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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #define DEBUG_SUBSYSTEM S_ECHO
  34
  35 #include <linux/user_namespace.h>
  36 #ifdef HAVE_UIDGID_HEADER
  37 # include <linux/uidgid.h>
  38 #endif
  39 #include <libcfs/libcfs.h>
  40
  41 #include <obd.h>
  42 #include <obd_support.h>
  43 #include <obd_class.h>
  44 #include <lustre_debug.h>
  45 #include <lprocfs_status.h>
  46 #include <cl_object.h>
  47 #include <lustre_fid.h>
  48 #include <lustre_lmv.h>
  49 #include <lustre_acl.h>
  50 #include <uapi/linux/lustre/lustre_ioctl.h>
  51 #include <lustre_net.h>
  52 #ifdef HAVE_SERVER_SUPPORT
  53 # include <md_object.h>
  54
  55 #define ETI_NAME_LEN    20
  56
  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; /* may be vmalloc'd */
 234         int                     eti_big_lmmsize;
 235         char                    eti_name[ETI_NAME_LEN];
 236         struct lu_buf           eti_buf;
 237         /* If we want to test large ACL, then need to enlarge the buffer. */
 238         char                    eti_xattr_buf[LUSTRE_POSIX_ACL_MAX_SIZE_OLD];
 239 #endif
 240 };
 241
 242 /* No session used right now */
 243 struct echo_session_info {
 244         unsigned long dummy;
 245 };
 246
 247 static struct kmem_cache *echo_lock_kmem;
 248 static struct kmem_cache *echo_object_kmem;
 249 static struct kmem_cache *echo_thread_kmem;
 250 static struct kmem_cache *echo_session_kmem;
 251 /* static struct kmem_cache *echo_req_kmem; */
 252
 253 static struct lu_kmem_descr echo_caches[] = {
 254         {
 255                 .ckd_cache = &echo_lock_kmem,
 256                 .ckd_name  = "echo_lock_kmem",
 257                 .ckd_size  = sizeof (struct echo_lock)
 258         },
 259         {
 260                 .ckd_cache = &echo_object_kmem,
 261                 .ckd_name  = "echo_object_kmem",
 262                 .ckd_size  = sizeof (struct echo_object)
 263         },
 264         {
 265                 .ckd_cache = &echo_thread_kmem,
 266                 .ckd_name  = "echo_thread_kmem",
 267                 .ckd_size  = sizeof (struct echo_thread_info)
 268         },
 269         {
 270                 .ckd_cache = &echo_session_kmem,
 271                 .ckd_name  = "echo_session_kmem",
 272                 .ckd_size  = sizeof (struct echo_session_info)
 273         },
 274         {
 275                 .ckd_cache = NULL
 276         }
 277 };
 278
 279 /** \defgroup echo_page Page operations
 280  *
 281  * Echo page operations.
 282  *
 283  * @{
 284  */
 285 static int echo_page_own(const struct lu_env *env,
 286                          const struct cl_page_slice *slice,
 287                          struct cl_io *io, int nonblock)
 288 {
 289         struct echo_page *ep = cl2echo_page(slice);
 290
 291         if (!nonblock)
 292                 mutex_lock(&ep->ep_lock);
 293         else if (!mutex_trylock(&ep->ep_lock))
 294                 return -EAGAIN;
 295         return 0;
 296 }
 297
 298 static void echo_page_disown(const struct lu_env *env,
 299                              const struct cl_page_slice *slice,
 300                              struct cl_io *io)
 301 {
 302         struct echo_page *ep = cl2echo_page(slice);
 303
 304         LASSERT(mutex_is_locked(&ep->ep_lock));
 305         mutex_unlock(&ep->ep_lock);
 306 }
 307
 308 static void echo_page_discard(const struct lu_env *env,
 309                               const struct cl_page_slice *slice,
 310                               struct cl_io *unused)
 311 {
 312         cl_page_delete(env, slice->cpl_page);
 313 }
 314
 315 static int echo_page_is_vmlocked(const struct lu_env *env,
 316                                  const struct cl_page_slice *slice)
 317 {
 318         if (mutex_is_locked(&cl2echo_page(slice)->ep_lock))
 319                 return -EBUSY;
 320         return -ENODATA;
 321 }
 322
 323 static void echo_page_completion(const struct lu_env *env,
 324                                  const struct cl_page_slice *slice,
 325                                  int ioret)
 326 {
 327         LASSERT(slice->cpl_page->cp_sync_io != NULL);
 328 }
 329
 330 static void echo_page_fini(const struct lu_env *env,
 331                            struct cl_page_slice *slice)
 332 {
 333         struct echo_object *eco = cl2echo_obj(slice->cpl_obj);
 334         ENTRY;
 335
 336         atomic_dec(&eco->eo_npages);
 337         put_page(slice->cpl_page->cp_vmpage);
 338         EXIT;
 339 }
 340
 341 static int echo_page_prep(const struct lu_env *env,
 342                           const struct cl_page_slice *slice,
 343                           struct cl_io *unused)
 344 {
 345         return 0;
 346 }
 347
 348 static int echo_page_print(const struct lu_env *env,
 349                            const struct cl_page_slice *slice,
 350                            void *cookie, lu_printer_t printer)
 351 {
 352         struct echo_page *ep = cl2echo_page(slice);
 353
 354         (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p %d vm@%p\n",
 355                    ep, mutex_is_locked(&ep->ep_lock),
 356                    slice->cpl_page->cp_vmpage);
 357         return 0;
 358 }
 359
 360 static const struct cl_page_operations echo_page_ops = {
 361         .cpo_own           = echo_page_own,
 362         .cpo_disown        = echo_page_disown,
 363         .cpo_discard       = echo_page_discard,
 364         .cpo_fini          = echo_page_fini,
 365         .cpo_print         = echo_page_print,
 366         .cpo_is_vmlocked   = echo_page_is_vmlocked,
 367         .io = {
 368                 [CRT_READ] = {
 369                         .cpo_prep        = echo_page_prep,
 370                         .cpo_completion  = echo_page_completion,
 371                 },
 372                 [CRT_WRITE] = {
 373                         .cpo_prep        = echo_page_prep,
 374                         .cpo_completion  = echo_page_completion,
 375                 }
 376         }
 377 };
 378 /** @} echo_page */
 379
 380 /** \defgroup echo_lock Locking
 381  *
 382  * echo lock operations
 383  *
 384  * @{
 385  */
 386 static void echo_lock_fini(const struct lu_env *env,
 387                            struct cl_lock_slice *slice)
 388 {
 389         struct echo_lock *ecl = cl2echo_lock(slice);
 390
 391         LASSERT(list_empty(&ecl->el_chain));
 392         OBD_SLAB_FREE_PTR(ecl, echo_lock_kmem);
 393 }
 394
 395 static struct cl_lock_operations echo_lock_ops = {
 396         .clo_fini      = echo_lock_fini,
 397 };
 398
 399 /** @} echo_lock */
 400
 401 /** \defgroup echo_cl_ops cl_object operations
 402  *
 403  * operations for cl_object
 404  *
 405  * @{
 406  */
 407 static int echo_page_init(const struct lu_env *env, struct cl_object *obj,
 408                           struct cl_page *page, pgoff_t index)
 409 {
 410         struct echo_page *ep = cl_object_page_slice(obj, page);
 411         struct echo_object *eco = cl2echo_obj(obj);
 412         ENTRY;
 413
 414         get_page(page->cp_vmpage);
 415         mutex_init(&ep->ep_lock);
 416         cl_page_slice_add(page, &ep->ep_cl, obj, index, &echo_page_ops);
 417         atomic_inc(&eco->eo_npages);
 418         RETURN(0);
 419 }
 420
 421 static int echo_io_init(const struct lu_env *env, struct cl_object *obj,
 422                         struct cl_io *io)
 423 {
 424         return 0;
 425 }
 426
 427 static int echo_lock_init(const struct lu_env *env,
 428                           struct cl_object *obj, struct cl_lock *lock,
 429                           const struct cl_io *unused)
 430 {
 431         struct echo_lock *el;
 432         ENTRY;
 433
 434         OBD_SLAB_ALLOC_PTR_GFP(el, echo_lock_kmem, GFP_NOFS);
 435         if (el != NULL) {
 436                 cl_lock_slice_add(lock, &el->el_cl, obj, &echo_lock_ops);
 437                 el->el_object = cl2echo_obj(obj);
 438                 INIT_LIST_HEAD(&el->el_chain);
 439                 atomic_set(&el->el_refcount, 0);
 440         }
 441         RETURN(el == NULL ? -ENOMEM : 0);
 442 }
 443
 444 static int echo_conf_set(const struct lu_env *env, struct cl_object *obj,
 445                          const struct cl_object_conf *conf)
 446 {
 447         return 0;
 448 }
 449
 450 static const struct cl_object_operations echo_cl_obj_ops = {
 451         .coo_page_init = echo_page_init,
 452         .coo_lock_init = echo_lock_init,
 453         .coo_io_init   = echo_io_init,
 454         .coo_conf_set  = echo_conf_set
 455 };
 456 /** @} echo_cl_ops */
 457
 458 /** \defgroup echo_lu_ops lu_object operations
 459  *
 460  * operations for echo lu object.
 461  *
 462  * @{
 463  */
 464 static int echo_object_init(const struct lu_env *env, struct lu_object *obj,
 465                             const struct lu_object_conf *conf)
 466 {
 467         struct echo_device *ed         = cl2echo_dev(lu2cl_dev(obj->lo_dev));
 468         struct echo_client_obd *ec     = ed->ed_ec;
 469         struct echo_object *eco        = cl2echo_obj(lu2cl(obj));
 470         ENTRY;
 471
 472         if (ed->ed_next) {
 473                 struct lu_object  *below;
 474                 struct lu_device  *under;
 475
 476                 under = ed->ed_next;
 477                 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header,
 478                                                         under);
 479                 if (below == NULL)
 480                         RETURN(-ENOMEM);
 481                 lu_object_add(obj, below);
 482         }
 483
 484         if (!ed->ed_next_ismd) {
 485                 const struct cl_object_conf *cconf = lu2cl_conf(conf);
 486                 struct echo_object_conf *econf = cl2echo_conf(cconf);
 487
 488                 LASSERT(econf->eoc_oinfo != NULL);
 489
 490                 /* Transfer the oinfo pointer to eco that it won't be
 491                  * freed. */
 492                 eco->eo_oinfo = *econf->eoc_oinfo;
 493                 *econf->eoc_oinfo = NULL;
 494         } else {
 495                 eco->eo_oinfo = NULL;
 496         }
 497
 498         eco->eo_dev = ed;
 499         atomic_set(&eco->eo_npages, 0);
 500         cl_object_page_init(lu2cl(obj), sizeof(struct echo_page));
 501
 502         spin_lock(&ec->ec_lock);
 503         list_add_tail(&eco->eo_obj_chain, &ec->ec_objects);
 504         spin_unlock(&ec->ec_lock);
 505
 506         RETURN(0);
 507 }
 508
 509 static void echo_object_free(const struct lu_env *env, struct lu_object *obj)
 510 {
 511         struct echo_object *eco    = cl2echo_obj(lu2cl(obj));
 512         struct echo_client_obd *ec = eco->eo_dev->ed_ec;
 513         ENTRY;
 514
 515         LASSERT(atomic_read(&eco->eo_npages) == 0);
 516
 517         spin_lock(&ec->ec_lock);
 518         list_del_init(&eco->eo_obj_chain);
 519         spin_unlock(&ec->ec_lock);
 520
 521         lu_object_fini(obj);
 522         lu_object_header_fini(obj->lo_header);
 523
 524         if (eco->eo_oinfo != NULL)
 525                 OBD_FREE_PTR(eco->eo_oinfo);
 526
 527         OBD_SLAB_FREE_PTR(eco, echo_object_kmem);
 528         EXIT;
 529 }
 530
 531 static int echo_object_print(const struct lu_env *env, void *cookie,
 532                             lu_printer_t p, const struct lu_object *o)
 533 {
 534         struct echo_object *obj = cl2echo_obj(lu2cl(o));
 535
 536         return (*p)(env, cookie, "echoclient-object@%p", obj);
 537 }
 538
 539 static const struct lu_object_operations echo_lu_obj_ops = {
 540         .loo_object_init      = echo_object_init,
 541         .loo_object_delete    = NULL,
 542         .loo_object_release   = NULL,
 543         .loo_object_free      = echo_object_free,
 544         .loo_object_print     = echo_object_print,
 545         .loo_object_invariant = NULL
 546 };
 547 /** @} echo_lu_ops */
 548
 549 /** \defgroup echo_lu_dev_ops  lu_device operations
 550  *
 551  * Operations for echo lu device.
 552  *
 553  * @{
 554  */
 555 static struct lu_object *echo_object_alloc(const struct lu_env *env,
 556                                            const struct lu_object_header *hdr,
 557                                            struct lu_device *dev)
 558 {
 559         struct echo_object *eco;
 560         struct lu_object *obj = NULL;
 561         ENTRY;
 562
 563         /* we're the top dev. */
 564         LASSERT(hdr == NULL);
 565         OBD_SLAB_ALLOC_PTR_GFP(eco, echo_object_kmem, GFP_NOFS);
 566         if (eco != NULL) {
 567                 struct cl_object_header *hdr = &eco->eo_hdr;
 568
 569                 obj = &echo_obj2cl(eco)->co_lu;
 570                 cl_object_header_init(hdr);
 571                 hdr->coh_page_bufsize = cfs_size_round(sizeof(struct cl_page));
 572
 573                 lu_object_init(obj, &hdr->coh_lu, dev);
 574                 lu_object_add_top(&hdr->coh_lu, obj);
 575
 576                 eco->eo_cl.co_ops = &echo_cl_obj_ops;
 577                 obj->lo_ops       = &echo_lu_obj_ops;
 578         }
 579         RETURN(obj);
 580 }
 581
 582 static struct lu_device_operations echo_device_lu_ops = {
 583         .ldo_object_alloc   = echo_object_alloc,
 584 };
 585
 586 /** @} echo_lu_dev_ops */
 587
 588 /** \defgroup echo_init Setup and teardown
 589  *
 590  * Init and fini functions for echo client.
