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, 2013, 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 <md_object.h>
  51 #include <lustre_fid.h>
  52 #include <lustre_acl.h>
  53 #include <lustre_net.h>
  54 #include <obd_lov.h>
  55
  56 #include "echo_internal.h"
  57
  58 /** \defgroup echo_client Echo Client
  59  * @{
  60  */
  61
  62 struct echo_device {
  63         struct cl_device        ed_cl;
  64         struct echo_client_obd *ed_ec;
  65
  66         struct cl_site          ed_site_myself;
  67         struct cl_site         *ed_site;
  68         struct lu_device       *ed_next;
  69         int                     ed_next_islov;
  70         int                     ed_next_ismd;
  71         struct lu_client_seq   *ed_cl_seq;
  72 };
  73
  74 struct echo_object {
  75         struct cl_object        eo_cl;
  76         struct cl_object_header eo_hdr;
  77
  78         struct echo_device     *eo_dev;
  79         cfs_list_t              eo_obj_chain;
  80         struct lov_stripe_md   *eo_lsm;
  81         cfs_atomic_t            eo_npages;
  82         int                     eo_deleted;
  83 };
  84
  85 struct echo_object_conf {
  86         struct cl_object_conf  eoc_cl;
  87         struct lov_stripe_md **eoc_md;
  88 };
  89
  90 struct echo_page {
  91         struct cl_page_slice   ep_cl;
  92         struct mutex            ep_lock;
  93         struct page            *ep_vmpage;
  94 };
  95
  96 struct echo_lock {
  97         struct cl_lock_slice   el_cl;
  98         cfs_list_t             el_chain;
  99         struct echo_object    *el_object;
 100         __u64                  el_cookie;
 101         cfs_atomic_t           el_refcount;
 102 };
 103
 104 struct echo_io {
 105         struct cl_io_slice     ei_cl;
 106 };
 107
 108 #if 0
 109 struct echo_req {
 110         struct cl_req_slice er_cl;
 111 };
 112 #endif
 113
 114 static int echo_client_setup(const struct lu_env *env,
 115                              struct obd_device *obddev,
 116                              struct lustre_cfg *lcfg);
 117 static int echo_client_cleanup(struct obd_device *obddev);
 118
 119
 120 /** \defgroup echo_helpers Helper functions
 121  * @{
 122  */
 123 static inline struct echo_device *cl2echo_dev(const struct cl_device *dev)
 124 {
 125         return container_of0(dev, struct echo_device, ed_cl);
 126 }
 127
 128 static inline struct cl_device *echo_dev2cl(struct echo_device *d)
 129 {
 130         return &d->ed_cl;
 131 }
 132
 133 static inline struct echo_device *obd2echo_dev(const struct obd_device *obd)
 134 {
 135         return cl2echo_dev(lu2cl_dev(obd->obd_lu_dev));
 136 }
 137
 138 static inline struct cl_object *echo_obj2cl(struct echo_object *eco)
 139 {
 140         return &eco->eo_cl;
 141 }
 142
 143 static inline struct echo_object *cl2echo_obj(const struct cl_object *o)
 144 {
 145         return container_of(o, struct echo_object, eo_cl);
 146 }
 147
 148 static inline struct echo_page *cl2echo_page(const struct cl_page_slice *s)
 149 {
 150         return container_of(s, struct echo_page, ep_cl);
 151 }
 152
 153 static inline struct echo_lock *cl2echo_lock(const struct cl_lock_slice *s)
 154 {
 155         return container_of(s, struct echo_lock, el_cl);
 156 }
 157
 158 static inline struct cl_lock *echo_lock2cl(const struct echo_lock *ecl)
 159 {
 160         return ecl->el_cl.cls_lock;
 161 }
 162
 163 static struct lu_context_key echo_thread_key;
 164 static inline struct echo_thread_info *echo_env_info(const struct lu_env *env)
 165 {
 166         struct echo_thread_info *info;
 167         info = lu_context_key_get(&env->le_ctx, &echo_thread_key);
 168         LASSERT(info != NULL);
 169         return info;
 170 }
 171
 172 static inline
 173 struct echo_object_conf *cl2echo_conf(const struct cl_object_conf *c)
 174 {
 175         return container_of(c, struct echo_object_conf, eoc_cl);
 176 }
 177
 178 /** @} echo_helpers */
 179
 180 static struct echo_object *cl_echo_object_find(struct echo_device *d,
 181                                                struct lov_stripe_md **lsm);
 182 static int cl_echo_object_put(struct echo_object *eco);
 183 static int cl_echo_enqueue   (struct echo_object *eco, obd_off start,
 184                               obd_off end, int mode, __u64 *cookie);
 185 static int cl_echo_cancel    (struct echo_device *d, __u64 cookie);
 186 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
 187                               struct page **pages, int npages, int async);
 188
 189 static struct echo_thread_info *echo_env_info(const struct lu_env *env);
 190
 191 struct echo_thread_info {
 192         struct echo_object_conf eti_conf;
 193         struct lustre_md        eti_md;
 194
 195         struct cl_2queue        eti_queue;
 196         struct cl_io            eti_io;
 197         struct cl_lock_descr    eti_descr;
 198         struct lu_fid           eti_fid;
 199         struct lu_fid           eti_fid2;
 200 #ifdef HAVE_SERVER_SUPPORT
 201         struct md_op_spec       eti_spec;
 202         struct lov_mds_md_v3    eti_lmm;
 203         struct lov_user_md_v3   eti_lum;
 204         struct md_attr          eti_ma;
 205         struct lu_name          eti_lname;
 206         /* per-thread values, can be re-used */
 207         void                    *eti_big_lmm;
 208         int                     eti_big_lmmsize;
 209         char                    eti_name[20];
 210         struct lu_buf           eti_buf;
 211         char                    eti_xattr_buf[LUSTRE_POSIX_ACL_MAX_SIZE];
 212 #endif
 213 };
 214
 215 /* No session used right now */
 216 struct echo_session_info {
 217         unsigned long dummy;
 218 };
 219
 220 static struct kmem_cache *echo_lock_kmem;
 221 static struct kmem_cache *echo_object_kmem;
 222 static struct kmem_cache *echo_thread_kmem;
 223 static struct kmem_cache *echo_session_kmem;
 224 /* static struct kmem_cache *echo_req_kmem; */
 225
 226 static struct lu_kmem_descr echo_caches[] = {
 227         {
 228                 .ckd_cache = &echo_lock_kmem,
 229                 .ckd_name  = "echo_lock_kmem",
 230                 .ckd_size  = sizeof (struct echo_lock)
 231         },
 232         {
 233                 .ckd_cache = &echo_object_kmem,
 234                 .ckd_name  = "echo_object_kmem",
 235                 .ckd_size  = sizeof (struct echo_object)
 236         },
 237         {
 238                 .ckd_cache = &echo_thread_kmem,
 239                 .ckd_name  = "echo_thread_kmem",
 240                 .ckd_size  = sizeof (struct echo_thread_info)
 241         },
 242         {
 243                 .ckd_cache = &echo_session_kmem,
 244                 .ckd_name  = "echo_session_kmem",
 245                 .ckd_size  = sizeof (struct echo_session_info)
 246         },
 247 #if 0
 248         {
 249                 .ckd_cache = &echo_req_kmem,
 250                 .ckd_name  = "echo_req_kmem",
 251                 .ckd_size  = sizeof (struct echo_req)
 252         },
 253 #endif
 254         {
 255                 .ckd_cache = NULL
 256         }
 257 };
 258
 259 /** \defgroup echo_page Page operations
 260  *
 261  * Echo page operations.
 262  *
 263  * @{
 264  */
 265 static struct page *echo_page_vmpage(const struct lu_env *env,
 266                                     const struct cl_page_slice *slice)
 267 {
 268         return cl2echo_page(slice)->ep_vmpage;
 269 }
 270
 271 static int echo_page_own(const struct lu_env *env,
 272                          const struct cl_page_slice *slice,
 273                          struct cl_io *io, int nonblock)
 274 {
 275         struct echo_page *ep = cl2echo_page(slice);
 276
 277         if (!nonblock)
 278                 mutex_lock(&ep->ep_lock);
 279         else if (!mutex_trylock(&ep->ep_lock))
 280                 return -EAGAIN;
 281         return 0;
 282 }
 283
 284 static void echo_page_disown(const struct lu_env *env,
 285                              const struct cl_page_slice *slice,
 286                              struct cl_io *io)
 287 {
 288         struct echo_page *ep = cl2echo_page(slice);
 289
 290         LASSERT(mutex_is_locked(&ep->ep_lock));
 291         mutex_unlock(&ep->ep_lock);
 292 }
 293
 294 static void echo_page_discard(const struct lu_env *env,
 295                               const struct cl_page_slice *slice,
 296                               struct cl_io *unused)
 297 {
 298         cl_page_delete(env, slice->cpl_page);
 299 }
 300
 301 static int echo_page_is_vmlocked(const struct lu_env *env,
 302                                  const struct cl_page_slice *slice)
 303 {
 304         if (mutex_is_locked(&cl2echo_page(slice)->ep_lock))
 305                 return -EBUSY;
 306         return -ENODATA;
 307 }
 308
 309 static void echo_page_completion(const struct lu_env *env,
 310                                  const struct cl_page_slice *slice,
 311                                  int ioret)
 312 {
 313         LASSERT(slice->cpl_page->cp_sync_io != NULL);
 314 }
 315
 316 static void echo_page_fini(const struct lu_env *env,
 317                            struct cl_page_slice *slice)
 318 {
 319         struct echo_object *eco = cl2echo_obj(slice->cpl_obj);
 320         ENTRY;
 321
 322         cfs_atomic_dec(&eco->eo_npages);
 323         page_cache_release(cl2echo_page(slice)->ep_vmpage);
 324         EXIT;
 325 }
 326
 327 static int echo_page_prep(const struct lu_env *env,
 328                           const struct cl_page_slice *slice,
 329                           struct cl_io *unused)
 330 {
 331         return 0;
 332 }
 333
 334 static int echo_page_print(const struct lu_env *env,
 335                            const struct cl_page_slice *slice,
 336                            void *cookie, lu_printer_t printer)
 337 {
 338         struct echo_page *ep = cl2echo_page(slice);
 339
 340         (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p %d vm@%p\n",
 341                    ep, mutex_is_locked(&ep->ep_lock), ep->ep_vmpage);
 342         return 0;
 343 }
 344
 345 static const struct cl_page_operations echo_page_ops = {
 346         .cpo_own           = echo_page_own,
 347         .cpo_disown        = echo_page_disown,
 348         .cpo_discard       = echo_page_discard,
 349         .cpo_vmpage        = echo_page_vmpage,
 350         .cpo_fini          = echo_page_fini,
 351         .cpo_print         = echo_page_print,
 352         .cpo_is_vmlocked   = echo_page_is_vmlocked,
 353         .io = {
 354                 [CRT_READ] = {
 355                         .cpo_prep        = echo_page_prep,
 356                         .cpo_completion  = echo_page_completion,
 357                 },
 358                 [CRT_WRITE] = {
 359                         .cpo_prep        = echo_page_prep,
 360                         .cpo_completion  = echo_page_completion,
 361                 }
 362         }
 363 };
 364 /** @} echo_page */
 365
 366 /** \defgroup echo_lock Locking
 367  *
 368  * echo lock operations
 369  *
 370  * @{
 371  */
 372 static void echo_lock_fini(const struct lu_env *env,
 373                            struct cl_lock_slice *slice)
 374 {
 375         struct echo_lock *ecl = cl2echo_lock(slice);
 376
 377         LASSERT(cfs_list_empty(&ecl->el_chain));
 378         OBD_SLAB_FREE_PTR(ecl, echo_lock_kmem);
 379 }
 380
 381 static void echo_lock_delete(const struct lu_env *env,
 382                              const struct cl_lock_slice *slice)
 383 {
 384         struct echo_lock *ecl      = cl2echo_lock(slice);
 385
 386         LASSERT(cfs_list_empty(&ecl->el_chain));
 387 }
 388
 389 static int echo_lock_fits_into(const struct lu_env *env,
 390                                const struct cl_lock_slice *slice,
 391                                const struct cl_lock_descr *need,
 392                                const struct cl_io *unused)
 393 {
 394         return 1;
 395 }
 396
 397 static struct cl_lock_operations echo_lock_ops = {
 398         .clo_fini      = echo_lock_fini,
 399         .clo_delete    = echo_lock_delete,
 400         .clo_fits_into = echo_lock_fits_into
 401 };
 402
 403 /** @} echo_lock */
 404
 405 /** \defgroup echo_cl_ops cl_object operations
 406  *
 407  * operations for cl_object
 408  *
 409  * @{
 410  */
 411 static int echo_page_init(const struct lu_env *env, struct cl_object *obj,
 412                         struct cl_page *page, struct page *vmpage)
 413 {
 414         struct echo_page *ep = cl_object_page_slice(obj, page);
 415         struct echo_object *eco = cl2echo_obj(obj);
 416         ENTRY;
 417
 418         ep->ep_vmpage = vmpage;
 419         page_cache_get(vmpage);
 420         mutex_init(&ep->ep_lock);
 421         cl_page_slice_add(page, &ep->ep_cl, obj, &echo_page_ops);
 422         cfs_atomic_inc(&eco->eo_npages);
 423         RETURN(0);
 424 }
 425
 426 static int echo_io_init(const struct lu_env *env, struct cl_object *obj,
 427                         struct cl_io *io)
 428 {
 429         return 0;
 430 }
 431
 432 static int echo_lock_init(const struct lu_env *env,
 433                           struct cl_object *obj, struct cl_lock *lock,
 434                           const struct cl_io *unused)
 435 {
 436         struct echo_lock *el;
 437         ENTRY;
 438
 439         OBD_SLAB_ALLOC_PTR_GFP(el, echo_lock_kmem, __GFP_IO);
 440         if (el != NULL) {
 441                 cl_lock_slice_add(lock, &el->el_cl, obj, &echo_lock_ops);
 442                 el->el_object = cl2echo_obj(obj);
 443                 CFS_INIT_LIST_HEAD(&el->el_chain);
 444                 cfs_atomic_set(&el->el_refcount, 0);
 445         }
 446         RETURN(el == NULL ? -ENOMEM : 0);
 447 }
 448
 449 static int echo_conf_set(const struct lu_env *env, struct cl_object *obj,
 450                          const struct cl_object_conf *conf)
 451 {
 452         return 0;
 453 }
 454
 455 static const struct cl_object_operations echo_cl_obj_ops = {
 456         .coo_page_init = echo_page_init,
 457         .coo_lock_init = echo_lock_init,
 458         .coo_io_init   = echo_io_init,
 459         .coo_conf_set  = echo_conf_set
 460 };
 461 /** @} echo_cl_ops */
 462
 463 /** \defgroup echo_lu_ops lu_object operations
 464  *
 465  * operations for echo lu object.
