drivers/staging/lustre/lustre/obdecho/echo_client.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_ECHO
  38 #include <linux/libcfs/libcfs.h>
  39
  40 #include <obd.h>
  41 #include <obd_support.h>
  42 #include <obd_class.h>
  43 #include <lustre_debug.h>
  44 #include <lprocfs_status.h>
  45 #include <cl_object.h>
  46 #include <md_object.h>
  47 #include <lustre_fid.h>
  48 #include <lustre_acl.h>
  49 #include <lustre_net.h>
  50
  51 #include "echo_internal.h"
  52
  53 /** \defgroup echo_client Echo Client
  54  * @{
  55  */
  56
  57 struct echo_device {
  58         struct cl_device        ed_cl;
  59         struct echo_client_obd *ed_ec;
  60
  61         struct cl_site    ed_site_myself;
  62         struct cl_site   *ed_site;
  63         struct lu_device       *ed_next;
  64         int                  ed_next_islov;
  65         int                  ed_next_ismd;
  66         struct lu_client_seq   *ed_cl_seq;
  67 };
  68
  69 struct echo_object {
  70         struct cl_object        eo_cl;
  71         struct cl_object_header eo_hdr;
  72
  73         struct echo_device     *eo_dev;
  74         struct list_head              eo_obj_chain;
  75         struct lov_stripe_md   *eo_lsm;
  76         atomic_t            eo_npages;
  77         int                  eo_deleted;
  78 };
  79
  80 struct echo_object_conf {
  81         struct cl_object_conf  eoc_cl;
  82         struct lov_stripe_md **eoc_md;
  83 };
  84
  85 struct echo_page {
  86         struct cl_page_slice   ep_cl;
  87         struct mutex            ep_lock;
  88         struct page         *ep_vmpage;
  89 };
  90
  91 struct echo_lock {
  92         struct cl_lock_slice   el_cl;
  93         struct list_head             el_chain;
  94         struct echo_object    *el_object;
  95         __u64             el_cookie;
  96         atomic_t           el_refcount;
  97 };
  98
  99 struct echo_io {
 100         struct cl_io_slice     ei_cl;
 101 };
 102
 103 #if 0
 104 struct echo_req {
 105         struct cl_req_slice er_cl;
 106 };
 107 #endif
 108
 109 static int echo_client_setup(const struct lu_env *env,
 110                              struct obd_device *obddev,
 111                              struct lustre_cfg *lcfg);
 112 static int echo_client_cleanup(struct obd_device *obddev);
 113
 114
 115 /** \defgroup echo_helpers Helper functions
 116  * @{
 117  */
 118 static inline struct echo_device *cl2echo_dev(const struct cl_device *dev)
 119 {
 120         return container_of0(dev, struct echo_device, ed_cl);
 121 }
 122
 123 static inline struct cl_device *echo_dev2cl(struct echo_device *d)
 124 {
 125         return &d->ed_cl;
 126 }
 127
 128 static inline struct echo_device *obd2echo_dev(const struct obd_device *obd)
 129 {
 130         return cl2echo_dev(lu2cl_dev(obd->obd_lu_dev));
 131 }
 132
 133 static inline struct cl_object *echo_obj2cl(struct echo_object *eco)
 134 {
 135         return &eco->eo_cl;
 136 }
 137
 138 static inline struct echo_object *cl2echo_obj(const struct cl_object *o)
 139 {
 140         return container_of(o, struct echo_object, eo_cl);
 141 }
 142
 143 static inline struct echo_page *cl2echo_page(const struct cl_page_slice *s)
 144 {
 145         return container_of(s, struct echo_page, ep_cl);
 146 }
 147
 148 static inline struct echo_lock *cl2echo_lock(const struct cl_lock_slice *s)
 149 {
 150         return container_of(s, struct echo_lock, el_cl);
 151 }
 152
 153 static inline struct cl_lock *echo_lock2cl(const struct echo_lock *ecl)
 154 {
 155         return ecl->el_cl.cls_lock;
 156 }
 157
 158 static struct lu_context_key echo_thread_key;
 159 static inline struct echo_thread_info *echo_env_info(const struct lu_env *env)
 160 {
 161         struct echo_thread_info *info;
 162         info = lu_context_key_get(&env->le_ctx, &echo_thread_key);
 163         LASSERT(info != NULL);
 164         return info;
 165 }
 166
 167 static inline
 168 struct echo_object_conf *cl2echo_conf(const struct cl_object_conf *c)
 169 {
 170         return container_of(c, struct echo_object_conf, eoc_cl);
 171 }
 172
 173 /** @} echo_helpers */
 174
 175 static struct echo_object *cl_echo_object_find(struct echo_device *d,
 176                                                struct lov_stripe_md **lsm);
 177 static int cl_echo_object_put(struct echo_object *eco);
 178 static int cl_echo_enqueue   (struct echo_object *eco, obd_off start,
 179                               obd_off end, int mode, __u64 *cookie);
 180 static int cl_echo_cancel    (struct echo_device *d, __u64 cookie);
 181 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
 182                               struct page **pages, int npages, int async);
 183
 184 static struct echo_thread_info *echo_env_info(const struct lu_env *env);
 185
 186 struct echo_thread_info {
 187         struct echo_object_conf eti_conf;
 188         struct lustre_md        eti_md;
 189
 190         struct cl_2queue        eti_queue;
 191         struct cl_io        eti_io;
 192         struct cl_lock_descr    eti_descr;
 193         struct lu_fid      eti_fid;
 194         struct lu_fid           eti_fid2;
 195         struct md_op_spec       eti_spec;
 196         struct lov_mds_md_v3    eti_lmm;
 197         struct lov_user_md_v3   eti_lum;
 198         struct md_attr    eti_ma;
 199         struct lu_name    eti_lname;
 200         /* per-thread values, can be re-used */
 201         void                    *eti_big_lmm;
 202         int                     eti_big_lmmsize;
 203         char                eti_name[20];
 204         struct lu_buf      eti_buf;
 205         char                eti_xattr_buf[LUSTRE_POSIX_ACL_MAX_SIZE];
 206 };
 207
 208 /* No session used right now */
 209 struct echo_session_info {
 210         unsigned long dummy;
 211 };
 212
 213 static struct kmem_cache *echo_lock_kmem;
 214 static struct kmem_cache *echo_object_kmem;
 215 static struct kmem_cache *echo_thread_kmem;
 216 static struct kmem_cache *echo_session_kmem;
 217 //static struct kmem_cache *echo_req_kmem;
 218
 219 static struct lu_kmem_descr echo_caches[] = {
 220         {
 221                 .ckd_cache = &echo_lock_kmem,
 222                 .ckd_name  = "echo_lock_kmem",
 223                 .ckd_size  = sizeof (struct echo_lock)
 224         },
 225         {
 226                 .ckd_cache = &echo_object_kmem,
 227                 .ckd_name  = "echo_object_kmem",
 228                 .ckd_size  = sizeof (struct echo_object)
 229         },
 230         {
 231                 .ckd_cache = &echo_thread_kmem,
 232                 .ckd_name  = "echo_thread_kmem",
 233                 .ckd_size  = sizeof (struct echo_thread_info)
 234         },
 235         {
 236                 .ckd_cache = &echo_session_kmem,
 237                 .ckd_name  = "echo_session_kmem",
 238                 .ckd_size  = sizeof (struct echo_session_info)
 239         },
 240 #if 0
 241         {
 242                 .ckd_cache = &echo_req_kmem,
 243                 .ckd_name  = "echo_req_kmem",
 244                 .ckd_size  = sizeof (struct echo_req)
 245         },
 246 #endif
 247         {
 248                 .ckd_cache = NULL
 249         }
 250 };
 251
 252 /** \defgroup echo_page Page operations
 253  *
 254  * Echo page operations.
 255  *
 256  * @{
 257  */
 258 static struct page *echo_page_vmpage(const struct lu_env *env,
 259                                     const struct cl_page_slice *slice)
 260 {
 261         return cl2echo_page(slice)->ep_vmpage;
 262 }
 263
 264 static int echo_page_own(const struct lu_env *env,
 265                          const struct cl_page_slice *slice,
 266                          struct cl_io *io, int nonblock)
 267 {
 268         struct echo_page *ep = cl2echo_page(slice);
 269
 270         if (!nonblock)
 271                 mutex_lock(&ep->ep_lock);
 272         else if (!mutex_trylock(&ep->ep_lock))
 273                 return -EAGAIN;
 274         return 0;
 275 }
 276
 277 static void echo_page_disown(const struct lu_env *env,
 278                              const struct cl_page_slice *slice,
 279                              struct cl_io *io)
 280 {
 281         struct echo_page *ep = cl2echo_page(slice);
 282
 283         LASSERT(mutex_is_locked(&ep->ep_lock));
 284         mutex_unlock(&ep->ep_lock);
 285 }
 286
 287 static void echo_page_discard(const struct lu_env *env,
 288                               const struct cl_page_slice *slice,
 289                               struct cl_io *unused)
 290 {
 291         cl_page_delete(env, slice->cpl_page);
 292 }
 293
 294 static int echo_page_is_vmlocked(const struct lu_env *env,
 295                                  const struct cl_page_slice *slice)
 296 {
 297         if (mutex_is_locked(&cl2echo_page(slice)->ep_lock))
 298                 return -EBUSY;
 299         return -ENODATA;
 300 }
 301
 302 static void echo_page_completion(const struct lu_env *env,
 303                                  const struct cl_page_slice *slice,
 304                                  int ioret)
 305 {
 306         LASSERT(slice->cpl_page->cp_sync_io != NULL);
 307 }
 308
 309 static void echo_page_fini(const struct lu_env *env,
 310                            struct cl_page_slice *slice)
 311 {
 312         struct echo_page *ep    = cl2echo_page(slice);
 313         struct echo_object *eco = cl2echo_obj(slice->cpl_obj);
 314         struct page *vmpage      = ep->ep_vmpage;
 315
 316         atomic_dec(&eco->eo_npages);
 317         page_cache_release(vmpage);
 318 }
 319
 320 static int echo_page_prep(const struct lu_env *env,
 321                           const struct cl_page_slice *slice,
 322                           struct cl_io *unused)
 323 {
 324         return 0;
 325 }
 326
 327 static int echo_page_print(const struct lu_env *env,
 328                            const struct cl_page_slice *slice,
 329                            void *cookie, lu_printer_t printer)
 330 {
 331         struct echo_page *ep = cl2echo_page(slice);
 332
 333         (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p %d vm@%p\n",
 334                    ep, mutex_is_locked(&ep->ep_lock), ep->ep_vmpage);
 335         return 0;
 336 }
 337
 338 static const struct cl_page_operations echo_page_ops = {
 339         .cpo_own           = echo_page_own,
 340         .cpo_disown     = echo_page_disown,
 341         .cpo_discard       = echo_page_discard,
 342         .cpo_vmpage     = echo_page_vmpage,
 343         .cpo_fini         = echo_page_fini,
 344         .cpo_print       = echo_page_print,
 345         .cpo_is_vmlocked   = echo_page_is_vmlocked,
 346         .io = {
 347                 [CRT_READ] = {
 348                         .cpo_prep       = echo_page_prep,
 349                         .cpo_completion  = echo_page_completion,
 350                 },
 351                 [CRT_WRITE] = {
 352                         .cpo_prep       = echo_page_prep,
 353                         .cpo_completion  = echo_page_completion,
 354                 }
 355         }
 356 };
 357 /** @} echo_page */
 358
 359 /** \defgroup echo_lock Locking
 360  *
 361  * echo lock operations
 362  *
 363  * @{
 364  */
 365 static void echo_lock_fini(const struct lu_env *env,
 366                            struct cl_lock_slice *slice)
 367 {
 368         struct echo_lock *ecl = cl2echo_lock(slice);
 369
 370         LASSERT(list_empty(&ecl->el_chain));
 371         OBD_SLAB_FREE_PTR(ecl, echo_lock_kmem);
 372 }
 373
 374 static void echo_lock_delete(const struct lu_env *env,
 375                              const struct cl_lock_slice *slice)
 376 {
 377         struct echo_lock *ecl      = cl2echo_lock(slice);
 378
 379         LASSERT(list_empty(&ecl->el_chain));
 380 }
 381
 382 static int echo_lock_fits_into(const struct lu_env *env,
 383                                const struct cl_lock_slice *slice,
 384                                const struct cl_lock_descr *need,
 385                                const struct cl_io *unused)
 386 {
 387         return 1;
 388 }
 389
 390 static struct cl_lock_operations echo_lock_ops = {
 391         .clo_fini      = echo_lock_fini,
 392         .clo_delete    = echo_lock_delete,
 393         .clo_fits_into = echo_lock_fits_into
 394 };
 395
 396 /** @} echo_lock */
 397
 398 /** \defgroup echo_cl_ops cl_object operations
 399  *
 400  * operations for cl_object
 401  *
 402  * @{
 403  */
 404 static int echo_page_init(const struct lu_env *env, struct cl_object *obj,
 405                         struct cl_page *page, struct page *vmpage)
 406 {
 407         struct echo_page *ep = cl_object_page_slice(obj, page);
 408         struct echo_object *eco = cl2echo_obj(obj);
 409
 410         ep->ep_vmpage = vmpage;
 411         page_cache_get(vmpage);
 412         mutex_init(&ep->ep_lock);
 413         cl_page_slice_add(page, &ep->ep_cl, obj, &echo_page_ops);
 414         atomic_inc(&eco->eo_npages);
 415         return 0;
 416 }
 417
 418 static int echo_io_init(const struct lu_env *env, struct cl_object *obj,
 419                         struct cl_io *io)
 420 {
 421         return 0;
 422 }
 423
 424 static int echo_lock_init(const struct lu_env *env,
 425                           struct cl_object *obj, struct cl_lock *lock,
 426                           const struct cl_io *unused)
 427 {
 428         struct echo_lock *el;
 429
 430         OBD_SLAB_ALLOC_PTR_GFP(el, echo_lock_kmem, GFP_NOFS);
 431         if (el != NULL) {
 432                 cl_lock_slice_add(lock, &el->el_cl, obj, &echo_lock_ops);
 433                 el->el_object = cl2echo_obj(obj);
 434                 INIT_LIST_HEAD(&el->el_chain);
 435                 atomic_set(&el->el_refcount, 0);
 436         }
 437         return el == NULL ? -ENOMEM : 0;
 438 }
 439
 440 static int echo_conf_set(const struct lu_env *env, struct cl_object *obj,
 441                          const struct cl_object_conf *conf)
 442 {
 443         return 0;
 444 }
 445
 446 static const struct cl_object_operations echo_cl_obj_ops = {
 447         .coo_page_init = echo_page_init,
 448         .coo_lock_init = echo_lock_init,
 449         .coo_io_init   = echo_io_init,
 450         .coo_conf_set  = echo_conf_set
 451 };
 452 /** @} echo_cl_ops */
 453
 454 /** \defgroup echo_lu_ops lu_object operations
 455  *
 456  * operations for echo lu object.