 591  *
 592  * @{
 593  */
 594 static int echo_site_init(const struct lu_env *env, struct echo_device *ed)
 595 {
 596         struct cl_site *site = &ed->ed_site_myself;
 597         int rc;
 598
 599         /* initialize site */
 600         rc = cl_site_init(site, &ed->ed_cl);
 601         if (rc) {
 602                 CERROR("Cannot initialize site for echo client(%d)\n", rc);
 603                 return rc;
 604         }
 605
 606         rc = lu_site_init_finish(&site->cs_lu);
 607         if (rc) {
 608                 cl_site_fini(site);
 609                 return rc;
 610         }
 611
 612         ed->ed_site = &site->cs_lu;
 613         return 0;
 614 }
 615
 616 static void echo_site_fini(const struct lu_env *env, struct echo_device *ed)
 617 {
 618         if (ed->ed_site) {
 619                 if (!ed->ed_next_ismd)
 620                         lu_site_fini(ed->ed_site);
 621                 ed->ed_site = NULL;
 622         }
 623 }
 624
 625 static void *echo_thread_key_init(const struct lu_context *ctx,
 626                                   struct lu_context_key *key)
 627 {
 628         struct echo_thread_info *info;
 629
 630         OBD_SLAB_ALLOC_PTR_GFP(info, echo_thread_kmem, GFP_NOFS);
 631         if (info == NULL)
 632                 info = ERR_PTR(-ENOMEM);
 633         return info;
 634 }
 635
 636 static void echo_thread_key_fini(const struct lu_context *ctx,
 637                          struct lu_context_key *key, void *data)
 638 {
 639         struct echo_thread_info *info = data;
 640         OBD_SLAB_FREE_PTR(info, echo_thread_kmem);
 641 }
 642
 643 static struct lu_context_key echo_thread_key = {
 644         .lct_tags = LCT_CL_THREAD,
 645         .lct_init = echo_thread_key_init,
 646         .lct_fini = echo_thread_key_fini,
 647 };
 648
 649 static void *echo_session_key_init(const struct lu_context *ctx,
 650                                   struct lu_context_key *key)
 651 {
 652         struct echo_session_info *session;
 653
 654         OBD_SLAB_ALLOC_PTR_GFP(session, echo_session_kmem, GFP_NOFS);
 655         if (session == NULL)
 656                 session = ERR_PTR(-ENOMEM);
 657         return session;
 658 }
 659
 660 static void echo_session_key_fini(const struct lu_context *ctx,
 661                                  struct lu_context_key *key, void *data)
 662 {
 663         struct echo_session_info *session = data;
 664         OBD_SLAB_FREE_PTR(session, echo_session_kmem);
 665 }
 666
 667 static struct lu_context_key echo_session_key = {
 668         .lct_tags = LCT_SESSION,
 669         .lct_init = echo_session_key_init,
 670         .lct_fini = echo_session_key_fini,
 671 };
 672
 673 LU_TYPE_INIT_FINI(echo, &echo_thread_key, &echo_session_key);
 674
 675 #ifdef HAVE_SERVER_SUPPORT
 676 # define ECHO_SEQ_WIDTH 0xffffffff
 677 static int echo_fid_init(struct echo_device *ed, char *obd_name,
 678                          struct seq_server_site *ss)
 679 {
 680         char *prefix;
 681         int rc;
 682         ENTRY;
 683
 684         OBD_ALLOC_PTR(ed->ed_cl_seq);
 685         if (ed->ed_cl_seq == NULL)
 686                 RETURN(-ENOMEM);
 687
 688         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
 689         if (prefix == NULL)
 690                 GOTO(out_free_seq, rc = -ENOMEM);
 691
 692         snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", obd_name);
 693
 694         /* Init client side sequence-manager */
 695         rc = seq_client_init(ed->ed_cl_seq, NULL,
 696                              LUSTRE_SEQ_METADATA,
 697                              prefix, ss->ss_server_seq);
 698         ed->ed_cl_seq->lcs_width = ECHO_SEQ_WIDTH;
 699         OBD_FREE(prefix, MAX_OBD_NAME + 5);
 700         if (rc)
 701                 GOTO(out_free_seq, rc);
 702
 703         RETURN(0);
 704
 705 out_free_seq:
 706         OBD_FREE_PTR(ed->ed_cl_seq);
 707         ed->ed_cl_seq = NULL;
 708         RETURN(rc);
 709 }
 710
 711 static int echo_fid_fini(struct obd_device *obddev)
 712 {
 713         struct echo_device *ed = obd2echo_dev(obddev);
 714         ENTRY;
 715
 716         if (ed->ed_cl_seq != NULL) {
 717                 seq_client_fini(ed->ed_cl_seq);
 718                 OBD_FREE_PTR(ed->ed_cl_seq);
 719                 ed->ed_cl_seq = NULL;
 720         }
 721
 722         RETURN(0);
 723 }
 724
 725 static void echo_ed_los_fini(const struct lu_env *env, struct echo_device *ed)
 726 {
 727         ENTRY;
 728
 729         if (ed != NULL && ed->ed_next_ismd && ed->ed_los != NULL) {
 730                 local_oid_storage_fini(env, ed->ed_los);
 731                 ed->ed_los = NULL;
 732         }
 733 }
 734
 735 static int
 736 echo_md_local_file_create(const struct lu_env *env, struct echo_md_device *emd,
 737                           struct local_oid_storage *los,
 738                           const struct lu_fid *pfid, const char *name,
 739                           __u32 mode, struct lu_fid *fid)
 740 {
 741         struct dt_object        *parent = NULL;
 742         struct dt_object        *dto = NULL;
 743         int                      rc = 0;
 744         ENTRY;
 745
 746         LASSERT(!fid_is_zero(pfid));
 747         parent = dt_locate(env, emd->emd_bottom, pfid);
 748         if (unlikely(IS_ERR(parent)))
 749                 RETURN(PTR_ERR(parent));
 750
 751         /* create local file with @fid */
 752         dto = local_file_find_or_create_with_fid(env, emd->emd_bottom, fid,
 753                                                  parent, name, mode);
 754         if (IS_ERR(dto))
 755                 GOTO(out_put, rc = PTR_ERR(dto));
 756
 757         *fid = *lu_object_fid(&dto->do_lu);
 758         /* since stack is not fully set up the local_storage uses own stack
 759          * and we should drop its object from cache */
 760         dt_object_put_nocache(env, dto);
 761
 762         EXIT;
 763 out_put:
 764         dt_object_put(env, parent);
 765         RETURN(rc);
 766 }
 767
 768 static int
 769 echo_md_root_get(const struct lu_env *env, struct echo_md_device *emd,
 770                  struct echo_device *ed)
 771 {
 772         struct lu_fid                    fid;
 773         int                              rc = 0;
 774         ENTRY;
 775
 776         /* Setup local dirs */
 777         fid.f_seq = FID_SEQ_LOCAL_NAME;
 778         fid.f_oid = 1;
 779         fid.f_ver = 0;
 780         rc = local_oid_storage_init(env, emd->emd_bottom, &fid, &ed->ed_los);
 781         if (rc != 0)
 782                 RETURN(rc);
 783
 784         lu_echo_root_fid(&fid);
 785         if (echo_md_seq_site(emd)->ss_node_id == 0) {
 786                 rc = echo_md_local_file_create(env, emd, ed->ed_los,
 787                                                &emd->emd_local_root_fid,
 788                                                echo_md_root_dir_name, S_IFDIR |
 789                                                S_IRUGO | S_IWUSR | S_IXUGO,
 790                                                &fid);
 791                 if (rc != 0) {
 792                         CERROR("%s: create md echo root fid failed: rc = %d\n",
 793                                emd2obd_dev(emd)->obd_name, rc);
 794                         GOTO(out_los, rc);
 795                 }
 796         }
 797         ed->ed_root_fid = fid;
 798
 799         RETURN(0);
 800 out_los:
 801         echo_ed_los_fini(env, ed);
 802
 803         RETURN(rc);
 804 }
 805 #endif /* HAVE_SERVER_SUPPORT */
 806
 807 static struct lu_device *echo_device_alloc(const struct lu_env *env,
 808                                            struct lu_device_type *t,
 809                                            struct lustre_cfg *cfg)
 810 {
 811         struct lu_device   *next;
 812         struct echo_device *ed;
 813         struct cl_device   *cd;
 814         struct obd_device  *obd = NULL; /* to keep compiler happy */
 815         struct obd_device  *tgt;
 816         const char *tgt_type_name;
 817         int rc;
 818         int cleanup = 0;
 819         ENTRY;
 820
 821         OBD_ALLOC_PTR(ed);
 822         if (ed == NULL)
 823                 GOTO(out, rc = -ENOMEM);
 824
 825         cleanup = 1;
 826         cd = &ed->ed_cl;
 827         rc = cl_device_init(cd, t);
 828         if (rc)
 829                 GOTO(out, rc);
 830
 831         cd->cd_lu_dev.ld_ops = &echo_device_lu_ops;
 832
 833         cleanup = 2;
 834         obd = class_name2obd(lustre_cfg_string(cfg, 0));
 835         LASSERT(obd != NULL);
 836         LASSERT(env != NULL);
 837
 838         tgt = class_name2obd(lustre_cfg_string(cfg, 1));
 839         if (tgt == NULL) {
 840                 CERROR("Can not find tgt device %s\n",
 841                         lustre_cfg_string(cfg, 1));
 842                 GOTO(out, rc = -ENODEV);
 843         }
 844
 845         next = tgt->obd_lu_dev;
 846
 847         if (strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME) == 0) {
 848                 ed->ed_next_ismd = 1;
 849         } else if (strcmp(tgt->obd_type->typ_name, LUSTRE_OST_NAME) == 0 ||
 850                    strcmp(tgt->obd_type->typ_name, LUSTRE_OSC_NAME) == 0) {
 851                 ed->ed_next_ismd = 0;
 852                 rc = echo_site_init(env, ed);
 853                 if (rc)
 854                         GOTO(out, rc);
 855         } else {
 856                 GOTO(out, rc = -EINVAL);
 857         }
 858
 859         cleanup = 3;
 860
 861         rc = echo_client_setup(env, obd, cfg);
 862         if (rc)
 863                 GOTO(out, rc);
 864
 865         ed->ed_ec = &obd->u.echo_client;
 866         cleanup = 4;
 867
 868         if (ed->ed_next_ismd) {
 869 #ifdef HAVE_SERVER_SUPPORT
 870                 /* Suppose to connect to some Metadata layer */
 871                 struct lu_site          *ls = NULL;
 872                 struct lu_device        *ld = NULL;
 873                 struct md_device        *md = NULL;
 874                 struct echo_md_device   *emd = NULL;
 875                 int                      found = 0;
 876
 877                 if (next == NULL) {
 878                         CERROR("%s is not lu device type!\n",
 879                                lustre_cfg_string(cfg, 1));
 880                         GOTO(out, rc = -EINVAL);
 881                 }
 882
 883                 tgt_type_name = lustre_cfg_string(cfg, 2);
 884                 if (!tgt_type_name) {
 885                         CERROR("%s no type name for echo %s setup\n",
 886                                 lustre_cfg_string(cfg, 1),
 887                                 tgt->obd_type->typ_name);
 888                         GOTO(out, rc = -EINVAL);
 889                 }
 890
 891                 ls = next->ld_site;
 892
 893                 spin_lock(&ls->ls_ld_lock);
 894                 list_for_each_entry(ld, &ls->ls_ld_linkage, ld_linkage) {
 895                         if (strcmp(ld->ld_type->ldt_name, tgt_type_name) == 0) {
 896                                 found = 1;
 897                                 break;
 898                         }
 899                 }
 900                 spin_unlock(&ls->ls_ld_lock);
 901
 902                 if (found == 0) {
 903                         CERROR("%s is not lu device type!\n",
 904                                lustre_cfg_string(cfg, 1));
 905                         GOTO(out, rc = -EINVAL);
 906                 }
 907
 908                 next = ld;
 909                 /* For MD echo client, it will use the site in MDS stack */
 910                 ed->ed_site = ls;
 911                 ed->ed_cl.cd_lu_dev.ld_site = ls;
 912                 rc = echo_fid_init(ed, obd->obd_name, lu_site2seq(ls));
 913                 if (rc) {
 914                         CERROR("echo fid init error %d\n", rc);
 915                         GOTO(out, rc);
 916                 }
 917
 918                 md = lu2md_dev(next);
 919                 emd = lu2emd_dev(&md->md_lu_dev);
 920                 rc = echo_md_root_get(env, emd, ed);
 921                 if (rc != 0) {
 922                         CERROR("%s: get root error: rc = %d\n",
 923                                 emd2obd_dev(emd)->obd_name, rc);
 924                         GOTO(out, rc);
 925                 }
 926 #else /* !HAVE_SERVER_SUPPORT */
 927                 CERROR("Local operations are NOT supported on client side. "
 928                        "Only remote operations are supported. Metadata client "
 929                        "must be run on server side.\n");
 930                 GOTO(out, rc = -EOPNOTSUPP);
 931 #endif /* HAVE_SERVER_SUPPORT */
 932         } else {
 933                  /* if echo client is to be stacked upon ost device, the next is
 934                   * NULL since ost is not a clio device so far */
 935                 if (next != NULL && !lu_device_is_cl(next))
 936                         next = NULL;
 937
 938                 tgt_type_name = tgt->obd_type->typ_name;
 939                 if (next != NULL) {
 940                         LASSERT(next != NULL);
 941                         if (next->ld_site != NULL)
 942                                 GOTO(out, rc = -EBUSY);
 943
 944                         next->ld_site = ed->ed_site;
 945                         rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
 946                                                      next->ld_type->ldt_name,
 947                                                      NULL);
 948                         if (rc)
 949                                 GOTO(out, rc);
 950                 } else
 951                         LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0);
 952         }
 953
 954         ed->ed_next = next;
 955         RETURN(&cd->cd_lu_dev);
 956 out:
 957         switch(cleanup) {
 958         case 4: {
 959                 int rc2;
 960                 rc2 = echo_client_cleanup(obd);
 961                 if (rc2)
 962                         CERROR("Cleanup obd device %s error(%d)\n",
 963                                obd->obd_name, rc2);
 964         }
 965
 966         case 3:
 967                 echo_site_fini(env, ed);
 968         case 2:
 969                 cl_device_fini(&ed->ed_cl);
 970         case 1:
 971                 OBD_FREE_PTR(ed);
 972         case 0:
 973         default:
 974                 break;
 975         }
 976         return(ERR_PTR(rc));
 977 }
 978
 979 static int echo_device_init(const struct lu_env *env, struct lu_device *d,
 980                           const char *name, struct lu_device *next)
 981 {
 982         LBUG();
 983         return 0;
 984 }
 985
 986 static struct lu_device *echo_device_fini(const struct lu_env *env,
 987                                           struct lu_device *d)
 988 {
 989         struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
 990         struct lu_device *next = ed->ed_next;
 991
 992         while (next && !ed->ed_next_ismd)
 993                 next = next->ld_type->ldt_ops->ldto_device_fini(env, next);
 994         return NULL;
 995 }
 996
 997 static void echo_lock_release(const struct lu_env *env,
 998                               struct echo_lock *ecl,
 999                               int still_used)
1000 {
1001         struct cl_lock *clk = echo_lock2cl(ecl);
1002
1003         cl_lock_release(env, clk);
1004 }
1005
1006 static struct lu_device *echo_device_free(const struct lu_env *env,
1007                                           struct lu_device *d)
1008 {
1009         struct echo_device     *ed   = cl2echo_dev(lu2cl_dev(d));
1010         struct echo_client_obd *ec   = ed->ed_ec;
1011         struct echo_object     *eco;
1012         struct lu_device       *next = ed->ed_next;
1013
1014         CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n",
1015                ed, next);
1016
1017         lu_site_purge(env, ed->ed_site, -1);
1018
1019         /* check if there are objects still alive.