 466  *
 467  * @{
 468  */
 469 static int echo_object_init(const struct lu_env *env, struct lu_object *obj,
 470                             const struct lu_object_conf *conf)
 471 {
 472         struct echo_device *ed         = cl2echo_dev(lu2cl_dev(obj->lo_dev));
 473         struct echo_client_obd *ec     = ed->ed_ec;
 474         struct echo_object *eco        = cl2echo_obj(lu2cl(obj));
 475         ENTRY;
 476
 477         if (ed->ed_next) {
 478                 struct lu_object  *below;
 479                 struct lu_device  *under;
 480
 481                 under = ed->ed_next;
 482                 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header,
 483                                                         under);
 484                 if (below == NULL)
 485                         RETURN(-ENOMEM);
 486                 lu_object_add(obj, below);
 487         }
 488
 489         if (!ed->ed_next_ismd) {
 490                 const struct cl_object_conf *cconf = lu2cl_conf(conf);
 491                 struct echo_object_conf *econf = cl2echo_conf(cconf);
 492
 493                 LASSERT(econf->eoc_md);
 494                 eco->eo_lsm = *econf->eoc_md;
 495                 /* clear the lsm pointer so that it won't get freed. */
 496                 *econf->eoc_md = NULL;
 497         } else {
 498                 eco->eo_lsm = NULL;
 499         }
 500
 501         eco->eo_dev = ed;
 502         cfs_atomic_set(&eco->eo_npages, 0);
 503         cl_object_page_init(lu2cl(obj), sizeof(struct echo_page));
 504
 505         spin_lock(&ec->ec_lock);
 506         cfs_list_add_tail(&eco->eo_obj_chain, &ec->ec_objects);
 507         spin_unlock(&ec->ec_lock);
 508
 509         RETURN(0);
 510 }
 511
 512 /* taken from osc_unpackmd() */
 513 static int echo_alloc_memmd(struct echo_device *ed,
 514                             struct lov_stripe_md **lsmp)
 515 {
 516         int lsm_size;
 517
 518         ENTRY;
 519
 520         /* If export is lov/osc then use their obd method */
 521         if (ed->ed_next != NULL)
 522                 return obd_alloc_memmd(ed->ed_ec->ec_exp, lsmp);
 523         /* OFD has no unpackmd method, do everything here */
 524         lsm_size = lov_stripe_md_size(1);
 525
 526         LASSERT(*lsmp == NULL);
 527         OBD_ALLOC(*lsmp, lsm_size);
 528         if (*lsmp == NULL)
 529                 RETURN(-ENOMEM);
 530
 531         OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
 532         if ((*lsmp)->lsm_oinfo[0] == NULL) {
 533                 OBD_FREE(*lsmp, lsm_size);
 534                 RETURN(-ENOMEM);
 535         }
 536
 537         loi_init((*lsmp)->lsm_oinfo[0]);
 538         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
 539         ostid_set_seq_echo(&(*lsmp)->lsm_oi);
 540
 541         RETURN(lsm_size);
 542 }
 543
 544 static int echo_free_memmd(struct echo_device *ed, struct lov_stripe_md **lsmp)
 545 {
 546         int lsm_size;
 547
 548         ENTRY;
 549
 550         /* If export is lov/osc then use their obd method */
 551         if (ed->ed_next != NULL)
 552                 return obd_free_memmd(ed->ed_ec->ec_exp, lsmp);
 553         /* OFD has no unpackmd method, do everything here */
 554         lsm_size = lov_stripe_md_size(1);
 555
 556         LASSERT(*lsmp != NULL);
 557         OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
 558         OBD_FREE(*lsmp, lsm_size);
 559         *lsmp = NULL;
 560         RETURN(0);
 561 }
 562
 563 static void echo_object_free(const struct lu_env *env, struct lu_object *obj)
 564 {
 565         struct echo_object *eco    = cl2echo_obj(lu2cl(obj));
 566         struct echo_client_obd *ec = eco->eo_dev->ed_ec;
 567         ENTRY;
 568
 569         LASSERT(cfs_atomic_read(&eco->eo_npages) == 0);
 570
 571         spin_lock(&ec->ec_lock);
 572         cfs_list_del_init(&eco->eo_obj_chain);
 573         spin_unlock(&ec->ec_lock);
 574
 575         lu_object_fini(obj);
 576         lu_object_header_fini(obj->lo_header);
 577
 578         if (eco->eo_lsm)
 579                 echo_free_memmd(eco->eo_dev, &eco->eo_lsm);
 580         OBD_SLAB_FREE_PTR(eco, echo_object_kmem);
 581         EXIT;
 582 }
 583
 584 static int echo_object_print(const struct lu_env *env, void *cookie,
 585                             lu_printer_t p, const struct lu_object *o)
 586 {
 587         struct echo_object *obj = cl2echo_obj(lu2cl(o));
 588
 589         return (*p)(env, cookie, "echoclient-object@%p", obj);
 590 }
 591
 592 static const struct lu_object_operations echo_lu_obj_ops = {
 593         .loo_object_init      = echo_object_init,
 594         .loo_object_delete    = NULL,
 595         .loo_object_release   = NULL,
 596         .loo_object_free      = echo_object_free,
 597         .loo_object_print     = echo_object_print,
 598         .loo_object_invariant = NULL
 599 };
 600 /** @} echo_lu_ops */
 601
 602 /** \defgroup echo_lu_dev_ops  lu_device operations
 603  *
 604  * Operations for echo lu device.
 605  *
 606  * @{
 607  */
 608 static struct lu_object *echo_object_alloc(const struct lu_env *env,
 609                                            const struct lu_object_header *hdr,
 610                                            struct lu_device *dev)
 611 {
 612         struct echo_object *eco;
 613         struct lu_object *obj = NULL;
 614         ENTRY;
 615
 616         /* we're the top dev. */
 617         LASSERT(hdr == NULL);
 618         OBD_SLAB_ALLOC_PTR_GFP(eco, echo_object_kmem, __GFP_IO);
 619         if (eco != NULL) {
 620                 struct cl_object_header *hdr = &eco->eo_hdr;
 621
 622                 obj = &echo_obj2cl(eco)->co_lu;
 623                 cl_object_header_init(hdr);
 624                 lu_object_init(obj, &hdr->coh_lu, dev);
 625                 lu_object_add_top(&hdr->coh_lu, obj);
 626
 627                 eco->eo_cl.co_ops = &echo_cl_obj_ops;
 628                 obj->lo_ops       = &echo_lu_obj_ops;
 629         }
 630         RETURN(obj);
 631 }
 632
 633 static struct lu_device_operations echo_device_lu_ops = {
 634         .ldo_object_alloc   = echo_object_alloc,
 635 };
 636
 637 /** @} echo_lu_dev_ops */
 638
 639 static struct cl_device_operations echo_device_cl_ops = {
 640 };
 641
 642 /** \defgroup echo_init Setup and teardown
 643  *
 644  * Init and fini functions for echo client.
 645  *
 646  * @{
 647  */
 648 static int echo_site_init(const struct lu_env *env, struct echo_device *ed)
 649 {
 650         struct cl_site *site = &ed->ed_site_myself;
 651         int rc;
 652
 653         /* initialize site */
 654         rc = cl_site_init(site, &ed->ed_cl);
 655         if (rc) {
 656                 CERROR("Cannot initilize site for echo client(%d)\n", rc);
 657                 return rc;
 658         }
 659
 660         rc = lu_site_init_finish(&site->cs_lu);
 661         if (rc)
 662                 return rc;
 663
 664         ed->ed_site = site;
 665         return 0;
 666 }
 667
 668 static void echo_site_fini(const struct lu_env *env, struct echo_device *ed)
 669 {
 670         if (ed->ed_site) {
 671                 if (!ed->ed_next_ismd)
 672                         cl_site_fini(ed->ed_site);
 673                 ed->ed_site = NULL;
 674         }
 675 }
 676
 677 static void *echo_thread_key_init(const struct lu_context *ctx,
 678                           struct lu_context_key *key)
 679 {
 680         struct echo_thread_info *info;
 681
 682         OBD_SLAB_ALLOC_PTR_GFP(info, echo_thread_kmem, __GFP_IO);
 683         if (info == NULL)
 684                 info = ERR_PTR(-ENOMEM);
 685         return info;
 686 }
 687
 688 static void echo_thread_key_fini(const struct lu_context *ctx,
 689                          struct lu_context_key *key, void *data)
 690 {
 691         struct echo_thread_info *info = data;
 692         OBD_SLAB_FREE_PTR(info, echo_thread_kmem);
 693 }
 694
 695 static void echo_thread_key_exit(const struct lu_context *ctx,
 696                          struct lu_context_key *key, void *data)
 697 {
 698 }
 699
 700 static struct lu_context_key echo_thread_key = {
 701         .lct_tags = LCT_CL_THREAD,
 702         .lct_init = echo_thread_key_init,
 703         .lct_fini = echo_thread_key_fini,
 704         .lct_exit = echo_thread_key_exit
 705 };
 706
 707 static void *echo_session_key_init(const struct lu_context *ctx,
 708                                   struct lu_context_key *key)
 709 {
 710         struct echo_session_info *session;
 711
 712         OBD_SLAB_ALLOC_PTR_GFP(session, echo_session_kmem, __GFP_IO);
 713         if (session == NULL)
 714                 session = ERR_PTR(-ENOMEM);
 715         return session;
 716 }
 717
 718 static void echo_session_key_fini(const struct lu_context *ctx,
 719                                  struct lu_context_key *key, void *data)
 720 {
 721         struct echo_session_info *session = data;
 722         OBD_SLAB_FREE_PTR(session, echo_session_kmem);
 723 }
 724
 725 static void echo_session_key_exit(const struct lu_context *ctx,
 726                                  struct lu_context_key *key, void *data)
 727 {
 728 }
 729
 730 static struct lu_context_key echo_session_key = {
 731         .lct_tags = LCT_SESSION,
 732         .lct_init = echo_session_key_init,
 733         .lct_fini = echo_session_key_fini,
 734         .lct_exit = echo_session_key_exit
 735 };
 736
 737 LU_TYPE_INIT_FINI(echo, &echo_thread_key, &echo_session_key);
 738
 739 #ifdef HAVE_SERVER_SUPPORT
 740 # define ECHO_SEQ_WIDTH 0xffffffff
 741 static int echo_fid_init(struct echo_device *ed, char *obd_name,
 742                          struct seq_server_site *ss)
 743 {
 744         char *prefix;
 745         int rc;
 746         ENTRY;
 747
 748         OBD_ALLOC_PTR(ed->ed_cl_seq);
 749         if (ed->ed_cl_seq == NULL)
 750                 RETURN(-ENOMEM);
 751
 752         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
 753         if (prefix == NULL)
 754                 GOTO(out_free_seq, rc = -ENOMEM);
 755
 756         snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", obd_name);
 757
 758         /* Init client side sequence-manager */
 759         rc = seq_client_init(ed->ed_cl_seq, NULL,
 760                              LUSTRE_SEQ_METADATA,
 761                              prefix, ss->ss_server_seq);
 762         ed->ed_cl_seq->lcs_width = ECHO_SEQ_WIDTH;
 763         OBD_FREE(prefix, MAX_OBD_NAME + 5);
 764         if (rc)
 765                 GOTO(out_free_seq, rc);
 766
 767         RETURN(0);
 768
 769 out_free_seq:
 770         OBD_FREE_PTR(ed->ed_cl_seq);
 771         ed->ed_cl_seq = NULL;
 772         RETURN(rc);
 773 }
 774
 775 static int echo_fid_fini(struct obd_device *obddev)
 776 {
 777         struct echo_device *ed = obd2echo_dev(obddev);
 778         ENTRY;
 779
 780         if (ed->ed_cl_seq != NULL) {
 781                 seq_client_fini(ed->ed_cl_seq);
 782                 OBD_FREE_PTR(ed->ed_cl_seq);
 783                 ed->ed_cl_seq = NULL;
 784         }
 785
 786         RETURN(0);
 787 }
 788 #endif /* HAVE_SERVER_SUPPORT */
 789
 790 static struct lu_device *echo_device_alloc(const struct lu_env *env,
 791                                            struct lu_device_type *t,
 792                                            struct lustre_cfg *cfg)
 793 {
 794         struct lu_device   *next;
 795         struct echo_device *ed;
 796         struct cl_device   *cd;
 797         struct obd_device  *obd = NULL; /* to keep compiler happy */
 798         struct obd_device  *tgt;
 799         const char *tgt_type_name;
 800         int rc;
 801         int cleanup = 0;
 802         ENTRY;
 803
 804         OBD_ALLOC_PTR(ed);
 805         if (ed == NULL)
 806                 GOTO(out, rc = -ENOMEM);
 807
 808         cleanup = 1;
 809         cd = &ed->ed_cl;
 810         rc = cl_device_init(cd, t);
 811         if (rc)
 812                 GOTO(out, rc);
 813
 814         cd->cd_lu_dev.ld_ops = &echo_device_lu_ops;
 815         cd->cd_ops = &echo_device_cl_ops;
 816
 817         cleanup = 2;
 818         obd = class_name2obd(lustre_cfg_string(cfg, 0));
 819         LASSERT(obd != NULL);
 820         LASSERT(env != NULL);
 821
 822         tgt = class_name2obd(lustre_cfg_string(cfg, 1));
 823         if (tgt == NULL) {
 824                 CERROR("Can not find tgt device %s\n",
 825                         lustre_cfg_string(cfg, 1));
 826                 GOTO(out, rc = -ENODEV);
 827         }
 828
 829         next = tgt->obd_lu_dev;
 830         if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
 831                 ed->ed_next_ismd = 1;
 832         } else {
 833                 ed->ed_next_ismd = 0;
 834                 rc = echo_site_init(env, ed);
 835                 if (rc)
 836                         GOTO(out, rc);
 837         }
 838         cleanup = 3;
 839
 840         rc = echo_client_setup(env, obd, cfg);
 841         if (rc)
 842                 GOTO(out, rc);
 843
 844         ed->ed_ec = &obd->u.echo_client;
 845         cleanup = 4;
 846
 847         if (ed->ed_next_ismd) {
 848 #ifdef HAVE_SERVER_SUPPORT
 849                 /* Suppose to connect to some Metadata layer */
 850                 struct lu_site *ls;
 851                 struct lu_device *ld;
 852                 int    found = 0;
 853
 854                 if (next == NULL) {
 855                         CERROR("%s is not lu device type!\n",
 856                                lustre_cfg_string(cfg, 1));
 857                         GOTO(out, rc = -EINVAL);
 858                 }
 859
 860                 tgt_type_name = lustre_cfg_string(cfg, 2);
 861                 if (!tgt_type_name) {
 862                         CERROR("%s no type name for echo %s setup\n",
 863                                 lustre_cfg_string(cfg, 1),
 864                                 tgt->obd_type->typ_name);
 865                         GOTO(out, rc = -EINVAL);
 866                 }
 867
 868                 ls = next->ld_site;
 869
 870                 spin_lock(&ls->ls_ld_lock);
 871                 cfs_list_for_each_entry(ld, &ls->ls_ld_linkage, ld_linkage) {
 872                         if (strcmp(ld->ld_type->ldt_name, tgt_type_name) == 0) {
 873                                 found = 1;
 874                                 break;
 875                         }
 876                 }
 877                 spin_unlock(&ls->ls_ld_lock);
 878
 879                 if (found == 0) {
