echo_obj(lu2cl(o));
 587
 588         return (*p)(env, cookie, "echoclient-object@%p", obj);
 589 }
 590
 591 static const struct lu_object_operations echo_lu_obj_ops = {
 592         .loo_object_init      = echo_object_init,
 593         .loo_object_delete    = NULL,
 594         .loo_object_release   = NULL,
 595         .loo_object_free      = echo_object_free,
 596         .loo_object_print     = echo_object_print,
 597         .loo_object_invariant = NULL
 598 };
 599 /** @} echo_lu_ops */
 600
 601 /** \defgroup echo_lu_dev_ops  lu_device operations
 602  *
 603  * Operations for echo lu device.
 604  *
 605  * @{
 606  */
 607 static struct lu_object *echo_object_alloc(const struct lu_env *env,
 608                                            const struct lu_object_header *hdr,
 609                                            struct lu_device *dev)
 610 {
 611         struct echo_object *eco;
 612         struct lu_object *obj = NULL;
 613         ENTRY;
 614
 615         /* we're the top dev. */
 616         LASSERT(hdr == NULL);
 617         OBD_SLAB_ALLOC_PTR_GFP(eco, echo_object_kmem, CFS_ALLOC_IO);
 618         if (eco != NULL) {
 619                 struct cl_object_header *hdr = &eco->eo_hdr;
 620
 621                 obj = &echo_obj2cl(eco)->co_lu;
 622                 cl_object_header_init(hdr);
 623                 lu_object_init(obj, &hdr->coh_lu, dev);
 624                 lu_object_add_top(&hdr->coh_lu, obj);
 625
 626                 eco->eo_cl.co_ops = &echo_cl_obj_ops;
 627                 obj->lo_ops       = &echo_lu_obj_ops;
 628         }
 629         RETURN(obj);
 630 }
 631
 632 static struct lu_device_operations echo_device_lu_ops = {
 633         .ldo_object_alloc   = echo_object_alloc,
 634 };
 635
 636 /** @} echo_lu_dev_ops */
 637
 638 static struct cl_device_operations echo_device_cl_ops = {
 639 };
 640
 641 /** \defgroup echo_init Setup and teardown
 642  *
 643  * Init and fini functions for echo client.
 644  *
 645  * @{
 646  */
 647 static int echo_site_init(const struct lu_env *env, struct echo_device *ed)
 648 {
 649         struct cl_site *site = &ed->ed_site_myself;
 650         int rc;
 651
 652         /* initialize site */
 653         rc = cl_site_init(site, &ed->ed_cl);
 654         if (rc) {
 655                 CERROR("Cannot initilize site for echo client(%d)\n", rc);
 656                 return rc;
 657         }
 658
 659         rc = lu_site_init_finish(&site->cs_lu);
 660         if (rc)
 661                 return rc;
 662
 663         ed->ed_site = site;
 664         return 0;
 665 }
 666
 667 static void echo_site_fini(const struct lu_env *env, struct echo_device *ed)
 668 {
 669         if (ed->ed_site) {
 670                 if (!ed->ed_next_ismd)
 671                         cl_site_fini(ed->ed_site);
 672                 ed->ed_site = NULL;
 673         }
 674 }
 675
 676 static void *echo_thread_key_init(const struct lu_context *ctx,
 677                           struct lu_context_key *key)
 678 {
 679         struct echo_thread_info *info;
 680
 681         OBD_SLAB_ALLOC_PTR_GFP(info, echo_thread_kmem, CFS_ALLOC_IO);
 682         if (info == NULL)
 683                 info = ERR_PTR(-ENOMEM);
 684         return info;
 685 }
 686
 687 static void echo_thread_key_fini(const struct lu_context *ctx,
 688                          struct lu_context_key *key, void *data)
 689 {
 690         struct echo_thread_info *info = data;
 691         OBD_SLAB_FREE_PTR(info, echo_thread_kmem);
 692 }
 693
 694 static void echo_thread_key_exit(const struct lu_context *ctx,
 695                          struct lu_context_key *key, void *data)
 696 {
 697 }
 698
 699 static struct lu_context_key echo_thread_key = {
 700         .lct_tags = LCT_CL_THREAD,
 701         .lct_init = echo_thread_key_init,
 702         .lct_fini = echo_thread_key_fini,
 703         .lct_exit = echo_thread_key_exit
 704 };
 705
 706 static void *echo_session_key_init(const struct lu_context *ctx,
 707                                   struct lu_context_key *key)
 708 {
 709         struct echo_session_info *session;
 710
 711         OBD_SLAB_ALLOC_PTR_GFP(session, echo_session_kmem, CFS_ALLOC_IO);
 712         if (session == NULL)
 713                 session = ERR_PTR(-ENOMEM);
 714         return session;
 715 }
 716
 717 static void echo_session_key_fini(const struct lu_context *ctx,
 718                                  struct lu_context_key *key, void *data)
 719 {
 720         struct echo_session_info *session = data;
 721         OBD_SLAB_FREE_PTR(session, echo_session_kmem);
 722 }
 723
 724 static void echo_session_key_exit(const struct lu_context *ctx,
 725                                  struct lu_context_key *key, void *data)
 726 {
 727 }
 728
 729 static struct lu_context_key echo_session_key = {
 730         .lct_tags = LCT_SESSION,
 731         .lct_init = echo_session_key_init,
 732         .lct_fini = echo_session_key_fini,
 733         .lct_exit = echo_session_key_exit
 734 };
 735
 736 LU_TYPE_INIT_FINI(echo, &echo_thread_key, &echo_session_key);
 737
 738 #define ECHO_SEQ_WIDTH 0xffffffff
 739 static int echo_fid_init(struct echo_device *ed, char *obd_name,
 740                          struct seq_server_site *ss)
 741 {
 742         char *prefix;
 743         int rc;
 744         ENTRY;
 745
 746         OBD_ALLOC_PTR(ed->ed_cl_seq);
 747         if (ed->ed_cl_seq == NULL)
 748                 RETURN(-ENOMEM);
 749
 750         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
 751         if (prefix == NULL)
 752                 GOTO(out_free_seq, rc = -ENOMEM);
 753
 754         snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", obd_name);
 755
 756         /* Init client side sequence-manager */
 757         rc = seq_client_init(ed->ed_cl_seq, NULL,
 758                              LUSTRE_SEQ_METADATA,
 759                              prefix, ss->ss_server_seq);
 760         ed->ed_cl_seq->lcs_width = ECHO_SEQ_WIDTH;
 761         OBD_FREE(prefix, MAX_OBD_NAME + 5);
 762         if (rc)
 763                 GOTO(out_free_seq, rc);
 764
 765         RETURN(0);
 766
 767 out_free_seq:
 768         OBD_FREE_PTR(ed->ed_cl_seq);
 769         ed->ed_cl_seq = NULL;
 770         RETURN(rc);
 771 }
 772
 773 static int echo_fid_fini(struct obd_device *obddev)
 774 {
 775         struct echo_device *ed = obd2echo_dev(obddev);
 776         ENTRY;
 777
 778         if (ed->ed_cl_seq != NULL) {
 779                 seq_client_fini(ed->ed_cl_seq);
 780                 OBD_FREE_PTR(ed->ed_cl_seq);
 781                 ed->ed_cl_seq = NULL;
 782         }
 783
 784         RETURN(0);
 785 }
 786
 787 static struct lu_device *echo_device_alloc(const struct lu_env *env,
 788                                            struct lu_device_type *t,
 789                                            struct lustre_cfg *cfg)
 790 {
 791         struct lu_device   *next;
 792         struct echo_device *ed;
 793         struct cl_device   *cd;
 794         struct obd_device  *obd = NULL; /* to keep compiler happy */
 795         struct obd_device  *tgt;
 796         const char *tgt_type_name;
 797         int rc;
 798         int cleanup = 0;
 799         ENTRY;
 800
 801         OBD_ALLOC_PTR(ed);
 802         if (ed == NULL)
 803                 GOTO(out, rc = -ENOMEM);
 804
 805         cleanup = 1;
 806         cd = &ed->ed_cl;
 807         rc = cl_device_init(cd, t);
 808         if (rc)
 809                 GOTO(out, rc);
 810
 811         cd->cd_lu_dev.ld_ops = &echo_device_lu_ops;
 812         cd->cd_ops = &echo_device_cl_ops;
 813
 814         cleanup = 2;
 815         obd = class_name2obd(lustre_cfg_string(cfg, 0));
 816         LASSERT(obd != NULL);
 817         LASSERT(env != NULL);
 818
 819         tgt = class_name2obd(lustre_cfg_string(cfg, 1));
 820         if (tgt == NULL) {
 821                 CERROR("Can not find tgt device %s\n",
 822                         lustre_cfg_string(cfg, 1));
 823                 GOTO(out, rc = -ENODEV);
 824         }
 825
 826         next = tgt->obd_lu_dev;
 827         if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
 828                 ed->ed_next_ismd = 1;
 829         } else {
 830                 ed->ed_next_ismd = 0;
 831                 rc = echo_site_init(env, ed);
 832                 if (rc)
 833                         GOTO(out, rc);
 834         }
 835         cleanup = 3;
 836
 837         rc = echo_client_setup(env, obd, cfg);
 838         if (rc)
 839                 GOTO(out, rc);
 840
 841         ed->ed_ec = &obd->u.echo_client;
 842         cleanup = 4;
 843
 844         if (ed->ed_next_ismd) {
 845                 /* Suppose to connect to some Metadata layer */
 846                 struct lu_site *ls;
 847                 struct lu_device *ld;
 848                 int    found = 0;
 849
 850                 if (next == NULL) {
 851                         CERROR("%s is not lu device type!\n",
 852                                lustre_cfg_string(cfg, 1));
 853                         GOTO(out, rc = -EINVAL);
 854                 }
 855
 856                 tgt_type_name = lustre_cfg_string(cfg, 2);
 857                 if (!tgt_type_name) {
 858                         CERROR("%s no type name for echo %s setup\n",
 859                                 lustre_cfg_string(cfg, 1),
 860                                 tgt->obd_type->typ_name);
 861                         GOTO(out, rc = -EINVAL);
 862                 }
 863
 864                 ls = next->ld_site;
 865
 866                 spin_lock(&ls->ls_ld_lock);
 867                 cfs_list_for_each_entry(ld, &ls->ls_ld_linkage, ld_linkage) {
 868                         if (strcmp(ld->ld_type->ldt_name, tgt_type_name) == 0) {
 869                                 found = 1;
 870                                 break;
 871                         }
 872                 }
 873                 spin_unlock(&ls->ls_ld_lock);
 874
 875                 if (found == 0) {
 876                         CERROR("%s is not lu device type!\n",
 877                                lustre_cfg_string(cfg, 1));
 878                         GOTO(out, rc = -EINVAL);
 879                 }
 880
 881                 next = ld;
 882                 /* For MD echo client, it will use the site in MDS stack */
 883                 ed->ed_site_myself.cs_lu = *ls;
 884                 ed->ed_site = &ed->ed_site_myself;
 885                 ed->ed_cl.cd_lu_dev.ld_site = &ed->ed_site_myself.cs_lu;
 886                 rc = echo_fid_init(ed, obd->obd_name, lu_site2seq(ls));
 887                 if (rc) {
 888                         CERROR("echo fid init error %d\n", rc);
 889                         GOTO(out, rc);
 890                 }
 891         } else {
 892                  /* if echo client is to be stacked upon ost device, the next is
 893                   * NULL since ost is not a clio device so far */
 894                 if (next != NULL && !lu_device_is_cl(next))
 895                         next = NULL;
 896
 897                 tgt_type_name = tgt->obd_type->typ_name;
 898                 if (next != NULL) {
 899                         LASSERT(next != NULL);
 900                         if (next->ld_site != NULL)
 901                                 GOTO(out, rc = -EBUSY);
 902
 903                         next->ld_site = &ed->ed_site->cs_lu;
 904                         rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
 905                                                      next->ld_type->ldt_name,
 906                                                      NULL);
 907                         if (rc)
 908                                 GOTO(out, rc);
 909
 910                         /* Tricky case, I have to determine the obd type since
 911                          * CLIO uses the different parameters to initialize
 912                          * objects for lov & osc. */
 913                         if (strcmp(tgt_type_name, LUSTRE_LOV_NAME) == 0)
 914                                 ed->ed_next_islov = 1;
 915                         else
 916                                 LASSERT(strcmp(tgt_type_name,
 917                                                LUSTRE_OSC_NAME) == 0);
 918                 } else
 919                         LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0);
 920         }
 921
 922         ed->ed_next = next;
 923         RETURN(&cd->cd_lu_dev);
 924 out:
 925         switch(cleanup) {
 926         case 4: {
 927                 int rc2;
 928                 rc2 = echo_client_cleanup(obd);
 929                 if (rc2)
 930                         CERROR("Cleanup obd device %s error(%d)\n",
 931                                obd->obd_name, rc2);
 932         }
 933
 934         case 3:
 935                 echo_site_fini(env, ed);
 936         case 2:
 937                 cl_device_fini(&ed->ed_cl);
 938         case 1:
 939                 OBD_FREE_PTR(ed);
 940         case 0:
 941         default:
 942                 break;
 943         }
 944         return(ERR_PTR(rc));
 945 }
 946
 947 static int echo_device_init(const struct lu_env *env, struct lu_device *d,
 948                           const char *name, struct lu_device *next)
 949 {
 950         LBUG();
 951         return 0;
 952 }
 953
 954 static struct lu_device *echo_device_fini(const struct lu_env *env,
 955                                           struct lu_device *d)
 956 {
 957         struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
 958         struct lu_device *next = ed->ed_next;
 959
 960         while (next && !ed->ed_next_ismd)
 961                 next = next->ld_type->ldt_ops->ldto_device_fini(env, next);
 962         return NULL;
 963 }
 964
 965 static void echo_lock_release(const struct lu_env *env,
 966                               struct echo_lock *ecl,
 967                               int still_used)
 968 {
 969         struct cl_lock *clk = echo_lock2cl(ecl);
 970
 971         cl_lock_get(clk);
 972         cl_unuse(env, clk);
 973         cl_lock_release(env, clk, "ec enqueue", ecl->el_object);
 974         if (!still_used) {
 975                 cl_lock_mutex_get(env, clk);
 976                 cl_lock_cancel(env, clk);
 977                 cl_lock_delete(env, clk);
 978                 cl_lock_mutex_put(env, clk);
 979         }
 980         cl_lock_put(env, clk);
 981 }
 982
 983 static struct lu_device *echo_device_free(const struct lu_env *env,
 984                                           struct lu_device *d)
 985 {
 986         struct echo_device     *ed   = cl2echo_dev(lu2cl_dev(d));
 987         struct echo_client_obd *ec   = ed->ed_ec;
 988         struct echo_object     *eco;
 989         struct lu_device       *next = ed->ed_next;
 990
 991         CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n",
 992                ed, next);
 993
 994         lu_site_purge(env, &ed->ed_site->cs_lu, -1);
 995
 996         /* check if there are objects still alive.
