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, 2014, 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
  39 #include <linux/user_namespace.h>
  40 #ifdef HAVE_UIDGID_HEADER
  41 # include <linux/uidgid.h>
  42 #endif
  43 #include <libcfs/libcfs.h>
  44
  45 #include <obd.h>
  46 #include <obd_support.h>
  47 #include <obd_class.h>
  48 #include <lustre_debug.h>
  49 #include <lprocfs_status.h>
  50 #include <cl_object.h>
  51 #include <lustre_fid.h>
  52 #include <lustre_acl.h>
  53 #include <lustre_ioctl.h>
  54 #include <lustre_net.h>
  55 #ifdef HAVE_SERVER_SUPPORT
  56 # include <md_object.h>
  57 #endif /* HAVE_SERVER_SUPPORT */
  58
  59 #include "echo_internal.h"
  60
  61 /** \defgroup echo_client Echo Client
  62  * @{
  63  */
  64
  65 struct echo_device {
  66         struct cl_device          ed_cl;
  67         struct echo_client_obd   *ed_ec;
  68
  69         struct cl_site            ed_site_myself;
  70         struct lu_site           *ed_site;
  71         struct lu_device         *ed_next;
  72         int                       ed_next_ismd;
  73         struct lu_client_seq     *ed_cl_seq;
  74 #ifdef HAVE_SERVER_SUPPORT
  75         struct local_oid_storage *ed_los;
  76         struct lu_fid             ed_root_fid;
  77 #endif /* HAVE_SERVER_SUPPORT */
  78 };
  79
  80 struct echo_object {
  81         struct cl_object        eo_cl;
  82         struct cl_object_header eo_hdr;
  83         struct echo_device     *eo_dev;
  84         struct list_head        eo_obj_chain;
  85         struct lov_oinfo       *eo_oinfo;
  86         atomic_t                eo_npages;
  87         int                     eo_deleted;
  88 };
  89
  90 struct echo_object_conf {
  91         struct cl_object_conf   eoc_cl;
  92         struct lov_oinfo      **eoc_oinfo;
  93 };
  94
  95 struct echo_page {
  96         struct cl_page_slice    ep_cl;
  97         struct mutex            ep_lock;
  98 };
  99
 100 struct echo_lock {
 101         struct cl_lock_slice    el_cl;
 102         struct list_head        el_chain;
 103         struct echo_object     *el_object;
 104         __u64                   el_cookie;
 105         atomic_t                el_refcount;
 106 };
 107
 108 #ifdef HAVE_SERVER_SUPPORT
 109 static const char echo_md_root_dir_name[] = "ROOT_ECHO";
 110
 111 /**
 112  * In order to use the values of members in struct mdd_device,
 113  * we define an alias structure here.
 114  */
 115 struct echo_md_device {
 116         struct md_device                 emd_md_dev;
 117         struct obd_export               *emd_child_exp;
 118         struct dt_device                *emd_child;
 119         struct dt_device                *emd_bottom;
 120         struct lu_fid                    emd_root_fid;
 121         struct lu_fid                    emd_local_root_fid;
 122 };
 123 #endif /* HAVE_SERVER_SUPPORT */
 124
 125 static int echo_client_setup(const struct lu_env *env,
 126                              struct obd_device *obddev,
 127                              struct lustre_cfg *lcfg);
 128 static int echo_client_cleanup(struct obd_device *obddev);
 129
 130
 131 /** \defgroup echo_helpers Helper functions
 132  * @{
 133  */
 134 static inline struct echo_device *cl2echo_dev(const struct cl_device *dev)
 135 {
 136         return container_of0(dev, struct echo_device, ed_cl);
 137 }
 138
 139 static inline struct cl_device *echo_dev2cl(struct echo_device *d)
 140 {
 141         return &d->ed_cl;
 142 }
 143
 144 static inline struct echo_device *obd2echo_dev(const struct obd_device *obd)
 145 {
 146         return cl2echo_dev(lu2cl_dev(obd->obd_lu_dev));
 147 }
 148
 149 static inline struct cl_object *echo_obj2cl(struct echo_object *eco)
 150 {
 151         return &eco->eo_cl;
 152 }
 153
 154 static inline struct echo_object *cl2echo_obj(const struct cl_object *o)
 155 {
 156         return container_of(o, struct echo_object, eo_cl);
 157 }
 158
 159 static inline struct echo_page *cl2echo_page(const struct cl_page_slice *s)
 160 {
 161         return container_of(s, struct echo_page, ep_cl);
 162 }
 163
 164 static inline struct echo_lock *cl2echo_lock(const struct cl_lock_slice *s)
 165 {
 166         return container_of(s, struct echo_lock, el_cl);
 167 }
 168
 169 static inline struct cl_lock *echo_lock2cl(const struct echo_lock *ecl)
 170 {
 171         return ecl->el_cl.cls_lock;
 172 }
 173
 174 static struct lu_context_key echo_thread_key;
 175 static inline struct echo_thread_info *echo_env_info(const struct lu_env *env)
 176 {
 177         struct echo_thread_info *info;
 178         info = lu_context_key_get(&env->le_ctx, &echo_thread_key);
 179         LASSERT(info != NULL);
 180         return info;
 181 }
 182
 183 static inline
 184 struct echo_object_conf *cl2echo_conf(const struct cl_object_conf *c)
 185 {
 186         return container_of(c, struct echo_object_conf, eoc_cl);
 187 }
 188
 189 #ifdef HAVE_SERVER_SUPPORT
 190 static inline struct echo_md_device *lu2emd_dev(struct lu_device *d)
 191 {
 192         return container_of0(d, struct echo_md_device, emd_md_dev.md_lu_dev);
 193 }
 194
 195 static inline struct lu_device *emd2lu_dev(struct echo_md_device *d)
 196 {
 197         return &d->emd_md_dev.md_lu_dev;
 198 }
 199
 200 static inline struct seq_server_site *echo_md_seq_site(struct echo_md_device *d)
 201 {
 202         return emd2lu_dev(d)->ld_site->ld_seq_site;
 203 }
 204
 205 static inline struct obd_device *emd2obd_dev(struct echo_md_device *d)
 206 {
 207         return d->emd_md_dev.md_lu_dev.ld_obd;
 208 }
 209 #endif /* HAVE_SERVER_SUPPORT */
 210
 211 /** @} echo_helpers */
 212
 213 static int cl_echo_object_put(struct echo_object *eco);
 214 static int cl_echo_object_brw(struct echo_object *eco, int rw, u64 offset,
 215                               struct page **pages, int npages, int async);
 216
 217 struct echo_thread_info {
 218         struct echo_object_conf eti_conf;
 219         struct lustre_md        eti_md;
 220
 221         struct cl_2queue        eti_queue;
 222         struct cl_io            eti_io;
 223         struct cl_lock          eti_lock;
 224         struct lu_fid           eti_fid;
 225         struct lu_fid           eti_fid2;
 226 #ifdef HAVE_SERVER_SUPPORT
 227         struct md_op_spec       eti_spec;
 228         struct lov_mds_md_v3    eti_lmm;
 229         struct lov_user_md_v3   eti_lum;
 230         struct md_attr          eti_ma;
 231         struct lu_name          eti_lname;
 232         /* per-thread values, can be re-used */
 233         void                    *eti_big_lmm;
 234         int                     eti_big_lmmsize;
 235         char                    eti_name[20];
 236         struct lu_buf           eti_buf;
 237         char                    eti_xattr_buf[LUSTRE_POSIX_ACL_MAX_SIZE];
 238 #endif
 239 };
 240
 241 /* No session used right now */
 242 struct echo_session_info {
 243         unsigned long dummy;
 244 };
 245
 246 static struct kmem_cache *echo_lock_kmem;
 247 static struct kmem_cache *echo_object_kmem;
 248 static struct kmem_cache *echo_thread_kmem;
 249 static struct kmem_cache *echo_session_kmem;
 250 /* static struct kmem_cache *echo_req_kmem; */
 251
 252 static struct lu_kmem_descr echo_caches[] = {
 253         {
 254                 .ckd_cache = &echo_lock_kmem,
 255                 .ckd_name  = "echo_lock_kmem",
 256                 .ckd_size  = sizeof (struct echo_lock)
 257         },
 258         {
 259                 .ckd_cache = &echo_object_kmem,
 260                 .ckd_name  = "echo_object_kmem",
 261                 .ckd_size  = sizeof (struct echo_object)
 262         },
 263         {
 264                 .ckd_cache = &echo_thread_kmem,
 265                 .ckd_name  = "echo_thread_kmem",
 266                 .ckd_size  = sizeof (struct echo_thread_info)
 267         },
 268         {
 269                 .ckd_cache = &echo_session_kmem,
 270                 .ckd_name  = "echo_session_kmem",
 271                 .ckd_size  = sizeof (struct echo_session_info)
 272         },
 273         {
 274                 .ckd_cache = NULL
 275         }
 276 };
 277
 278 /** \defgroup echo_page Page operations
 279  *
 280  * Echo page operations.
 281  *
 282  * @{
 283  */
 284 static int echo_page_own(const struct lu_env *env,
 285                          const struct cl_page_slice *slice,
 286                          struct cl_io *io, int nonblock)
 287 {
 288         struct echo_page *ep = cl2echo_page(slice);
 289
 290         if (!nonblock)
 291                 mutex_lock(&ep->ep_lock);
 292         else if (!mutex_trylock(&ep->ep_lock))
 293                 return -EAGAIN;
 294         return 0;
 295 }
 296
 297 static void echo_page_disown(const struct lu_env *env,
 298                              const struct cl_page_slice *slice,
 299                              struct cl_io *io)
 300 {
 301         struct echo_page *ep = cl2echo_page(slice);
 302
 303         LASSERT(mutex_is_locked(&ep->ep_lock));
 304         mutex_unlock(&ep->ep_lock);
 305 }
 306
 307 static void echo_page_discard(const struct lu_env *env,
 308                               const struct cl_page_slice *slice,
 309                               struct cl_io *unused)
 310 {
 311         cl_page_delete(env, slice->cpl_page);
 312 }
 313
 314 static int echo_page_is_vmlocked(const struct lu_env *env,
 315                                  const struct cl_page_slice *slice)
 316 {
 317         if (mutex_is_locked(&cl2echo_page(slice)->ep_lock))
 318                 return -EBUSY;
 319         return -ENODATA;
 320 }
 321
 322 static void echo_page_completion(const struct lu_env *env,
 323                                  const struct cl_page_slice *slice,
 324                                  int ioret)
 325 {
 326         LASSERT(slice->cpl_page->cp_sync_io != NULL);
 327 }
 328
 329 static void echo_page_fini(const struct lu_env *env,
 330                            struct cl_page_slice *slice)
 331 {
 332         struct echo_object *eco = cl2echo_obj(slice->cpl_obj);
 333         ENTRY;
 334
 335         atomic_dec(&eco->eo_npages);
 336         page_cache_release(slice->cpl_page->cp_vmpage);
 337         EXIT;
 338 }
 339
 340 static int echo_page_prep(const struct lu_env *env,
 341                           const struct cl_page_slice *slice,
 342                           struct cl_io *unused)
 343 {
 344         return 0;
 345 }
 346
 347 static int echo_page_print(const struct lu_env *env,
 348                            const struct cl_page_slice *slice,
 349                            void *cookie, lu_printer_t printer)
 350 {
 351         struct echo_page *ep = cl2echo_page(slice);
 352
 353         (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p %d vm@%p\n",
 354                    ep, mutex_is_locked(&ep->ep_lock),
 355                    slice->cpl_page->cp_vmpage);
 356         return 0;
 357 }
 358
 359 static const struct cl_page_operations echo_page_ops = {
 360         .cpo_own           = echo_page_own,
 361         .cpo_disown        = echo_page_disown,
 362         .cpo_discard       = echo_page_discard,
 363         .cpo_fini          = echo_page_fini,
 364         .cpo_print         = echo_page_print,
 365         .cpo_is_vmlocked   = echo_page_is_vmlocked,
 366         .io = {
 367                 [CRT_READ] = {
 368                         .cpo_prep        = echo_page_prep,
 369                         .cpo_completion  = echo_page_completion,
 370                 },
 371                 [CRT_WRITE] = {
 372                         .cpo_prep        = echo_page_prep,
 373                         .cpo_completion  = echo_page_completion,
 374                 }
 375         }
 376 };
 377 /** @} echo_page */
 378
 379 /** \defgroup echo_lock Locking
 380  *
 381  * echo lock operations
 382  *
 383  * @{
 384  */
 385 static void echo_lock_fini(const struct lu_env *env,
 386                            struct cl_lock_slice *slice)
 387 {
 388         struct echo_lock *ecl = cl2echo_lock(slice);
 389
 390         LASSERT(list_empty(&ecl->el_chain));
 391         OBD_SLAB_FREE_PTR(ecl, echo_lock_kmem);
 392 }
 393
 394 static struct cl_lock_operations echo_lock_ops = {
 395         .clo_fini      = echo_lock_fini,
 396 };
 397
 398 /** @} echo_lock */
 399
 400 /** \defgroup echo_cl_ops cl_object operations
 401  *
 402  * operations for cl_object
 403  *
 404  * @{
 405  */
 406 static int echo_page_init(const struct lu_env *env, struct cl_object *obj,
 407                           struct cl_page *page, pgoff_t index)
 408 {
 409         struct echo_page *ep = cl_object_page_slice(obj, page);
 410         struct echo_object *eco = cl2echo_obj(obj);
 411         ENTRY;
 412
 413         page_cache_get(page->cp_vmpage);
 414         mutex_init(&ep->ep_lock);
 415         cl_page_slice_add(page, &ep->ep_cl, obj, index, &echo_page_ops);
 416         atomic_inc(&eco->eo_npages);
 417         RETURN(0);
 418 }
 419
 420 static int echo_io_init(const struct lu_env *env, struct cl_object *obj,
 421                         struct cl_io *io)
 422 {
 423         return 0;
 424 }
 425
 426 static int echo_lock_init(const struct lu_env *env,
 427                           struct cl_object *obj, struct cl_lock *lock,
 428                           const struct cl_io *unused)
 429 {
 430         struct echo_lock *el;
 431         ENTRY;
 432
 433         OBD_SLAB_ALLOC_PTR_GFP(el, echo_lock_kmem, GFP_NOFS);
 434         if (el != NULL) {
 435                 cl_lock_slice_add(lock, &el->el_cl, obj, &echo_lock_ops);
 436                 el->el_object = cl2echo_obj(obj);
 437                 INIT_LIST_HEAD(&el->el_chain);
 438                 atomic_set(&el->el_refcount, 0);
 439         }
 440         RETURN(el == NULL ? -ENOMEM : 0);
 441 }
 442
 443 static int echo_conf_set(const struct lu_env *env, struct cl_object *obj,
 444                          const struct cl_object_conf *conf)
 445 {
 446         return 0;
 447 }
 448
 449 static const struct cl_object_operations echo_cl_obj_ops = {
 450         .coo_page_init = echo_page_init,
 451         .coo_lock_init = echo_lock_init,
 452         .coo_io_init   = echo_io_init,
 453         .coo_conf_set  = echo_conf_set
 454 };
 455 /** @} echo_cl_ops */
 456
 457 /** \defgroup echo_lu_ops lu_object operations
 458  *
 459  * operations for echo lu object.