 457  *
 458  * @{
 459  */
 460 static int echo_object_init(const struct lu_env *env, struct lu_object *obj,
 461                             const struct lu_object_conf *conf)
 462 {
 463         struct echo_device *ed   = cl2echo_dev(lu2cl_dev(obj->lo_dev));
 464         struct echo_client_obd *ec     = ed->ed_ec;
 465         struct echo_object *eco = cl2echo_obj(lu2cl(obj));
 466
 467         if (ed->ed_next) {
 468                 struct lu_object  *below;
 469                 struct lu_device  *under;
 470
 471                 under = ed->ed_next;
 472                 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header,
 473                                                         under);
 474                 if (below == NULL)
 475                         return -ENOMEM;
 476                 lu_object_add(obj, below);
 477         }
 478
 479         if (!ed->ed_next_ismd) {
 480                 const struct cl_object_conf *cconf = lu2cl_conf(conf);
 481                 struct echo_object_conf *econf = cl2echo_conf(cconf);
 482
 483                 LASSERT(econf->eoc_md);
 484                 eco->eo_lsm = *econf->eoc_md;
 485                 /* clear the lsm pointer so that it won't get freed. */
 486                 *econf->eoc_md = NULL;
 487         } else {
 488                 eco->eo_lsm = NULL;
 489         }
 490
 491         eco->eo_dev = ed;
 492         atomic_set(&eco->eo_npages, 0);
 493         cl_object_page_init(lu2cl(obj), sizeof(struct echo_page));
 494
 495         spin_lock(&ec->ec_lock);
 496         list_add_tail(&eco->eo_obj_chain, &ec->ec_objects);
 497         spin_unlock(&ec->ec_lock);
 498
 499         return 0;
 500 }
 501
 502 /* taken from osc_unpackmd() */
 503 static int echo_alloc_memmd(struct echo_device *ed,
 504                             struct lov_stripe_md **lsmp)
 505 {
 506         int lsm_size;
 507
 508         /* If export is lov/osc then use their obd method */
 509         if (ed->ed_next != NULL)
 510                 return obd_alloc_memmd(ed->ed_ec->ec_exp, lsmp);
 511         /* OFD has no unpackmd method, do everything here */
 512         lsm_size = lov_stripe_md_size(1);
 513
 514         LASSERT(*lsmp == NULL);
 515         OBD_ALLOC(*lsmp, lsm_size);
 516         if (*lsmp == NULL)
 517                 return -ENOMEM;
 518
 519         OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
 520         if ((*lsmp)->lsm_oinfo[0] == NULL) {
 521                 OBD_FREE(*lsmp, lsm_size);
 522                 return -ENOMEM;
 523         }
 524
 525         loi_init((*lsmp)->lsm_oinfo[0]);
 526         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
 527         ostid_set_seq_echo(&(*lsmp)->lsm_oi);
 528
 529         return lsm_size;
 530 }
 531
 532 static int echo_free_memmd(struct echo_device *ed, struct lov_stripe_md **lsmp)
 533 {
 534         int lsm_size;
 535
 536         /* If export is lov/osc then use their obd method */
 537         if (ed->ed_next != NULL)
 538                 return obd_free_memmd(ed->ed_ec->ec_exp, lsmp);
 539         /* OFD has no unpackmd method, do everything here */
 540         lsm_size = lov_stripe_md_size(1);
 541
 542         LASSERT(*lsmp != NULL);
 543         OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
 544         OBD_FREE(*lsmp, lsm_size);
 545         *lsmp = NULL;
 546         return 0;
 547 }
 548
 549 static void echo_object_free(const struct lu_env *env, struct lu_object *obj)
 550 {
 551         struct echo_object *eco    = cl2echo_obj(lu2cl(obj));
 552         struct echo_client_obd *ec = eco->eo_dev->ed_ec;
 553
 554         LASSERT(atomic_read(&eco->eo_npages) == 0);
 555
 556         spin_lock(&ec->ec_lock);
 557         list_del_init(&eco->eo_obj_chain);
 558         spin_unlock(&ec->ec_lock);
 559
 560         lu_object_fini(obj);
 561         lu_object_header_fini(obj->lo_header);
 562
 563         if (eco->eo_lsm)
 564                 echo_free_memmd(eco->eo_dev, &eco->eo_lsm);
 565         OBD_SLAB_FREE_PTR(eco, echo_object_kmem);
 566 }
 567
 568 static int echo_object_print(const struct lu_env *env, void *cookie,
 569                             lu_printer_t p, const struct lu_object *o)
 570 {
 571         struct echo_object *obj = cl2echo_obj(lu2cl(o));
 572
 573         return (*p)(env, cookie, "echoclient-object@%p", obj);
 574 }
 575
 576 static const struct lu_object_operations echo_lu_obj_ops = {
 577         .loo_object_init      = echo_object_init,
 578         .loo_object_delete    = NULL,
 579         .loo_object_release   = NULL,
 580         .loo_object_free      = echo_object_free,
 581         .loo_object_print     = echo_object_print,
 582         .loo_object_invariant = NULL
 583 };
 584 /** @} echo_lu_ops */
 585
 586 /** \defgroup echo_lu_dev_ops  lu_device operations
 587  *
 588  * Operations for echo lu device.
 589  *
 590  * @{
 591  */
 592 static struct lu_object *echo_object_alloc(const struct lu_env *env,
 593                                            const struct lu_object_header *hdr,
 594                                            struct lu_device *dev)
 595 {
 596         struct echo_object *eco;
 597         struct lu_object *obj = NULL;
 598
 599         /* we're the top dev. */
 600         LASSERT(hdr == NULL);
 601         OBD_SLAB_ALLOC_PTR_GFP(eco, echo_object_kmem, GFP_NOFS);
 602         if (eco != NULL) {
 603                 struct cl_object_header *hdr = &eco->eo_hdr;
 604
 605                 obj = &echo_obj2cl(eco)->co_lu;
 606                 cl_object_header_init(hdr);
 607                 lu_object_init(obj, &hdr->coh_lu, dev);
 608                 lu_object_add_top(&hdr->coh_lu, obj);
 609
 610                 eco->eo_cl.co_ops = &echo_cl_obj_ops;
 611                 obj->lo_ops       = &echo_lu_obj_ops;
 612         }
 613         return obj;
 614 }
 615
 616 static struct lu_device_operations echo_device_lu_ops = {
 617         .ldo_object_alloc   = echo_object_alloc,
 618 };
 619
 620 /** @} echo_lu_dev_ops */
 621
 622 static struct cl_device_operations echo_device_cl_ops = {
 623 };
 624
 625 /** \defgroup echo_init Setup and teardown
 626  *
 627  * Init and fini functions for echo client.
 628  *
 629  * @{
 630  */
 631 static int echo_site_init(const struct lu_env *env, struct echo_device *ed)
 632 {
 633         struct cl_site *site = &ed->ed_site_myself;
 634         int rc;
 635
 636         /* initialize site */
 637         rc = cl_site_init(site, &ed->ed_cl);
 638         if (rc) {
 639                 CERROR("Cannot initialize site for echo client(%d)\n", rc);
 640                 return rc;
 641         }
 642
 643         rc = lu_site_init_finish(&site->cs_lu);
 644         if (rc)
 645                 return rc;
 646
 647         ed->ed_site = site;
 648         return 0;
 649 }
 650
 651 static void echo_site_fini(const struct lu_env *env, struct echo_device *ed)
 652 {
 653         if (ed->ed_site) {
 654                 if (!ed->ed_next_ismd)
 655                         cl_site_fini(ed->ed_site);
 656                 ed->ed_site = NULL;
 657         }
 658 }
 659
 660 static void *echo_thread_key_init(const struct lu_context *ctx,
 661                           struct lu_context_key *key)
 662 {
 663         struct echo_thread_info *info;
 664
 665         OBD_SLAB_ALLOC_PTR_GFP(info, echo_thread_kmem, GFP_NOFS);
 666         if (info == NULL)
 667                 info = ERR_PTR(-ENOMEM);
 668         return info;
 669 }
 670
 671 static void echo_thread_key_fini(const struct lu_context *ctx,
 672                          struct lu_context_key *key, void *data)
 673 {
 674         struct echo_thread_info *info = data;
 675         OBD_SLAB_FREE_PTR(info, echo_thread_kmem);
 676 }
 677
 678 static void echo_thread_key_exit(const struct lu_context *ctx,
 679                          struct lu_context_key *key, void *data)
 680 {
 681 }
 682
 683 static struct lu_context_key echo_thread_key = {
 684         .lct_tags = LCT_CL_THREAD,
 685         .lct_init = echo_thread_key_init,
 686         .lct_fini = echo_thread_key_fini,
 687         .lct_exit = echo_thread_key_exit
 688 };
 689
 690 static void *echo_session_key_init(const struct lu_context *ctx,
 691                                   struct lu_context_key *key)
 692 {
 693         struct echo_session_info *session;
 694
 695         OBD_SLAB_ALLOC_PTR_GFP(session, echo_session_kmem, GFP_NOFS);
 696         if (session == NULL)
 697                 session = ERR_PTR(-ENOMEM);
 698         return session;
 699 }
 700
 701 static void echo_session_key_fini(const struct lu_context *ctx,
 702                                  struct lu_context_key *key, void *data)
 703 {
 704         struct echo_session_info *session = data;
 705         OBD_SLAB_FREE_PTR(session, echo_session_kmem);
 706 }
 707
 708 static void echo_session_key_exit(const struct lu_context *ctx,
 709                                  struct lu_context_key *key, void *data)
 710 {
 711 }
 712
 713 static struct lu_context_key echo_session_key = {
 714         .lct_tags = LCT_SESSION,
 715         .lct_init = echo_session_key_init,
 716         .lct_fini = echo_session_key_fini,
 717         .lct_exit = echo_session_key_exit
 718 };
 719
 720 LU_TYPE_INIT_FINI(echo, &echo_thread_key, &echo_session_key);
 721
 722 #define ECHO_SEQ_WIDTH 0xffffffff
 723 static int echo_fid_init(struct echo_device *ed, char *obd_name,
 724                          struct seq_server_site *ss)
 725 {
 726         char *prefix;
 727         int rc;
 728
 729         OBD_ALLOC_PTR(ed->ed_cl_seq);
 730         if (ed->ed_cl_seq == NULL)
 731                 return -ENOMEM;
 732
 733         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
 734         if (prefix == NULL)
 735                 GOTO(out_free_seq, rc = -ENOMEM);
 736
 737         snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", obd_name);
 738
 739         /* Init client side sequence-manager */
 740         rc = seq_client_init(ed->ed_cl_seq, NULL,
 741                              LUSTRE_SEQ_METADATA,
 742                              prefix, ss->ss_server_seq);
 743         ed->ed_cl_seq->lcs_width = ECHO_SEQ_WIDTH;
 744         OBD_FREE(prefix, MAX_OBD_NAME + 5);
 745         if (rc)
 746                 GOTO(out_free_seq, rc);
 747
 748         return 0;
 749
 750 out_free_seq:
 751         OBD_FREE_PTR(ed->ed_cl_seq);
 752         ed->ed_cl_seq = NULL;
 753         return rc;
 754 }
 755
 756 static int echo_fid_fini(struct obd_device *obddev)
 757 {
 758         struct echo_device *ed = obd2echo_dev(obddev);
 759
 760         if (ed->ed_cl_seq != NULL) {
 761                 seq_client_fini(ed->ed_cl_seq);
 762                 OBD_FREE_PTR(ed->ed_cl_seq);
 763                 ed->ed_cl_seq = NULL;
 764         }
 765
 766         return 0;
 767 }
 768
 769 static struct lu_device *echo_device_alloc(const struct lu_env *env,
 770                                            struct lu_device_type *t,
 771                                            struct lustre_cfg *cfg)
 772 {
 773         struct lu_device   *next;
 774         struct echo_device *ed;
 775         struct cl_device   *cd;
 776         struct obd_device  *obd = NULL; /* to keep compiler happy */
 777         struct obd_device  *tgt;
 778         const char *tgt_type_name;
 779         int rc;
 780         int cleanup = 0;
 781
 782         OBD_ALLOC_PTR(ed);
 783         if (ed == NULL)
 784                 GOTO(out, rc = -ENOMEM);
 785
 786         cleanup = 1;
 787         cd = &ed->ed_cl;
 788         rc = cl_device_init(cd, t);
 789         if (rc)
 790                 GOTO(out, rc);
 791
 792         cd->cd_lu_dev.ld_ops = &echo_device_lu_ops;
 793         cd->cd_ops = &echo_device_cl_ops;
 794
 795         cleanup = 2;
 796         obd = class_name2obd(lustre_cfg_string(cfg, 0));
 797         LASSERT(obd != NULL);
 798         LASSERT(env != NULL);
 799
 800         tgt = class_name2obd(lustre_cfg_string(cfg, 1));
 801         if (tgt == NULL) {
 802                 CERROR("Can not find tgt device %s\n",
 803                         lustre_cfg_string(cfg, 1));
 804                 GOTO(out, rc = -ENODEV);
 805         }
 806
 807         next = tgt->obd_lu_dev;
 808         if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
 809                 ed->ed_next_ismd = 1;
 810         } else {
 811                 ed->ed_next_ismd = 0;
 812                 rc = echo_site_init(env, ed);
 813                 if (rc)
 814                         GOTO(out, rc);
 815         }
 816         cleanup = 3;
 817
 818         rc = echo_client_setup(env, obd, cfg);
 819         if (rc)
 820                 GOTO(out, rc);
 821
 822         ed->ed_ec = &obd->u.echo_client;
 823         cleanup = 4;
 824
 825         if (ed->ed_next_ismd) {
 826                 /* Suppose to connect to some Metadata layer */
 827                 struct lu_site *ls;
 828                 struct lu_device *ld;
 829                 int    found = 0;
 830
 831                 if (next == NULL) {
 832                         CERROR("%s is not lu device type!\n",
 833                                lustre_cfg_string(cfg, 1));
 834                         GOTO(out, rc = -EINVAL);
 835                 }
 836
 837                 tgt_type_name = lustre_cfg_string(cfg, 2);
 838                 if (!tgt_type_name) {
 839                         CERROR("%s no type name for echo %s setup\n",
 840                                 lustre_cfg_string(cfg, 1),
 841                                 tgt->obd_type->typ_name);
 842                         GOTO(out, rc = -EINVAL);
 843                 }
 844
 845                 ls = next->ld_site;
 846
 847                 spin_lock(&ls->ls_ld_lock);
 848                 list_for_each_entry(ld, &ls->ls_ld_linkage, ld_linkage) {
 849                         if (strcmp(ld->ld_type->ldt_name, tgt_type_name) == 0) {
 850                                 found = 1;
 851                                 break;
 852                         }
 853                 }
 854                 spin_unlock(&ls->ls_ld_lock);
 855