 997          * It shouldn't have any object because lu_site_purge would cleanup
 998          * all of cached objects. Anyway, probably the echo device is being
 999          * parallelly accessed.
1000          */
1001         spin_lock(&ec->ec_lock);
1002         cfs_list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain)
1003                 eco->eo_deleted = 1;
1004         spin_unlock(&ec->ec_lock);
1005
1006         /* purge again */
1007         lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1008
1009         CDEBUG(D_INFO,
1010                "Waiting for the reference of echo object to be dropped\n");
1011
1012         /* Wait for the last reference to be dropped. */
1013         spin_lock(&ec->ec_lock);
1014         while (!cfs_list_empty(&ec->ec_objects)) {
1015                 spin_unlock(&ec->ec_lock);
1016                 CERROR("echo_client still has objects at cleanup time, "
1017                        "wait for 1 second\n");
1018                 cfs_schedule_timeout_and_set_state(CFS_TASK_UNINT,
1019                                                    cfs_time_seconds(1));
1020                 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1021                 spin_lock(&ec->ec_lock);
1022         }
1023         spin_unlock(&ec->ec_lock);
1024
1025         LASSERT(cfs_list_empty(&ec->ec_locks));
1026
1027         CDEBUG(D_INFO, "No object exists, exiting...\n");
1028
1029         echo_client_cleanup(d->ld_obd);
1030         echo_fid_fini(d->ld_obd);
1031         while (next && !ed->ed_next_ismd)
1032                 next = next->ld_type->ldt_ops->ldto_device_free(env, next);
1033
1034         LASSERT(ed->ed_site == lu2cl_site(d->ld_site));
1035         echo_site_fini(env, ed);
1036         cl_device_fini(&ed->ed_cl);
1037         OBD_FREE_PTR(ed);
1038
1039         return NULL;
1040 }
1041
1042 static const struct lu_device_type_operations echo_device_type_ops = {
1043         .ldto_init = echo_type_init,
1044         .ldto_fini = echo_type_fini,
1045
1046         .ldto_start = echo_type_start,
1047         .ldto_stop  = echo_type_stop,
1048
1049         .ldto_device_alloc = echo_device_alloc,
1050         .ldto_device_free  = echo_device_free,
1051         .ldto_device_init  = echo_device_init,
1052         .ldto_device_fini  = echo_device_fini
1053 };
1054
1055 static struct lu_device_type echo_device_type = {
1056         .ldt_tags     = LU_DEVICE_CL,
1057         .ldt_name     = LUSTRE_ECHO_CLIENT_NAME,
1058         .ldt_ops      = &echo_device_type_ops,
1059         .ldt_ctx_tags = LCT_CL_THREAD | LCT_MD_THREAD | LCT_DT_THREAD,
1060 };
1061 /** @} echo_init */
1062
1063 /** \defgroup echo_exports Exported operations
1064  *
1065  * exporting functions to echo client
1066  *
1067  * @{
1068  */
1069
1070 /* Interfaces to echo client obd device */
1071 static struct echo_object *cl_echo_object_find(struct echo_device *d,
1072                                                struct lov_stripe_md **lsmp)
1073 {
1074         struct lu_env *env;
1075         struct echo_thread_info *info;
1076         struct echo_object_conf *conf;
1077         struct lov_stripe_md    *lsm;
1078         struct echo_object *eco;
1079         struct cl_object   *obj;
1080         struct lu_fid *fid;
1081         int refcheck;
1082         ENTRY;
1083
1084         LASSERT(lsmp);
1085         lsm = *lsmp;
1086         LASSERT(lsm);
1087         LASSERT(ostid_id(&lsm->lsm_oi) != 0);
1088         LASSERT(ostid_seq(&lsm->lsm_oi) == FID_SEQ_ECHO);
1089
1090         /* Never return an object if the obd is to be freed. */
1091         if (echo_dev2cl(d)->cd_lu_dev.ld_obd->obd_stopping)
1092                 RETURN(ERR_PTR(-ENODEV));
1093
1094         env = cl_env_get(&refcheck);
1095         if (IS_ERR(env))
1096                 RETURN((void *)env);
1097
1098         info = echo_env_info(env);
1099         conf = &info->eti_conf;
1100         if (d->ed_next) {
1101                 if (!d->ed_next_islov) {
1102                         struct lov_oinfo *oinfo = lsm->lsm_oinfo[0];
1103                         LASSERT(oinfo != NULL);
1104                         oinfo->loi_oi = lsm->lsm_oi;
1105                         conf->eoc_cl.u.coc_oinfo = oinfo;
1106                 } else {
1107                         struct lustre_md *md;
1108                         md = &info->eti_md;
1109                         memset(md, 0, sizeof *md);
1110                         md->lsm = lsm;
1111                         conf->eoc_cl.u.coc_md = md;
1112                 }
1113         }
1114         conf->eoc_md = lsmp;
1115
1116         fid  = &info->eti_fid;
1117         ostid_to_fid(fid, &lsm->lsm_oi, 0);
1118
1119         /* In the function below, .hs_keycmp resolves to
1120          * lu_obj_hop_keycmp() */
1121         /* coverity[overrun-buffer-val] */
1122         obj = cl_object_find(env, echo_dev2cl(d), fid, &conf->eoc_cl);
1123         if (IS_ERR(obj))
1124                 GOTO(out, eco = (void*)obj);
1125
1126         eco = cl2echo_obj(obj);
1127         if (eco->eo_deleted) {
1128                 cl_object_put(env, obj);
1129                 eco = ERR_PTR(-EAGAIN);
1130         }
1131
1132 out:
1133         cl_env_put(env, &refcheck);
1134         RETURN(eco);
1135 }
1136
1137 static int cl_echo_object_put(struct echo_object *eco)
1138 {
1139         struct lu_env *env;
1140         struct cl_object *obj = echo_obj2cl(eco);
1141         int refcheck;
1142         ENTRY;
1143
1144         env = cl_env_get(&refcheck);
1145         if (IS_ERR(env))
1146                 RETURN(PTR_ERR(env));
1147
1148         /* an external function to kill an object? */
1149         if (eco->eo_deleted) {
1150                 struct lu_object_header *loh = obj->co_lu.lo_header;
1151                 LASSERT(&eco->eo_hdr == luh2coh(loh));
1152                 set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags);
1153         }
1154
1155         cl_object_put(env, obj);
1156         cl_env_put(env, &refcheck);
1157         RETURN(0);
1158 }
1159
1160 static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco,
1161                             obd_off start, obd_off end, int mode,
1162                             __u64 *cookie , __u32 enqflags)
1163 {
1164         struct cl_io *io;
1165         struct cl_lock *lck;
1166         struct cl_object *obj;
1167         struct cl_lock_descr *descr;
1168         struct echo_thread_info *info;
1169         int rc = -ENOMEM;
1170         ENTRY;
1171
1172         info = echo_env_info(env);
1173         io = &info->eti_io;
1174         descr = &info->eti_descr;
1175         obj = echo_obj2cl(eco);
1176
1177         descr->cld_obj   = obj;
1178         descr->cld_start = cl_index(obj, start);
1179         descr->cld_end   = cl_index(obj, end);
1180         descr->cld_mode  = mode == LCK_PW ? CLM_WRITE : CLM_READ;
1181         descr->cld_enq_flags = enqflags;
1182         io->ci_obj = obj;
1183
1184         lck = cl_lock_request(env, io, descr, "ec enqueue", eco);
1185         if (lck) {
1186                 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
1187                 struct echo_lock *el;
1188
1189                 rc = cl_wait(env, lck);
1190                 if (rc == 0) {
1191                         el = cl2echo_lock(cl_lock_at(lck, &echo_device_type));
1192                         spin_lock(&ec->ec_lock);
1193                         if (cfs_list_empty(&el->el_chain)) {
1194                                 cfs_list_add(&el->el_chain, &ec->ec_locks);
1195                                 el->el_cookie = ++ec->ec_unique;
1196                         }
1197                         cfs_atomic_inc(&el->el_refcount);
1198                         *cookie = el->el_cookie;
1199                         spin_unlock(&ec->ec_lock);
1200                 } else {
1201                         cl_lock_release(env, lck, "ec enqueue", cfs_current());
1202                 }
1203         }
1204         RETURN(rc);
1205 }
1206
1207 static int cl_echo_enqueue(struct echo_object *eco, obd_off start, obd_off end,
1208                            int mode, __u64 *cookie)
1209 {
1210         struct echo_thread_info *info;
1211         struct lu_env *env;
1212         struct cl_io *io;
1213         int refcheck;
1214         int result;
1215         ENTRY;
1216
1217         env = cl_env_get(&refcheck);
1218         if (IS_ERR(env))
1219                 RETURN(PTR_ERR(env));
1220
1221         info = echo_env_info(env);
1222         io = &info->eti_io;
1223
1224         io->ci_ignore_layout = 1;
1225         result = cl_io_init(env, io, CIT_MISC, echo_obj2cl(eco));
1226         if (result < 0)
1227                 GOTO(out, result);
1228         LASSERT(result == 0);
1229
1230         result = cl_echo_enqueue0(env, eco, start, end, mode, cookie, 0);
1231         cl_io_fini(env, io);
1232
1233         EXIT;
1234 out:
1235         cl_env_put(env, &refcheck);
1236         return result;
1237 }
1238
1239 static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed,
1240                            __u64 cookie)
1241 {
1242         struct echo_client_obd *ec = ed->ed_ec;
1243         struct echo_lock       *ecl = NULL;
1244         cfs_list_t             *el;
1245         int found = 0, still_used = 0;
1246         ENTRY;
1247
1248         LASSERT(ec != NULL);
1249         spin_lock(&ec->ec_lock);
1250         cfs_list_for_each (el, &ec->ec_locks) {
1251                 ecl = cfs_list_entry (el, struct echo_lock, el_chain);
1252                 CDEBUG(D_INFO, "ecl: %p, cookie: "LPX64"\n", ecl, ecl->el_cookie);
1253                 found = (ecl->el_cookie == cookie);
1254                 if (found) {
1255                         if (cfs_atomic_dec_and_test(&ecl->el_refcount))
1256                                 cfs_list_del_init(&ecl->el_chain);
1257                         else
1258                                 still_used = 1;
1259                         break;
1260                 }
1261         }
1262         spin_unlock(&ec->ec_lock);
1263
1264         if (!found)
1265                 RETURN(-ENOENT);
1266
1267         echo_lock_release(env, ecl, still_used);
1268         RETURN(0);
1269 }
1270
1271 static int cl_echo_cancel(struct echo_device *ed, __u64 cookie)
1272 {
1273         struct lu_env *env;
1274         int refcheck;
1275         int rc;
1276         ENTRY;
1277
1278         env = cl_env_get(&refcheck);
1279         if (IS_ERR(env))
1280                 RETURN(PTR_ERR(env));
1281
1282         rc = cl_echo_cancel0(env, ed, cookie);
1283
1284         cl_env_put(env, &refcheck);
1285         RETURN(rc);
1286 }
1287
1288 static int cl_echo_async_brw(const struct lu_env *env, struct cl_io *io,
1289                              enum cl_req_type unused, struct cl_2queue *queue)
1290 {
1291         struct cl_page *clp;
1292         struct cl_page *temp;
1293         int result = 0;
1294         ENTRY;
1295
1296         cl_page_list_for_each_safe(clp, temp, &queue->c2_qin) {
1297                 int rc;
1298                 rc = cl_page_cache_add(env, io, clp, CRT_WRITE);
1299                 if (rc == 0)
1300                         continue;
1301                 result = result ?: rc;
1302         }
1303         RETURN(result);
1304 }
1305
1306 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
1307                               cfs_page_t **pages, int npages, int async)
1308 {
1309         struct lu_env           *env;
1310         struct echo_thread_info *info;
1311         struct cl_object        *obj = echo_obj2cl(eco);
1312         struct echo_device      *ed  = eco->eo_dev;
1313         struct cl_2queue        *queue;
1314         struct cl_io            *io;