1020          * It shouldn't have any object because lu_site_purge would cleanup
1021          * all of cached objects. Anyway, probably the echo device is being
1022          * parallelly accessed.
1023          */
1024         spin_lock(&ec->ec_lock);
1025         list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain)
1026                 eco->eo_deleted = 1;
1027         spin_unlock(&ec->ec_lock);
1028
1029         /* purge again */
1030         lu_site_purge(env, ed->ed_site, -1);
1031
1032         CDEBUG(D_INFO,
1033                "Waiting for the reference of echo object to be dropped\n");
1034
1035         /* Wait for the last reference to be dropped. */
1036         spin_lock(&ec->ec_lock);
1037         while (!list_empty(&ec->ec_objects)) {
1038                 spin_unlock(&ec->ec_lock);
1039                 CERROR("echo_client still has objects at cleanup time, "
1040                        "wait for 1 second\n");
1041                 set_current_state(TASK_UNINTERRUPTIBLE);
1042                 schedule_timeout(cfs_time_seconds(1));
1043                 lu_site_purge(env, ed->ed_site, -1);
1044                 spin_lock(&ec->ec_lock);
1045         }
1046         spin_unlock(&ec->ec_lock);
1047
1048         LASSERT(list_empty(&ec->ec_locks));
1049
1050         CDEBUG(D_INFO, "No object exists, exiting...\n");
1051
1052         echo_client_cleanup(d->ld_obd);
1053 #ifdef HAVE_SERVER_SUPPORT
1054         echo_fid_fini(d->ld_obd);
1055         echo_ed_los_fini(env, ed);
1056 #endif
1057         while (next && !ed->ed_next_ismd)
1058                 next = next->ld_type->ldt_ops->ldto_device_free(env, next);
1059
1060         LASSERT(ed->ed_site == d->ld_site);
1061         echo_site_fini(env, ed);
1062         cl_device_fini(&ed->ed_cl);
1063         OBD_FREE_PTR(ed);
1064
1065         cl_env_cache_purge(~0);
1066
1067         return NULL;
1068 }
1069
1070 static const struct lu_device_type_operations echo_device_type_ops = {
1071         .ldto_init = echo_type_init,
1072         .ldto_fini = echo_type_fini,
1073
1074         .ldto_start = echo_type_start,
1075         .ldto_stop  = echo_type_stop,
1076
1077         .ldto_device_alloc = echo_device_alloc,
1078         .ldto_device_free  = echo_device_free,
1079         .ldto_device_init  = echo_device_init,
1080         .ldto_device_fini  = echo_device_fini
1081 };
1082
1083 static struct lu_device_type echo_device_type = {
1084         .ldt_tags     = LU_DEVICE_CL,
1085         .ldt_name     = LUSTRE_ECHO_CLIENT_NAME,
1086         .ldt_ops      = &echo_device_type_ops,
1087         .ldt_ctx_tags = LCT_CL_THREAD | LCT_MD_THREAD | LCT_DT_THREAD,
1088 };
1089 /** @} echo_init */
1090
1091 /** \defgroup echo_exports Exported operations
1092  *
1093  * exporting functions to echo client
1094  *
1095  * @{
1096  */
1097
1098 /* Interfaces to echo client obd device */
1099 static struct echo_object *
1100 cl_echo_object_find(struct echo_device *d, const struct ost_id *oi)
1101 {
1102         struct lu_env *env;
1103         struct echo_thread_info *info;
1104         struct echo_object_conf *conf;
1105         struct echo_object *eco;
1106         struct cl_object *obj;
1107         struct lov_oinfo *oinfo = NULL;
1108         struct lu_fid *fid;
1109         __u16  refcheck;
1110         int rc;
1111         ENTRY;
1112
1113         LASSERTF(ostid_id(oi) != 0, DOSTID"\n", POSTID(oi));
1114         LASSERTF(ostid_seq(oi) == FID_SEQ_ECHO, DOSTID"\n", POSTID(oi));
1115
1116         /* Never return an object if the obd is to be freed. */
1117         if (echo_dev2cl(d)->cd_lu_dev.ld_obd->obd_stopping)
1118                 RETURN(ERR_PTR(-ENODEV));
1119
1120         env = cl_env_get(&refcheck);
1121         if (IS_ERR(env))
1122                 RETURN((void *)env);
1123
1124         info = echo_env_info(env);
1125         conf = &info->eti_conf;
1126         if (d->ed_next) {
1127                 OBD_ALLOC_PTR(oinfo);
1128                 if (oinfo == NULL)
1129                         GOTO(out, eco = ERR_PTR(-ENOMEM));
1130
1131                 oinfo->loi_oi = *oi;
1132                 conf->eoc_cl.u.coc_oinfo = oinfo;
1133         }
1134
1135         /* If echo_object_init() is successful then ownership of oinfo
1136          * is transferred to the object. */
1137         conf->eoc_oinfo = &oinfo;
1138
1139         fid = &info->eti_fid;
1140         rc = ostid_to_fid(fid, oi, 0);
1141         if (rc != 0)
1142                 GOTO(out, eco = ERR_PTR(rc));
1143
1144         /* In the function below, .hs_keycmp resolves to
1145          * lu_obj_hop_keycmp() */
1146         /* coverity[overrun-buffer-val] */
1147         obj = cl_object_find(env, echo_dev2cl(d), fid, &conf->eoc_cl);
1148         if (IS_ERR(obj))
1149                 GOTO(out, eco = (void*)obj);
1150
1151         eco = cl2echo_obj(obj);
1152         if (eco->eo_deleted) {
1153                 cl_object_put(env, obj);
1154                 eco = ERR_PTR(-EAGAIN);
1155         }
1156
1157 out:
1158         if (oinfo != NULL)
1159                 OBD_FREE_PTR(oinfo);
1160
1161         cl_env_put(env, &refcheck);
1162         RETURN(eco);
1163 }
1164
1165 static int cl_echo_object_put(struct echo_object *eco)
1166 {
1167         struct lu_env *env;
1168         struct cl_object *obj = echo_obj2cl(eco);
1169         __u16  refcheck;
1170         ENTRY;
1171
1172         env = cl_env_get(&refcheck);
1173         if (IS_ERR(env))
1174                 RETURN(PTR_ERR(env));
1175
1176         /* an external function to kill an object? */
1177         if (eco->eo_deleted) {
1178                 struct lu_object_header *loh = obj->co_lu.lo_header;
1179                 LASSERT(&eco->eo_hdr == luh2coh(loh));
1180                 set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags);
1181         }
1182
1183         cl_object_put(env, obj);
1184         cl_env_put(env, &refcheck);
1185         RETURN(0);
1186 }
1187
1188 static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco,
1189                             u64 start, u64 end, int mode,
1190                             __u64 *cookie , __u32 enqflags)
1191 {
1192         struct cl_io *io;
1193         struct cl_lock *lck;
1194         struct cl_object *obj;
1195         struct cl_lock_descr *descr;
1196         struct echo_thread_info *info;
1197         int rc = -ENOMEM;
1198         ENTRY;
1199
1200         info = echo_env_info(env);
1201         io = &info->eti_io;
1202         lck = &info->eti_lock;
1203         obj = echo_obj2cl(eco);
1204
1205         memset(lck, 0, sizeof(*lck));
1206         descr = &lck->cll_descr;
1207         descr->cld_obj   = obj;
1208         descr->cld_start = cl_index(obj, start);
1209         descr->cld_end   = cl_index(obj, end);
1210         descr->cld_mode  = mode == LCK_PW ? CLM_WRITE : CLM_READ;
1211         descr->cld_enq_flags = enqflags;
1212         io->ci_obj = obj;
1213
1214         rc = cl_lock_request(env, io, lck);
1215         if (rc == 0) {
1216                 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
1217                 struct echo_lock *el;
1218
1219                 el = cl2echo_lock(cl_lock_at(lck, &echo_device_type));
1220                 spin_lock(&ec->ec_lock);
1221                 if (list_empty(&el->el_chain)) {
1222                         list_add(&el->el_chain, &ec->ec_locks);
1223                         el->el_cookie = ++ec->ec_unique;
1224                 }
1225                 atomic_inc(&el->el_refcount);
1226                 *cookie = el->el_cookie;
1227                 spin_unlock(&ec->ec_lock);
1228         }
1229         RETURN(rc);
1230 }
1231
1232 static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed,
1233                            __u64 cookie)
1234 {
1235         struct echo_client_obd *ec = ed->ed_ec;
1236         struct echo_lock       *ecl = NULL;
1237         struct list_head        *el;
1238         int found = 0, still_used = 0;
1239         ENTRY;
1240
1241         LASSERT(ec != NULL);
1242         spin_lock(&ec->ec_lock);
1243         list_for_each(el, &ec->ec_locks) {
1244                 ecl = list_entry(el, struct echo_lock, el_chain);
1245                 CDEBUG(D_INFO, "ecl: %p, cookie: %#llx\n", ecl, ecl->el_cookie);
1246                 found = (ecl->el_cookie == cookie);
1247                 if (found) {
1248                         if (atomic_dec_and_test(&ecl->el_refcount))
1249                                 list_del_init(&ecl->el_chain);
1250                         else
1251                                 still_used = 1;
1252                         break;
1253                 }
1254         }
1255         spin_unlock(&ec->ec_lock);
1256
1257         if (!found)
1258                 RETURN(-ENOENT);
1259
1260         echo_lock_release(env, ecl, still_used);
1261         RETURN(0);
1262 }
1263
1264 static void echo_commit_callback(const struct lu_env *env, struct cl_io *io,
1265                                 struct cl_page *page)
1266 {
1267         struct echo_thread_info *info;
1268         struct cl_2queue        *queue;
1269
1270         info = echo_env_info(env);
1271         LASSERT(io == &info->eti_io);
1272
1273         queue = &info->eti_queue;
1274         cl_page_list_add(&queue->c2_qout, page);
1275 }
1276
1277 static int cl_echo_object_brw(struct echo_object *eco, int rw, u64 offset,
1278                               struct page **pages, int npages, int async)
1279 {
1280         struct lu_env           *env;
1281         struct echo_thread_info *info;
1282         struct cl_object        *obj = echo_obj2cl(eco);
1283         struct echo_device      *ed  = eco->eo_dev;
1284         struct cl_2queue        *queue;
1285         struct cl_io            *io;
1286         struct cl_page          *clp;
1287         struct lustre_handle    lh = { 0 };
1288         int page_size = cl_page_size(obj);
1289         int rc;
1290         int i;
1291         __u16 refcheck;
1292         ENTRY;
1293
1294         LASSERT((offset & ~PAGE_MASK) == 0);
1295         LASSERT(ed->ed_next != NULL);
1296         env = cl_env_get(&refcheck);
1297         if (IS_ERR(env))
1298                 RETURN(PTR_ERR(env));
1299
1300         info    = echo_env_info(env);
1301         io      = &info->eti_io;
1302         queue   = &info->eti_queue;
1303
1304         cl_2queue_init(queue);
1305
1306         io->ci_ignore_layout = 1;
