lustre/obdecho/echo_client.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_ECHO
  38 #ifdef __KERNEL__
  39 #include <libcfs/libcfs.h>
  40 #else
  41 #include <liblustre.h>
  42 #endif
  43
  44 #include <obd.h>
  45 #include <obd_support.h>
  46 #include <obd_class.h>
  47 #include <lustre_debug.h>
  48 #include <lprocfs_status.h>
  49 #include <cl_object.h>
  50 #include <lustre_fid.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_net.h>
  53
  54 #include "echo_internal.h"
  55
  56 /** \defgroup echo_client Echo Client
  57  * @{
  58  */
  59
  60 struct echo_device {
  61         struct cl_device        ed_cl;
  62         struct echo_client_obd *ed_ec;
  63
  64         struct cl_site          ed_site_myself;
  65         struct cl_site         *ed_site;
  66         struct lu_device       *ed_next;
  67         int                     ed_next_islov;
  68         int                     ed_next_ismd;
  69         struct lu_client_seq   *ed_cl_seq;
  70 };
  71
  72 struct echo_object {
  73         struct cl_object        eo_cl;
  74         struct cl_object_header eo_hdr;
  75
  76         struct echo_device     *eo_dev;
  77         cfs_list_t              eo_obj_chain;
  78         struct lov_stripe_md   *eo_lsm;
  79         cfs_atomic_t            eo_npages;
  80         int                     eo_deleted;
  81 };
  82
  83 struct echo_object_conf {
  84         struct cl_object_conf  eoc_cl;
  85         struct lov_stripe_md **eoc_md;
  86 };
  87
  88 struct echo_page {
  89         struct cl_page_slice   ep_cl;
  90         cfs_mutex_t            ep_lock;
  91         cfs_page_t            *ep_vmpage;
  92 };
  93
  94 struct echo_lock {
  95         struct cl_lock_slice   el_cl;
  96         cfs_list_t             el_chain;
  97         struct echo_object    *el_object;
  98         __u64                  el_cookie;
  99         cfs_atomic_t           el_refcount;
 100 };
 101
 102 struct echo_io {
 103         struct cl_io_slice     ei_cl;
 104 };
 105
 106 #if 0
 107 struct echo_req {
 108         struct cl_req_slice er_cl;
 109 };
 110 #endif
 111
 112 static int echo_client_setup(const struct lu_env *env,
 113                              struct obd_device *obddev,
 114                              struct lustre_cfg *lcfg);
 115 static int echo_client_cleanup(struct obd_device *obddev);
 116
 117
 118 /** \defgroup echo_helpers Helper functions
 119  * @{
 120  */
 121 static inline struct echo_device *cl2echo_dev(const struct cl_device *dev)
 122 {
 123         return container_of0(dev, struct echo_device, ed_cl);
 124 }
 125
 126 static inline struct cl_device *echo_dev2cl(struct echo_device *d)
 127 {
 128         return &d->ed_cl;
 129 }
 130
 131 static inline struct echo_device *obd2echo_dev(const struct obd_device *obd)
 132 {
 133         return cl2echo_dev(lu2cl_dev(obd->obd_lu_dev));
 134 }
 135
 136 static inline struct cl_object *echo_obj2cl(struct echo_object *eco)
 137 {
 138         return &eco->eo_cl;
 139 }
 140
 141 static inline struct echo_object *cl2echo_obj(const struct cl_object *o)
 142 {
 143         return container_of(o, struct echo_object, eo_cl);
 144 }
 145
 146 static inline struct echo_page *cl2echo_page(const struct cl_page_slice *s)
 147 {
 148         return container_of(s, struct echo_page, ep_cl);
 149 }
 150
 151 static inline struct echo_lock *cl2echo_lock(const struct cl_lock_slice *s)
 152 {
 153         return container_of(s, struct echo_lock, el_cl);
 154 }
 155
 156 static inline struct cl_lock *echo_lock2cl(const struct echo_lock *ecl)
 157 {
 158         return ecl->el_cl.cls_lock;
 159 }
 160
 161 static struct lu_context_key echo_thread_key;
 162 static inline struct echo_thread_info *echo_env_info(const struct lu_env *env)
 163 {
 164         struct echo_thread_info *info;
 165         info = lu_context_key_get(&env->le_ctx, &echo_thread_key);
 166         LASSERT(info != NULL);
 167         return info;
 168 }
 169
 170 static inline
 171 struct echo_object_conf *cl2echo_conf(const struct cl_object_conf *c)
 172 {
 173         return container_of(c, struct echo_object_conf, eoc_cl);
 174 }
 175
 176 static inline void lsm2fid(struct lov_stripe_md *lsm, struct lu_fid *fid)
 177 {
 178         fid_zero(fid);
 179         fid->f_seq = FID_SEQ_ECHO;
 180         /* truncated to 32 bits by assignment */
 181         fid->f_oid = lsm->lsm_object_id;
 182         fid->f_ver = lsm->lsm_object_id >> 32;
 183 }
 184 /** @} echo_helpers */
 185
 186 static struct echo_object *cl_echo_object_find(struct echo_device *d,
 187                                                struct lov_stripe_md **lsm);
 188 static int cl_echo_object_put(struct echo_object *eco);
 189 static int cl_echo_enqueue   (struct echo_object *eco, obd_off start,
 190                               obd_off end, int mode, __u64 *cookie);
 191 static int cl_echo_cancel    (struct echo_device *d, __u64 cookie);
 192 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
 193                               cfs_page_t **pages, int npages, int async);
 194
 195 static struct echo_thread_info *echo_env_info(const struct lu_env *env);
 196
 197 struct echo_thread_info {
 198         struct echo_object_conf eti_conf;
 199         struct lustre_md        eti_md;
 200
 201         struct cl_2queue        eti_queue;
 202         struct cl_io            eti_io;
 203         struct cl_lock_descr    eti_descr;
 204         struct lu_fid           eti_fid;
 205         struct lu_fid           eti_fid2;
 206         struct md_op_spec       eti_spec;
 207         struct lov_mds_md_v3    eti_lmm;
 208         struct lov_user_md_v3   eti_lum;
 209         struct md_attr          eti_ma;
 210         struct lu_name          eti_lname;
 211         char                    eti_name[20];
 212         struct lu_buf           eti_buf;
 213         char                    eti_xattr_buf[LUSTRE_POSIX_ACL_MAX_SIZE];
 214 };
 215
 216 /* No session used right now */
 217 struct echo_session_info {
 218         unsigned long dummy;
 219 };
 220
 221 static cfs_mem_cache_t *echo_page_kmem;
 222 static cfs_mem_cache_t *echo_lock_kmem;
 223 static cfs_mem_cache_t *echo_object_kmem;
 224 static cfs_mem_cache_t *echo_thread_kmem;
 225 static cfs_mem_cache_t *echo_session_kmem;
 226 //static cfs_mem_cache_t *echo_req_kmem;
 227
 228 static struct lu_kmem_descr echo_caches[] = {
 229         {
 230                 .ckd_cache = &echo_page_kmem,
 231                 .ckd_name  = "echo_page_kmem",
 232                 .ckd_size  = sizeof (struct echo_page)
 233         },
 234         {
 235                 .ckd_cache = &echo_lock_kmem,
 236                 .ckd_name  = "echo_lock_kmem",
 237                 .ckd_size  = sizeof (struct echo_lock)
 238         },
 239         {
 240                 .ckd_cache = &echo_object_kmem,
 241                 .ckd_name  = "echo_object_kmem",
 242                 .ckd_size  = sizeof (struct echo_object)
 243         },
 244         {
 245                 .ckd_cache = &echo_thread_kmem,
 246                 .ckd_name  = "echo_thread_kmem",
 247                 .ckd_size  = sizeof (struct echo_thread_info)
 248         },
 249         {
 250                 .ckd_cache = &echo_session_kmem,
 251                 .ckd_name  = "echo_session_kmem",
 252                 .ckd_size  = sizeof (struct echo_session_info)
 253         },
 254 #if 0
 255         {
 256                 .ckd_cache = &echo_req_kmem,
 257                 .ckd_name  = "echo_req_kmem",
 258                 .ckd_size  = sizeof (struct echo_req)
 259         },
 260 #endif
 261         {
 262                 .ckd_cache = NULL
 263         }
 264 };
 265
 266 /** \defgroup echo_page Page operations
 267  *
 268  * Echo page operations.
 269  *
 270  * @{
 271  */
 272 static cfs_page_t *echo_page_vmpage(const struct lu_env *env,
 273                                     const struct cl_page_slice *slice)
 274 {
 275         return cl2echo_page(slice)->ep_vmpage;
 276 }
 277
 278 static int echo_page_own(const struct lu_env *env,
 279                          const struct cl_page_slice *slice,
 280                          struct cl_io *io, int nonblock)
 281 {
 282         struct echo_page *ep = cl2echo_page(slice);
 283
 284         if (!nonblock)
 285                 cfs_mutex_lock(&ep->ep_lock);
 286         else if (!cfs_mutex_trylock(&ep->ep_lock))
 287                 return -EAGAIN;
 288         return 0;
 289 }
 290
 291 static void echo_page_disown(const struct lu_env *env,
 292                              const struct cl_page_slice *slice,
 293                              struct cl_io *io)
 294 {
 295         struct echo_page *ep = cl2echo_page(slice);
 296
 297         LASSERT(cfs_mutex_is_locked(&ep->ep_lock));
 298         cfs_mutex_unlock(&ep->ep_lock);
 299 }
 300
 301 static void echo_page_discard(const struct lu_env *env,
 302                               const struct cl_page_slice *slice,
 303                               struct cl_io *unused)
 304 {
 305         cl_page_delete(env, slice->cpl_page);
 306 }
 307
 308 static int echo_page_is_vmlocked(const struct lu_env *env,
 309                                  const struct cl_page_slice *slice)
 310 {
 311         if (cfs_mutex_is_locked(&cl2echo_page(slice)->ep_lock))
 312                 return -EBUSY;
 313         return -ENODATA;
 314 }
 315
 316 static void echo_page_completion(const struct lu_env *env,
 317                                  const struct cl_page_slice *slice,
 318                                  int ioret)
 319 {
 320         LASSERT(slice->cpl_page->cp_sync_io != NULL);
 321 }
 322
 323 static void echo_page_fini(const struct lu_env *env,
 324                            struct cl_page_slice *slice)
 325 {
 326         struct echo_page *ep    = cl2echo_page(slice);
 327         struct echo_object *eco = cl2echo_obj(slice->cpl_obj);
 328         cfs_page_t *vmpage      = ep->ep_vmpage;
 329         ENTRY;
 330
 331         cfs_atomic_dec(&eco->eo_npages);
 332         page_cache_release(vmpage);
 333         OBD_SLAB_FREE_PTR(ep, echo_page_kmem);
 334         EXIT;
 335 }
 336
 337 static int echo_page_prep(const struct lu_env *env,
 338                           const struct cl_page_slice *slice,
 339                           struct cl_io *unused)
 340 {
 341         return 0;
 342 }
 343
 344 static int echo_page_print(const struct lu_env *env,
 345                            const struct cl_page_slice *slice,
 346                            void *cookie, lu_printer_t printer)
 347 {
 348         struct echo_page *ep = cl2echo_page(slice);
 349
 350         (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p %d vm@%p\n",
 351                    ep, cfs_mutex_is_locked(&ep->ep_lock), ep->ep_vmpage);
 352         return 0;
 353 }
 354
 355 static const struct cl_page_operations echo_page_ops = {
 356         .cpo_own           = echo_page_own,
 357         .cpo_disown        = echo_page_disown,
 358         .cpo_discard       = echo_page_discard,
 359         .cpo_vmpage        = echo_page_vmpage,
 360         .cpo_fini          = echo_page_fini,
 361         .cpo_print         = echo_page_print,
 362         .cpo_is_vmlocked   = echo_page_is_vmlocked,
 363         .io = {
 364                 [CRT_READ] = {
 365                         .cpo_prep        = echo_page_prep,
 366                         .cpo_completion  = echo_page_completion,
 367                 },
 368                 [CRT_WRITE] = {
 369                         .cpo_prep        = echo_page_prep,
 370                         .cpo_completion  = echo_page_completion,
 371                 }
 372         }
 373 };
 374 /** @} echo_page */
 375
 376 /** \defgroup echo_lock Locking
 377  *
 378  * echo lock operations
 379  *
 380  * @{
 381  */
 382 static void echo_lock_fini(const struct lu_env *env,
 383                            struct cl_lock_slice *slice)
 384 {
 385         struct echo_lock *ecl = cl2echo_lock(slice);
 386
 387         LASSERT(cfs_list_empty(&ecl->el_chain));
 388         OBD_SLAB_FREE_PTR(ecl, echo_lock_kmem);
 389 }
 390
 391 static void echo_lock_delete(const struct lu_env *env,
 392                              const struct cl_lock_slice *slice)
 393 {
 394         struct echo_lock *ecl      = cl2echo_lock(slice);
 395
 396         LASSERT(cfs_list_empty(&ecl->el_chain));
 397 }
 398
 399 static int echo_lock_fits_into(const struct lu_env *env,
 400                                const struct cl_lock_slice *slice,
 401                                const struct cl_lock_descr *need,
 402                                const struct cl_io *unused)
 403 {
 404         return 1;
 405 }
 406
 407 static struct cl_lock_operations echo_lock_ops = {
 408         .clo_fini      = echo_lock_fini,
 409         .clo_delete    = echo_lock_delete,
 410         .clo_fits_into = echo_lock_fits_into
 411 };
 412
 413 /** @} echo_lock */
 414
 415 /** \defgroup echo_cl_ops cl_object operations
 416  *
 417  * operations for cl_object
 418  *
 419  * @{
 420  */
 421 static struct cl_page *echo_page_init(const struct lu_env *env,
 422                                       struct cl_object *obj,
 423                                       struct cl_page *page, cfs_page_t *vmpage)
 424 {
 425         struct echo_page *ep;
 426         ENTRY;
 427
 428         OBD_SLAB_ALLOC_PTR_GFP(ep, echo_page_kmem, CFS_ALLOC_IO);
 429         if (ep != NULL) {
 430                 struct echo_object *eco = cl2echo_obj(obj);
 431                 ep->ep_vmpage = vmpage;
 432                 page_cache_get(vmpage);
 433                 cfs_mutex_init(&ep->ep_lock);
 434                 cl_page_slice_add(page, &ep->ep_cl, obj, &echo_page_ops);
 435                 cfs_atomic_inc(&eco->eo_npages);
 436         }
 437         RETURN(ERR_PTR(ep ? 0 : -ENOMEM));
 438 }
 439
 440 static int echo_io_init(const struct lu_env *env, struct cl_object *obj,
 441                         struct cl_io *io)
 442 {
 443         return 0;
 444 }
 445
 446 static int echo_lock_init(const struct lu_env *env,
 447                           struct cl_object *obj, struct cl_lock *lock,
 448                           const struct cl_io *unused)
 449 {
 450         struct echo_lock *el;
 451         ENTRY;
 452
 453         OBD_SLAB_ALLOC_PTR_GFP(el, echo_lock_kmem, CFS_ALLOC_IO);
 454         if (el != NULL) {
 455                 cl_lock_slice_add(lock, &el->el_cl, obj, &echo_lock_ops);
 456                 el->el_object = cl2echo_obj(obj);
 457                 CFS_INIT_LIST_HEAD(&el->el_chain);
 458                 cfs_atomic_set(&el->el_refcount, 0);
 459         }
 460         RETURN(el == NULL ? -ENOMEM : 0);
 461 }
 462
 463 static int echo_conf_set(const struct lu_env *env, struct cl_object *obj,
 464                          const struct cl_object_conf *conf)
 465 {
 466         return 0;
 467 }
 468
 469 static const struct cl_object_operations echo_cl_obj_ops = {
 470         .coo_page_init = echo_page_init,
 471         .coo_lock_init = echo_lock_init,
 472         .coo_io_init   = echo_io_init,
 473         .coo_conf_set  = echo_conf_set
 474 };
 475 /** @} echo_cl_ops */
 476
 477 /** \defgroup echo_lu_ops lu_object operations
 478  *
 479  * operations for echo lu object.