1315         struct cl_page          *clp;
1316         struct lustre_handle    lh = { 0 };
1317         int page_size = cl_page_size(obj);
1318         int refcheck;
1319         int rc;
1320         int i;
1321         ENTRY;
1322
1323         LASSERT((offset & ~CFS_PAGE_MASK) == 0);
1324         LASSERT(ed->ed_next != NULL);
1325         env = cl_env_get(&refcheck);
1326         if (IS_ERR(env))
1327                 RETURN(PTR_ERR(env));
1328
1329         info    = echo_env_info(env);
1330         io      = &info->eti_io;
1331         queue   = &info->eti_queue;
1332
1333         cl_2queue_init(queue);
1334
1335         io->ci_ignore_layout = 1;
1336         rc = cl_io_init(env, io, CIT_MISC, obj);
1337         if (rc < 0)
1338                 GOTO(out, rc);
1339         LASSERT(rc == 0);
1340
1341
1342         rc = cl_echo_enqueue0(env, eco, offset,
1343                               offset + npages * CFS_PAGE_SIZE - 1,
1344                               rw == READ ? LCK_PR : LCK_PW, &lh.cookie,
1345                               CEF_NEVER);
1346         if (rc < 0)
1347                 GOTO(error_lock, rc);
1348
1349         for (i = 0; i < npages; i++) {
1350                 LASSERT(pages[i]);
1351                 clp = cl_page_find(env, obj, cl_index(obj, offset),
1352                                    pages[i], CPT_TRANSIENT);
1353                 if (IS_ERR(clp)) {
1354                         rc = PTR_ERR(clp);
1355                         break;
1356                 }
1357                 LASSERT(clp->cp_type == CPT_TRANSIENT);
1358
1359                 rc = cl_page_own(env, io, clp);
1360                 if (rc) {
1361                         LASSERT(clp->cp_state == CPS_FREEING);
1362                         cl_page_put(env, clp);
1363                         break;
1364                 }
1365
1366                 cl_2queue_add(queue, clp);
1367
1368                 /* drop the reference count for cl_page_find, so that the page
1369                  * will be freed in cl_2queue_fini. */
1370                 cl_page_put(env, clp);
1371                 cl_page_clip(env, clp, 0, page_size);
1372
1373                 offset += page_size;
1374         }
1375
1376         if (rc == 0) {
1377                 enum cl_req_type typ = rw == READ ? CRT_READ : CRT_WRITE;
1378
1379                 async = async && (typ == CRT_WRITE);
1380                 if (async)
1381                         rc = cl_echo_async_brw(env, io, typ, queue);
1382                 else
1383                         rc = cl_io_submit_sync(env, io, typ, queue, 0);
1384                 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
1385                        async ? "async" : "sync", rc);
1386         }
1387
1388         cl_echo_cancel0(env, ed, lh.cookie);
1389         EXIT;
1390 error_lock:
1391         cl_2queue_discard(env, io, queue);
1392         cl_2queue_disown(env, io, queue);
1393         cl_2queue_fini(env, queue);
1394         cl_io_fini(env, io);
1395 out:
1396         cl_env_put(env, &refcheck);
1397         return rc;
1398 }
1399 /** @} echo_exports */
1400
1401
1402 static obd_id last_object_id;
1403
1404 static int
1405 echo_copyout_lsm (struct lov_stripe_md *lsm, void *_ulsm, int ulsm_nob)
1406 {
1407         struct lov_stripe_md *ulsm = _ulsm;
1408         int nob, i;
1409
1410         nob = offsetof (struct lov_stripe_md, lsm_oinfo[lsm->lsm_stripe_count]);
1411         if (nob > ulsm_nob)
1412                 return (-EINVAL);
1413
1414         if (cfs_copy_to_user (ulsm, lsm, sizeof(ulsm)))
1415                 return (-EFAULT);
1416
1417         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1418                 if (cfs_copy_to_user (ulsm->lsm_oinfo[i], lsm->lsm_oinfo[i],
1419                                       sizeof(lsm->lsm_oinfo[0])))
1420                         return (-EFAULT);
1421         }
1422         return 0;
1423 }
1424
1425 static int
1426 echo_copyin_lsm (struct echo_device *ed, struct lov_stripe_md *lsm,
1427                  void *ulsm, int ulsm_nob)
1428 {
1429         struct echo_client_obd *ec = ed->ed_ec;
1430         int                     i;
1431
1432         if (ulsm_nob < sizeof (*lsm))
1433                 return (-EINVAL);
1434
1435         if (cfs_copy_from_user (lsm, ulsm, sizeof (*lsm)))
1436                 return (-EFAULT);
1437
1438         if (lsm->lsm_stripe_count > ec->ec_nstripes ||
1439             lsm->lsm_magic != LOV_MAGIC ||
1440             (lsm->lsm_stripe_size & (~CFS_PAGE_MASK)) != 0 ||
1441             ((__u64)lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL))
1442                 return (-EINVAL);
1443
1444
1445         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1446                 if (cfs_copy_from_user(lsm->lsm_oinfo[i],
1447                                        ((struct lov_stripe_md *)ulsm)-> \
1448                                        lsm_oinfo[i],
1449                                        sizeof(lsm->lsm_oinfo[0])))
1450                         return (-EFAULT);
1451         }
1452         return (0);
1453 }
1454
1455 static inline void echo_md_build_name(struct lu_name *lname, char *name,
1456                                       __u64 id)
1457 {
1458         sprintf(name, LPU64, id);
1459         lname->ln_name = name;
1460         lname->ln_namelen = strlen(name);
1461 }
1462
1463 /* similar to mdt_attr_get_complex */
1464 static int echo_big_lmm_get(const struct lu_env *env, struct md_object *o,
1465                             struct md_attr *ma)
1466 {
1467         struct echo_thread_info *info = echo_env_info(env);
1468         int                      rc;
1469
1470         ENTRY;
1471
1472         LASSERT(ma->ma_lmm_size > 0);
1473
1474         rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LOV);
1475         if (rc < 0)
1476                 RETURN(rc);
1477
1478         /* big_lmm may need to be grown */
1479         if (info->eti_big_lmmsize < rc) {
1480                 int size = size_roundup_power2(rc);
1481
1482                 if (info->eti_big_lmmsize > 0) {
1483                         /* free old buffer */
1484                         LASSERT(info->eti_big_lmm);
1485                         OBD_FREE_LARGE(info->eti_big_lmm,
1486                                        info->eti_big_lmmsize);
1487                         info->eti_big_lmm = NULL;
1488                         info->eti_big_lmmsize = 0;
1489                 }
1490
1491                 OBD_ALLOC_LARGE(info->eti_big_lmm, size);
1492                 if (info->eti_big_lmm == NULL)
1493                         RETURN(-ENOMEM);
1494                 info->eti_big_lmmsize = size;
1495         }
1496         LASSERT(info->eti_big_lmmsize >= rc);
1497
1498         info->eti_buf.lb_buf = info->eti_big_lmm;
1499         info->eti_buf.lb_len = info->eti_big_lmmsize;
1500         rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LOV);
1501         if (rc < 0)
1502                 RETURN(rc);
1503
1504         ma->ma_valid |= MA_LOV;
1505         ma->ma_lmm = info->eti_big_lmm;
1506         ma->ma_lmm_size = rc;
1507
1508         RETURN(0);
1509 }
1510
1511 int echo_attr_get_complex(const struct lu_env *env, struct md_object *next,
1512                           struct md_attr *ma)
1513 {
1514         struct echo_thread_info *info = echo_env_info(env);
1515         struct lu_buf           *buf = &info->eti_buf;
1516         cfs_umode_t              mode = lu_object_attr(&next->mo_lu);
1517         int                      need = ma->ma_need;
1518         int                      rc = 0, rc2;
1519
1520         ENTRY;
1521
1522         ma->ma_valid = 0;
1523
1524         if (need & MA_INODE) {
1525                 ma->ma_need = MA_INODE;
1526                 rc = mo_attr_get(env, next, ma);
1527                 if (rc)
1528                         GOTO(out, rc);
1529                 ma->ma_valid |= MA_INODE;
1530         }
1531
1532         if (need & MA_LOV) {
1533                 if (S_ISREG(mode) || S_ISDIR(mode)) {
1534                         LASSERT(ma->ma_lmm_size > 0);
1535                         buf->lb_buf = ma->ma_lmm;
1536                         buf->lb_len = ma->ma_lmm_size;
1537                         rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LOV);
1538                         if (rc2 > 0) {
1539                                 ma->ma_lmm_size = rc2;
1540                                 ma->ma_valid |= MA_LOV;
1541                         } else if (rc2 == -ENODATA) {
1542                                 /* no LOV EA */
1543                                 ma->ma_lmm_size = 0;
1544                         } else if (rc2 == -ERANGE) {
1545                                 rc2 = echo_big_lmm_get(env, next, ma);
1546                                 if (rc2 < 0)
1547                                         GOTO(out, rc = rc2);
1548                         } else {
1549                                 GOTO(out, rc = rc2);
1550                         }
1551                 }
1552         }
1553
1554 #ifdef CONFIG_FS_POSIX_ACL
1555         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1556                 buf->lb_buf = ma->ma_acl;
1557                 buf->lb_len = ma->ma_acl_size;
1558                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1559                 if (rc2 > 0) {
1560                         ma->ma_acl_size = rc2;
1561                         ma->ma_valid |= MA_ACL_DEF;
1562                 } else if (rc2 == -ENODATA) {
1563                         /* no ACLs */
1564                         ma->ma_acl_size = 0;
1565                 } else {
1566                         GOTO(out, rc = rc2);
1567                 }
1568         }
1569 #endif
1570 out:
1571         ma->ma_need = need;
1572         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
1573                rc, ma->ma_valid, ma->ma_lmm);
1574         RETURN(rc);
1575 }
1576
1577 static int
1578 echo_md_create_internal(const struct lu_env *env, struct echo_device *ed,
1579                         struct md_object *parent, struct lu_fid *fid,
1580                         struct lu_name *lname, struct md_op_spec *spec,
1581                         struct md_attr *ma)
1582 {
1583         struct lu_object        *ec_child, *child;
1584         struct lu_device        *ld = ed->ed_next;
1585         struct echo_thread_info *info = echo_env_info(env);
1586         struct lu_fid           *fid2 = &info->eti_fid2;
1587         struct lu_object_conf    conf = { .loc_flags = LOC_F_NEW };
1588         int                      rc;
1589
1590         ENTRY;
1591
1592         rc = mdo_lookup(env, parent, lname, fid2, spec);
1593         if (rc == 0)
1594                 return -EEXIST;
1595         else if (rc != -ENOENT)
1596                 return rc;
1597
1598         ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev,
1599                                      fid, &conf);
1600         if (IS_ERR(ec_child)) {
1601                 CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid),
1602                         PTR_ERR(ec_child));
1603                 RETURN(PTR_ERR(ec_child));
1604         }
1605
1606         child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1607         if (child == NULL) {
1608                 CERROR("Can not locate the child "DFID"\n", PFID(fid));
1609                 GOTO(out_put, rc = -EINVAL);
1610         }
1611
1612         CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n",
1613                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1614
1615         /*
1616          * Do not perform lookup sanity check. We know that name does not exist.