1307         rc = cl_io_init(env, io, CIT_MISC, obj);
1308         if (rc < 0)
1309                 GOTO(out, rc);
1310         LASSERT(rc == 0);
1311
1312
1313         rc = cl_echo_enqueue0(env, eco, offset,
1314                               offset + npages * PAGE_SIZE - 1,
1315                               rw == READ ? LCK_PR : LCK_PW, &lh.cookie,
1316                               CEF_NEVER);
1317         if (rc < 0)
1318                 GOTO(error_lock, rc);
1319
1320         for (i = 0; i < npages; i++) {
1321                 LASSERT(pages[i]);
1322                 clp = cl_page_find(env, obj, cl_index(obj, offset),
1323                                    pages[i], CPT_TRANSIENT);
1324                 if (IS_ERR(clp)) {
1325                         rc = PTR_ERR(clp);
1326                         break;
1327                 }
1328                 LASSERT(clp->cp_type == CPT_TRANSIENT);
1329
1330                 rc = cl_page_own(env, io, clp);
1331                 if (rc) {
1332                         LASSERT(clp->cp_state == CPS_FREEING);
1333                         cl_page_put(env, clp);
1334                         break;
1335                 }
1336
1337                 cl_2queue_add(queue, clp);
1338
1339                 /* drop the reference count for cl_page_find, so that the page
1340                  * will be freed in cl_2queue_fini. */
1341                 cl_page_put(env, clp);
1342                 cl_page_clip(env, clp, 0, page_size);
1343
1344                 offset += page_size;
1345         }
1346
1347         if (rc == 0) {
1348                 enum cl_req_type typ = rw == READ ? CRT_READ : CRT_WRITE;
1349
1350                 async = async && (typ == CRT_WRITE);
1351                 if (async)
1352                         rc = cl_io_commit_async(env, io, &queue->c2_qin,
1353                                                 0, PAGE_SIZE,
1354                                                 echo_commit_callback);
1355                 else
1356                         rc = cl_io_submit_sync(env, io, typ, queue, 0);
1357                 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
1358                        async ? "async" : "sync", rc);
1359         }
1360
1361         cl_echo_cancel0(env, ed, lh.cookie);
1362         EXIT;
1363 error_lock:
1364         cl_2queue_discard(env, io, queue);
1365         cl_2queue_disown(env, io, queue);
1366         cl_2queue_fini(env, queue);
1367         cl_io_fini(env, io);
1368 out:
1369         cl_env_put(env, &refcheck);
1370         return rc;
1371 }
1372 /** @} echo_exports */
1373
1374
1375 static u64 last_object_id;
1376
1377 #ifdef HAVE_SERVER_SUPPORT
1378 static inline void echo_md_build_name(struct lu_name *lname, char *name,
1379                                       __u64 id)
1380 {
1381         snprintf(name, ETI_NAME_LEN, "%llu", id);
1382         lname->ln_name = name;
1383         lname->ln_namelen = strlen(name);
1384 }
1385
1386 /* similar to mdt_attr_get_complex */
1387 static int echo_big_lmm_get(const struct lu_env *env, struct md_object *o,
1388                             struct md_attr *ma)
1389 {
1390         struct echo_thread_info *info = echo_env_info(env);
1391         int                      rc;
1392
1393         ENTRY;
1394
1395         LASSERT(ma->ma_lmm_size > 0);
1396
1397         LASSERT(ma->ma_need & (MA_LOV | MA_LMV));
1398         if (ma->ma_need & MA_LOV)
1399                 rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LOV);
1400         else
1401                 rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LMV);
1402
1403         if (rc < 0)
1404                 RETURN(rc);
1405
1406         /* big_lmm may need to be grown */
1407         if (info->eti_big_lmmsize < rc) {
1408                 int size = size_roundup_power2(rc);
1409
1410                 if (info->eti_big_lmmsize > 0) {
1411                         /* free old buffer */
1412                         LASSERT(info->eti_big_lmm);
1413                         OBD_FREE_LARGE(info->eti_big_lmm,
1414                                        info->eti_big_lmmsize);
1415                         info->eti_big_lmm = NULL;
1416                         info->eti_big_lmmsize = 0;
1417                 }
1418
1419                 OBD_ALLOC_LARGE(info->eti_big_lmm, size);
1420                 if (info->eti_big_lmm == NULL)
1421                         RETURN(-ENOMEM);
1422                 info->eti_big_lmmsize = size;
1423         }
1424         LASSERT(info->eti_big_lmmsize >= rc);
1425
1426         info->eti_buf.lb_buf = info->eti_big_lmm;
1427         info->eti_buf.lb_len = info->eti_big_lmmsize;
1428         if (ma->ma_need & MA_LOV)
1429                 rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LOV);
1430         else
1431                 rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LMV);
1432         if (rc < 0)
1433                 RETURN(rc);
1434
1435         if (ma->ma_need & MA_LOV)
1436                 ma->ma_valid |= MA_LOV;
1437         else
1438                 ma->ma_valid |= MA_LMV;
1439
1440         ma->ma_lmm = info->eti_big_lmm;
1441         ma->ma_lmm_size = rc;
1442
1443         RETURN(0);
1444 }
1445
1446 static int echo_attr_get_complex(const struct lu_env *env,
1447                                  struct md_object *next,
1448                                  struct md_attr *ma)
1449 {
1450         struct echo_thread_info *info = echo_env_info(env);
1451         struct lu_buf           *buf = &info->eti_buf;
1452         umode_t                  mode = lu_object_attr(&next->mo_lu);
1453         int                      rc = 0, rc2;
1454
1455         ENTRY;
1456
1457         ma->ma_valid = 0;
1458
1459         if (ma->ma_need & MA_INODE) {
1460                 rc = mo_attr_get(env, next, ma);
1461                 if (rc)
1462                         GOTO(out, rc);
1463                 ma->ma_valid |= MA_INODE;
1464         }
1465
1466         if ((ma->ma_need & MA_LOV) && (S_ISREG(mode) || S_ISDIR(mode))) {
1467                 LASSERT(ma->ma_lmm_size > 0);
1468                 buf->lb_buf = ma->ma_lmm;
1469                 buf->lb_len = ma->ma_lmm_size;
1470                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LOV);
1471                 if (rc2 > 0) {
1472                         ma->ma_lmm_size = rc2;
1473                         ma->ma_valid |= MA_LOV;
1474                 } else if (rc2 == -ENODATA) {
1475                         /* no LOV EA */
1476                         ma->ma_lmm_size = 0;
1477                 } else if (rc2 == -ERANGE) {
1478                         rc2 = echo_big_lmm_get(env, next, ma);
1479                         if (rc2 < 0)
1480                                 GOTO(out, rc = rc2);
1481                 } else {
1482                         GOTO(out, rc = rc2);
1483                 }
1484         }
1485
1486         if ((ma->ma_need & MA_LMV) && S_ISDIR(mode)) {
1487                 LASSERT(ma->ma_lmm_size > 0);
1488                 buf->lb_buf = ma->ma_lmm;
1489                 buf->lb_len = ma->ma_lmm_size;
1490                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LMV);
1491                 if (rc2 > 0) {
1492                         ma->ma_lmm_size = rc2;
1493                         ma->ma_valid |= MA_LMV;
1494                 } else if (rc2 == -ENODATA) {
1495                         /* no LMV EA */
1496                         ma->ma_lmm_size = 0;
1497                 } else if (rc2 == -ERANGE) {
1498                         rc2 = echo_big_lmm_get(env, next, ma);
1499                         if (rc2 < 0)
1500                                 GOTO(out, rc = rc2);
1501                 } else {
1502                         GOTO(out, rc = rc2);
1503                 }
1504         }
1505
1506 #ifdef CONFIG_FS_POSIX_ACL
1507         if ((ma->ma_need & MA_ACL_DEF) && S_ISDIR(mode)) {
1508                 buf->lb_buf = ma->ma_acl;
1509                 buf->lb_len = ma->ma_acl_size;
1510                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1511                 if (rc2 > 0) {
1512                         ma->ma_acl_size = rc2;
1513                         ma->ma_valid |= MA_ACL_DEF;
1514                 } else if (rc2 == -ENODATA) {
1515                         /* no ACLs */
1516                         ma->ma_acl_size = 0;
1517                 } else {
1518                         GOTO(out, rc = rc2);
1519                 }
1520         }
1521 #endif
1522 out:
1523         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1524                rc, ma->ma_valid, ma->ma_lmm);
1525         RETURN(rc);
1526 }
1527
1528 static int
1529 echo_md_create_internal(const struct lu_env *env, struct echo_device *ed,
1530                         struct md_object *parent, struct lu_fid *fid,
1531                         struct lu_name *lname, struct md_op_spec *spec,
1532                         struct md_attr *ma)
1533 {
1534         struct lu_object        *ec_child, *child;
1535         struct lu_device        *ld = ed->ed_next;
1536         struct echo_thread_info *info = echo_env_info(env);
1537         struct lu_fid           *fid2 = &info->eti_fid2;
1538         struct lu_object_conf    conf = { .loc_flags = LOC_F_NEW };
1539         int                      rc;
1540
1541         ENTRY;
1542
1543         rc = mdo_lookup(env, parent, lname, fid2, spec);
1544         if (rc == 0)
1545                 return -EEXIST;
1546         else if (rc != -ENOENT)
1547                 return rc;
1548
1549         ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev,
1550                                      fid, &conf);
1551         if (IS_ERR(ec_child)) {
1552                 CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid),
1553                         PTR_ERR(ec_child));
1554                 RETURN(PTR_ERR(ec_child));
1555         }
1556
1557         child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1558         if (child == NULL) {
1559                 CERROR("Can not locate the child "DFID"\n", PFID(fid));
1560                 GOTO(out_put, rc = -EINVAL);
1561         }
1562
1563         CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n",
1564                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1565
1566         /*
1567          * Do not perform lookup sanity check. We know that name does not exist.