 480  *
 481  * @{
 482  */
 483 static int echo_object_init(const struct lu_env *env, struct lu_object *obj,
 484                             const struct lu_object_conf *conf)
 485 {
 486         struct echo_device *ed         = cl2echo_dev(lu2cl_dev(obj->lo_dev));
 487         struct echo_client_obd *ec     = ed->ed_ec;
 488         struct echo_object *eco        = cl2echo_obj(lu2cl(obj));
 489         ENTRY;
 490
 491         if (ed->ed_next) {
 492                 struct lu_object  *below;
 493                 struct lu_device  *under;
 494
 495                 under = ed->ed_next;
 496                 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header,
 497                                                         under);
 498                 if (below == NULL)
 499                         RETURN(-ENOMEM);
 500                 lu_object_add(obj, below);
 501         }
 502
 503         if (!ed->ed_next_ismd) {
 504                 const struct cl_object_conf *cconf = lu2cl_conf(conf);
 505                 struct echo_object_conf *econf = cl2echo_conf(cconf);
 506
 507                 LASSERT(econf->eoc_md);
 508                 eco->eo_lsm = *econf->eoc_md;
 509                 /* clear the lsm pointer so that it won't get freed. */
 510                 *econf->eoc_md = NULL;
 511         } else {
 512                 eco->eo_lsm = NULL;
 513         }
 514
 515         eco->eo_dev = ed;
 516         cfs_atomic_set(&eco->eo_npages, 0);
 517
 518         cfs_spin_lock(&ec->ec_lock);
 519         cfs_list_add_tail(&eco->eo_obj_chain, &ec->ec_objects);
 520         cfs_spin_unlock(&ec->ec_lock);
 521
 522         RETURN(0);
 523 }
 524
 525 static void echo_object_free(const struct lu_env *env, struct lu_object *obj)
 526 {
 527         struct echo_object *eco    = cl2echo_obj(lu2cl(obj));
 528         struct echo_client_obd *ec = eco->eo_dev->ed_ec;
 529         ENTRY;
 530
 531         LASSERT(cfs_atomic_read(&eco->eo_npages) == 0);
 532
 533         cfs_spin_lock(&ec->ec_lock);
 534         cfs_list_del_init(&eco->eo_obj_chain);
 535         cfs_spin_unlock(&ec->ec_lock);
 536
 537         lu_object_fini(obj);
 538         lu_object_header_fini(obj->lo_header);
 539
 540         if (eco->eo_lsm)
 541                 obd_free_memmd(ec->ec_exp, &eco->eo_lsm);
 542         OBD_SLAB_FREE_PTR(eco, echo_object_kmem);
 543         EXIT;
 544 }
 545
 546 static int echo_object_print(const struct lu_env *env, void *cookie,
 547                             lu_printer_t p, const struct lu_object *o)
 548 {
 549         struct echo_object *obj = cl2echo_obj(lu2cl(o));
 550
 551         return (*p)(env, cookie, "echoclient-object@%p", obj);
 552 }
 553
 554 static const struct lu_object_operations echo_lu_obj_ops = {
 555         .loo_object_init      = echo_object_init,
 556         .loo_object_delete    = NULL,
 557         .loo_object_release   = NULL,
 558         .loo_object_free      = echo_object_free,
 559         .loo_object_print     = echo_object_print,
 560         .loo_object_invariant = NULL
 561 };
 562 /** @} echo_lu_ops */
 563
 564 /** \defgroup echo_lu_dev_ops  lu_device operations
 565  *
 566  * Operations for echo lu device.
 567  *
 568  * @{
 569  */
 570 static struct lu_object *echo_object_alloc(const struct lu_env *env,
 571                                            const struct lu_object_header *hdr,
 572                                            struct lu_device *dev)
 573 {
 574         struct echo_object *eco;
 575         struct lu_object *obj = NULL;
 576         ENTRY;
 577
 578         /* we're the top dev. */
 579         LASSERT(hdr == NULL);
 580         OBD_SLAB_ALLOC_PTR_GFP(eco, echo_object_kmem, CFS_ALLOC_IO);
 581         if (eco != NULL) {
 582                 struct cl_object_header *hdr = &eco->eo_hdr;
 583
 584                 obj = &echo_obj2cl(eco)->co_lu;
 585                 cl_object_header_init(hdr);
 586                 lu_object_init(obj, &hdr->coh_lu, dev);
 587                 lu_object_add_top(&hdr->coh_lu, obj);
 588
 589                 eco->eo_cl.co_ops = &echo_cl_obj_ops;
 590                 obj->lo_ops       = &echo_lu_obj_ops;
 591         }
 592         RETURN(obj);
 593 }
 594
 595 static struct lu_device_operations echo_device_lu_ops = {
 596         .ldo_object_alloc   = echo_object_alloc,
 597 };
 598
 599 /** @} echo_lu_dev_ops */
 600
 601 static struct cl_device_operations echo_device_cl_ops = {
 602 };
 603
 604 /** \defgroup echo_init Setup and teardown
 605  *
 606  * Init and fini functions for echo client.
 607  *
 608  * @{
 609  */
 610 static int echo_site_init(const struct lu_env *env, struct echo_device *ed)
 611 {
 612         struct cl_site *site = &ed->ed_site_myself;
 613         int rc;
 614
 615         /* initialize site */
 616         rc = cl_site_init(site, &ed->ed_cl);
 617         if (rc) {
 618                 CERROR("Cannot initilize site for echo client(%d)\n", rc);
 619                 return rc;
 620         }
 621
 622         rc = lu_site_init_finish(&site->cs_lu);
 623         if (rc)
 624                 return rc;
 625
 626         ed->ed_site = site;
 627         return 0;
 628 }
 629
 630 static void echo_site_fini(const struct lu_env *env, struct echo_device *ed)
 631 {
 632         if (ed->ed_site) {
 633                 if (!ed->ed_next_ismd)
 634                         cl_site_fini(ed->ed_site);
 635                 ed->ed_site = NULL;
 636         }
 637 }
 638
 639 static void *echo_thread_key_init(const struct lu_context *ctx,
 640                           struct lu_context_key *key)
 641 {
 642         struct echo_thread_info *info;
 643
 644         OBD_SLAB_ALLOC_PTR_GFP(info, echo_thread_kmem, CFS_ALLOC_IO);
 645         if (info == NULL)
 646                 info = ERR_PTR(-ENOMEM);
 647         return info;
 648 }
 649
 650 static void echo_thread_key_fini(const struct lu_context *ctx,
 651                          struct lu_context_key *key, void *data)
 652 {
 653         struct echo_thread_info *info = data;
 654         OBD_SLAB_FREE_PTR(info, echo_thread_kmem);
 655 }
 656
 657 static void echo_thread_key_exit(const struct lu_context *ctx,
 658                          struct lu_context_key *key, void *data)
 659 {
 660 }
 661
 662 static struct lu_context_key echo_thread_key = {
 663         .lct_tags = LCT_CL_THREAD,
 664         .lct_init = echo_thread_key_init,
 665         .lct_fini = echo_thread_key_fini,
 666         .lct_exit = echo_thread_key_exit
 667 };
 668
 669 static void *echo_session_key_init(const struct lu_context *ctx,
 670                                   struct lu_context_key *key)
 671 {
 672         struct echo_session_info *session;
 673
 674         OBD_SLAB_ALLOC_PTR_GFP(session, echo_session_kmem, CFS_ALLOC_IO);
 675         if (session == NULL)
 676                 session = ERR_PTR(-ENOMEM);
 677         return session;
 678 }
 679
 680 static void echo_session_key_fini(const struct lu_context *ctx,
 681                                  struct lu_context_key *key, void *data)
 682 {
 683         struct echo_session_info *session = data;
 684         OBD_SLAB_FREE_PTR(session, echo_session_kmem);
 685 }
 686
 687 static void echo_session_key_exit(const struct lu_context *ctx,
 688                                  struct lu_context_key *key, void *data)
 689 {
 690 }
 691
 692 static struct lu_context_key echo_session_key = {
 693         .lct_tags = LCT_SESSION,
 694         .lct_init = echo_session_key_init,
 695         .lct_fini = echo_session_key_fini,
 696         .lct_exit = echo_session_key_exit
 697 };
 698
 699 LU_TYPE_INIT_FINI(echo, &echo_thread_key, &echo_session_key);
 700
 701 #define ECHO_SEQ_WIDTH 0xffffffff
 702 static int echo_fid_init(struct echo_device *ed, char *obd_name,
 703                          struct md_site *ms)
 704 {
 705         char *prefix;
 706         int rc;
 707         ENTRY;
 708
 709         OBD_ALLOC_PTR(ed->ed_cl_seq);
 710         if (ed->ed_cl_seq == NULL)
 711                 RETURN(-ENOMEM);
 712
 713         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
 714         if (prefix == NULL)
 715                 GOTO(out_free_seq, rc = -ENOMEM);
 716
 717         snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", obd_name);
 718
 719         /* Init client side sequence-manager */
 720         rc = seq_client_init(ed->ed_cl_seq, NULL,
 721                              LUSTRE_SEQ_METADATA,
 722                              prefix, ms->ms_server_seq);
 723         ed->ed_cl_seq->lcs_width = ECHO_SEQ_WIDTH;
 724         OBD_FREE(prefix, MAX_OBD_NAME + 5);
 725         if (rc)
 726                 GOTO(out_free_seq, rc);
 727
 728         RETURN(0);
 729
 730 out_free_seq:
 731         OBD_FREE_PTR(ed->ed_cl_seq);
 732         ed->ed_cl_seq = NULL;
 733         RETURN(rc);
 734 }
 735
 736 static int echo_fid_fini(struct obd_device *obddev)
 737 {
 738         struct echo_device *ed = obd2echo_dev(obddev);
 739         ENTRY;
 740
 741         if (ed->ed_cl_seq != NULL) {
 742                 seq_client_fini(ed->ed_cl_seq);
 743                 OBD_FREE_PTR(ed->ed_cl_seq);
 744                 ed->ed_cl_seq = NULL;
 745         }
 746
 747         RETURN(0);
 748 }
 749
 750 static struct lu_device *echo_device_alloc(const struct lu_env *env,
 751                                            struct lu_device_type *t,
 752                                            struct lustre_cfg *cfg)
 753 {
 754         struct lu_device   *next;
 755         struct echo_device *ed;
 756         struct cl_device   *cd;
 757         struct obd_device  *obd = NULL; /* to keep compiler happy */
 758         struct obd_device  *tgt;
 759         const char *tgt_type_name;
 760         int rc;
 761         int cleanup = 0;
 762         ENTRY;
 763
 764         OBD_ALLOC_PTR(ed);
 765         if (ed == NULL)
 766                 GOTO(out, rc = -ENOMEM);
 767
 768         cleanup = 1;
 769         cd = &ed->ed_cl;
 770         rc = cl_device_init(cd, t);
 771         if (rc)
 772                 GOTO(out, rc);
 773
 774         cd->cd_lu_dev.ld_ops = &echo_device_lu_ops;
 775         cd->cd_ops = &echo_device_cl_ops;
 776
 777         cleanup = 2;
 778         obd = class_name2obd(lustre_cfg_string(cfg, 0));
 779         LASSERT(obd != NULL);
 780         LASSERT(env != NULL);
 781
 782         tgt = class_name2obd(lustre_cfg_string(cfg, 1));
 783         if (tgt == NULL) {
 784                 CERROR("Can not find tgt device %s\n",
 785                         lustre_cfg_string(cfg, 1));
 786                 GOTO(out, rc = -ENODEV);
 787         }
 788
 789         next = tgt->obd_lu_dev;
 790         if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
 791                 ed->ed_next_ismd = 1;
 792         } else {
 793                 ed->ed_next_ismd = 0;
 794                 rc = echo_site_init(env, ed);
 795                 if (rc)
 796                         GOTO(out, rc);
 797         }
 798         cleanup = 3;
 799
 800         rc = echo_client_setup(env, obd, cfg);
 801         if (rc)
 802                 GOTO(out, rc);
 803
 804         ed->ed_ec = &obd->u.echo_client;
 805         cleanup = 4;
 806
 807         if (ed->ed_next_ismd) {
 808                 /* Suppose to connect to some Metadata layer */
 809                 struct lu_site *ls;
 810                 struct lu_device *ld;
 811                 int    found = 0;
 812
 813                 if (next == NULL) {
 814                         CERROR("%s is not lu device type!\n",
 815                                lustre_cfg_string(cfg, 1));
 816                         GOTO(out, rc = -EINVAL);
 817                 }
 818
 819                 tgt_type_name = lustre_cfg_string(cfg, 2);
 820                 if (!tgt_type_name) {
 821                         CERROR("%s no type name for echo %s setup\n",
 822                                 lustre_cfg_string(cfg, 1),
 823                                 tgt->obd_type->typ_name);
 824                         GOTO(out, rc = -EINVAL);
 825                 }
 826