 460  *
 461  * @{
 462  */
 463 static int echo_object_init(const struct lu_env *env, struct lu_object *obj,
 464                             const struct lu_object_conf *conf)
 465 {
 466         struct echo_device *ed         = cl2echo_dev(lu2cl_dev(obj->lo_dev));
 467         struct echo_client_obd *ec     = ed->ed_ec;
 468         struct echo_object *eco        = cl2echo_obj(lu2cl(obj));
 469         ENTRY;
 470
 471         if (ed->ed_next) {
 472                 struct lu_object  *below;
 473                 struct lu_device  *under;
 474
 475                 under = ed->ed_next;
 476                 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header,
 477                                                         under);
 478                 if (below == NULL)
 479                         RETURN(-ENOMEM);
 480                 lu_object_add(obj, below);
 481         }
 482
 483         if (!ed->ed_next_ismd) {
 484                 const struct cl_object_conf *cconf = lu2cl_conf(conf);
 485                 struct echo_object_conf *econf = cl2echo_conf(cconf);
 486
 487                 LASSERT(econf->eoc_oinfo != NULL);
 488
 489                 /* Transfer the oinfo pointer to eco that it won't be
 490                  * freed. */
 491                 eco->eo_oinfo = *econf->eoc_oinfo;
 492                 *econf->eoc_oinfo = NULL;
 493         } else {
 494                 eco->eo_oinfo = NULL;
 495         }
 496
 497         eco->eo_dev = ed;
 498         atomic_set(&eco->eo_npages, 0);
 499         cl_object_page_init(lu2cl(obj), sizeof(struct echo_page));
 500
 501         spin_lock(&ec->ec_lock);
 502         list_add_tail(&eco->eo_obj_chain, &ec->ec_objects);
 503         spin_unlock(&ec->ec_lock);
 504
 505         RETURN(0);
 506 }
 507
 508 static void echo_object_free(const struct lu_env *env, struct lu_object *obj)
 509 {
 510         struct echo_object *eco    = cl2echo_obj(lu2cl(obj));
 511         struct echo_client_obd *ec = eco->eo_dev->ed_ec;
 512         ENTRY;
 513
 514         LASSERT(atomic_read(&eco->eo_npages) == 0);
 515
 516         spin_lock(&ec->ec_lock);
 517         list_del_init(&eco->eo_obj_chain);
 518         spin_unlock(&ec->ec_lock);
 519
 520         lu_object_fini(obj);
 521         lu_object_header_fini(obj->lo_header);
 522
 523         if (eco->eo_oinfo != NULL)
 524                 OBD_FREE_PTR(eco->eo_oinfo);
 525
 526         OBD_SLAB_FREE_PTR(eco, echo_object_kmem);
 527         EXIT;
 528 }
 529
 530 static int echo_object_print(const struct lu_env *env, void *cookie,
 531                             lu_printer_t p, const struct lu_object *o)
 532 {
 533         struct echo_object *obj = cl2echo_obj(lu2cl(o));
 534
 535         return (*p)(env, cookie, "echoclient-object@%p", obj);
 536 }
 537
 538 static const struct lu_object_operations echo_lu_obj_ops = {
 539         .loo_object_init      = echo_object_init,
 540         .loo_object_delete    = NULL,
 541         .loo_object_release   = NULL,
 542         .loo_object_free      = echo_object_free,
 543         .loo_object_print     = echo_object_print,
 544         .loo_object_invariant = NULL
 545 };
 546 /** @} echo_lu_ops */
 547
 548 /** \defgroup echo_lu_dev_ops  lu_device operations
 549  *
 550  * Operations for echo lu device.
 551  *
 552  * @{
 553  */
 554 static struct lu_object *echo_object_alloc(const struct lu_env *env,
 555                                            const struct lu_object_header *hdr,
 556                                            struct lu_device *dev)
 557 {
 558         struct echo_object *eco;
 559         struct lu_object *obj = NULL;
 560         ENTRY;
 561
 562         /* we're the top dev. */
 563         LASSERT(hdr == NULL);
 564         OBD_SLAB_ALLOC_PTR_GFP(eco, echo_object_kmem, GFP_NOFS);
 565         if (eco != NULL) {
 566                 struct cl_object_header *hdr = &eco->eo_hdr;
 567
 568                 obj = &echo_obj2cl(eco)->co_lu;
 569                 cl_object_header_init(hdr);
 570                 hdr->coh_page_bufsize = cfs_size_round(sizeof(struct cl_page));
 571
 572                 lu_object_init(obj, &hdr->coh_lu, dev);
 573                 lu_object_add_top(&hdr->coh_lu, obj);
 574
 575                 eco->eo_cl.co_ops = &echo_cl_obj_ops;
 576                 obj->lo_ops       = &echo_lu_obj_ops;
 577         }
 578         RETURN(obj);
 579 }
 580
 581 static struct lu_device_operations echo_device_lu_ops = {
 582         .ldo_object_alloc   = echo_object_alloc,
 583 };
 584
 585 /** @} echo_lu_dev_ops */
 586
 587 static struct cl_device_operations echo_device_cl_ops = {
 588 };
 589
 590 /** \defgroup echo_init Setup and teardown
 591  *
 592  * Init and fini functions for echo client.
 593  *
 594  * @{
 595  */
 596 static int echo_site_init(const struct lu_env *env, struct echo_device *ed)
 597 {
 598         struct cl_site *site = &ed->ed_site_myself;
 599         int rc;
 600
 601         /* initialize site */
 602         rc = cl_site_init(site, &ed->ed_cl);
 603         if (rc) {
 604                 CERROR("Cannot initialize site for echo client(%d)\n", rc);
 605                 return rc;
 606         }
 607
 608         rc = lu_site_init_finish(&site->cs_lu);
 609         if (rc) {
 610                 cl_site_fini(site);
 611                 return rc;
 612         }
 613
 614         ed->ed_site = &site->cs_lu;
 615         return 0;
 616 }
 617
 618 static void echo_site_fini(const struct lu_env *env, struct echo_device *ed)
 619 {
 620         if (ed->ed_site) {
 621                 if (!ed->ed_next_ismd)
 622                         lu_site_fini(ed->ed_site);
 623                 ed->ed_site = NULL;
 624         }
 625 }
 626
 627 static void *echo_thread_key_init(const struct lu_context *ctx,
 628                                   struct lu_context_key *key)
 629 {
 630         struct echo_thread_info *info;
 631
 632         OBD_SLAB_ALLOC_PTR_GFP(info, echo_thread_kmem, GFP_NOFS);
 633         if (info == NULL)
 634                 info = ERR_PTR(-ENOMEM);
 635         return info;
 636 }
 637
 638 static void echo_thread_key_fini(const struct lu_context *ctx,
 639                          struct lu_context_key *key, void *data)
 640 {
 641         struct echo_thread_info *info = data;
 642         OBD_SLAB_FREE_PTR(info, echo_thread_kmem);
 643 }
 644
 645 static void echo_thread_key_exit(const struct lu_context *ctx,
 646                          struct lu_context_key *key, void *data)
 647 {
 648 }
 649
 650 static struct lu_context_key echo_thread_key = {
 651         .lct_tags = LCT_CL_THREAD,
 652         .lct_init = echo_thread_key_init,
 653         .lct_fini = echo_thread_key_fini,
 654         .lct_exit = echo_thread_key_exit
 655 };
 656
 657 static void *echo_session_key_init(const struct lu_context *ctx,
 658                                   struct lu_context_key *key)
 659 {
 660         struct echo_session_info *session;
 661
 662         OBD_SLAB_ALLOC_PTR_GFP(session, echo_session_kmem, GFP_NOFS);
 663         if (session == NULL)
 664                 session = ERR_PTR(-ENOMEM);
 665         return session;
 666 }
 667
 668 static void echo_session_key_fini(const struct lu_context *ctx,
 669                                  struct lu_context_key *key, void *data)
 670 {
 671         struct echo_session_info *session = data;
 672         OBD_SLAB_FREE_PTR(session, echo_session_kmem);
 673 }
 674
 675 static void echo_session_key_exit(const struct lu_context *ctx,
 676                                  struct lu_context_key *key, void *data)
 677 {
 678 }
 679
 680 static struct lu_context_key echo_session_key = {
 681         .lct_tags = LCT_SESSION,
 682         .lct_init = echo_session_key_init,
 683         .lct_fini = echo_session_key_fini,
 684         .lct_exit = echo_session_key_exit
 685 };
 686
 687 LU_TYPE_INIT_FINI(echo, &echo_thread_key, &echo_session_key);
 688
 689 #ifdef HAVE_SERVER_SUPPORT
 690 # define ECHO_SEQ_WIDTH 0xffffffff
 691 static int echo_fid_init(struct echo_device *ed, char *obd_name,
 692                          struct seq_server_site *ss)
 693 {
 694         char *prefix;
 695         int rc;
 696         ENTRY;
 697
 698         OBD_ALLOC_PTR(ed->ed_cl_seq);
 699         if (ed->ed_cl_seq == NULL)
 700                 RETURN(-ENOMEM);
 701
 702         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
 703         if (prefix == NULL)
 704                 GOTO(out_free_seq, rc = -ENOMEM);
 705
 706         snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", obd_name);
 707
 708         /* Init client side sequence-manager */
 709         rc = seq_client_init(ed->ed_cl_seq, NULL,
 710                              LUSTRE_SEQ_METADATA,
 711                              prefix, ss->ss_server_seq);
 712         ed->ed_cl_seq->lcs_width = ECHO_SEQ_WIDTH;
 713         OBD_FREE(prefix, MAX_OBD_NAME + 5);
 714         if (rc)
 715                 GOTO(out_free_seq, rc);
 716
 717         RETURN(0);
 718
 719 out_free_seq:
 720         OBD_FREE_PTR(ed->ed_cl_seq);
 721         ed->ed_cl_seq = NULL;
 722         RETURN(rc);
 723 }
 724
 725 static int echo_fid_fini(struct obd_device *obddev)
 726 {
 727         struct echo_device *ed = obd2echo_dev(obddev);
 728         ENTRY;
 729
 730         if (ed->ed_cl_seq != NULL) {
 731                 seq_client_fini(ed->ed_cl_seq);
 732                 OBD_FREE_PTR(ed->ed_cl_seq);
 733                 ed->ed_cl_seq = NULL;
 734         }
 735
 736         RETURN(0);
 737 }
 738
 739 static void echo_ed_los_fini(const struct lu_env *env, struct echo_device *ed)
 740 {
 741         ENTRY;
 742
 743         if (ed != NULL && ed->ed_next_ismd && ed->ed_los != NULL) {
 744                 local_oid_storage_fini(env, ed->ed_los);
 745                 ed->ed_los = NULL;
 746         }
 747 }
 748
 749 static int
 750 echo_md_local_file_create(const struct lu_env *env, struct echo_md_device *emd,
 751                           struct local_oid_storage *los,
 752                           const struct lu_fid *pfid, const char *name,
 753                           __u32 mode, struct lu_fid *fid)
 754 {
 755         struct dt_object        *parent = NULL;
 756         struct dt_object        *dto = NULL;
 757         int                      rc = 0;
 758         ENTRY;
 759
 760         LASSERT(!fid_is_zero(pfid));
 761         parent = dt_locate(env, emd->emd_bottom, pfid);
 762         if (unlikely(IS_ERR(parent)))
 763                 RETURN(PTR_ERR(parent));
 764
 765         /* create local file with @fid */
 766         dto = local_file_find_or_create_with_fid(env, emd->emd_bottom, fid,
 767                                                  parent, name, mode);