1617          */
1618         spec->sp_cr_lookup = 0;
1619         rc = mdo_create(env, parent, lname, lu2md(child), spec, ma);
1620         if (rc) {
1621                 CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc);
1622                 GOTO(out_put, rc);
1623         }
1624         CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc  = %d\n",
1625                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc);
1626         EXIT;
1627 out_put:
1628         lu_object_put(env, ec_child);
1629         return rc;
1630 }
1631
1632 static int echo_set_lmm_size(const struct lu_env *env, struct lu_device *ld,
1633                              struct md_attr *ma)
1634 {
1635         struct echo_thread_info *info = echo_env_info(env);
1636
1637         if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
1638                 ma->ma_lmm = (void *)&info->eti_lmm;
1639                 ma->ma_lmm_size = sizeof(info->eti_lmm);
1640         } else {
1641                 LASSERT(info->eti_big_lmmsize);
1642                 ma->ma_lmm = info->eti_big_lmm;
1643                 ma->ma_lmm_size = info->eti_big_lmmsize;
1644         }
1645
1646         return 0;
1647 }
1648
1649 static int echo_create_md_object(const struct lu_env *env,
1650                                  struct echo_device *ed,
1651                                  struct lu_object *ec_parent,
1652                                  struct lu_fid *fid,
1653                                  char *name, int namelen,
1654                                  __u64 id, __u32 mode, int count,
1655                                  int stripe_count, int stripe_offset)
1656 {
1657         struct lu_object        *parent;
1658         struct echo_thread_info *info = echo_env_info(env);
1659         struct lu_name          *lname = &info->eti_lname;
1660         struct md_op_spec       *spec = &info->eti_spec;
1661         struct md_attr          *ma = &info->eti_ma;
1662         struct lu_device        *ld = ed->ed_next;
1663         int                      rc = 0;
1664         int                      i;
1665
1666         ENTRY;
1667
1668         if (ec_parent == NULL)
1669                 return -1;
1670         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1671         if (parent == NULL)
1672                 RETURN(-ENXIO);
1673
1674         memset(ma, 0, sizeof(*ma));
1675         memset(spec, 0, sizeof(*spec));
1676         if (stripe_count != 0) {
1677                 spec->sp_cr_flags |= FMODE_WRITE;
1678                 echo_set_lmm_size(env, ld, ma);
1679                 if (stripe_count != -1) {
1680                         struct lov_user_md_v3 *lum = &info->eti_lum;
1681
1682                         lum->lmm_magic = LOV_USER_MAGIC_V3;
1683                         lum->lmm_stripe_count = stripe_count;
1684                         lum->lmm_stripe_offset = stripe_offset;
1685                         lum->lmm_pattern = 0;
1686                         spec->u.sp_ea.eadata = lum;
1687                         spec->u.sp_ea.eadatalen = sizeof(*lum);
1688                         spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1689                 }
1690         }
1691
1692         ma->ma_attr.la_mode = mode;
1693         ma->ma_attr.la_valid = LA_CTIME | LA_MODE;
1694         ma->ma_attr.la_ctime = cfs_time_current_64();
1695
1696         if (name != NULL) {
1697                 lname->ln_name = name;
1698                 lname->ln_namelen = namelen;
1699                 /* If name is specified, only create one object by name */
1700                 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1701                                              spec, ma);
1702                 RETURN(rc);
1703         }
1704
1705         /* Create multiple object sequenced by id */
1706         for (i = 0; i < count; i++) {
1707                 char *tmp_name = info->eti_name;
1708
1709                 echo_md_build_name(lname, tmp_name, id);
1710
1711                 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1712                                              spec, ma);
1713                 if (rc) {
1714                         CERROR("Can not create child %s: rc = %d\n", tmp_name,
1715                                 rc);
1716                         break;
1717                 }
1718                 id++;
1719                 fid->f_oid++;
1720         }
1721
1722         RETURN(rc);
1723 }
1724
1725 static struct lu_object *echo_md_lookup(const struct lu_env *env,
1726                                         struct echo_device *ed,
1727                                         struct md_object *parent,
1728                                         struct lu_name *lname)
1729 {
1730         struct echo_thread_info *info = echo_env_info(env);
1731         struct lu_fid           *fid = &info->eti_fid;
1732         struct lu_object        *child;
1733         int    rc;
1734         ENTRY;
1735
1736         CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name,
1737                PFID(fid), parent);
1738         rc = mdo_lookup(env, parent, lname, fid, NULL);
1739         if (rc) {
1740                 CERROR("lookup %s: rc = %d\n", lname->ln_name, rc);
1741                 RETURN(ERR_PTR(rc));
1742         }
1743
1744         /* In the function below, .hs_keycmp resolves to
1745          * lu_obj_hop_keycmp() */
1746         /* coverity[overrun-buffer-val] */
1747         child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1748
1749         RETURN(child);
1750 }
1751
1752 static int echo_setattr_object(const struct lu_env *env,
1753                                struct echo_device *ed,
1754                                struct lu_object *ec_parent,
1755                                __u64 id, int count)
1756 {
1757         struct lu_object        *parent;
1758         struct echo_thread_info *info = echo_env_info(env);
1759         struct lu_name          *lname = &info->eti_lname;
1760         char                    *name = info->eti_name;
1761         struct lu_device        *ld = ed->ed_next;
1762         struct lu_buf           *buf = &info->eti_buf;
1763         int                      rc = 0;
1764         int                      i;
1765
1766         ENTRY;
1767
1768         if (ec_parent == NULL)
1769                 return -1;
1770         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1771         if (parent == NULL)
1772                 RETURN(-ENXIO);
1773
1774         for (i = 0; i < count; i++) {
1775                 struct lu_object *ec_child, *child;
1776
1777                 echo_md_build_name(lname, name, id);
1778
1779                 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1780                 if (IS_ERR(ec_child)) {
1781                         CERROR("Can't find child %s: rc = %ld\n",
1782                                 lname->ln_name, PTR_ERR(ec_child));
1783                         RETURN(PTR_ERR(ec_child));
1784                 }
1785
1786                 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1787                 if (child == NULL) {
1788                         CERROR("Can not locate the child %s\n", lname->ln_name);
1789                         lu_object_put(env, ec_child);
1790                         rc = -EINVAL;
1791                         break;
1792                 }
1793
1794                 CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n",
1795                        PFID(lu_object_fid(child)));
1796
1797                 buf->lb_buf = info->eti_xattr_buf;
1798                 buf->lb_len = sizeof(info->eti_xattr_buf);
1799
1800                 sprintf(name, "%s.test1", XATTR_USER_PREFIX);
1801                 rc = mo_xattr_set(env, lu2md(child), buf, name,
1802                                   LU_XATTR_CREATE);
1803                 if (rc < 0) {
1804                         CERROR("Can not setattr child "DFID": rc = %d\n",
1805                                 PFID(lu_object_fid(child)), rc);
1806                         lu_object_put(env, ec_child);
1807                         break;
1808                 }
1809                 CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n",
1810                        PFID(lu_object_fid(child)));
1811                 id++;
1812                 lu_object_put(env, ec_child);
1813         }
1814         RETURN(rc);
1815 }
1816
1817 static int echo_getattr_object(const struct lu_env *env,
1818                                struct echo_device *ed,
1819                                struct lu_object *ec_parent,
1820                                __u64 id, int count)
1821 {
1822         struct lu_object        *parent;
1823         struct echo_thread_info *info = echo_env_info(env);
1824         struct lu_name          *lname = &info->eti_lname;
1825         char                    *name = info->eti_name;
1826         struct md_attr          *ma = &info->eti_ma;
1827         struct lu_device        *ld = ed->ed_next;
1828         int                      rc = 0;
1829         int                      i;
1830
1831         ENTRY;
1832
1833         if (ec_parent == NULL)
1834                 return -1;
1835         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1836         if (parent == NULL)
1837                 RETURN(-ENXIO);
1838
1839         memset(ma, 0, sizeof(*ma));
1840         ma->ma_need |= MA_INODE | MA_LOV | MA_PFID | MA_HSM | MA_ACL_DEF;
1841         ma->ma_acl = info->eti_xattr_buf;
1842         ma->ma_acl_size = sizeof(info->eti_xattr_buf);
1843
1844         for (i = 0; i < count; i++) {
1845                 struct lu_object *ec_child, *child;
1846
1847                 ma->ma_valid = 0;
1848                 echo_md_build_name(lname, name, id);
1849                 echo_set_lmm_size(env, ld, ma);
1850
1851                 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1852                 if (IS_ERR(ec_child)) {
1853                         CERROR("Can't find child %s: rc = %ld\n",
1854                                lname->ln_name, PTR_ERR(ec_child));
1855                         RETURN(PTR_ERR(ec_child));
1856                 }
1857
1858                 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1859                 if (child == NULL) {
1860                         CERROR("Can not locate the child %s\n", lname->ln_name);
1861                         lu_object_put(env, ec_child);
1862                         RETURN(-EINVAL);
1863                 }
1864
1865                 CDEBUG(D_RPCTRACE, "Start getattr object "DFID"\n",
1866                        PFID(lu_object_fid(child)));
1867                 rc = echo_attr_get_complex(env, lu2md(child), ma);
1868                 if (rc) {
1869                         CERROR("Can not getattr child "DFID": rc = %d\n",
1870                                 PFID(lu_object_fid(child)), rc);
1871                         lu_object_put(env, ec_child);
1872                         break;
1873                 }
1874                 CDEBUG(D_RPCTRACE, "End getattr object "DFID"\n",
1875                        PFID(lu_object_fid(child)));
1876                 id++;
1877                 lu_object_put(env, ec_child);
1878         }
1879
1880         RETURN(rc);
1881 }
1882
1883 static int echo_lookup_object(const struct lu_env *env,
1884                               struct echo_device *ed,
1885                               struct lu_object *ec_parent,
1886                               __u64 id, int count)
1887 {
1888         struct lu_object        *parent;
1889         struct echo_thread_info *info = echo_env_info(env);
1890         struct lu_name          *lname = &info->eti_lname;
1891         char                    *name = info->eti_name;
1892         struct lu_fid           *fid = &info->eti_fid;
1893         struct lu_device        *ld = ed->ed_next;
1894         int                      rc = 0;
1895         int                      i;
1896
1897         if (ec_parent == NULL)
1898                 return -1;
1899         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1900         if (parent == NULL)
1901                 return -ENXIO;
1902
1903         /*prepare the requests*/
1904         for (i = 0; i < count; i++) {
1905                 echo_md_build_name(lname, name, id);
1906
1907                 CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n",
1908                        PFID(lu_object_fid(parent)), lname->ln_name, parent);
1909
1910                 rc = mdo_lookup(env, lu2md(parent), lname, fid, NULL);
1911                 if (rc) {
1912                         CERROR("Can not lookup child %s: rc = %d\n", name, rc);
1913                         break;
1914                 }
1915                 CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n",