 827                 ls = next->ld_site;
 828
 829                 cfs_spin_lock(&ls->ls_ld_lock);
 830                 cfs_list_for_each_entry(ld, &ls->ls_ld_linkage, ld_linkage) {
 831                         if (strcmp(ld->ld_type->ldt_name, tgt_type_name) == 0) {
 832                                 found = 1;
 833                                 break;
 834                         }
 835                 }
 836                 cfs_spin_unlock(&ls->ls_ld_lock);
 837
 838                 if (found == 0) {
 839                         CERROR("%s is not lu device type!\n",
 840                                lustre_cfg_string(cfg, 1));
 841                         GOTO(out, rc = -EINVAL);
 842                 }
 843
 844                 next = ld;
 845                 /* For MD echo client, it will use the site in MDS stack */
 846                 ed->ed_site_myself.cs_lu = *ls;
 847                 ed->ed_site = &ed->ed_site_myself;
 848                 ed->ed_cl.cd_lu_dev.ld_site = &ed->ed_site_myself.cs_lu;
 849                 rc = echo_fid_init(ed, obd->obd_name, lu_site2md(ls));
 850                 if (rc) {
 851                         CERROR("echo fid init error %d\n", rc);
 852                         GOTO(out, rc);
 853                 }
 854         } else {
 855                  /* if echo client is to be stacked upon ost device, the next is
 856                   * NULL since ost is not a clio device so far */
 857                 if (next != NULL && !lu_device_is_cl(next))
 858                         next = NULL;
 859
 860                 tgt_type_name = tgt->obd_type->typ_name;
 861                 if (next != NULL) {
 862                         LASSERT(next != NULL);
 863                         if (next->ld_site != NULL)
 864                                 GOTO(out, rc = -EBUSY);
 865
 866                         next->ld_site = &ed->ed_site->cs_lu;
 867                         rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
 868                                                      next->ld_type->ldt_name,
 869                                                      NULL);
 870                         if (rc)
 871                                 GOTO(out, rc);
 872
 873                         /* Tricky case, I have to determine the obd type since
 874                          * CLIO uses the different parameters to initialize
 875                          * objects for lov & osc. */
 876                         if (strcmp(tgt_type_name, LUSTRE_LOV_NAME) == 0)
 877                                 ed->ed_next_islov = 1;
 878                         else
 879                                 LASSERT(strcmp(tgt_type_name,
 880                                                LUSTRE_OSC_NAME) == 0);
 881                 } else
 882                         LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0);
 883         }
 884
 885         ed->ed_next = next;
 886         RETURN(&cd->cd_lu_dev);
 887 out:
 888         switch(cleanup) {
 889         case 4: {
 890                 int rc2;
 891                 rc2 = echo_client_cleanup(obd);
 892                 if (rc2)
 893                         CERROR("Cleanup obd device %s error(%d)\n",
 894                                obd->obd_name, rc2);
 895         }
 896
 897         case 3:
 898                 echo_site_fini(env, ed);
 899         case 2:
 900                 cl_device_fini(&ed->ed_cl);
 901         case 1:
 902                 OBD_FREE_PTR(ed);
 903         case 0:
 904         default:
 905                 break;
 906         }
 907         return(ERR_PTR(rc));
 908 }
 909
 910 static int echo_device_init(const struct lu_env *env, struct lu_device *d,
 911                           const char *name, struct lu_device *next)
 912 {
 913         LBUG();
 914         return 0;
 915 }
 916
 917 static struct lu_device *echo_device_fini(const struct lu_env *env,
 918                                           struct lu_device *d)
 919 {
 920         struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
 921         struct lu_device *next = ed->ed_next;
 922
 923         while (next && !ed->ed_next_ismd)
 924                 next = next->ld_type->ldt_ops->ldto_device_fini(env, next);
 925         return NULL;
 926 }
 927
 928 static void echo_lock_release(const struct lu_env *env,
 929                               struct echo_lock *ecl,
 930                               int still_used)
 931 {
 932         struct cl_lock *clk = echo_lock2cl(ecl);
 933
 934         cl_lock_get(clk);
 935         cl_unuse(env, clk);
 936         cl_lock_release(env, clk, "ec enqueue", ecl->el_object);
 937         if (!still_used) {
 938                 cl_lock_mutex_get(env, clk);
 939                 cl_lock_cancel(env, clk);
 940                 cl_lock_delete(env, clk);
 941                 cl_lock_mutex_put(env, clk);
 942         }
 943         cl_lock_put(env, clk);
 944 }
 945
 946 static struct lu_device *echo_device_free(const struct lu_env *env,
 947                                           struct lu_device *d)
 948 {
 949         struct echo_device     *ed   = cl2echo_dev(lu2cl_dev(d));
 950         struct echo_client_obd *ec   = ed->ed_ec;
 951         struct echo_object     *eco;
 952         struct lu_device       *next = ed->ed_next;
 953
 954         CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n",
 955                ed, next);
 956
 957         lu_site_purge(env, &ed->ed_site->cs_lu, -1);
 958
 959         /* check if there are objects still alive.
 960          * It shouldn't have any object because lu_site_purge would cleanup
 961          * all of cached objects. Anyway, probably the echo device is being
 962          * parallelly accessed.
 963          */
 964         cfs_spin_lock(&ec->ec_lock);
 965         cfs_list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain)
 966                 eco->eo_deleted = 1;
 967         cfs_spin_unlock(&ec->ec_lock);
 968
 969         /* purge again */
 970         lu_site_purge(env, &ed->ed_site->cs_lu, -1);
 971
 972         CDEBUG(D_INFO,
 973                "Waiting for the reference of echo object to be dropped\n");
 974
 975         /* Wait for the last reference to be dropped. */
 976         cfs_spin_lock(&ec->ec_lock);
 977         while (!cfs_list_empty(&ec->ec_objects)) {
 978                 cfs_spin_unlock(&ec->ec_lock);
 979                 CERROR("echo_client still has objects at cleanup time, "
 980                        "wait for 1 second\n");
 981                 cfs_schedule_timeout_and_set_state(CFS_TASK_UNINT,
 982                                                    cfs_time_seconds(1));
 983                 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
 984                 cfs_spin_lock(&ec->ec_lock);
 985         }
 986         cfs_spin_unlock(&ec->ec_lock);
 987
 988         LASSERT(cfs_list_empty(&ec->ec_locks));
 989
 990         CDEBUG(D_INFO, "No object exists, exiting...\n");
 991
 992         echo_client_cleanup(d->ld_obd);
 993         echo_fid_fini(d->ld_obd);
 994         while (next && !ed->ed_next_ismd)
 995                 next = next->ld_type->ldt_ops->ldto_device_free(env, next);
 996
 997         LASSERT(ed->ed_site == lu2cl_site(d->ld_site));
 998         echo_site_fini(env, ed);
 999         cl_device_fini(&ed->ed_cl);
1000         OBD_FREE_PTR(ed);
1001
1002         return NULL;
1003 }
1004
1005 static const struct lu_device_type_operations echo_device_type_ops = {
1006         .ldto_init = echo_type_init,
1007         .ldto_fini = echo_type_fini,
1008
1009         .ldto_start = echo_type_start,
1010         .ldto_stop  = echo_type_stop,
1011
1012         .ldto_device_alloc = echo_device_alloc,
1013         .ldto_device_free  = echo_device_free,
1014         .ldto_device_init  = echo_device_init,
1015         .ldto_device_fini  = echo_device_fini
1016 };
1017
1018 static struct lu_device_type echo_device_type = {
1019         .ldt_tags     = LU_DEVICE_CL,
1020         .ldt_name     = LUSTRE_ECHO_CLIENT_NAME,
1021         .ldt_ops      = &echo_device_type_ops,
1022         .ldt_ctx_tags = LCT_CL_THREAD | LCT_MD_THREAD | LCT_DT_THREAD,
1023 };
1024 /** @} echo_init */
1025
1026 /** \defgroup echo_exports Exported operations
1027  *
1028  * exporting functions to echo client
1029  *
1030  * @{
1031  */
1032
1033 /* Interfaces to echo client obd device */
1034 static struct echo_object *cl_echo_object_find(struct echo_device *d,
1035                                                struct lov_stripe_md **lsmp)
1036 {
1037         struct lu_env *env;
1038         struct echo_thread_info *info;
1039         struct echo_object_conf *conf;
1040         struct lov_stripe_md    *lsm;
1041         struct echo_object *eco;
1042         struct cl_object   *obj;
1043         struct lu_fid *fid;
1044         int refcheck;
1045         ENTRY;
1046
1047         LASSERT(lsmp);
1048         lsm = *lsmp;
1049         LASSERT(lsm);
1050         LASSERT(lsm->lsm_object_id);
1051
1052         /* Never return an object if the obd is to be freed. */
1053         if (echo_dev2cl(d)->cd_lu_dev.ld_obd->obd_stopping)
1054                 RETURN(ERR_PTR(-ENODEV));
1055
1056         env = cl_env_get(&refcheck);
1057         if (IS_ERR(env))
1058                 RETURN((void *)env);
1059
1060         info = echo_env_info(env);
1061         conf = &info->eti_conf;
1062         if (d->ed_next) {
1063                 if (!d->ed_next_islov) {
1064                         struct lov_oinfo *oinfo = lsm->lsm_oinfo[0];
1065                         LASSERT(oinfo != NULL);
1066                         oinfo->loi_id = lsm->lsm_object_id;
1067                         oinfo->loi_seq = lsm->lsm_object_seq;
1068                         conf->eoc_cl.u.coc_oinfo = oinfo;
1069                 } else {
1070                         struct lustre_md *md;
1071                         md = &info->eti_md;
1072                         memset(md, 0, sizeof *md);
1073                         md->lsm = lsm;
1074                         conf->eoc_cl.u.coc_md = md;
1075                 }
1076         }
1077         conf->eoc_md = lsmp;
1078
1079         fid  = &info->eti_fid;
1080         lsm2fid(lsm, fid);
1081
1082         obj = cl_object_find(env, echo_dev2cl(d), fid, &conf->eoc_cl);
1083         if (IS_ERR(obj))
1084                 GOTO(out, eco = (void*)obj);
1085
1086         eco = cl2echo_obj(obj);
1087         if (eco->eo_deleted) {
1088                 cl_object_put(env, obj);
1089                 eco = ERR_PTR(-EAGAIN);
1090         }
1091
1092 out:
1093         cl_env_put(env, &refcheck);
1094         RETURN(eco);
1095 }
1096
1097 static int cl_echo_object_put(struct echo_object *eco)
1098 {
1099         struct lu_env *env;
1100         struct cl_object *obj = echo_obj2cl(eco);
1101         int refcheck;
1102         ENTRY;
1103
1104         env = cl_env_get(&refcheck);
1105         if (IS_ERR(env))
1106                 RETURN(PTR_ERR(env));
1107
1108         /* an external function to kill an object? */
1109         if (eco->eo_deleted) {
1110                 struct lu_object_header *loh = obj->co_lu.lo_header;
1111                 LASSERT(&eco->eo_hdr == luh2coh(loh));
1112                 cfs_set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags);
1113         }
1114
1115         cl_object_put(env, obj);
1116         cl_env_put(env, &refcheck);
1117         RETURN(0);
1118 }
1119
1120 static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco,
1121                             obd_off start, obd_off end, int mode,
1122                             __u64 *cookie , __u32 enqflags)
1123 {
1124         struct cl_io *io;
1125         struct cl_lock *lck;
1126         struct cl_object *obj;
1127         struct cl_lock_descr *descr;
1128         struct echo_thread_info *info;
1129         int rc = -ENOMEM;
1130         ENTRY;
1131
1132         info = echo_env_info(env);
1133         io = &info->eti_io;
1134         descr = &info->eti_descr;
1135         obj = echo_obj2cl(eco);
1136
1137         descr->cld_obj   = obj;
