lustre/obdecho/echo.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2010, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/obdecho/echo.c
  33  *
  34  * Author: Peter Braam <braam@clusterfs.com>
  35  * Author: Andreas Dilger <adilger@clusterfs.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_ECHO
  39
  40 #include <obd_support.h>
  41 #include <obd_class.h>
  42 #include <lustre_dlm.h>
  43 #include <lprocfs_status.h>
  44
  45 #include "echo_internal.h"
  46
  47 /*
  48  * The echo objid needs to be below 2^32, because regular FID numbers are
  49  * limited to 2^32 objects in f_oid for the FID_SEQ_ECHO range. b=23335
  50  */
  51 #define ECHO_INIT_OID        0x10000000ULL
  52 #define ECHO_HANDLE_MAGIC    0xabcd0123fedc9876ULL
  53
  54 #define ECHO_PERSISTENT_PAGES (ECHO_PERSISTENT_SIZE >> PAGE_SHIFT)
  55 static struct page *echo_persistent_pages[ECHO_PERSISTENT_PAGES];
  56
  57 enum {
  58         LPROC_ECHO_READ_BYTES = 1,
  59         LPROC_ECHO_WRITE_BYTES = 2,
  60         LPROC_ECHO_LAST = LPROC_ECHO_WRITE_BYTES + 1
  61 };
  62
  63 struct echo_srv_device {
  64         struct lu_device esd_dev;
  65         struct lu_target esd_lut;
  66 };
  67
  68 static inline struct echo_srv_device *echo_srv_dev(struct lu_device *d)
  69 {
  70         return container_of_safe(d, struct echo_srv_device, esd_dev);
  71 }
  72
  73 static inline struct obd_device *echo_srv_obd(struct echo_srv_device *esd)
  74 {
  75         return esd->esd_dev.ld_obd;
  76 }
  77
  78 static int echo_connect(const struct lu_env *env,
  79                         struct obd_export **exp, struct obd_device *obd,
  80                         struct obd_uuid *cluuid, struct obd_connect_data *data,
  81                         void *localdata)
  82 {
  83         struct lustre_handle conn = { 0 };
  84         int rc;
  85
  86         data->ocd_connect_flags &= ECHO_CONNECT_SUPPORTED;
  87
  88         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
  89                 data->ocd_connect_flags2 &= ECHO_CONNECT_SUPPORTED2;
  90
  91         rc = class_connect(&conn, obd, cluuid);
  92         if (rc) {
  93                 CERROR("can't connect %d\n", rc);
  94                 return rc;
  95         }
  96         *exp = class_conn2export(&conn);
  97
  98         return 0;
  99 }
 100
 101 static int echo_disconnect(struct obd_export *exp)
 102 {
 103         LASSERT(exp != NULL);
 104
 105         return server_disconnect_export(exp);
 106 }
 107
 108 static int echo_init_export(struct obd_export *exp)
 109 {
 110         return ldlm_init_export(exp);
 111 }
 112
 113 static int echo_destroy_export(struct obd_export *exp)
 114 {
 115         ENTRY;
 116
 117         target_destroy_export(exp);
 118         ldlm_destroy_export(exp);
 119
 120         RETURN(0);
 121 }
 122
 123 static u64 echo_next_id(struct obd_device *obd)
 124 {
 125         u64 id;
 126
 127         spin_lock(&obd->u.echo.eo_lock);
 128         id = ++obd->u.echo.eo_lastino;
 129         spin_unlock(&obd->u.echo.eo_lock);
 130
 131         return id;
 132 }
 133
 134 static void
 135 echo_page_debug_setup(struct page *page, int rw, u64 id,
 136                       __u64 offset, int len)
 137 {
 138         int   page_offset = offset & ~PAGE_MASK;
 139         char *addr        = ((char *)kmap(page)) + page_offset;
 140
 141         if (len % OBD_ECHO_BLOCK_SIZE != 0)
 142                 CERROR("Unexpected block size %d\n", len);
 143
 144         while (len > 0) {
 145                 if (rw & OBD_BRW_READ)
 146                         block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE,
 147                                           offset, id);
 148                 else
 149                         block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE,
 150                                           0xecc0ecc0ecc0ecc0ULL,
 151                                           0xecc0ecc0ecc0ecc0ULL);
 152
 153                 addr   += OBD_ECHO_BLOCK_SIZE;
 154                 offset += OBD_ECHO_BLOCK_SIZE;
 155                 len    -= OBD_ECHO_BLOCK_SIZE;
 156         }
 157
 158         kunmap(page);
 159 }
 160
 161 static int