1916                        PFID(lu_object_fid(parent)), lname->ln_name, parent);
1917
1918                 id++;
1919         }
1920         return rc;
1921 }
1922
1923 static int echo_md_destroy_internal(const struct lu_env *env,
1924                                     struct echo_device *ed,
1925                                     struct md_object *parent,
1926                                     struct lu_name *lname,
1927                                     struct md_attr *ma)
1928 {
1929         struct lu_device   *ld = ed->ed_next;
1930         struct lu_object   *ec_child;
1931         struct lu_object   *child;
1932         int                 rc;
1933
1934         ENTRY;
1935
1936         ec_child = echo_md_lookup(env, ed, parent, lname);
1937         if (IS_ERR(ec_child)) {
1938                 CERROR("Can't find child %s: rc = %ld\n", lname->ln_name,
1939                         PTR_ERR(ec_child));
1940                 RETURN(PTR_ERR(ec_child));
1941         }
1942
1943         child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1944         if (child == NULL) {
1945                 CERROR("Can not locate the child %s\n", lname->ln_name);
1946                 GOTO(out_put, rc = -EINVAL);
1947         }
1948
1949         CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n",
1950                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1951
1952         rc = mdo_unlink(env, parent, lu2md(child), lname, ma, 0);
1953         if (rc) {
1954                 CERROR("Can not unlink child %s: rc = %d\n",
1955                         lname->ln_name, rc);
1956                 GOTO(out_put, rc);
1957         }
1958         CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n",
1959                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1960 out_put:
1961         lu_object_put(env, ec_child);
1962         return rc;
1963 }
1964
1965 static int echo_destroy_object(const struct lu_env *env,
1966                                struct echo_device *ed,
1967                                struct lu_object *ec_parent,
1968                                char *name, int namelen,
1969                                __u64 id, __u32 mode,
1970                                int count)
1971 {
1972         struct echo_thread_info *info = echo_env_info(env);
1973         struct lu_name          *lname = &info->eti_lname;
1974         struct md_attr          *ma = &info->eti_ma;
1975         struct lu_device        *ld = ed->ed_next;
1976         struct lu_object        *parent;
1977         int                      rc = 0;
1978         int                      i;
1979         ENTRY;
1980
1981         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1982         if (parent == NULL)
1983                 RETURN(-EINVAL);
1984
1985         memset(ma, 0, sizeof(*ma));
1986         ma->ma_attr.la_mode = mode;
1987         ma->ma_attr.la_valid = LA_CTIME;
1988         ma->ma_attr.la_ctime = cfs_time_current_64();
1989         ma->ma_need = MA_INODE;
1990         ma->ma_valid = 0;
1991
1992         if (name != NULL) {
1993                 lname->ln_name = name;
1994                 lname->ln_namelen = namelen;
1995                 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
1996                                               ma);
1997                 RETURN(rc);
1998         }
1999
2000         /*prepare the requests*/
2001         for (i = 0; i < count; i++) {
2002                 char *tmp_name = info->eti_name;
2003
2004                 ma->ma_valid = 0;
2005                 echo_md_build_name(lname, tmp_name, id);
2006
2007                 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
2008                                               ma);
2009                 if (rc) {
2010                         CERROR("Can not unlink child %s: rc = %d\n", name, rc);
2011                         break;
2012                 }
2013                 id++;
2014         }
2015
2016         RETURN(rc);
2017 }
2018
2019 static struct lu_object *echo_resolve_path(const struct lu_env *env,
2020                                            struct echo_device *ed, char *path,
2021                                            int path_len)
2022 {
2023         struct lu_device        *ld = ed->ed_next;
2024         struct md_device        *md = lu2md_dev(ld);
2025         struct echo_thread_info *info = echo_env_info(env);
2026         struct lu_fid           *fid = &info->eti_fid;
2027         struct lu_name          *lname = &info->eti_lname;
2028         struct lu_object        *parent = NULL;
2029         struct lu_object        *child = NULL;
2030         int rc = 0;
2031         ENTRY;
2032
2033         /*Only support MDD layer right now*/
2034         rc = md->md_ops->mdo_root_get(env, md, fid);
2035         if (rc) {
2036                 CERROR("get root error: rc = %d\n", rc);
2037                 RETURN(ERR_PTR(rc));
2038         }
2039
2040         /* In the function below, .hs_keycmp resolves to
2041          * lu_obj_hop_keycmp() */
2042         /* coverity[overrun-buffer-val] */
2043         parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
2044         if (IS_ERR(parent)) {
2045                 CERROR("Can not find the parent "DFID": rc = %ld\n",
2046                         PFID(fid), PTR_ERR(parent));
2047                 RETURN(parent);
2048         }
2049
2050         while (1) {
2051                 struct lu_object *ld_parent;
2052                 char *e;
2053
2054                 e = strsep(&path, "/");
2055                 if (e == NULL)
2056                         break;
2057
2058                 if (e[0] == 0) {
2059                         if (!path || path[0] == '\0')
2060                                 break;
2061                         continue;
2062                 }
2063
2064                 lname->ln_name = e;
2065                 lname->ln_namelen = strlen(e);
2066
2067                 ld_parent = lu_object_locate(parent->lo_header, ld->ld_type);
2068                 if (ld_parent == NULL) {
2069                         lu_object_put(env, parent);
2070                         rc = -EINVAL;
2071                         break;
2072                 }
2073
2074                 child = echo_md_lookup(env, ed, lu2md(ld_parent), lname);
2075                 lu_object_put(env, parent);
2076                 if (IS_ERR(child)) {
2077                         rc = (int)PTR_ERR(child);
2078                         CERROR("lookup %s under parent "DFID": rc = %d\n",
2079                                 lname->ln_name, PFID(lu_object_fid(ld_parent)),
2080                                 rc);
2081                         break;
2082                 }
2083                 parent = child;
2084         }
2085         if (rc)
2086                 RETURN(ERR_PTR(rc));
2087
2088         RETURN(parent);
2089 }
2090
2091 static void echo_ucred_init(struct lu_env *env)
2092 {
2093         struct lu_ucred *ucred = lu_ucred(env);
2094
2095         ucred->uc_valid = UCRED_INVALID;
2096
2097         ucred->uc_suppgids[0] = -1;
2098         ucred->uc_suppgids[1] = -1;
2099
2100         ucred->uc_uid   = ucred->uc_o_uid   = cfs_curproc_uid();
2101         ucred->uc_gid   = ucred->uc_o_gid   = cfs_curproc_gid();
2102         ucred->uc_fsuid = ucred->uc_o_fsuid = cfs_curproc_fsuid();
2103         ucred->uc_fsgid = ucred->uc_o_fsgid = cfs_curproc_fsgid();
2104         ucred->uc_cap   = cfs_curproc_cap_pack();
2105
2106         /* remove fs privilege for non-root user. */
2107         if (ucred->uc_fsuid)
2108                 ucred->uc_cap &= ~CFS_CAP_FS_MASK;
2109         ucred->uc_valid = UCRED_NEW;
2110 }
2111
2112 static void echo_ucred_fini(struct lu_env *env)
2113 {
2114         struct lu_ucred *ucred = lu_ucred(env);
2115         ucred->uc_valid = UCRED_INIT;
2116 }
2117
2118 #define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_MD_THREAD)
2119 #define ECHO_MD_SES_TAG (LCT_REMEMBER | LCT_SESSION)
2120 static int echo_md_handler(struct echo_device *ed, int command,
2121                            char *path, int path_len, __u64 id, int count,
2122                            struct obd_ioctl_data *data)
2123 {
2124         struct echo_thread_info *info;
2125         struct lu_device      *ld = ed->ed_next;
2126         struct lu_env         *env;
2127         int                    refcheck;
2128         struct lu_object      *parent;
2129         char                  *name = NULL;
2130         int                    namelen = data->ioc_plen2;
2131         int                    rc = 0;
2132         ENTRY;
2133
2134         if (ld == NULL) {
2135                 CERROR("MD echo client is not being initialized properly\n");
2136                 RETURN(-EINVAL);
2137         }
2138
2139         if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
2140                 CERROR("Only support MDD layer right now!\n");
2141                 RETURN(-EINVAL);
2142         }
2143
2144         env = cl_env_get(&refcheck);
2145         if (IS_ERR(env))
2146                 RETURN(PTR_ERR(env));
2147
2148         rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG);
2149         if (rc != 0)
2150                 GOTO(out_env, rc);
2151
2152         /* init big_lmm buffer */
2153         info = echo_env_info(env);
2154         LASSERT(info->eti_big_lmm == NULL);
2155         OBD_ALLOC_LARGE(info->eti_big_lmm, MIN_MD_SIZE);
2156         if (info->eti_big_lmm == NULL)
2157                 GOTO(out_env, rc = -ENOMEM);
2158         info->eti_big_lmmsize = MIN_MD_SIZE;
2159
2160         parent = echo_resolve_path(env, ed, path, path_len);
2161         if (IS_ERR(parent)) {
2162                 CERROR("Can not resolve the path %s: rc = %ld\n", path,
2163                         PTR_ERR(parent));
2164                 GOTO(out_free, rc = PTR_ERR(parent));
2165         }
2166
2167         if (namelen > 0) {
2168                 OBD_ALLOC(name, namelen + 1);
2169                 if (name == NULL)
2170                         GOTO(out_put, rc = -ENOMEM);
2171                 if (cfs_copy_from_user(name, data->ioc_pbuf2, namelen))
2172                         GOTO(out_name, rc = -EFAULT);
2173         }
2174
2175         echo_ucred_init(env);
2176
2177         switch (command) {
2178         case ECHO_MD_CREATE:
2179         case ECHO_MD_MKDIR: {
2180                 struct echo_thread_info *info = echo_env_info(env);
2181                 __u32 mode = data->ioc_obdo2.o_mode;
2182                 struct lu_fid *fid = &info->eti_fid;
2183                 int stripe_count = (int)data->ioc_obdo2.o_misc;
2184                 int stripe_index = (int)data->ioc_obdo2.o_stripe_idx;
2185
2186                 ostid_to_fid(fid, &data->ioc_obdo1.o_oi, 0);
2187                 /* In the function below, .hs_keycmp resolves to
2188                  * lu_obj_hop_keycmp() */
2189                 /* coverity[overrun-buffer-val] */
2190                 rc = echo_create_md_object(env, ed, parent, fid, name, namelen,
2191                                            id, mode, count, stripe_count,
2192                                            stripe_index);
2193                 break;
2194         }
2195         case ECHO_MD_DESTROY:
2196         case ECHO_MD_RMDIR: {
2197                 __u32 mode = data->ioc_obdo2.o_mode;
2198
2199                 rc = echo_destroy_object(env, ed, parent, name, namelen,
2200                                          id, mode, count);
2201                 break;
2202         }
2203         case ECHO_MD_LOOKUP:
2204                 rc = echo_lookup_object(env, ed, parent, id, count);
2205                 break;
2206         case ECHO_MD_GETATTR:
2207                 rc = echo_getattr_object(env, ed, parent, id, count);
2208                 break;
2209         case ECHO_MD_SETATTR:
2210                 rc = echo_setattr_object(env, ed, parent, id, count);
2211                 break;
2212         default:
2213                 CERROR("unknown command %d\n", command);
2214                 rc = -EINVAL;
2215                 break;
2216         }
2217         echo_ucred_fini(env);
2218
2219 out_name:
2220         if (name != NULL)
2221                 OBD_FREE(name, namelen + 1);
2222 out_put:
2223         lu_object_put(env, parent);
2224 out_free:
2225         LASSERT(info->eti_big_lmm);
2226         OBD_FREE_LARGE(info->eti_big_lmm, info->eti_big_lmmsize);
2227         info->eti_big_lmm = NULL;
2228         info->eti_big_lmmsize = 0;
2229 out_env:
2230         cl_env_put(env, &refcheck);
2231         return rc;
2232 }
2233
2234 static int echo_create_object(const struct lu_env *env, struct echo_device *ed,
2235                               int on_target, struct obdo *oa, void *ulsm,