 768         if (IS_ERR(dto))
 769                 GOTO(out_put, rc = PTR_ERR(dto));
 770
 771         *fid = *lu_object_fid(&dto->do_lu);
 772         /* since stack is not fully set up the local_storage uses own stack
 773          * and we should drop its object from cache */
 774         lu_object_put_nocache(env, &dto->do_lu);
 775
 776         EXIT;
 777 out_put:
 778         lu_object_put(env, &parent->do_lu);
 779         RETURN(rc);
 780 }
 781
 782 static int
 783 echo_md_root_get(const struct lu_env *env, struct echo_md_device *emd,
 784                  struct echo_device *ed)
 785 {
 786         struct lu_fid                    fid;
 787         int                              rc = 0;
 788         ENTRY;
 789
 790         /* Setup local dirs */
 791         fid.f_seq = FID_SEQ_LOCAL_NAME;
 792         fid.f_oid = 1;
 793         fid.f_ver = 0;
 794         rc = local_oid_storage_init(env, emd->emd_bottom, &fid, &ed->ed_los);
 795         if (rc != 0)
 796                 RETURN(rc);
 797
 798         lu_echo_root_fid(&fid);
 799         if (echo_md_seq_site(emd)->ss_node_id == 0) {
 800                 rc = echo_md_local_file_create(env, emd, ed->ed_los,
 801                                                &emd->emd_local_root_fid,
 802                                                echo_md_root_dir_name, S_IFDIR |
 803                                                S_IRUGO | S_IWUSR | S_IXUGO,
 804                                                &fid);
 805                 if (rc != 0) {
 806                         CERROR("%s: create md echo root fid failed: rc = %d\n",
 807                                emd2obd_dev(emd)->obd_name, rc);
 808                         GOTO(out_los, rc);
 809                 }
 810         }
 811         ed->ed_root_fid = fid;
 812
 813         RETURN(0);
 814 out_los:
 815         echo_ed_los_fini(env, ed);
 816
 817         RETURN(rc);
 818 }
 819 #endif /* HAVE_SERVER_SUPPORT */
 820
 821 static struct lu_device *echo_device_alloc(const struct lu_env *env,
 822                                            struct lu_device_type *t,
 823                                            struct lustre_cfg *cfg)
 824 {
 825         struct lu_device   *next;
 826         struct echo_device *ed;
 827         struct cl_device   *cd;
 828         struct obd_device  *obd = NULL; /* to keep compiler happy */
 829         struct obd_device  *tgt;
 830         const char *tgt_type_name;
 831         int rc;
 832         int cleanup = 0;
 833         ENTRY;
 834
 835         OBD_ALLOC_PTR(ed);
 836         if (ed == NULL)
 837                 GOTO(out, rc = -ENOMEM);
 838
 839         cleanup = 1;
 840         cd = &ed->ed_cl;
 841         rc = cl_device_init(cd, t);
 842         if (rc)
 843                 GOTO(out, rc);
 844
 845         cd->cd_lu_dev.ld_ops = &echo_device_lu_ops;
 846         cd->cd_ops = &echo_device_cl_ops;
 847
 848         cleanup = 2;
 849         obd = class_name2obd(lustre_cfg_string(cfg, 0));
 850         LASSERT(obd != NULL);
 851         LASSERT(env != NULL);
 852
 853         tgt = class_name2obd(lustre_cfg_string(cfg, 1));
 854         if (tgt == NULL) {
 855                 CERROR("Can not find tgt device %s\n",
 856                         lustre_cfg_string(cfg, 1));
 857                 GOTO(out, rc = -ENODEV);
 858         }
 859
 860         next = tgt->obd_lu_dev;
 861
 862         if (strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME) == 0) {
 863                 ed->ed_next_ismd = 1;
 864         } else if (strcmp(tgt->obd_type->typ_name, LUSTRE_OST_NAME) == 0 ||
 865                    strcmp(tgt->obd_type->typ_name, LUSTRE_OSC_NAME) == 0) {
 866                 ed->ed_next_ismd = 0;
 867                 rc = echo_site_init(env, ed);
 868                 if (rc)
 869                         GOTO(out, rc);
 870         } else {
 871                 GOTO(out, rc = -EINVAL);
 872         }
 873
 874         cleanup = 3;
 875
 876         rc = echo_client_setup(env, obd, cfg);
 877         if (rc)
 878                 GOTO(out, rc);
 879
 880         ed->ed_ec = &obd->u.echo_client;
 881         cleanup = 4;
 882
 883         if (ed->ed_next_ismd) {
 884 #ifdef HAVE_SERVER_SUPPORT
 885                 /* Suppose to connect to some Metadata layer */
 886                 struct lu_site          *ls = NULL;
 887                 struct lu_device        *ld = NULL;
 888                 struct md_device        *md = NULL;
 889                 struct echo_md_device   *emd = NULL;
 890                 int                      found = 0;
 891
 892                 if (next == NULL) {
 893                         CERROR("%s is not lu device type!\n",
 894                                lustre_cfg_string(cfg, 1));
 895                         GOTO(out, rc = -EINVAL);
 896                 }
 897
 898                 tgt_type_name = lustre_cfg_string(cfg, 2);
 899                 if (!tgt_type_name) {
 900                         CERROR("%s no type name for echo %s setup\n",
 901                                 lustre_cfg_string(cfg, 1),
 902                                 tgt->obd_type->typ_name);
 903                         GOTO(out, rc = -EINVAL);
 904                 }
 905
 906                 ls = next->ld_site;
 907
 908                 spin_lock(&ls->ls_ld_lock);
 909                 list_for_each_entry(ld, &ls->ls_ld_linkage, ld_linkage) {
 910                         if (strcmp(ld->ld_type->ldt_name, tgt_type_name) == 0) {
 911                                 found = 1;
 912                                 break;
 913                         }
 914                 }
 915                 spin_unlock(&ls->ls_ld_lock);
 916
 917                 if (found == 0) {
 918                         CERROR("%s is not lu device type!\n",
 919                                lustre_cfg_string(cfg, 1));
 920                         GOTO(out, rc = -EINVAL);
 921                 }
 922
 923                 next = ld;
 924                 /* For MD echo client, it will use the site in MDS stack */
 925                 ed->ed_site = ls;
 926                 ed->ed_cl.cd_lu_dev.ld_site = ls;
 927                 rc = echo_fid_init(ed, obd->obd_name, lu_site2seq(ls));
 928                 if (rc) {
 929                         CERROR("echo fid init error %d\n", rc);
 930                         GOTO(out, rc);
 931                 }
 932
 933                 md = lu2md_dev(next);
 934                 emd = lu2emd_dev(&md->md_lu_dev);
 935                 rc = echo_md_root_get(env, emd, ed);
 936                 if (rc != 0) {
 937                         CERROR("%s: get root error: rc = %d\n",
 938                                 emd2obd_dev(emd)->obd_name, rc);
 939                         GOTO(out, rc);
 940                 }
 941 #else /* !HAVE_SERVER_SUPPORT */
 942                 CERROR("Local operations are NOT supported on client side. "
 943                        "Only remote operations are supported. Metadata client "
 944                        "must be run on server side.\n");
 945                 GOTO(out, rc = -EOPNOTSUPP);
 946 #endif /* HAVE_SERVER_SUPPORT */
 947         } else {
 948                  /* if echo client is to be stacked upon ost device, the next is
 949                   * NULL since ost is not a clio device so far */
 950                 if (next != NULL && !lu_device_is_cl(next))
 951                         next = NULL;
 952
 953                 tgt_type_name = tgt->obd_type->typ_name;
 954                 if (next != NULL) {
 955                         LASSERT(next != NULL);
 956                         if (next->ld_site != NULL)
 957                                 GOTO(out, rc = -EBUSY);
 958
 959                         next->ld_site = ed->ed_site;
 960                         rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
 961                                                      next->ld_type->ldt_name,
 962                                                      NULL);
 963                         if (rc)
 964                                 GOTO(out, rc);
 965                 } else
 966                         LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0);
 967         }
 968
 969         ed->ed_next = next;
 970         RETURN(&cd->cd_lu_dev);
 971 out:
 972         switch(cleanup) {
 973         case 4: {
 974                 int rc2;
 975                 rc2 = echo_client_cleanup(obd);
 976                 if (rc2)
 977                         CERROR("Cleanup obd device %s error(%d)\n",
 978                                obd->obd_name, rc2);
 979         }
 980
 981         case 3:
 982                 echo_site_fini(env, ed);
 983         case 2:
 984                 cl_device_fini(&ed->ed_cl);
 985         case 1:
 986                 OBD_FREE_PTR(ed);
 987         case 0:
 988         default:
 989                 break;
 990         }
 991         return(ERR_PTR(rc));
 992 }
 993
 994 static int echo_device_init(const struct lu_env *env, struct lu_device *d,
 995                           const char *name, struct lu_device *next)
 996 {
 997         LBUG();
 998         return 0;
 999 }
1000
1001 static struct lu_device *echo_device_fini(const struct lu_env *env,
1002                                           struct lu_device *d)
1003 {
1004         struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
1005         struct lu_device *next = ed->ed_next;
1006
1007         while (next && !ed->ed_next_ismd)
1008                 next = next->ld_type->ldt_ops->ldto_device_fini(env, next);
1009         return NULL;
1010 }
1011
1012 static void echo_lock_release(const struct lu_env *env,
1013                               struct echo_lock *ecl,
1014                               int still_used)
1015 {
1016         struct cl_lock *clk = echo_lock2cl(ecl);
1017
1018         cl_lock_release(env, clk);
1019 }
1020
1021 static struct lu_device *echo_device_free(const struct lu_env *env,
1022                                           struct lu_device *d)
1023 {
1024         struct echo_device     *ed   = cl2echo_dev(lu2cl_dev(d));
1025         struct echo_client_obd *ec   = ed->ed_ec;
1026         struct echo_object     *eco;
1027         struct lu_device       *next = ed->ed_next;
1028
1029         CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n",
1030                ed, next);
1031
1032         lu_site_purge(env, ed->ed_site, -1);
1033
1034         /* check if there are objects still alive.
1035          * It shouldn't have any object because lu_site_purge would cleanup
1036          * all of cached objects. Anyway, probably the echo device is being
1037          * parallelly accessed.