 162 echo_page_debug_check(struct page *page, u64 id,
 163                       __u64 offset, int len)
 164 {
 165         int   page_offset = offset & ~PAGE_MASK;
 166         char *addr        = ((char *)kmap(page)) + page_offset;
 167         int   rc          = 0;
 168         int   rc2;
 169
 170         if (len % OBD_ECHO_BLOCK_SIZE != 0)
 171                 CERROR("Unexpected block size %d\n", len);
 172
 173         while (len > 0) {
 174                 rc2 = block_debug_check("echo", addr, OBD_ECHO_BLOCK_SIZE,
 175                                         offset, id);
 176
 177                 if (rc2 != 0 && rc == 0)
 178                         rc = rc2;
 179
 180                 addr   += OBD_ECHO_BLOCK_SIZE;
 181                 offset += OBD_ECHO_BLOCK_SIZE;
 182                 len    -= OBD_ECHO_BLOCK_SIZE;
 183         }
 184
 185         kunmap(page);
 186
 187         return rc;
 188 }
 189
 190 static int echo_map_nb_to_lb(struct obdo *oa, struct obd_ioobj *obj,
 191                              struct niobuf_remote *nb, int *pages,
 192                              struct niobuf_local *lb, int cmd, int *left)
 193 {
 194         gfp_t gfp_mask = (ostid_id(&obj->ioo_oid) & 1) ?
 195                         GFP_HIGHUSER : GFP_KERNEL;
 196         int ispersistent = ostid_id(&obj->ioo_oid) == ECHO_PERSISTENT_OBJID;
 197         int debug_setup = (!ispersistent &&
 198                            (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
 199                            (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
 200         struct niobuf_local *res = lb;
 201         u64 offset = nb->rnb_offset;
 202         int len = nb->rnb_len;
 203
 204         while (len > 0) {
 205                 int plen = PAGE_SIZE - (offset & (PAGE_SIZE - 1));
 206
 207                 if (len < plen)
 208                         plen = len;
 209
 210                 /* check for local buf overflow */
 211                 if (*left == 0)
 212                         return -EINVAL;
 213
 214                 res->lnb_file_offset = offset;
 215                 res->lnb_len = plen;
 216                 LASSERT((res->lnb_file_offset & ~PAGE_MASK) +
 217                         res->lnb_len <= PAGE_SIZE);
 218
 219                 if (ispersistent &&
 220                     ((res->lnb_file_offset >> PAGE_SHIFT) <
 221                       ECHO_PERSISTENT_PAGES)) {
 222                         res->lnb_page =
 223                                 echo_persistent_pages[res->lnb_file_offset >>
 224                                                       PAGE_SHIFT];
 225                         /* Take extra ref so __free_pages() can be called OK */
 226                         get_page(res->lnb_page);
 227                 } else {
 228                         res->lnb_page = alloc_page(gfp_mask);
 229                         if (!res->lnb_page) {
 230                                 CERROR("can't get page for id " DOSTID"\n",
 231                                        POSTID(&obj->ioo_oid));
 232                                 return -ENOMEM;
 233                         }
 234                         /* set mapping so page is not considered encrypted */
 235                         res->lnb_page->mapping = ECHO_MAPPING_UNENCRYPTED;
 236                 }
 237
 238                 CDEBUG(D_PAGE, "$$$$ get page %p @ %llu for %d\n",
 239                        res->lnb_page, res->lnb_file_offset, res->lnb_len);
 240
 241                 if (cmd & OBD_BRW_READ)
 242                         res->lnb_rc = res->lnb_len;
 243
 244                 if (debug_setup)
 245                         echo_page_debug_setup(res->lnb_page, cmd,
 246                                               ostid_id(&obj->ioo_oid),
 247                                               res->lnb_file_offset,
 248                                               res->lnb_len);
 249
 250                 offset += plen;
 251                 len -= plen;
 252                 res++;
 253
 254                 (*left)--;
 255                 (*pages)++;
 256         }
 257
 258         return 0;
 259 }
 260
 261 static int echo_finalize_lb(struct obdo *oa, struct obd_ioobj *obj,