 856                 if (found == 0) {
 857                         CERROR("%s is not lu device type!\n",
 858                                lustre_cfg_string(cfg, 1));
 859                         GOTO(out, rc = -EINVAL);
 860                 }
 861
 862                 next = ld;
 863                 /* For MD echo client, it will use the site in MDS stack */
 864                 ed->ed_site_myself.cs_lu = *ls;
 865                 ed->ed_site = &ed->ed_site_myself;
 866                 ed->ed_cl.cd_lu_dev.ld_site = &ed->ed_site_myself.cs_lu;
 867                 rc = echo_fid_init(ed, obd->obd_name, lu_site2seq(ls));
 868                 if (rc) {
 869                         CERROR("echo fid init error %d\n", rc);
 870                         GOTO(out, rc);
 871                 }
 872         } else {
 873                  /* if echo client is to be stacked upon ost device, the next is
 874                   * NULL since ost is not a clio device so far */
 875                 if (next != NULL && !lu_device_is_cl(next))
 876                         next = NULL;
 877
 878                 tgt_type_name = tgt->obd_type->typ_name;
 879                 if (next != NULL) {
 880                         LASSERT(next != NULL);
 881                         if (next->ld_site != NULL)
 882                                 GOTO(out, rc = -EBUSY);
 883
 884                         next->ld_site = &ed->ed_site->cs_lu;
 885                         rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
 886                                                      next->ld_type->ldt_name,
 887                                                      NULL);
 888                         if (rc)
 889                                 GOTO(out, rc);
 890
 891                         /* Tricky case, I have to determine the obd type since
 892                          * CLIO uses the different parameters to initialize
 893                          * objects for lov & osc. */
 894                         if (strcmp(tgt_type_name, LUSTRE_LOV_NAME) == 0)
 895                                 ed->ed_next_islov = 1;
 896                         else
 897                                 LASSERT(strcmp(tgt_type_name,
 898                                                LUSTRE_OSC_NAME) == 0);
 899                 } else
 900                         LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0);
 901         }
 902
 903         ed->ed_next = next;
 904         return &cd->cd_lu_dev;
 905 out:
 906         switch(cleanup) {
 907         case 4: {
 908                 int rc2;
 909                 rc2 = echo_client_cleanup(obd);
 910                 if (rc2)
 911                         CERROR("Cleanup obd device %s error(%d)\n",
 912                                obd->obd_name, rc2);
 913         }
 914
 915         case 3:
 916                 echo_site_fini(env, ed);
 917         case 2:
 918                 cl_device_fini(&ed->ed_cl);
 919         case 1:
 920                 OBD_FREE_PTR(ed);
 921         case 0:
 922         default:
 923                 break;
 924         }
 925         return(ERR_PTR(rc));
 926 }
 927
 928 static int echo_device_init(const struct lu_env *env, struct lu_device *d,
 929                           const char *name, struct lu_device *next)
 930 {
 931         LBUG();
 932         return 0;
 933 }
 934
 935 static struct lu_device *echo_device_fini(const struct lu_env *env,
 936                                           struct lu_device *d)
 937 {
 938         struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
 939         struct lu_device *next = ed->ed_next;
 940
 941         while (next && !ed->ed_next_ismd)
 942                 next = next->ld_type->ldt_ops->ldto_device_fini(env, next);
 943         return NULL;
 944 }
 945
 946 static void echo_lock_release(const struct lu_env *env,
 947                               struct echo_lock *ecl,
 948                               int still_used)
 949 {
 950         struct cl_lock *clk = echo_lock2cl(ecl);
 951
 952         cl_lock_get(clk);
 953         cl_unuse(env, clk);
 954         cl_lock_release(env, clk, "ec enqueue", ecl->el_object);
 955         if (!still_used) {
 956                 cl_lock_mutex_get(env, clk);
 957                 cl_lock_cancel(env, clk);
 958                 cl_lock_delete(env, clk);
 959                 cl_lock_mutex_put(env, clk);
 960         }
 961         cl_lock_put(env, clk);
 962 }
 963
 964 static struct lu_device *echo_device_free(const struct lu_env *env,
 965                                           struct lu_device *d)
 966 {
 967         struct echo_device     *ed   = cl2echo_dev(lu2cl_dev(d));
 968         struct echo_client_obd *ec   = ed->ed_ec;
 969         struct echo_object     *eco;
 970         struct lu_device       *next = ed->ed_next;
 971
 972         CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n",
 973                ed, next);
 974
 975         lu_site_purge(env, &ed->ed_site->cs_lu, -1);
 976
 977         /* check if there are objects still alive.
 978          * It shouldn't have any object because lu_site_purge would cleanup
 979          * all of cached objects. Anyway, probably the echo device is being
 980          * parallelly accessed.
 981          */
 982         spin_lock(&ec->ec_lock);
 983         list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain)
 984                 eco->eo_deleted = 1;
 985         spin_unlock(&ec->ec_lock);
 986
 987         /* purge again */
 988         lu_site_purge(env, &ed->ed_site->cs_lu, -1);
 989
 990         CDEBUG(D_INFO,
 991                "Waiting for the reference of echo object to be dropped\n");
 992
 993         /* Wait for the last reference to be dropped. */
 994         spin_lock(&ec->ec_lock);
 995         while (!list_empty(&ec->ec_objects)) {
 996                 spin_unlock(&ec->ec_lock);
 997                 CERROR("echo_client still has objects at cleanup time, "
 998                        "wait for 1 second\n");
 999                 set_current_state(TASK_UNINTERRUPTIBLE);
1000                 schedule_timeout(cfs_time_seconds(1));
1001                 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1002                 spin_lock(&ec->ec_lock);
1003         }
1004         spin_unlock(&ec->ec_lock);
1005
1006         LASSERT(list_empty(&ec->ec_locks));
1007
1008         CDEBUG(D_INFO, "No object exists, exiting...\n");
1009
1010         echo_client_cleanup(d->ld_obd);
1011         echo_fid_fini(d->ld_obd);
1012         while (next && !ed->ed_next_ismd)
1013                 next = next->ld_type->ldt_ops->ldto_device_free(env, next);
1014
1015         LASSERT(ed->ed_site == lu2cl_site(d->ld_site));
1016         echo_site_fini(env, ed);
1017         cl_device_fini(&ed->ed_cl);
1018         OBD_FREE_PTR(ed);
1019
1020         return NULL;
1021 }
1022
1023 static const struct lu_device_type_operations echo_device_type_ops = {
1024         .ldto_init = echo_type_init,
1025         .ldto_fini = echo_type_fini,
1026
1027         .ldto_start = echo_type_start,
1028         .ldto_stop  = echo_type_stop,
1029
1030         .ldto_device_alloc = echo_device_alloc,
1031         .ldto_device_free  = echo_device_free,
1032         .ldto_device_init  = echo_device_init,
1033         .ldto_device_fini  = echo_device_fini
1034 };
1035
1036 static struct lu_device_type echo_device_type = {
1037         .ldt_tags     = LU_DEVICE_CL,
1038         .ldt_name     = LUSTRE_ECHO_CLIENT_NAME,
1039         .ldt_ops      = &echo_device_type_ops,
1040         .ldt_ctx_tags = LCT_CL_THREAD | LCT_MD_THREAD | LCT_DT_THREAD,
1041 };
1042 /** @} echo_init */
1043
1044 /** \defgroup echo_exports Exported operations
1045  *
1046  * exporting functions to echo client
1047  *
1048  * @{
1049  */
1050
1051 /* Interfaces to echo client obd device */
1052 static struct echo_object *cl_echo_object_find(struct echo_device *d,
1053                                                struct lov_stripe_md **lsmp)
1054 {
1055         struct lu_env *env;
1056         struct echo_thread_info *info;
1057         struct echo_object_conf *conf;
1058         struct lov_stripe_md    *lsm;
1059         struct echo_object *eco;
1060         struct cl_object   *obj;
1061         struct lu_fid *fid;
1062         int refcheck;
1063         int rc;
1064
1065         LASSERT(lsmp);
1066         lsm = *lsmp;
1067         LASSERT(lsm);
1068         LASSERTF(ostid_id(&lsm->lsm_oi) != 0, DOSTID"\n", POSTID(&lsm->lsm_oi));
1069         LASSERTF(ostid_seq(&lsm->lsm_oi) == FID_SEQ_ECHO, DOSTID"\n",
1070                  POSTID(&lsm->lsm_oi));
1071
1072         /* Never return an object if the obd is to be freed. */
1073         if (echo_dev2cl(d)->cd_lu_dev.ld_obd->obd_stopping)
1074                 return ERR_PTR(-ENODEV);
1075
1076         env = cl_env_get(&refcheck);
1077         if (IS_ERR(env))
1078                 return (void *)env;
1079
1080         info = echo_env_info(env);
1081         conf = &info->eti_conf;
1082         if (d->ed_next) {
1083                 if (!d->ed_next_islov) {
1084                         struct lov_oinfo *oinfo = lsm->lsm_oinfo[0];
1085                         LASSERT(oinfo != NULL);
1086                         oinfo->loi_oi = lsm->lsm_oi;
1087                         conf->eoc_cl.u.coc_oinfo = oinfo;
1088                 } else {
1089                         struct lustre_md *md;
1090                         md = &info->eti_md;
1091                         memset(md, 0, sizeof(*md));
1092                         md->lsm = lsm;
1093                         conf->eoc_cl.u.coc_md = md;
1094                 }
1095         }
1096         conf->eoc_md = lsmp;
1097
1098         fid  = &info->eti_fid;
1099         rc = ostid_to_fid(fid, &lsm->lsm_oi, 0);
1100         if (rc != 0)
1101                 GOTO(out, eco = ERR_PTR(rc));
1102
1103         /* In the function below, .hs_keycmp resolves to
1104          * lu_obj_hop_keycmp() */
1105         /* coverity[overrun-buffer-val] */
1106         obj = cl_object_find(env, echo_dev2cl(d), fid, &conf->eoc_cl);
1107         if (IS_ERR(obj))
1108                 GOTO(out, eco = (void *)obj);
1109
1110         eco = cl2echo_obj(obj);
1111         if (eco->eo_deleted) {
1112                 cl_object_put(env, obj);
1113                 eco = ERR_PTR(-EAGAIN);
1114         }
1115
1116 out:
1117         cl_env_put(env, &refcheck);
1118         return eco;
1119 }
1120
1121 static int cl_echo_object_put(struct echo_object *eco)
1122 {
1123         struct lu_env *env;
1124         struct cl_object *obj = echo_obj2cl(eco);
1125         int refcheck;
1126
1127         env = cl_env_get(&refcheck);
1128         if (IS_ERR(env))
1129                 return PTR_ERR(env);
1130
1131         /* an external function to kill an object? */
1132         if (eco->eo_deleted) {
1133                 struct lu_object_header *loh = obj->co_lu.lo_header;
1134                 LASSERT(&eco->eo_hdr == luh2coh(loh));
1135                 set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags);
1136         }
1137
1138         cl_object_put(env, obj);
1139         cl_env_put(env, &refcheck);
1140         return 0;
1141 }
1142
1143 static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco,
1144                             obd_off start, obd_off end, int mode,
1145                             __u64 *cookie , __u32 enqflags)
1146 {
1147         struct cl_io *io;
1148         struct cl_lock *lck;
1149         struct cl_object *obj;
1150         struct cl_lock_descr *descr;
1151         struct echo_thread_info *info;
1152         int rc = -ENOMEM;
1153
1154         info = echo_env_info(env);
1155         io = &info->eti_io;
1156         descr = &info->eti_descr;
1157         obj = echo_obj2cl(eco);
1158
1159         descr->cld_obj   = obj;
1160         descr->cld_start = cl_index(obj, start);
1161         descr->cld_end   = cl_index(obj, end);
1162         descr->cld_mode  = mode == LCK_PW ? CLM_WRITE : CLM_READ;
1163         descr->cld_enq_flags = enqflags;
1164         io->ci_obj = obj;
1165
1166         lck = cl_lock_request(env, io, descr, "ec enqueue", eco);
1167         if (lck) {
1168                 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
1169                 struct echo_lock *el;
1170
1171                 rc = cl_wait(env, lck);
1172                 if (rc == 0) {
1173                         el = cl2echo_lock(cl_lock_at(lck, &echo_device_type));
1174                         spin_lock(&ec->ec_lock);
1175                         if (list_empty(&el->el_chain)) {
1176                                 list_add(&el->el_chain, &ec->ec_locks);
1177                                 el->el_cookie = ++ec->ec_unique;
1178                         }
1179                         atomic_inc(&el->el_refcount);
1180                         *cookie = el->el_cookie;
1181                         spin_unlock(&ec->ec_lock);
1182                 } else {
1183                         cl_lock_release(env, lck, "ec enqueue", current);
1184                 }
1185         }
1186         return rc;
1187 }
1188
1189 static int cl_echo_enqueue(struct echo_object *eco, obd_off start, obd_off end,
1190                            int mode, __u64 *cookie)
1191 {
1192         struct echo_thread_info *info;
1193         struct lu_env *env;
1194         struct cl_io *io;
1195         int refcheck;
1196         int result;
1197
1198         env = cl_env_get(&refcheck);
1199         if (IS_ERR(env))
1200                 return PTR_ERR(env);
1201
1202         info = echo_env_info(env);
1203         io = &info->eti_io;
1204
1205         io->ci_ignore_layout = 1;
1206         result = cl_io_init(env, io, CIT_MISC, echo_obj2cl(eco));
1207         if (result < 0)
1208                 GOTO(out, result);
1209         LASSERT(result == 0);
1210
1211         result = cl_echo_enqueue0(env, eco, start, end, mode, cookie, 0);
1212         cl_io_fini(env, io);
1213
1214 out:
1215         cl_env_put(env, &refcheck);
1216         return result;
1217 }
1218
1219 static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed,