1568          */
1569         spec->sp_cr_lookup = 0;
1570         rc = mdo_create(env, parent, lname, lu2md(child), spec, ma);
1571         if (rc) {
1572                 CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc);
1573                 GOTO(out_put, rc);
1574         }
1575         CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc  = %d\n",
1576                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc);
1577         EXIT;
1578 out_put:
1579         lu_object_put(env, ec_child);
1580         return rc;
1581 }
1582
1583 static int echo_set_lmm_size(const struct lu_env *env, struct lu_device *ld,
1584                              struct md_attr *ma)
1585 {
1586         struct echo_thread_info *info = echo_env_info(env);
1587
1588         if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
1589                 ma->ma_lmm = (void *)&info->eti_lmm;
1590                 ma->ma_lmm_size = sizeof(info->eti_lmm);
1591         } else {
1592                 LASSERT(info->eti_big_lmmsize);
1593                 ma->ma_lmm = info->eti_big_lmm;
1594                 ma->ma_lmm_size = info->eti_big_lmmsize;
1595         }
1596
1597         return 0;
1598 }
1599
1600 static int
1601 echo_md_dir_stripe_choose(const struct lu_env *env, struct echo_device *ed,
1602                           struct lu_object *obj, const char *name,
1603                           unsigned int namelen, __u64 id,
1604                           struct lu_object **new_parent)
1605 {
1606         struct echo_thread_info *info = echo_env_info(env);
1607         struct md_attr          *ma = &info->eti_ma;
1608         struct lmv_mds_md_v1    *lmv;
1609         struct lu_device        *ld = ed->ed_next;
1610         unsigned int            idx;
1611         struct lu_name          tmp_ln_name;
1612         struct lu_fid           stripe_fid;
1613         struct lu_object        *stripe_obj;
1614         int                     rc;
1615
1616         LASSERT(obj != NULL);
1617         LASSERT(S_ISDIR(obj->lo_header->loh_attr));
1618
1619         memset(ma, 0, sizeof(*ma));
1620         echo_set_lmm_size(env, ld, ma);
1621         ma->ma_need = MA_LMV;
1622         rc = echo_attr_get_complex(env, lu2md(obj), ma);
1623         if (rc) {
1624                 CERROR("Can not getattr child "DFID": rc = %d\n",
1625                         PFID(lu_object_fid(obj)), rc);
1626                 return rc;
1627         }
1628
1629         if (!(ma->ma_valid & MA_LMV)) {
1630                 *new_parent = obj;
1631                 return 0;
1632         }
1633
1634         lmv = (struct lmv_mds_md_v1 *)ma->ma_lmm;
1635         if (le32_to_cpu(lmv->lmv_magic) != LMV_MAGIC_V1) {
1636                 rc = -EINVAL;
1637                 CERROR("Invalid mds md magic %x "DFID": rc = %d\n",
1638                        le32_to_cpu(lmv->lmv_magic), PFID(lu_object_fid(obj)),
1639                        rc);
1640                 return rc;
1641         }
1642
1643         if (name != NULL) {
1644                 tmp_ln_name.ln_name = name;
1645                 tmp_ln_name.ln_namelen = namelen;
1646         } else {
1647                 LASSERT(id != -1);
1648                 echo_md_build_name(&tmp_ln_name, info->eti_name, id);
1649         }
1650
1651         idx = lmv_name_to_stripe_index(LMV_HASH_TYPE_FNV_1A_64,
1652                                 le32_to_cpu(lmv->lmv_stripe_count),
1653                                 tmp_ln_name.ln_name, tmp_ln_name.ln_namelen);
1654
1655         LASSERT(idx < le32_to_cpu(lmv->lmv_stripe_count));
1656         fid_le_to_cpu(&stripe_fid, &lmv->lmv_stripe_fids[idx]);
1657
1658         stripe_obj = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, &stripe_fid,
1659                                        NULL);
1660         if (IS_ERR(stripe_obj)) {
1661                 rc = PTR_ERR(stripe_obj);
1662                 CERROR("Can not find the parent "DFID": rc = %d\n",
1663                        PFID(&stripe_fid), rc);
1664                 return rc;
1665         }
1666
1667         *new_parent = lu_object_locate(stripe_obj->lo_header, ld->ld_type);
1668         if (*new_parent == NULL) {
1669                 lu_object_put(env, stripe_obj);
1670                 RETURN(-ENXIO);
1671         }
1672
1673         return rc;
1674 }
1675
1676 static int echo_create_md_object(const struct lu_env *env,
1677                                  struct echo_device *ed,
1678                                  struct lu_object *ec_parent,
1679                                  struct lu_fid *fid,
1680                                  char *name, int namelen,
1681                                  __u64 id, __u32 mode, int count,
1682                                  int stripe_count, int stripe_offset)
1683 {
1684         struct lu_object        *parent;
1685         struct lu_object        *new_parent;
1686         struct echo_thread_info *info = echo_env_info(env);
1687         struct lu_name          *lname = &info->eti_lname;
1688         struct md_op_spec       *spec = &info->eti_spec;
1689         struct md_attr          *ma = &info->eti_ma;
1690         struct lu_device        *ld = ed->ed_next;
1691         int                      rc = 0;
1692         int                      i;
1693
1694         ENTRY;
1695
1696         if (ec_parent == NULL)
1697                 return -1;
1698         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1699         if (parent == NULL)
1700                 RETURN(-ENXIO);
1701
1702         rc = echo_md_dir_stripe_choose(env, ed, parent, name, namelen,
1703                                        id, &new_parent);
1704         if (rc != 0)
1705                 RETURN(rc);
1706
1707         LASSERT(new_parent != NULL);
1708         memset(ma, 0, sizeof(*ma));
1709         memset(spec, 0, sizeof(*spec));
1710         echo_set_lmm_size(env, ld, ma);
1711         if (stripe_count != 0) {
1712                 spec->sp_cr_flags |= MDS_FMODE_WRITE;
1713                 if (stripe_count != -1) {
1714                         if (S_ISDIR(mode)) {
1715                                 struct lmv_user_md *lmu;
1716
1717                                 lmu = (struct lmv_user_md *)&info->eti_lum;
1718                                 lmu->lum_magic = LMV_USER_MAGIC;
1719                                 lmu->lum_stripe_offset = stripe_offset;
1720                                 lmu->lum_stripe_count = stripe_count;
1721                                 lmu->lum_hash_type = LMV_HASH_TYPE_FNV_1A_64;
1722                                 spec->u.sp_ea.eadata = lmu;
1723                                 spec->u.sp_ea.eadatalen = sizeof(*lmu);
1724                         } else {
1725                                 struct lov_user_md_v3 *lum = &info->eti_lum;
1726
1727                                 lum->lmm_magic = LOV_USER_MAGIC_V3;
1728                                 lum->lmm_stripe_count = stripe_count;
1729                                 lum->lmm_stripe_offset = stripe_offset;
1730                                 lum->lmm_pattern = LOV_PATTERN_NONE;
1731                                 spec->u.sp_ea.eadata = lum;
1732                                 spec->u.sp_ea.eadatalen = sizeof(*lum);
1733                         }
1734                         spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1735                 }
1736         }
1737
1738         ma->ma_attr.la_mode = mode;
1739         ma->ma_attr.la_valid = LA_CTIME | LA_MODE;
1740         ma->ma_attr.la_ctime = ktime_get_real_seconds();
1741
1742         if (name != NULL) {
1743                 lname->ln_name = name;
1744                 lname->ln_namelen = namelen;
1745                 /* If name is specified, only create one object by name */
1746                 rc = echo_md_create_internal(env, ed, lu2md(new_parent), fid,
1747                                              lname, spec, ma);
1748                 GOTO(out_put, rc);
1749         }
1750
1751         /* Create multiple object sequenced by id */
1752         for (i = 0; i < count; i++) {
1753                 char *tmp_name = info->eti_name;
1754
1755                 echo_md_build_name(lname, tmp_name, id);
1756
1757                 rc = echo_md_create_internal(env, ed, lu2md(new_parent),
1758                                              fid, lname, spec, ma);
1759                 if (rc) {
1760                         CERROR("Can not create child %s: rc = %d\n", tmp_name,
1761                                 rc);
1762                         break;
1763                 }
1764                 id++;
1765                 fid->f_oid++;
1766         }
1767
1768 out_put:
1769         if (new_parent != parent)
1770                 lu_object_put(env, new_parent);
1771
1772         RETURN(rc);
1773 }
1774
1775 static struct lu_object *echo_md_lookup(const struct lu_env *env,
1776                                         struct echo_device *ed,
1777                                         struct md_object *parent,
1778                                         struct lu_name *lname)
1779 {
1780         struct echo_thread_info *info = echo_env_info(env);
1781         struct lu_fid           *fid = &info->eti_fid;
1782         struct lu_object        *child;
1783         int    rc;
1784         ENTRY;
1785
1786         CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name,
1787                PFID(fid), parent);
1788
1789         rc = mdo_lookup(env, parent, lname, fid, NULL);
1790         if (rc) {
1791                 CERROR("lookup %s: rc = %d\n", lname->ln_name, rc);
1792                 RETURN(ERR_PTR(rc));
1793         }
1794
1795         /* In the function below, .hs_keycmp resolves to
1796          * lu_obj_hop_keycmp() */
1797         /* coverity[overrun-buffer-val] */
1798         child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1799
1800         RETURN(child);
1801 }
1802
1803 static int echo_setattr_object(const struct lu_env *env,
1804                                struct echo_device *ed,
1805                                struct lu_object *ec_parent,
1806                                __u64 id, int count)
1807 {
1808         struct lu_object        *parent;
1809         struct lu_object        *new_parent;
1810         struct echo_thread_info *info = echo_env_info(env);
1811         struct lu_name          *lname = &info->eti_lname;
1812         char                    *name = info->eti_name;
1813         struct lu_device        *ld = ed->ed_next;
1814         struct lu_buf           *buf = &info->eti_buf;
1815         int                      rc = 0;
1816         int                      i;
1817
1818         ENTRY;
1819
1820         if (ec_parent == NULL)
1821                 return -1;
1822         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1823         if (parent == NULL)
1824                 RETURN(-ENXIO);
1825
1826         rc = echo_md_dir_stripe_choose(env, ed, parent, NULL, 0, id,
1827                                        &new_parent);
1828         if (rc != 0)
1829                 RETURN(rc);
1830
1831         for (i = 0; i < count; i++) {
1832                 struct lu_object *ec_child, *child;
1833
1834                 echo_md_build_name(lname, name, id);
1835
1836                 ec_child = echo_md_lookup(env, ed, lu2md(new_parent), lname);
1837                 if (IS_ERR(ec_child)) {
1838                         rc = PTR_ERR(ec_child);
1839                         CERROR("Can't find child %s: rc = %d\n",
1840                                 lname->ln_name, rc);
1841                         break;
1842                 }
1843
1844                 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1845                 if (child == NULL) {
1846                         CERROR("Can not locate the child %s\n", lname->ln_name);
1847                         lu_object_put(env, ec_child);
1848                         rc = -EINVAL;
1849                         break;
1850                 }
1851
1852                 CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n",
1853                        PFID(lu_object_fid(child)));
1854
1855                 buf->lb_buf = info->eti_xattr_buf;
1856                 buf->lb_len = sizeof(info->eti_xattr_buf);
1857
1858                 sprintf(name, "%s.test1", XATTR_USER_PREFIX);
1859                 rc = mo_xattr_set(env, lu2md(child), buf, name,
1860                                   LU_XATTR_CREATE);
1861                 if (rc < 0) {
1862                         CERROR("Can not setattr child "DFID": rc = %d\n",
1863                                 PFID(lu_object_fid(child)), rc);
1864                         lu_object_put(env, ec_child);
1865                         break;
1866                 }
1867                 CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n",
1868                        PFID(lu_object_fid(child)));
1869                 id++;
1870                 lu_object_put(env, ec_child);
1871         }
1872
1873         if (new_parent != parent)
1874                 lu_object_put(env, new_parent);
1875
1876         RETURN(rc);
1877 }
1878
1879 static int echo_getattr_object(const struct lu_env *env,
1880                                struct echo_device *ed,
1881                                struct lu_object *ec_parent,
1882                                __u64 id, int count)
1883 {
1884         struct lu_object        *parent;
1885         struct lu_object        *new_parent;
1886         struct echo_thread_info *info = echo_env_info(env);
1887         struct lu_name          *lname = &info->eti_lname;
1888         char                    *name = info->eti_name;
1889         struct md_attr          *ma = &info->eti_ma;
1890         struct lu_device        *ld = ed->ed_next;
1891         int                      rc = 0;
1892         int                      i;
1893
1894         ENTRY;
1895
1896         if (ec_parent == NULL)
1897                 return -1;
1898         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1899         if (parent == NULL)
1900                 RETURN(-ENXIO);
1901
1902         rc = echo_md_dir_stripe_choose(env, ed, parent, NULL, 0, id,
1903                                        &new_parent);
1904         if (rc != 0)
1905                 RETURN(rc);
1906
1907         memset(ma, 0, sizeof(*ma));
1908         ma->ma_need |= MA_INODE | MA_LOV | MA_PFID | MA_HSM | MA_ACL_DEF;
1909         ma->ma_acl = info->eti_xattr_buf;
1910         ma->ma_acl_size = sizeof(info->eti_xattr_buf);
1911
1912         for (i = 0; i < count; i++) {
1913                 struct lu_object *ec_child, *child;
1914
1915                 ma->ma_valid = 0;
1916                 echo_md_build_name(lname, name, id);
1917                 echo_set_lmm_size(env, ld, ma);
1918
1919                 ec_child = echo_md_lookup(env, ed, lu2md(new_parent), lname);
1920                 if (IS_ERR(ec_child)) {
1921                         CERROR("Can't find child %s: rc = %ld\n",
1922                                lname->ln_name, PTR_ERR(ec_child));
1923                         RETURN(PTR_ERR(ec_child));