 262                             struct niobuf_remote *rb, int *pgs,
 263                             struct niobuf_local *lb, int verify)
 264 {
 265         struct niobuf_local *res = lb;
 266         u64 start = rb->rnb_offset >> PAGE_SHIFT;
 267         u64 end   = (rb->rnb_offset + rb->rnb_len + PAGE_SIZE - 1) >>
 268                     PAGE_SHIFT;
 269         int     count  = (int)(end - start);
 270         int     rc     = 0;
 271         int     i;
 272
 273         for (i = 0; i < count; i++, (*pgs) ++, res++) {
 274                 struct page *page = res->lnb_page;
 275                 void       *addr;
 276
 277                 if (!page) {
 278                         CERROR("null page objid %llu:%p, buf %d/%d\n",
 279                                ostid_id(&obj->ioo_oid), page, i,
 280                                obj->ioo_bufcnt);
 281                         return -EFAULT;
 282                 }
 283
 284                 addr = kmap(page);
 285
 286                 CDEBUG(D_PAGE, "$$$$ use page %p, addr %p@%llu\n",
 287                        res->lnb_page, addr, res->lnb_file_offset);
 288
 289                 if (verify) {
 290                         int vrc = echo_page_debug_check(page,
 291                                                         ostid_id(&obj->ioo_oid),
 292                                                         res->lnb_file_offset,
 293                                                         res->lnb_len);
 294                         /* check all the pages always */
 295                         if (vrc != 0 && rc == 0)
 296                                 rc = vrc;
 297                 }
 298
 299                 kunmap(page);
 300                 /* NB see comment above regarding persistent pages */
 301                 __free_page(page);
 302         }
 303
 304         return rc;
 305 }
 306
 307 static int echo_preprw(const struct lu_env *env, int cmd,
 308                        struct obd_export *export, struct obdo *oa,
 309                        int objcount, struct obd_ioobj *obj,
 310                        struct niobuf_remote *nb, int *pages,
 311                        struct niobuf_local *res)
 312 {
 313         struct obd_device *obd;
 314         int tot_bytes = 0;
 315         int rc = 0;
 316         int i, left;
 317
 318         ENTRY;
 319
 320         obd = export->exp_obd;
 321         if (!obd)
 322                 RETURN(-EINVAL);
 323
 324         /* Temp fix to stop falling foul of osc_announce_cached() */
 325         oa->o_valid &= ~(OBD_MD_FLBLOCKS | OBD_MD_FLGRANT);
 326
 327         memset(res, 0, sizeof(*res) * *pages);
 328
 329         CDEBUG(D_PAGE, "%s %d obdos with %d IOs\n",
 330                cmd == OBD_BRW_READ ? "reading" : "writing", objcount, *pages);
 331
 332         left = *pages;
 333         *pages = 0;
 334
 335         for (i = 0; i < objcount; i++, obj++) {
 336                 int j;
 337
 338                 for (j = 0 ; j < obj->ioo_bufcnt ; j++, nb++) {
 339                         rc = echo_map_nb_to_lb(oa, obj, nb, pages,
 340                                                res + *pages, cmd, &left);
 341                         if (rc)
 342                                 GOTO(preprw_cleanup, rc);
 343
 344                         tot_bytes += nb->rnb_len;
 345                 }
 346         }
 347
 348         atomic_add(*pages, &obd->u.echo.eo_prep);
 349
 350         if (cmd & OBD_BRW_READ)
 351                 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_READ_BYTES,
 352                                     tot_bytes);
 353         else
 354                 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
 355                                     tot_bytes);
 356
 357         CDEBUG(D_PAGE, "%d pages allocated after prep\n",
 358                atomic_read(&obd->u.echo.eo_prep));
 359
 360         RETURN(0);
 361
 362 preprw_cleanup:
 363         /*
 364          * It is possible that we would rather handle errors by  allow
 365          * any already-set-up pages to complete, rather than tearing them
 366          * all down again.  I believe that this is what the in-kernel
 367          * prep/commit operations do.