1038          */
1039         spin_lock(&ec->ec_lock);
1040         list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain)
1041                 eco->eo_deleted = 1;
1042         spin_unlock(&ec->ec_lock);
1043
1044         /* purge again */
1045         lu_site_purge(env, ed->ed_site, -1);
1046
1047         CDEBUG(D_INFO,
1048                "Waiting for the reference of echo object to be dropped\n");
1049
1050         /* Wait for the last reference to be dropped. */
1051         spin_lock(&ec->ec_lock);
1052         while (!list_empty(&ec->ec_objects)) {
1053                 spin_unlock(&ec->ec_lock);
1054                 CERROR("echo_client still has objects at cleanup time, "
1055                        "wait for 1 second\n");
1056                 set_current_state(TASK_UNINTERRUPTIBLE);
1057                 schedule_timeout(cfs_time_seconds(1));
1058                 lu_site_purge(env, ed->ed_site, -1);
1059                 spin_lock(&ec->ec_lock);
1060         }
1061         spin_unlock(&ec->ec_lock);
1062
1063         LASSERT(list_empty(&ec->ec_locks));
1064
1065         CDEBUG(D_INFO, "No object exists, exiting...\n");
1066
1067         echo_client_cleanup(d->ld_obd);
1068 #ifdef HAVE_SERVER_SUPPORT
1069         echo_fid_fini(d->ld_obd);
1070         echo_ed_los_fini(env, ed);
1071 #endif
1072         while (next && !ed->ed_next_ismd)
1073                 next = next->ld_type->ldt_ops->ldto_device_free(env, next);
1074
1075         LASSERT(ed->ed_site == d->ld_site);
1076         echo_site_fini(env, ed);
1077         cl_device_fini(&ed->ed_cl);
1078         OBD_FREE_PTR(ed);
1079
1080         return NULL;
1081 }
1082
1083 static const struct lu_device_type_operations echo_device_type_ops = {
1084         .ldto_init = echo_type_init,
1085         .ldto_fini = echo_type_fini,
1086
1087         .ldto_start = echo_type_start,
1088         .ldto_stop  = echo_type_stop,
1089
1090         .ldto_device_alloc = echo_device_alloc,
1091         .ldto_device_free  = echo_device_free,
1092         .ldto_device_init  = echo_device_init,
1093         .ldto_device_fini  = echo_device_fini
1094 };
1095
1096 static struct lu_device_type echo_device_type = {
1097         .ldt_tags     = LU_DEVICE_CL,
1098         .ldt_name     = LUSTRE_ECHO_CLIENT_NAME,
1099         .ldt_ops      = &echo_device_type_ops,
1100         .ldt_ctx_tags = LCT_CL_THREAD | LCT_MD_THREAD | LCT_DT_THREAD,
1101 };
1102 /** @} echo_init */
1103
1104 /** \defgroup echo_exports Exported operations
1105  *
1106  * exporting functions to echo client
1107  *
1108  * @{
1109  */
1110
1111 /* Interfaces to echo client obd device */
1112 static struct echo_object *
1113 cl_echo_object_find(struct echo_device *d, const struct ost_id *oi)
1114 {
1115         struct lu_env *env;
1116         struct echo_thread_info *info;
1117         struct echo_object_conf *conf;
1118         struct echo_object *eco;
1119         struct cl_object *obj;
1120         struct lov_oinfo *oinfo = NULL;
1121         struct lu_fid *fid;
1122         int refcheck;
1123         int rc;
1124         ENTRY;
1125
1126         LASSERTF(ostid_id(oi) != 0, DOSTID"\n", POSTID(oi));
1127         LASSERTF(ostid_seq(oi) == FID_SEQ_ECHO, DOSTID"\n", POSTID(oi));
1128
1129         /* Never return an object if the obd is to be freed. */
1130         if (echo_dev2cl(d)->cd_lu_dev.ld_obd->obd_stopping)
1131                 RETURN(ERR_PTR(-ENODEV));
1132
1133         env = cl_env_get(&refcheck);
1134         if (IS_ERR(env))
1135                 RETURN((void *)env);
1136
1137         info = echo_env_info(env);
1138         conf = &info->eti_conf;
1139         if (d->ed_next) {
1140                 OBD_ALLOC_PTR(oinfo);
1141                 if (oinfo == NULL)
1142                         GOTO(out, eco = ERR_PTR(-ENOMEM));
1143
1144                 oinfo->loi_oi = *oi;
1145                 conf->eoc_cl.u.coc_oinfo = oinfo;
1146         }
1147
1148         /* If echo_object_init() is successful then ownership of oinfo
1149          * is transferred to the object. */
1150         conf->eoc_oinfo = &oinfo;
1151
1152         fid = &info->eti_fid;
1153         rc = ostid_to_fid(fid, oi, 0);
1154         if (rc != 0)
1155                 GOTO(out, eco = ERR_PTR(rc));
1156
1157         /* In the function below, .hs_keycmp resolves to
1158          * lu_obj_hop_keycmp() */
1159         /* coverity[overrun-buffer-val] */
1160         obj = cl_object_find(env, echo_dev2cl(d), fid, &conf->eoc_cl);
1161         if (IS_ERR(obj))
1162                 GOTO(out, eco = (void*)obj);
1163
1164         eco = cl2echo_obj(obj);
1165         if (eco->eo_deleted) {
1166                 cl_object_put(env, obj);
1167                 eco = ERR_PTR(-EAGAIN);
1168         }
1169
1170 out:
1171         if (oinfo != NULL)
1172                 OBD_FREE_PTR(oinfo);
1173
1174         cl_env_put(env, &refcheck);
1175         RETURN(eco);
1176 }
1177
1178 static int cl_echo_object_put(struct echo_object *eco)
1179 {
1180         struct lu_env *env;
1181         struct cl_object *obj = echo_obj2cl(eco);
1182         int refcheck;
1183         ENTRY;
1184
1185         env = cl_env_get(&refcheck);
1186         if (IS_ERR(env))
1187                 RETURN(PTR_ERR(env));
1188
1189         /* an external function to kill an object? */
1190         if (eco->eo_deleted) {
1191                 struct lu_object_header *loh = obj->co_lu.lo_header;
1192                 LASSERT(&eco->eo_hdr == luh2coh(loh));
1193                 set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags);
1194         }
1195
1196         cl_object_put(env, obj);
1197         cl_env_put(env, &refcheck);
1198         RETURN(0);
1199 }
1200
1201 static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco,
1202                             u64 start, u64 end, int mode,
1203                             __u64 *cookie , __u32 enqflags)
1204 {
1205         struct cl_io *io;
1206         struct cl_lock *lck;
1207         struct cl_object *obj;
1208         struct cl_lock_descr *descr;
1209         struct echo_thread_info *info;
1210         int rc = -ENOMEM;
1211         ENTRY;
1212
1213         info = echo_env_info(env);
1214         io = &info->eti_io;
1215         lck = &info->eti_lock;
1216         obj = echo_obj2cl(eco);
1217
1218         memset(lck, 0, sizeof(*lck));
1219         descr = &lck->cll_descr;
1220         descr->cld_obj   = obj;
1221         descr->cld_start = cl_index(obj, start);
1222         descr->cld_end   = cl_index(obj, end);
1223         descr->cld_mode  = mode == LCK_PW ? CLM_WRITE : CLM_READ;
1224         descr->cld_enq_flags = enqflags;
1225         io->ci_obj = obj;
1226
1227         rc = cl_lock_request(env, io, lck);
1228         if (rc == 0) {
1229                 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
1230                 struct echo_lock *el;
1231
1232                 el = cl2echo_lock(cl_lock_at(lck, &echo_device_type));
1233                 spin_lock(&ec->ec_lock);
1234                 if (list_empty(&el->el_chain)) {
1235                         list_add(&el->el_chain, &ec->ec_locks);
1236                         el->el_cookie = ++ec->ec_unique;
1237                 }
1238                 atomic_inc(&el->el_refcount);
1239                 *cookie = el->el_cookie;
1240                 spin_unlock(&ec->ec_lock);
1241         }
1242         RETURN(rc);
1243 }
1244
1245 static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed,
1246                            __u64 cookie)
1247 {
1248         struct echo_client_obd *ec = ed->ed_ec;
1249         struct echo_lock       *ecl = NULL;
1250         struct list_head        *el;
1251         int found = 0, still_used = 0;
1252         ENTRY;
1253
1254         LASSERT(ec != NULL);
1255         spin_lock(&ec->ec_lock);
1256         list_for_each(el, &ec->ec_locks) {
1257                 ecl = list_entry(el, struct echo_lock, el_chain);
1258                 CDEBUG(D_INFO, "ecl: %p, cookie: "LPX64"\n", ecl, ecl->el_cookie);
1259                 found = (ecl->el_cookie == cookie);
1260                 if (found) {
1261                         if (atomic_dec_and_test(&ecl->el_refcount))
1262                                 list_del_init(&ecl->el_chain);
1263                         else
1264                                 still_used = 1;
1265                         break;
1266                 }
1267         }
1268         spin_unlock(&ec->ec_lock);
1269
1270         if (!found)
1271                 RETURN(-ENOENT);
1272
1273         echo_lock_release(env, ecl, still_used);
1274         RETURN(0);
1275 }
1276
1277 static void echo_commit_callback(const struct lu_env *env, struct cl_io *io,
1278                                 struct cl_page *page)
1279 {
1280         struct echo_thread_info *info;
1281         struct cl_2queue        *queue;
1282
1283         info = echo_env_info(env);
1284         LASSERT(io == &info->eti_io);
1285
1286         queue = &info->eti_queue;
1287         cl_page_list_add(&queue->c2_qout, page);
1288 }
1289
1290 static int cl_echo_object_brw(struct echo_object *eco, int rw, u64 offset,
1291                               struct page **pages, int npages, int async)
1292 {
1293         struct lu_env           *env;
1294         struct echo_thread_info *info;
1295         struct cl_object        *obj = echo_obj2cl(eco);
1296         struct echo_device      *ed  = eco->eo_dev;
1297         struct cl_2queue        *queue;
1298         struct cl_io            *io;
1299         struct cl_page          *clp;
1300         struct lustre_handle    lh = { 0 };
1301         int page_size = cl_page_size(obj);
1302         int refcheck;
1303         int rc;
1304         int i;
1305         ENTRY;
1306
1307         LASSERT((offset & ~PAGE_MASK) == 0);
1308         LASSERT(ed->ed_next != NULL);
1309         env = cl_env_get(&refcheck);
1310         if (IS_ERR(env))
1311                 RETURN(PTR_ERR(env));
1312
1313         info    = echo_env_info(env);
1314         io      = &info->eti_io;
1315         queue   = &info->eti_queue;
1316
1317         cl_2queue_init(queue);
1318
1319         io->ci_ignore_layout = 1;
1320         rc = cl_io_init(env, io, CIT_MISC, obj);
1321         if (rc < 0)
1322                 GOTO(out, rc);
1323         LASSERT(rc == 0);
1324
1325
1326         rc = cl_echo_enqueue0(env, eco, offset,
1327                               offset + npages * PAGE_CACHE_SIZE - 1,
1328                               rw == READ ? LCK_PR : LCK_PW, &lh.cookie,
1329                               CEF_NEVER);
1330         if (rc < 0)
1331                 GOTO(error_lock, rc);
1332
1333         for (i = 0; i < npages; i++) {
1334                 LASSERT(pages[i]);
1335                 clp = cl_page_find(env, obj, cl_index(obj, offset),
1336                                    pages[i], CPT_TRANSIENT);
1337                 if (IS_ERR(clp)) {
1338                         rc = PTR_ERR(clp);
1339                         break;
1340                 }
1341                 LASSERT(clp->cp_type == CPT_TRANSIENT);
1342
1343                 rc = cl_page_own(env, io, clp);
1344                 if (rc) {
1345                         LASSERT(clp->cp_state == CPS_FREEING);
1346                         cl_page_put(env, clp);
1347                         break;
1348                 }
1349
1350                 cl_2queue_add(queue, clp);
1351
1352                 /* drop the reference count for cl_page_find, so that the page
1353                  * will be freed in cl_2queue_fini. */
1354                 cl_page_put(env, clp);
1355                 cl_page_clip(env, clp, 0, page_size);
1356
1357                 offset += page_size;
1358         }
1359
1360         if (rc == 0) {
1361                 enum cl_req_type typ = rw == READ ? CRT_READ : CRT_WRITE;
1362
1363                 async = async && (typ == CRT_WRITE);
1364                 if (async)
1365                         rc = cl_io_commit_async(env, io, &queue->c2_qin,
1366                                                 0, PAGE_SIZE,
1367                                                 echo_commit_callback);
1368                 else
1369                         rc = cl_io_submit_sync(env, io, typ, queue, 0);
1370                 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
1371                        async ? "async" : "sync", rc);
1372         }
1373
1374         cl_echo_cancel0(env, ed, lh.cookie);
1375         EXIT;
1376 error_lock:
1377         cl_2queue_discard(env, io, queue);
1378         cl_2queue_disown(env, io, queue);
1379         cl_2queue_fini(env, queue);
1380         cl_io_fini(env, io);
1381 out:
1382         cl_env_put(env, &refcheck);
1383         return rc;
1384 }
1385 /** @} echo_exports */
1386
1387
1388 static u64 last_object_id;
1389
1390 #ifdef HAVE_SERVER_SUPPORT
1391 static inline void echo_md_build_name(struct lu_name *lname, char *name,
1392                                       __u64 id)
1393 {
1394         sprintf(name, LPU64, id);
1395         lname->ln_name = name;
1396         lname->ln_namelen = strlen(name);
1397 }
1398
1399 /* similar to mdt_attr_get_complex */
1400 static int echo_big_lmm_get(const struct lu_env *env, struct md_object *o,
1401                             struct md_attr *ma)
1402 {
1403         struct echo_thread_info *info = echo_env_info(env);
1404         int                      rc;
1405
1406         ENTRY;
1407
1408         LASSERT(ma->ma_lmm_size > 0);
1409
1410         rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LOV);
1411         if (rc < 0)
1412                 RETURN(rc);
1413
1414         /* big_lmm may need to be grown */
1415         if (info->eti_big_lmmsize < rc) {
1416                 int size = size_roundup_power2(rc);
1417
1418                 if (info->eti_big_lmmsize > 0) {
1419                         /* free old buffer */
1420                         LASSERT(info->eti_big_lmm);
1421                         OBD_FREE_LARGE(info->eti_big_lmm,
1422                                        info->eti_big_lmmsize);
1423                         info->eti_big_lmm = NULL;
1424                         info->eti_big_lmmsize = 0;
1425                 }
1426
1427                 OBD_ALLOC_LARGE(info->eti_big_lmm, size);
1428                 if (info->eti_big_lmm == NULL)
1429                         RETURN(-ENOMEM);
1430                 info->eti_big_lmmsize = size;
1431         }
1432         LASSERT(info->eti_big_lmmsize >= rc);
1433
1434         info->eti_buf.lb_buf = info->eti_big_lmm;
1435         info->eti_buf.lb_len = info->eti_big_lmmsize;
1436         rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LOV);
1437         if (rc < 0)
1438                 RETURN(rc);
1439
1440         ma->ma_valid |= MA_LOV;
1441         ma->ma_lmm = info->eti_big_lmm;
1442         ma->ma_lmm_size = rc;
1443
1444         RETURN(0);
1445 }
1446
1447 static int echo_attr_get_complex(const struct lu_env *env,
1448                                  struct md_object *next,
1449                                  struct md_attr *ma)
1450 {
1451         struct echo_thread_info *info = echo_env_info(env);
1452         struct lu_buf           *buf = &info->eti_buf;
1453         umode_t          mode = lu_object_attr(&next->mo_lu);
1454         int                      need = ma->ma_need;
1455         int                      rc = 0, rc2;
1456
1457         ENTRY;
1458
1459         ma->ma_valid = 0;
1460
1461         if (need & MA_INODE) {
1462                 ma->ma_need = MA_INODE;
1463                 rc = mo_attr_get(env, next, ma);
1464                 if (rc)
1465                         GOTO(out, rc);
1466                 ma->ma_valid |= MA_INODE;
1467         }
1468
1469         if (need & MA_LOV) {
1470                 if (S_ISREG(mode) || S_ISDIR(mode)) {
1471                         LASSERT(ma->ma_lmm_size > 0);
1472                         buf->lb_buf = ma->ma_lmm;
1473                         buf->lb_len = ma->ma_lmm_size;
1474                         rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LOV);
1475                         if (rc2 > 0) {
1476                                 ma->ma_lmm_size = rc2;
1477                                 ma->ma_valid |= MA_LOV;
1478                         } else if (rc2 == -ENODATA) {
1479                                 /* no LOV EA */
1480                                 ma->ma_lmm_size = 0;
1481                         } else if (rc2 == -ERANGE) {
1482                                 rc2 = echo_big_lmm_get(env, next, ma);
1483                                 if (rc2 < 0)
1484                                         GOTO(out, rc = rc2);
1485                         } else {
1486                                 GOTO(out, rc = rc2);
1487                         }
1488                 }
1489         }
1490
1491 #ifdef CONFIG_FS_POSIX_ACL
1492         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1493                 buf->lb_buf = ma->ma_acl;
1494                 buf->lb_len = ma->ma_acl_size;
1495                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1496                 if (rc2 > 0) {
1497                         ma->ma_acl_size = rc2;
1498                         ma->ma_valid |= MA_ACL_DEF;
1499                 } else if (rc2 == -ENODATA) {
1500                         /* no ACLs */
1501                         ma->ma_acl_size = 0;
1502                 } else {
1503                         GOTO(out, rc = rc2);
1504                 }
1505         }
1506 #endif
1507 out:
1508         ma->ma_need = need;
1509         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
1510                rc, ma->ma_valid, ma->ma_lmm);
1511         RETURN(rc);
1512 }
1513
1514 static int
1515 echo_md_create_internal(const struct lu_env *env, struct echo_device *ed,
1516                         struct md_object *parent, struct lu_fid *fid,
1517                         struct lu_name *lname, struct md_op_spec *spec,
1518                         struct md_attr *ma)
1519 {
1520         struct lu_object        *ec_child, *child;
1521         struct lu_device        *ld = ed->ed_next;
1522         struct echo_thread_info *info = echo_env_info(env);
1523         struct lu_fid           *fid2 = &info->eti_fid2;
1524         struct lu_object_conf    conf = { .loc_flags = LOC_F_NEW };
1525         int                      rc;
1526
1527         ENTRY;
1528
1529         rc = mdo_lookup(env, parent, lname, fid2, spec);
1530         if (rc == 0)
1531                 return -EEXIST;
1532         else if (rc != -ENOENT)
1533                 return rc;
1534
1535         ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev,
1536                                      fid, &conf);
1537         if (IS_ERR(ec_child)) {
1538                 CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid),
1539                         PTR_ERR(ec_child));
1540                 RETURN(PTR_ERR(ec_child));
1541         }
1542
1543         child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1544         if (child == NULL) {
1545                 CERROR("Can not locate the child "DFID"\n", PFID(fid));
1546                 GOTO(out_put, rc = -EINVAL);
1547         }
1548
1549         CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n",
1550                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1551
1552         /*
1553          * Do not perform lookup sanity check. We know that name does not exist.