1220                            __u64 cookie)
1221 {
1222         struct echo_client_obd *ec = ed->ed_ec;
1223         struct echo_lock       *ecl = NULL;
1224         struct list_head             *el;
1225         int found = 0, still_used = 0;
1226
1227         LASSERT(ec != NULL);
1228         spin_lock(&ec->ec_lock);
1229         list_for_each (el, &ec->ec_locks) {
1230                 ecl = list_entry (el, struct echo_lock, el_chain);
1231                 CDEBUG(D_INFO, "ecl: %p, cookie: "LPX64"\n", ecl, ecl->el_cookie);
1232                 found = (ecl->el_cookie == cookie);
1233                 if (found) {
1234                         if (atomic_dec_and_test(&ecl->el_refcount))
1235                                 list_del_init(&ecl->el_chain);
1236                         else
1237                                 still_used = 1;
1238                         break;
1239                 }
1240         }
1241         spin_unlock(&ec->ec_lock);
1242
1243         if (!found)
1244                 return -ENOENT;
1245
1246         echo_lock_release(env, ecl, still_used);
1247         return 0;
1248 }
1249
1250 static int cl_echo_cancel(struct echo_device *ed, __u64 cookie)
1251 {
1252         struct lu_env *env;
1253         int refcheck;
1254         int rc;
1255
1256         env = cl_env_get(&refcheck);
1257         if (IS_ERR(env))
1258                 return PTR_ERR(env);
1259
1260         rc = cl_echo_cancel0(env, ed, cookie);
1261
1262         cl_env_put(env, &refcheck);
1263         return rc;
1264 }
1265
1266 static int cl_echo_async_brw(const struct lu_env *env, struct cl_io *io,
1267                              enum cl_req_type unused, struct cl_2queue *queue)
1268 {
1269         struct cl_page *clp;
1270         struct cl_page *temp;
1271         int result = 0;
1272
1273         cl_page_list_for_each_safe(clp, temp, &queue->c2_qin) {
1274                 int rc;
1275                 rc = cl_page_cache_add(env, io, clp, CRT_WRITE);
1276                 if (rc == 0)
1277                         continue;
1278                 result = result ?: rc;
1279         }
1280         return result;
1281 }
1282
1283 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
1284                               struct page **pages, int npages, int async)
1285 {
1286         struct lu_env      *env;
1287         struct echo_thread_info *info;
1288         struct cl_object        *obj = echo_obj2cl(eco);
1289         struct echo_device      *ed  = eco->eo_dev;
1290         struct cl_2queue        *queue;
1291         struct cl_io        *io;
1292         struct cl_page    *clp;
1293         struct lustre_handle    lh = { 0 };
1294         int page_size = cl_page_size(obj);
1295         int refcheck;
1296         int rc;
1297         int i;
1298
1299         LASSERT((offset & ~CFS_PAGE_MASK) == 0);
1300         LASSERT(ed->ed_next != NULL);
1301         env = cl_env_get(&refcheck);
1302         if (IS_ERR(env))
1303                 return PTR_ERR(env);
1304
1305         info    = echo_env_info(env);
1306         io      = &info->eti_io;
1307         queue   = &info->eti_queue;
1308
1309         cl_2queue_init(queue);
1310
1311         io->ci_ignore_layout = 1;
1312         rc = cl_io_init(env, io, CIT_MISC, obj);
1313         if (rc < 0)
1314                 GOTO(out, rc);
1315         LASSERT(rc == 0);
1316
1317
1318         rc = cl_echo_enqueue0(env, eco, offset,
1319                               offset + npages * PAGE_CACHE_SIZE - 1,
1320                               rw == READ ? LCK_PR : LCK_PW, &lh.cookie,
1321                               CEF_NEVER);
1322         if (rc < 0)
1323                 GOTO(error_lock, rc);
1324
1325         for (i = 0; i < npages; i++) {
1326                 LASSERT(pages[i]);
1327                 clp = cl_page_find(env, obj, cl_index(obj, offset),
1328                                    pages[i], CPT_TRANSIENT);
1329                 if (IS_ERR(clp)) {
1330                         rc = PTR_ERR(clp);
1331                         break;
1332                 }
1333                 LASSERT(clp->cp_type == CPT_TRANSIENT);
1334
1335                 rc = cl_page_own(env, io, clp);
1336                 if (rc) {
1337                         LASSERT(clp->cp_state == CPS_FREEING);
1338                         cl_page_put(env, clp);
1339                         break;
1340                 }
1341
1342                 cl_2queue_add(queue, clp);
1343
1344                 /* drop the reference count for cl_page_find, so that the page
1345                  * will be freed in cl_2queue_fini. */
1346                 cl_page_put(env, clp);
1347                 cl_page_clip(env, clp, 0, page_size);
1348
1349                 offset += page_size;
1350         }
1351
1352         if (rc == 0) {
1353                 enum cl_req_type typ = rw == READ ? CRT_READ : CRT_WRITE;
1354
1355                 async = async && (typ == CRT_WRITE);
1356                 if (async)
1357                         rc = cl_echo_async_brw(env, io, typ, queue);
1358                 else
1359                         rc = cl_io_submit_sync(env, io, typ, queue, 0);
1360                 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
1361                        async ? "async" : "sync", rc);
1362         }
1363
1364         cl_echo_cancel0(env, ed, lh.cookie);
1365 error_lock:
1366         cl_2queue_discard(env, io, queue);
1367         cl_2queue_disown(env, io, queue);
1368         cl_2queue_fini(env, queue);
1369         cl_io_fini(env, io);
1370 out:
1371         cl_env_put(env, &refcheck);
1372         return rc;
1373 }
1374 /** @} echo_exports */
1375
1376
1377 static obd_id last_object_id;
1378
1379 static int
1380 echo_copyout_lsm (struct lov_stripe_md *lsm, void *_ulsm, int ulsm_nob)
1381 {
1382         struct lov_stripe_md *ulsm = _ulsm;
1383         int nob, i;
1384
1385         nob = offsetof (struct lov_stripe_md, lsm_oinfo[lsm->lsm_stripe_count]);
1386         if (nob > ulsm_nob)
1387                 return (-EINVAL);
1388
1389         if (copy_to_user (ulsm, lsm, sizeof(*ulsm)))
1390                 return (-EFAULT);
1391
1392         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1393                 if (copy_to_user (ulsm->lsm_oinfo[i], lsm->lsm_oinfo[i],
1394                                       sizeof(lsm->lsm_oinfo[0])))
1395                         return (-EFAULT);
1396         }
1397         return 0;
1398 }
1399
1400 static int
1401 echo_copyin_lsm (struct echo_device *ed, struct lov_stripe_md *lsm,
1402                  void *ulsm, int ulsm_nob)
1403 {
1404         struct echo_client_obd *ec = ed->ed_ec;
1405         int                  i;
1406
1407         if (ulsm_nob < sizeof (*lsm))
1408                 return (-EINVAL);
1409
1410         if (copy_from_user (lsm, ulsm, sizeof (*lsm)))
1411                 return (-EFAULT);
1412
1413         if (lsm->lsm_stripe_count > ec->ec_nstripes ||
1414             lsm->lsm_magic != LOV_MAGIC ||
1415             (lsm->lsm_stripe_size & (~CFS_PAGE_MASK)) != 0 ||
1416             ((__u64)lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL))
1417                 return (-EINVAL);
1418
1419
1420         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1421                 if (copy_from_user(lsm->lsm_oinfo[i],
1422                                        ((struct lov_stripe_md *)ulsm)-> \
1423                                        lsm_oinfo[i],
1424                                        sizeof(lsm->lsm_oinfo[0])))
1425                         return (-EFAULT);
1426         }
1427         return (0);
1428 }
1429
1430 static inline void echo_md_build_name(struct lu_name *lname, char *name,
1431                                       __u64 id)
1432 {
1433         sprintf(name, LPU64, id);
1434         lname->ln_name = name;
1435         lname->ln_namelen = strlen(name);
1436 }
1437
1438 /* similar to mdt_attr_get_complex */
1439 static int echo_big_lmm_get(const struct lu_env *env, struct md_object *o,
1440                             struct md_attr *ma)
1441 {
1442         struct echo_thread_info *info = echo_env_info(env);
1443         int                      rc;
1444
1445         LASSERT(ma->ma_lmm_size > 0);
1446
1447         rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LOV);
1448         if (rc < 0)
1449                 return rc;
1450
1451         /* big_lmm may need to be grown */
1452         if (info->eti_big_lmmsize < rc) {
1453                 int size = size_roundup_power2(rc);
1454
1455                 if (info->eti_big_lmmsize > 0) {
1456                         /* free old buffer */
1457                         LASSERT(info->eti_big_lmm);
1458                         OBD_FREE_LARGE(info->eti_big_lmm,
1459                                        info->eti_big_lmmsize);
1460                         info->eti_big_lmm = NULL;
1461                         info->eti_big_lmmsize = 0;
1462                 }
1463
1464                 OBD_ALLOC_LARGE(info->eti_big_lmm, size);
1465                 if (info->eti_big_lmm == NULL)
1466                         return -ENOMEM;
1467                 info->eti_big_lmmsize = size;
1468         }
1469         LASSERT(info->eti_big_lmmsize >= rc);
1470
1471         info->eti_buf.lb_buf = info->eti_big_lmm;
1472         info->eti_buf.lb_len = info->eti_big_lmmsize;
1473         rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LOV);
1474         if (rc < 0)
1475                 return rc;
1476
1477         ma->ma_valid |= MA_LOV;
1478         ma->ma_lmm = info->eti_big_lmm;
1479         ma->ma_lmm_size = rc;
1480
1481         return 0;
1482 }
1483
1484 int echo_attr_get_complex(const struct lu_env *env, struct md_object *next,
1485                           struct md_attr *ma)
1486 {
1487         struct echo_thread_info *info = echo_env_info(env);
1488         struct lu_buf           *buf = &info->eti_buf;
1489         umode_t          mode = lu_object_attr(&next->mo_lu);
1490         int                      need = ma->ma_need;
1491         int                      rc = 0, rc2;
1492
1493         ma->ma_valid = 0;
1494
1495         if (need & MA_INODE) {
1496                 ma->ma_need = MA_INODE;
1497                 rc = mo_attr_get(env, next, ma);
1498                 if (rc)
1499                         GOTO(out, rc);
1500                 ma->ma_valid |= MA_INODE;
1501         }
1502
1503         if (need & MA_LOV) {
1504                 if (S_ISREG(mode) || S_ISDIR(mode)) {
1505                         LASSERT(ma->ma_lmm_size > 0);
1506                         buf->lb_buf = ma->ma_lmm;
1507                         buf->lb_len = ma->ma_lmm_size;
1508                         rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LOV);
1509                         if (rc2 > 0) {
1510                                 ma->ma_lmm_size = rc2;
1511                                 ma->ma_valid |= MA_LOV;
1512                         } else if (rc2 == -ENODATA) {
1513                                 /* no LOV EA */
1514                                 ma->ma_lmm_size = 0;
1515                         } else if (rc2 == -ERANGE) {
1516                                 rc2 = echo_big_lmm_get(env, next, ma);
1517                                 if (rc2 < 0)
1518                                         GOTO(out, rc = rc2);
1519                         } else {
1520                                 GOTO(out, rc = rc2);
1521                         }
1522                 }
1523         }
1524
1525 #ifdef CONFIG_FS_POSIX_ACL
1526         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1527                 buf->lb_buf = ma->ma_acl;
1528                 buf->lb_len = ma->ma_acl_size;
1529                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1530                 if (rc2 > 0) {
1531                         ma->ma_acl_size = rc2;
1532                         ma->ma_valid |= MA_ACL_DEF;
1533                 } else if (rc2 == -ENODATA) {
1534                         /* no ACLs */
1535                         ma->ma_acl_size = 0;
1536                 } else {
1537                         GOTO(out, rc = rc2);
1538                 }
1539         }
1540 #endif
1541 out:
1542         ma->ma_need = need;
1543         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
1544                rc, ma->ma_valid, ma->ma_lmm);
1545         return rc;
1546 }
1547
1548 static int
1549 echo_md_create_internal(const struct lu_env *env, struct echo_device *ed,
1550                         struct md_object *parent, struct lu_fid *fid,
1551                         struct lu_name *lname, struct md_op_spec *spec,
1552                         struct md_attr *ma)
1553 {
1554         struct lu_object        *ec_child, *child;
1555         struct lu_device        *ld = ed->ed_next;
1556         struct echo_thread_info *info = echo_env_info(env);
1557         struct lu_fid           *fid2 = &info->eti_fid2;
1558         struct lu_object_conf    conf = { .loc_flags = LOC_F_NEW };
1559         int                      rc;
1560
1561         rc = mdo_lookup(env, parent, lname, fid2, spec);
1562         if (rc == 0)
1563                 return -EEXIST;
1564         else if (rc != -ENOENT)
1565                 return rc;
1566
1567         ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev,
1568                                      fid, &conf);
1569         if (IS_ERR(ec_child)) {
1570                 CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid),
1571                         PTR_ERR(ec_child));
1572                 return PTR_ERR(ec_child);
1573         }
1574
1575         child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1576         if (child == NULL) {
1577                 CERROR("Can not locate the child "DFID"\n", PFID(fid));
1578                 GOTO(out_put, rc = -EINVAL);
1579         }
1580
1581         CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n",
1582                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1583
1584         /*
1585          * Do not perform lookup sanity check. We know that name does not exist.