1138         descr->cld_start = cl_index(obj, start);
1139         descr->cld_end   = cl_index(obj, end);
1140         descr->cld_mode  = mode == LCK_PW ? CLM_WRITE : CLM_READ;
1141         descr->cld_enq_flags = enqflags;
1142         io->ci_obj = obj;
1143
1144         lck = cl_lock_request(env, io, descr, "ec enqueue", eco);
1145         if (lck) {
1146                 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
1147                 struct echo_lock *el;
1148
1149                 rc = cl_wait(env, lck);
1150                 if (rc == 0) {
1151                         el = cl2echo_lock(cl_lock_at(lck, &echo_device_type));
1152                         cfs_spin_lock(&ec->ec_lock);
1153                         if (cfs_list_empty(&el->el_chain)) {
1154                                 cfs_list_add(&el->el_chain, &ec->ec_locks);
1155                                 el->el_cookie = ++ec->ec_unique;
1156                         }
1157                         cfs_atomic_inc(&el->el_refcount);
1158                         *cookie = el->el_cookie;
1159                         cfs_spin_unlock(&ec->ec_lock);
1160                 } else
1161                         cl_lock_release(env, lck, "ec enqueue", cfs_current());
1162         }
1163         RETURN(rc);
1164 }
1165
1166 static int cl_echo_enqueue(struct echo_object *eco, obd_off start, obd_off end,
1167                            int mode, __u64 *cookie)
1168 {
1169         struct echo_thread_info *info;
1170         struct lu_env *env;
1171         struct cl_io *io;
1172         int refcheck;
1173         int result;
1174         ENTRY;
1175
1176         env = cl_env_get(&refcheck);
1177         if (IS_ERR(env))
1178                 RETURN(PTR_ERR(env));
1179
1180         info = echo_env_info(env);
1181         io = &info->eti_io;
1182
1183         io->ci_ignore_layout = 1;
1184         result = cl_io_init(env, io, CIT_MISC, echo_obj2cl(eco));
1185         if (result < 0)
1186                 GOTO(out, result);
1187         LASSERT(result == 0);
1188
1189         result = cl_echo_enqueue0(env, eco, start, end, mode, cookie, 0);
1190         cl_io_fini(env, io);
1191
1192         EXIT;
1193 out:
1194         cl_env_put(env, &refcheck);
1195         return result;
1196 }
1197
1198 static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed,
1199                            __u64 cookie)
1200 {
1201         struct echo_client_obd *ec = ed->ed_ec;
1202         struct echo_lock       *ecl = NULL;
1203         cfs_list_t             *el;
1204         int found = 0, still_used = 0;
1205         ENTRY;
1206
1207         LASSERT(ec != NULL);
1208         cfs_spin_lock (&ec->ec_lock);
1209         cfs_list_for_each (el, &ec->ec_locks) {
1210                 ecl = cfs_list_entry (el, struct echo_lock, el_chain);
1211                 CDEBUG(D_INFO, "ecl: %p, cookie: "LPX64"\n", ecl, ecl->el_cookie);
1212                 found = (ecl->el_cookie == cookie);
1213                 if (found) {
1214                         if (cfs_atomic_dec_and_test(&ecl->el_refcount))
1215                                 cfs_list_del_init(&ecl->el_chain);
1216                         else
1217                                 still_used = 1;
1218                         break;
1219                 }
1220         }
1221         cfs_spin_unlock (&ec->ec_lock);
1222
1223         if (!found)
1224                 RETURN(-ENOENT);
1225
1226         echo_lock_release(env, ecl, still_used);
1227         RETURN(0);
1228 }
1229
1230 static int cl_echo_cancel(struct echo_device *ed, __u64 cookie)
1231 {
1232         struct lu_env *env;
1233         int refcheck;
1234         int rc;
1235         ENTRY;
1236
1237         env = cl_env_get(&refcheck);
1238         if (IS_ERR(env))
1239                 RETURN(PTR_ERR(env));
1240
1241         rc = cl_echo_cancel0(env, ed, cookie);
1242
1243         cl_env_put(env, &refcheck);
1244         RETURN(rc);
1245 }
1246
1247 static int cl_echo_async_brw(const struct lu_env *env, struct cl_io *io,
1248                              enum cl_req_type unused, struct cl_2queue *queue)
1249 {
1250         struct cl_page *clp;
1251         struct cl_page *temp;
1252         int result = 0;
1253         ENTRY;
1254
1255         cl_page_list_for_each_safe(clp, temp, &queue->c2_qin) {
1256                 int rc;
1257                 rc = cl_page_cache_add(env, io, clp, CRT_WRITE);
1258                 if (rc == 0)
1259                         continue;
1260                 result = result ?: rc;
1261         }
1262         RETURN(result);
1263 }
1264
1265 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
1266                               cfs_page_t **pages, int npages, int async)
1267 {
1268         struct lu_env           *env;
1269         struct echo_thread_info *info;
1270         struct cl_object        *obj = echo_obj2cl(eco);
1271         struct echo_device      *ed  = eco->eo_dev;
1272         struct cl_2queue        *queue;
1273         struct cl_io            *io;
1274         struct cl_page          *clp;
1275         struct lustre_handle    lh = { 0 };
1276         int page_size = cl_page_size(obj);
1277         int refcheck;
1278         int rc;
1279         int i;
1280         ENTRY;
1281
1282         LASSERT((offset & ~CFS_PAGE_MASK) == 0);
1283         LASSERT(ed->ed_next != NULL);
1284         env = cl_env_get(&refcheck);
1285         if (IS_ERR(env))
1286                 RETURN(PTR_ERR(env));
1287
1288         info    = echo_env_info(env);
1289         io      = &info->eti_io;
1290         queue   = &info->eti_queue;
1291
1292         cl_2queue_init(queue);
1293
1294         io->ci_ignore_layout = 1;
1295         rc = cl_io_init(env, io, CIT_MISC, obj);
1296         if (rc < 0)
1297                 GOTO(out, rc);
1298         LASSERT(rc == 0);
1299
1300
1301         rc = cl_echo_enqueue0(env, eco, offset,
1302                               offset + npages * CFS_PAGE_SIZE - 1,
1303                               rw == READ ? LCK_PR : LCK_PW, &lh.cookie,
1304                               CEF_NEVER);
1305         if (rc < 0)
1306                 GOTO(error_lock, rc);
1307
1308         for (i = 0; i < npages; i++) {
1309                 LASSERT(pages[i]);
1310                 clp = cl_page_find(env, obj, cl_index(obj, offset),
1311                                    pages[i], CPT_TRANSIENT);
1312                 if (IS_ERR(clp)) {
1313                         rc = PTR_ERR(clp);
1314                         break;
1315                 }
1316                 LASSERT(clp->cp_type == CPT_TRANSIENT);
1317
1318                 rc = cl_page_own(env, io, clp);
1319                 if (rc) {
1320                         LASSERT(clp->cp_state == CPS_FREEING);
1321                         cl_page_put(env, clp);
1322                         break;
1323                 }
1324
1325                 cl_2queue_add(queue, clp);
1326
1327                 /* drop the reference count for cl_page_find, so that the page
1328                  * will be freed in cl_2queue_fini. */
1329                 cl_page_put(env, clp);
1330                 cl_page_clip(env, clp, 0, page_size);
1331
1332                 offset += page_size;
1333         }
1334
1335         if (rc == 0) {
1336                 enum cl_req_type typ = rw == READ ? CRT_READ : CRT_WRITE;
1337
1338                 async = async && (typ == CRT_WRITE);
1339                 if (async)
1340                         rc = cl_echo_async_brw(env, io, typ, queue);
1341                 else
1342                         rc = cl_io_submit_sync(env, io, typ, queue, 0);
1343                 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
1344                        async ? "async" : "sync", rc);
1345         }
1346
1347         cl_echo_cancel0(env, ed, lh.cookie);
1348         EXIT;
1349 error_lock:
1350         cl_2queue_discard(env, io, queue);
1351         cl_2queue_disown(env, io, queue);
1352         cl_2queue_fini(env, queue);
1353         cl_io_fini(env, io);
1354 out:
1355         cl_env_put(env, &refcheck);
1356         return rc;
1357 }
1358 /** @} echo_exports */
1359
1360
1361 static obd_id last_object_id;
1362
1363 static int
1364 echo_copyout_lsm (struct lov_stripe_md *lsm, void *_ulsm, int ulsm_nob)
1365 {
1366         struct lov_stripe_md *ulsm = _ulsm;
1367         int nob, i;
1368
1369         nob = offsetof (struct lov_stripe_md, lsm_oinfo[lsm->lsm_stripe_count]);
1370         if (nob > ulsm_nob)
1371                 return (-EINVAL);
1372
1373         if (cfs_copy_to_user (ulsm, lsm, sizeof(ulsm)))
1374                 return (-EFAULT);
1375
1376         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1377                 if (cfs_copy_to_user (ulsm->lsm_oinfo[i], lsm->lsm_oinfo[i],
1378                                       sizeof(lsm->lsm_oinfo[0])))
1379                         return (-EFAULT);
1380         }
1381         return 0;
1382 }
1383
1384 static int
1385 echo_copyin_lsm (struct echo_device *ed, struct lov_stripe_md *lsm,
1386                  void *ulsm, int ulsm_nob)
1387 {
1388         struct echo_client_obd *ec = ed->ed_ec;
1389         int                     i;
1390
1391         if (ulsm_nob < sizeof (*lsm))
1392                 return (-EINVAL);
1393
1394         if (cfs_copy_from_user (lsm, ulsm, sizeof (*lsm)))
1395                 return (-EFAULT);
1396
1397         if (lsm->lsm_stripe_count > ec->ec_nstripes ||
1398             lsm->lsm_magic != LOV_MAGIC ||
1399             (lsm->lsm_stripe_size & (~CFS_PAGE_MASK)) != 0 ||
1400             ((__u64)lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL))
1401                 return (-EINVAL);
1402
1403
1404         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1405                 if (cfs_copy_from_user(lsm->lsm_oinfo[i],
1406                                        ((struct lov_stripe_md *)ulsm)-> \
1407                                        lsm_oinfo[i],
1408                                        sizeof(lsm->lsm_oinfo[0])))
1409                         return (-EFAULT);
1410         }
1411         return (0);
1412 }
1413
1414 static inline void echo_md_build_name(struct lu_name *lname, char *name,
1415                                       __u64 id)
1416 {
1417         sprintf(name, LPU64, id);
1418         lname->ln_name = name;
1419         lname->ln_namelen = strlen(name);
1420 }
1421
1422 static int
1423 echo_md_create_internal(const struct lu_env *env, struct echo_device *ed,
1424                         struct md_object *parent, struct lu_fid *fid,
1425                         struct lu_name *lname, struct md_op_spec *spec,
1426                         struct md_attr *ma)
1427 {
1428         struct lu_object        *ec_child, *child;
1429         struct lu_device        *ld = ed->ed_next;
1430         struct echo_thread_info *info = echo_env_info(env);
1431         struct lu_fid           *fid2 = &info->eti_fid2;
1432         struct lu_object_conf    conf = { .loc_flags = LOC_F_NEW };
1433         int                      rc;
1434
1435         rc = mdo_lookup(env, parent, lname, fid2, spec);
1436         if (rc == 0)
1437                 return -EEXIST;
1438         else if (rc != -ENOENT)
1439                 return rc;
1440
1441         ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev,
1442                                      fid, &conf);
1443         if (IS_ERR(ec_child)) {
1444                 CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid),
1445                         PTR_ERR(ec_child));
1446                 return PTR_ERR(ec_child);
1447         }
1448
1449         child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1450         if (child == NULL) {
1451                 CERROR("Can not locate the child "DFID"\n", PFID(fid));
1452                 GOTO(out_put, rc = -EINVAL);
1453         }
1454
1455         CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n",
1456                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1457
1458         /*
1459          * Do not perform lookup sanity check. We know that name does not exist.