 368          */
 369         CERROR("cleaning up %u pages (%d obdos)\n", *pages, objcount);
 370         for (i = 0; i < *pages; i++) {
 371                 kunmap(res[i].lnb_page);
 372                 /*
 373                  * NB if this is a persistent page, __free_page() will just
 374                  * lose the extra ref gained above
 375                  */
 376                 __free_page(res[i].lnb_page);
 377                 res[i].lnb_page = NULL;
 378                 atomic_dec(&obd->u.echo.eo_prep);
 379         }
 380
 381         return rc;
 382 }
 383
 384 static int echo_commitrw(const struct lu_env *env, int cmd,
 385                          struct obd_export *export, struct obdo *oa,
 386                          int objcount, struct obd_ioobj *obj,
 387                          struct niobuf_remote *rb, int niocount,
 388                          struct niobuf_local *res, int rc)
 389 {
 390         struct obd_device *obd;
 391         int pgs = 0;
 392         int i;
 393
 394         ENTRY;
 395
 396         obd = export->exp_obd;
 397         if (!obd)
 398                 RETURN(-EINVAL);
 399
 400         if (rc)
 401                 GOTO(commitrw_cleanup, rc);
 402
 403         if ((cmd & OBD_BRW_RWMASK) == OBD_BRW_READ) {
 404                 CDEBUG(D_PAGE, "reading %d obdos with %d IOs\n",
 405                        objcount, niocount);
 406         } else {
 407                 CDEBUG(D_PAGE, "writing %d obdos with %d IOs\n",
 408                        objcount, niocount);
 409         }
 410
 411         if (niocount && !res) {
 412                 CERROR("NULL res niobuf with niocount %d\n", niocount);
 413                 RETURN(-EINVAL);
 414         }
 415
 416         for (i = 0; i < objcount; i++, obj++) {
 417                 int verify = (rc == 0 &&
 418                              ostid_id(&obj->ioo_oid) != ECHO_PERSISTENT_OBJID &&
 419                               (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
 420                               (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
 421                 int j;
 422
 423                 for (j = 0 ; j < obj->ioo_bufcnt ; j++, rb++) {
 424                         int vrc = echo_finalize_lb(oa, obj, rb, &pgs, &res[pgs],
 425                                                    verify);
 426                         if (vrc == 0)
 427                                 continue;
 428
 429                         if (vrc == -EFAULT)
 430                                 GOTO(commitrw_cleanup, rc = vrc);
 431
 432                         if (rc == 0)
 433                                 rc = vrc;
 434                 }
 435         }
 436
 437         atomic_sub(pgs, &obd->u.echo.eo_prep);
 438
 439         CDEBUG(D_PAGE, "%d pages remain after commit\n",
 440                atomic_read(&obd->u.echo.eo_prep));
 441         RETURN(rc);
 442
 443 commitrw_cleanup:
 444         atomic_sub(pgs, &obd->u.echo.eo_prep);
 445
 446         CERROR("cleaning up %d pages (%d obdos)\n",
 447                niocount - pgs - 1, objcount);
 448
 449         while (pgs < niocount) {
 450                 struct page *page = res[pgs++].lnb_page;
 451
 452                 if (!page)
 453                         continue;
 454
 455                 /* NB see comment above regarding persistent pages */
 456                 __free_page(page);
 457                 atomic_dec(&obd->u.echo.eo_prep);
 458         }
 459         return rc;
 460 }
 461
 462 LPROC_SEQ_FOPS_RO_TYPE(echo, uuid);
 463 static struct lprocfs_vars lprocfs_echo_obd_vars[] = {
 464         { .name =       "uuid",
 465           .fops =       &echo_uuid_fops         },
 466         { NULL }
 467 };
 468
 469 const struct obd_ops echo_obd_ops = {
 470         .o_owner           = THIS_MODULE,
 471         .o_connect         = echo_connect,
 472         .o_disconnect      = echo_disconnect,
 473         .o_init_export     = echo_init_export,
 474         .o_destroy_export  = echo_destroy_export,
 475         .o_preprw          = echo_preprw,
 476         .o_commitrw        = echo_commitrw,
 477 };
 478
 479 /**
 480  * Echo Server request handler for OST_CREATE RPC.