1554          */
1555         spec->sp_cr_lookup = 0;
1556         rc = mdo_create(env, parent, lname, lu2md(child), spec, ma);
1557         if (rc) {
1558                 CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc);
1559                 GOTO(out_put, rc);
1560         }
1561         CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc  = %d\n",
1562                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc);
1563         EXIT;
1564 out_put:
1565         lu_object_put(env, ec_child);
1566         return rc;
1567 }
1568
1569 static int echo_set_lmm_size(const struct lu_env *env, struct lu_device *ld,
1570                              struct md_attr *ma)
1571 {
1572         struct echo_thread_info *info = echo_env_info(env);
1573
1574         if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
1575                 ma->ma_lmm = (void *)&info->eti_lmm;
1576                 ma->ma_lmm_size = sizeof(info->eti_lmm);
1577         } else {
1578                 LASSERT(info->eti_big_lmmsize);
1579                 ma->ma_lmm = info->eti_big_lmm;
1580                 ma->ma_lmm_size = info->eti_big_lmmsize;
1581         }
1582
1583         return 0;
1584 }
1585
1586 static int echo_create_md_object(const struct lu_env *env,
1587                                  struct echo_device *ed,
1588                                  struct lu_object *ec_parent,
1589                                  struct lu_fid *fid,
1590                                  char *name, int namelen,
1591                                  __u64 id, __u32 mode, int count,
1592                                  int stripe_count, int stripe_offset)
1593 {
1594         struct lu_object        *parent;
1595         struct echo_thread_info *info = echo_env_info(env);
1596         struct lu_name          *lname = &info->eti_lname;
1597         struct md_op_spec       *spec = &info->eti_spec;
1598         struct md_attr          *ma = &info->eti_ma;
1599         struct lu_device        *ld = ed->ed_next;
1600         int                      rc = 0;
1601         int                      i;
1602
1603         ENTRY;
1604
1605         if (ec_parent == NULL)
1606                 return -1;
1607         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1608         if (parent == NULL)
1609                 RETURN(-ENXIO);
1610
1611         memset(ma, 0, sizeof(*ma));
1612         memset(spec, 0, sizeof(*spec));
1613         if (stripe_count != 0) {
1614                 spec->sp_cr_flags |= FMODE_WRITE;
1615                 echo_set_lmm_size(env, ld, ma);
1616                 if (stripe_count != -1) {
1617                         struct lov_user_md_v3 *lum = &info->eti_lum;
1618
1619                         lum->lmm_magic = LOV_USER_MAGIC_V3;
1620                         lum->lmm_stripe_count = stripe_count;
1621                         lum->lmm_stripe_offset = stripe_offset;
1622                         lum->lmm_pattern = 0;
1623                         spec->u.sp_ea.eadata = lum;
1624                         spec->u.sp_ea.eadatalen = sizeof(*lum);
1625                         spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1626                 }
1627         }
1628
1629         ma->ma_attr.la_mode = mode;
1630         ma->ma_attr.la_valid = LA_CTIME | LA_MODE;
1631         ma->ma_attr.la_ctime = cfs_time_current_64();
1632
1633         if (name != NULL) {
1634                 lname->ln_name = name;
1635                 lname->ln_namelen = namelen;
1636                 /* If name is specified, only create one object by name */
1637                 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1638                                              spec, ma);
1639                 RETURN(rc);
1640         }
1641
1642         /* Create multiple object sequenced by id */
1643         for (i = 0; i < count; i++) {
1644                 char *tmp_name = info->eti_name;
1645
1646                 echo_md_build_name(lname, tmp_name, id);
1647
1648                 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1649                                              spec, ma);
1650                 if (rc) {
1651                         CERROR("Can not create child %s: rc = %d\n", tmp_name,
1652                                 rc);
1653                         break;
1654                 }
1655                 id++;
1656                 fid->f_oid++;
1657         }
1658
1659         RETURN(rc);
1660 }
1661
1662 static struct lu_object *echo_md_lookup(const struct lu_env *env,
1663                                         struct echo_device *ed,
1664                                         struct md_object *parent,
1665                                         struct lu_name *lname)
1666 {
1667         struct echo_thread_info *info = echo_env_info(env);
1668         struct lu_fid           *fid = &info->eti_fid;
1669         struct lu_object        *child;
1670         int    rc;
1671         ENTRY;
1672
1673         CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name,
1674                PFID(fid), parent);
1675         rc = mdo_lookup(env, parent, lname, fid, NULL);
1676         if (rc) {
1677                 CERROR("lookup %s: rc = %d\n", lname->ln_name, rc);
1678                 RETURN(ERR_PTR(rc));
1679         }
1680
1681         /* In the function below, .hs_keycmp resolves to
1682          * lu_obj_hop_keycmp() */
1683         /* coverity[overrun-buffer-val] */
1684         child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1685
1686         RETURN(child);
1687 }
1688
1689 static int echo_setattr_object(const struct lu_env *env,
1690                                struct echo_device *ed,
1691                                struct lu_object *ec_parent,
1692                                __u64 id, int count)
1693 {
1694         struct lu_object        *parent;
1695         struct echo_thread_info *info = echo_env_info(env);
1696         struct lu_name          *lname = &info->eti_lname;
1697         char                    *name = info->eti_name;
1698         struct lu_device        *ld = ed->ed_next;
1699         struct lu_buf           *buf = &info->eti_buf;
1700         int                      rc = 0;
1701         int                      i;
1702
1703         ENTRY;
1704
1705         if (ec_parent == NULL)
1706                 return -1;
1707         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1708         if (parent == NULL)
1709                 RETURN(-ENXIO);
1710
1711         for (i = 0; i < count; i++) {
1712                 struct lu_object *ec_child, *child;
1713
1714                 echo_md_build_name(lname, name, id);
1715
1716                 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1717                 if (IS_ERR(ec_child)) {
1718                         CERROR("Can't find child %s: rc = %ld\n",
1719                                 lname->ln_name, PTR_ERR(ec_child));
1720                         RETURN(PTR_ERR(ec_child));
1721                 }
1722
1723                 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1724                 if (child == NULL) {
1725                         CERROR("Can not locate the child %s\n", lname->ln_name);
1726                         lu_object_put(env, ec_child);
1727                         rc = -EINVAL;
1728                         break;
1729                 }
1730
1731                 CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n",
1732                        PFID(lu_object_fid(child)));
1733
1734                 buf->lb_buf = info->eti_xattr_buf;
1735                 buf->lb_len = sizeof(info->eti_xattr_buf);
1736
1737                 sprintf(name, "%s.test1", XATTR_USER_PREFIX);
1738                 rc = mo_xattr_set(env, lu2md(child), buf, name,
1739                                   LU_XATTR_CREATE);
1740                 if (rc < 0) {
1741                         CERROR("Can not setattr child "DFID": rc = %d\n",
1742                                 PFID(lu_object_fid(child)), rc);
1743                         lu_object_put(env, ec_child);
1744                         break;
1745                 }
1746                 CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n",
1747                        PFID(lu_object_fid(child)));
1748                 id++;
1749                 lu_object_put(env, ec_child);
1750         }
1751         RETURN(rc);
1752 }
1753
1754 static int echo_getattr_object(const struct lu_env *env,
1755                                struct echo_device *ed,
1756                                struct lu_object *ec_parent,
1757                                __u64 id, int count)
1758 {
1759         struct lu_object        *parent;
1760         struct echo_thread_info *info = echo_env_info(env);
1761         struct lu_name          *lname = &info->eti_lname;
1762         char                    *name = info->eti_name;
1763         struct md_attr          *ma = &info->eti_ma;
1764         struct lu_device        *ld = ed->ed_next;
1765         int                      rc = 0;
1766         int                      i;
1767
1768         ENTRY;
1769
1770         if (ec_parent == NULL)
1771                 return -1;
1772         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1773         if (parent == NULL)
1774                 RETURN(-ENXIO);
1775
1776         memset(ma, 0, sizeof(*ma));
1777         ma->ma_need |= MA_INODE | MA_LOV | MA_PFID | MA_HSM | MA_ACL_DEF;
1778         ma->ma_acl = info->eti_xattr_buf;
1779         ma->ma_acl_size = sizeof(info->eti_xattr_buf);
1780
1781         for (i = 0; i < count; i++) {
1782                 struct lu_object *ec_child, *child;
1783
1784                 ma->ma_valid = 0;
1785                 echo_md_build_name(lname, name, id);
1786                 echo_set_lmm_size(env, ld, ma);
1787
1788                 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1789                 if (IS_ERR(ec_child)) {
1790                         CERROR("Can't find child %s: rc = %ld\n",
1791                                lname->ln_name, PTR_ERR(ec_child));
1792                         RETURN(PTR_ERR(ec_child));
1793                 }
1794
1795                 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1796                 if (child == NULL) {
1797                         CERROR("Can not locate the child %s\n", lname->ln_name);
1798                         lu_object_put(env, ec_child);
1799                         RETURN(-EINVAL);
1800                 }
1801
1802                 CDEBUG(D_RPCTRACE, "Start getattr object "DFID"\n",
1803                        PFID(lu_object_fid(child)));
1804                 rc = echo_attr_get_complex(env, lu2md(child), ma);
1805                 if (rc) {
1806                         CERROR("Can not getattr child "DFID": rc = %d\n",
1807                                 PFID(lu_object_fid(child)), rc);
1808                         lu_object_put(env, ec_child);
1809                         break;
1810                 }
1811                 CDEBUG(D_RPCTRACE, "End getattr object "DFID"\n",
1812                        PFID(lu_object_fid(child)));
1813                 id++;
1814                 lu_object_put(env, ec_child);
1815         }
1816
1817         RETURN(rc);
1818 }
1819
1820 static int echo_lookup_object(const struct lu_env *env,
1821                               struct echo_device *ed,
1822                               struct lu_object *ec_parent,
1823                               __u64 id, int count)
1824 {
1825         struct lu_object        *parent;
1826         struct echo_thread_info *info = echo_env_info(env);
1827         struct lu_name          *lname = &info->eti_lname;
1828         char                    *name = info->eti_name;
1829         struct lu_fid           *fid = &info->eti_fid;
1830         struct lu_device        *ld = ed->ed_next;
1831         int                      rc = 0;
1832         int                      i;
1833
1834         if (ec_parent == NULL)
1835                 return -1;
1836         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1837         if (parent == NULL)
1838                 return -ENXIO;
1839
1840         /*prepare the requests*/
1841         for (i = 0; i < count; i++) {
1842                 echo_md_build_name(lname, name, id);
1843
1844                 CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n",
1845                        PFID(lu_object_fid(parent)), lname->ln_name, parent);
1846
1847                 rc = mdo_lookup(env, lu2md(parent), lname, fid, NULL);
1848                 if (rc) {
1849                         CERROR("Can not lookup child %s: rc = %d\n", name, rc);
1850                         break;
1851                 }
1852                 CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n",
1853                        PFID(lu_object_fid(parent)), lname->ln_name, parent);
1854
1855                 id++;
1856         }
1857         return rc;
1858 }
1859
1860 static int echo_md_destroy_internal(const struct lu_env *env,
1861                                     struct echo_device *ed,
1862                                     struct md_object *parent,
1863                                     struct lu_name *lname,
1864                                     struct md_attr *ma)
1865 {
1866         struct lu_device   *ld = ed->ed_next;
1867         struct lu_object   *ec_child;
1868         struct lu_object   *child;
1869         int                 rc;
1870
1871         ENTRY;
1872
1873         ec_child = echo_md_lookup(env, ed, parent, lname);
1874         if (IS_ERR(ec_child)) {
1875                 CERROR("Can't find child %s: rc = %ld\n", lname->ln_name,