2236                               int ulsm_nob, struct obd_trans_info *oti)
2237 {
2238         struct echo_object     *eco;
2239         struct echo_client_obd *ec = ed->ed_ec;
2240         struct lov_stripe_md   *lsm = NULL;
2241         int                     rc;
2242         int                     created = 0;
2243         ENTRY;
2244
2245         if ((oa->o_valid & OBD_MD_FLID) == 0 && /* no obj id */
2246             (on_target ||                       /* set_stripe */
2247              ec->ec_nstripes != 0)) {           /* LOV */
2248                 CERROR ("No valid oid\n");
2249                 RETURN(-EINVAL);
2250         }
2251
2252         rc = echo_alloc_memmd(ed, &lsm);
2253         if (rc < 0) {
2254                 CERROR("Cannot allocate md: rc = %d\n", rc);
2255                 GOTO(failed, rc);
2256         }
2257
2258         if (ulsm != NULL) {
2259                 int i, idx;
2260
2261                 rc = echo_copyin_lsm (ed, lsm, ulsm, ulsm_nob);
2262                 if (rc != 0)
2263                         GOTO(failed, rc);
2264
2265                 if (lsm->lsm_stripe_count == 0)
2266                         lsm->lsm_stripe_count = ec->ec_nstripes;
2267
2268                 if (lsm->lsm_stripe_size == 0)
2269                         lsm->lsm_stripe_size = CFS_PAGE_SIZE;
2270
2271                 idx = cfs_rand();
2272
2273                 /* setup stripes: indices + default ids if required */
2274                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2275                         if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) == 0)
2276                                 lsm->lsm_oinfo[i]->loi_oi = lsm->lsm_oi;
2277
2278                         lsm->lsm_oinfo[i]->loi_ost_idx =
2279                                 (idx + i) % ec->ec_nstripes;
2280                 }
2281         }
2282
2283         /* setup object ID here for !on_target and LOV hint */
2284         if (oa->o_valid & OBD_MD_FLID) {
2285                 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2286                 lsm->lsm_oi = oa->o_oi;
2287         }
2288
2289         if (ostid_id(&lsm->lsm_oi) == 0)
2290                 ostid_set_id(&lsm->lsm_oi, ++last_object_id);
2291
2292         rc = 0;
2293         if (on_target) {
2294                 /* Only echo objects are allowed to be created */
2295                 LASSERT((oa->o_valid & OBD_MD_FLGROUP) &&
2296                         (ostid_seq(&oa->o_oi) == FID_SEQ_ECHO));
2297                 rc = obd_create(env, ec->ec_exp, oa, &lsm, oti);
2298                 if (rc != 0) {
2299                         CERROR("Cannot create objects: rc = %d\n", rc);
2300                         GOTO(failed, rc);
2301                 }
2302                 created = 1;
2303         }
2304
2305         /* See what object ID we were given */
2306         oa->o_oi = lsm->lsm_oi;
2307         oa->o_valid |= OBD_MD_FLID;
2308
2309         eco = cl_echo_object_find(ed, &lsm);
2310         if (IS_ERR(eco))
2311                 GOTO(failed, rc = PTR_ERR(eco));
2312         cl_echo_object_put(eco);
2313
2314         CDEBUG(D_INFO, "oa oid "DOSTID"\n", POSTID(&oa->o_oi));
2315         EXIT;
2316
2317  failed:
2318         if (created && rc)
2319                 obd_destroy(env, ec->ec_exp, oa, lsm, oti, NULL, NULL);
2320         if (lsm)
2321                 echo_free_memmd(ed, &lsm);
2322         if (rc)
2323                 CERROR("create object failed with: rc = %d\n", rc);
2324         return (rc);
2325 }
2326
2327 static int echo_get_object(struct echo_object **ecop, struct echo_device *ed,
2328                            struct obdo *oa)
2329 {
2330         struct lov_stripe_md   *lsm = NULL;
2331         struct echo_object     *eco;
2332         int                     rc;
2333         ENTRY;
2334
2335         if ((oa->o_valid & OBD_MD_FLID) == 0 || ostid_id(&oa->o_oi) == 0) {
2336                 /* disallow use of object id 0 */
2337                 CERROR ("No valid oid\n");
2338                 RETURN(-EINVAL);
2339         }
2340
2341         rc = echo_alloc_memmd(ed, &lsm);
2342         if (rc < 0)
2343                 RETURN(rc);
2344
2345         lsm->lsm_oi = oa->o_oi;
2346         if (!(oa->o_valid & OBD_MD_FLGROUP))
2347                 ostid_set_seq_echo(&lsm->lsm_oi);
2348
2349         rc = 0;
2350         eco = cl_echo_object_find(ed, &lsm);
2351         if (!IS_ERR(eco))
2352                 *ecop = eco;
2353         else
2354                 rc = PTR_ERR(eco);
2355         if (lsm)
2356                 echo_free_memmd(ed, &lsm);
2357         RETURN(rc);
2358 }
2359
2360 static void echo_put_object(struct echo_object *eco)
2361 {
2362         if (cl_echo_object_put(eco))
2363                 CERROR("echo client: drop an object failed");
2364 }
2365
2366 static void
2367 echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp)
2368 {
2369         unsigned long stripe_count;
2370         unsigned long stripe_size;
2371         unsigned long width;
2372         unsigned long woffset;
2373         int           stripe_index;
2374         obd_off       offset;
2375
2376         if (lsm->lsm_stripe_count <= 1)
2377                 return;
2378
2379         offset       = *offp;
2380         stripe_size  = lsm->lsm_stripe_size;
2381         stripe_count = lsm->lsm_stripe_count;
2382
2383         /* width = # bytes in all stripes */
2384         width = stripe_size * stripe_count;
2385
2386         /* woffset = offset within a width; offset = whole number of widths */
2387         woffset = do_div (offset, width);
2388
2389         stripe_index = woffset / stripe_size;
2390
2391         *idp = ostid_id(&lsm->lsm_oinfo[stripe_index]->loi_oi);
2392         *offp = offset * stripe_size + woffset % stripe_size;
2393 }
2394
2395 static void
2396 echo_client_page_debug_setup(struct lov_stripe_md *lsm,
2397                              cfs_page_t *page, int rw, obd_id id,
2398                              obd_off offset, obd_off count)
2399 {
2400         char    *addr;
2401         obd_off  stripe_off;
2402         obd_id   stripe_id;
2403         int      delta;
2404
2405         /* no partial pages on the client */
2406         LASSERT(count == CFS_PAGE_SIZE);
2407
2408         addr = cfs_kmap(page);
2409
2410         for (delta = 0; delta < CFS_PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2411                 if (rw == OBD_BRW_WRITE) {
2412                         stripe_off = offset + delta;
2413                         stripe_id = id;
2414                         echo_get_stripe_off_id(lsm, &stripe_off, &stripe_id);
2415                 } else {
2416                         stripe_off = 0xdeadbeef00c0ffeeULL;
2417                         stripe_id = 0xdeadbeef00c0ffeeULL;
2418                 }
2419                 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
2420                                   stripe_off, stripe_id);
2421         }
2422
2423         cfs_kunmap(page);
2424 }
2425
2426 static int echo_client_page_debug_check(struct lov_stripe_md *lsm,
2427                                         cfs_page_t *page, obd_id id,
2428                                         obd_off offset, obd_off count)
2429 {
2430         obd_off stripe_off;
2431         obd_id  stripe_id;
2432         char   *addr;
2433         int     delta;
2434         int     rc;
2435         int     rc2;
2436
2437         /* no partial pages on the client */
2438         LASSERT(count == CFS_PAGE_SIZE);
2439
2440         addr = cfs_kmap(page);
2441
2442         for (rc = delta = 0; delta < CFS_PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2443                 stripe_off = offset + delta;
2444                 stripe_id = id;
2445                 echo_get_stripe_off_id (lsm, &stripe_off, &stripe_id);
2446
2447                 rc2 = block_debug_check("test_brw",
2448                                         addr + delta, OBD_ECHO_BLOCK_SIZE,
2449                                         stripe_off, stripe_id);
2450                 if (rc2 != 0) {
2451                         CERROR ("Error in echo object "LPX64"\n", id);
2452                         rc = rc2;
2453                 }
2454         }
2455
2456         cfs_kunmap(page);
2457         return rc;
2458 }
2459
2460 static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa,
2461                             struct echo_object *eco, obd_off offset,
2462                             obd_size count, int async,
2463                             struct obd_trans_info *oti)
2464 {
2465         struct lov_stripe_md   *lsm = eco->eo_lsm;
2466         obd_count               npages;
2467         struct brw_page        *pga;
2468         struct brw_page        *pgp;
2469         cfs_page_t            **pages;
2470         obd_off                 off;
2471         int                     i;
2472         int                     rc;
2473         int                     verify;
2474         int                     gfp_mask;
2475         int                     brw_flags = 0;
2476         ENTRY;
2477
2478         verify = (ostid_id(&oa->o_oi) != ECHO_PERSISTENT_OBJID &&
2479                   (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
2480                   (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
2481
2482         gfp_mask = ((ostid_id(&oa->o_oi) & 2) == 0) ? CFS_ALLOC_STD : CFS_ALLOC_HIGHUSER;
2483
2484         LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
2485         LASSERT(lsm != NULL);
2486         LASSERT(ostid_id(&lsm->lsm_oi) == ostid_id(&oa->o_oi));
2487
2488         if (count <= 0 ||
2489             (count & (~CFS_PAGE_MASK)) != 0)
2490                 RETURN(-EINVAL);
2491
2492         /* XXX think again with misaligned I/O */
2493         npages = count >> CFS_PAGE_SHIFT;
2494
2495         if (rw == OBD_BRW_WRITE)
2496                 brw_flags = OBD_BRW_ASYNC;
2497
2498         OBD_ALLOC(pga, npages * sizeof(*pga));
2499         if (pga == NULL)
2500                 RETURN(-ENOMEM);
2501
2502         OBD_ALLOC(pages, npages * sizeof(*pages));
2503         if (pages == NULL) {
2504                 OBD_FREE(pga, npages * sizeof(*pga));
2505                 RETURN(-ENOMEM);
2506         }
2507
2508         for (i = 0, pgp = pga, off = offset;
2509              i < npages;
2510              i++, pgp++, off += CFS_PAGE_SIZE) {
2511
2512                 LASSERT (pgp->pg == NULL);      /* for cleanup */
2513
2514                 rc = -ENOMEM;
2515                 OBD_PAGE_ALLOC(pgp->pg, gfp_mask);
2516                 if (pgp->pg == NULL)
2517                         goto out;
2518
2519                 pages[i] = pgp->pg;
2520                 pgp->count = CFS_PAGE_SIZE;
2521                 pgp->off = off;
2522                 pgp->flag = brw_flags;
2523
2524                 if (verify)
2525                         echo_client_page_debug_setup(lsm, pgp->pg, rw,
2526                                                      ostid_id(&oa->o_oi), off,
2527                                                      pgp->count);
2528         }
2529
2530         /* brw mode can only be used at client */
2531         LASSERT(ed->ed_next != NULL);
2532         rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async);
2533
2534  out:
2535         if (rc != 0 || rw != OBD_BRW_READ)
2536                 verify = 0;
2537
2538         for (i = 0, pgp = pga; i < npages; i++, pgp++) {
2539                 if (pgp->pg == NULL)
2540                         continue;
2541
2542                 if (verify) {
2543                         int vrc;
2544                         vrc = echo_client_page_debug_check(lsm, pgp->pg,
2545                                                            ostid_id(&oa->o_oi),
2546                                                            pgp->off, pgp->count);
2547                         if (vrc != 0 && rc == 0)
2548                                 rc = vrc;
2549                 }
2550                 OBD_PAGE_FREE(pgp->pg);
2551         }
2552         OBD_FREE(pga, npages * sizeof(*pga));
2553         OBD_FREE(pages, npages * sizeof(*pages));
2554         RETURN(rc);
2555 }
2556
2557 static int echo_client_prep_commit(const struct lu_env *env,
2558                                    struct obd_export *exp, int rw,
2559                                    struct obdo *oa, struct echo_object *eco,
2560                                    obd_off offset, obd_size count,
2561                                    obd_size batch, struct obd_trans_info *oti,
2562                                    int async)
2563 {
2564         struct lov_stripe_md *lsm = eco->eo_lsm;
2565         struct obd_ioobj ioo;
2566         struct niobuf_local *lnb;
2567         struct niobuf_remote *rnb;