 481  *
 482  * This is part of request processing. Its simulates the object
 483  * creation on OST.
 484  *
 485  * \param[in] tsi       target session environment for this request
 486  *
 487  * \retval              0 if successful
 488  * \retval              negative value on error
 489  */
 490 static int esd_create_hdl(struct tgt_session_info *tsi)
 491 {
 492         const struct obdo *oa = &tsi->tsi_ost_body->oa;
 493         struct obd_device *obd = tsi->tsi_exp->exp_obd;
 494         struct ost_body *repbody;
 495         struct obdo *rep_oa;
 496
 497         ENTRY;
 498
 499         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
 500         if (!repbody)
 501                 RETURN(-ENOMEM);
 502
 503         if (!(oa->o_mode & S_IFMT)) {
 504                 CERROR("%s: no type is set in obdo!\n",
 505                        tsi->tsi_exp->exp_obd->obd_name);
 506                 RETURN(-ENOENT);
 507         }
 508
 509         if (!(oa->o_valid & OBD_MD_FLTYPE)) {
 510                 CERROR("%s: invalid o_valid in obdo: %#llx\n",
 511                        tsi->tsi_exp->exp_obd->obd_name, oa->o_valid);
 512                 RETURN(-EINVAL);
 513         }
 514
 515         rep_oa = &repbody->oa;
 516
 517         if (!fid_seq_is_echo(ostid_seq(&oa->o_oi))) {
 518                 CERROR("%s: invalid seq %#llx\n",
 519                        tsi->tsi_exp->exp_obd->obd_name, ostid_seq(&oa->o_oi));
 520                 return -EINVAL;
 521         }
 522
 523         ostid_set_seq_echo(&rep_oa->o_oi);
 524         ostid_set_id(&rep_oa->o_oi, echo_next_id(obd));
 525
 526         CDEBUG(D_INFO, "%s: Create object "DOSTID"\n",
 527                tsi->tsi_exp->exp_obd->obd_name, POSTID(&rep_oa->o_oi));
 528
 529         rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
 530
 531         RETURN(0);
 532 }
 533
 534 /**
 535  * Echo Server request handler for OST_DESTROY RPC.
 536  *
 537  * This is Echo Server part of request handling. It simulates the objects
 538  * destroy on OST.
 539  *
 540  * \param[in] tsi       target session environment for this request
 541  *
 542  * \retval              0 if successful
 543  * \retval              negative value on error
 544  */
 545 static int esd_destroy_hdl(struct tgt_session_info *tsi)
 546 {
 547         const struct obdo *oa = &tsi->tsi_ost_body->oa;
 548         struct obd_device *obd = tsi->tsi_exp->exp_obd;
 549         struct ost_body *repbody;
 550         u64 oid;
 551
 552         ENTRY;
 553
 554         oid = ostid_id(&oa->o_oi);
 555         LASSERT(oid != 0);
 556
 557         if (!(oa->o_valid & OBD_MD_FLID)) {
 558                 CERROR("%s: obdo missing FLID valid flag: %#llx\n",
 559                        tsi->tsi_exp->exp_obd->obd_name, oa->o_valid);
 560                 RETURN(-EINVAL);
 561         }
 562
 563         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
 564
 565         if (ostid_id(&oa->o_oi) > obd->u.echo.eo_lastino ||
 566             ostid_id(&oa->o_oi) < ECHO_INIT_OID) {
 567                 CERROR("%s: bad objid to destroy: "DOSTID"\n",
 568                        tsi->tsi_exp->exp_obd->obd_name, POSTID(&oa->o_oi));
 569                 RETURN(-EINVAL);
 570         }
 571
 572         CDEBUG(D_INFO, "%s: Destroy object "DOSTID"\n",
 573                tsi->tsi_exp->exp_obd->obd_name, POSTID(&oa->o_oi));
 574
 575         repbody->oa.o_oi = oa->o_oi;
 576         RETURN(0);
 577 }
 578
 579 /**
 580  * Echo Server request handler for OST_GETATTR RPC.