1876                         PTR_ERR(ec_child));
1877                 RETURN(PTR_ERR(ec_child));
1878         }
1879
1880         child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1881         if (child == NULL) {
1882                 CERROR("Can not locate the child %s\n", lname->ln_name);
1883                 GOTO(out_put, rc = -EINVAL);
1884         }
1885
1886         if (lu_object_remote(child)) {
1887                 CERROR("Can not destroy remote object %s: rc = %d\n",
1888                        lname->ln_name, -EPERM);
1889                 GOTO(out_put, rc = -EPERM);
1890         }
1891         CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n",
1892                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1893
1894         rc = mdo_unlink(env, parent, lu2md(child), lname, ma, 0);
1895         if (rc) {
1896                 CERROR("Can not unlink child %s: rc = %d\n",
1897                         lname->ln_name, rc);
1898                 GOTO(out_put, rc);
1899         }
1900         CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n",
1901                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1902 out_put:
1903         lu_object_put(env, ec_child);
1904         return rc;
1905 }
1906
1907 static int echo_destroy_object(const struct lu_env *env,
1908                                struct echo_device *ed,
1909                                struct lu_object *ec_parent,
1910                                char *name, int namelen,
1911                                __u64 id, __u32 mode,
1912                                int count)
1913 {
1914         struct echo_thread_info *info = echo_env_info(env);
1915         struct lu_name          *lname = &info->eti_lname;
1916         struct md_attr          *ma = &info->eti_ma;
1917         struct lu_device        *ld = ed->ed_next;
1918         struct lu_object        *parent;
1919         int                      rc = 0;
1920         int                      i;
1921         ENTRY;
1922
1923         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1924         if (parent == NULL)
1925                 RETURN(-EINVAL);
1926
1927         memset(ma, 0, sizeof(*ma));
1928         ma->ma_attr.la_mode = mode;
1929         ma->ma_attr.la_valid = LA_CTIME;
1930         ma->ma_attr.la_ctime = cfs_time_current_64();
1931         ma->ma_need = MA_INODE;
1932         ma->ma_valid = 0;
1933
1934         if (name != NULL) {
1935                 lname->ln_name = name;
1936                 lname->ln_namelen = namelen;
1937                 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
1938                                               ma);
1939                 RETURN(rc);
1940         }
1941
1942         /*prepare the requests*/
1943         for (i = 0; i < count; i++) {
1944                 char *tmp_name = info->eti_name;
1945
1946                 ma->ma_valid = 0;
1947                 echo_md_build_name(lname, tmp_name, id);
1948
1949                 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
1950                                               ma);
1951                 if (rc) {
1952                         CERROR("Can not unlink child %s: rc = %d\n", name, rc);
1953                         break;
1954                 }
1955                 id++;
1956         }
1957
1958         RETURN(rc);
1959 }
1960
1961 static struct lu_object *echo_resolve_path(const struct lu_env *env,
1962                                            struct echo_device *ed, char *path,
1963                                            int path_len)
1964 {
1965         struct lu_device        *ld = ed->ed_next;
1966         struct echo_thread_info *info = echo_env_info(env);
1967         struct lu_fid           *fid = &info->eti_fid;
1968         struct lu_name          *lname = &info->eti_lname;
1969         struct lu_object        *parent = NULL;
1970         struct lu_object        *child = NULL;
1971         int                      rc = 0;
1972         ENTRY;
1973
1974         *fid = ed->ed_root_fid;
1975
1976         /* In the function below, .hs_keycmp resolves to
1977          * lu_obj_hop_keycmp() */
1978         /* coverity[overrun-buffer-val] */
1979         parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1980         if (IS_ERR(parent)) {
1981                 CERROR("Can not find the parent "DFID": rc = %ld\n",
1982                         PFID(fid), PTR_ERR(parent));
1983                 RETURN(parent);
1984         }
1985
1986         while (1) {
1987                 struct lu_object *ld_parent;
1988                 char *e;
1989
1990                 e = strsep(&path, "/");
1991                 if (e == NULL)
1992                         break;
1993
1994                 if (e[0] == 0) {
1995                         if (!path || path[0] == '\0')
1996                                 break;
1997                         continue;
1998                 }
1999
2000                 lname->ln_name = e;
2001                 lname->ln_namelen = strlen(e);
2002
2003                 ld_parent = lu_object_locate(parent->lo_header, ld->ld_type);
2004                 if (ld_parent == NULL) {
2005                         lu_object_put(env, parent);
2006                         rc = -EINVAL;
2007                         break;
2008                 }
2009
2010                 child = echo_md_lookup(env, ed, lu2md(ld_parent), lname);
2011                 lu_object_put(env, parent);
2012                 if (IS_ERR(child)) {
2013                         rc = (int)PTR_ERR(child);
2014                         CERROR("lookup %s under parent "DFID": rc = %d\n",
2015                                 lname->ln_name, PFID(lu_object_fid(ld_parent)),
2016                                 rc);
2017                         break;
2018                 }
2019                 parent = child;
2020         }
2021         if (rc)
2022                 RETURN(ERR_PTR(rc));
2023
2024         RETURN(parent);
2025 }
2026
2027 static void echo_ucred_init(struct lu_env *env)
2028 {
2029         struct lu_ucred *ucred = lu_ucred(env);
2030
2031         ucred->uc_valid = UCRED_INVALID;
2032
2033         ucred->uc_suppgids[0] = -1;
2034         ucred->uc_suppgids[1] = -1;
2035
2036         ucred->uc_uid = ucred->uc_o_uid  =
2037                                 from_kuid(&init_user_ns, current_uid());
2038         ucred->uc_gid = ucred->uc_o_gid  =
2039                                 from_kgid(&init_user_ns, current_gid());
2040         ucred->uc_fsuid = ucred->uc_o_fsuid =
2041                                 from_kuid(&init_user_ns, current_fsuid());
2042         ucred->uc_fsgid = ucred->uc_o_fsgid =
2043                                 from_kgid(&init_user_ns, current_fsgid());
2044         ucred->uc_cap = cfs_curproc_cap_pack();
2045
2046         /* remove fs privilege for non-root user. */
2047         if (ucred->uc_fsuid)
2048                 ucred->uc_cap &= ~CFS_CAP_FS_MASK;
2049         ucred->uc_valid = UCRED_NEW;
2050 }
2051
2052 static void echo_ucred_fini(struct lu_env *env)
2053 {
2054         struct lu_ucred *ucred = lu_ucred(env);
2055         ucred->uc_valid = UCRED_INIT;
2056 }
2057
2058 #define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_MD_THREAD)
2059 #define ECHO_MD_SES_TAG (LCT_REMEMBER | LCT_SESSION | LCT_SERVER_SESSION)
2060 static int echo_md_handler(struct echo_device *ed, int command,
2061                            char *path, int path_len, __u64 id, int count,
2062                            struct obd_ioctl_data *data)
2063 {
2064         struct echo_thread_info *info;
2065         struct lu_device      *ld = ed->ed_next;
2066         struct lu_env         *env;
2067         int                    refcheck;
2068         struct lu_object      *parent;
2069         char                  *name = NULL;
2070         int                    namelen = data->ioc_plen2;
2071         int                    rc = 0;
2072         ENTRY;
2073
2074         if (ld == NULL) {
2075                 CERROR("MD echo client is not being initialized properly\n");
2076                 RETURN(-EINVAL);
2077         }
2078
2079         if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
2080                 CERROR("Only support MDD layer right now!\n");
2081                 RETURN(-EINVAL);
2082         }
2083
2084         env = cl_env_get(&refcheck);
2085         if (IS_ERR(env))
2086                 RETURN(PTR_ERR(env));
2087
2088         rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG);
2089         if (rc != 0)
2090                 GOTO(out_env, rc);
2091
2092         /* init big_lmm buffer */
2093         info = echo_env_info(env);
2094         LASSERT(info->eti_big_lmm == NULL);
2095         OBD_ALLOC_LARGE(info->eti_big_lmm, MIN_MD_SIZE);
2096         if (info->eti_big_lmm == NULL)
2097                 GOTO(out_env, rc = -ENOMEM);
2098         info->eti_big_lmmsize = MIN_MD_SIZE;
2099
2100         parent = echo_resolve_path(env, ed, path, path_len);
2101         if (IS_ERR(parent)) {
2102                 CERROR("Can not resolve the path %s: rc = %ld\n", path,
2103                         PTR_ERR(parent));
2104                 GOTO(out_free, rc = PTR_ERR(parent));
2105         }
2106
2107         if (namelen > 0) {
2108                 OBD_ALLOC(name, namelen + 1);
2109                 if (name == NULL)
2110                         GOTO(out_put, rc = -ENOMEM);
2111                 if (copy_from_user(name, data->ioc_pbuf2, namelen))
2112                         GOTO(out_name, rc = -EFAULT);
2113         }
2114
2115         echo_ucred_init(env);
2116
2117         switch (command) {
2118         case ECHO_MD_CREATE:
2119         case ECHO_MD_MKDIR: {
2120                 struct echo_thread_info *info = echo_env_info(env);
2121                 __u32 mode = data->ioc_obdo2.o_mode;
2122                 struct lu_fid *fid = &info->eti_fid;
2123                 int stripe_count = (int)data->ioc_obdo2.o_misc;
2124                 int stripe_index = (int)data->ioc_obdo2.o_stripe_idx;
2125
2126                 rc = ostid_to_fid(fid, &data->ioc_obdo1.o_oi, 0);
2127                 if (rc != 0)
2128                         break;
2129
2130                 /* In the function below, .hs_keycmp resolves to
2131                  * lu_obj_hop_keycmp() */
2132                 /* coverity[overrun-buffer-val] */
2133                 rc = echo_create_md_object(env, ed, parent, fid, name, namelen,
2134                                            id, mode, count, stripe_count,
2135                                            stripe_index);
2136                 break;
2137         }
2138         case ECHO_MD_DESTROY:
2139         case ECHO_MD_RMDIR: {
2140                 __u32 mode = data->ioc_obdo2.o_mode;
2141
2142                 rc = echo_destroy_object(env, ed, parent, name, namelen,
2143                                          id, mode, count);
2144                 break;
2145         }
2146         case ECHO_MD_LOOKUP:
2147                 rc = echo_lookup_object(env, ed, parent, id, count);
2148                 break;
2149         case ECHO_MD_GETATTR:
2150                 rc = echo_getattr_object(env, ed, parent, id, count);
2151                 break;
2152         case ECHO_MD_SETATTR:
2153                 rc = echo_setattr_object(env, ed, parent, id, count);
2154                 break;
2155         default:
2156                 CERROR("unknown command %d\n", command);
2157                 rc = -EINVAL;
2158                 break;
2159         }
2160         echo_ucred_fini(env);
2161
2162 out_name:
2163         if (name != NULL)
2164                 OBD_FREE(name, namelen + 1);
2165 out_put:
2166         lu_object_put(env, parent);
2167 out_free:
2168         LASSERT(info->eti_big_lmm);
2169         OBD_FREE_LARGE(info->eti_big_lmm, info->eti_big_lmmsize);
2170         info->eti_big_lmm = NULL;
2171         info->eti_big_lmmsize = 0;
2172 out_env:
2173         cl_env_put(env, &refcheck);
2174         return rc;
2175 }
2176 #endif /* HAVE_SERVER_SUPPORT */
2177
2178 static int echo_create_object(const struct lu_env *env, struct echo_device *ed,
2179                               struct obdo *oa)
2180 {
2181         struct echo_object      *eco;
2182         struct echo_client_obd  *ec = ed->ed_ec;
2183         int created = 0;
2184         int rc;
2185         ENTRY;
2186
2187         if (!(oa->o_valid & OBD_MD_FLID) ||
2188             !(oa->o_valid & OBD_MD_FLGROUP) ||
2189             !fid_seq_is_echo(ostid_seq(&oa->o_oi))) {
2190                 CERROR("invalid oid "DOSTID"\n", POSTID(&oa->o_oi));
2191                 RETURN(-EINVAL);
2192         }
2193
2194         if (ostid_id(&oa->o_oi) == 0)
2195                 ostid_set_id(&oa->o_oi, ++last_object_id);
2196
2197         rc = obd_create(env, ec->ec_exp, oa);
2198         if (rc != 0) {
2199                 CERROR("Cannot create objects: rc = %d\n", rc);
2200                 GOTO(failed, rc);
2201         }
2202
2203         created = 1;
2204
2205         oa->o_valid |= OBD_MD_FLID;
2206
2207         eco = cl_echo_object_find(ed, &oa->o_oi);
2208         if (IS_ERR(eco))
2209                 GOTO(failed, rc = PTR_ERR(eco));
2210         cl_echo_object_put(eco);
2211
2212         CDEBUG(D_INFO, "oa oid "DOSTID"\n", POSTID(&oa->o_oi));
2213         EXIT;
2214
2215 failed:
2216         if (created && rc != 0)
2217                 obd_destroy(env, ec->ec_exp, oa);
2218
2219         if (rc != 0)
2220                 CERROR("create object failed with: rc = %d\n", rc);
2221
2222         return rc;
2223 }
2224
2225 static int echo_get_object(struct echo_object **ecop, struct echo_device *ed,
2226                            struct obdo *oa)