1924                 }
1925
1926                 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1927                 if (child == NULL) {
1928                         CERROR("Can not locate the child %s\n", lname->ln_name);
1929                         lu_object_put(env, ec_child);
1930                         RETURN(-EINVAL);
1931                 }
1932
1933                 CDEBUG(D_RPCTRACE, "Start getattr object "DFID"\n",
1934                        PFID(lu_object_fid(child)));
1935                 rc = echo_attr_get_complex(env, lu2md(child), ma);
1936                 if (rc) {
1937                         CERROR("Can not getattr child "DFID": rc = %d\n",
1938                                 PFID(lu_object_fid(child)), rc);
1939                         lu_object_put(env, ec_child);
1940                         break;
1941                 }
1942                 CDEBUG(D_RPCTRACE, "End getattr object "DFID"\n",
1943                        PFID(lu_object_fid(child)));
1944                 id++;
1945                 lu_object_put(env, ec_child);
1946         }
1947
1948         if (new_parent != parent)
1949                 lu_object_put(env, new_parent);
1950
1951         RETURN(rc);
1952 }
1953
1954 static int echo_lookup_object(const struct lu_env *env,
1955                               struct echo_device *ed,
1956                               struct lu_object *ec_parent,
1957                               __u64 id, int count)
1958 {
1959         struct lu_object        *parent;
1960         struct lu_object        *new_parent;
1961         struct echo_thread_info *info = echo_env_info(env);
1962         struct lu_name          *lname = &info->eti_lname;
1963         char                    *name = info->eti_name;
1964         struct lu_fid           *fid = &info->eti_fid;
1965         struct lu_device        *ld = ed->ed_next;
1966         int                      rc = 0;
1967         int                      i;
1968
1969         if (ec_parent == NULL)
1970                 return -1;
1971         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1972         if (parent == NULL)
1973                 return -ENXIO;
1974
1975         rc = echo_md_dir_stripe_choose(env, ed, parent, NULL, 0, id,
1976                                        &new_parent);
1977         if (rc != 0)
1978                 RETURN(rc);
1979
1980         /*prepare the requests*/
1981         for (i = 0; i < count; i++) {
1982                 echo_md_build_name(lname, name, id);
1983
1984                 CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n",
1985                        PFID(lu_object_fid(new_parent)), lname->ln_name,
1986                        new_parent);
1987
1988                 rc = mdo_lookup(env, lu2md(new_parent), lname, fid, NULL);
1989                 if (rc) {
1990                         CERROR("Can not lookup child %s: rc = %d\n", name, rc);
1991                         break;
1992                 }
1993
1994                 CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n",
1995                        PFID(lu_object_fid(new_parent)), lname->ln_name,
1996                        new_parent);
1997
1998                 id++;
1999         }
2000
2001         if (new_parent != parent)
2002                 lu_object_put(env, new_parent);
2003
2004         return rc;
2005 }
2006
2007 static int echo_md_destroy_internal(const struct lu_env *env,
2008                                     struct echo_device *ed,
2009                                     struct md_object *parent,
2010                                     struct lu_name *lname,
2011                                     struct md_attr *ma)
2012 {
2013         struct lu_device   *ld = ed->ed_next;
2014         struct lu_object   *ec_child;
2015         struct lu_object   *child;
2016         int                 rc;
2017
2018         ENTRY;
2019
2020         ec_child = echo_md_lookup(env, ed, parent, lname);
2021         if (IS_ERR(ec_child)) {
2022                 CERROR("Can't find child %s: rc = %ld\n", lname->ln_name,
2023                         PTR_ERR(ec_child));
2024                 RETURN(PTR_ERR(ec_child));
2025         }
2026
2027         child = lu_object_locate(ec_child->lo_header, ld->ld_type);
2028         if (child == NULL) {
2029                 CERROR("Can not locate the child %s\n", lname->ln_name);
2030                 GOTO(out_put, rc = -EINVAL);
2031         }
2032
2033         if (lu_object_remote(child)) {
2034                 CERROR("Can not destroy remote object %s: rc = %d\n",
2035                        lname->ln_name, -EPERM);
2036                 GOTO(out_put, rc = -EPERM);
2037         }
2038         CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n",
2039                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
2040
2041         rc = mdo_unlink(env, parent, lu2md(child), lname, ma, 0);
2042         if (rc) {
2043                 CERROR("Can not unlink child %s: rc = %d\n",
2044                         lname->ln_name, rc);
2045                 GOTO(out_put, rc);
2046         }
2047         CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n",
2048                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
2049 out_put:
2050         lu_object_put(env, ec_child);
2051         return rc;
2052 }
2053
2054 static int echo_destroy_object(const struct lu_env *env,
2055                                struct echo_device *ed,
2056                                struct lu_object *ec_parent,
2057                                char *name, int namelen,
2058                                __u64 id, __u32 mode,
2059                                int count)
2060 {
2061         struct echo_thread_info *info = echo_env_info(env);
2062         struct lu_name          *lname = &info->eti_lname;
2063         struct md_attr          *ma = &info->eti_ma;
2064         struct lu_device        *ld = ed->ed_next;
2065         struct lu_object        *parent;
2066         struct lu_object        *new_parent;
2067         int                      rc = 0;
2068         int                      i;
2069         ENTRY;
2070
2071         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
2072         if (parent == NULL)
2073                 RETURN(-EINVAL);
2074
2075         rc = echo_md_dir_stripe_choose(env, ed, parent, name, namelen,
2076                                        id, &new_parent);
2077         if (rc != 0)
2078                 RETURN(rc);
2079
2080         memset(ma, 0, sizeof(*ma));
2081         ma->ma_attr.la_mode = mode;
2082         ma->ma_attr.la_valid = LA_CTIME;
2083         ma->ma_attr.la_ctime = ktime_get_real_seconds();
2084         ma->ma_need = MA_INODE;
2085         ma->ma_valid = 0;
2086
2087         if (name != NULL) {
2088                 lname->ln_name = name;
2089                 lname->ln_namelen = namelen;
2090                 rc = echo_md_destroy_internal(env, ed, lu2md(new_parent), lname,
2091                                               ma);
2092                 GOTO(out_put, rc);
2093         }
2094
2095         /*prepare the requests*/
2096         for (i = 0; i < count; i++) {
2097                 char *tmp_name = info->eti_name;
2098
2099                 ma->ma_valid = 0;
2100                 echo_md_build_name(lname, tmp_name, id);
2101
2102                 rc = echo_md_destroy_internal(env, ed, lu2md(new_parent), lname,
2103                                               ma);
2104                 if (rc) {
2105                         CERROR("Can not unlink child %s: rc = %d\n", name, rc);
2106                         break;
2107                 }
2108                 id++;
2109         }
2110
2111 out_put:
2112         if (new_parent != parent)
2113                 lu_object_put(env, new_parent);
2114
2115         RETURN(rc);
2116 }
2117
2118 static struct lu_object *echo_resolve_path(const struct lu_env *env,
2119                                            struct echo_device *ed, char *path,
2120                                            int path_len)
2121 {
2122         struct lu_device        *ld = ed->ed_next;
2123         struct echo_thread_info *info = echo_env_info(env);
2124         struct lu_fid           *fid = &info->eti_fid;
2125         struct lu_name          *lname = &info->eti_lname;
2126         struct lu_object        *parent = NULL;
2127         struct lu_object        *child = NULL;
2128         int                      rc = 0;
2129         ENTRY;
2130
2131         *fid = ed->ed_root_fid;
2132
2133         /* In the function below, .hs_keycmp resolves to
2134          * lu_obj_hop_keycmp() */
2135         /* coverity[overrun-buffer-val] */
2136         parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
2137         if (IS_ERR(parent)) {
2138                 CERROR("Can not find the parent "DFID": rc = %ld\n",
2139                         PFID(fid), PTR_ERR(parent));
2140                 RETURN(parent);
2141         }
2142
2143         while (1) {
2144                 struct lu_object *ld_parent;
2145                 char *e;
2146
2147                 e = strsep(&path, "/");
2148                 if (e == NULL)
2149                         break;
2150
2151                 if (e[0] == 0) {
2152                         if (!path || path[0] == '\0')
2153                                 break;
2154                         continue;
2155                 }
2156
2157                 lname->ln_name = e;
2158                 lname->ln_namelen = strlen(e);
2159
2160                 ld_parent = lu_object_locate(parent->lo_header, ld->ld_type);
2161                 if (ld_parent == NULL) {
2162                         lu_object_put(env, parent);
2163                         rc = -EINVAL;
2164                         break;
2165                 }
2166
2167                 child = echo_md_lookup(env, ed, lu2md(ld_parent), lname);
2168                 lu_object_put(env, parent);
2169                 if (IS_ERR(child)) {
2170                         rc = (int)PTR_ERR(child);
2171                         CERROR("lookup %s under parent "DFID": rc = %d\n",
2172                                 lname->ln_name, PFID(lu_object_fid(ld_parent)),
2173                                 rc);
2174                         break;
2175                 }
2176                 parent = child;
2177         }
2178         if (rc)
2179                 RETURN(ERR_PTR(rc));
2180
2181         RETURN(parent);
2182 }
2183
2184 static void echo_ucred_init(struct lu_env *env)
2185 {
2186         struct lu_ucred *ucred = lu_ucred(env);
2187
2188         ucred->uc_valid = UCRED_INVALID;
2189
2190         ucred->uc_suppgids[0] = -1;
2191         ucred->uc_suppgids[1] = -1;
2192
2193         ucred->uc_uid = ucred->uc_o_uid  =
2194                                 from_kuid(&init_user_ns, current_uid());
2195         ucred->uc_gid = ucred->uc_o_gid  =
2196                                 from_kgid(&init_user_ns, current_gid());
2197         ucred->uc_fsuid = ucred->uc_o_fsuid =
2198                                 from_kuid(&init_user_ns, current_fsuid());
2199         ucred->uc_fsgid = ucred->uc_o_fsgid =
2200                                 from_kgid(&init_user_ns, current_fsgid());
2201         ucred->uc_cap = cfs_curproc_cap_pack();
2202
2203         /* remove fs privilege for non-root user. */
2204         if (ucred->uc_fsuid)
2205                 ucred->uc_cap &= ~CFS_CAP_FS_MASK;
2206         ucred->uc_valid = UCRED_NEW;
2207 }
2208
2209 static void echo_ucred_fini(struct lu_env *env)
2210 {
2211         struct lu_ucred *ucred = lu_ucred(env);
2212         ucred->uc_valid = UCRED_INIT;
2213 }
2214
2215 #define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_MD_THREAD)
2216 #define ECHO_MD_SES_TAG (LCT_REMEMBER | LCT_SESSION | LCT_SERVER_SESSION)
2217 static int echo_md_handler(struct echo_device *ed, int command,
2218                            char *path, int path_len, __u64 id, int count,
2219                            struct obd_ioctl_data *data)
2220 {
2221         struct echo_thread_info *info;
2222         struct lu_device      *ld = ed->ed_next;
2223         struct lu_env         *env;
2224         __u16                  refcheck;
2225         struct lu_object      *parent;
2226         char                  *name = NULL;
2227         int                    namelen = data->ioc_plen2;
2228         int                    rc = 0;
2229         ENTRY;
2230
2231         if (ld == NULL) {
2232                 CERROR("MD echo client is not being initialized properly\n");
2233                 RETURN(-EINVAL);
2234         }
2235
2236         if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
2237                 CERROR("Only support MDD layer right now!\n");
2238                 RETURN(-EINVAL);
2239         }
2240
2241         env = cl_env_get(&refcheck);
2242         if (IS_ERR(env))
2243                 RETURN(PTR_ERR(env));
2244
2245         rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG);
2246         if (rc != 0)
2247                 GOTO(out_env, rc);
2248
2249         /* init big_lmm buffer */
2250         info = echo_env_info(env);
2251         LASSERT(info->eti_big_lmm == NULL);
2252         OBD_ALLOC_LARGE(info->eti_big_lmm, MIN_MD_SIZE);
2253         if (info->eti_big_lmm == NULL)
2254                 GOTO(out_env, rc = -ENOMEM);
2255         info->eti_big_lmmsize = MIN_MD_SIZE;
2256
2257         parent = echo_resolve_path(env, ed, path, path_len);
2258         if (IS_ERR(parent)) {
2259                 CERROR("Can not resolve the path %s: rc = %ld\n", path,
2260                         PTR_ERR(parent));
2261                 GOTO(out_free, rc = PTR_ERR(parent));
2262         }
2263
2264         if (namelen > 0) {
2265                 OBD_ALLOC(name, namelen + 1);
2266                 if (name == NULL)
2267                         GOTO(out_put, rc = -ENOMEM);
2268                 if (copy_from_user(name, data->ioc_pbuf2, namelen))
2269                         GOTO(out_name, rc = -EFAULT);
2270         }
2271
2272         echo_ucred_init(env);
2273
2274         switch (command) {
2275         case ECHO_MD_CREATE:
2276         case ECHO_MD_MKDIR: {
2277                 struct echo_thread_info *info = echo_env_info(env);
2278                 __u32 mode = data->ioc_obdo2.o_mode;
2279                 struct lu_fid *fid = &info->eti_fid;
2280                 int stripe_count = (int)data->ioc_obdo2.o_misc;
2281                 int stripe_index = (int)data->ioc_obdo2.o_stripe_idx;
2282
2283                 rc = ostid_to_fid(fid, &data->ioc_obdo1.o_oi, 0);
2284                 if (rc != 0)
2285                         break;
2286
2287                 /* In the function below, .hs_keycmp resolves to
2288                  * lu_obj_hop_keycmp() */
2289                 /* coverity[overrun-buffer-val] */
2290                 rc = echo_create_md_object(env, ed, parent, fid, name, namelen,
2291                                            id, mode, count, stripe_count,
2292                                            stripe_index);
2293                 break;