1586          */
1587         spec->sp_cr_lookup = 0;
1588         rc = mdo_create(env, parent, lname, lu2md(child), spec, ma);
1589         if (rc) {
1590                 CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc);
1591                 GOTO(out_put, rc);
1592         }
1593         CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc  = %d\n",
1594                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc);
1595 out_put:
1596         lu_object_put(env, ec_child);
1597         return rc;
1598 }
1599
1600 static int echo_set_lmm_size(const struct lu_env *env, struct lu_device *ld,
1601                              struct md_attr *ma)
1602 {
1603         struct echo_thread_info *info = echo_env_info(env);
1604
1605         if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
1606                 ma->ma_lmm = (void *)&info->eti_lmm;
1607                 ma->ma_lmm_size = sizeof(info->eti_lmm);
1608         } else {
1609                 LASSERT(info->eti_big_lmmsize);
1610                 ma->ma_lmm = info->eti_big_lmm;
1611                 ma->ma_lmm_size = info->eti_big_lmmsize;
1612         }
1613
1614         return 0;
1615 }
1616
1617 static int echo_create_md_object(const struct lu_env *env,
1618                                  struct echo_device *ed,
1619                                  struct lu_object *ec_parent,
1620                                  struct lu_fid *fid,
1621                                  char *name, int namelen,
1622                                  __u64 id, __u32 mode, int count,
1623                                  int stripe_count, int stripe_offset)
1624 {
1625         struct lu_object        *parent;
1626         struct echo_thread_info *info = echo_env_info(env);
1627         struct lu_name    *lname = &info->eti_lname;
1628         struct md_op_spec       *spec = &info->eti_spec;
1629         struct md_attr    *ma = &info->eti_ma;
1630         struct lu_device        *ld = ed->ed_next;
1631         int                   rc = 0;
1632         int                   i;
1633
1634         if (ec_parent == NULL)
1635                 return -1;
1636         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1637         if (parent == NULL)
1638                 return -ENXIO;
1639
1640         memset(ma, 0, sizeof(*ma));
1641         memset(spec, 0, sizeof(*spec));
1642         if (stripe_count != 0) {
1643                 spec->sp_cr_flags |= FMODE_WRITE;
1644                 echo_set_lmm_size(env, ld, ma);
1645                 if (stripe_count != -1) {
1646                         struct lov_user_md_v3 *lum = &info->eti_lum;
1647
1648                         lum->lmm_magic = LOV_USER_MAGIC_V3;
1649                         lum->lmm_stripe_count = stripe_count;
1650                         lum->lmm_stripe_offset = stripe_offset;
1651                         lum->lmm_pattern = 0;
1652                         spec->u.sp_ea.eadata = lum;
1653                         spec->u.sp_ea.eadatalen = sizeof(*lum);
1654                         spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1655                 }
1656         }
1657
1658         ma->ma_attr.la_mode = mode;
1659         ma->ma_attr.la_valid = LA_CTIME | LA_MODE;
1660         ma->ma_attr.la_ctime = cfs_time_current_64();
1661
1662         if (name != NULL) {
1663                 lname->ln_name = name;
1664                 lname->ln_namelen = namelen;
1665                 /* If name is specified, only create one object by name */
1666                 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1667                                              spec, ma);
1668                 return rc;
1669         }
1670
1671         /* Create multiple object sequenced by id */
1672         for (i = 0; i < count; i++) {
1673                 char *tmp_name = info->eti_name;
1674
1675                 echo_md_build_name(lname, tmp_name, id);
1676
1677                 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1678                                              spec, ma);
1679                 if (rc) {
1680                         CERROR("Can not create child %s: rc = %d\n", tmp_name,
1681                                 rc);
1682                         break;
1683                 }
1684                 id++;
1685                 fid->f_oid++;
1686         }
1687
1688         return rc;
1689 }
1690
1691 static struct lu_object *echo_md_lookup(const struct lu_env *env,
1692                                         struct echo_device *ed,
1693                                         struct md_object *parent,
1694                                         struct lu_name *lname)
1695 {
1696         struct echo_thread_info *info = echo_env_info(env);
1697         struct lu_fid      *fid = &info->eti_fid;
1698         struct lu_object        *child;
1699         int    rc;
1700
1701         CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name,
1702                PFID(fid), parent);
1703         rc = mdo_lookup(env, parent, lname, fid, NULL);
1704         if (rc) {
1705                 CERROR("lookup %s: rc = %d\n", lname->ln_name, rc);
1706                 return ERR_PTR(rc);
1707         }
1708
1709         /* In the function below, .hs_keycmp resolves to
1710          * lu_obj_hop_keycmp() */
1711         /* coverity[overrun-buffer-val] */
1712         child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1713
1714         return child;
1715 }
1716
1717 static int echo_setattr_object(const struct lu_env *env,
1718                                struct echo_device *ed,
1719                                struct lu_object *ec_parent,
1720                                __u64 id, int count)
1721 {
1722         struct lu_object        *parent;
1723         struct echo_thread_info *info = echo_env_info(env);
1724         struct lu_name    *lname = &info->eti_lname;
1725         char                *name = info->eti_name;
1726         struct lu_device        *ld = ed->ed_next;
1727         struct lu_buf      *buf = &info->eti_buf;
1728         int                   rc = 0;
1729         int                   i;
1730
1731         if (ec_parent == NULL)
1732                 return -1;
1733         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1734         if (parent == NULL)
1735                 return -ENXIO;
1736
1737         for (i = 0; i < count; i++) {
1738                 struct lu_object *ec_child, *child;
1739
1740                 echo_md_build_name(lname, name, id);
1741
1742                 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1743                 if (IS_ERR(ec_child)) {
1744                         CERROR("Can't find child %s: rc = %ld\n",
1745                                 lname->ln_name, PTR_ERR(ec_child));
1746                         return PTR_ERR(ec_child);
1747                 }
1748
1749                 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1750                 if (child == NULL) {
1751                         CERROR("Can not locate the child %s\n", lname->ln_name);
1752                         lu_object_put(env, ec_child);
1753                         rc = -EINVAL;
1754                         break;
1755                 }
1756
1757                 CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n",
1758                        PFID(lu_object_fid(child)));
1759
1760                 buf->lb_buf = info->eti_xattr_buf;
1761                 buf->lb_len = sizeof(info->eti_xattr_buf);
1762
1763                 sprintf(name, "%s.test1", XATTR_USER_PREFIX);
1764                 rc = mo_xattr_set(env, lu2md(child), buf, name,
1765                                   LU_XATTR_CREATE);
1766                 if (rc < 0) {
1767                         CERROR("Can not setattr child "DFID": rc = %d\n",
1768                                 PFID(lu_object_fid(child)), rc);
1769                         lu_object_put(env, ec_child);
1770                         break;
1771                 }
1772                 CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n",
1773                        PFID(lu_object_fid(child)));
1774                 id++;
1775                 lu_object_put(env, ec_child);
1776         }
1777         return rc;
1778 }
1779
1780 static int echo_getattr_object(const struct lu_env *env,
1781                                struct echo_device *ed,
1782                                struct lu_object *ec_parent,
1783                                __u64 id, int count)
1784 {
1785         struct lu_object        *parent;
1786         struct echo_thread_info *info = echo_env_info(env);
1787         struct lu_name    *lname = &info->eti_lname;
1788         char                *name = info->eti_name;
1789         struct md_attr    *ma = &info->eti_ma;
1790         struct lu_device        *ld = ed->ed_next;
1791         int                   rc = 0;
1792         int                   i;
1793
1794         if (ec_parent == NULL)
1795                 return -1;
1796         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1797         if (parent == NULL)
1798                 return -ENXIO;
1799
1800         memset(ma, 0, sizeof(*ma));
1801         ma->ma_need |= MA_INODE | MA_LOV | MA_PFID | MA_HSM | MA_ACL_DEF;
1802         ma->ma_acl = info->eti_xattr_buf;
1803         ma->ma_acl_size = sizeof(info->eti_xattr_buf);
1804
1805         for (i = 0; i < count; i++) {
1806                 struct lu_object *ec_child, *child;
1807
1808                 ma->ma_valid = 0;
1809                 echo_md_build_name(lname, name, id);
1810                 echo_set_lmm_size(env, ld, ma);
1811
1812                 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1813                 if (IS_ERR(ec_child)) {
1814                         CERROR("Can't find child %s: rc = %ld\n",
1815                                lname->ln_name, PTR_ERR(ec_child));
1816                         return PTR_ERR(ec_child);
1817                 }
1818
1819                 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1820                 if (child == NULL) {
1821                         CERROR("Can not locate the child %s\n", lname->ln_name);
1822                         lu_object_put(env, ec_child);
1823                         return -EINVAL;
1824                 }
1825
1826                 CDEBUG(D_RPCTRACE, "Start getattr object "DFID"\n",
1827                        PFID(lu_object_fid(child)));
1828                 rc = echo_attr_get_complex(env, lu2md(child), ma);
1829                 if (rc) {
1830                         CERROR("Can not getattr child "DFID": rc = %d\n",
1831                                 PFID(lu_object_fid(child)), rc);
1832                         lu_object_put(env, ec_child);
1833                         break;
1834                 }
1835                 CDEBUG(D_RPCTRACE, "End getattr object "DFID"\n",
1836                        PFID(lu_object_fid(child)));
1837                 id++;
1838                 lu_object_put(env, ec_child);
1839         }
1840
1841         return rc;
1842 }
1843
1844 static int echo_lookup_object(const struct lu_env *env,
1845                               struct echo_device *ed,
1846                               struct lu_object *ec_parent,
1847                               __u64 id, int count)
1848 {
1849         struct lu_object        *parent;
1850         struct echo_thread_info *info = echo_env_info(env);
1851         struct lu_name    *lname = &info->eti_lname;
1852         char                *name = info->eti_name;
1853         struct lu_fid      *fid = &info->eti_fid;
1854         struct lu_device        *ld = ed->ed_next;
1855         int                   rc = 0;
1856         int                   i;
1857
1858         if (ec_parent == NULL)
1859                 return -1;
1860         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1861         if (parent == NULL)
1862                 return -ENXIO;
1863
1864         /*prepare the requests*/
1865         for (i = 0; i < count; i++) {
1866                 echo_md_build_name(lname, name, id);
1867
1868                 CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n",
1869                        PFID(lu_object_fid(parent)), lname->ln_name, parent);
1870
1871                 rc = mdo_lookup(env, lu2md(parent), lname, fid, NULL);
1872                 if (rc) {
1873                         CERROR("Can not lookup child %s: rc = %d\n", name, rc);
1874                         break;
1875                 }
1876                 CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n",
1877                        PFID(lu_object_fid(parent)), lname->ln_name, parent);
1878
1879                 id++;
1880         }
1881         return rc;
1882 }
1883
1884 static int echo_md_destroy_internal(const struct lu_env *env,
1885                                     struct echo_device *ed,
1886                                     struct md_object *parent,
1887                                     struct lu_name *lname,
1888                                     struct md_attr *ma)
1889 {
1890         struct lu_device   *ld = ed->ed_next;
1891         struct lu_object   *ec_child;
1892         struct lu_object   *child;
1893         int              rc;
1894
1895         ec_child = echo_md_lookup(env, ed, parent, lname);
1896         if (IS_ERR(ec_child)) {
1897                 CERROR("Can't find child %s: rc = %ld\n", lname->ln_name,
1898                         PTR_ERR(ec_child));
1899                 return PTR_ERR(ec_child);
1900         }
1901
1902         child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1903         if (child == NULL) {
1904                 CERROR("Can not locate the child %s\n", lname->ln_name);
1905                 GOTO(out_put, rc = -EINVAL);
1906         }
1907
1908         CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n",
1909                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1910
1911         rc = mdo_unlink(env, parent, lu2md(child), lname, ma, 0);
1912         if (rc) {
1913                 CERROR("Can not unlink child %s: rc = %d\n",
1914                         lname->ln_name, rc);
1915                 GOTO(out_put, rc);
1916         }
1917         CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n",
1918                PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1919 out_put:
1920         lu_object_put(env, ec_child);
1921         return rc;
1922 }
1923
1924 static int echo_destroy_object(const struct lu_env *env,
1925                                struct echo_device *ed,
1926                                struct lu_object *ec_parent,
1927                                char *name, int namelen,
1928                                __u64 id, __u32 mode,
1929                                int count)
1930 {
1931         struct echo_thread_info *info = echo_env_info(env);
1932         struct lu_name    *lname = &info->eti_lname;
1933         struct md_attr    *ma = &info->eti_ma;
1934         struct lu_device        *ld = ed->ed_next;
1935         struct lu_object        *parent;
1936         int                   rc = 0;
1937         int                   i;
1938
1939         parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1940         if (parent == NULL)
1941                 return -EINVAL;
1942
1943         memset(ma, 0, sizeof(*ma));
1944         ma->ma_attr.la_mode = mode;
1945         ma->ma_attr.la_valid = LA_CTIME;
1946         ma->ma_attr.la_ctime = cfs_time_current_64();
1947         ma->ma_need = MA_INODE;
1948         ma->ma_valid = 0;
1949
1950         if (name != NULL) {
1951                 lname->ln_name = name;