2227 {
2228         struct echo_object *eco;
2229         int rc;
2230         ENTRY;
2231
2232         if (!(oa->o_valid & OBD_MD_FLID) ||
2233             !(oa->o_valid & OBD_MD_FLGROUP) ||
2234             ostid_id(&oa->o_oi) == 0) {
2235                 CERROR("invalid oid "DOSTID"\n", POSTID(&oa->o_oi));
2236                 RETURN(-EINVAL);
2237         }
2238
2239         rc = 0;
2240         eco = cl_echo_object_find(ed, &oa->o_oi);
2241         if (!IS_ERR(eco))
2242                 *ecop = eco;
2243         else
2244                 rc = PTR_ERR(eco);
2245
2246         RETURN(rc);
2247 }
2248
2249 static void echo_put_object(struct echo_object *eco)
2250 {
2251         int rc;
2252
2253         rc = cl_echo_object_put(eco);
2254         if (rc)
2255                 CERROR("%s: echo client drop an object failed: rc = %d\n",
2256                        eco->eo_dev->ed_ec->ec_exp->exp_obd->obd_name, rc);
2257 }
2258
2259 static void echo_client_page_debug_setup(struct page *page, int rw, u64 id,
2260                                          u64 offset, u64 count)
2261 {
2262         char    *addr;
2263         u64      stripe_off;
2264         u64      stripe_id;
2265         int      delta;
2266
2267         /* no partial pages on the client */
2268         LASSERT(count == PAGE_CACHE_SIZE);
2269
2270         addr = kmap(page);
2271
2272         for (delta = 0; delta < PAGE_CACHE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2273                 if (rw == OBD_BRW_WRITE) {
2274                         stripe_off = offset + delta;
2275                         stripe_id = id;
2276                 } else {
2277                         stripe_off = 0xdeadbeef00c0ffeeULL;
2278                         stripe_id = 0xdeadbeef00c0ffeeULL;
2279                 }
2280                 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
2281                                   stripe_off, stripe_id);
2282         }
2283
2284         kunmap(page);
2285 }
2286
2287 static int
2288 echo_client_page_debug_check(struct page *page, u64 id, u64 offset, u64 count)
2289 {
2290         u64      stripe_off;
2291         u64      stripe_id;
2292         char   *addr;
2293         int     delta;
2294         int     rc;
2295         int     rc2;
2296
2297         /* no partial pages on the client */
2298         LASSERT(count == PAGE_CACHE_SIZE);
2299
2300         addr = kmap(page);
2301
2302         for (rc = delta = 0; delta < PAGE_CACHE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2303                 stripe_off = offset + delta;
2304                 stripe_id = id;
2305
2306                 rc2 = block_debug_check("test_brw",
2307                                         addr + delta, OBD_ECHO_BLOCK_SIZE,
2308                                         stripe_off, stripe_id);
2309                 if (rc2 != 0) {
2310                         CERROR ("Error in echo object "LPX64"\n", id);
2311                         rc = rc2;
2312                 }
2313         }
2314
2315         kunmap(page);
2316         return rc;
2317 }
2318
2319 static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa,
2320                             struct echo_object *eco, u64 offset,
2321                             u64 count, int async)
2322 {
2323         size_t                  npages;
2324         struct brw_page        *pga;
2325         struct brw_page        *pgp;
2326         struct page            **pages;
2327         u64                      off;
2328         size_t                  i;
2329         int                     rc;
2330         int                     verify;
2331         gfp_t                   gfp_mask;
2332         u32                     brw_flags = 0;
2333         ENTRY;
2334
2335         verify = (ostid_id(&oa->o_oi) != ECHO_PERSISTENT_OBJID &&
2336                   (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
2337                   (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
2338
2339         gfp_mask = ((ostid_id(&oa->o_oi) & 2) == 0) ? GFP_IOFS : GFP_HIGHUSER;
2340
2341         LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
2342
2343         if ((count & (~PAGE_MASK)) != 0)
2344                 RETURN(-EINVAL);
2345
2346         /* XXX think again with misaligned I/O */
2347         npages = count >> PAGE_CACHE_SHIFT;
2348
2349         if (rw == OBD_BRW_WRITE)
2350                 brw_flags = OBD_BRW_ASYNC;
2351
2352         OBD_ALLOC(pga, npages * sizeof(*pga));
2353         if (pga == NULL)
2354                 RETURN(-ENOMEM);
2355
2356         OBD_ALLOC(pages, npages * sizeof(*pages));
2357         if (pages == NULL) {
2358                 OBD_FREE(pga, npages * sizeof(*pga));
2359                 RETURN(-ENOMEM);
2360         }
2361
2362         for (i = 0, pgp = pga, off = offset;
2363              i < npages;
2364              i++, pgp++, off += PAGE_CACHE_SIZE) {
2365
2366                 LASSERT(pgp->pg == NULL);       /* for cleanup */
2367
2368                 rc = -ENOMEM;
2369                 pgp->pg = alloc_page(gfp_mask);
2370                 if (pgp->pg == NULL)
2371                         goto out;
2372
2373                 pages[i] = pgp->pg;
2374                 pgp->count = PAGE_CACHE_SIZE;
2375                 pgp->off = off;
2376                 pgp->flag = brw_flags;
2377
2378                 if (verify)
2379                         echo_client_page_debug_setup(pgp->pg, rw,
2380                                                      ostid_id(&oa->o_oi), off,
2381                                                      pgp->count);
2382         }
2383
2384         /* brw mode can only be used at client */
2385         LASSERT(ed->ed_next != NULL);
2386         rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async);
2387
2388  out:
2389         if (rc != 0 || rw != OBD_BRW_READ)
2390                 verify = 0;
2391
2392         for (i = 0, pgp = pga; i < npages; i++, pgp++) {
2393                 if (pgp->pg == NULL)
2394                         continue;
2395
2396                 if (verify) {
2397                         int vrc;
2398                         vrc = echo_client_page_debug_check(pgp->pg,
2399                                                            ostid_id(&oa->o_oi),
2400                                                            pgp->off, pgp->count);
2401                         if (vrc != 0 && rc == 0)
2402                                 rc = vrc;
2403                 }
2404                 __free_page(pgp->pg);
2405         }
2406         OBD_FREE(pga, npages * sizeof(*pga));
2407         OBD_FREE(pages, npages * sizeof(*pages));
2408         RETURN(rc);
2409 }
2410
2411 static int echo_client_prep_commit(const struct lu_env *env,
2412                                    struct obd_export *exp, int rw,
2413                                    struct obdo *oa, struct echo_object *eco,
2414                                    u64 offset, u64 count,
2415                                    u64 batch, int async)
2416 {
2417         struct obd_ioobj         ioo;
2418         struct niobuf_local     *lnb;
2419         struct niobuf_remote     rnb;
2420         u64                      off;
2421         u64                      npages, tot_pages;
2422         int i, ret = 0, brw_flags = 0;
2423
2424         ENTRY;
2425
2426         if (count <= 0 || (count & ~PAGE_CACHE_MASK) != 0)
2427                 RETURN(-EINVAL);
2428
2429         npages = batch >> PAGE_CACHE_SHIFT;
2430         tot_pages = count >> PAGE_CACHE_SHIFT;
2431
2432         OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
2433         if (lnb == NULL)
2434                 GOTO(out, ret = -ENOMEM);
2435
2436         if (rw == OBD_BRW_WRITE && async)
2437                 brw_flags |= OBD_BRW_ASYNC;
2438
2439         obdo_to_ioobj(oa, &ioo);
2440
2441         off = offset;
2442
2443         for (; tot_pages > 0; tot_pages -= npages) {
2444                 int lpages;
2445
2446                 if (tot_pages < npages)
2447                         npages = tot_pages;
2448
2449                 rnb.rnb_offset = off;
2450                 rnb.rnb_len = npages * PAGE_CACHE_SIZE;
2451                 rnb.rnb_flags = brw_flags;
2452                 ioo.ioo_bufcnt = 1;
2453                 off += npages * PAGE_CACHE_SIZE;
2454
2455                 lpages = npages;
2456                 ret = obd_preprw(env, rw, exp, oa, 1, &ioo, &rnb, &lpages, lnb);
2457                 if (ret != 0)
2458                         GOTO(out, ret);
2459
2460                 for (i = 0; i < lpages; i++) {
2461                         struct page *page = lnb[i].lnb_page;
2462
2463                         /* read past eof? */
2464                         if (page == NULL && lnb[i].lnb_rc == 0)
2465                                 continue;
2466
2467                         if (async)
2468                                 lnb[i].lnb_flags |= OBD_BRW_ASYNC;
2469
2470                         if (ostid_id(&oa->o_oi) == ECHO_PERSISTENT_OBJID ||
2471                             (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
2472                             (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
2473                                 continue;
2474
2475                         if (rw == OBD_BRW_WRITE)
2476                                 echo_client_page_debug_setup(page, rw,
2477                                                         ostid_id(&oa->o_oi),
2478                                                         lnb[i].lnb_file_offset,
2479                                                         lnb[i].lnb_len);
2480                         else
2481                                 echo_client_page_debug_check(page,
2482                                                         ostid_id(&oa->o_oi),
2483                                                         lnb[i].lnb_file_offset,
2484                                                         lnb[i].lnb_len);
2485                 }
2486
2487                 ret = obd_commitrw(env, rw, exp, oa, 1, &ioo, &rnb, npages, lnb,
2488                                    ret);
2489                 if (ret != 0)
2490                         GOTO(out, ret);
2491
2492                 /* Reuse env context. */
2493                 lu_context_exit((struct lu_context *)&env->le_ctx);
2494                 lu_context_enter((struct lu_context *)&env->le_ctx);
2495         }
2496
2497 out:
2498         if (lnb)
2499                 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
2500         RETURN(ret);
2501 }
2502
2503 static int echo_client_brw_ioctl(const struct lu_env *env, int rw,
2504                                  struct obd_export *exp,
2505                                  struct obd_ioctl_data *data)
2506 {
2507         struct obd_device *obd = class_exp2obd(exp);
2508         struct echo_device *ed = obd2echo_dev(obd);
2509         struct echo_client_obd *ec = ed->ed_ec;
2510         struct obdo *oa = &data->ioc_obdo1;
2511         struct echo_object *eco;
2512         int rc;
2513         int async = 0;
2514         long test_mode;
2515         ENTRY;
2516
2517         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2518
2519         rc = echo_get_object(&eco, ed, oa);
2520         if (rc)
2521                 RETURN(rc);
2522
2523         oa->o_valid &= ~OBD_MD_FLHANDLE;
2524
2525         /* OFD/obdfilter works only via prep/commit */
2526         test_mode = (long)data->ioc_pbuf1;
2527         if (ed->ed_next == NULL && test_mode != 3) {
2528                 test_mode = 3;
2529                 data->ioc_plen1 = data->ioc_count;
2530         }
2531
2532         if (test_mode == 3)
2533                 async = 1;
2534
2535         /* Truncate batch size to maximum */
2536         if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE)
2537                 data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE;
2538
2539         switch (test_mode) {
2540         case 1:
2541                 /* fall through */
2542         case 2:
2543                 rc = echo_client_kbrw(ed, rw, oa, eco, data->ioc_offset,
2544                                       data->ioc_count, async);
2545                 break;
2546         case 3:
2547                 rc = echo_client_prep_commit(env, ec->ec_exp, rw, oa, eco,
2548                                              data->ioc_offset, data->ioc_count,
2549                                              data->ioc_plen1, async);
2550                 break;
2551         default:
2552                 rc = -EINVAL;
2553         }
2554
2555         echo_put_object(eco);
2556
2557         RETURN(rc);
2558 }
2559
2560 static int
2561 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2562                       void *karg, void *uarg)
2563 {
2564 #ifdef HAVE_SERVER_SUPPORT
2565         struct tgt_session_info *tsi;
2566 #endif
2567         struct obd_device      *obd = exp->exp_obd;
2568         struct echo_device     *ed = obd2echo_dev(obd);
2569         struct echo_client_obd *ec = ed->ed_ec;
2570         struct echo_object     *eco;
2571         struct obd_ioctl_data  *data = karg;
2572         struct lu_env          *env;
2573         struct obdo            *oa;
2574         struct lu_fid           fid;
2575         int                     rw = OBD_BRW_READ;
2576         int                     rc = 0;
2577 #ifdef HAVE_SERVER_SUPPORT
2578         struct lu_context        echo_session;
2579 #endif
2580         ENTRY;
2581
2582         oa = &data->ioc_obdo1;
2583         if (!(oa->o_valid & OBD_MD_FLGROUP)) {
2584                 oa->o_valid |= OBD_MD_FLGROUP;
2585                 ostid_set_seq_echo(&oa->o_oi);
2586         }