 880                         CERROR("%s is not lu device type!\n",
 881                                lustre_cfg_string(cfg, 1));
 882                         GOTO(out, rc = -EINVAL);
 883                 }
 884
 885                 next = ld;
 886                 /* For MD echo client, it will use the site in MDS stack */
 887                 ed->ed_site_myself.cs_lu = *ls;
 888                 ed->ed_site = &ed->ed_site_myself;
 889                 ed->ed_cl.cd_lu_dev.ld_site = &ed->ed_site_myself.cs_lu;
 890                 rc = echo_fid_init(ed, obd->obd_name, lu_site2seq(ls));
 891                 if (rc) {
 892                         CERROR("echo fid init error %d\n", rc);
 893                         GOTO(out, rc);
 894                 }
 895 #else /* !HAVE_SERVER_SUPPORT */
 896                 CERROR("Local operations are NOT supported on client side. "
 897                        "Only remote operations are supported. Metadata client "
 898                        "must be run on server side.\n");
 899                 GOTO(out, rc = -EOPNOTSUPP);
 900 #endif
 901         } else {
 902                  /* if echo client is to be stacked upon ost device, the next is
 903                   * NULL since ost is not a clio device so far */
 904                 if (next != NULL && !lu_device_is_cl(next))
 905                         next = NULL;
 906
 907                 tgt_type_name = tgt->obd_type->typ_name;
 908                 if (next != NULL) {
 909                         LASSERT(next != NULL);
 910                         if (next->ld_site != NULL)
 911                                 GOTO(out, rc = -EBUSY);
 912
 913                         next->ld_site = &ed->ed_site->cs_lu;
 914                         rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
 915                                                      next->ld_type->ldt_name,
 916                                                      NULL);
 917                         if (rc)
 918                                 GOTO(out, rc);
 919
 920                         /* Tricky case, I have to determine the obd type since
 921                          * CLIO uses the different parameters to initialize
 922                          * objects for lov & osc. */
 923                         if (strcmp(tgt_type_name, LUSTRE_LOV_NAME) == 0)
 924                                 ed->ed_next_islov = 1;
 925                         else
 926                                 LASSERT(strcmp(tgt_type_name,
 927                                                LUSTRE_OSC_NAME) == 0);
 928                 } else
 929                         LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0);
 930         }
 931
 932         ed->ed_next = next;
 933         RETURN(&cd->cd_lu_dev);
 934 out:
 935         switch(cleanup) {
 936         case 4: {
 937                 int rc2;
 938                 rc2 = echo_client_cleanup(obd);
 939                 if (rc2)
 940                         CERROR("Cleanup obd device %s error(%d)\n",
 941                                obd->obd_name, rc2);
 942         }
 943
 944         case 3:
 945                 echo_site_fini(env, ed);
 946         case 2:
 947                 cl_device_fini(&ed->ed_cl);
 948         case 1:
 949                 OBD_FREE_PTR(ed);
 950         case 0:
 951         default:
 952                 break;
 953         }
 954         return(ERR_PTR(rc));
 955 }
 956
 957 static int echo_device_init(const struct lu_env *env, struct lu_device *d,
 958                           const char *name, struct lu_device *next)
 959 {
 960         LBUG();
 961         return 0;
 962 }
 963
 964 static struct lu_device *echo_device_fini(const struct lu_env *env,
 965                                           struct lu_device *d)
 966 {
 967         struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
 968         struct lu_device *next = ed->ed_next;
 969
 970         while (next && !ed->ed_next_ismd)
 971                 next = next->ld_type->ldt_ops->ldto_device_fini(env, next);
 972         return NULL;
 973 }
 974
 975 static void echo_lock_release(const struct lu_env *env,
 976                               struct echo_lock *ecl,
 977                               int still_used)
 978 {
 979         struct cl_lock *clk = echo_lock2cl(ecl);
 980
 981         cl_lock_get(clk);
 982         cl_unuse(env, clk);
 983         cl_lock_release(env, clk, "ec enqueue", ecl->el_object);
 984         if (!still_used) {
 985                 cl_lock_mutex_get(env, clk);
 986                 cl_lock_cancel(env, clk);
 987                 cl_lock_delete(env, clk);
 988                 cl_lock_mutex_put(env, clk);
 989         }
 990         cl_lock_put(env, clk);
 991 }
 992
 993 static struct lu_device *echo_device_free(const struct lu_env *env,
 994                                           struct lu_device *d)
 995 {
 996         struct echo_device     *ed   = cl2echo_dev(lu2cl_dev(d));
 997         struct echo_client_obd *ec   = ed->ed_ec;
 998         struct echo_object     *eco;
 999         struct lu_device       *next = ed->ed_next;
1000
1001         CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n",
1002                ed, next);
1003
1004         lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1005
1006         /* check if there are objects still alive.
1007          * It shouldn't have any object because lu_site_purge would cleanup
1008          * all of cached objects. Anyway, probably the echo device is being
1009          * parallelly accessed.
1010          */
1011         spin_lock(&ec->ec_lock);
1012         cfs_list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain)
1013                 eco->eo_deleted = 1;
1014         spin_unlock(&ec->ec_lock);
1015
1016         /* purge again */
1017         lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1018
1019         CDEBUG(D_INFO,
1020                "Waiting for the reference of echo object to be dropped\n");
1021
1022         /* Wait for the last reference to be dropped. */
1023         spin_lock(&ec->ec_lock);
1024         while (!cfs_list_empty(&ec->ec_objects)) {
1025                 spin_unlock(&ec->ec_lock);
1026                 CERROR("echo_client still has objects at cleanup time, "
1027                        "wait for 1 second\n");
1028                 schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE,
1029                                                    cfs_time_seconds(1));
1030                 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1031                 spin_lock(&ec->ec_lock);
1032         }
1033         spin_unlock(&ec->ec_lock);
1034
1035         LASSERT(cfs_list_empty(&ec->ec_locks));
1036
1037         CDEBUG(D_INFO, "No object exists, exiting...\n");
1038
1039         echo_client_cleanup(d->ld_obd);
1040 #ifdef HAVE_SERVER_SUPPORT
1041         echo_fid_fini(d->ld_obd);
1042 #endif
1043         while (next && !ed->ed_next_ismd)
1044                 next = next->ld_type->ldt_ops->ldto_device_free(env, next);
1045
1046         LASSERT(ed->ed_site == lu2cl_site(d->ld_site));
1047         echo_site_fini(env, ed);
1048         cl_device_fini(&ed->ed_cl);
1049         OBD_FREE_PTR(ed);
1050
1051         return NULL;
1052 }
1053
1054 static const struct lu_device_type_operations echo_device_type_ops = {
1055         .ldto_init = echo_type_init,
1056         .ldto_fini = echo_type_fini,
1057
1058         .ldto_start = echo_type_start,
1059         .ldto_stop  = echo_type_stop,
1060
1061         .ldto_device_alloc = echo_device_alloc,
1062         .ldto_device_free  = echo_device_free,
1063         .ldto_device_init  = echo_device_init,
1064         .ldto_device_fini  = echo_device_fini
1065 };
1066
1067 static struct lu_device_type echo_device_type = {
1068         .ldt_tags     = LU_DEVICE_CL,
1069         .ldt_name     = LUSTRE_ECHO_CLIENT_NAME,
1070         .ldt_ops      = &echo_device_type_ops,
1071         .ldt_ctx_tags = LCT_CL_THREAD | LCT_MD_THREAD | LCT_DT_THREAD,
1072 };
1073 /** @} echo_init */
1074
1075 /** \defgroup echo_exports Exported operations
1076  *
1077  * exporting functions to echo client
1078  *
1079  * @{
1080  */
1081
1082 /* Interfaces to echo client obd device */
1083 static struct echo_object *cl_echo_object_find(struct echo_device *d,
1084                                                struct lov_stripe_md **lsmp)
1085 {
1086         struct lu_env *env;
1087         struct echo_thread_info *info;
1088         struct echo_object_conf *conf;
1089         struct lov_stripe_md    *lsm;
1090         struct echo_object *eco;
1091         struct cl_object   *obj;
1092         struct lu_fid *fid;
1093         int refcheck;
1094         int rc;
1095         ENTRY;
1096
1097         LASSERT(lsmp);
1098         lsm = *lsmp;
1099         LASSERT(lsm);
1100         LASSERTF(ostid_id(&lsm->lsm_oi) != 0, DOSTID"\n", POSTID(&lsm->lsm_oi));
1101         LASSERTF(ostid_seq(&lsm->lsm_oi) == FID_SEQ_ECHO, DOSTID"\n",
1102                  POSTID(&lsm->lsm_oi));
1103
1104         /* Never return an object if the obd is to be freed. */
1105         if (echo_dev2cl(d)->cd_lu_dev.ld_obd->obd_stopping)
1106                 RETURN(ERR_PTR(-ENODEV));
1107
1108         env = cl_env_get(&refcheck);
1109         if (IS_ERR(env))
1110                 RETURN((void *)env);
1111
1112         info = echo_env_info(env);
1113         conf = &info->eti_conf;
1114         if (d->ed_next) {
1115                 if (!d->ed_next_islov) {
1116                         struct lov_oinfo *oinfo = lsm->lsm_oinfo[0];
1117                         LASSERT(oinfo != NULL);
1118                         oinfo->loi_oi = lsm->lsm_oi;
1119                         conf->eoc_cl.u.coc_oinfo = oinfo;
1120                 } else {
1121                         struct lustre_md *md;
1122                         md = &info->eti_md;
1123                         memset(md, 0, sizeof *md);
1124                         md->lsm = lsm;
1125                         conf->eoc_cl.u.coc_md = md;
1126                 }
1127         }
1128         conf->eoc_md = lsmp;
1129
1130         fid  = &info->eti_fid;
1131         rc = ostid_to_fid(fid, &lsm->lsm_oi, 0);
1132         if (rc != 0)
1133                 GOTO(out, eco = ERR_PTR(rc));
1134
1135         /* In the function below, .hs_keycmp resolves to
1136          * lu_obj_hop_keycmp() */
1137         /* coverity[overrun-buffer-val] */
1138         obj = cl_object_find(env, echo_dev2cl(d), fid, &conf->eoc_cl);
1139         if (IS_ERR(obj))
1140                 GOTO(out, eco = (void*)obj);
1141
1142         eco = cl2echo_obj(obj);
1143         if (eco->eo_deleted) {
1144                 cl_object_put(env, obj);
1145                 eco = ERR_PTR(-EAGAIN);
1146         }
1147
1148 out:
1149         cl_env_put(env, &refcheck);
1150         RETURN(eco);
1151 }
1152
1153 static int cl_echo_object_put(struct echo_object *eco)
1154 {
1155         struct lu_env *env;
1156         struct cl_object *obj = echo_obj2cl(eco);
1157         int refcheck;
1158         ENTRY;
1159
1160         env = cl_env_get(&refcheck);
1161         if (IS_ERR(env))
1162                 RETURN(PTR_ERR(env));
1163
1164         /* an external function to kill an object? */
1165         if (eco->eo_deleted) {
1166                 struct lu_object_header *loh = obj->co_lu.lo_header;
1167                 LASSERT(&eco->eo_hdr == luh2coh(loh));
1168                 set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags);
1169         }
1170
1171         cl_object_put(env, obj);
1172         cl_env_put(env, &refcheck);
1173         RETURN(0);
1174 }
1175
1176 static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco,
1177                             obd_off start, obd_off end, int mode,
1178                             __u64 *cookie , __u32 enqflags)
1179 {
1180         struct cl_io *io;
1181         struct cl_lock *lck;
1182         struct cl_object *obj;
1183         struct cl_lock_descr *descr;
1184         struct echo_thread_info *info;
1185         int rc = -ENOMEM;
1186         ENTRY;
1187
1188         info = echo_env_info(env);
1189         io = &info->eti_io;
1190         descr = &info->eti_descr;
1191         obj = echo_obj2cl(eco);
1192
1193         descr->cld_obj   = obj;
1194         descr->cld_start = cl_index(obj, start);
1195         descr->cld_end   = cl_index(obj, end);
1196         descr->cld_mode  = mode == LCK_PW ? CLM_WRITE : CLM_READ;
1197         descr->cld_enq_flags = enqflags;
1198         io->ci_obj = obj;
1199
1200         lck = cl_lock_request(env, io, descr, "ec enqueue", eco);
1201         if (lck) {
1202                 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
1203                 struct echo_lock *el;
1204
1205                 rc = cl_wait(env, lck);
1206                 if (rc == 0) {
1207                         el = cl2echo_lock(cl_lock_at(lck, &echo_device_type));
1208                         spin_lock(&ec->ec_lock);
1209                         if (cfs_list_empty(&el->el_chain)) {
1210                                 cfs_list_add(&el->el_chain, &ec->ec_locks);
1211                                 el->el_cookie = ++ec->ec_unique;
1212                         }
1213                         cfs_atomic_inc(&el->el_refcount);
1214                         *cookie = el->el_cookie;
1215                         spin_unlock(&ec->ec_lock);
1216                 } else {
1217                         cl_lock_release(env, lck, "ec enqueue", current);
1218                 }
1219         }
1220         RETURN(rc);
1221 }
1222
1223 static int cl_echo_enqueue(struct echo_object *eco, obd_off start, obd_off end,
1224                            int mode, __u64 *cookie)
1225 {
1226         struct echo_thread_info *info;
1227         struct lu_env *env;
1228         struct cl_io *io;
1229         int refcheck;
1230         int result;
1231         ENTRY;
1232
1233         env = cl_env_get(&refcheck);
1234         if (IS_ERR(env))
1235                 RETURN(PTR_ERR(env));
1236
1237         info = echo_env_info(env);
1238         io = &info->eti_io;
1239
1240         io->ci_ignore_layout = 1;
1241         result = cl_io_init(env, io, CIT_MISC, echo_obj2cl(eco));
1242         if (result < 0)
1243                 GOTO(out, result);
1244         LASSERT(result == 0);
1245
1246         result = cl_echo_enqueue0(env, eco, start, end, mode, cookie, 0);
1247         cl_io_fini(env, io);
1248
1249         EXIT;
1250 out:
1251         cl_env_put(env, &refcheck);
1252         return result;
1253 }
1254
1255 static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed,
1256                            __u64 cookie)
1257 {
1258         struct echo_client_obd *ec = ed->ed_ec;
1259         struct echo_lock       *ecl = NULL;
1260         cfs_list_t             *el;
1261         int found = 0, still_used = 0;
1262         ENTRY;
1263
1264         LASSERT(ec != NULL);
1265         spin_lock(&ec->ec_lock);
1266         cfs_list_for_each (el, &ec->ec_locks) {
1267                 ecl = cfs_list_entry (el, struct echo_lock, el_chain);
1268                 CDEBUG(D_INFO, "ecl: %p, cookie: "LPX64"\n", ecl, ecl->el_cookie);
1269                 found = (ecl->el_cookie == cookie);
1270                 if (found) {
1271                         if (cfs_atomic_dec_and_test(&ecl->el_refcount))
1272                                 cfs_list_del_init(&ecl->el_chain);
1273                         else
1274                                 still_used = 1;
1275                         break;
1276                 }
1277         }
1278         spin_unlock(&ec->ec_lock);
1279
1280         if (!found)
1281                 RETURN(-ENOENT);
1282
1283         echo_lock_release(env, ecl, still_used);
1284         RETURN(0);
1285 }
1286
1287 static int cl_echo_cancel(struct echo_device *ed, __u64 cookie)
1288 {
1289         struct lu_env *env;
1290         int refcheck;
1291         int rc;
1292         ENTRY;
1293
1294         env = cl_env_get(&refcheck);
1295         if (IS_ERR(env))
1296                 RETURN(PTR_ERR(env));
1297
1298         rc = cl_echo_cancel0(env, ed, cookie);
1299
1300         cl_env_put(env, &refcheck);
1301         RETURN(rc);
1302 }
1303
1304 static int cl_echo_async_brw(const struct lu_env *env, struct cl_io *io,
1305                              enum cl_req_type unused, struct cl_2queue *queue)
1306 {
1307         struct cl_page *clp;
1308         struct cl_page *temp;
1309         int result = 0;
1310         ENTRY;
1311
1312         cl_page_list_for_each_safe(clp, temp, &queue->c2_qin) {
1313                 int rc;
1314                 rc = cl_page_cache_add(env, io, clp, CRT_WRITE);
1315                 if (rc == 0)
1316                         continue;
1317                 result = result ?: rc;
1318         }
1319         RETURN(result);
1320 }
1321
1322 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
1323                               struct page **pages, int npages, int async)
1324 {
1325         struct lu_env           *env;
1326         struct echo_thread_info *info;
1327         struct cl_object        *obj = echo_obj2cl(eco);
1328         struct echo_device      *ed  = eco->eo_dev;
1329         struct cl_2queue        *queue;
1330         struct cl_io            *io;
1331         struct cl_page          *clp;
1332         struct lustre_handle    lh = { 0 };
1333         int page_size = cl_page_size(obj);
1334         int refcheck;
1335         int rc;
1336         int i;
1337         ENTRY;
1338
1339         LASSERT((offset & ~CFS_PAGE_MASK) == 0);
1340         LASSERT(ed->ed_next != NULL);
1341         env = cl_env_get(&refcheck);
1342         if (IS_ERR(env))
1343                 RETURN(PTR_ERR(env));
1344
1345         info    = echo_env_info(env);
1346         io      = &info->eti_io;
1347         queue   = &info->eti_queue;
1348
1349         cl_2queue_init(queue);
1350
1351         io->ci_ignore_layout = 1;
1352         rc = cl_io_init(env, io, CIT_MISC, obj);
1353         if (rc < 0)
1354                 GOTO(out, rc);
1355         LASSERT(rc == 0);
1356
1357
1358         rc = cl_echo_enqueue0(env, eco, offset,
1359                               offset + npages * PAGE_CACHE_SIZE - 1,
1360                               rw == READ ? LCK_PR : LCK_PW, &lh.cookie,
1361                               CEF_NEVER);
1362         if (rc < 0)
1363                 GOTO(error_lock, rc);
1364
1365         for (i = 0; i < npages; i++) {
1366                 LASSERT(pages[i]);
1367                 clp = cl_page_find(env, obj, cl_index(obj, offset),
1368                                    pages[i], CPT_TRANSIENT);
1369                 if (IS_ERR(clp)) {
1370                         rc = PTR_ERR(clp);
1371                         break;
1372                 }
1373                 LASSERT(clp->cp_type == CPT_TRANSIENT);
1374
1375                 rc = cl_page_own(env, io, clp);
1376                 if (rc) {
1377                         LASSERT(clp->cp_state == CPS_FREEING);
1378                         cl_page_put(env, clp);
1379                         break;
1380                 }
1381
1382                 cl_2queue_add(queue, clp);
1383
1384                 /* drop the reference count for cl_page_find, so that the page
1385                  * will be freed in cl_2queue_fini. */
1386                 cl_page_put(env, clp);
1387                 cl_page_clip(env, clp, 0, page_size);
1388
1389                 offset += page_size;
1390         }
1391
1392         if (rc == 0) {
1393                 enum cl_req_type typ = rw == READ ? CRT_READ : CRT_WRITE;
1394
1395                 async = async && (typ == CRT_WRITE);
1396                 if (async)
1397                         rc = cl_echo_async_brw(env, io, typ, queue);
1398                 else
1399                         rc = cl_io_submit_sync(env, io, typ, queue, 0);
1400                 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
1401                        async ? "async" : "sync", rc);
1402         }
1403
1404         cl_echo_cancel0(env, ed, lh.cookie);
1405         EXIT;
1406 error_lock:
1407         cl_2queue_discard(env, io, queue);
1408         cl_2queue_disown(env, io, queue);
1409         cl_2queue_fini(env, queue);
1410         cl_io_fini(env, io);
1411 out:
1412         cl_env_put(env, &refcheck);
1413         return rc;
1414 }
1415 /** @} echo_exports */
1416
1417
1418 static obd_id last_object_id;
1419
1420 static int
1421 echo_copyout_lsm (struct lov_stripe_md *lsm, void *_ulsm, int ulsm_nob)
1422 {
1423         struct lov_stripe_md *ulsm = _ulsm;
1424         int nob, i;
1425
1426         nob = offsetof (struct lov_stripe_md, lsm_oinfo[lsm->lsm_stripe_count]);
1427         if (nob > ulsm_nob)
1428                 return (-EINVAL);
1429
1430         if (copy_to_user (ulsm, lsm, sizeof(ulsm)))
1431                 return (-EFAULT);
1432
1433         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1434                 if (copy_to_user (ulsm->lsm_oinfo[i], lsm->lsm_oinfo[i],
1435                                       sizeof(lsm->lsm_oinfo[0])))
1436                         return (-EFAULT);
1437         }
1438         return 0;
1439 }
1440
1441 static int
1442 echo_copyin_lsm (struct echo_device *ed, struct lov_stripe_md *lsm,
1443                  void *ulsm, int ulsm_nob)
1444 {
1445         struct echo_client_obd *ec = ed->ed_ec;
1446         int                     i;
1447
1448         if (ulsm_nob < sizeof (*lsm))
1449                 return (-EINVAL);
1450
1451         if (copy_from_user (lsm, ulsm, sizeof (*lsm)))
1452                 return (-EFAULT);
1453
1454         if (lsm->lsm_stripe_count > ec->ec_nstripes ||
1455             lsm->lsm_magic != LOV_MAGIC ||
1456             (lsm->lsm_stripe_size & (~CFS_PAGE_MASK)) != 0 ||
1457             ((__u64)lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL))
1458                 return (-EINVAL);
1459
1460
1461         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1462                 if (copy_from_user(lsm->lsm_oinfo[i],
1463                                        ((struct lov_stripe_md *)ulsm)-> \
1464                                        lsm_oinfo[i],
1465                                        sizeof(lsm->lsm_oinfo[0])))
1466                         return (-EFAULT);
1467         }
1468         return (0);
1469 }
1470
1471 #ifdef HAVE_SERVER_SUPPORT
1472 static inline void echo_md_build_name(struct lu_name *lname, char *name,
1473                                       __u64 id)
1474 {
1475         sprintf(name, LPU64, id);
1476         lname->ln_name = name;
1477         lname->ln_namelen = strlen(name);
1478 }
1479
1480 /* similar to mdt_attr_get_complex */
1481 static int echo_big_lmm_get(const struct lu_env *env, struct md_object *o,
1482                             struct md_attr *ma)
1483 {
1484         struct echo_thread_info *info = echo_env_info(env);
1485         int                      rc;
1486
1487         ENTRY;
1488
1489         LASSERT(ma->ma_lmm_size > 0);
1490
1491         rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LOV);
1492         if (rc < 0)
1493                 RETURN(rc);
1494
1495         /* big_lmm may need to be grown */
1496         if (info->eti_big_lmmsize < rc) {
1497                 int size = size_roundup_power2(rc);
1498
1499                 if (info->eti_big_lmmsize > 0) {
1500                         /* free old buffer */
1501                         LASSERT(info->eti_big_lmm);
1502                         OBD_FREE_LARGE(info->eti_big_lmm,
1503                                        info->eti_big_lmmsize);
1504                         info->eti_big_lmm = NULL;
1505                         info->eti_big_lmmsize = 0;
1506                 }
1507
1508                 OBD_ALLOC_LARGE(info->eti_big_lmm, size);
1509                 if (info->eti_big_lmm == NULL)
1510                         RETURN(-ENOMEM);
1511                 info->eti_big_lmmsize = size;
1512         }
1513         LASSERT(info->eti_big_lmmsize >= rc);
1514
1515         info->eti_buf.lb_buf = info->eti_big_lmm;
1516         info->eti_buf.lb_len = info->eti_big_lmmsize;
1517         rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LOV);
1518         if (rc < 0)
1519                 RETURN(rc);
1520
1521         ma->ma_valid |= MA_LOV;
1522         ma->ma_lmm = info->eti_big_lmm;
1523         ma->ma_lmm_size = rc;
1524
1525         RETURN(0);
1526 }
1527
1528 int echo_attr_get_complex(const struct lu_env *env, struct md_object *next,
1529                           struct md_attr *ma)
1530 {
1531         struct echo_thread_info *info = echo_env_info(env);
1532         struct lu_buf           *buf = &info->eti_buf;
1533         umode_t          mode = lu_object_attr(&next->mo_lu);
1534         int                      need = ma->ma_need;
1535         int                      rc = 0, rc2;
1536
1537         ENTRY;
1538
1539         ma->ma_valid = 0;
1540
1541         if (need & MA_INODE) {
1542                 ma->ma_need = MA_INODE;
1543                 rc = mo_attr_get(env, next, ma);
1544                 if (rc)
1545                         GOTO(out, rc);
1546                 ma->ma_valid |= MA_INODE;
1547         }
1548
1549         if (need & MA_LOV) {
1550                 if (S_ISREG(mode) || S_ISDIR(mode)) {
1551                         LASSERT(ma->ma_lmm_size > 0);
1552                         buf->lb_buf = ma->ma_lmm;
1553                         buf->lb_len = ma->ma_lmm_size;
1554                         rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LOV);
1555                         if (rc2 > 0) {
1556                                 ma->ma_lmm_size = rc2;
1557                                 ma->ma_valid |= MA_LOV;
1558                         } else if (rc2 == -ENODATA) {
1559                                 /* no LOV EA */
1560                                 ma->ma_lmm_size = 0;
1561                         } else if (rc2 == -ERANGE) {
1562                                 rc2 = echo_big_lmm_get(env, next, ma);
1563                                 if (rc2 < 0)
1564                                         GOTO(out, rc = rc2);
1565                         } else {
1566                                 GOTO(out, rc = rc2);
1567                         }
1568                 }
1569         }
1570
1571 #ifdef CONFIG_FS_POSIX_ACL
1572         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1573                 buf->lb_buf = ma->ma_acl;
1574                 buf->lb_len = ma->ma_acl_size;
1575                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1576                 if (rc2 > 0) {
1577                         ma->ma_acl_size = rc2;
1578                         ma->ma_valid |= MA_ACL_DEF;
1579                 } else if (rc2 == -ENODATA) {
1580                         /* no ACLs */
1581                         ma->ma_acl_size = 0;
1582                 } else {
1583                         GOTO(out, rc = rc2);
1584                 }
1585         }
1586 #endif
1587 out:
1588         ma->ma_need = need;
1589         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
1590                rc, ma->ma_valid, ma->ma_lmm);
1591         RETURN(rc);
1592 }
1593
1594 static int
1595 echo_md_create_internal(const struct lu_env *env, struct echo_device *ed,
1596                         struct md_object *parent, struct lu_fid *fid,
1597                         struct lu_name *lname, struct md_op_spec *spec,
1598                         struct md_attr *ma)
1599 {
1600         struct lu_object        *ec_child, *child;
1601         struct lu_device        *ld = ed->ed_next;
1602         struct echo_thread_info *info = echo_env_info(env);
1603         struct lu_fid           *fid2 = &info->eti_fid2;
1604         struct lu_object_conf    conf = { .loc_flags = LOC_F_NEW };
1605         int                      rc;
1606
1607         ENTRY;
1608
1609         rc = mdo_lookup(env, parent, lname, fid2, spec);
1610         if (rc == 0)
1611                 return -EEXIST;
1612         else if (rc != -ENOENT)
1613                 return rc;
1614
1615         ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev,
1616                                      fid, &conf);
1617         if (IS_ERR(ec_child)) {
1618                 CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid),
1619                         PTR_ERR(ec_child));
1620                 RETURN(PTR_ERR(ec_child));
1621         }
1622
1623         child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1624         if (child == NULL) {
1625                 CERROR("Can not locate the child "DFID"\n", PFID(fid));
1626                 GOTO(out_put, rc = -EINVAL);
1627         }
1628
1629         CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n",
1630                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1631
1632         /*
1633          * Do not perform lookup sanity check. We know that name does not exist.