1952                 lname->ln_namelen = namelen;
1953                 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
1954                                               ma);
1955                 return rc;
1956         }
1957
1958         /*prepare the requests*/
1959         for (i = 0; i < count; i++) {
1960                 char *tmp_name = info->eti_name;
1961
1962                 ma->ma_valid = 0;
1963                 echo_md_build_name(lname, tmp_name, id);
1964
1965                 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
1966                                               ma);
1967                 if (rc) {
1968                         CERROR("Can not unlink child %s: rc = %d\n", name, rc);
1969                         break;
1970                 }
1971                 id++;
1972         }
1973
1974         return rc;
1975 }
1976
1977 static struct lu_object *echo_resolve_path(const struct lu_env *env,
1978                                            struct echo_device *ed, char *path,
1979                                            int path_len)
1980 {
1981         struct lu_device        *ld = ed->ed_next;
1982         struct md_device        *md = lu2md_dev(ld);
1983         struct echo_thread_info *info = echo_env_info(env);
1984         struct lu_fid      *fid = &info->eti_fid;
1985         struct lu_name    *lname = &info->eti_lname;
1986         struct lu_object        *parent = NULL;
1987         struct lu_object        *child = NULL;
1988         int rc = 0;
1989
1990         /*Only support MDD layer right now*/
1991         rc = md->md_ops->mdo_root_get(env, md, fid);
1992         if (rc) {
1993                 CERROR("get root error: rc = %d\n", rc);
1994                 return ERR_PTR(rc);
1995         }
1996
1997         /* In the function below, .hs_keycmp resolves to
1998          * lu_obj_hop_keycmp() */
1999         /* coverity[overrun-buffer-val] */
2000         parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
2001         if (IS_ERR(parent)) {
2002                 CERROR("Can not find the parent "DFID": rc = %ld\n",
2003                         PFID(fid), PTR_ERR(parent));
2004                 return parent;
2005         }
2006
2007         while (1) {
2008                 struct lu_object *ld_parent;
2009                 char *e;
2010
2011                 e = strsep(&path, "/");
2012                 if (e == NULL)
2013                         break;
2014
2015                 if (e[0] == 0) {
2016                         if (!path || path[0] == '\0')
2017                                 break;
2018                         continue;
2019                 }
2020
2021                 lname->ln_name = e;
2022                 lname->ln_namelen = strlen(e);
2023
2024                 ld_parent = lu_object_locate(parent->lo_header, ld->ld_type);
2025                 if (ld_parent == NULL) {
2026                         lu_object_put(env, parent);
2027                         rc = -EINVAL;
2028                         break;
2029                 }
2030
2031                 child = echo_md_lookup(env, ed, lu2md(ld_parent), lname);
2032                 lu_object_put(env, parent);
2033                 if (IS_ERR(child)) {
2034                         rc = (int)PTR_ERR(child);
2035                         CERROR("lookup %s under parent "DFID": rc = %d\n",
2036                                 lname->ln_name, PFID(lu_object_fid(ld_parent)),
2037                                 rc);
2038                         break;
2039                 }
2040                 parent = child;
2041         }
2042         if (rc)
2043                 return ERR_PTR(rc);
2044
2045         return parent;
2046 }
2047
2048 static void echo_ucred_init(struct lu_env *env)
2049 {
2050         struct lu_ucred *ucred = lu_ucred(env);
2051
2052         ucred->uc_valid = UCRED_INVALID;
2053
2054         ucred->uc_suppgids[0] = -1;
2055         ucred->uc_suppgids[1] = -1;
2056
2057         ucred->uc_uid   = ucred->uc_o_uid   =
2058                                 from_kuid(&init_user_ns, current_uid());
2059         ucred->uc_gid   = ucred->uc_o_gid   =
2060                                 from_kgid(&init_user_ns, current_gid());
2061         ucred->uc_fsuid = ucred->uc_o_fsuid =
2062                                 from_kuid(&init_user_ns, current_fsuid());
2063         ucred->uc_fsgid = ucred->uc_o_fsgid =
2064                                 from_kgid(&init_user_ns, current_fsgid());
2065         ucred->uc_cap   = cfs_curproc_cap_pack();
2066
2067         /* remove fs privilege for non-root user. */
2068         if (ucred->uc_fsuid)
2069                 ucred->uc_cap &= ~CFS_CAP_FS_MASK;
2070         ucred->uc_valid = UCRED_NEW;
2071 }
2072
2073 static void echo_ucred_fini(struct lu_env *env)
2074 {
2075         struct lu_ucred *ucred = lu_ucred(env);
2076         ucred->uc_valid = UCRED_INIT;
2077 }
2078
2079 #define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_MD_THREAD)
2080 #define ECHO_MD_SES_TAG (LCT_REMEMBER | LCT_SESSION)
2081 static int echo_md_handler(struct echo_device *ed, int command,
2082                            char *path, int path_len, __u64 id, int count,
2083                            struct obd_ioctl_data *data)
2084 {
2085         struct echo_thread_info *info;
2086         struct lu_device      *ld = ed->ed_next;
2087         struct lu_env    *env;
2088         int                 refcheck;
2089         struct lu_object      *parent;
2090         char              *name = NULL;
2091         int                 namelen = data->ioc_plen2;
2092         int                 rc = 0;
2093
2094         if (ld == NULL) {
2095                 CERROR("MD echo client is not being initialized properly\n");
2096                 return -EINVAL;
2097         }
2098
2099         if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
2100                 CERROR("Only support MDD layer right now!\n");
2101                 return -EINVAL;
2102         }
2103
2104         env = cl_env_get(&refcheck);
2105         if (IS_ERR(env))
2106                 return PTR_ERR(env);
2107
2108         rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG);
2109         if (rc != 0)
2110                 GOTO(out_env, rc);
2111
2112         /* init big_lmm buffer */
2113         info = echo_env_info(env);
2114         LASSERT(info->eti_big_lmm == NULL);
2115         OBD_ALLOC_LARGE(info->eti_big_lmm, MIN_MD_SIZE);
2116         if (info->eti_big_lmm == NULL)
2117                 GOTO(out_env, rc = -ENOMEM);
2118         info->eti_big_lmmsize = MIN_MD_SIZE;
2119
2120         parent = echo_resolve_path(env, ed, path, path_len);
2121         if (IS_ERR(parent)) {
2122                 CERROR("Can not resolve the path %s: rc = %ld\n", path,
2123                         PTR_ERR(parent));
2124                 GOTO(out_free, rc = PTR_ERR(parent));
2125         }
2126
2127         if (namelen > 0) {
2128                 OBD_ALLOC(name, namelen + 1);
2129                 if (name == NULL)
2130                         GOTO(out_put, rc = -ENOMEM);
2131                 if (copy_from_user(name, data->ioc_pbuf2, namelen))
2132                         GOTO(out_name, rc = -EFAULT);
2133         }
2134
2135         echo_ucred_init(env);
2136
2137         switch (command) {
2138         case ECHO_MD_CREATE:
2139         case ECHO_MD_MKDIR: {
2140                 struct echo_thread_info *info = echo_env_info(env);
2141                 __u32 mode = data->ioc_obdo2.o_mode;
2142                 struct lu_fid *fid = &info->eti_fid;
2143                 int stripe_count = (int)data->ioc_obdo2.o_misc;
2144                 int stripe_index = (int)data->ioc_obdo2.o_stripe_idx;
2145
2146                 rc = ostid_to_fid(fid, &data->ioc_obdo1.o_oi, 0);
2147                 if (rc != 0)
2148                         break;
2149
2150                 /* In the function below, .hs_keycmp resolves to
2151                  * lu_obj_hop_keycmp() */
2152                 /* coverity[overrun-buffer-val] */
2153                 rc = echo_create_md_object(env, ed, parent, fid, name, namelen,
2154                                            id, mode, count, stripe_count,
2155                                            stripe_index);
2156                 break;
2157         }
2158         case ECHO_MD_DESTROY:
2159         case ECHO_MD_RMDIR: {
2160                 __u32 mode = data->ioc_obdo2.o_mode;
2161
2162                 rc = echo_destroy_object(env, ed, parent, name, namelen,
2163                                          id, mode, count);
2164                 break;
2165         }
2166         case ECHO_MD_LOOKUP:
2167                 rc = echo_lookup_object(env, ed, parent, id, count);
2168                 break;
2169         case ECHO_MD_GETATTR:
2170                 rc = echo_getattr_object(env, ed, parent, id, count);
2171                 break;
2172         case ECHO_MD_SETATTR:
2173                 rc = echo_setattr_object(env, ed, parent, id, count);
2174                 break;
2175         default:
2176                 CERROR("unknown command %d\n", command);
2177                 rc = -EINVAL;
2178                 break;
2179         }
2180         echo_ucred_fini(env);
2181
2182 out_name:
2183         if (name != NULL)
2184                 OBD_FREE(name, namelen + 1);
2185 out_put:
2186         lu_object_put(env, parent);
2187 out_free:
2188         LASSERT(info->eti_big_lmm);
2189         OBD_FREE_LARGE(info->eti_big_lmm, info->eti_big_lmmsize);
2190         info->eti_big_lmm = NULL;
2191         info->eti_big_lmmsize = 0;
2192 out_env:
2193         cl_env_put(env, &refcheck);
2194         return rc;
2195 }
2196
2197 static int echo_create_object(const struct lu_env *env, struct echo_device *ed,
2198                               int on_target, struct obdo *oa, void *ulsm,
2199                               int ulsm_nob, struct obd_trans_info *oti)
2200 {
2201         struct echo_object     *eco;
2202         struct echo_client_obd *ec = ed->ed_ec;
2203         struct lov_stripe_md   *lsm = NULL;
2204         int                  rc;
2205         int                  created = 0;
2206
2207         if ((oa->o_valid & OBD_MD_FLID) == 0 && /* no obj id */
2208             (on_target ||                      /* set_stripe */
2209              ec->ec_nstripes != 0)) {      /* LOV */
2210                 CERROR ("No valid oid\n");
2211                 return -EINVAL;
2212         }
2213
2214         rc = echo_alloc_memmd(ed, &lsm);
2215         if (rc < 0) {
2216                 CERROR("Cannot allocate md: rc = %d\n", rc);
2217                 GOTO(failed, rc);
2218         }
2219
2220         if (ulsm != NULL) {
2221                 int i, idx;
2222
2223                 rc = echo_copyin_lsm (ed, lsm, ulsm, ulsm_nob);
2224                 if (rc != 0)
2225                         GOTO(failed, rc);
2226
2227                 if (lsm->lsm_stripe_count == 0)
2228                         lsm->lsm_stripe_count = ec->ec_nstripes;
2229
2230                 if (lsm->lsm_stripe_size == 0)
2231                         lsm->lsm_stripe_size = PAGE_CACHE_SIZE;
2232
2233                 idx = cfs_rand();
2234
2235                 /* setup stripes: indices + default ids if required */
2236                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2237                         if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) == 0)
2238                                 lsm->lsm_oinfo[i]->loi_oi = lsm->lsm_oi;
2239
2240                         lsm->lsm_oinfo[i]->loi_ost_idx =
2241                                 (idx + i) % ec->ec_nstripes;
2242                 }
2243         }
2244
2245         /* setup object ID here for !on_target and LOV hint */
2246         if (oa->o_valid & OBD_MD_FLID) {
2247                 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2248                 lsm->lsm_oi = oa->o_oi;
2249         }
2250
2251         if (ostid_id(&lsm->lsm_oi) == 0)
2252                 ostid_set_id(&lsm->lsm_oi, ++last_object_id);
2253
2254         rc = 0;
2255         if (on_target) {
2256                 /* Only echo objects are allowed to be created */
2257                 LASSERT((oa->o_valid & OBD_MD_FLGROUP) &&
2258                         (ostid_seq(&oa->o_oi) == FID_SEQ_ECHO));
2259                 rc = obd_create(env, ec->ec_exp, oa, &lsm, oti);
2260                 if (rc != 0) {
2261                         CERROR("Cannot create objects: rc = %d\n", rc);
2262                         GOTO(failed, rc);
2263                 }
2264                 created = 1;
2265         }
2266
2267         /* See what object ID we were given */
2268         oa->o_oi = lsm->lsm_oi;
2269         oa->o_valid |= OBD_MD_FLID;
2270
2271         eco = cl_echo_object_find(ed, &lsm);
2272         if (IS_ERR(eco))
2273                 GOTO(failed, rc = PTR_ERR(eco));
2274         cl_echo_object_put(eco);
2275
2276         CDEBUG(D_INFO, "oa oid "DOSTID"\n", POSTID(&oa->o_oi));
2277
2278  failed:
2279         if (created && rc)
2280                 obd_destroy(env, ec->ec_exp, oa, lsm, oti, NULL, NULL);
2281         if (lsm)
2282                 echo_free_memmd(ed, &lsm);
2283         if (rc)
2284                 CERROR("create object failed with: rc = %d\n", rc);
2285         return (rc);
2286 }
2287
2288 static int echo_get_object(struct echo_object **ecop, struct echo_device *ed,
2289                            struct obdo *oa)
2290 {
2291         struct lov_stripe_md   *lsm = NULL;
2292         struct echo_object     *eco;
2293         int                  rc;
2294
2295         if ((oa->o_valid & OBD_MD_FLID) == 0 || ostid_id(&oa->o_oi) == 0) {
2296                 /* disallow use of object id 0 */
2297                 CERROR ("No valid oid\n");
2298                 return -EINVAL;
2299         }
2300
2301         rc = echo_alloc_memmd(ed, &lsm);
2302         if (rc < 0)
2303                 return rc;
2304
2305         lsm->lsm_oi = oa->o_oi;
2306         if (!(oa->o_valid & OBD_MD_FLGROUP))
2307                 ostid_set_seq_echo(&lsm->lsm_oi);
2308
2309         rc = 0;
2310         eco = cl_echo_object_find(ed, &lsm);
2311         if (!IS_ERR(eco))
2312                 *ecop = eco;
2313         else
2314                 rc = PTR_ERR(eco);
2315         if (lsm)
2316                 echo_free_memmd(ed, &lsm);
2317         return rc;
2318 }
2319
2320 static void echo_put_object(struct echo_object *eco)
2321 {
2322         if (cl_echo_object_put(eco))
2323                 CERROR("echo client: drop an object failed");
2324 }
2325
2326 static void
2327 echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp)
2328 {
2329         unsigned long stripe_count;
2330         unsigned long stripe_size;
2331         unsigned long width;
2332         unsigned long woffset;
2333         int        stripe_index;
2334         obd_off       offset;
2335
2336         if (lsm->lsm_stripe_count <= 1)
2337                 return;
2338
2339         offset       = *offp;
2340         stripe_size  = lsm->lsm_stripe_size;
2341         stripe_count = lsm->lsm_stripe_count;
2342
2343         /* width = # bytes in all stripes */
2344         width = stripe_size * stripe_count;
2345
2346         /* woffset = offset within a width; offset = whole number of widths */
2347         woffset = do_div (offset, width);
2348
2349         stripe_index = woffset / stripe_size;
2350
2351         *idp = ostid_id(&lsm->lsm_oinfo[stripe_index]->loi_oi);
2352         *offp = offset * stripe_size + woffset % stripe_size;
2353 }
2354
2355 static void
2356 echo_client_page_debug_setup(struct lov_stripe_md *lsm,
2357                              struct page *page, int rw, obd_id id,