2568         obd_off off;
2569         obd_size npages, tot_pages;
2570         int i, ret = 0, brw_flags = 0;
2571
2572         ENTRY;
2573
2574         if (count <= 0 || (count & (~CFS_PAGE_MASK)) != 0 ||
2575             (lsm != NULL && ostid_id(&lsm->lsm_oi) != ostid_id(&oa->o_oi)))
2576                 RETURN(-EINVAL);
2577
2578         npages = batch >> CFS_PAGE_SHIFT;
2579         tot_pages = count >> CFS_PAGE_SHIFT;
2580
2581         OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
2582         OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote));
2583
2584         if (lnb == NULL || rnb == NULL)
2585                 GOTO(out, ret = -ENOMEM);
2586
2587         if (rw == OBD_BRW_WRITE && async)
2588                 brw_flags |= OBD_BRW_ASYNC;
2589
2590         obdo_to_ioobj(oa, &ioo);
2591
2592         off = offset;
2593
2594         for(; tot_pages; tot_pages -= npages) {
2595                 int lpages;
2596
2597                 if (tot_pages < npages)
2598                         npages = tot_pages;
2599
2600                 for (i = 0; i < npages; i++, off += CFS_PAGE_SIZE) {
2601                         rnb[i].offset = off;
2602                         rnb[i].len = CFS_PAGE_SIZE;
2603                         rnb[i].flags = brw_flags;
2604                 }
2605
2606                 ioo.ioo_bufcnt = npages;
2607                 oti->oti_transno = 0;
2608
2609                 lpages = npages;
2610                 ret = obd_preprw(env, rw, exp, oa, 1, &ioo, rnb, &lpages,
2611                                  lnb, oti, NULL);
2612                 if (ret != 0)
2613                         GOTO(out, ret);
2614                 LASSERT(lpages == npages);
2615
2616                 for (i = 0; i < lpages; i++) {
2617                         cfs_page_t *page = lnb[i].page;
2618
2619                         /* read past eof? */
2620                         if (page == NULL && lnb[i].rc == 0)
2621                                 continue;
2622
2623                         if (async)
2624                                 lnb[i].flags |= OBD_BRW_ASYNC;
2625
2626                         if (ostid_id(&oa->o_oi) == ECHO_PERSISTENT_OBJID ||
2627                             (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
2628                             (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
2629                                 continue;
2630
2631                         if (rw == OBD_BRW_WRITE)
2632                                 echo_client_page_debug_setup(lsm, page, rw,
2633                                                             ostid_id(&oa->o_oi),
2634                                                              rnb[i].offset,
2635                                                              rnb[i].len);
2636                         else
2637                                 echo_client_page_debug_check(lsm, page,
2638                                                             ostid_id(&oa->o_oi),
2639                                                              rnb[i].offset,
2640                                                              rnb[i].len);
2641                 }
2642
2643                 ret = obd_commitrw(env, rw, exp, oa, 1, &ioo,
2644                                    rnb, npages, lnb, oti, ret);
2645                 if (ret != 0)
2646                         GOTO(out, ret);
2647
2648                 /* Reset oti otherwise it would confuse ldiskfs. */
2649                 memset(oti, 0, sizeof(*oti));
2650         }
2651
2652 out:
2653         if (lnb)
2654                 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
2655         if (rnb)
2656                 OBD_FREE(rnb, npages * sizeof(struct niobuf_remote));
2657         RETURN(ret);
2658 }
2659
2660 static int echo_client_brw_ioctl(const struct lu_env *env, int rw,
2661                                  struct obd_export *exp,
2662                                  struct obd_ioctl_data *data,
2663                                  struct obd_trans_info *dummy_oti)
2664 {
2665         struct obd_device *obd = class_exp2obd(exp);
2666         struct echo_device *ed = obd2echo_dev(obd);
2667         struct echo_client_obd *ec = ed->ed_ec;
2668         struct obdo *oa = &data->ioc_obdo1;
2669         struct echo_object *eco;
2670         int rc;
2671         int async = 1;
2672         long test_mode;
2673         ENTRY;
2674
2675         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2676
2677         rc = echo_get_object(&eco, ed, oa);
2678         if (rc)
2679                 RETURN(rc);
2680
2681         oa->o_valid &= ~OBD_MD_FLHANDLE;
2682
2683         /* OFD/obdfilter works only via prep/commit */
2684         test_mode = (long)data->ioc_pbuf1;
2685         if (test_mode == 1)
2686                 async = 0;
2687
2688         if (ed->ed_next == NULL && test_mode != 3) {
2689                 test_mode = 3;
2690                 data->ioc_plen1 = data->ioc_count;
2691         }
2692
2693         /* Truncate batch size to maximum */
2694         if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE)
2695                 data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE;
2696
2697         switch (test_mode) {
2698         case 1:
2699                 /* fall through */
2700         case 2:
2701                 rc = echo_client_kbrw(ed, rw, oa,
2702                                       eco, data->ioc_offset,
2703                                       data->ioc_count, async, dummy_oti);
2704                 break;
2705         case 3:
2706                 rc = echo_client_prep_commit(env, ec->ec_exp, rw, oa,
2707                                              eco, data->ioc_offset,
2708                                              data->ioc_count, data->ioc_plen1,
2709                                              dummy_oti, async);
2710                 break;
2711         default:
2712                 rc = -EINVAL;
2713         }
2714         echo_put_object(eco);
2715         RETURN(rc);
2716 }
2717
2718 static int
2719 echo_client_enqueue(struct obd_export *exp, struct obdo *oa,
2720                     int mode, obd_off offset, obd_size nob)
2721 {
2722         struct echo_device     *ed = obd2echo_dev(exp->exp_obd);
2723         struct lustre_handle   *ulh = &oa->o_handle;
2724         struct echo_object     *eco;
2725         obd_off                 end;
2726         int                     rc;
2727         ENTRY;
2728
2729         if (ed->ed_next == NULL)
2730                 RETURN(-EOPNOTSUPP);
2731
2732         if (!(mode == LCK_PR || mode == LCK_PW))
2733                 RETURN(-EINVAL);
2734
2735         if ((offset & (~CFS_PAGE_MASK)) != 0 ||
2736             (nob & (~CFS_PAGE_MASK)) != 0)
2737                 RETURN(-EINVAL);
2738
2739         rc = echo_get_object (&eco, ed, oa);
2740         if (rc != 0)
2741                 RETURN(rc);
2742
2743         end = (nob == 0) ? ((obd_off) -1) : (offset + nob - 1);
2744         rc = cl_echo_enqueue(eco, offset, end, mode, &ulh->cookie);
2745         if (rc == 0) {
2746                 oa->o_valid |= OBD_MD_FLHANDLE;
2747                 CDEBUG(D_INFO, "Cookie is "LPX64"\n", ulh->cookie);
2748         }
2749         echo_put_object(eco);
2750         RETURN(rc);
2751 }
2752
2753 static int
2754 echo_client_cancel(struct obd_export *exp, struct obdo *oa)
2755 {
2756         struct echo_device *ed     = obd2echo_dev(exp->exp_obd);
2757         __u64               cookie = oa->o_handle.cookie;
2758
2759         if ((oa->o_valid & OBD_MD_FLHANDLE) == 0)
2760                 return -EINVAL;
2761
2762         CDEBUG(D_INFO, "Cookie is "LPX64"\n", cookie);
2763         return cl_echo_cancel(ed, cookie);
2764 }
2765
2766 static int
2767 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2768                       void *karg, void *uarg)
2769 {
2770         struct obd_device      *obd = exp->exp_obd;
2771         struct echo_device     *ed = obd2echo_dev(obd);
2772         struct echo_client_obd *ec = ed->ed_ec;
2773         struct echo_object     *eco;
2774         struct obd_ioctl_data  *data = karg;
2775         struct obd_trans_info   dummy_oti;
2776         struct lu_env          *env;
2777         struct oti_req_ack_lock *ack_lock;
2778         struct obdo            *oa;
2779         struct lu_fid           fid;
2780         int                     rw = OBD_BRW_READ;
2781         int                     rc = 0;
2782         int                     i;
2783         ENTRY;
2784
2785         memset(&dummy_oti, 0, sizeof(dummy_oti));
2786
2787         oa = &data->ioc_obdo1;
2788         if (!(oa->o_valid & OBD_MD_FLGROUP)) {
2789                 oa->o_valid |= OBD_MD_FLGROUP;
2790                 ostid_set_seq_echo(&oa->o_oi);
2791         }
2792
2793         /* This FID is unpacked just for validation at this point */
2794         rc = ostid_to_fid(&fid, &oa->o_oi, 0);
2795         if (rc < 0)
2796                 RETURN(rc);
2797
2798         OBD_ALLOC_PTR(env);
2799         if (env == NULL)
2800                 RETURN(-ENOMEM);
2801
2802         rc = lu_env_init(env, LCT_DT_THREAD);
2803         if (rc)
2804                 GOTO(out, rc = -ENOMEM);
2805
2806         switch (cmd) {
2807         case OBD_IOC_CREATE:                    /* may create echo object */
2808                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2809                         GOTO (out, rc = -EPERM);
2810
2811                 rc = echo_create_object(env, ed, 1, oa, data->ioc_pbuf1,
2812                                         data->ioc_plen1, &dummy_oti);
2813                 GOTO(out, rc);
2814
2815         case OBD_IOC_ECHO_MD: {
2816                 int count;
2817                 int cmd;
2818                 char *dir = NULL;
2819                 int dirlen;
2820                 __u64 id;
2821
2822                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2823                         GOTO(out, rc = -EPERM);
2824
2825                 count = data->ioc_count;
2826                 cmd = data->ioc_command;
2827
2828                 id = ostid_id(&data->ioc_obdo2.o_oi);
2829
2830                 dirlen = data->ioc_plen1;
2831                 OBD_ALLOC(dir, dirlen + 1);
2832                 if (dir == NULL)
2833                         GOTO(out, rc = -ENOMEM);
2834
2835                 if (cfs_copy_from_user(dir, data->ioc_pbuf1, dirlen)) {
2836                         OBD_FREE(dir, data->ioc_plen1 + 1);
2837                         GOTO(out, rc = -EFAULT);
2838                 }
2839
2840                 rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data);
2841                 OBD_FREE(dir, dirlen + 1);
2842                 GOTO(out, rc);
2843         }
2844         case OBD_IOC_ECHO_ALLOC_SEQ: {
2845                 struct lu_env   *cl_env;
2846                 int              refcheck;
2847                 __u64            seq;
2848                 int              max_count;
2849
2850                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2851                         GOTO(out, rc = -EPERM);
2852
2853                 cl_env = cl_env_get(&refcheck);
2854                 if (IS_ERR(cl_env))
2855                         GOTO(out, rc = PTR_ERR(cl_env));
2856
2857                 rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG,
2858                                             ECHO_MD_SES_TAG);
2859                 if (rc != 0) {
2860                         cl_env_put(cl_env, &refcheck);
2861                         GOTO(out, rc);
2862                 }
2863
2864                 rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq);
2865                 cl_env_put(cl_env, &refcheck);
2866                 if (rc < 0) {
2867                         CERROR("%s: Can not alloc seq: rc = %d\n",
2868                                obd->obd_name, rc);
2869                         GOTO(out, rc);
2870                 }
2871
2872                 if (cfs_copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1))
2873                         return -EFAULT;
2874
2875                 max_count = LUSTRE_METADATA_SEQ_MAX_WIDTH;
2876                 if (cfs_copy_to_user(data->ioc_pbuf2, &max_count,
2877                                      data->ioc_plen2))
2878                         return -EFAULT;
2879                 GOTO(out, rc);
2880         }
2881         case OBD_IOC_DESTROY:
2882                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2883                         GOTO (out, rc = -EPERM);
2884
2885                 rc = echo_get_object(&eco, ed, oa);
2886                 if (rc == 0) {
2887                         rc = obd_destroy(env, ec->ec_exp, oa, eco->eo_lsm,
2888                                          &dummy_oti, NULL, NULL);