 581  *
 582  * This is Echo Server part of request handling. It returns an object
 583  * attributes to the client. All objects have the same attributes in
 584  * Echo Server.
 585  *
 586  * \param[in] tsi       target session environment for this request
 587  *
 588  * \retval              0 if successful
 589  * \retval              negative value on error
 590  */
 591 static int esd_getattr_hdl(struct tgt_session_info *tsi)
 592 {
 593         const struct obdo *oa = &tsi->tsi_ost_body->oa;
 594         struct obd_device *obd = tsi->tsi_exp->exp_obd;
 595         struct ost_body *repbody;
 596
 597         ENTRY;
 598
 599         if (!(oa->o_valid & OBD_MD_FLID)) {
 600                 CERROR("%s: obdo missing FLID valid flag: %#llx\n",
 601                        tsi->tsi_exp->exp_obd->obd_name, oa->o_valid);
 602                 RETURN(-EINVAL);
 603         }
 604
 605         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
 606         if (!repbody)
 607                 RETURN(-ENOMEM);
 608
 609         repbody->oa.o_oi = oa->o_oi;
 610         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
 611
 612         obdo_cpy_md(&repbody->oa, &obd->u.echo.eo_oa, oa->o_valid);
 613
 614         repbody->oa.o_valid |= OBD_MD_FLFLAGS;
 615         repbody->oa.o_flags = OBD_FL_FLUSH;
 616
 617         RETURN(0);
 618 }
 619
 620 /**
 621  * Echo Server request handler for OST_SETATTR RPC.
 622  *
 623  * This is Echo Server part of request handling. It sets common
 624  * attributes from request to the Echo Server objects.
 625  *
 626  * \param[in] tsi       target session environment for this request
 627  *
 628  * \retval              0 if successful
 629  * \retval              negative value on error
 630  */
 631 static int esd_setattr_hdl(struct tgt_session_info *tsi)
 632 {
 633         struct ost_body *body = tsi->tsi_ost_body;
 634         struct obd_device *obd = tsi->tsi_exp->exp_obd;
 635         struct ost_body *repbody;
 636
 637         ENTRY;
 638
 639         if (!(body->oa.o_valid & OBD_MD_FLID)) {
 640                 CERROR("%s: obdo missing FLID valid flag: %#llx\n",
 641                        tsi->tsi_exp->exp_obd->obd_name,
 642                        body->oa.o_valid);
 643                 RETURN(-EINVAL);
 644         }
 645
 646         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
 647         if (!repbody)
 648                 RETURN(-ENOMEM);
 649
 650         repbody->oa.o_oi = body->oa.o_oi;
 651         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
 652
 653         obd->u.echo.eo_oa = body->oa;
 654
 655         RETURN(0);
 656 }
 657
 658 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
 659 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
 660 #define OST_BRW_READ    OST_READ
 661 #define OST_BRW_WRITE   OST_WRITE
 662
 663 /**
 664  * Table of Echo Server specific request handlers
 665  *
 666  * This table contains all opcodes accepted by Echo Server and
 667  * specifies handlers for them. The tgt_request_handler()
 668  * uses such table from each target to process incoming
 669  * requests.