1634          */
1635         spec->sp_cr_lookup = 0;
1636         rc = mdo_create(env, parent, lname, lu2md(child), spec, ma);
1637         if (rc) {
1638                 CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc);
1639                 GOTO(out_put, rc);
1640         }
1641         CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc  = %d\n",
1642                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc);
1643         EXIT;
1644 out_put:
1645         lu_object_put(env, ec_child);
1646         return rc;
1647 }
1648
1649 static int echo_set_lmm_size(const struct lu_env *env, struct lu_device *ld,
1650                              struct md_attr *ma)
1651 {
1652         struct echo_thread_info *info = echo_env_info(env);
1653
1654         if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
1655                 ma->ma_lmm = (void *)&info->eti_lmm;
1656                 ma->ma_lmm_size = sizeof(info->eti_lmm);
1657         } else {
1658                 LASSERT(info->eti_big_lmmsize);
1659                 ma->ma_lmm = info->eti_big_lmm;
1660                 ma->ma_lmm_size = info->eti_big_lmmsize;
1661         }
1662
1663         return 0;
1664 }
1665
1666 static int echo_create_md_object(const struct lu_env *env,
1667                                  struct echo_device *ed,
1668                                  struct lu_object *ec_parent,
1669                                  struct lu_fid *fid,
1670                                  char *name, int namelen,
1671                                  __u64 id, __u32 mode, int count,
1672                                  int stripe_count, int stripe_offset)
1673 {
1674         struct lu_object        *parent;
1675         struct echo_thread_info *info = echo_env_info(env);
1676         struct lu_name          *lname = &info->eti_lname;
1677         struct md_op_spec       *spec = &info->eti_spec;
1678         struct md_attr          *ma = &info->eti_ma;
1679         struct lu_device        *ld = ed->ed_next;
1680         int                      rc = 0;
1681         int                      i;
1682
1683         ENTRY;
1684
1685         if (ec_parent == NULL)
1686                 return -1;
1687         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1688         if (parent == NULL)
1689                 RETURN(-ENXIO);
1690
1691         memset(ma, 0, sizeof(*ma));
1692         memset(spec, 0, sizeof(*spec));
1693         if (stripe_count != 0) {
1694                 spec->sp_cr_flags |= FMODE_WRITE;
1695                 echo_set_lmm_size(env, ld, ma);
1696                 if (stripe_count != -1) {
1697                         struct lov_user_md_v3 *lum = &info->eti_lum;
1698
1699                         lum->lmm_magic = LOV_USER_MAGIC_V3;
1700                         lum->lmm_stripe_count = stripe_count;
1701                         lum->lmm_stripe_offset = stripe_offset;
1702                         lum->lmm_pattern = 0;
1703                         spec->u.sp_ea.eadata = lum;
1704                         spec->u.sp_ea.eadatalen = sizeof(*lum);
1705                         spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1706                 }
1707         }
1708
1709         ma->ma_attr.la_mode = mode;
1710         ma->ma_attr.la_valid = LA_CTIME | LA_MODE;
1711         ma->ma_attr.la_ctime = cfs_time_current_64();
1712
1713         if (name != NULL) {
1714                 lname->ln_name = name;
1715                 lname->ln_namelen = namelen;
1716                 /* If name is specified, only create one object by name */
1717                 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1718                                              spec, ma);
1719                 RETURN(rc);
1720         }
1721
1722         /* Create multiple object sequenced by id */
1723         for (i = 0; i < count; i++) {
1724                 char *tmp_name = info->eti_name;
1725
1726                 echo_md_build_name(lname, tmp_name, id);
1727
1728                 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1729                                              spec, ma);
1730                 if (rc) {
1731                         CERROR("Can not create child %s: rc = %d\n", tmp_name,
1732                                 rc);
1733                         break;
1734                 }
1735                 id++;
1736                 fid->f_oid++;
1737         }
1738
1739         RETURN(rc);
1740 }
1741
1742 static struct lu_object *echo_md_lookup(const struct lu_env *env,
1743                                         struct echo_device *ed,
1744                                         struct md_object *parent,
1745                                         struct lu_name *lname)
1746 {
1747         struct echo_thread_info *info = echo_env_info(env);
1748         struct lu_fid           *fid = &info->eti_fid;
1749         struct lu_object        *child;
1750         int    rc;
1751         ENTRY;
1752
1753         CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name,
1754                PFID(fid), parent);
1755         rc = mdo_lookup(env, parent, lname, fid, NULL);
1756         if (rc) {
1757                 CERROR("lookup %s: rc = %d\n", lname->ln_name, rc);
1758                 RETURN(ERR_PTR(rc));
1759         }
1760
1761         /* In the function below, .hs_keycmp resolves to
1762          * lu_obj_hop_keycmp() */
1763         /* coverity[overrun-buffer-val] */
1764         child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1765
1766         RETURN(child);
1767 }
1768
1769 static int echo_setattr_object(const struct lu_env *env,
1770                                struct echo_device *ed,
1771                                struct lu_object *ec_parent,
1772                                __u64 id, int count)
1773 {
1774         struct lu_object        *parent;
1775         struct echo_thread_info *info = echo_env_info(env);
1776         struct lu_name          *lname = &info->eti_lname;
1777         char                    *name = info->eti_name;
1778         struct lu_device        *ld = ed->ed_next;
1779         struct lu_buf           *buf = &info->eti_buf;
1780         int                      rc = 0;
1781         int                      i;
1782
1783         ENTRY;
1784
1785         if (ec_parent == NULL)
1786                 return -1;
1787         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1788         if (parent == NULL)
1789                 RETURN(-ENXIO);
1790
1791         for (i = 0; i < count; i++) {
1792                 struct lu_object *ec_child, *child;
1793
1794                 echo_md_build_name(lname, name, id);
1795
1796                 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1797                 if (IS_ERR(ec_child)) {
1798                         CERROR("Can't find child %s: rc = %ld\n",
1799                                 lname->ln_name, PTR_ERR(ec_child));
1800                         RETURN(PTR_ERR(ec_child));
1801                 }
1802
1803                 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1804                 if (child == NULL) {
1805                         CERROR("Can not locate the child %s\n", lname->ln_name);
1806                         lu_object_put(env, ec_child);
1807                         rc = -EINVAL;
1808                         break;
1809                 }
1810
1811                 CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n",
1812                        PFID(lu_object_fid(child)));
1813
1814                 buf->lb_buf = info->eti_xattr_buf;
1815                 buf->lb_len = sizeof(info->eti_xattr_buf);
1816
1817                 sprintf(name, "%s.test1", XATTR_USER_PREFIX);
1818                 rc = mo_xattr_set(env, lu2md(child), buf, name,
1819                                   LU_XATTR_CREATE);
1820                 if (rc < 0) {
1821                         CERROR("Can not setattr child "DFID": rc = %d\n",
1822                                 PFID(lu_object_fid(child)), rc);
1823                         lu_object_put(env, ec_child);
1824                         break;
1825                 }
1826                 CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n",
1827                        PFID(lu_object_fid(child)));
1828                 id++;
1829                 lu_object_put(env, ec_child);
1830         }
1831         RETURN(rc);
1832 }
1833
1834 static int echo_getattr_object(const struct lu_env *env,
1835                                struct echo_device *ed,
1836                                struct lu_object *ec_parent,
1837                                __u64 id, int count)
1838 {
1839         struct lu_object        *parent;
1840         struct echo_thread_info *info = echo_env_info(env);
1841         struct lu_name          *lname = &info->eti_lname;
1842         char                    *name = info->eti_name;
1843         struct md_attr          *ma = &info->eti_ma;
1844         struct lu_device        *ld = ed->ed_next;
1845         int                      rc = 0;
1846         int                      i;
1847
1848         ENTRY;
1849
1850         if (ec_parent == NULL)
1851                 return -1;
1852         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1853         if (parent == NULL)
1854                 RETURN(-ENXIO);
1855
1856         memset(ma, 0, sizeof(*ma));
1857         ma->ma_need |= MA_INODE | MA_LOV | MA_PFID | MA_HSM | MA_ACL_DEF;
1858         ma->ma_acl = info->eti_xattr_buf;
1859         ma->ma_acl_size = sizeof(info->eti_xattr_buf);
1860
1861         for (i = 0; i < count; i++) {
1862                 struct lu_object *ec_child, *child;
1863
1864                 ma->ma_valid = 0;
1865                 echo_md_build_name(lname, name, id);
1866                 echo_set_lmm_size(env, ld, ma);
1867
1868                 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1869                 if (IS_ERR(ec_child)) {
1870                         CERROR("Can't find child %s: rc = %ld\n",
1871                                lname->ln_name, PTR_ERR(ec_child));
1872                         RETURN(PTR_ERR(ec_child));
1873                 }
1874
1875                 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1876                 if (child == NULL) {
1877                         CERROR("Can not locate the child %s\n", lname->ln_name);
1878                         lu_object_put(env, ec_child);
1879                         RETURN(-EINVAL);
1880                 }
1881
1882                 CDEBUG(D_RPCTRACE, "Start getattr object "DFID"\n",
1883                        PFID(lu_object_fid(child)));
1884                 rc = echo_attr_get_complex(env, lu2md(child), ma);
1885                 if (rc) {
1886                         CERROR("Can not getattr child "DFID": rc = %d\n",
1887                                 PFID(lu_object_fid(child)), rc);
1888                         lu_object_put(env, ec_child);
1889                         break;
1890                 }
1891                 CDEBUG(D_RPCTRACE, "End getattr object "DFID"\n",
1892                        PFID(lu_object_fid(child)));
1893                 id++;
1894                 lu_object_put(env, ec_child);
1895         }
1896
1897         RETURN(rc);
1898 }
1899
1900 static int echo_lookup_object(const struct lu_env *env,
1901                               struct echo_device *ed,
1902                               struct lu_object *ec_parent,
1903                               __u64 id, int count)
1904 {
1905         struct lu_object        *parent;
1906         struct echo_thread_info *info = echo_env_info(env);
1907         struct lu_name          *lname = &info->eti_lname;
1908         char                    *name = info->eti_name;
1909         struct lu_fid           *fid = &info->eti_fid;
1910         struct lu_device        *ld = ed->ed_next;
1911         int                      rc = 0;
1912         int                      i;
1913
1914         if (ec_parent == NULL)
1915                 return -1;
1916         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1917         if (parent == NULL)
1918                 return -ENXIO;
1919
1920         /*prepare the requests*/
1921         for (i = 0; i < count; i++) {
1922                 echo_md_build_name(lname, name, id);
1923
1924                 CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n",
1925                        PFID(lu_object_fid(parent)), lname->ln_name, parent);
1926
1927                 rc = mdo_lookup(env, lu2md(parent), lname, fid, NULL);
1928                 if (rc) {
1929                         CERROR("Can not lookup child %s: rc = %d\n", name, rc);
1930                         break;
1931                 }
1932                 CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n",
1933                        PFID(lu_object_fid(parent)), lname->ln_name, parent);
1934
1935                 id++;
1936         }
1937         return rc;
1938 }
1939
1940 static int echo_md_destroy_internal(const struct lu_env *env,
1941                                     struct echo_device *ed,
1942                                     struct md_object *parent,
1943                                     struct lu_name *lname,
1944                                     struct md_attr *ma)
1945 {
1946         struct lu_device   *ld = ed->ed_next;
1947         struct lu_object   *ec_child;
1948         struct lu_object   *child;
1949         int                 rc;
1950
1951         ENTRY;
1952
1953         ec_child = echo_md_lookup(env, ed, parent, lname);
1954         if (IS_ERR(ec_child)) {
1955                 CERROR("Can't find child %s: rc = %ld\n", lname->ln_name,
1956                         PTR_ERR(ec_child));
1957                 RETURN(PTR_ERR(ec_child));
1958         }
1959
1960         child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1961         if (child == NULL) {
1962                 CERROR("Can not locate the child %s\n", lname->ln_name);
1963                 GOTO(out_put, rc = -EINVAL);
1964         }
1965
1966         if (lu_object_remote(child)) {
1967                 CERROR("Can not destroy remote object %s: rc = %d\n",
1968                        lname->ln_name, -EPERM);
1969                 GOTO(out_put, rc = -EPERM);
1970         }
1971         CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n",
1972                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1973
1974         rc = mdo_unlink(env, parent, lu2md(child), lname, ma, 0);
1975         if (rc) {
1976                 CERROR("Can not unlink child %s: rc = %d\n",
1977                         lname->ln_name, rc);
1978                 GOTO(out_put, rc);
1979         }
1980         CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n",
1981                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1982 out_put:
1983         lu_object_put(env, ec_child);
1984         return rc;
1985 }
1986
1987 static int echo_destroy_object(const struct lu_env *env,
1988                                struct echo_device *ed,
1989                                struct lu_object *ec_parent,
1990                                char *name, int namelen,
1991                                __u64 id, __u32 mode,
1992                                int count)
1993 {
1994         struct echo_thread_info *info = echo_env_info(env);
1995         struct lu_name          *lname = &info->eti_lname;
1996         struct md_attr          *ma = &info->eti_ma;
1997         struct lu_device        *ld = ed->ed_next;
1998         struct lu_object        *parent;
1999         int                      rc = 0;
2000         int                      i;
2001         ENTRY;
2002
2003         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
2004         if (parent == NULL)
2005                 RETURN(-EINVAL);
2006
2007         memset(ma, 0, sizeof(*ma));
2008         ma->ma_attr.la_mode = mode;
2009         ma->ma_attr.la_valid = LA_CTIME;
2010         ma->ma_attr.la_ctime = cfs_time_current_64();
2011         ma->ma_need = MA_INODE;
2012         ma->ma_valid = 0;
2013
2014         if (name != NULL) {
2015                 lname->ln_name = name;
2016                 lname->ln_namelen = namelen;
2017                 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
2018                                               ma);
2019                 RETURN(rc);
2020         }
2021
2022         /*prepare the requests*/