2294         }
2295         case ECHO_MD_DESTROY:
2296         case ECHO_MD_RMDIR: {
2297                 __u32 mode = data->ioc_obdo2.o_mode;
2298
2299                 rc = echo_destroy_object(env, ed, parent, name, namelen,
2300                                          id, mode, count);
2301                 break;
2302         }
2303         case ECHO_MD_LOOKUP:
2304                 rc = echo_lookup_object(env, ed, parent, id, count);
2305                 break;
2306         case ECHO_MD_GETATTR:
2307                 rc = echo_getattr_object(env, ed, parent, id, count);
2308                 break;
2309         case ECHO_MD_SETATTR:
2310                 rc = echo_setattr_object(env, ed, parent, id, count);
2311                 break;
2312         default:
2313                 CERROR("unknown command %d\n", command);
2314                 rc = -EINVAL;
2315                 break;
2316         }
2317         echo_ucred_fini(env);
2318
2319 out_name:
2320         if (name != NULL)
2321                 OBD_FREE(name, namelen + 1);
2322 out_put:
2323         lu_object_put(env, parent);
2324 out_free:
2325         LASSERT(info->eti_big_lmm);
2326         OBD_FREE_LARGE(info->eti_big_lmm, info->eti_big_lmmsize);
2327         info->eti_big_lmm = NULL;
2328         info->eti_big_lmmsize = 0;
2329 out_env:
2330         cl_env_put(env, &refcheck);
2331         return rc;
2332 }
2333 #endif /* HAVE_SERVER_SUPPORT */
2334
2335 static int echo_create_object(const struct lu_env *env, struct echo_device *ed,
2336                               struct obdo *oa)
2337 {
2338         struct echo_object      *eco;
2339         struct echo_client_obd  *ec = ed->ed_ec;
2340         int created = 0;
2341         int rc;
2342         ENTRY;
2343
2344         if (!(oa->o_valid & OBD_MD_FLID) ||
2345             !(oa->o_valid & OBD_MD_FLGROUP) ||
2346             !fid_seq_is_echo(ostid_seq(&oa->o_oi))) {
2347                 CERROR("invalid oid "DOSTID"\n", POSTID(&oa->o_oi));
2348                 RETURN(-EINVAL);
2349         }
2350
2351         if (ostid_id(&oa->o_oi) == 0) {
2352                 rc = ostid_set_id(&oa->o_oi, ++last_object_id);
2353                 if (rc)
2354                         GOTO(failed, rc);
2355         }
2356
2357         rc = obd_create(env, ec->ec_exp, oa);
2358         if (rc != 0) {
2359                 CERROR("Cannot create objects: rc = %d\n", rc);
2360                 GOTO(failed, rc);
2361         }
2362
2363         created = 1;
2364
2365         oa->o_valid |= OBD_MD_FLID;
2366
2367         eco = cl_echo_object_find(ed, &oa->o_oi);
2368         if (IS_ERR(eco))
2369                 GOTO(failed, rc = PTR_ERR(eco));
2370         cl_echo_object_put(eco);
2371
2372         CDEBUG(D_INFO, "oa oid "DOSTID"\n", POSTID(&oa->o_oi));
2373         EXIT;
2374
2375 failed:
2376         if (created && rc != 0)
2377                 obd_destroy(env, ec->ec_exp, oa);
2378
2379         if (rc != 0)
2380                 CERROR("create object failed with: rc = %d\n", rc);
2381
2382         return rc;
2383 }
2384
2385 static int echo_get_object(struct echo_object **ecop, struct echo_device *ed,
2386                            struct obdo *oa)
2387 {
2388         struct echo_object *eco;
2389         int rc;
2390         ENTRY;
2391
2392         if (!(oa->o_valid & OBD_MD_FLID) ||
2393             !(oa->o_valid & OBD_MD_FLGROUP) ||
2394             ostid_id(&oa->o_oi) == 0) {
2395                 CERROR("invalid oid "DOSTID"\n", POSTID(&oa->o_oi));
2396                 RETURN(-EINVAL);
2397         }
2398
2399         rc = 0;
2400         eco = cl_echo_object_find(ed, &oa->o_oi);
2401         if (!IS_ERR(eco))
2402                 *ecop = eco;
2403         else
2404                 rc = PTR_ERR(eco);
2405
2406         RETURN(rc);
2407 }
2408
2409 static void echo_put_object(struct echo_object *eco)
2410 {
2411         int rc;
2412
2413         rc = cl_echo_object_put(eco);
2414         if (rc)
2415                 CERROR("%s: echo client drop an object failed: rc = %d\n",
2416                        eco->eo_dev->ed_ec->ec_exp->exp_obd->obd_name, rc);
2417 }
2418
2419 static void echo_client_page_debug_setup(struct page *page, int rw, u64 id,
2420                                          u64 offset, u64 count)
2421 {
2422         char    *addr;
2423         u64      stripe_off;
2424         u64      stripe_id;
2425         int      delta;
2426
2427         /* no partial pages on the client */
2428         LASSERT(count == PAGE_SIZE);
2429
2430         addr = kmap(page);
2431
2432         for (delta = 0; delta < PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2433                 if (rw == OBD_BRW_WRITE) {
2434                         stripe_off = offset + delta;
2435                         stripe_id = id;
2436                 } else {
2437                         stripe_off = 0xdeadbeef00c0ffeeULL;
2438                         stripe_id = 0xdeadbeef00c0ffeeULL;
2439                 }
2440                 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
2441                                   stripe_off, stripe_id);
2442         }
2443
2444         kunmap(page);
2445 }
2446
2447 static int
2448 echo_client_page_debug_check(struct page *page, u64 id, u64 offset, u64 count)
2449 {
2450         u64      stripe_off;
2451         u64      stripe_id;
2452         char   *addr;
2453         int     delta;
2454         int     rc;
2455         int     rc2;
2456
2457         /* no partial pages on the client */
2458         LASSERT(count == PAGE_SIZE);
2459
2460         addr = kmap(page);
2461
2462         for (rc = delta = 0; delta < PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2463                 stripe_off = offset + delta;
2464                 stripe_id = id;
2465
2466                 rc2 = block_debug_check("test_brw",
2467                                         addr + delta, OBD_ECHO_BLOCK_SIZE,
2468                                         stripe_off, stripe_id);
2469                 if (rc2 != 0) {
2470                         CERROR("Error in echo object %#llx\n", id);
2471                         rc = rc2;
2472                 }
2473         }
2474
2475         kunmap(page);
2476         return rc;
2477 }
2478
2479 static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa,
2480                             struct echo_object *eco, u64 offset,
2481                             u64 count, int async)
2482 {
2483         size_t                  npages;
2484         struct brw_page        *pga;
2485         struct brw_page        *pgp;
2486         struct page            **pages;
2487         u64                      off;
2488         size_t                  i;
2489         int                     rc;
2490         int                     verify;
2491         gfp_t                   gfp_mask;
2492         u32                     brw_flags = 0;
2493         ENTRY;
2494
2495         verify = (ostid_id(&oa->o_oi) != ECHO_PERSISTENT_OBJID &&
2496                   (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
2497                   (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
2498
2499         gfp_mask = ((ostid_id(&oa->o_oi) & 2) == 0) ? GFP_KERNEL : GFP_HIGHUSER;
2500
2501         LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
2502
2503         if ((count & (~PAGE_MASK)) != 0)
2504                 RETURN(-EINVAL);
2505
2506         /* XXX think again with misaligned I/O */
2507         npages = count >> PAGE_SHIFT;
2508
2509         if (rw == OBD_BRW_WRITE)
2510                 brw_flags = OBD_BRW_ASYNC;
2511
2512         OBD_ALLOC(pga, npages * sizeof(*pga));
2513         if (pga == NULL)
2514                 RETURN(-ENOMEM);
2515
2516         OBD_ALLOC(pages, npages * sizeof(*pages));
2517         if (pages == NULL) {
2518                 OBD_FREE(pga, npages * sizeof(*pga));
2519                 RETURN(-ENOMEM);
2520         }
2521
2522         for (i = 0, pgp = pga, off = offset;
2523              i < npages;
2524              i++, pgp++, off += PAGE_SIZE) {
2525
2526                 LASSERT(pgp->pg == NULL);       /* for cleanup */
2527
2528                 rc = -ENOMEM;
2529                 pgp->pg = alloc_page(gfp_mask);
2530                 if (pgp->pg == NULL)
2531                         goto out;
2532
2533                 pages[i] = pgp->pg;
2534                 pgp->count = PAGE_SIZE;
2535                 pgp->off = off;
2536                 pgp->flag = brw_flags;
2537
2538                 if (verify)
2539                         echo_client_page_debug_setup(pgp->pg, rw,
2540                                                      ostid_id(&oa->o_oi), off,
2541                                                      pgp->count);
2542         }
2543
2544         /* brw mode can only be used at client */
2545         LASSERT(ed->ed_next != NULL);
2546         rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async);
2547
2548  out:
2549         if (rc != 0 || rw != OBD_BRW_READ)
2550                 verify = 0;
2551
2552         for (i = 0, pgp = pga; i < npages; i++, pgp++) {
2553                 if (pgp->pg == NULL)
2554                         continue;
2555
2556                 if (verify) {
2557                         int vrc;
2558                         vrc = echo_client_page_debug_check(pgp->pg,
2559                                                            ostid_id(&oa->o_oi),
2560                                                            pgp->off, pgp->count);
2561                         if (vrc != 0 && rc == 0)
2562                                 rc = vrc;
2563                 }
2564                 __free_page(pgp->pg);
2565         }
2566         OBD_FREE(pga, npages * sizeof(*pga));
2567         OBD_FREE(pages, npages * sizeof(*pages));
2568         RETURN(rc);
2569 }
2570
2571 static int echo_client_prep_commit(const struct lu_env *env,
2572                                    struct obd_export *exp, int rw,
2573                                    struct obdo *oa, struct echo_object *eco,
2574                                    u64 offset, u64 count,
2575                                    u64 batch, int async)
2576 {
2577         struct obd_ioobj ioo;
2578         struct niobuf_local *lnb;
2579         struct niobuf_remote rnb;
2580         u64 off;
2581         u64 npages, tot_pages, apc;
2582         int i, ret = 0, brw_flags = 0;
2583
2584         ENTRY;
2585
2586         if (count <= 0 || (count & ~PAGE_MASK) != 0)
2587                 RETURN(-EINVAL);
2588
2589         apc = npages = batch >> PAGE_SHIFT;
2590         tot_pages = count >> PAGE_SHIFT;
2591
2592         OBD_ALLOC_LARGE(lnb, apc * sizeof(struct niobuf_local));
2593         if (lnb == NULL)
2594                 RETURN(-ENOMEM);
2595
2596         if (rw == OBD_BRW_WRITE && async)
2597                 brw_flags |= OBD_BRW_ASYNC;
2598
2599         obdo_to_ioobj(oa, &ioo);
2600
2601         off = offset;
2602
2603         for (; tot_pages > 0; tot_pages -= npages) {
2604                 int lpages;
2605
2606                 if (tot_pages < npages)
2607                         npages = tot_pages;
2608
2609                 rnb.rnb_offset = off;
2610                 rnb.rnb_len = npages * PAGE_SIZE;
2611                 rnb.rnb_flags = brw_flags;
2612                 ioo.ioo_bufcnt = 1;
2613                 off += npages * PAGE_SIZE;
2614
2615                 lpages = npages;
2616                 ret = obd_preprw(env, rw, exp, oa, 1, &ioo, &rnb, &lpages, lnb);
2617                 if (ret != 0)
2618                         GOTO(out, ret);
2619
2620                 for (i = 0; i < lpages; i++) {
2621                         struct page *page = lnb[i].lnb_page;
2622
2623                         /* read past eof? */
2624                         if (page == NULL && lnb[i].lnb_rc == 0)
2625                                 continue;
2626
2627                         if (async)
2628                                 lnb[i].lnb_flags |= OBD_BRW_ASYNC;
2629
2630                         if (ostid_id(&oa->o_oi) == ECHO_PERSISTENT_OBJID ||
2631                             (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
2632                             (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
2633                                 continue;
2634
2635                         if (rw == OBD_BRW_WRITE)
2636                                 echo_client_page_debug_setup(page, rw,
2637                                                         ostid_id(&oa->o_oi),
2638                                                         lnb[i].lnb_file_offset,
2639                                                         lnb[i].lnb_len);
2640                         else
2641                                 echo_client_page_debug_check(page,
2642                                                         ostid_id(&oa->o_oi),
2643                                                         lnb[i].lnb_file_offset,
2644                                                         lnb[i].lnb_len);
2645                 }
2646
2647                 ret = obd_commitrw(env, rw, exp, oa, 1, &ioo, &rnb, npages, lnb,
2648                                    ret);
2649                 if (ret != 0)
2650                         break;
2651
2652                 /* Reuse env context. */
2653                 lu_context_exit((struct lu_context *)&env->le_ctx);
2654                 lu_context_enter((struct lu_context *)&env->le_ctx);
2655         }
2656
2657 out:
2658         OBD_FREE_LARGE(lnb, apc * sizeof(struct niobuf_local));
2659
2660         RETURN(ret);
2661 }
2662
2663 static int echo_client_brw_ioctl(const struct lu_env *env, int rw,
2664                                  struct obd_export *exp,
2665                                  struct obd_ioctl_data *data)
2666 {
2667         struct obd_device *obd = class_exp2obd(exp);
2668         struct echo_device *ed = obd2echo_dev(obd);
2669         struct echo_client_obd *ec = ed->ed_ec;
2670         struct obdo *oa = &data->ioc_obdo1;
2671         struct echo_object *eco;
2672         int rc;
2673         int async = 0;
2674         long test_mode;
2675         ENTRY;
2676
2677         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2678
2679         rc = echo_get_object(&eco, ed, oa);
2680         if (rc)
2681                 RETURN(rc);
2682
2683         oa->o_valid &= ~OBD_MD_FLHANDLE;
2684
2685         /* OFD/obdfilter works only via prep/commit */
2686         test_mode = (long)data->ioc_pbuf1;
2687         if (ed->ed_next == NULL && test_mode != 3) {
2688                 test_mode = 3;
2689                 data->ioc_plen1 = data->ioc_count;
2690         }
2691
2692         if (test_mode == 3)
2693                 async = 1;
2694
2695         /* Truncate batch size to maximum */
2696         if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE)
2697                 data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE;
2698
2699         switch (test_mode) {
2700         case 1:
2701                 /* fall through */
2702         case 2:
2703                 rc = echo_client_kbrw(ed, rw, oa, eco, data->ioc_offset,
2704                                       data->ioc_count, async);
2705                 break;
2706         case 3:
2707                 rc = echo_client_prep_commit(env, ec->ec_exp, rw, oa, eco,
2708                                              data->ioc_offset, data->ioc_count,
2709                                              data->ioc_plen1, async);
2710                 break;
2711         default:
2712                 rc = -EINVAL;
2713         }
2714
2715         echo_put_object(eco);
2716
2717         RETURN(rc);
2718 }
2719
2720 static int
2721 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2722                       void *karg, void __user *uarg)
2723 {
2724 #ifdef HAVE_SERVER_SUPPORT
2725         struct tgt_session_info *tsi;
2726 #endif
2727         struct obd_device      *obd = exp->exp_obd;
2728         struct echo_device     *ed = obd2echo_dev(obd);
2729         struct echo_client_obd *ec = ed->ed_ec;
2730         struct echo_object     *eco;
2731         struct obd_ioctl_data  *data = karg;
2732         struct lu_env          *env;
2733         struct obdo            *oa;
2734         struct lu_fid           fid;
2735         int                     rw = OBD_BRW_READ;
2736         int                     rc = 0;
2737 #ifdef HAVE_SERVER_SUPPORT
2738         struct lu_context        echo_session;
2739 #endif
2740         ENTRY;
2741
2742         oa = &data->ioc_obdo1;
2743         if (!(oa->o_valid & OBD_MD_FLGROUP)) {
2744                 oa->o_valid |= OBD_MD_FLGROUP;
2745                 ostid_set_seq_echo(&oa->o_oi);
2746         }
2747
2748         /* This FID is unpacked just for validation at this point */
2749         rc = ostid_to_fid(&fid, &oa->o_oi, 0);
2750         if (rc < 0)
2751                 RETURN(rc);
2752
2753         OBD_ALLOC_PTR(env);
2754         if (env == NULL)
2755                 RETURN(-ENOMEM);
2756
2757         rc = lu_env_init(env, LCT_DT_THREAD);
2758         if (rc)
2759                 GOTO(out_alloc, rc = -ENOMEM);
2760
2761 #ifdef HAVE_SERVER_SUPPORT
2762         env->le_ses = &echo_session;
2763         rc = lu_context_init(env->le_ses, LCT_SERVER_SESSION | LCT_NOREF);
2764         if (unlikely(rc < 0))
2765                 GOTO(out_env, rc);
2766         lu_context_enter(env->le_ses);
2767
2768         tsi = tgt_ses_info(env);
2769         tsi->tsi_exp = ec->ec_exp;
2770         tsi->tsi_jobid = NULL;
2771 #endif
2772         switch (cmd) {
2773         case OBD_IOC_CREATE:                    /* may create echo object */
2774                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2775                         GOTO (out, rc = -EPERM);
2776
2777                 rc = echo_create_object(env, ed, oa);
2778                 GOTO(out, rc);
2779
2780 #ifdef HAVE_SERVER_SUPPORT
2781         case OBD_IOC_ECHO_MD: {
2782                 int count;
2783                 int cmd;
2784                 char *dir = NULL;
2785                 int dirlen;
2786                 __u64 id;
2787
2788                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2789                         GOTO(out, rc = -EPERM);
2790
2791                 count = data->ioc_count;
2792                 cmd = data->ioc_command;
2793
2794                 id = data->ioc_obdo2.o_oi.oi.oi_id;
2795                 dirlen = data->ioc_plen1;
2796                 OBD_ALLOC(dir, dirlen + 1);
2797                 if (dir == NULL)
2798                         GOTO(out, rc = -ENOMEM);
2799
2800                 if (copy_from_user(dir, data->ioc_pbuf1, dirlen)) {
2801                         OBD_FREE(dir, data->ioc_plen1 + 1);
2802                         GOTO(out, rc = -EFAULT);
2803                 }
2804
2805                 rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data);
2806                 OBD_FREE(dir, dirlen + 1);
2807                 GOTO(out, rc);
2808         }
2809         case OBD_IOC_ECHO_ALLOC_SEQ: {
2810                 struct lu_env   *cl_env;
2811                 __u16            refcheck;
2812                 __u64            seq;
2813                 int              max_count;
2814
2815                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2816                         GOTO(out, rc = -EPERM);
2817
2818                 cl_env = cl_env_get(&refcheck);
2819                 if (IS_ERR(cl_env))
2820                         GOTO(out, rc = PTR_ERR(cl_env));
2821
2822                 rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG,
2823                                             ECHO_MD_SES_TAG);
2824                 if (rc != 0) {
2825                         cl_env_put(cl_env, &refcheck);
2826                         GOTO(out, rc);
2827                 }
2828
2829                 rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq);
2830                 cl_env_put(cl_env, &refcheck);
2831                 if (rc < 0) {
2832                         CERROR("%s: Can not alloc seq: rc = %d\n",
2833                                obd->obd_name, rc);
2834                         GOTO(out, rc);
2835                 }
2836
2837                 if (copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1))
2838                         return -EFAULT;
2839
2840                 max_count = LUSTRE_METADATA_SEQ_MAX_WIDTH;
2841                 if (copy_to_user(data->ioc_pbuf2, &max_count,
2842                                      data->ioc_plen2))
2843                         return -EFAULT;
2844                 GOTO(out, rc);
2845         }
2846 #endif /* HAVE_SERVER_SUPPORT */
2847         case OBD_IOC_DESTROY:
2848                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2849                         GOTO (out, rc = -EPERM);
2850
2851                 rc = echo_get_object(&eco, ed, oa);
2852                 if (rc == 0) {
2853                         rc = obd_destroy(env, ec->ec_exp, oa);
2854                         if (rc == 0)
2855                                 eco->eo_deleted = 1;
2856                         echo_put_object(eco);
2857                 }
2858                 GOTO(out, rc);
2859
2860         case OBD_IOC_GETATTR:
2861                 rc = echo_get_object(&eco, ed, oa);
2862                 if (rc == 0) {
2863                         rc = obd_getattr(env, ec->ec_exp, oa);
2864                         echo_put_object(eco);
2865                 }
2866                 GOTO(out, rc);
2867
2868         case OBD_IOC_SETATTR:
2869                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2870                         GOTO (out, rc = -EPERM);
2871
2872                 rc = echo_get_object(&eco, ed, oa);
2873                 if (rc == 0) {
2874                         rc = obd_setattr(env, ec->ec_exp, oa);
2875                         echo_put_object(eco);
2876                 }
2877                 GOTO(out, rc);
2878
2879         case OBD_IOC_BRW_WRITE:
2880                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2881                         GOTO (out, rc = -EPERM);
2882
2883                 rw = OBD_BRW_WRITE;
2884                 /* fall through */
2885         case OBD_IOC_BRW_READ:
2886                 rc = echo_client_brw_ioctl(env, rw, exp, data);
2887                 GOTO(out, rc);
2888
2889         default:
2890                 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
2891                 GOTO (out, rc = -ENOTTY);
2892         }
2893
2894         EXIT;
2895 out:
2896 #ifdef HAVE_SERVER_SUPPORT
2897         lu_context_exit(env->le_ses);
2898         lu_context_fini(env->le_ses);
2899 out_env:
2900 #endif
2901         lu_env_fini(env);
2902 out_alloc:
2903         OBD_FREE_PTR(env);
2904
2905         return rc;
2906 }
2907
2908 static int echo_client_setup(const struct lu_env *env,
2909                              struct obd_device *obddev, struct lustre_cfg *lcfg)
2910 {
2911         struct echo_client_obd *ec = &obddev->u.echo_client;
2912         struct obd_device *tgt;
2913         struct obd_uuid echo_uuid = { "ECHO_UUID" };
2914         struct obd_connect_data *ocd = NULL;
2915         int rc;
2916         ENTRY;
2917
2918         if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2919                 CERROR("requires a TARGET OBD name\n");
2920                 RETURN(-EINVAL);
2921         }
2922
2923         tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
2924         if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
2925                 CERROR("device not attached or not set up (%s)\n",
2926                        lustre_cfg_string(lcfg, 1));
2927                 RETURN(-EINVAL);
2928         }
2929
2930         spin_lock_init(&ec->ec_lock);
2931         INIT_LIST_HEAD(&ec->ec_objects);
2932         INIT_LIST_HEAD(&ec->ec_locks);
2933         ec->ec_unique = 0;
2934
2935         if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
2936 #ifdef HAVE_SERVER_SUPPORT
2937                 lu_context_tags_update(ECHO_MD_CTX_TAG);
2938                 lu_session_tags_update(ECHO_MD_SES_TAG);
2939 #else
2940                 CERROR("Local operations are NOT supported on client side. "
2941                        "Only remote operations are supported. Metadata client "
2942                        "must be run on server side.\n");
2943 #endif
2944                 RETURN(0);
2945         }
2946
2947         OBD_ALLOC(ocd, sizeof(*ocd));
2948         if (ocd == NULL) {
2949                 CERROR("Can't alloc ocd connecting to %s\n",
2950                        lustre_cfg_string(lcfg, 1));
2951                 return -ENOMEM;
2952         }
2953
2954         ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL |
2955                                  OBD_CONNECT_BRW_SIZE |
2956                                  OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 |
2957                                  OBD_CONNECT_64BITHASH | OBD_CONNECT_LVB_TYPE |
2958                                  OBD_CONNECT_FID;
2959         ocd->ocd_brw_size = DT_MAX_BRW_SIZE;
2960         ocd->ocd_version = LUSTRE_VERSION_CODE;
2961         ocd->ocd_group = FID_SEQ_ECHO;
2962
2963         rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL);
2964         if (rc == 0) {
2965                 /* Turn off pinger because it connects to tgt obd directly. */
2966                 spin_lock(&tgt->obd_dev_lock);
2967                 list_del_init(&ec->ec_exp->exp_obd_chain_timed);
2968                 spin_unlock(&tgt->obd_dev_lock);
2969         }
2970
2971         OBD_FREE(ocd, sizeof(*ocd));
2972
2973         if (rc != 0) {
2974                 CERROR("fail to connect to device %s\n",
2975                        lustre_cfg_string(lcfg, 1));
2976                 return (rc);
2977         }
2978
2979         RETURN(rc);
2980 }
2981
2982 static int echo_client_cleanup(struct obd_device *obddev)
2983 {
2984         struct echo_device *ed = obd2echo_dev(obddev);
2985         struct echo_client_obd *ec = &obddev->u.echo_client;
2986         int rc;
2987         ENTRY;
2988
2989         /*Do nothing for Metadata echo client*/
2990         if (ed == NULL )
2991                 RETURN(0);
2992
2993         if (ed->ed_next_ismd) {
2994 #ifdef HAVE_SERVER_SUPPORT
2995                 lu_context_tags_clear(ECHO_MD_CTX_TAG);
2996                 lu_session_tags_clear(ECHO_MD_SES_TAG);
2997 #else
2998                 CERROR("This is client-side only module, does not support "
2999                         "metadata echo client.\n");
3000 #endif
3001                 RETURN(0);
3002         }
3003
3004         if (!list_empty(&obddev->obd_exports)) {
3005                 CERROR("still has clients!\n");
3006                 RETURN(-EBUSY);
3007         }
3008
3009         LASSERT(atomic_read(&ec->ec_exp->exp_refcount) > 0);
3010         rc = obd_disconnect(ec->ec_exp);
3011         if (rc != 0)
3012                 CERROR("fail to disconnect device: %d\n", rc);
3013
3014         RETURN(rc);
3015 }
3016
3017 static int echo_client_connect(const struct lu_env *env,
3018                                struct obd_export **exp,
3019                                struct obd_device *src, struct obd_uuid *cluuid,
3020                                struct obd_connect_data *data, void *localdata)
3021 {
3022         int                rc;
3023         struct lustre_handle conn = { 0 };
3024
3025         ENTRY;
3026         rc = class_connect(&conn, src, cluuid);
3027         if (rc == 0) {
3028                 *exp = class_conn2export(&conn);
3029         }
3030
3031         RETURN (rc);
3032 }
3033
3034 static int echo_client_disconnect(struct obd_export *exp)
3035 {
3036         int                     rc;
3037         ENTRY;
3038
3039         if (exp == NULL)
3040                 GOTO(out, rc = -EINVAL);
3041
3042         rc = class_disconnect(exp);
3043         GOTO(out, rc);
3044  out:
3045         return rc;
3046 }
3047
3048 static struct obd_ops echo_client_obd_ops = {
3049         .o_owner       = THIS_MODULE,
3050         .o_iocontrol   = echo_client_iocontrol,
3051         .o_connect     = echo_client_connect,
3052         .o_disconnect  = echo_client_disconnect
3053 };
3054
3055 static int __init obdecho_init(void)
3056 {
3057         int rc;
3058
3059         ENTRY;
3060         LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
3061
3062         LASSERT(PAGE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
3063
3064 # ifdef HAVE_SERVER_SUPPORT
3065         rc = echo_persistent_pages_init();
3066         if (rc != 0)
3067                 goto failed_0;
3068
3069         rc = class_register_type(&echo_obd_ops, NULL, true, NULL,
3070                                  LUSTRE_ECHO_NAME, &echo_srv_type);
3071         if (rc != 0)
3072                 goto failed_1;
3073 # endif
3074
3075         rc = lu_kmem_init(echo_caches);
3076         if (rc == 0) {
3077                 rc = class_register_type(&echo_client_obd_ops, NULL, false,
3078                                          NULL, LUSTRE_ECHO_CLIENT_NAME,
3079                                          &echo_device_type);
3080                 if (rc)
3081                         lu_kmem_fini(echo_caches);
3082         }
3083
3084 # ifdef HAVE_SERVER_SUPPORT
3085         if (rc == 0)
3086                 RETURN(0);
3087
3088         class_unregister_type(LUSTRE_ECHO_NAME);
3089 failed_1:
3090         echo_persistent_pages_fini();
3091 failed_0:
3092 # endif
3093         RETURN(rc);
3094 }
3095
3096 static void __exit obdecho_exit(void)
3097 {
3098         class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
3099         lu_kmem_fini(echo_caches);
3100
3101 #ifdef HAVE_SERVER_SUPPORT
3102         class_unregister_type(LUSTRE_ECHO_NAME);
3103         echo_persistent_pages_fini();
3104 #endif
3105 }
3106
3107 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3108 MODULE_DESCRIPTION("Lustre Echo Client test driver");
3109 MODULE_VERSION(LUSTRE_VERSION_STRING);
3110 MODULE_LICENSE("GPL");
3111
3112 module_init(obdecho_init);
3113 module_exit(obdecho_exit);
3114
3115 /** @} echo_client */