2889                         if (rc == 0)
2890                                 eco->eo_deleted = 1;
2891                         echo_put_object(eco);
2892                 }
2893                 GOTO(out, rc);
2894
2895         case OBD_IOC_GETATTR:
2896                 rc = echo_get_object(&eco, ed, oa);
2897                 if (rc == 0) {
2898                         struct obd_info oinfo = { { { 0 } } };
2899                         oinfo.oi_md = eco->eo_lsm;
2900                         oinfo.oi_oa = oa;
2901                         rc = obd_getattr(env, ec->ec_exp, &oinfo);
2902                         echo_put_object(eco);
2903                 }
2904                 GOTO(out, rc);
2905
2906         case OBD_IOC_SETATTR:
2907                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2908                         GOTO (out, rc = -EPERM);
2909
2910                 rc = echo_get_object(&eco, ed, oa);
2911                 if (rc == 0) {
2912                         struct obd_info oinfo = { { { 0 } } };
2913                         oinfo.oi_oa = oa;
2914                         oinfo.oi_md = eco->eo_lsm;
2915
2916                         rc = obd_setattr(env, ec->ec_exp, &oinfo, NULL);
2917                         echo_put_object(eco);
2918                 }
2919                 GOTO(out, rc);
2920
2921         case OBD_IOC_BRW_WRITE:
2922                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2923                         GOTO (out, rc = -EPERM);
2924
2925                 rw = OBD_BRW_WRITE;
2926                 /* fall through */
2927         case OBD_IOC_BRW_READ:
2928                 rc = echo_client_brw_ioctl(env, rw, exp, data, &dummy_oti);
2929                 GOTO(out, rc);
2930
2931         case ECHO_IOC_GET_STRIPE:
2932                 rc = echo_get_object(&eco, ed, oa);
2933                 if (rc == 0) {
2934                         rc = echo_copyout_lsm(eco->eo_lsm, data->ioc_pbuf1,
2935                                               data->ioc_plen1);
2936                         echo_put_object(eco);
2937                 }
2938                 GOTO(out, rc);
2939
2940         case ECHO_IOC_SET_STRIPE:
2941                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2942                         GOTO (out, rc = -EPERM);
2943
2944                 if (data->ioc_pbuf1 == NULL) {  /* unset */
2945                         rc = echo_get_object(&eco, ed, oa);
2946                         if (rc == 0) {
2947                                 eco->eo_deleted = 1;
2948                                 echo_put_object(eco);
2949                         }
2950                 } else {
2951                         rc = echo_create_object(env, ed, 0, oa,
2952                                                 data->ioc_pbuf1,
2953                                                 data->ioc_plen1, &dummy_oti);
2954                 }
2955                 GOTO (out, rc);
2956
2957         case ECHO_IOC_ENQUEUE:
2958                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2959                         GOTO (out, rc = -EPERM);
2960
2961                 rc = echo_client_enqueue(exp, oa,
2962                                          data->ioc_conn1, /* lock mode */
2963                                          data->ioc_offset,
2964                                          data->ioc_count);/*extent*/
2965                 GOTO (out, rc);
2966
2967         case ECHO_IOC_CANCEL:
2968                 rc = echo_client_cancel(exp, oa);
2969                 GOTO (out, rc);
2970
2971         default:
2972                 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
2973                 GOTO (out, rc = -ENOTTY);
2974         }
2975
2976         EXIT;
2977 out:
2978         lu_env_fini(env);
2979         OBD_FREE_PTR(env);
2980
2981         /* XXX this should be in a helper also called by target_send_reply */
2982         for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4;
2983              i++, ack_lock++) {
2984                 if (!ack_lock->mode)
2985                         break;
2986                 ldlm_lock_decref(&ack_lock->lock, ack_lock->mode);
2987         }
2988
2989         return rc;
2990 }
2991
2992 static int echo_client_setup(const struct lu_env *env,
2993                              struct obd_device *obddev, struct lustre_cfg *lcfg)
2994 {
2995         struct echo_client_obd *ec = &obddev->u.echo_client;
2996         struct obd_device *tgt;
2997         struct obd_uuid echo_uuid = { "ECHO_UUID" };
2998         struct obd_connect_data *ocd = NULL;
2999         int rc;
3000         ENTRY;
3001
3002         if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
3003                 CERROR("requires a TARGET OBD name\n");
3004                 RETURN(-EINVAL);
3005         }
3006
3007         tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
3008         if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
3009                 CERROR("device not attached or not set up (%s)\n",
3010                        lustre_cfg_string(lcfg, 1));
3011                 RETURN(-EINVAL);
3012         }
3013
3014         spin_lock_init(&ec->ec_lock);
3015         CFS_INIT_LIST_HEAD (&ec->ec_objects);
3016         CFS_INIT_LIST_HEAD (&ec->ec_locks);
3017         ec->ec_unique = 0;
3018         ec->ec_nstripes = 0;
3019
3020         if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
3021                 lu_context_tags_update(ECHO_MD_CTX_TAG);
3022                 lu_session_tags_update(ECHO_MD_SES_TAG);
3023                 RETURN(0);
3024         }
3025
3026         OBD_ALLOC(ocd, sizeof(*ocd));
3027         if (ocd == NULL) {
3028                 CERROR("Can't alloc ocd connecting to %s\n",
3029                        lustre_cfg_string(lcfg, 1));
3030                 return -ENOMEM;
3031         }
3032
3033         ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL |
3034                                  OBD_CONNECT_BRW_SIZE |
3035                                  OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 |
3036                                  OBD_CONNECT_64BITHASH | OBD_CONNECT_LVB_TYPE;
3037         ocd->ocd_brw_size = DT_MAX_BRW_SIZE;
3038         ocd->ocd_version = LUSTRE_VERSION_CODE;
3039         ocd->ocd_group = FID_SEQ_ECHO;
3040
3041         rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL);
3042         if (rc == 0) {
3043                 /* Turn off pinger because it connects to tgt obd directly. */
3044                 spin_lock(&tgt->obd_dev_lock);
3045                 cfs_list_del_init(&ec->ec_exp->exp_obd_chain_timed);
3046                 spin_unlock(&tgt->obd_dev_lock);
3047         }
3048
3049         OBD_FREE(ocd, sizeof(*ocd));
3050
3051         if (rc != 0) {
3052                 CERROR("fail to connect to device %s\n",
3053                        lustre_cfg_string(lcfg, 1));
3054                 return (rc);
3055         }
3056
3057         RETURN(rc);
3058 }
3059
3060 static int echo_client_cleanup(struct obd_device *obddev)
3061 {
3062         struct echo_device *ed = obd2echo_dev(obddev);
3063         struct echo_client_obd *ec = &obddev->u.echo_client;
3064         int rc;
3065         ENTRY;
3066
3067         /*Do nothing for Metadata echo client*/
3068         if (ed == NULL )
3069                 RETURN(0);
3070
3071         if (ed->ed_next_ismd) {
3072                 lu_context_tags_clear(ECHO_MD_CTX_TAG);
3073                 lu_session_tags_clear(ECHO_MD_SES_TAG);
3074                 RETURN(0);
3075         }
3076
3077         if (!cfs_list_empty(&obddev->obd_exports)) {
3078                 CERROR("still has clients!\n");
3079                 RETURN(-EBUSY);
3080         }
3081
3082         LASSERT(cfs_atomic_read(&ec->ec_exp->exp_refcount) > 0);
3083         rc = obd_disconnect(ec->ec_exp);
3084         if (rc != 0)
3085                 CERROR("fail to disconnect device: %d\n", rc);
3086
3087         RETURN(rc);
3088 }
3089
3090 static int echo_client_connect(const struct lu_env *env,
3091                                struct obd_export **exp,
3092                                struct obd_device *src, struct obd_uuid *cluuid,
3093                                struct obd_connect_data *data, void *localdata)
3094 {
3095         int                rc;
3096         struct lustre_handle conn = { 0 };
3097
3098         ENTRY;
3099         rc = class_connect(&conn, src, cluuid);
3100         if (rc == 0) {
3101                 *exp = class_conn2export(&conn);
3102         }
3103
3104         RETURN (rc);
3105 }
3106
3107 static int echo_client_disconnect(struct obd_export *exp)
3108 {
3109 #if 0
3110         struct obd_device      *obd;
3111         struct echo_client_obd *ec;
3112         struct ec_lock         *ecl;
3113 #endif
3114         int                     rc;
3115         ENTRY;
3116
3117         if (exp == NULL)
3118                 GOTO(out, rc = -EINVAL);
3119
3120 #if 0
3121         obd = exp->exp_obd;
3122         ec = &obd->u.echo_client;
3123
3124         /* no more contention on export's lock list */
3125         while (!cfs_list_empty (&exp->exp_ec_data.eced_locks)) {
3126                 ecl = cfs_list_entry (exp->exp_ec_data.eced_locks.next,
3127                                       struct ec_lock, ecl_exp_chain);
3128                 cfs_list_del (&ecl->ecl_exp_chain);
3129
3130                 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm,
3131                                  ecl->ecl_mode, &ecl->ecl_lock_handle);
3132
3133                 CDEBUG (D_INFO, "Cancel lock on object "LPX64" on disconnect "
3134                         "(%d)\n", ecl->ecl_object->eco_id, rc);
3135
3136                 echo_put_object (ecl->ecl_object);
3137                 OBD_FREE (ecl, sizeof (*ecl));
3138         }
3139 #endif
3140
3141         rc = class_disconnect(exp);
3142         GOTO(out, rc);
3143  out:
3144         return rc;
3145 }
3146
3147 static struct obd_ops echo_client_obd_ops = {
3148         .o_owner       = THIS_MODULE,
3149
3150 #if 0
3151         .o_setup       = echo_client_setup,
3152         .o_cleanup     = echo_client_cleanup,
3153 #endif
3154
3155         .o_iocontrol   = echo_client_iocontrol,
3156         .o_connect     = echo_client_connect,
3157         .o_disconnect  = echo_client_disconnect
3158 };
3159
3160 int echo_client_init(void)
3161 {
3162         struct lprocfs_static_vars lvars = { 0 };
3163         int rc;
3164
3165         lprocfs_echo_init_vars(&lvars);
3166
3167         rc = lu_kmem_init(echo_caches);
3168         if (rc == 0) {
3169                 rc = class_register_type(&echo_client_obd_ops, NULL,
3170                                          lvars.module_vars,
3171                                          LUSTRE_ECHO_CLIENT_NAME,
3172                                          &echo_device_type);
3173                 if (rc)
3174                         lu_kmem_fini(echo_caches);
3175         }
3176         return rc;
3177 }
3178
3179 void echo_client_exit(void)
3180 {
3181         class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
3182         lu_kmem_fini(echo_caches);
3183 }
3184
3185 #ifdef __KERNEL__
3186 static int __init obdecho_init(void)
3187 {
3188         struct lprocfs_static_vars lvars;
3189         int rc;
3190
3191         ENTRY;
3192         LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
3193
3194         LASSERT(CFS_PAGE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
3195
3196         lprocfs_echo_init_vars(&lvars);
3197
3198 # ifdef HAVE_SERVER_SUPPORT
3199         rc = echo_persistent_pages_init();
3200         if (rc != 0)
3201                 goto failed_0;
3202
3203         rc = class_register_type(&echo_obd_ops, NULL, lvars.module_vars,
3204                                  LUSTRE_ECHO_NAME, NULL);
3205         if (rc != 0)
3206                 goto failed_1;
3207 # endif
3208
3209         rc = echo_client_init();
3210
3211 # ifdef HAVE_SERVER_SUPPORT
3212         if (rc == 0)
3213                 RETURN(0);
3214
3215         class_unregister_type(LUSTRE_ECHO_NAME);
3216 failed_1:
3217         echo_persistent_pages_fini();
3218 failed_0:
3219 # endif
3220         RETURN(rc);
3221 }
3222
3223 static void /*__exit*/ obdecho_exit(void)
3224 {
3225         echo_client_exit();
3226
3227 # ifdef HAVE_SERVER_SUPPORT
3228         class_unregister_type(LUSTRE_ECHO_NAME);
3229         echo_persistent_pages_fini();
3230 # endif
3231 }
3232
3233 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3234 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
3235 MODULE_LICENSE("GPL");
3236
3237 cfs_module(obdecho, LUSTRE_VERSION_STRING, obdecho_init, obdecho_exit);
3238 #endif /* __KERNEL__ */
3239
3240 /** @} echo_client */