2587
2588         /* This FID is unpacked just for validation at this point */
2589         rc = ostid_to_fid(&fid, &oa->o_oi, 0);
2590         if (rc < 0)
2591                 RETURN(rc);
2592
2593         OBD_ALLOC_PTR(env);
2594         if (env == NULL)
2595                 RETURN(-ENOMEM);
2596
2597         rc = lu_env_init(env, LCT_DT_THREAD);
2598         if (rc)
2599                 GOTO(out_alloc, rc = -ENOMEM);
2600
2601 #ifdef HAVE_SERVER_SUPPORT
2602         env->le_ses = &echo_session;
2603         rc = lu_context_init(env->le_ses, LCT_SERVER_SESSION | LCT_NOREF);
2604         if (unlikely(rc < 0))
2605                 GOTO(out_env, rc);
2606         lu_context_enter(env->le_ses);
2607
2608         tsi = tgt_ses_info(env);
2609         tsi->tsi_exp = ec->ec_exp;
2610         tsi->tsi_jobid = NULL;
2611 #endif
2612         switch (cmd) {
2613         case OBD_IOC_CREATE:                    /* may create echo object */
2614                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2615                         GOTO (out, rc = -EPERM);
2616
2617                 rc = echo_create_object(env, ed, oa);
2618                 GOTO(out, rc);
2619
2620 #ifdef HAVE_SERVER_SUPPORT
2621         case OBD_IOC_ECHO_MD: {
2622                 int count;
2623                 int cmd;
2624                 char *dir = NULL;
2625                 int dirlen;
2626                 __u64 id;
2627
2628                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2629                         GOTO(out, rc = -EPERM);
2630
2631                 count = data->ioc_count;
2632                 cmd = data->ioc_command;
2633
2634                 id = data->ioc_obdo2.o_oi.oi.oi_id;
2635                 dirlen = data->ioc_plen1;
2636                 OBD_ALLOC(dir, dirlen + 1);
2637                 if (dir == NULL)
2638                         GOTO(out, rc = -ENOMEM);
2639
2640                 if (copy_from_user(dir, data->ioc_pbuf1, dirlen)) {
2641                         OBD_FREE(dir, data->ioc_plen1 + 1);
2642                         GOTO(out, rc = -EFAULT);
2643                 }
2644
2645                 rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data);
2646                 OBD_FREE(dir, dirlen + 1);
2647                 GOTO(out, rc);
2648         }
2649         case OBD_IOC_ECHO_ALLOC_SEQ: {
2650                 struct lu_env   *cl_env;
2651                 int              refcheck;
2652                 __u64            seq;
2653                 int              max_count;
2654
2655                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2656                         GOTO(out, rc = -EPERM);
2657
2658                 cl_env = cl_env_get(&refcheck);
2659                 if (IS_ERR(cl_env))
2660                         GOTO(out, rc = PTR_ERR(cl_env));
2661
2662                 rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG,
2663                                             ECHO_MD_SES_TAG);
2664                 if (rc != 0) {
2665                         cl_env_put(cl_env, &refcheck);
2666                         GOTO(out, rc);
2667                 }
2668
2669                 rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq);
2670                 cl_env_put(cl_env, &refcheck);
2671                 if (rc < 0) {
2672                         CERROR("%s: Can not alloc seq: rc = %d\n",
2673                                obd->obd_name, rc);
2674                         GOTO(out, rc);
2675                 }
2676
2677                 if (copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1))
2678                         return -EFAULT;
2679
2680                 max_count = LUSTRE_METADATA_SEQ_MAX_WIDTH;
2681                 if (copy_to_user(data->ioc_pbuf2, &max_count,
2682                                      data->ioc_plen2))
2683                         return -EFAULT;
2684                 GOTO(out, rc);
2685         }
2686 #endif /* HAVE_SERVER_SUPPORT */
2687         case OBD_IOC_DESTROY:
2688                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2689                         GOTO (out, rc = -EPERM);
2690
2691                 rc = echo_get_object(&eco, ed, oa);
2692                 if (rc == 0) {
2693                         rc = obd_destroy(env, ec->ec_exp, oa);
2694                         if (rc == 0)
2695                                 eco->eo_deleted = 1;
2696                         echo_put_object(eco);
2697                 }
2698                 GOTO(out, rc);
2699
2700         case OBD_IOC_GETATTR:
2701                 rc = echo_get_object(&eco, ed, oa);
2702                 if (rc == 0) {
2703                         rc = obd_getattr(env, ec->ec_exp, oa);
2704                         echo_put_object(eco);
2705                 }
2706                 GOTO(out, rc);
2707
2708         case OBD_IOC_SETATTR:
2709                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2710                         GOTO (out, rc = -EPERM);
2711
2712                 rc = echo_get_object(&eco, ed, oa);
2713                 if (rc == 0) {
2714                         rc = obd_setattr(env, ec->ec_exp, oa);
2715                         echo_put_object(eco);
2716                 }
2717                 GOTO(out, rc);
2718
2719         case OBD_IOC_BRW_WRITE:
2720                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2721                         GOTO (out, rc = -EPERM);
2722
2723                 rw = OBD_BRW_WRITE;
2724                 /* fall through */
2725         case OBD_IOC_BRW_READ:
2726                 rc = echo_client_brw_ioctl(env, rw, exp, data);
2727                 GOTO(out, rc);
2728
2729         default:
2730                 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
2731                 GOTO (out, rc = -ENOTTY);
2732         }
2733
2734         EXIT;
2735 out:
2736 #ifdef HAVE_SERVER_SUPPORT
2737         lu_context_exit(env->le_ses);
2738         lu_context_fini(env->le_ses);
2739 out_env:
2740 #endif
2741         lu_env_fini(env);
2742 out_alloc:
2743         OBD_FREE_PTR(env);
2744
2745         return rc;
2746 }
2747
2748 static int echo_client_setup(const struct lu_env *env,
2749                              struct obd_device *obddev, struct lustre_cfg *lcfg)
2750 {
2751         struct echo_client_obd *ec = &obddev->u.echo_client;
2752         struct obd_device *tgt;
2753         struct obd_uuid echo_uuid = { "ECHO_UUID" };
2754         struct obd_connect_data *ocd = NULL;
2755         int rc;
2756         ENTRY;
2757
2758         if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2759                 CERROR("requires a TARGET OBD name\n");
2760                 RETURN(-EINVAL);
2761         }
2762
2763         tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
2764         if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
2765                 CERROR("device not attached or not set up (%s)\n",
2766                        lustre_cfg_string(lcfg, 1));
2767                 RETURN(-EINVAL);
2768         }
2769
2770         spin_lock_init(&ec->ec_lock);
2771         INIT_LIST_HEAD(&ec->ec_objects);
2772         INIT_LIST_HEAD(&ec->ec_locks);
2773         ec->ec_unique = 0;
2774
2775         if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
2776 #ifdef HAVE_SERVER_SUPPORT
2777                 lu_context_tags_update(ECHO_MD_CTX_TAG);
2778                 lu_session_tags_update(ECHO_MD_SES_TAG);
2779 #else
2780                 CERROR("Local operations are NOT supported on client side. "
2781                        "Only remote operations are supported. Metadata client "
2782                        "must be run on server side.\n");
2783 #endif
2784                 RETURN(0);
2785         }
2786
2787         OBD_ALLOC(ocd, sizeof(*ocd));
2788         if (ocd == NULL) {
2789                 CERROR("Can't alloc ocd connecting to %s\n",
2790                        lustre_cfg_string(lcfg, 1));
2791                 return -ENOMEM;
2792         }
2793
2794         ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL |
2795                                  OBD_CONNECT_BRW_SIZE |
2796                                  OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 |
2797                                  OBD_CONNECT_64BITHASH | OBD_CONNECT_LVB_TYPE |
2798                                  OBD_CONNECT_FID;
2799         ocd->ocd_brw_size = DT_MAX_BRW_SIZE;
2800         ocd->ocd_version = LUSTRE_VERSION_CODE;
2801         ocd->ocd_group = FID_SEQ_ECHO;
2802
2803         rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL);
2804         if (rc == 0) {
2805                 /* Turn off pinger because it connects to tgt obd directly. */
2806                 spin_lock(&tgt->obd_dev_lock);
2807                 list_del_init(&ec->ec_exp->exp_obd_chain_timed);
2808                 spin_unlock(&tgt->obd_dev_lock);
2809         }
2810
2811         OBD_FREE(ocd, sizeof(*ocd));
2812
2813         if (rc != 0) {
2814                 CERROR("fail to connect to device %s\n",
2815                        lustre_cfg_string(lcfg, 1));
2816                 return (rc);
2817         }
2818
2819         RETURN(rc);
2820 }
2821
2822 static int echo_client_cleanup(struct obd_device *obddev)
2823 {
2824         struct echo_device *ed = obd2echo_dev(obddev);
2825         struct echo_client_obd *ec = &obddev->u.echo_client;
2826         int rc;
2827         ENTRY;
2828
2829         /*Do nothing for Metadata echo client*/
2830         if (ed == NULL )
2831                 RETURN(0);
2832
2833         if (ed->ed_next_ismd) {
2834 #ifdef HAVE_SERVER_SUPPORT
2835                 lu_context_tags_clear(ECHO_MD_CTX_TAG);
2836                 lu_session_tags_clear(ECHO_MD_SES_TAG);
2837 #else
2838                 CERROR("This is client-side only module, does not support "
2839                         "metadata echo client.\n");
2840 #endif
2841                 RETURN(0);
2842         }
2843
2844         if (!list_empty(&obddev->obd_exports)) {
2845                 CERROR("still has clients!\n");
2846                 RETURN(-EBUSY);
2847         }
2848
2849         LASSERT(atomic_read(&ec->ec_exp->exp_refcount) > 0);
2850         rc = obd_disconnect(ec->ec_exp);
2851         if (rc != 0)
2852                 CERROR("fail to disconnect device: %d\n", rc);
2853
2854         RETURN(rc);
2855 }
2856
2857 static int echo_client_connect(const struct lu_env *env,
2858                                struct obd_export **exp,
2859                                struct obd_device *src, struct obd_uuid *cluuid,
2860                                struct obd_connect_data *data, void *localdata)
2861 {
2862         int                rc;
2863         struct lustre_handle conn = { 0 };
2864
2865         ENTRY;
2866         rc = class_connect(&conn, src, cluuid);
2867         if (rc == 0) {
2868                 *exp = class_conn2export(&conn);
2869         }
2870
2871         RETURN (rc);
2872 }
2873
2874 static int echo_client_disconnect(struct obd_export *exp)
2875 {
2876         int                     rc;
2877         ENTRY;
2878
2879         if (exp == NULL)
2880                 GOTO(out, rc = -EINVAL);
2881
2882         rc = class_disconnect(exp);
2883         GOTO(out, rc);
2884  out:
2885         return rc;
2886 }
2887
2888 static struct obd_ops echo_client_obd_ops = {
2889         .o_owner       = THIS_MODULE,
2890         .o_iocontrol   = echo_client_iocontrol,
2891         .o_connect     = echo_client_connect,
2892         .o_disconnect  = echo_client_disconnect
2893 };
2894
2895 static int echo_client_init(void)
2896 {
2897         int rc;
2898
2899         rc = lu_kmem_init(echo_caches);
2900         if (rc == 0) {
2901                 rc = class_register_type(&echo_client_obd_ops, NULL, true, NULL,
2902                                          LUSTRE_ECHO_CLIENT_NAME,
2903                                          &echo_device_type);
2904                 if (rc)
2905                         lu_kmem_fini(echo_caches);
2906         }
2907         return rc;
2908 }
2909
2910 static void echo_client_exit(void)
2911 {
2912         class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
2913         lu_kmem_fini(echo_caches);
2914 }
2915
2916 static int __init obdecho_init(void)
2917 {
2918         int rc;
2919
2920         ENTRY;
2921         LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
2922
2923         LASSERT(PAGE_CACHE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
2924
2925 # ifdef HAVE_SERVER_SUPPORT
2926         rc = echo_persistent_pages_init();
2927         if (rc != 0)
2928                 goto failed_0;
2929
2930         rc = class_register_type(&echo_obd_ops, NULL, true, NULL,
2931                                  LUSTRE_ECHO_NAME, NULL);
2932         if (rc != 0)
2933                 goto failed_1;
2934 # endif
2935
2936         rc = echo_client_init();
2937
2938 # ifdef HAVE_SERVER_SUPPORT
2939         if (rc == 0)
2940                 RETURN(0);
2941
2942         class_unregister_type(LUSTRE_ECHO_NAME);
2943 failed_1:
2944         echo_persistent_pages_fini();
2945 failed_0:
2946 # endif
2947         RETURN(rc);
2948 }
2949
2950 static void /*__exit*/ obdecho_exit(void)
2951 {
2952         echo_client_exit();
2953
2954 # ifdef HAVE_SERVER_SUPPORT
2955         class_unregister_type(LUSTRE_ECHO_NAME);
2956         echo_persistent_pages_fini();
2957 # endif
2958 }
2959
2960 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2961 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
2962 MODULE_VERSION(LUSTRE_VERSION_STRING);
2963 MODULE_LICENSE("GPL");
2964
2965 module_init(obdecho_init);
2966 module_exit(obdecho_exit);
2967
2968 /** @} echo_client */