 670  */
 671 static struct tgt_handler esd_tgt_handlers[] = {
 672 TGT_RPC_HANDLER(OST_FIRST_OPC, 0, OST_CONNECT, tgt_connect,
 673                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
 674 TGT_RPC_HANDLER(OST_FIRST_OPC, 0, OST_DISCONNECT, tgt_disconnect,
 675                 &RQF_OST_DISCONNECT, LUSTRE_OBD_VERSION),
 676 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_GETATTR, esd_getattr_hdl),
 677 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_SETATTR,
 678             esd_setattr_hdl),
 679 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE, OST_CREATE, esd_create_hdl),
 680 TGT_OST_HDL(HAS_REPLY | IS_MUTABLE, OST_DESTROY, esd_destroy_hdl),
 681 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read),
 682 TGT_OST_HDL(HAS_BODY | IS_MUTABLE, OST_BRW_WRITE, tgt_brw_write),
 683 };
 684
 685 static struct tgt_opc_slice esd_common_slice[] = {
 686         {
 687                 .tos_opc_start  = OST_FIRST_OPC,
 688                 .tos_opc_end    = OST_LAST_OPC,
 689                 .tos_hs         = esd_tgt_handlers
 690         },
 691         {
 692                 .tos_opc_start  = OBD_FIRST_OPC,
 693                 .tos_opc_end    = OBD_LAST_OPC,
 694                 .tos_hs         = tgt_obd_handlers
 695         },
 696         {
 697                 .tos_opc_start  = LDLM_FIRST_OPC,
 698                 .tos_opc_end    = LDLM_LAST_OPC,
 699                 .tos_hs         = tgt_dlm_handlers
 700         },
 701         {
 702                 .tos_opc_start  = SEC_FIRST_OPC,
 703                 .tos_opc_end    = SEC_LAST_OPC,
 704                 .tos_hs         = tgt_sec_ctx_handlers
 705         },
 706         {
 707                 .tos_hs         = NULL
 708         }
 709 };
 710
 711 /**
 712  * lu_device_operations matrix for ECHO SRV device is NULL,
 713  * this device is just serving incoming requests immediately
 714  * without building a stack of lu_devices.
 715  */
 716 static struct lu_device_operations echo_srv_lu_ops = { 0 };
 717
 718 /**
 719  * Initialize Echo Server device with parameters in the config log \a cfg.
 720  *
 721  * This is the main starting point of Echo Server initialization. It fills all
 722  * parameters with their initial values and starts Echo Server.
 723  *
 724  * \param[in] env       execution environment
 725  * \param[in] m         Echo Server device
 726  * \param[in] ldt       LU device type of Echo Server
 727  * \param[in] cfg       configuration log
 728  *
 729  * \retval              0 if successful
 730  * \retval              negative value on error
 731  */
 732 static int echo_srv_init0(const struct lu_env *env,
 733                           struct echo_srv_device *esd,
 734                           struct lu_device_type *ldt, struct lustre_cfg *cfg)
 735 {
 736         const char *dev = lustre_cfg_string(cfg, 0);
 737         struct obd_device *obd;
 738         char ns_name[48];
 739         int rc;
 740
 741         ENTRY;
 742
 743         obd = class_name2obd(dev);
 744         if (!obd) {
 745                 CERROR("Cannot find obd with name %s\n", dev);
 746                 RETURN(-ENODEV);
 747         }
 748
 749         spin_lock_init(&obd->u.echo.eo_lock);
 750         obd->u.echo.eo_lastino = ECHO_INIT_OID;
 751
 752         esd->esd_dev.ld_ops = &echo_srv_lu_ops;
 753         esd->esd_dev.ld_obd = obd;
 754         /* set this lu_device to obd, because error handling need it */
 755         obd->obd_lu_dev = &esd->esd_dev;
 756
 757         /* No connection accepted until configurations will finish */
 758         spin_lock(&obd->obd_dev_lock);
 759         obd->obd_no_conn = 1;
 760         spin_unlock(&obd->obd_dev_lock);
 761
 762         /* non-replayable target */
 763         obd->obd_replayable = 0;
 764
 765         snprintf(ns_name, sizeof(ns_name), "echotgt-%s", obd->obd_uuid.uuid);
 766         obd->obd_namespace = ldlm_namespace_new(obd, ns_name,
 767                                                 LDLM_NAMESPACE_SERVER,
 768                                                 LDLM_NAMESPACE_MODEST,
 769                                                 LDLM_NS_TYPE_OST);
 770         if (IS_ERR(obd->obd_namespace)) {
 771                 rc = PTR_ERR(obd->obd_namespace);