2023         for (i = 0; i < count; i++) {
2024                 char *tmp_name = info->eti_name;
2025
2026                 ma->ma_valid = 0;
2027                 echo_md_build_name(lname, tmp_name, id);
2028
2029                 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
2030                                               ma);
2031                 if (rc) {
2032                         CERROR("Can not unlink child %s: rc = %d\n", name, rc);
2033                         break;
2034                 }
2035                 id++;
2036         }
2037
2038         RETURN(rc);
2039 }
2040
2041 static struct lu_object *echo_resolve_path(const struct lu_env *env,
2042                                            struct echo_device *ed, char *path,
2043                                            int path_len)
2044 {
2045         struct lu_device        *ld = ed->ed_next;
2046         struct md_device        *md = lu2md_dev(ld);
2047         struct echo_thread_info *info = echo_env_info(env);
2048         struct lu_fid           *fid = &info->eti_fid;
2049         struct lu_name          *lname = &info->eti_lname;
2050         struct lu_object        *parent = NULL;
2051         struct lu_object        *child = NULL;
2052         int rc = 0;
2053         ENTRY;
2054
2055         /*Only support MDD layer right now*/
2056         rc = md->md_ops->mdo_root_get(env, md, fid);
2057         if (rc) {
2058                 CERROR("get root error: rc = %d\n", rc);
2059                 RETURN(ERR_PTR(rc));
2060         }
2061
2062         /* In the function below, .hs_keycmp resolves to
2063          * lu_obj_hop_keycmp() */
2064         /* coverity[overrun-buffer-val] */
2065         parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
2066         if (IS_ERR(parent)) {
2067                 CERROR("Can not find the parent "DFID": rc = %ld\n",
2068                         PFID(fid), PTR_ERR(parent));
2069                 RETURN(parent);
2070         }
2071
2072         while (1) {
2073                 struct lu_object *ld_parent;
2074                 char *e;
2075
2076                 e = strsep(&path, "/");
2077                 if (e == NULL)
2078                         break;
2079
2080                 if (e[0] == 0) {
2081                         if (!path || path[0] == '\0')
2082                                 break;
2083                         continue;
2084                 }
2085
2086                 lname->ln_name = e;
2087                 lname->ln_namelen = strlen(e);
2088
2089                 ld_parent = lu_object_locate(parent->lo_header, ld->ld_type);
2090                 if (ld_parent == NULL) {
2091                         lu_object_put(env, parent);
2092                         rc = -EINVAL;
2093                         break;
2094                 }
2095
2096                 child = echo_md_lookup(env, ed, lu2md(ld_parent), lname);
2097                 lu_object_put(env, parent);
2098                 if (IS_ERR(child)) {
2099                         rc = (int)PTR_ERR(child);
2100                         CERROR("lookup %s under parent "DFID": rc = %d\n",
2101                                 lname->ln_name, PFID(lu_object_fid(ld_parent)),
2102                                 rc);
2103                         break;
2104                 }
2105                 parent = child;
2106         }
2107         if (rc)
2108                 RETURN(ERR_PTR(rc));
2109
2110         RETURN(parent);
2111 }
2112
2113 static void echo_ucred_init(struct lu_env *env)
2114 {
2115         struct lu_ucred *ucred = lu_ucred(env);
2116
2117         ucred->uc_valid = UCRED_INVALID;
2118
2119         ucred->uc_suppgids[0] = -1;
2120         ucred->uc_suppgids[1] = -1;
2121
2122         ucred->uc_uid   = ucred->uc_o_uid   = current_uid();
2123         ucred->uc_gid   = ucred->uc_o_gid   = current_gid();
2124         ucred->uc_fsuid = ucred->uc_o_fsuid = current_fsuid();
2125         ucred->uc_fsgid = ucred->uc_o_fsgid = current_fsgid();
2126         ucred->uc_cap   = cfs_curproc_cap_pack();
2127
2128         /* remove fs privilege for non-root user. */
2129         if (ucred->uc_fsuid)
2130                 ucred->uc_cap &= ~CFS_CAP_FS_MASK;
2131         ucred->uc_valid = UCRED_NEW;
2132 }
2133
2134 static void echo_ucred_fini(struct lu_env *env)
2135 {
2136         struct lu_ucred *ucred = lu_ucred(env);
2137         ucred->uc_valid = UCRED_INIT;
2138 }
2139
2140 #define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_MD_THREAD)
2141 #define ECHO_MD_SES_TAG (LCT_REMEMBER | LCT_SESSION | LCT_SERVER_SESSION)
2142 static int echo_md_handler(struct echo_device *ed, int command,
2143                            char *path, int path_len, __u64 id, int count,
2144                            struct obd_ioctl_data *data)
2145 {
2146         struct echo_thread_info *info;
2147         struct lu_device      *ld = ed->ed_next;
2148         struct lu_env         *env;
2149         int                    refcheck;
2150         struct lu_object      *parent;
2151         char                  *name = NULL;
2152         int                    namelen = data->ioc_plen2;
2153         int                    rc = 0;
2154         ENTRY;
2155
2156         if (ld == NULL) {
2157                 CERROR("MD echo client is not being initialized properly\n");
2158                 RETURN(-EINVAL);
2159         }
2160
2161         if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
2162                 CERROR("Only support MDD layer right now!\n");
2163                 RETURN(-EINVAL);
2164         }
2165
2166         env = cl_env_get(&refcheck);
2167         if (IS_ERR(env))
2168                 RETURN(PTR_ERR(env));
2169
2170         rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG);
2171         if (rc != 0)
2172                 GOTO(out_env, rc);
2173
2174         /* init big_lmm buffer */
2175         info = echo_env_info(env);
2176         LASSERT(info->eti_big_lmm == NULL);
2177         OBD_ALLOC_LARGE(info->eti_big_lmm, MIN_MD_SIZE);
2178         if (info->eti_big_lmm == NULL)
2179                 GOTO(out_env, rc = -ENOMEM);
2180         info->eti_big_lmmsize = MIN_MD_SIZE;
2181
2182         parent = echo_resolve_path(env, ed, path, path_len);
2183         if (IS_ERR(parent)) {
2184                 CERROR("Can not resolve the path %s: rc = %ld\n", path,
2185                         PTR_ERR(parent));
2186                 GOTO(out_free, rc = PTR_ERR(parent));
2187         }
2188
2189         if (namelen > 0) {
2190                 OBD_ALLOC(name, namelen + 1);
2191                 if (name == NULL)
2192                         GOTO(out_put, rc = -ENOMEM);
2193                 if (copy_from_user(name, data->ioc_pbuf2, namelen))
2194                         GOTO(out_name, rc = -EFAULT);
2195         }
2196
2197         echo_ucred_init(env);
2198
2199         switch (command) {
2200         case ECHO_MD_CREATE:
2201         case ECHO_MD_MKDIR: {
2202                 struct echo_thread_info *info = echo_env_info(env);
2203                 __u32 mode = data->ioc_obdo2.o_mode;
2204                 struct lu_fid *fid = &info->eti_fid;
2205                 int stripe_count = (int)data->ioc_obdo2.o_misc;
2206                 int stripe_index = (int)data->ioc_obdo2.o_stripe_idx;
2207
2208                 rc = ostid_to_fid(fid, &data->ioc_obdo1.o_oi, 0);
2209                 if (rc != 0)
2210                         break;
2211
2212                 /* In the function below, .hs_keycmp resolves to
2213                  * lu_obj_hop_keycmp() */
2214                 /* coverity[overrun-buffer-val] */
2215                 rc = echo_create_md_object(env, ed, parent, fid, name, namelen,
2216                                            id, mode, count, stripe_count,
2217                                            stripe_index);
2218                 break;
2219         }
2220         case ECHO_MD_DESTROY:
2221         case ECHO_MD_RMDIR: {
2222                 __u32 mode = data->ioc_obdo2.o_mode;
2223
2224                 rc = echo_destroy_object(env, ed, parent, name, namelen,
2225                                          id, mode, count);
2226                 break;
2227         }
2228         case ECHO_MD_LOOKUP:
2229                 rc = echo_lookup_object(env, ed, parent, id, count);
2230                 break;
2231         case ECHO_MD_GETATTR:
2232                 rc = echo_getattr_object(env, ed, parent, id, count);
2233                 break;
2234         case ECHO_MD_SETATTR:
2235                 rc = echo_setattr_object(env, ed, parent, id, count);
2236                 break;
2237         default:
2238                 CERROR("unknown command %d\n", command);
2239                 rc = -EINVAL;
2240                 break;
2241         }
2242         echo_ucred_fini(env);
2243
2244 out_name:
2245         if (name != NULL)
2246                 OBD_FREE(name, namelen + 1);
2247 out_put:
2248         lu_object_put(env, parent);
2249 out_free:
2250         LASSERT(info->eti_big_lmm);
2251         OBD_FREE_LARGE(info->eti_big_lmm, info->eti_big_lmmsize);
2252         info->eti_big_lmm = NULL;
2253         info->eti_big_lmmsize = 0;
2254 out_env:
2255         cl_env_put(env, &refcheck);
2256         return rc;
2257 }
2258 #endif /* HAVE_SERVER_SUPPORT */
2259
2260 static int echo_create_object(const struct lu_env *env, struct echo_device *ed,
2261                               int on_target, struct obdo *oa, void *ulsm,
2262                               int ulsm_nob, struct obd_trans_info *oti)
2263 {
2264         struct echo_object     *eco;
2265         struct echo_client_obd *ec = ed->ed_ec;
2266         struct lov_stripe_md   *lsm = NULL;
2267         int                     rc;
2268         int                     created = 0;
2269         ENTRY;
2270
2271         if ((oa->o_valid & OBD_MD_FLID) == 0 && /* no obj id */
2272             (on_target ||                       /* set_stripe */
2273              ec->ec_nstripes != 0)) {           /* LOV */
2274                 CERROR ("No valid oid\n");
2275                 RETURN(-EINVAL);
2276         }
2277
2278         rc = echo_alloc_memmd(ed, &lsm);
2279         if (rc < 0) {
2280                 CERROR("Cannot allocate md: rc = %d\n", rc);
2281                 GOTO(failed, rc);
2282         }
2283
2284         if (ulsm != NULL) {
2285                 int i, idx;
2286
2287                 rc = echo_copyin_lsm (ed, lsm, ulsm, ulsm_nob);
2288                 if (rc != 0)
2289                         GOTO(failed, rc);
2290
2291                 if (lsm->lsm_stripe_count == 0)
2292                         lsm->lsm_stripe_count = ec->ec_nstripes;
2293
2294                 if (lsm->lsm_stripe_size == 0)
2295                         lsm->lsm_stripe_size = PAGE_CACHE_SIZE;
2296
2297                 idx = cfs_rand();
2298
2299                 /* setup stripes: indices + default ids if required */
2300                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2301                         if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) == 0)
2302                                 lsm->lsm_oinfo[i]->loi_oi = lsm->lsm_oi;
2303
2304                         lsm->lsm_oinfo[i]->loi_ost_idx =
2305                                 (idx + i) % ec->ec_nstripes;
2306                 }
2307         }
2308
2309         /* setup object ID here for !on_target and LOV hint */
2310         if (oa->o_valid & OBD_MD_FLID) {
2311                 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2312                 lsm->lsm_oi = oa->o_oi;
2313         }
2314
2315         if (ostid_id(&lsm->lsm_oi) == 0)
2316                 ostid_set_id(&lsm->lsm_oi, ++last_object_id);
2317
2318         rc = 0;
2319         if (on_target) {
2320                 /* Only echo objects are allowed to be created */
2321                 LASSERT((oa->o_valid & OBD_MD_FLGROUP) &&
2322                         (ostid_seq(&oa->o_oi) == FID_SEQ_ECHO));
2323                 rc = obd_create(env, ec->ec_exp, oa, &lsm, oti);
2324                 if (rc != 0) {
2325                         CERROR("Cannot create objects: rc = %d\n", rc);
2326                         GOTO(failed, rc);
2327                 }
2328                 created = 1;
2329         }
2330
2331         /* See what object ID we were given */
2332         oa->o_oi = lsm->lsm_oi;
2333         oa->o_valid |= OBD_MD_FLID;
2334
2335         eco = cl_echo_object_find(ed, &lsm);
2336         if (IS_ERR(eco))
2337                 GOTO(failed, rc = PTR_ERR(eco));
2338         cl_echo_object_put(eco);
2339
2340         CDEBUG(D_INFO, "oa oid "DOSTID"\n", POSTID(&oa->o_oi));
2341         EXIT;
2342
2343  failed:
2344         if (created && rc)
2345                 obd_destroy(env, ec->ec_exp, oa, lsm, oti, NULL, NULL);
2346         if (lsm)
2347                 echo_free_memmd(ed, &lsm);
2348         if (rc)
2349                 CERROR("create object failed with: rc = %d\n", rc);
2350         return (rc);
2351 }
2352
2353 static int echo_get_object(struct echo_object **ecop, struct echo_device *ed,
2354                            struct obdo *oa)
2355 {
2356         struct lov_stripe_md   *lsm = NULL;
2357         struct echo_object     *eco;
2358         int                     rc;
2359         ENTRY;
2360
2361         if ((oa->o_valid & OBD_MD_FLID) == 0 || ostid_id(&oa->o_oi) == 0) {
2362                 /* disallow use of object id 0 */
2363                 CERROR ("No valid oid\n");
2364                 RETURN(-EINVAL);
2365         }
2366
2367         rc = echo_alloc_memmd(ed, &lsm);
2368         if (rc < 0)
2369                 RETURN(rc);
2370
2371         lsm->lsm_oi = oa->o_oi;
2372         if (!(oa->o_valid & OBD_MD_FLGROUP))
2373                 ostid_set_seq_echo(&lsm->lsm_oi);
2374
2375         rc = 0;
2376         eco = cl_echo_object_find(ed, &lsm);
2377         if (!IS_ERR(eco))
2378                 *ecop = eco;
2379         else
2380                 rc = PTR_ERR(eco);
2381         if (lsm)
2382                 echo_free_memmd(ed, &lsm);
2383         RETURN(rc);
2384 }
2385
2386 static void echo_put_object(struct echo_object *eco)
2387 {
2388         if (cl_echo_object_put(eco))
2389                 CERROR("echo client: drop an object failed");
2390 }
2391
2392 static void
2393 echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp)
2394 {
2395         unsigned long stripe_count;
2396         unsigned long stripe_size;
2397         unsigned long width;
2398         unsigned long woffset;
2399         int           stripe_index;
2400         obd_off       offset;
2401
2402         if (lsm->lsm_stripe_count <= 1)
2403                 return;
2404
2405         offset       = *offp;
2406         stripe_size  = lsm->lsm_stripe_size;
2407         stripe_count = lsm->lsm_stripe_count;
2408
2409         /* width = # bytes in all stripes */
2410         width = stripe_size * stripe_count;
2411
2412         /* woffset = offset within a width; offset = whole number of widths */
2413         woffset = do_div (offset, width);
2414
2415         stripe_index = woffset / stripe_size;
2416
2417         *idp = ostid_id(&lsm->lsm_oinfo[stripe_index]->loi_oi);
2418         *offp = offset * stripe_size + woffset % stripe_size;
2419 }
2420
2421 static void
2422 echo_client_page_debug_setup(struct lov_stripe_md *lsm,
2423                              struct page *page, int rw, obd_id id,
2424                              obd_off offset, obd_off count)
2425 {
2426         char    *addr;
2427         obd_off  stripe_off;
2428         obd_id   stripe_id;
2429         int      delta;
2430
2431         /* no partial pages on the client */
2432         LASSERT(count == PAGE_CACHE_SIZE);
2433
2434         addr = kmap(page);
2435
2436         for (delta = 0; delta < PAGE_CACHE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2437                 if (rw == OBD_BRW_WRITE) {
2438                         stripe_off = offset + delta;
2439                         stripe_id = id;
2440                         echo_get_stripe_off_id(lsm, &stripe_off, &stripe_id);
2441                 } else {
2442                         stripe_off = 0xdeadbeef00c0ffeeULL;
2443                         stripe_id = 0xdeadbeef00c0ffeeULL;
2444                 }
2445                 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
2446                                   stripe_off, stripe_id);
2447         }
2448