1460          */
1461         spec->sp_cr_lookup = 0;
1462         rc = mdo_create(env, parent, lname, lu2md(child), spec, ma);
1463         if (rc) {
1464                 CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc);
1465                 GOTO(out_put, rc);
1466         }
1467         CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc  = %d\n",
1468                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc);
1469 out_put:
1470         lu_object_put(env, ec_child);
1471         return rc;
1472 }
1473
1474 static int echo_set_lmm_size(const struct lu_env *env,
1475                              struct lu_device *ld,
1476                              struct md_attr *ma)
1477 {
1478         struct md_device *md = lu2md_dev(ld);
1479         int lmm_size, cookie_size, rc;
1480         ENTRY;
1481
1482         md = lu2md_dev(ld);
1483         rc = md->md_ops->mdo_maxsize_get(env, md,
1484                                          &lmm_size, &cookie_size);
1485         if (rc)
1486                 RETURN(rc);
1487
1488         ma->ma_lmm_size = lmm_size;
1489         if (lmm_size > 0) {
1490                 OBD_ALLOC(ma->ma_lmm, lmm_size);
1491                 if (ma->ma_lmm == NULL) {
1492                         ma->ma_lmm_size = 0;
1493                         RETURN(-ENOMEM);
1494                 }
1495         }
1496
1497         ma->ma_cookie_size = cookie_size;
1498         if (cookie_size > 0) {
1499                 OBD_ALLOC(ma->ma_cookie, cookie_size);
1500                 if (ma->ma_cookie == NULL) {
1501                         ma->ma_cookie_size = 0;
1502                         RETURN(-ENOMEM);
1503                 }
1504         }
1505
1506         RETURN(0);
1507 }
1508
1509 static int echo_create_md_object(const struct lu_env *env,
1510                                  struct echo_device *ed,
1511                                  struct lu_object *ec_parent,
1512                                  struct lu_fid *fid,
1513                                  char *name, int namelen,
1514                                  __u64 id, __u32 mode, int count,
1515                                  int stripe_count, int stripe_offset)
1516 {
1517         struct lu_object        *parent;
1518         struct echo_thread_info *info = echo_env_info(env);
1519         struct lu_name          *lname = &info->eti_lname;
1520         struct md_op_spec       *spec = &info->eti_spec;
1521         struct md_attr          *ma = &info->eti_ma;
1522         struct lu_device        *ld = ed->ed_next;
1523         int                      rc = 0;
1524         int                      i;
1525
1526         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1527         if (ec_parent == NULL) {
1528                 lu_object_put(env, ec_parent);
1529                 RETURN(PTR_ERR(parent));
1530         }
1531
1532         memset(ma, 0, sizeof(*ma));
1533         memset(spec, 0, sizeof(*spec));
1534         if (stripe_count != 0) {
1535                 spec->sp_cr_flags |= FMODE_WRITE;
1536                 rc = echo_set_lmm_size(env, ld, ma);
1537                 if (rc)
1538                         GOTO(out_free, rc);
1539                 if (stripe_count != -1) {
1540                         struct lov_user_md_v3 *lum = &info->eti_lum;
1541                         lum->lmm_magic = LOV_USER_MAGIC_V3;
1542                         lum->lmm_stripe_count = stripe_count;
1543                         lum->lmm_stripe_offset = stripe_offset;
1544                         lum->lmm_pattern = 0;
1545                         spec->u.sp_ea.eadata = lum;
1546                         spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1547                 }
1548         }
1549
1550         ma->ma_attr.la_mode = mode;
1551         ma->ma_attr.la_valid = LA_CTIME;
1552         ma->ma_attr.la_ctime = cfs_time_current_64();
1553
1554         if (name != NULL) {
1555                 lname->ln_name = name;
1556                 lname->ln_namelen = namelen;
1557                 /* If name is specified, only create one object by name */
1558                 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1559                                              spec, ma);
1560                 GOTO(out_free, rc);
1561         }
1562
1563         /* Create multiple object sequenced by id */
1564         for (i = 0; i < count; i++) {
1565                 char *tmp_name = info->eti_name;
1566
1567                 echo_md_build_name(lname, tmp_name, id);
1568
1569                 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1570                                              spec, ma);
1571                 if (rc) {
1572                         CERROR("Can not create child %s: rc = %d\n", tmp_name,
1573                                 rc);
1574                         break;
1575                 }
1576                 id++;
1577                 fid->f_oid++;
1578         }
1579
1580 out_free:
1581         if (ma->ma_lmm_size > 0 && ma->ma_lmm != NULL)
1582                 OBD_FREE(ma->ma_lmm, ma->ma_lmm_size);
1583         if (ma->ma_cookie_size > 0 && ma->ma_cookie != NULL)
1584                 OBD_FREE(ma->ma_cookie, ma->ma_cookie_size);
1585
1586         return rc;
1587 }
1588
1589 static struct lu_object *echo_md_lookup(const struct lu_env *env,
1590                                         struct echo_device *ed,
1591                                         struct md_object *parent,
1592                                         struct lu_name *lname)
1593 {
1594         struct echo_thread_info *info = echo_env_info(env);
1595         struct lu_fid           *fid = &info->eti_fid;
1596         struct lu_object        *child;
1597         int    rc;
1598         ENTRY;
1599
1600         CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name,
1601                PFID(fid), parent);
1602         rc = mdo_lookup(env, parent, lname, fid, NULL);
1603         if (rc) {
1604                 CERROR("lookup %s: rc = %d\n", lname->ln_name, rc);
1605                 RETURN(ERR_PTR(rc));
1606         }
1607
1608         child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1609
1610         RETURN(child);
1611 }
1612
1613 static int echo_setattr_object(const struct lu_env *env,
1614                                struct echo_device *ed,
1615                                struct lu_object *ec_parent,
1616                                __u64 id, int count)
1617 {
1618         struct lu_object        *parent;
1619         struct echo_thread_info *info = echo_env_info(env);
1620         struct lu_name          *lname = &info->eti_lname;
1621         char                    *name = info->eti_name;
1622         struct lu_device        *ld = ed->ed_next;
1623         struct lu_buf           *buf = &info->eti_buf;
1624         int                      rc = 0;
1625         int                      i;
1626
1627         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1628         if (ec_parent == NULL) {
1629                 lu_object_put(env, ec_parent);
1630                 return PTR_ERR(parent);
1631         }
1632
1633         buf->lb_buf = info->eti_xattr_buf;
1634         buf->lb_len = sizeof(info->eti_xattr_buf);
1635         for (i = 0; i < count; i++) {
1636                 struct lu_object *ec_child, *child;
1637
1638                 echo_md_build_name(lname, name, id);
1639
1640                 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1641                 if (IS_ERR(ec_child)) {
1642                         CERROR("Can't find child %s: rc = %ld\n",
1643                                 lname->ln_name, PTR_ERR(ec_child));
1644                         RETURN(PTR_ERR(ec_child));
1645                 }
1646
1647                 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1648                 if (child == NULL) {
1649                         CERROR("Can not locate the child %s\n", lname->ln_name);
1650                         lu_object_put(env, ec_child);
1651                         rc = -EINVAL;
1652                         break;
1653                 }
1654
1655                 CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n",
1656                        PFID(lu_object_fid(child)));
1657
1658                 sprintf(name, "%s.test1", XATTR_USER_PREFIX);
1659                 rc = mo_xattr_set(env, lu2md(child), buf, name,
1660                                   LU_XATTR_CREATE);
1661                 if (rc) {
1662                         CERROR("Can not setattr child "DFID": rc = %d\n",
1663                                 PFID(lu_object_fid(child)), rc);
1664                         lu_object_put(env, ec_child);
1665                         break;
1666                 }
1667                 CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n",
1668                        PFID(lu_object_fid(child)));
1669                 id++;
1670                 lu_object_put(env, ec_child);
1671         }
1672         return rc;
1673 }
1674
1675 static int echo_getattr_object(const struct lu_env *env,
1676                                struct echo_device *ed,
1677                                struct lu_object *ec_parent,
1678                                __u64 id, int count)
1679 {
1680         struct lu_object        *parent;
1681         struct echo_thread_info *info = echo_env_info(env);
1682         struct lu_name          *lname = &info->eti_lname;
1683         char                    *name = info->eti_name;
1684         struct md_attr          *ma = &info->eti_ma;
1685         struct lu_device        *ld = ed->ed_next;
1686         int                      rc = 0;
1687         int                      i;
1688
1689         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1690         if (ec_parent == NULL) {
1691                 lu_object_put(env, ec_parent);
1692                 return PTR_ERR(parent);
1693         }
1694
1695         memset(ma, 0, sizeof(*ma));
1696         rc = echo_set_lmm_size(env, ld, ma);
1697         if (rc)
1698                 GOTO(out_free, rc);
1699
1700         ma->ma_need |= MA_INODE | MA_LOV | MA_PFID | MA_HSM | MA_ACL_DEF;
1701         ma->ma_acl = info->eti_xattr_buf;
1702         ma->ma_acl_size = sizeof(info->eti_xattr_buf);
1703
1704         for (i = 0; i < count; i++) {
1705                 struct lu_object *ec_child, *child;
1706
1707                 ma->ma_valid = 0;
1708                 echo_md_build_name(lname, name, id);
1709
1710                 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1711                 if (IS_ERR(ec_child)) {
1712                         CERROR("Can't find child %s: rc = %ld\n",
1713                                lname->ln_name, PTR_ERR(ec_child));
1714                         RETURN(PTR_ERR(ec_child));
1715                 }
1716
1717                 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1718                 if (child == NULL) {
1719                         CERROR("Can not locate the child %s\n", lname->ln_name);
1720                         lu_object_put(env, ec_child);
1721                         GOTO(out_free, rc = -EINVAL);
1722                 }
1723
1724                 CDEBUG(D_RPCTRACE, "Start getattr object "DFID"\n",
1725                        PFID(lu_object_fid(child)));
1726                 rc = mo_attr_get(env, lu2md(child), ma);
1727                 if (rc) {
1728                         CERROR("Can not getattr child "DFID": rc = %d\n",
1729                                 PFID(lu_object_fid(child)), rc);
1730                         lu_object_put(env, ec_child);
1731                         break;
1732                 }
1733                 CDEBUG(D_RPCTRACE, "End getattr object "DFID"\n",
1734                        PFID(lu_object_fid(child)));
1735                 id++;
1736                 lu_object_put(env, ec_child);
1737         }
1738
1739 out_free:
1740         if (ma->ma_lmm_size > 0 && ma->ma_lmm != NULL)
1741                 OBD_FREE(ma->ma_lmm, ma->ma_lmm_size);
1742         if (ma->ma_cookie_size > 0 && ma->ma_cookie != NULL)
1743                 OBD_FREE(ma->ma_cookie, ma->ma_cookie_size);
1744         return rc;
1745 }
1746
1747 static int echo_lookup_object(const struct lu_env *env,
1748                               struct echo_device *ed,
1749                               struct lu_object *ec_parent,
1750                               __u64 id, int count)
1751 {
1752         struct lu_object        *parent;
1753         struct echo_thread_info *info = echo_env_info(env);
1754         struct lu_name          *lname = &info->eti_lname;
1755         char                    *name = info->eti_name;
1756         struct lu_fid           *fid = &info->eti_fid;
1757         struct lu_device        *ld = ed->ed_next;
1758         int                      rc = 0;
1759         int                      i;
1760
1761         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1762         if (ec_parent == NULL) {
1763                 lu_object_put(env, ec_parent);
1764                 return PTR_ERR(parent);
1765         }
1766
1767         /*prepare the requests*/
1768         for (i = 0; i < count; i++) {
1769                 echo_md_build_name(lname, name, id);
1770
1771                 CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n",
1772                        PFID(lu_object_fid(parent)), lname->ln_name, parent);
1773
1774                 rc = mdo_lookup(env, lu2md(parent), lname, fid, NULL);
1775                 if (rc) {
1776                         CERROR("Can not lookup child %s: rc = %d\n", name, rc);
1777                         break;
1778                 }
1779                 CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n",
1780                        PFID(lu_object_fid(parent)), lname->ln_name, parent);
1781
1782                 id++;
1783         }
1784         return rc;
1785 }
1786
1787 static int echo_md_destroy_internal(const struct lu_env *env,
1788                                     struct echo_device *ed,
1789                                     struct md_object *parent,
1790                                     struct lu_name *lname,
1791                                     struct md_attr *ma)
1792 {
1793         struct lu_device   *ld = ed->ed_next;
1794         struct lu_object   *ec_child;
1795         struct lu_object   *child;
1796         int                 rc;
1797
1798         ec_child = echo_md_lookup(env, ed, parent, lname);
1799         if (IS_ERR(ec_child)) {
1800                 CERROR("Can't find child %s: rc = %ld\n", lname->ln_name,
1801                         PTR_ERR(ec_child));
1802                 RETURN(PTR_ERR(ec_child));
1803         }
1804
1805         child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1806         if (child == NULL) {
1807                 CERROR("Can not locate the child %s\n", lname->ln_name);
1808                 GOTO(out_put, rc = -EINVAL);
1809         }
1810
1811         CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n",
1812                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1813
1814 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,3,50,0)
1815         /* After 2.4, MDT will send destroy RPC to OST directly, so no need
1816          * this flag */
1817         ma->ma_valid |= MA_FLAGS;
1818         ma->ma_attr_flags |= MDS_UNLINK_DESTROY;
1819 #else
1820 #warning "Please remove this after 2.4 (LOD/OSP)"
1821 #endif
1822         rc = mdo_unlink(env, parent, lu2md(child), lname, ma);
1823         if (rc) {
1824                 CERROR("Can not unlink child %s: rc = %d\n",
1825                         lname->ln_name, rc);
1826                 GOTO(out_put, rc);
1827         }
1828         CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n",