2358                              obd_off offset, obd_off count)
2359 {
2360         char    *addr;
2361         obd_off  stripe_off;
2362         obd_id   stripe_id;
2363         int      delta;
2364
2365         /* no partial pages on the client */
2366         LASSERT(count == PAGE_CACHE_SIZE);
2367
2368         addr = kmap(page);
2369
2370         for (delta = 0; delta < PAGE_CACHE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2371                 if (rw == OBD_BRW_WRITE) {
2372                         stripe_off = offset + delta;
2373                         stripe_id = id;
2374                         echo_get_stripe_off_id(lsm, &stripe_off, &stripe_id);
2375                 } else {
2376                         stripe_off = 0xdeadbeef00c0ffeeULL;
2377                         stripe_id = 0xdeadbeef00c0ffeeULL;
2378                 }
2379                 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
2380                                   stripe_off, stripe_id);
2381         }
2382
2383         kunmap(page);
2384 }
2385
2386 static int echo_client_page_debug_check(struct lov_stripe_md *lsm,
2387                                         struct page *page, obd_id id,
2388                                         obd_off offset, obd_off count)
2389 {
2390         obd_off stripe_off;
2391         obd_id  stripe_id;
2392         char   *addr;
2393         int     delta;
2394         int     rc;
2395         int     rc2;
2396
2397         /* no partial pages on the client */
2398         LASSERT(count == PAGE_CACHE_SIZE);
2399
2400         addr = kmap(page);
2401
2402         for (rc = delta = 0; delta < PAGE_CACHE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2403                 stripe_off = offset + delta;
2404                 stripe_id = id;
2405                 echo_get_stripe_off_id (lsm, &stripe_off, &stripe_id);
2406
2407                 rc2 = block_debug_check("test_brw",
2408                                         addr + delta, OBD_ECHO_BLOCK_SIZE,
2409                                         stripe_off, stripe_id);
2410                 if (rc2 != 0) {
2411                         CERROR ("Error in echo object "LPX64"\n", id);
2412                         rc = rc2;
2413                 }
2414         }
2415
2416         kunmap(page);
2417         return rc;
2418 }
2419
2420 static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa,
2421                             struct echo_object *eco, obd_off offset,
2422                             obd_size count, int async,
2423                             struct obd_trans_info *oti)
2424 {
2425         struct lov_stripe_md   *lsm = eco->eo_lsm;
2426         obd_count              npages;
2427         struct brw_page *pga;
2428         struct brw_page *pgp;
2429         struct page         **pages;
2430         obd_off          off;
2431         int                  i;
2432         int                  rc;
2433         int                  verify;
2434         gfp_t                gfp_mask;
2435         int                  brw_flags = 0;
2436
2437         verify = (ostid_id(&oa->o_oi) != ECHO_PERSISTENT_OBJID &&
2438                   (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
2439                   (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
2440
2441         gfp_mask = ((ostid_id(&oa->o_oi) & 2) == 0) ? GFP_IOFS : GFP_HIGHUSER;
2442
2443         LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
2444         LASSERT(lsm != NULL);
2445         LASSERT(ostid_id(&lsm->lsm_oi) == ostid_id(&oa->o_oi));
2446
2447         if (count <= 0 ||
2448             (count & (~CFS_PAGE_MASK)) != 0)
2449                 return -EINVAL;
2450
2451         /* XXX think again with misaligned I/O */
2452         npages = count >> PAGE_CACHE_SHIFT;
2453
2454         if (rw == OBD_BRW_WRITE)
2455                 brw_flags = OBD_BRW_ASYNC;
2456
2457         OBD_ALLOC(pga, npages * sizeof(*pga));
2458         if (pga == NULL)
2459                 return -ENOMEM;
2460
2461         OBD_ALLOC(pages, npages * sizeof(*pages));
2462         if (pages == NULL) {
2463                 OBD_FREE(pga, npages * sizeof(*pga));
2464                 return -ENOMEM;
2465         }
2466
2467         for (i = 0, pgp = pga, off = offset;
2468              i < npages;
2469              i++, pgp++, off += PAGE_CACHE_SIZE) {
2470
2471                 LASSERT (pgp->pg == NULL);      /* for cleanup */
2472
2473                 rc = -ENOMEM;
2474                 OBD_PAGE_ALLOC(pgp->pg, gfp_mask);
2475                 if (pgp->pg == NULL)
2476                         goto out;
2477
2478                 pages[i] = pgp->pg;
2479                 pgp->count = PAGE_CACHE_SIZE;
2480                 pgp->off = off;
2481                 pgp->flag = brw_flags;
2482
2483                 if (verify)
2484                         echo_client_page_debug_setup(lsm, pgp->pg, rw,
2485                                                      ostid_id(&oa->o_oi), off,
2486                                                      pgp->count);
2487         }
2488
2489         /* brw mode can only be used at client */
2490         LASSERT(ed->ed_next != NULL);
2491         rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async);
2492
2493  out:
2494         if (rc != 0 || rw != OBD_BRW_READ)
2495                 verify = 0;
2496
2497         for (i = 0, pgp = pga; i < npages; i++, pgp++) {
2498                 if (pgp->pg == NULL)
2499                         continue;
2500
2501                 if (verify) {
2502                         int vrc;
2503                         vrc = echo_client_page_debug_check(lsm, pgp->pg,
2504                                                            ostid_id(&oa->o_oi),
2505                                                            pgp->off, pgp->count);
2506                         if (vrc != 0 && rc == 0)
2507                                 rc = vrc;
2508                 }
2509                 OBD_PAGE_FREE(pgp->pg);
2510         }
2511         OBD_FREE(pga, npages * sizeof(*pga));
2512         OBD_FREE(pages, npages * sizeof(*pages));
2513         return rc;
2514 }
2515
2516 static int echo_client_prep_commit(const struct lu_env *env,
2517                                    struct obd_export *exp, int rw,
2518                                    struct obdo *oa, struct echo_object *eco,
2519                                    obd_off offset, obd_size count,
2520                                    obd_size batch, struct obd_trans_info *oti,
2521                                    int async)
2522 {
2523         struct lov_stripe_md *lsm = eco->eo_lsm;
2524         struct obd_ioobj ioo;
2525         struct niobuf_local *lnb;
2526         struct niobuf_remote *rnb;
2527         obd_off off;
2528         obd_size npages, tot_pages;
2529         int i, ret = 0, brw_flags = 0;
2530
2531         if (count <= 0 || (count & (~CFS_PAGE_MASK)) != 0 ||
2532             (lsm != NULL && ostid_id(&lsm->lsm_oi) != ostid_id(&oa->o_oi)))
2533                 return -EINVAL;
2534
2535         npages = batch >> PAGE_CACHE_SHIFT;
2536         tot_pages = count >> PAGE_CACHE_SHIFT;
2537
2538         OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
2539         OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote));
2540
2541         if (lnb == NULL || rnb == NULL)
2542                 GOTO(out, ret = -ENOMEM);
2543
2544         if (rw == OBD_BRW_WRITE && async)
2545                 brw_flags |= OBD_BRW_ASYNC;
2546
2547         obdo_to_ioobj(oa, &ioo);
2548
2549         off = offset;
2550
2551         for(; tot_pages; tot_pages -= npages) {
2552                 int lpages;
2553
2554                 if (tot_pages < npages)
2555                         npages = tot_pages;
2556
2557                 for (i = 0; i < npages; i++, off += PAGE_CACHE_SIZE) {
2558                         rnb[i].offset = off;
2559                         rnb[i].len = PAGE_CACHE_SIZE;
2560                         rnb[i].flags = brw_flags;
2561                 }
2562
2563                 ioo.ioo_bufcnt = npages;
2564                 oti->oti_transno = 0;
2565
2566                 lpages = npages;
2567                 ret = obd_preprw(env, rw, exp, oa, 1, &ioo, rnb, &lpages,
2568                                  lnb, oti, NULL);
2569                 if (ret != 0)
2570                         GOTO(out, ret);
2571                 LASSERT(lpages == npages);
2572
2573                 for (i = 0; i < lpages; i++) {
2574                         struct page *page = lnb[i].page;
2575
2576                         /* read past eof? */
2577                         if (page == NULL && lnb[i].rc == 0)
2578                                 continue;
2579
2580                         if (async)
2581                                 lnb[i].flags |= OBD_BRW_ASYNC;
2582
2583                         if (ostid_id(&oa->o_oi) == ECHO_PERSISTENT_OBJID ||
2584                             (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
2585                             (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
2586                                 continue;
2587
2588                         if (rw == OBD_BRW_WRITE)
2589                                 echo_client_page_debug_setup(lsm, page, rw,
2590                                                             ostid_id(&oa->o_oi),
2591                                                              rnb[i].offset,
2592                                                              rnb[i].len);
2593                         else
2594                                 echo_client_page_debug_check(lsm, page,
2595                                                             ostid_id(&oa->o_oi),
2596                                                              rnb[i].offset,
2597                                                              rnb[i].len);
2598                 }
2599
2600                 ret = obd_commitrw(env, rw, exp, oa, 1, &ioo,
2601                                    rnb, npages, lnb, oti, ret);
2602                 if (ret != 0)
2603                         GOTO(out, ret);
2604
2605                 /* Reset oti otherwise it would confuse ldiskfs. */
2606                 memset(oti, 0, sizeof(*oti));
2607
2608                 /* Reuse env context. */
2609                 lu_context_exit((struct lu_context *)&env->le_ctx);
2610                 lu_context_enter((struct lu_context *)&env->le_ctx);
2611         }
2612
2613 out:
2614         if (lnb)
2615                 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
2616         if (rnb)
2617                 OBD_FREE(rnb, npages * sizeof(struct niobuf_remote));
2618         return ret;
2619 }
2620
2621 static int echo_client_brw_ioctl(const struct lu_env *env, int rw,
2622                                  struct obd_export *exp,
2623                                  struct obd_ioctl_data *data,
2624                                  struct obd_trans_info *dummy_oti)
2625 {
2626         struct obd_device *obd = class_exp2obd(exp);
2627         struct echo_device *ed = obd2echo_dev(obd);
2628         struct echo_client_obd *ec = ed->ed_ec;
2629         struct obdo *oa = &data->ioc_obdo1;
2630         struct echo_object *eco;
2631         int rc;
2632         int async = 1;
2633         long test_mode;
2634
2635         LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2636
2637         rc = echo_get_object(&eco, ed, oa);
2638         if (rc)
2639                 return rc;
2640
2641         oa->o_valid &= ~OBD_MD_FLHANDLE;
2642
2643         /* OFD/obdfilter works only via prep/commit */
2644         test_mode = (long)data->ioc_pbuf1;
2645         if (test_mode == 1)
2646                 async = 0;
2647
2648         if (ed->ed_next == NULL && test_mode != 3) {
2649                 test_mode = 3;
2650                 data->ioc_plen1 = data->ioc_count;
2651         }
2652
2653         /* Truncate batch size to maximum */
2654         if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE)
2655                 data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE;
2656
2657         switch (test_mode) {
2658         case 1:
2659                 /* fall through */
2660         case 2:
2661                 rc = echo_client_kbrw(ed, rw, oa,
2662                                       eco, data->ioc_offset,
2663                                       data->ioc_count, async, dummy_oti);
2664                 break;
2665         case 3:
2666                 rc = echo_client_prep_commit(env, ec->ec_exp, rw, oa,
2667                                              eco, data->ioc_offset,
2668                                              data->ioc_count, data->ioc_plen1,
2669                                              dummy_oti, async);
2670                 break;
2671         default:
2672                 rc = -EINVAL;
2673         }
2674         echo_put_object(eco);
2675         return rc;
2676 }
2677
2678 static int
2679 echo_client_enqueue(struct obd_export *exp, struct obdo *oa,
2680                     int mode, obd_off offset, obd_size nob)
2681 {
2682         struct echo_device     *ed = obd2echo_dev(exp->exp_obd);
2683         struct lustre_handle   *ulh = &oa->o_handle;
2684         struct echo_object     *eco;
2685         obd_off          end;
2686         int                  rc;
2687
2688         if (ed->ed_next == NULL)
2689                 return -EOPNOTSUPP;
2690
2691         if (!(mode == LCK_PR || mode == LCK_PW))
2692                 return -EINVAL;
2693
2694         if ((offset & (~CFS_PAGE_MASK)) != 0 ||
2695             (nob & (~CFS_PAGE_MASK)) != 0)
2696                 return -EINVAL;
2697
2698         rc = echo_get_object (&eco, ed, oa);
2699         if (rc != 0)
2700                 return rc;
2701
2702         end = (nob == 0) ? ((obd_off) -1) : (offset + nob - 1);
2703         rc = cl_echo_enqueue(eco, offset, end, mode, &ulh->cookie);
2704         if (rc == 0) {
2705                 oa->o_valid |= OBD_MD_FLHANDLE;
2706                 CDEBUG(D_INFO, "Cookie is "LPX64"\n", ulh->cookie);
2707         }
2708         echo_put_object(eco);
2709         return rc;
2710 }
2711
2712 static int
2713 echo_client_cancel(struct obd_export *exp, struct obdo *oa)
2714 {
2715         struct echo_device *ed     = obd2echo_dev(exp->exp_obd);
2716         __u64          cookie = oa->o_handle.cookie;
2717
2718         if ((oa->o_valid & OBD_MD_FLHANDLE) == 0)
2719                 return -EINVAL;
2720
2721         CDEBUG(D_INFO, "Cookie is "LPX64"\n", cookie);
2722         return cl_echo_cancel(ed, cookie);
2723 }
2724
2725 static int
2726 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2727                       void *karg, void *uarg)
2728 {
2729         struct obd_device      *obd = exp->exp_obd;
2730         struct echo_device     *ed = obd2echo_dev(obd);
2731         struct echo_client_obd *ec = ed->ed_ec;
2732         struct echo_object     *eco;
2733         struct obd_ioctl_data  *data = karg;
2734         struct obd_trans_info   dummy_oti;
2735         struct lu_env     *env;
2736         struct oti_req_ack_lock *ack_lock;
2737         struct obdo         *oa;
2738         struct lu_fid      fid;
2739         int                  rw = OBD_BRW_READ;
2740         int                  rc = 0;
2741         int                  i;
2742
2743         memset(&dummy_oti, 0, sizeof(dummy_oti));
2744
2745         oa = &data->ioc_obdo1;
2746         if (!(oa->o_valid & OBD_MD_FLGROUP)) {
2747                 oa->o_valid |= OBD_MD_FLGROUP;
2748                 ostid_set_seq_echo(&oa->o_oi);
2749         }
2750