2449         kunmap(page);
2450 }
2451
2452 static int echo_client_page_debug_check(struct lov_stripe_md *lsm,
2453                                         struct page *page, obd_id id,
2454                                         obd_off offset, obd_off count)
2455 {
2456         obd_off stripe_off;
2457         obd_id  stripe_id;
2458         char   *addr;
2459         int     delta;
2460         int     rc;
2461         int     rc2;
2462
2463         /* no partial pages on the client */
2464         LASSERT(count == PAGE_CACHE_SIZE);
2465
2466         addr = kmap(page);
2467
2468         for (rc = delta = 0; delta < PAGE_CACHE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2469                 stripe_off = offset + delta;
2470                 stripe_id = id;
2471                 echo_get_stripe_off_id (lsm, &stripe_off, &stripe_id);
2472
2473                 rc2 = block_debug_check("test_brw",
2474                                         addr + delta, OBD_ECHO_BLOCK_SIZE,
2475                                         stripe_off, stripe_id);
2476                 if (rc2 != 0) {
2477                         CERROR ("Error in echo object "LPX64"\n", id);
2478                         rc = rc2;
2479                 }
2480         }
2481
2482         kunmap(page);
2483         return rc;
2484 }
2485
2486 static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa,
2487                             struct echo_object *eco, obd_off offset,
2488                             obd_size count, int async,
2489                             struct obd_trans_info *oti)
2490 {
2491         struct lov_stripe_md   *lsm = eco->eo_lsm;
2492         obd_count               npages;
2493         struct brw_page        *pga;
2494         struct brw_page        *pgp;
2495         struct page            **pages;
2496         obd_off                 off;
2497         int                     i;
2498         int                     rc;
2499         int                     verify;
2500         int                     gfp_mask;
2501         int                     brw_flags = 0;
2502         ENTRY;
2503
2504         verify = (ostid_id(&oa->o_oi) != ECHO_PERSISTENT_OBJID &&
2505                   (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
2506                   (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
2507
2508         gfp_mask = ((ostid_id(&oa->o_oi) & 2) == 0) ? GFP_IOFS : GFP_HIGHUSER;
2509
2510         LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
2511         LASSERT(lsm != NULL);
2512         LASSERT(ostid_id(&lsm->lsm_oi) == ostid_id(&oa->o_oi));
2513
2514         if (count <= 0 ||
2515             (count & (~CFS_PAGE_MASK)) != 0)
2516                 RETURN(-EINVAL);
2517
2518         /* XXX think again with misaligned I/O */
2519         npages = count >> PAGE_CACHE_SHIFT;
2520
2521         if (rw == OBD_BRW_WRITE)
2522                 brw_flags = OBD_BRW_ASYNC;
2523
2524         OBD_ALLOC(pga, npages * sizeof(*pga));
2525         if (pga == NULL)
2526                 RETURN(-ENOMEM);
2527
2528         OBD_ALLOC(pages, npages * sizeof(*pages));
2529         if (pages == NULL) {
2530                 OBD_FREE(pga, npages * sizeof(*pga));
2531                 RETURN(-ENOMEM);
2532         }
2533
2534         for (i = 0, pgp = pga, off = offset;
2535              i < npages;
2536              i++, pgp++, off += PAGE_CACHE_SIZE) {
2537
2538                 LASSERT (pgp->pg == NULL);      /* for cleanup */
2539
2540                 rc = -ENOMEM;
2541                 OBD_PAGE_ALLOC(pgp->pg, gfp_mask);
2542                 if (pgp->pg == NULL)
2543                         goto out;
2544
2545                 pages[i] = pgp->pg;
2546                 pgp->count = PAGE_CACHE_SIZE;
2547                 pgp->off = off;
2548                 pgp->flag = brw_flags;
2549
2550                 if (verify)
2551                         echo_client_page_debug_setup(lsm, pgp->pg, rw,
2552                                                      ostid_id(&oa->o_oi), off,
2553                                                      pgp->count);
2554         }
2555
2556         /* brw mode can only be used at client */
2557         LASSERT(ed->ed_next != NULL);
2558         rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async);
2559
2560  out:
2561         if (rc != 0 || rw != OBD_BRW_READ)
2562                 verify = 0;
2563
2564         for (i = 0, pgp = pga; i < npages; i++, pgp++) {
2565                 if (pgp->pg == NULL)
2566                         continue;
2567
2568                 if (verify) {
2569                         int vrc;
2570                         vrc = echo_client_page_debug_check(lsm, pgp->pg,
2571                                                            ostid_id(&oa->o_oi),
2572                                                            pgp->off, pgp->count);
2573                         if (vrc != 0 && rc == 0)
2574                                 rc = vrc;
2575                 }
2576                 OBD_PAGE_FREE(pgp->pg);
2577         }
2578         OBD_FREE(pga, npages * sizeof(*pga));
2579         OBD_FREE(pages, npages * sizeof(*pages));
2580         RETURN(rc);
2581 }
2582
2583 static int echo_client_prep_commit(const struct lu_env *env,
2584                                    struct obd_export *exp, int rw,
2585                                    struct obdo *oa, struct echo_object *eco,
2586                                    obd_off offset, obd_size count,
2587                                    obd_size batch, struct obd_trans_info *oti,
2588                                    int async)
2589 {
2590         struct lov_stripe_md *lsm = eco->eo_lsm;
2591         struct obd_ioobj ioo;
2592         struct niobuf_local *lnb;
2593         struct niobuf_remote *rnb;
2594         obd_off off;
2595         obd_size npages, tot_pages;
2596         int i, ret = 0, brw_flags = 0;
2597
2598         ENTRY;
2599
2600         if (count <= 0 || (count & (~CFS_PAGE_MASK)) != 0 ||
2601             (lsm != NULL && ostid_id(&lsm->lsm_oi) != ostid_id(&oa->o_oi)))
2602                 RETURN(-EINVAL);
2603
2604         npages = batch >> PAGE_CACHE_SHIFT;
2605         tot_pages = count >> PAGE_CACHE_SHIFT;
2606
2607         OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
2608         OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote));
2609
2610         if (lnb == NULL || rnb == NULL)
2611                 GOTO(out, ret = -ENOMEM);
2612
2613         if (rw == OBD_BRW_WRITE && async)
2614                 brw_flags |= OBD_BRW_ASYNC;
2615
2616         obdo_to_ioobj(oa, &ioo);
2617
2618         off = offset;
2619
2620         for(; tot_pages; tot_pages -= npages) {
2621                 int lpages;
2622
2623                 if (tot_pages < npages)
2624                         npages = tot_pages;
2625
2626                 for (i = 0; i < npages; i++, off += PAGE_CACHE_SIZE) {
2627                         rnb[i].offset = off;
2628                         rnb[i].len = PAGE_CACHE_SIZE;
2629                         rnb[i].flags = brw_flags;
2630                 }
2631
2632                 ioo.ioo_bufcnt = npages;
2633                 oti->oti_transno = 0;
2634
2635                 lpages = npages;
2636                 ret = obd_preprw(env, rw, exp, oa, 1, &ioo, rnb, &lpages,
2637                                  lnb, oti, NULL);
2638                 if (ret != 0)
2639                         GOTO(out, ret);
2640                 LASSERT(lpages == npages);
2641
2642                 for (i = 0; i < lpages; i++) {
2643                         struct page *page = lnb[i].page;
2644
2645                         /* read past eof? */
2646                         if (page == NULL && lnb[i].rc == 0)
2647                                 continue;
2648
2649                         if (async)
2650                                 lnb[i].flags |= OBD_BRW_ASYNC;
2651
2652                         if (ostid_id(&oa->o_oi) == ECHO_PERSISTENT_OBJID ||
2653                             (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
2654                             (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
2655                                 continue;
2656
2657                         if (rw == OBD_BRW_WRITE)
2658                                 echo_client_page_debug_setup(lsm, page, rw,
2659                                                             ostid_id(&oa->o_oi),
2660                                                              rnb[i].offset,
2661                                                              rnb[i].len);
2662                         else
2663                                 echo_client_page_debug_check(lsm, page,
2664                                                             ostid_id(&oa->o_oi),
2665                                                              rnb[i].offset,
2666                                                              rnb[i].len);
2667                 }
2668
2669                 ret = obd_commitrw(env, rw, exp, oa, 1, &ioo,
2670                                    rnb, npages, lnb, oti, ret);
2671                 if (ret != 0)
2672                         GOTO(out, ret);
2673
2674                 /* Reset oti otherwise it would confuse ldiskfs. */
2675                 memset(oti, 0, sizeof(*oti));
2676
2677                 /* Reuse env context. */
2678                 lu_context_exit((struct lu_context *)&env->le_ctx);
2679                 lu_context_enter((struct lu_context *)&env->le_ctx);
2680         }
2681
2682 out:
2683         if (lnb)
2684                 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
2685         if (rnb)
2686                 OBD_FREE(rnb, npages * sizeof(struct niobuf_remote));
2687         RETURN(ret);
2688 }
2689
2690 static int echo_client_brw_ioctl(const struct lu_env *env, int rw,
2691                                  struct obd_export *exp,
2692                                  struct obd_ioctl_data *data,
2693                                  struct obd_trans_info *dummy_oti)
2694 {
2695         struct obd_device *obd = class_exp2obd(exp);
2696         struct echo_device *ed = obd2echo_dev(obd);
2697         struct echo_client_obd *ec = ed->ed_ec;
2698         struct obdo *oa = &data->ioc_obdo1;
2699         struct echo_object *eco;
2700         int rc;
2701         int async = 1;
2702         long test_mode;
2703         ENTRY;
2704
2705         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2706
2707         rc = echo_get_object(&eco, ed, oa);
2708         if (rc)
2709                 RETURN(rc);
2710
2711         oa->o_valid &= ~OBD_MD_FLHANDLE;
2712
2713         /* OFD/obdfilter works only via prep/commit */
2714         test_mode = (long)data->ioc_pbuf1;
2715         if (test_mode == 1)
2716                 async = 0;
2717
2718         if (ed->ed_next == NULL && test_mode != 3) {
2719                 test_mode = 3;
2720                 data->ioc_plen1 = data->ioc_count;
2721         }
2722
2723         /* Truncate batch size to maximum */
2724         if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE)
2725                 data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE;
2726
2727         switch (test_mode) {
2728         case 1:
2729                 /* fall through */
2730         case 2:
2731                 rc = echo_client_kbrw(ed, rw, oa,
2732                                       eco, data->ioc_offset,
2733                                       data->ioc_count, async, dummy_oti);
2734                 break;
2735         case 3:
2736                 rc = echo_client_prep_commit(env, ec->ec_exp, rw, oa,
2737                                              eco, data->ioc_offset,
2738                                              data->ioc_count, data->ioc_plen1,
2739                                              dummy_oti, async);
2740                 break;
2741         default:
2742                 rc = -EINVAL;
2743         }
2744         echo_put_object(eco);
2745         RETURN(rc);
2746 }
2747
2748 static int
2749 echo_client_enqueue(struct obd_export *exp, struct obdo *oa,
2750                     int mode, obd_off offset, obd_size nob)
2751 {
2752         struct echo_device     *ed = obd2echo_dev(exp->exp_obd);
2753         struct lustre_handle   *ulh = &oa->o_handle;
2754         struct echo_object     *eco;
2755         obd_off                 end;
2756         int                     rc;
2757         ENTRY;
2758
2759         if (ed->ed_next == NULL)
2760                 RETURN(-EOPNOTSUPP);
2761
2762         if (!(mode == LCK_PR || mode == LCK_PW))
2763                 RETURN(-EINVAL);
2764
2765         if ((offset & (~CFS_PAGE_MASK)) != 0 ||
2766             (nob & (~CFS_PAGE_MASK)) != 0)
2767                 RETURN(-EINVAL);
2768
2769         rc = echo_get_object (&eco, ed, oa);
2770         if (rc != 0)
2771                 RETURN(rc);
2772
2773         end = (nob == 0) ? ((obd_off) -1) : (offset + nob - 1);
2774         rc = cl_echo_enqueue(eco, offset, end, mode, &ulh->cookie);
2775         if (rc == 0) {
2776                 oa->o_valid |= OBD_MD_FLHANDLE;
2777                 CDEBUG(D_INFO, "Cookie is "LPX64"\n", ulh->cookie);
2778         }
2779         echo_put_object(eco);
2780         RETURN(rc);
2781 }
2782
2783 static int
2784 echo_client_cancel(struct obd_export *exp, struct obdo *oa)
2785 {
2786         struct echo_device *ed     = obd2echo_dev(exp->exp_obd);
2787         __u64               cookie = oa->o_handle.cookie;
2788
2789         if ((oa->o_valid & OBD_MD_FLHANDLE) == 0)
2790                 return -EINVAL;
2791
2792         CDEBUG(D_INFO, "Cookie is "LPX64"\n", cookie);
2793         return cl_echo_cancel(ed, cookie);
2794 }
2795
2796 static int
2797 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2798                       void *karg, void *uarg)
2799 {
2800 #ifdef HAVE_SERVER_SUPPORT
2801         struct tgt_session_info *tsi;
2802 #endif
2803         struct obd_device      *obd = exp->exp_obd;
2804         struct echo_device     *ed = obd2echo_dev(obd);
2805         struct echo_client_obd *ec = ed->ed_ec;
2806         struct echo_object     *eco;
2807         struct obd_ioctl_data  *data = karg;
2808         struct obd_trans_info   dummy_oti;
2809         struct lu_env          *env;
2810         struct oti_req_ack_lock *ack_lock;
2811         struct obdo            *oa;
2812         struct lu_fid           fid;
2813         int                     rw = OBD_BRW_READ;
2814         int                     rc = 0;
2815         int                     i;
2816 #ifdef HAVE_SERVER_SUPPORT
2817         struct lu_context        echo_session;
2818 #endif
2819         ENTRY;
2820
2821         memset(&dummy_oti, 0, sizeof(dummy_oti));
2822
2823         oa = &data->ioc_obdo1;
2824         if (!(oa->o_valid & OBD_MD_FLGROUP)) {
2825                 oa->o_valid |= OBD_MD_FLGROUP;
2826                 ostid_set_seq_echo(&oa->o_oi);
2827         }
2828
2829         /* This FID is unpacked just for validation at this point */
2830         rc = ostid_to_fid(&fid, &oa->o_oi, 0);
2831         if (rc < 0)
2832                 RETURN(rc);
2833
2834         OBD_ALLOC_PTR(env);
2835         if (env == NULL)
2836                 RETURN(-ENOMEM);
2837
2838         rc = lu_env_init(env, LCT_DT_THREAD);
2839         if (rc)
2840                 GOTO(out_alloc, rc = -ENOMEM);
2841
2842 #ifdef HAVE_SERVER_SUPPORT
2843         env->le_ses = &echo_session;
2844         rc = lu_context_init(env->le_ses, LCT_SERVER_SESSION | LCT_NOREF);
2845         if (unlikely(rc < 0))
2846                 GOTO(out_env, rc);
2847         lu_context_enter(env->le_ses);
2848
2849         tsi = tgt_ses_info(env);
2850         tsi->tsi_exp = ec->ec_exp;
2851 #endif
2852         switch (cmd) {
2853         case OBD_IOC_CREATE:                    /* may create echo object */
2854                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2855                         GOTO (out, rc = -EPERM);
2856
2857                 rc = echo_create_object(env, ed, 1, oa, data->ioc_pbuf1,
2858                                         data->ioc_plen1, &dummy_oti);
2859                 GOTO(out, rc);
2860
2861 #ifdef HAVE_SERVER_SUPPORT
2862         case OBD_IOC_ECHO_MD: {
2863                 int count;
2864                 int cmd;
2865                 char *dir = NULL;
2866                 int dirlen;