1829                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1830 out_put:
1831         lu_object_put(env, ec_child);
1832         return rc;
1833 }
1834
1835 static int echo_destroy_object(const struct lu_env *env,
1836                                struct echo_device *ed,
1837                                struct lu_object *ec_parent,
1838                                char *name, int namelen,
1839                                __u64 id, __u32 mode,
1840                                int count)
1841 {
1842         struct echo_thread_info *info = echo_env_info(env);
1843         struct lu_name          *lname = &info->eti_lname;
1844         struct md_attr          *ma = &info->eti_ma;
1845         struct lu_device        *ld = ed->ed_next;
1846         struct lu_object        *parent;
1847         int                      rc = 0;
1848         int                      i;
1849         ENTRY;
1850
1851         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1852         if (parent == NULL)
1853                 RETURN(-EINVAL);
1854
1855         memset(ma, 0, sizeof(*ma));
1856         ma->ma_attr.la_mode = mode;
1857         ma->ma_attr.la_valid = LA_CTIME;
1858         ma->ma_attr.la_ctime = cfs_time_current_64();
1859         ma->ma_need = MA_INODE;
1860         ma->ma_valid = 0;
1861
1862         rc = echo_set_lmm_size(env, ld, ma);
1863         if (rc)
1864                 GOTO(out_free, rc);
1865         if (name != NULL) {
1866                 lname->ln_name = name;
1867                 lname->ln_namelen = namelen;
1868                 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
1869                                               ma);
1870                 GOTO(out_free, rc);
1871         }
1872
1873         /*prepare the requests*/
1874         for (i = 0; i < count; i++) {
1875                 char *tmp_name = info->eti_name;
1876
1877                 ma->ma_need |= MA_LOV;
1878                 ma->ma_valid = 0;
1879                 echo_md_build_name(lname, tmp_name, id);
1880
1881                 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
1882                                               ma);
1883                 if (rc) {
1884                         CERROR("Can not unlink child %s: rc = %d\n", name, rc);
1885                         break;
1886                 }
1887                 id++;
1888         }
1889
1890 out_free:
1891         if (ma->ma_lmm_size > 0 && ma->ma_lmm != NULL)
1892                 OBD_FREE(ma->ma_lmm, ma->ma_lmm_size);
1893         if (ma->ma_cookie_size > 0 && ma->ma_cookie != NULL)
1894                 OBD_FREE(ma->ma_cookie, ma->ma_cookie_size);
1895         RETURN(rc);
1896 }
1897
1898 static struct lu_object *echo_resolve_path(const struct lu_env *env,
1899                                            struct echo_device *ed, char *path,
1900                                            int path_len)
1901 {
1902         struct lu_device        *ld = ed->ed_next;
1903         struct md_device        *md = lu2md_dev(ld);
1904         struct echo_thread_info *info = echo_env_info(env);
1905         struct lu_fid           *fid = &info->eti_fid;
1906         struct lu_name          *lname = &info->eti_lname;
1907         struct lu_object        *parent = NULL;
1908         struct lu_object        *child = NULL;
1909         int rc = 0;
1910         ENTRY;
1911
1912         /*Only support MDD layer right now*/
1913         rc = md->md_ops->mdo_root_get(env, md, fid);
1914         if (rc) {
1915                 CERROR("get root error: rc = %d\n", rc);
1916                 RETURN(ERR_PTR(rc));
1917         }
1918
1919         parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1920         if (IS_ERR(parent)) {
1921                 CERROR("Can not find the parent "DFID": rc = %ld\n",
1922                         PFID(fid), PTR_ERR(parent));
1923                 RETURN(parent);
1924         }
1925
1926         while (1) {
1927                 struct lu_object *ld_parent;
1928                 char *e;
1929
1930                 e = strsep(&path, "/");
1931                 if (e == NULL)
1932                         break;
1933
1934                 if (e[0] == 0) {
1935                         if (!path || path[0] == '\0')
1936                                 break;
1937                         continue;
1938                 }
1939
1940                 lname->ln_name = e;
1941                 lname->ln_namelen = strlen(e);
1942
1943                 ld_parent = lu_object_locate(parent->lo_header, ld->ld_type);
1944                 if (ld_parent == NULL) {
1945                         lu_object_put(env, parent);
1946                         rc = -EINVAL;
1947                         break;
1948                 }
1949
1950                 child = echo_md_lookup(env, ed, lu2md(ld_parent), lname);
1951                 lu_object_put(env, parent);
1952                 if (IS_ERR(child)) {
1953                         rc = (int)PTR_ERR(child);
1954                         CERROR("lookup %s under parent "DFID": rc = %d\n",
1955                                 lname->ln_name, PFID(lu_object_fid(ld_parent)),
1956                                 rc);
1957                         break;
1958                 }
1959                 parent = child;
1960         }
1961         if (rc)
1962                 RETURN(ERR_PTR(rc));
1963
1964         RETURN(parent);
1965 }
1966
1967 #define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_NOREF | LCT_MD_THREAD)
1968 #define ECHO_MD_SES_TAG (LCT_SESSION | LCT_REMEMBER | LCT_NOREF)
1969
1970 static int echo_md_handler(struct echo_device *ed, int command,
1971                            char *path, int path_len, int id, int count,
1972                            struct obd_ioctl_data *data)
1973 {
1974         struct lu_device      *ld = ed->ed_next;
1975         struct lu_env         *env;
1976         int                    refcheck;
1977         struct lu_object      *parent;
1978         char                  *name = NULL;
1979         int                    namelen = data->ioc_plen2;
1980         int                    rc = 0;
1981         ENTRY;
1982
1983         if (ld == NULL) {
1984                 CERROR("MD echo client is not being initialized properly\n");
1985                 RETURN(-EINVAL);
1986         }
1987
1988         if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
1989                 CERROR("Only support MDD layer right now!\n");
1990                 RETURN(-EINVAL);
1991         }
1992
1993         env = cl_env_get(&refcheck);
1994         if (IS_ERR(env))
1995                 RETURN(PTR_ERR(env));
1996
1997         rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG);
1998         if (rc != 0) {
1999                 cl_env_put(env, &refcheck);
2000                 RETURN(rc);
2001         }
2002
2003         parent = echo_resolve_path(env, ed, path, path_len);
2004         if (IS_ERR(parent)) {
2005                 CERROR("Can not resolve the path %s: rc = %ld\n", path,
2006                         PTR_ERR(parent));
2007                 cl_env_put(env, &refcheck);
2008                 RETURN(PTR_ERR(parent));
2009         }
2010
2011         if (namelen > 0) {
2012                 OBD_ALLOC(name, namelen + 1);
2013                 if (name == NULL)
2014                         RETURN(-ENOMEM);
2015                 if (cfs_copy_from_user(name, data->ioc_pbuf2, namelen)) {
2016                         OBD_FREE(name, namelen + 1);
2017                         RETURN(-EFAULT);
2018                 }
2019         }
2020
2021         switch (command) {
2022         case ECHO_MD_CREATE:
2023         case ECHO_MD_MKDIR: {
2024                 struct echo_thread_info *info = echo_env_info(env);
2025                 __u32 mode = data->ioc_obdo2.o_mode;
2026                 struct lu_fid *fid = &info->eti_fid;
2027                 int stripe_count = (int)data->ioc_obdo2.o_misc;
2028                 int stripe_index = (int)data->ioc_obdo2.o_stripe_idx;
2029
2030                 fid->f_seq = data->ioc_obdo1.o_seq;
2031                 fid->f_oid = (__u32)data->ioc_obdo1.o_id;
2032                 fid->f_ver = 0;
2033                 rc = echo_create_md_object(env, ed, parent, fid, name, namelen,
2034                                            id, mode, count, stripe_count,
2035                                            stripe_index);
2036                 break;
2037         }
2038         case ECHO_MD_DESTROY:
2039         case ECHO_MD_RMDIR: {
2040                 __u32 mode = data->ioc_obdo2.o_mode;
2041
2042                 rc = echo_destroy_object(env, ed, parent, name, namelen,
2043                                          id, mode, count);
2044                 break;
2045         }
2046         case ECHO_MD_LOOKUP:
2047                 rc = echo_lookup_object(env, ed, parent, id, count);
2048                 break;
2049         case ECHO_MD_GETATTR:
2050                 rc = echo_getattr_object(env, ed, parent, id, count);
2051                 break;
2052         case ECHO_MD_SETATTR:
2053                 rc = echo_setattr_object(env, ed, parent, id, count);
2054                 break;
2055         default:
2056                 CERROR("unknown command %d\n", command);
2057                 rc = -EINVAL;
2058                 break;
2059         }
2060         if (name != NULL)
2061                 OBD_FREE(name, namelen + 1);
2062         lu_object_put(env, parent);
2063         cl_env_put(env, &refcheck);
2064         return rc;
2065 }
2066
2067 static int echo_create_object(const struct lu_env *env, struct echo_device *ed,
2068                               int on_target, struct obdo *oa, void *ulsm,
2069                               int ulsm_nob, struct obd_trans_info *oti)
2070 {
2071         struct echo_object     *eco;
2072         struct echo_client_obd *ec = ed->ed_ec;
2073         struct lov_stripe_md   *lsm = NULL;
2074         int                     rc;
2075         int                     created = 0;
2076         ENTRY;
2077
2078         if ((oa->o_valid & OBD_MD_FLID) == 0 && /* no obj id */
2079             (on_target ||                       /* set_stripe */
2080              ec->ec_nstripes != 0)) {           /* LOV */
2081                 CERROR ("No valid oid\n");
2082                 RETURN(-EINVAL);
2083         }
2084
2085         rc = obd_alloc_memmd(ec->ec_exp, &lsm);
2086         if (rc < 0) {
2087                 CERROR("Cannot allocate md: rc = %d\n", rc);
2088                 GOTO(failed, rc);
2089         }
2090
2091         if (ulsm != NULL) {
2092                 int i, idx;
2093
2094                 rc = echo_copyin_lsm (ed, lsm, ulsm, ulsm_nob);
2095                 if (rc != 0)
2096                         GOTO(failed, rc);
2097
2098                 if (lsm->lsm_stripe_count == 0)
2099                         lsm->lsm_stripe_count = ec->ec_nstripes;
2100
2101                 if (lsm->lsm_stripe_size == 0)
2102                         lsm->lsm_stripe_size = CFS_PAGE_SIZE;
2103
2104                 idx = cfs_rand();
2105
2106                 /* setup stripes: indices + default ids if required */
2107                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2108                         if (lsm->lsm_oinfo[i]->loi_id == 0)
2109                                 lsm->lsm_oinfo[i]->loi_id = lsm->lsm_object_id;
2110
2111                         lsm->lsm_oinfo[i]->loi_ost_idx =
2112                                 (idx + i) % ec->ec_nstripes;
2113                 }
2114         }
2115
2116         /* setup object ID here for !on_target and LOV hint */
2117         if (oa->o_valid & OBD_MD_FLID)
2118                 lsm->lsm_object_id = oa->o_id;
2119
2120         if (lsm->lsm_object_id == 0)
2121                 lsm->lsm_object_id = ++last_object_id;
2122
2123         rc = 0;
2124         if (on_target) {
2125                 /* Only echo objects are allowed to be created */
2126                 LASSERT((oa->o_valid & OBD_MD_FLGROUP) &&
2127                         (oa->o_seq == FID_SEQ_ECHO));
2128                 rc = obd_create(env, ec->ec_exp, oa, &lsm, oti);
2129                 if (rc != 0) {
2130                         CERROR("Cannot create objects: rc = %d\n", rc);
2131                         GOTO(failed, rc);
2132                 }
2133                 created = 1;
2134         }
2135
2136         /* See what object ID we were given */
2137         oa->o_id = lsm->lsm_object_id;
2138         oa->o_valid |= OBD_MD_FLID;
2139
2140         eco = cl_echo_object_find(ed, &lsm);
2141         if (IS_ERR(eco))
2142                 GOTO(failed, rc = PTR_ERR(eco));
2143         cl_echo_object_put(eco);
2144
2145         CDEBUG(D_INFO, "oa->o_id = %lx\n", (long)oa->o_id);
2146         EXIT;
2147
2148  failed:
2149         if (created && rc)
2150                 obd_destroy(env, ec->ec_exp, oa, lsm, oti, NULL, NULL);
2151         if (lsm)
2152                 obd_free_memmd(ec->ec_exp, &lsm);
2153         if (rc)
2154                 CERROR("create object failed with: rc = %d\n", rc);
2155         return (rc);
2156 }
2157
2158 static int echo_get_object(struct echo_object **ecop, struct echo_device *ed,
2159                            struct obdo *oa)
2160 {
2161         struct echo_client_obd *ec  = ed->ed_ec;
2162         struct lov_stripe_md   *lsm = NULL;
2163         struct echo_object     *eco;
2164         int                     rc;
2165         ENTRY;
2166
2167         if ((oa->o_valid & OBD_MD_FLID) == 0 ||
2168             oa->o_id == 0)  /* disallow use of object id 0 */
2169         {
2170                 CERROR ("No valid oid\n");
2171                 RETURN(-EINVAL);
2172         }
2173
2174         rc = obd_alloc_memmd(ec->ec_exp, &lsm);
2175         if (rc < 0)
2176                 RETURN(rc);
2177
2178         lsm->lsm_object_id = oa->o_id;
2179         if (oa->o_valid & OBD_MD_FLGROUP)
2180                 lsm->lsm_object_seq = oa->o_seq;
2181         else
2182                 lsm->lsm_object_seq = FID_SEQ_ECHO;
2183
2184         rc = 0;
2185         eco = cl_echo_object_find(ed, &lsm);
2186         if (!IS_ERR(eco))
2187                 *ecop = eco;
2188         else
2189                 rc = PTR_ERR(eco);
2190         if (lsm)
2191                 obd_free_memmd(ec->ec_exp, &lsm);
2192         RETURN(rc);
2193 }
2194
2195 static void echo_put_object(struct echo_object *eco)
2196 {
2197         if (cl_echo_object_put(eco))
2198                 CERROR("echo client: drop an object failed");
2199 }
2200
2201 static void
2202 echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp)
2203 {
2204         unsigned long stripe_count;
2205         unsigned long stripe_size;
2206         unsigned long width;
2207         unsigned long woffset;
2208         int           stripe_index;
2209         obd_off       offset;
2210
2211         if (lsm->lsm_stripe_count <= 1)
2212                 return;
2213
2214         offset       = *offp;
2215         stripe_size  = lsm->lsm_stripe_size;
2216         stripe_count = lsm->lsm_stripe_count;
2217
2218         /* width = # bytes in all stripes */
2219         width = stripe_size * stripe_count;
2220
2221         /* woffset = offset within a width; offset = whole number of widths */
2222         woffset = do_div (offset, width);
2223
2224         stripe_index = woffset / stripe_size;
2225
2226         *idp = lsm->lsm_oinfo[stripe_index]->loi_id;
2227         *offp = offset * stripe_size + woffset % stripe_size;
2228 }
2229
2230 static void
2231 echo_client_page_debug_setup(struct lov_stripe_md *lsm,
2232                              cfs_page_t *page, int rw, obd_id id,
2233                              obd_off offset, obd_off count)
2234 {
2235         char    *addr;