2751         /* This FID is unpacked just for validation at this point */
2752         rc = ostid_to_fid(&fid, &oa->o_oi, 0);
2753         if (rc < 0)
2754                 return rc;
2755
2756         OBD_ALLOC_PTR(env);
2757         if (env == NULL)
2758                 return -ENOMEM;
2759
2760         rc = lu_env_init(env, LCT_DT_THREAD);
2761         if (rc)
2762                 GOTO(out, rc = -ENOMEM);
2763
2764         switch (cmd) {
2765         case OBD_IOC_CREATE:                /* may create echo object */
2766                 if (!capable(CFS_CAP_SYS_ADMIN))
2767                         GOTO (out, rc = -EPERM);
2768
2769                 rc = echo_create_object(env, ed, 1, oa, data->ioc_pbuf1,
2770                                         data->ioc_plen1, &dummy_oti);
2771                 GOTO(out, rc);
2772
2773         case OBD_IOC_ECHO_MD: {
2774                 int count;
2775                 int cmd;
2776                 char *dir = NULL;
2777                 int dirlen;
2778                 __u64 id;
2779
2780                 if (!capable(CFS_CAP_SYS_ADMIN))
2781                         GOTO(out, rc = -EPERM);
2782
2783                 count = data->ioc_count;
2784                 cmd = data->ioc_command;
2785
2786                 id = ostid_id(&data->ioc_obdo2.o_oi);
2787
2788                 dirlen = data->ioc_plen1;
2789                 OBD_ALLOC(dir, dirlen + 1);
2790                 if (dir == NULL)
2791                         GOTO(out, rc = -ENOMEM);
2792
2793                 if (copy_from_user(dir, data->ioc_pbuf1, dirlen)) {
2794                         OBD_FREE(dir, data->ioc_plen1 + 1);
2795                         GOTO(out, rc = -EFAULT);
2796                 }
2797
2798                 rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data);
2799                 OBD_FREE(dir, dirlen + 1);
2800                 GOTO(out, rc);
2801         }
2802         case OBD_IOC_ECHO_ALLOC_SEQ: {
2803                 struct lu_env   *cl_env;
2804                 int           refcheck;
2805                 __u64       seq;
2806                 int           max_count;
2807
2808                 if (!capable(CFS_CAP_SYS_ADMIN))
2809                         GOTO(out, rc = -EPERM);
2810
2811                 cl_env = cl_env_get(&refcheck);
2812                 if (IS_ERR(cl_env))
2813                         GOTO(out, rc = PTR_ERR(cl_env));
2814
2815                 rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG,
2816                                             ECHO_MD_SES_TAG);
2817                 if (rc != 0) {
2818                         cl_env_put(cl_env, &refcheck);
2819                         GOTO(out, rc);
2820                 }
2821
2822                 rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq);
2823                 cl_env_put(cl_env, &refcheck);
2824                 if (rc < 0) {
2825                         CERROR("%s: Can not alloc seq: rc = %d\n",
2826                                obd->obd_name, rc);
2827                         GOTO(out, rc);
2828                 }
2829
2830                 if (copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1))
2831                         return -EFAULT;
2832
2833                 max_count = LUSTRE_METADATA_SEQ_MAX_WIDTH;
2834                 if (copy_to_user(data->ioc_pbuf2, &max_count,
2835                                      data->ioc_plen2))
2836                         return -EFAULT;
2837                 GOTO(out, rc);
2838         }
2839         case OBD_IOC_DESTROY:
2840                 if (!capable(CFS_CAP_SYS_ADMIN))
2841                         GOTO (out, rc = -EPERM);
2842
2843                 rc = echo_get_object(&eco, ed, oa);
2844                 if (rc == 0) {
2845                         rc = obd_destroy(env, ec->ec_exp, oa, eco->eo_lsm,
2846                                          &dummy_oti, NULL, NULL);
2847                         if (rc == 0)
2848                                 eco->eo_deleted = 1;
2849                         echo_put_object(eco);
2850                 }
2851                 GOTO(out, rc);
2852
2853         case OBD_IOC_GETATTR:
2854                 rc = echo_get_object(&eco, ed, oa);
2855                 if (rc == 0) {
2856                         struct obd_info oinfo = { { { 0 } } };
2857                         oinfo.oi_md = eco->eo_lsm;
2858                         oinfo.oi_oa = oa;
2859                         rc = obd_getattr(env, ec->ec_exp, &oinfo);
2860                         echo_put_object(eco);
2861                 }
2862                 GOTO(out, rc);
2863
2864         case OBD_IOC_SETATTR:
2865                 if (!capable(CFS_CAP_SYS_ADMIN))
2866                         GOTO (out, rc = -EPERM);
2867
2868                 rc = echo_get_object(&eco, ed, oa);
2869                 if (rc == 0) {
2870                         struct obd_info oinfo = { { { 0 } } };
2871                         oinfo.oi_oa = oa;
2872                         oinfo.oi_md = eco->eo_lsm;
2873
2874                         rc = obd_setattr(env, ec->ec_exp, &oinfo, NULL);
2875                         echo_put_object(eco);
2876                 }
2877                 GOTO(out, rc);
2878
2879         case OBD_IOC_BRW_WRITE:
2880                 if (!capable(CFS_CAP_SYS_ADMIN))
2881                         GOTO (out, rc = -EPERM);
2882
2883                 rw = OBD_BRW_WRITE;
2884                 /* fall through */
2885         case OBD_IOC_BRW_READ:
2886                 rc = echo_client_brw_ioctl(env, rw, exp, data, &dummy_oti);
2887                 GOTO(out, rc);
2888
2889         case ECHO_IOC_GET_STRIPE:
2890                 rc = echo_get_object(&eco, ed, oa);
2891                 if (rc == 0) {
2892                         rc = echo_copyout_lsm(eco->eo_lsm, data->ioc_pbuf1,
2893                                               data->ioc_plen1);
2894                         echo_put_object(eco);
2895                 }
2896                 GOTO(out, rc);
2897
2898         case ECHO_IOC_SET_STRIPE:
2899                 if (!capable(CFS_CAP_SYS_ADMIN))
2900                         GOTO (out, rc = -EPERM);
2901
2902                 if (data->ioc_pbuf1 == NULL) {  /* unset */
2903                         rc = echo_get_object(&eco, ed, oa);
2904                         if (rc == 0) {
2905                                 eco->eo_deleted = 1;
2906                                 echo_put_object(eco);
2907                         }
2908                 } else {
2909                         rc = echo_create_object(env, ed, 0, oa,
2910                                                 data->ioc_pbuf1,
2911                                                 data->ioc_plen1, &dummy_oti);
2912                 }
2913                 GOTO (out, rc);
2914
2915         case ECHO_IOC_ENQUEUE:
2916                 if (!capable(CFS_CAP_SYS_ADMIN))
2917                         GOTO (out, rc = -EPERM);
2918
2919                 rc = echo_client_enqueue(exp, oa,
2920                                          data->ioc_conn1, /* lock mode */
2921                                          data->ioc_offset,
2922                                          data->ioc_count);/*extent*/
2923                 GOTO (out, rc);
2924
2925         case ECHO_IOC_CANCEL:
2926                 rc = echo_client_cancel(exp, oa);
2927                 GOTO (out, rc);
2928
2929         default:
2930                 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
2931                 GOTO (out, rc = -ENOTTY);
2932         }
2933
2934 out:
2935         lu_env_fini(env);
2936         OBD_FREE_PTR(env);
2937
2938         /* XXX this should be in a helper also called by target_send_reply */
2939         for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4;
2940              i++, ack_lock++) {
2941                 if (!ack_lock->mode)
2942                         break;
2943                 ldlm_lock_decref(&ack_lock->lock, ack_lock->mode);
2944         }
2945
2946         return rc;
2947 }
2948
2949 static int echo_client_setup(const struct lu_env *env,
2950                              struct obd_device *obddev, struct lustre_cfg *lcfg)
2951 {
2952         struct echo_client_obd *ec = &obddev->u.echo_client;
2953         struct obd_device *tgt;
2954         struct obd_uuid echo_uuid = { "ECHO_UUID" };
2955         struct obd_connect_data *ocd = NULL;
2956         int rc;
2957
2958         if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2959                 CERROR("requires a TARGET OBD name\n");
2960                 return -EINVAL;
2961         }
2962
2963         tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
2964         if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
2965                 CERROR("device not attached or not set up (%s)\n",
2966                        lustre_cfg_string(lcfg, 1));
2967                 return -EINVAL;
2968         }
2969
2970         spin_lock_init(&ec->ec_lock);
2971         INIT_LIST_HEAD (&ec->ec_objects);
2972         INIT_LIST_HEAD (&ec->ec_locks);
2973         ec->ec_unique = 0;
2974         ec->ec_nstripes = 0;
2975
2976         if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
2977                 lu_context_tags_update(ECHO_MD_CTX_TAG);
2978                 lu_session_tags_update(ECHO_MD_SES_TAG);
2979                 return 0;
2980         }
2981
2982         OBD_ALLOC(ocd, sizeof(*ocd));
2983         if (ocd == NULL) {
2984                 CERROR("Can't alloc ocd connecting to %s\n",
2985                        lustre_cfg_string(lcfg, 1));
2986                 return -ENOMEM;
2987         }
2988
2989         ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL |
2990                                  OBD_CONNECT_BRW_SIZE |
2991                                  OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 |
2992                                  OBD_CONNECT_64BITHASH | OBD_CONNECT_LVB_TYPE |
2993                                  OBD_CONNECT_FID;
2994         ocd->ocd_brw_size = DT_MAX_BRW_SIZE;
2995         ocd->ocd_version = LUSTRE_VERSION_CODE;
2996         ocd->ocd_group = FID_SEQ_ECHO;
2997
2998         rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL);
2999         if (rc == 0) {
3000                 /* Turn off pinger because it connects to tgt obd directly. */
3001                 spin_lock(&tgt->obd_dev_lock);
3002                 list_del_init(&ec->ec_exp->exp_obd_chain_timed);
3003                 spin_unlock(&tgt->obd_dev_lock);
3004         }
3005
3006         OBD_FREE(ocd, sizeof(*ocd));
3007
3008         if (rc != 0) {
3009                 CERROR("fail to connect to device %s\n",
3010                        lustre_cfg_string(lcfg, 1));
3011                 return (rc);
3012         }
3013
3014         return rc;
3015 }
3016
3017 static int echo_client_cleanup(struct obd_device *obddev)
3018 {
3019         struct echo_device *ed = obd2echo_dev(obddev);
3020         struct echo_client_obd *ec = &obddev->u.echo_client;
3021         int rc;
3022
3023         /*Do nothing for Metadata echo client*/
3024         if (ed == NULL )
3025                 return 0;
3026
3027         if (ed->ed_next_ismd) {
3028                 lu_context_tags_clear(ECHO_MD_CTX_TAG);
3029                 lu_session_tags_clear(ECHO_MD_SES_TAG);
3030                 return 0;
3031         }
3032
3033         if (!list_empty(&obddev->obd_exports)) {
3034                 CERROR("still has clients!\n");
3035                 return -EBUSY;
3036         }
3037
3038         LASSERT(atomic_read(&ec->ec_exp->exp_refcount) > 0);
3039         rc = obd_disconnect(ec->ec_exp);
3040         if (rc != 0)
3041                 CERROR("fail to disconnect device: %d\n", rc);
3042
3043         return rc;
3044 }
3045
3046 static int echo_client_connect(const struct lu_env *env,
3047                                struct obd_export **exp,
3048                                struct obd_device *src, struct obd_uuid *cluuid,
3049                                struct obd_connect_data *data, void *localdata)
3050 {
3051         int             rc;
3052         struct lustre_handle conn = { 0 };
3053
3054         rc = class_connect(&conn, src, cluuid);
3055         if (rc == 0) {
3056                 *exp = class_conn2export(&conn);
3057         }
3058
3059         return rc;
3060 }
3061
3062 static int echo_client_disconnect(struct obd_export *exp)
3063 {
3064 #if 0
3065         struct obd_device      *obd;
3066         struct echo_client_obd *ec;
3067         struct ec_lock   *ecl;
3068 #endif
3069         int                  rc;
3070
3071         if (exp == NULL)
3072                 GOTO(out, rc = -EINVAL);
3073
3074 #if 0
3075         obd = exp->exp_obd;
3076         ec = &obd->u.echo_client;
3077
3078         /* no more contention on export's lock list */
3079         while (!list_empty (&exp->exp_ec_data.eced_locks)) {
3080                 ecl = list_entry (exp->exp_ec_data.eced_locks.next,
3081                                       struct ec_lock, ecl_exp_chain);
3082                 list_del (&ecl->ecl_exp_chain);
3083
3084                 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm,
3085                                  ecl->ecl_mode, &ecl->ecl_lock_handle);
3086
3087                 CDEBUG (D_INFO, "Cancel lock on object "LPX64" on disconnect "
3088                         "(%d)\n", ecl->ecl_object->eco_id, rc);
3089
3090                 echo_put_object (ecl->ecl_object);
3091                 OBD_FREE (ecl, sizeof (*ecl));
3092         }
3093 #endif
3094
3095         rc = class_disconnect(exp);
3096         GOTO(out, rc);
3097  out:
3098         return rc;
3099 }
3100
3101 static struct obd_ops echo_client_obd_ops = {
3102         .o_owner       = THIS_MODULE,
3103
3104 #if 0
3105         .o_setup       = echo_client_setup,
3106         .o_cleanup     = echo_client_cleanup,
3107 #endif
3108
3109         .o_iocontrol   = echo_client_iocontrol,
3110         .o_connect     = echo_client_connect,
3111         .o_disconnect  = echo_client_disconnect
3112 };
3113
3114 int echo_client_init(void)
3115 {
3116         struct lprocfs_static_vars lvars = { 0 };
3117         int rc;
3118
3119         lprocfs_echo_init_vars(&lvars);
3120
3121         rc = lu_kmem_init(echo_caches);
3122         if (rc == 0) {
3123                 rc = class_register_type(&echo_client_obd_ops, NULL,
3124                                          lvars.module_vars,
3125                                          LUSTRE_ECHO_CLIENT_NAME,
3126                                          &echo_device_type);
3127                 if (rc)
3128                         lu_kmem_fini(echo_caches);
3129         }
3130         return rc;
3131 }
3132
3133 void echo_client_exit(void)
3134 {
3135         class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
3136         lu_kmem_fini(echo_caches);
3137 }
3138
3139 static int __init obdecho_init(void)
3140 {
3141         struct lprocfs_static_vars lvars;
3142         int rc;
3143
3144         LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
3145
3146         LASSERT(PAGE_CACHE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
3147
3148         lprocfs_echo_init_vars(&lvars);
3149
3150
3151         rc = echo_client_init();
3152
3153         return rc;
3154 }
3155
3156 static void /*__exit*/ obdecho_exit(void)
3157 {
3158         echo_client_exit();
3159
3160 }
3161
3162 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3163 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
3164 MODULE_LICENSE("GPL");
3165 MODULE_VERSION(LUSTRE_VERSION_STRING);
3166
3167 module_init(obdecho_init);
3168 module_exit(obdecho_exit);
3169
3170 /** @} echo_client */