2867                 __u64 id;
2868
2869                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2870                         GOTO(out, rc = -EPERM);
2871
2872                 count = data->ioc_count;
2873                 cmd = data->ioc_command;
2874
2875                 id = data->ioc_obdo2.o_oi.oi.oi_id;
2876                 dirlen = data->ioc_plen1;
2877                 OBD_ALLOC(dir, dirlen + 1);
2878                 if (dir == NULL)
2879                         GOTO(out, rc = -ENOMEM);
2880
2881                 if (copy_from_user(dir, data->ioc_pbuf1, dirlen)) {
2882                         OBD_FREE(dir, data->ioc_plen1 + 1);
2883                         GOTO(out, rc = -EFAULT);
2884                 }
2885
2886                 rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data);
2887                 OBD_FREE(dir, dirlen + 1);
2888                 GOTO(out, rc);
2889         }
2890         case OBD_IOC_ECHO_ALLOC_SEQ: {
2891                 struct lu_env   *cl_env;
2892                 int              refcheck;
2893                 __u64            seq;
2894                 int              max_count;
2895
2896                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2897                         GOTO(out, rc = -EPERM);
2898
2899                 cl_env = cl_env_get(&refcheck);
2900                 if (IS_ERR(cl_env))
2901                         GOTO(out, rc = PTR_ERR(cl_env));
2902
2903                 rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG,
2904                                             ECHO_MD_SES_TAG);
2905                 if (rc != 0) {
2906                         cl_env_put(cl_env, &refcheck);
2907                         GOTO(out, rc);
2908                 }
2909
2910                 rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq);
2911                 cl_env_put(cl_env, &refcheck);
2912                 if (rc < 0) {
2913                         CERROR("%s: Can not alloc seq: rc = %d\n",
2914                                obd->obd_name, rc);
2915                         GOTO(out, rc);
2916                 }
2917
2918                 if (copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1))
2919                         return -EFAULT;
2920
2921                 max_count = LUSTRE_METADATA_SEQ_MAX_WIDTH;
2922                 if (copy_to_user(data->ioc_pbuf2, &max_count,
2923                                      data->ioc_plen2))
2924                         return -EFAULT;
2925                 GOTO(out, rc);
2926         }
2927 #endif /* HAVE_SERVER_SUPPORT */
2928         case OBD_IOC_DESTROY:
2929                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2930                         GOTO (out, rc = -EPERM);
2931
2932                 rc = echo_get_object(&eco, ed, oa);
2933                 if (rc == 0) {
2934                         rc = obd_destroy(env, ec->ec_exp, oa, eco->eo_lsm,
2935                                          &dummy_oti, NULL, NULL);
2936                         if (rc == 0)
2937                                 eco->eo_deleted = 1;
2938                         echo_put_object(eco);
2939                 }
2940                 GOTO(out, rc);
2941
2942         case OBD_IOC_GETATTR:
2943                 rc = echo_get_object(&eco, ed, oa);
2944                 if (rc == 0) {
2945                         struct obd_info oinfo = { { { 0 } } };
2946                         oinfo.oi_md = eco->eo_lsm;
2947                         oinfo.oi_oa = oa;
2948                         rc = obd_getattr(env, ec->ec_exp, &oinfo);
2949                         echo_put_object(eco);
2950                 }
2951                 GOTO(out, rc);
2952
2953         case OBD_IOC_SETATTR:
2954                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2955                         GOTO (out, rc = -EPERM);
2956
2957                 rc = echo_get_object(&eco, ed, oa);
2958                 if (rc == 0) {
2959                         struct obd_info oinfo = { { { 0 } } };
2960                         oinfo.oi_oa = oa;
2961                         oinfo.oi_md = eco->eo_lsm;
2962
2963                         rc = obd_setattr(env, ec->ec_exp, &oinfo, NULL);
2964                         echo_put_object(eco);
2965                 }
2966                 GOTO(out, rc);
2967
2968         case OBD_IOC_BRW_WRITE:
2969                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2970                         GOTO (out, rc = -EPERM);
2971
2972                 rw = OBD_BRW_WRITE;
2973                 /* fall through */
2974         case OBD_IOC_BRW_READ:
2975                 rc = echo_client_brw_ioctl(env, rw, exp, data, &dummy_oti);
2976                 GOTO(out, rc);
2977
2978         case ECHO_IOC_GET_STRIPE:
2979                 rc = echo_get_object(&eco, ed, oa);
2980                 if (rc == 0) {
2981                         rc = echo_copyout_lsm(eco->eo_lsm, data->ioc_pbuf1,
2982                                               data->ioc_plen1);
2983                         echo_put_object(eco);
2984                 }
2985                 GOTO(out, rc);
2986
2987         case ECHO_IOC_SET_STRIPE:
2988                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2989                         GOTO (out, rc = -EPERM);
2990
2991                 if (data->ioc_pbuf1 == NULL) {  /* unset */
2992                         rc = echo_get_object(&eco, ed, oa);
2993                         if (rc == 0) {
2994                                 eco->eo_deleted = 1;
2995                                 echo_put_object(eco);
2996                         }
2997                 } else {
2998                         rc = echo_create_object(env, ed, 0, oa,
2999                                                 data->ioc_pbuf1,
3000                                                 data->ioc_plen1, &dummy_oti);
3001                 }
3002                 GOTO (out, rc);
3003
3004         case ECHO_IOC_ENQUEUE:
3005                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
3006                         GOTO (out, rc = -EPERM);
3007
3008                 rc = echo_client_enqueue(exp, oa,
3009                                          data->ioc_conn1, /* lock mode */
3010                                          data->ioc_offset,
3011                                          data->ioc_count);/*extent*/
3012                 GOTO (out, rc);
3013
3014         case ECHO_IOC_CANCEL:
3015                 rc = echo_client_cancel(exp, oa);
3016                 GOTO (out, rc);
3017
3018         default:
3019                 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
3020                 GOTO (out, rc = -ENOTTY);
3021         }
3022
3023         EXIT;
3024 out:
3025 #ifdef HAVE_SERVER_SUPPORT
3026         lu_context_exit(env->le_ses);
3027         lu_context_fini(env->le_ses);
3028 out_env:
3029 #endif
3030         lu_env_fini(env);
3031 out_alloc:
3032         OBD_FREE_PTR(env);
3033
3034         /* XXX this should be in a helper also called by target_send_reply */
3035         for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4;
3036              i++, ack_lock++) {
3037                 if (!ack_lock->mode)
3038                         break;
3039                 ldlm_lock_decref(&ack_lock->lock, ack_lock->mode);
3040         }
3041
3042         return rc;
3043 }
3044
3045 static int echo_client_setup(const struct lu_env *env,
3046                              struct obd_device *obddev, struct lustre_cfg *lcfg)
3047 {
3048         struct echo_client_obd *ec = &obddev->u.echo_client;
3049         struct obd_device *tgt;
3050         struct obd_uuid echo_uuid = { "ECHO_UUID" };
3051         struct obd_connect_data *ocd = NULL;
3052         int rc;
3053         ENTRY;
3054
3055         if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
3056                 CERROR("requires a TARGET OBD name\n");
3057                 RETURN(-EINVAL);
3058         }
3059
3060         tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
3061         if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
3062                 CERROR("device not attached or not set up (%s)\n",
3063                        lustre_cfg_string(lcfg, 1));
3064                 RETURN(-EINVAL);
3065         }
3066
3067         spin_lock_init(&ec->ec_lock);
3068         CFS_INIT_LIST_HEAD (&ec->ec_objects);
3069         CFS_INIT_LIST_HEAD (&ec->ec_locks);
3070         ec->ec_unique = 0;
3071         ec->ec_nstripes = 0;
3072
3073         if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
3074 #ifdef HAVE_SERVER_SUPPORT
3075                 lu_context_tags_update(ECHO_MD_CTX_TAG);
3076                 lu_session_tags_update(ECHO_MD_SES_TAG);
3077 #else
3078                 CERROR("Local operations are NOT supported on client side. "
3079                        "Only remote operations are supported. Metadata client "
3080                        "must be run on server side.\n");
3081 #endif
3082                 RETURN(0);
3083         }
3084
3085         OBD_ALLOC(ocd, sizeof(*ocd));
3086         if (ocd == NULL) {
3087                 CERROR("Can't alloc ocd connecting to %s\n",
3088                        lustre_cfg_string(lcfg, 1));
3089                 return -ENOMEM;
3090         }
3091
3092         ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL |
3093                                  OBD_CONNECT_BRW_SIZE |
3094                                  OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 |
3095                                  OBD_CONNECT_64BITHASH | OBD_CONNECT_LVB_TYPE |
3096                                  OBD_CONNECT_FID;
3097         ocd->ocd_brw_size = DT_MAX_BRW_SIZE;
3098         ocd->ocd_version = LUSTRE_VERSION_CODE;
3099         ocd->ocd_group = FID_SEQ_ECHO;
3100
3101         rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL);
3102         if (rc == 0) {
3103                 /* Turn off pinger because it connects to tgt obd directly. */
3104                 spin_lock(&tgt->obd_dev_lock);
3105                 cfs_list_del_init(&ec->ec_exp->exp_obd_chain_timed);
3106                 spin_unlock(&tgt->obd_dev_lock);
3107         }
3108
3109         OBD_FREE(ocd, sizeof(*ocd));
3110
3111         if (rc != 0) {
3112                 CERROR("fail to connect to device %s\n",
3113                        lustre_cfg_string(lcfg, 1));
3114                 return (rc);
3115         }
3116
3117         RETURN(rc);
3118 }
3119
3120 static int echo_client_cleanup(struct obd_device *obddev)
3121 {
3122         struct echo_device *ed = obd2echo_dev(obddev);
3123         struct echo_client_obd *ec = &obddev->u.echo_client;
3124         int rc;
3125         ENTRY;
3126
3127         /*Do nothing for Metadata echo client*/
3128         if (ed == NULL )
3129                 RETURN(0);
3130
3131         if (ed->ed_next_ismd) {
3132 #ifdef HAVE_SERVER_SUPPORT
3133                 lu_context_tags_clear(ECHO_MD_CTX_TAG);
3134                 lu_session_tags_clear(ECHO_MD_SES_TAG);
3135 #else
3136                 CERROR("This is client-side only module, does not support "
3137                         "metadata echo client.\n");
3138 #endif
3139                 RETURN(0);
3140         }
3141
3142         if (!cfs_list_empty(&obddev->obd_exports)) {
3143                 CERROR("still has clients!\n");
3144                 RETURN(-EBUSY);
3145         }
3146
3147         LASSERT(cfs_atomic_read(&ec->ec_exp->exp_refcount) > 0);
3148         rc = obd_disconnect(ec->ec_exp);
3149         if (rc != 0)
3150                 CERROR("fail to disconnect device: %d\n", rc);
3151
3152         RETURN(rc);
3153 }
3154
3155 static int echo_client_connect(const struct lu_env *env,
3156                                struct obd_export **exp,
3157                                struct obd_device *src, struct obd_uuid *cluuid,
3158                                struct obd_connect_data *data, void *localdata)
3159 {
3160         int                rc;
3161         struct lustre_handle conn = { 0 };
3162
3163         ENTRY;
3164         rc = class_connect(&conn, src, cluuid);
3165         if (rc == 0) {
3166                 *exp = class_conn2export(&conn);
3167         }
3168
3169         RETURN (rc);
3170 }
3171
3172 static int echo_client_disconnect(struct obd_export *exp)
3173 {
3174 #if 0
3175         struct obd_device      *obd;
3176         struct echo_client_obd *ec;
3177         struct ec_lock         *ecl;
3178 #endif
3179         int                     rc;
3180         ENTRY;
3181
3182         if (exp == NULL)
3183                 GOTO(out, rc = -EINVAL);
3184
3185 #if 0
3186         obd = exp->exp_obd;
3187         ec = &obd->u.echo_client;
3188
3189         /* no more contention on export's lock list */
3190         while (!cfs_list_empty (&exp->exp_ec_data.eced_locks)) {
3191                 ecl = cfs_list_entry (exp->exp_ec_data.eced_locks.next,
3192                                       struct ec_lock, ecl_exp_chain);
3193                 cfs_list_del (&ecl->ecl_exp_chain);
3194
3195                 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm,
3196                                  ecl->ecl_mode, &ecl->ecl_lock_handle);
3197
3198                 CDEBUG (D_INFO, "Cancel lock on object "LPX64" on disconnect "
3199                         "(%d)\n", ecl->ecl_object->eco_id, rc);
3200
3201                 echo_put_object (ecl->ecl_object);
3202                 OBD_FREE (ecl, sizeof (*ecl));
3203         }
3204 #endif
3205
3206         rc = class_disconnect(exp);
3207         GOTO(out, rc);
3208  out:
3209         return rc;
3210 }
3211
3212 static struct obd_ops echo_client_obd_ops = {
3213         .o_owner       = THIS_MODULE,
3214
3215 #if 0
3216         .o_setup       = echo_client_setup,
3217         .o_cleanup     = echo_client_cleanup,
3218 #endif
3219
3220         .o_iocontrol   = echo_client_iocontrol,
3221         .o_connect     = echo_client_connect,
3222         .o_disconnect  = echo_client_disconnect
3223 };
3224
3225 int echo_client_init(void)
3226 {
3227         struct lprocfs_static_vars lvars = { 0 };
3228         int rc;
3229
3230         lprocfs_echo_init_vars(&lvars);
3231
3232         rc = lu_kmem_init(echo_caches);
3233         if (rc == 0) {
3234                 rc = class_register_type(&echo_client_obd_ops, NULL,
3235                                          lvars.module_vars,
3236                                          LUSTRE_ECHO_CLIENT_NAME,
3237                                          &echo_device_type);
3238                 if (rc)
3239                         lu_kmem_fini(echo_caches);
3240         }
3241         return rc;
3242 }
3243
3244 void echo_client_exit(void)
3245 {
3246         class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
3247         lu_kmem_fini(echo_caches);
3248 }
3249
3250 #ifdef __KERNEL__
3251 static int __init obdecho_init(void)
3252 {
3253         struct lprocfs_static_vars lvars;
3254         int rc;
3255
3256         ENTRY;
3257         LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
3258
3259         LASSERT(PAGE_CACHE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
3260
3261         lprocfs_echo_init_vars(&lvars);
3262
3263 # ifdef HAVE_SERVER_SUPPORT
3264         rc = echo_persistent_pages_init();
3265         if (rc != 0)
3266                 goto failed_0;
3267
3268         rc = class_register_type(&echo_obd_ops, NULL, lvars.module_vars,
3269                                  LUSTRE_ECHO_NAME, NULL);
3270         if (rc != 0)
3271                 goto failed_1;
3272 # endif
3273
3274         rc = echo_client_init();
3275
3276 # ifdef HAVE_SERVER_SUPPORT
3277         if (rc == 0)
3278                 RETURN(0);
3279
3280         class_unregister_type(LUSTRE_ECHO_NAME);
3281 failed_1:
3282         echo_persistent_pages_fini();
3283 failed_0:
3284 # endif
3285         RETURN(rc);
3286 }
3287
3288 static void /*__exit*/ obdecho_exit(void)
3289 {
3290         echo_client_exit();
3291
3292 # ifdef HAVE_SERVER_SUPPORT
3293         class_unregister_type(LUSTRE_ECHO_NAME);
3294         echo_persistent_pages_fini();
3295 # endif
3296 }
3297
3298 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3299 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
3300 MODULE_LICENSE("GPL");
3301
3302 cfs_module(obdecho, LUSTRE_VERSION_STRING, obdecho_init, obdecho_exit);
3303 #endif /* __KERNEL__ */
3304
3305 /** @} echo_client */