2236         obd_off  stripe_off;
2237         obd_id   stripe_id;
2238         int      delta;
2239
2240         /* no partial pages on the client */
2241         LASSERT(count == CFS_PAGE_SIZE);
2242
2243         addr = cfs_kmap(page);
2244
2245         for (delta = 0; delta < CFS_PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2246                 if (rw == OBD_BRW_WRITE) {
2247                         stripe_off = offset + delta;
2248                         stripe_id = id;
2249                         echo_get_stripe_off_id(lsm, &stripe_off, &stripe_id);
2250                 } else {
2251                         stripe_off = 0xdeadbeef00c0ffeeULL;
2252                         stripe_id = 0xdeadbeef00c0ffeeULL;
2253                 }
2254                 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
2255                                   stripe_off, stripe_id);
2256         }
2257
2258         cfs_kunmap(page);
2259 }
2260
2261 static int echo_client_page_debug_check(struct lov_stripe_md *lsm,
2262                                         cfs_page_t *page, obd_id id,
2263                                         obd_off offset, obd_off count)
2264 {
2265         obd_off stripe_off;
2266         obd_id  stripe_id;
2267         char   *addr;
2268         int     delta;
2269         int     rc;
2270         int     rc2;
2271
2272         /* no partial pages on the client */
2273         LASSERT(count == CFS_PAGE_SIZE);
2274
2275         addr = cfs_kmap(page);
2276
2277         for (rc = delta = 0; delta < CFS_PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2278                 stripe_off = offset + delta;
2279                 stripe_id = id;
2280                 echo_get_stripe_off_id (lsm, &stripe_off, &stripe_id);
2281
2282                 rc2 = block_debug_check("test_brw",
2283                                         addr + delta, OBD_ECHO_BLOCK_SIZE,
2284                                         stripe_off, stripe_id);
2285                 if (rc2 != 0) {
2286                         CERROR ("Error in echo object "LPX64"\n", id);
2287                         rc = rc2;
2288                 }
2289         }
2290
2291         cfs_kunmap(page);
2292         return rc;
2293 }
2294
2295 static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa,
2296                             struct echo_object *eco, obd_off offset,
2297                             obd_size count, int async,
2298                             struct obd_trans_info *oti)
2299 {
2300         struct lov_stripe_md   *lsm = eco->eo_lsm;
2301         obd_count               npages;
2302         struct brw_page        *pga;
2303         struct brw_page        *pgp;
2304         cfs_page_t            **pages;
2305         obd_off                 off;
2306         int                     i;
2307         int                     rc;
2308         int                     verify;
2309         int                     gfp_mask;
2310         int                     brw_flags = 0;
2311         ENTRY;
2312
2313         verify = ((oa->o_id) != ECHO_PERSISTENT_OBJID &&
2314                   (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
2315                   (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
2316
2317         gfp_mask = ((oa->o_id & 2) == 0) ? CFS_ALLOC_STD : CFS_ALLOC_HIGHUSER;
2318
2319         LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
2320         LASSERT(lsm != NULL);
2321         LASSERT(lsm->lsm_object_id == oa->o_id);
2322
2323         if (count <= 0 ||
2324             (count & (~CFS_PAGE_MASK)) != 0)
2325                 RETURN(-EINVAL);
2326
2327         /* XXX think again with misaligned I/O */
2328         npages = count >> CFS_PAGE_SHIFT;
2329
2330         if (rw == OBD_BRW_WRITE)
2331                 brw_flags = OBD_BRW_ASYNC;
2332
2333         OBD_ALLOC(pga, npages * sizeof(*pga));
2334         if (pga == NULL)
2335                 RETURN(-ENOMEM);
2336
2337         OBD_ALLOC(pages, npages * sizeof(*pages));
2338         if (pages == NULL) {
2339                 OBD_FREE(pga, npages * sizeof(*pga));
2340                 RETURN(-ENOMEM);
2341         }
2342
2343         for (i = 0, pgp = pga, off = offset;
2344              i < npages;
2345              i++, pgp++, off += CFS_PAGE_SIZE) {
2346
2347                 LASSERT (pgp->pg == NULL);      /* for cleanup */
2348
2349                 rc = -ENOMEM;
2350                 OBD_PAGE_ALLOC(pgp->pg, gfp_mask);
2351                 if (pgp->pg == NULL)
2352                         goto out;
2353
2354                 pages[i] = pgp->pg;
2355                 pgp->count = CFS_PAGE_SIZE;
2356                 pgp->off = off;
2357                 pgp->flag = brw_flags;
2358
2359                 if (verify)
2360                         echo_client_page_debug_setup(lsm, pgp->pg, rw,
2361                                                      oa->o_id, off, pgp->count);
2362         }
2363
2364         /* brw mode can only be used at client */
2365         LASSERT(ed->ed_next != NULL);
2366         rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async);
2367
2368  out:
2369         if (rc != 0 || rw != OBD_BRW_READ)
2370                 verify = 0;
2371
2372         for (i = 0, pgp = pga; i < npages; i++, pgp++) {
2373                 if (pgp->pg == NULL)
2374                         continue;
2375
2376                 if (verify) {
2377                         int vrc;
2378                         vrc = echo_client_page_debug_check(lsm, pgp->pg, oa->o_id,
2379                                                            pgp->off, pgp->count);
2380                         if (vrc != 0 && rc == 0)
2381                                 rc = vrc;
2382                 }
2383                 OBD_PAGE_FREE(pgp->pg);
2384         }
2385         OBD_FREE(pga, npages * sizeof(*pga));
2386         OBD_FREE(pages, npages * sizeof(*pages));
2387         RETURN(rc);
2388 }
2389
2390 static int echo_client_prep_commit(struct obd_export *exp, int rw,
2391                                    struct obdo *oa, struct echo_object *eco,
2392                                    obd_off offset, obd_size count,
2393                                    obd_size batch, struct obd_trans_info *oti,
2394                                    int async)
2395 {
2396         struct lov_stripe_md *lsm = eco->eo_lsm;
2397         struct obd_ioobj ioo;
2398         struct niobuf_local *lnb;
2399         struct niobuf_remote *rnb;
2400         obd_off off;
2401         obd_size npages, tot_pages;
2402         int i, ret = 0;
2403         ENTRY;
2404
2405         if (count <= 0 || (count & (~CFS_PAGE_MASK)) != 0 ||
2406             (lsm != NULL && lsm->lsm_object_id != oa->o_id))
2407                 RETURN(-EINVAL);
2408
2409         npages = batch >> CFS_PAGE_SHIFT;
2410         tot_pages = count >> CFS_PAGE_SHIFT;
2411
2412         OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
2413         OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote));
2414
2415         if (lnb == NULL || rnb == NULL)
2416                 GOTO(out, ret = -ENOMEM);
2417
2418         obdo_to_ioobj(oa, &ioo);
2419
2420         off = offset;
2421
2422         for(; tot_pages; tot_pages -= npages) {
2423                 int lpages;
2424
2425                 if (tot_pages < npages)
2426                         npages = tot_pages;
2427
2428                 for (i = 0; i < npages; i++, off += CFS_PAGE_SIZE) {
2429                         rnb[i].offset = off;
2430                         rnb[i].len = CFS_PAGE_SIZE;
2431                 }
2432
2433                 ioo.ioo_bufcnt = npages;
2434                 oti->oti_transno = 0;
2435
2436                 lpages = npages;
2437                 ret = obd_preprw(NULL, rw, exp, oa, 1, &ioo, rnb, &lpages,
2438                                  lnb, oti, NULL);
2439                 if (ret != 0)
2440                         GOTO(out, ret);
2441                 LASSERT(lpages == npages);
2442
2443                 for (i = 0; i < lpages; i++) {
2444                         cfs_page_t *page = lnb[i].page;
2445
2446                         /* read past eof? */
2447                         if (page == NULL && lnb[i].rc == 0)
2448                                 continue;
2449
2450                         if (async)
2451                                 lnb[i].flags |= OBD_BRW_ASYNC;
2452
2453                         if (oa->o_id == ECHO_PERSISTENT_OBJID ||
2454                             (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
2455                             (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
2456                                 continue;
2457
2458                         if (rw == OBD_BRW_WRITE)
2459                                 echo_client_page_debug_setup(lsm, page, rw,
2460                                                              oa->o_id,
2461                                                              rnb[i].offset,
2462                                                              rnb[i].len);
2463                         else
2464                                 echo_client_page_debug_check(lsm, page,
2465                                                              oa->o_id,
2466                                                              rnb[i].offset,
2467                                                              rnb[i].len);
2468                 }
2469
2470                 ret = obd_commitrw(NULL, rw, exp, oa, 1, &ioo,
2471                                    rnb, npages, lnb, oti, ret);
2472                 if (ret != 0)
2473                         GOTO(out, ret);
2474
2475                 /* Reset oti otherwise it would confuse ldiskfs. */
2476                 memset(oti, 0, sizeof(*oti));
2477         }
2478
2479 out:
2480         if (lnb)
2481                 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
2482         if (rnb)
2483                 OBD_FREE(rnb, npages * sizeof(struct niobuf_remote));
2484         RETURN(ret);
2485 }
2486
2487 static int echo_client_brw_ioctl(int rw, struct obd_export *exp,
2488                                  struct obd_ioctl_data *data)
2489 {
2490         struct obd_device *obd = class_exp2obd(exp);
2491         struct echo_device *ed = obd2echo_dev(obd);
2492         struct echo_client_obd *ec = ed->ed_ec;
2493         struct obd_trans_info dummy_oti = { 0 };
2494         struct obdo *oa = &data->ioc_obdo1;
2495         struct echo_object *eco;
2496         int rc;
2497         int async = 1;
2498         long test_mode;
2499         ENTRY;
2500
2501         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2502
2503         rc = echo_get_object(&eco, ed, oa);
2504         if (rc)
2505                 RETURN(rc);
2506
2507         oa->o_valid &= ~OBD_MD_FLHANDLE;
2508
2509         /* obdfilter doesn't support obd_brw now, simulate via prep + commit */
2510         test_mode = (long)data->ioc_pbuf1;
2511         if (test_mode == 1)
2512                 async = 0;
2513
2514         if (ed->ed_next == NULL && test_mode != 3) {
2515                 test_mode = 3;
2516                 data->ioc_plen1 = data->ioc_count;
2517         }
2518
2519         /* Truncate batch size to maximum */
2520         if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE)
2521                 data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE;
2522
2523         switch (test_mode) {
2524         case 1:
2525                 /* fall through */
2526         case 2:
2527                 rc = echo_client_kbrw(ed, rw, oa,
2528                                       eco, data->ioc_offset,
2529                                       data->ioc_count, async, &dummy_oti);
2530                 break;
2531         case 3:
2532                 rc = echo_client_prep_commit(ec->ec_exp, rw, oa,
2533                                             eco, data->ioc_offset,
2534                                             data->ioc_count, data->ioc_plen1,
2535                                             &dummy_oti, async);
2536                 break;
2537         default:
2538                 rc = -EINVAL;
2539         }
2540         echo_put_object(eco);
2541         RETURN(rc);
2542 }
2543
2544 static int
2545 echo_client_enqueue(struct obd_export *exp, struct obdo *oa,
2546                     int mode, obd_off offset, obd_size nob)
2547 {
2548         struct echo_device     *ed = obd2echo_dev(exp->exp_obd);
2549         struct lustre_handle   *ulh = &oa->o_handle;
2550         struct echo_object     *eco;
2551         obd_off                 end;
2552         int                     rc;
2553         ENTRY;
2554
2555         if (ed->ed_next == NULL)
2556                 RETURN(-EOPNOTSUPP);
2557
2558         if (!(mode == LCK_PR || mode == LCK_PW))
2559                 RETURN(-EINVAL);
2560
2561         if ((offset & (~CFS_PAGE_MASK)) != 0 ||
2562             (nob & (~CFS_PAGE_MASK)) != 0)
2563                 RETURN(-EINVAL);
2564
2565         rc = echo_get_object (&eco, ed, oa);
2566         if (rc != 0)
2567                 RETURN(rc);
2568
2569         end = (nob == 0) ? ((obd_off) -1) : (offset + nob - 1);
2570         rc = cl_echo_enqueue(eco, offset, end, mode, &ulh->cookie);
2571         if (rc == 0) {
2572                 oa->o_valid |= OBD_MD_FLHANDLE;
2573                 CDEBUG(D_INFO, "Cookie is "LPX64"\n", ulh->cookie);
2574         }
2575         echo_put_object(eco);
2576         RETURN(rc);
2577 }
2578
2579 static int
2580 echo_client_cancel(struct obd_export *exp, struct obdo *oa)
2581 {
2582         struct echo_device *ed     = obd2echo_dev(exp->exp_obd);
2583         __u64               cookie = oa->o_handle.cookie;
2584
2585         if ((oa->o_valid & OBD_MD_FLHANDLE) == 0)
2586                 return -EINVAL;
2587
2588         CDEBUG(D_INFO, "Cookie is "LPX64"\n", cookie);
2589         return cl_echo_cancel(ed, cookie);
2590 }
2591
2592 static int
2593 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2594                       void *karg, void *uarg)
2595 {
2596         struct obd_device      *obd = exp->exp_obd;
2597         struct echo_device     *ed = obd2echo_dev(obd);
2598         struct echo_client_obd *ec = ed->ed_ec;
2599         struct echo_object     *eco;
2600         struct obd_ioctl_data  *data = karg;
2601         struct obd_trans_info   dummy_oti;
2602         struct lu_env          *env;
2603         struct oti_req_ack_lock *ack_lock;
2604         struct obdo            *oa;
2605         struct lu_fid           fid;
2606         int                     rw = OBD_BRW_READ;
2607         int                     rc = 0;
2608         int                     i;
2609         ENTRY;
2610
2611         memset(&dummy_oti, 0, sizeof(dummy_oti));
2612
2613         oa = &data->ioc_obdo1;
2614         if (!(oa->o_valid & OBD_MD_FLGROUP)) {
2615                 oa->o_valid |= OBD_MD_FLGROUP;
2616                 oa->o_seq = FID_SEQ_ECHO;
2617         }
2618
2619         /* This FID is unpacked just for validation at this point */
2620         rc = fid_ostid_unpack(&fid, &oa->o_oi, 0);
2621         if (rc < 0)
2622                 RETURN(rc);
2623
2624         OBD_ALLOC_PTR(env);
2625         if (env == NULL)
2626                 RETURN(-ENOMEM);
2627
2628         rc = lu_env_init(env, LCT_DT_THREAD);
2629         if (rc)
2630                 GOTO(out, rc = -ENOMEM);
2631
2632         switch (cmd) {
2633         case OBD_IOC_CREATE:                    /* may create echo object */
2634                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2635                         GOTO (out, rc = -EPERM);
2636
2637                 rc = echo_create_object(env, ed, 1, oa, data->ioc_pbuf1,
2638                                         data->ioc_plen1, &dummy_oti);
2639                 GOTO(out, rc);
2640
2641         case OBD_IOC_ECHO_MD: {
2642                 int count;
2643                 int cmd;
2644                 char *dir = NULL;
2645                 int dirlen;
2646                 __u64 id;
2647
2648                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2649                         GOTO(out, rc = -EPERM);
2650
2651                 count = data->ioc_count;
2652                 cmd = data->ioc_command;
2653
2654                 id = data->ioc_obdo2.o_id;
2655
2656                 dirlen = data->ioc_plen1;
2657                 OBD_ALLOC(dir, dirlen + 1);
2658                 if (dir == NULL)
2659                         GOTO(out, rc = -ENOMEM);
2660
2661                 if (cfs_copy_from_user(dir, data->ioc_pbuf1, dirlen)) {
2662                         OBD_FREE(dir, data->ioc_plen1 + 1);
2663                         GOTO(out, rc = -EFAULT);
2664                 }
2665
2666                 rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data);
2667                 OBD_FREE(dir, dirlen + 1);
2668                 GOTO(out, rc);
2669         }
2670         case OBD_IOC_ECHO_ALLOC_SEQ: {
2671                 struct lu_env   *cl_env;
2672                 int              refcheck;
2673                 __u64            seq;
2674                 int              max_count;
2675
2676                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2677                         GOTO(out, rc = -EPERM);
2678
2679                 cl_env = cl_env_get(&refcheck);
2680                 if (IS_ERR(cl_env))
2681                         GOTO(out, rc = PTR_ERR(cl_env));
2682
2683                 rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG,
2684                                             ECHO_MD_SES_TAG);
2685                 if (rc != 0) {
2686                         cl_env_put(cl_env, &refcheck);
2687                         GOTO(out, rc);
2688                 }
2689
2690                 rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq);
2691                 cl_env_put(cl_env, &refcheck);
2692                 if (rc < 0) {
2693                         CERROR("%s: Can not alloc seq: rc = %d\n",
2694                                obd->obd_name, rc);
2695                         GOTO(out, rc);
2696                 }
2697
2698                 if (cfs_copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1))
2699                         return -EFAULT;
2700
2701                 max_count = LUSTRE_SEQ_MAX_WIDTH;
2702                 if (cfs_copy_to_user(data->ioc_pbuf2, &max_count,
2703                                      data->ioc_plen2))
2704                         return -EFAULT;
2705                 GOTO(out, rc);
2706         }
2707         case OBD_IOC_DESTROY:
2708                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2709                         GOTO (out, rc = -EPERM);
2710
2711                 rc = echo_get_object(&eco, ed, oa);
2712                 if (rc == 0) {
2713                         rc = obd_destroy(env, ec->ec_exp, oa, eco->eo_lsm,
2714                                          &dummy_oti, NULL, NULL);
2715                         if (rc == 0)
2716                                 eco->eo_deleted = 1;
2717                         echo_put_object(eco);
2718                 }
2719                 GOTO(out, rc);
2720
2721         case OBD_IOC_GETATTR:
2722                 rc = echo_get_object(&eco, ed, oa);
2723                 if (rc == 0) {
2724                         struct obd_info oinfo = { { { 0 } } };
2725                         oinfo.oi_md = eco->eo_lsm;
2726                         oinfo.oi_oa = oa;
2727                         rc = obd_getattr(env, ec->ec_exp, &oinfo);
2728                         echo_put_object(eco);
2729                 }
2730                 GOTO(out, rc);
2731
2732         case OBD_IOC_SETATTR:
2733                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2734                         GOTO (out, rc = -EPERM);
2735
2736                 rc = echo_get_object(&eco, ed, oa);
2737                 if (rc == 0) {
2738                         struct obd_info oinfo = { { { 0 } } };
2739                         oinfo.oi_oa = oa;
2740                         oinfo.oi_md = eco->eo_lsm;
2741
2742                         rc = obd_setattr(env, ec->ec_exp, &oinfo, NULL);
2743                         echo_put_object(eco);
2744                 }
2745                 GOTO(out, rc);
2746
2747         case OBD_IOC_BRW_WRITE:
2748                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2749                         GOTO (out, rc = -EPERM);
2750
2751                 rw = OBD_BRW_WRITE;
2752                 /* fall through */
2753         case OBD_IOC_BRW_READ:
2754                 rc = echo_client_brw_ioctl(rw, exp, data);
2755                 GOTO(out, rc);
2756
2757         case ECHO_IOC_GET_STRIPE:
2758                 rc = echo_get_object(&eco, ed, oa);
2759                 if (rc == 0) {
2760                         rc = echo_copyout_lsm(eco->eo_lsm, data->ioc_pbuf1,
2761                                               data->ioc_plen1);
2762                         echo_put_object(eco);
2763                 }
2764                 GOTO(out, rc);
2765
2766         case ECHO_IOC_SET_STRIPE:
2767                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2768                         GOTO (out, rc = -EPERM);
2769
2770                 if (data->ioc_pbuf1 == NULL) {  /* unset */
2771                         rc = echo_get_object(&eco, ed, oa);
2772                         if (rc == 0) {
2773                                 eco->eo_deleted = 1;
2774                                 echo_put_object(eco);
2775                         }
2776                 } else {
2777                         rc = echo_create_object(env, ed, 0, oa,
2778                                                 data->ioc_pbuf1,
2779                                                 data->ioc_plen1, &dummy_oti);
2780                 }
2781                 GOTO (out, rc);
2782
2783         case ECHO_IOC_ENQUEUE:
2784                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2785                         GOTO (out, rc = -EPERM);
2786
2787                 rc = echo_client_enqueue(exp, oa,
2788                                          data->ioc_conn1, /* lock mode */
2789                                          data->ioc_offset,
2790                                          data->ioc_count);/*extent*/
2791                 GOTO (out, rc);
2792
2793         case ECHO_IOC_CANCEL:
2794                 rc = echo_client_cancel(exp, oa);
2795                 GOTO (out, rc);
2796
2797         default:
2798                 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
2799                 GOTO (out, rc = -ENOTTY);
2800         }
2801
2802         EXIT;
2803 out:
2804         lu_env_fini(env);
2805         OBD_FREE_PTR(env);
2806
2807         /* XXX this should be in a helper also called by target_send_reply */
2808         for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4;
2809              i++, ack_lock++) {
2810                 if (!ack_lock->mode)
2811                         break;
2812                 ldlm_lock_decref(&ack_lock->lock, ack_lock->mode);
2813         }
2814
2815         return rc;
2816 }
2817
2818 static int echo_client_setup(const struct lu_env *env,
2819                              struct obd_device *obddev, struct lustre_cfg *lcfg)
2820 {
2821         struct echo_client_obd *ec = &obddev->u.echo_client;
2822         struct obd_device *tgt;
2823         struct obd_uuid echo_uuid = { "ECHO_UUID" };
2824         struct obd_connect_data *ocd = NULL;
2825         int rc;
2826         ENTRY;
2827
2828         if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2829                 CERROR("requires a TARGET OBD name\n");
2830                 RETURN(-EINVAL);
2831         }
2832
2833         tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
2834         if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
2835                 CERROR("device not attached or not set up (%s)\n",
2836                        lustre_cfg_string(lcfg, 1));
2837                 RETURN(-EINVAL);
2838         }
2839
2840         cfs_spin_lock_init (&ec->ec_lock);
2841         CFS_INIT_LIST_HEAD (&ec->ec_objects);
2842         CFS_INIT_LIST_HEAD (&ec->ec_locks);
2843         ec->ec_unique = 0;
2844         ec->ec_nstripes = 0;
2845
2846         if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
2847                 lu_context_tags_update(ECHO_MD_CTX_TAG);
2848                 lu_session_tags_update(ECHO_MD_SES_TAG);
2849                 RETURN(0);
2850         }
2851
2852         OBD_ALLOC(ocd, sizeof(*ocd));
2853         if (ocd == NULL) {
2854                 CERROR("Can't alloc ocd connecting to %s\n",
2855                        lustre_cfg_string(lcfg, 1));
2856                 return -ENOMEM;
2857         }
2858
2859         ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL |
2860                                  OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 |
2861                                  OBD_CONNECT_64BITHASH;
2862         ocd->ocd_version = LUSTRE_VERSION_CODE;
2863         ocd->ocd_group = FID_SEQ_ECHO;
2864
2865         rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL);
2866         if (rc == 0) {
2867                 /* Turn off pinger because it connects to tgt obd directly. */
2868                 cfs_spin_lock(&tgt->obd_dev_lock);
2869                 cfs_list_del_init(&ec->ec_exp->exp_obd_chain_timed);
2870                 cfs_spin_unlock(&tgt->obd_dev_lock);
2871         }
2872
2873         OBD_FREE(ocd, sizeof(*ocd));
2874
2875         if (rc != 0) {
2876                 CERROR("fail to connect to device %s\n",
2877                        lustre_cfg_string(lcfg, 1));
2878                 return (rc);
2879         }
2880
2881         RETURN(rc);
2882 }
2883
2884 static int echo_client_cleanup(struct obd_device *obddev)
2885 {
2886         struct echo_device *ed = obd2echo_dev(obddev);
2887         struct echo_client_obd *ec = &obddev->u.echo_client;
2888         int rc;
2889         ENTRY;
2890
2891         /*Do nothing for Metadata echo client*/
2892         if (ed == NULL )
2893                 RETURN(0);
2894
2895         if (ed->ed_next_ismd) {
2896                 lu_context_tags_clear(ECHO_MD_CTX_TAG);
2897                 lu_session_tags_clear(ECHO_MD_SES_TAG);
2898                 RETURN(0);
2899         }
2900
2901         if (!cfs_list_empty(&obddev->obd_exports)) {
2902                 CERROR("still has clients!\n");
2903                 RETURN(-EBUSY);
2904         }
2905
2906         LASSERT(cfs_atomic_read(&ec->ec_exp->exp_refcount) > 0);
2907         rc = obd_disconnect(ec->ec_exp);
2908         if (rc != 0)
2909                 CERROR("fail to disconnect device: %d\n", rc);
2910
2911         RETURN(rc);
2912 }
2913
2914 static int echo_client_connect(const struct lu_env *env,
2915                                struct obd_export **exp,
2916                                struct obd_device *src, struct obd_uuid *cluuid,
2917                                struct obd_connect_data *data, void *localdata)
2918 {
2919         int                rc;
2920         struct lustre_handle conn = { 0 };
2921
2922         ENTRY;
2923         rc = class_connect(&conn, src, cluuid);
2924         if (rc == 0) {
2925                 *exp = class_conn2export(&conn);
2926         }
2927
2928         RETURN (rc);
2929 }
2930
2931 static int echo_client_disconnect(struct obd_export *exp)
2932 {
2933 #if 0
2934         struct obd_device      *obd;
2935         struct echo_client_obd *ec;
2936         struct ec_lock         *ecl;
2937 #endif
2938         int                     rc;
2939         ENTRY;
2940
2941         if (exp == NULL)
2942                 GOTO(out, rc = -EINVAL);
2943
2944 #if 0
2945         obd = exp->exp_obd;
2946         ec = &obd->u.echo_client;
2947
2948         /* no more contention on export's lock list */
2949         while (!cfs_list_empty (&exp->exp_ec_data.eced_locks)) {
2950                 ecl = cfs_list_entry (exp->exp_ec_data.eced_locks.next,
2951                                       struct ec_lock, ecl_exp_chain);
2952                 cfs_list_del (&ecl->ecl_exp_chain);
2953
2954                 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm,
2955                                  ecl->ecl_mode, &ecl->ecl_lock_handle);
2956
2957                 CDEBUG (D_INFO, "Cancel lock on object "LPX64" on disconnect "
2958                         "(%d)\n", ecl->ecl_object->eco_id, rc);
2959
2960                 echo_put_object (ecl->ecl_object);
2961                 OBD_FREE (ecl, sizeof (*ecl));
2962         }
2963 #endif
2964
2965         rc = class_disconnect(exp);
2966         GOTO(out, rc);
2967  out:
2968         return rc;
2969 }
2970
2971 static struct obd_ops echo_client_obd_ops = {
2972         .o_owner       = THIS_MODULE,
2973
2974 #if 0
2975         .o_setup       = echo_client_setup,
2976         .o_cleanup     = echo_client_cleanup,
2977 #endif
2978
2979         .o_iocontrol   = echo_client_iocontrol,
2980         .o_connect     = echo_client_connect,
2981         .o_disconnect  = echo_client_disconnect
2982 };
2983
2984 int echo_client_init(void)
2985 {
2986         struct lprocfs_static_vars lvars = { 0 };
2987         int rc;
2988
2989         lprocfs_echo_init_vars(&lvars);
2990
2991         rc = lu_kmem_init(echo_caches);
2992         if (rc == 0) {
2993                 rc = class_register_type(&echo_client_obd_ops, NULL,
2994                                          lvars.module_vars,
2995                                          LUSTRE_ECHO_CLIENT_NAME,
2996                                          &echo_device_type);
2997                 if (rc)
2998                         lu_kmem_fini(echo_caches);
2999         }
3000         return rc;
3001 }
3002
3003 void echo_client_exit(void)
3004 {
3005         class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
3006         lu_kmem_fini(echo_caches);
3007 }
3008
3009 #ifdef __KERNEL__
3010 static int __init obdecho_init(void)
3011 {
3012         struct lprocfs_static_vars lvars;
3013         int rc;
3014
3015         ENTRY;
3016         LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
3017
3018         LASSERT(CFS_PAGE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
3019
3020         lprocfs_echo_init_vars(&lvars);
3021
3022 # ifdef HAVE_SERVER_SUPPORT
3023         rc = echo_persistent_pages_init();
3024         if (rc != 0)
3025                 goto failed_0;
3026
3027         rc = class_register_type(&echo_obd_ops, NULL, lvars.module_vars,
3028                                  LUSTRE_ECHO_NAME, NULL);
3029         if (rc != 0)
3030                 goto failed_1;
3031 # endif
3032
3033         rc = echo_client_init();
3034
3035 # ifdef HAVE_SERVER_SUPPORT
3036         if (rc == 0)
3037                 RETURN(0);
3038
3039         class_unregister_type(LUSTRE_ECHO_NAME);
3040 failed_1:
3041         echo_persistent_pages_fini();
3042 failed_0:
3043 # endif
3044         RETURN(rc);
3045 }
3046
3047 static void /*__exit*/ obdecho_exit(void)
3048 {
3049         echo_client_exit();
3050
3051 # ifdef HAVE_SERVER_SUPPORT
3052         class_unregister_type(LUSTRE_ECHO_NAME);
3053         echo_persistent_pages_fini();
3054 # endif
3055 }
3056
3057 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3058 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
3059 MODULE_LICENSE("GPL");
3060
3061 cfs_module(obdecho, LUSTRE_VERSION_STRING, obdecho_init, obdecho_exit);
3062 #endif /* __KERNEL__ */
3063
3064 /** @} echo_client */