 772                 CERROR("%s: unable to create server namespace: rc = %d\n",
 773                        obd->obd_name, rc);
 774                 obd->obd_namespace = NULL;
 775                 RETURN(rc);
 776         }
 777
 778         obd->obd_vars = lprocfs_echo_obd_vars;
 779         if (!lprocfs_obd_setup(obd, true) &&
 780             lprocfs_alloc_obd_stats(obd, LPROC_ECHO_LAST) == 0) {
 781                 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_READ_BYTES,
 782                                      LPROCFS_CNTR_AVGMINMAX,
 783                                      "read_bytes", "bytes");
 784                 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
 785                                      LPROCFS_CNTR_AVGMINMAX,
 786                                      "write_bytes", "bytes");
 787         }
 788
 789         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
 790                            "echo_ldlm_cb_client", &obd->obd_ldlm_client);
 791
 792         rc = tgt_init(env, &esd->esd_lut, obd, NULL, esd_common_slice,
 793                       OBD_FAIL_OST_ALL_REQUEST_NET,
 794                       OBD_FAIL_OST_ALL_REPLY_NET);
 795         if (rc)
 796                 GOTO(err_out, rc);
 797
 798         spin_lock(&obd->obd_dev_lock);
 799         obd->obd_no_conn = 0;
 800         spin_unlock(&obd->obd_dev_lock);
 801
 802         RETURN(0);
 803
 804 err_out:
 805         ldlm_namespace_free(obd->obd_namespace, NULL, obd->obd_force);
 806         obd->obd_namespace = NULL;
 807
 808         lprocfs_obd_cleanup(obd);
 809         lprocfs_free_obd_stats(obd);
 810         RETURN(rc);
 811 }
 812
 813 /**
 814  * Stop the Echo Server device.
 815  *
 816  * This function stops the Echo Server device and all its subsystems.
 817  * This is the end of Echo Server lifecycle.
 818  *
 819  * \param[in] env       execution environment
 820  * \param[in] esd               ESD device
 821  */
 822 static void echo_srv_fini(const struct lu_env *env,
 823                           struct echo_srv_device *esd)
 824 {
 825         struct obd_device *obd = echo_srv_obd(esd);
 826         struct lu_device *d = &esd->esd_dev;
 827         int leaked;
 828
 829         ENTRY;
 830
 831         class_disconnect_exports(obd);
 832         if (obd->obd_namespace)
 833                 ldlm_namespace_free_prior(obd->obd_namespace, NULL,
 834                                           obd->obd_force);
 835
 836         obd_exports_barrier(obd);
 837         obd_zombie_barrier();
 838
 839         tgt_fini(env, &esd->esd_lut);
 840
 841         if (obd->obd_namespace) {
 842                 ldlm_namespace_free_post(obd->obd_namespace);
 843                 obd->obd_namespace = NULL;
 844         }
 845
 846         lprocfs_obd_cleanup(obd);
 847         lprocfs_free_obd_stats(obd);
 848
 849         leaked = atomic_read(&obd->u.echo.eo_prep);
 850         if (leaked != 0)
 851                 CERROR("%d prep/commitrw pages leaked\n", leaked);
 852
 853         LASSERT(atomic_read(&d->ld_ref) == 0);
 854         EXIT;
 855 }
 856
 857 /**
 858  * Implementation of lu_device_type_operations::ldto_device_fini.
 859  *
 860  * Finalize device. Dual to echo_srv_device_init(). It is called from
 861  * obd_precleanup() and stops the current device.
 862  *
 863  * \param[in] env       execution environment
 864  * \param[in] d         LU device of ESD
 865  *
 866  * \retval              NULL
 867  */
 868 static struct lu_device *echo_srv_device_fini(const struct lu_env *env,
 869                                               struct lu_device *d)
 870 {
 871         ENTRY;
 872         echo_srv_fini(env, echo_srv_dev(d));
 873         RETURN(NULL);
 874 }
 875
 876 /**
 877  * Implementation of lu_device_type_operations::ldto_device_free.
 878  *
 879  * Free Echo Server device. Dual to echo_srv_device_alloc().
 880  *
 881  * \param[in] env       execution environment
 882  * \param[in] d         LU device of ESD
 883  *
 884  * \retval              NULL
 885  */
 886 static struct lu_device *echo_srv_device_free(const struct lu_env *env,
 887                                               struct lu_device *d)
 888 {
 889         struct echo_srv_device *esd = echo_srv_dev(d);
 890
 891         lu_device_fini(&esd->esd_dev);
 892         OBD_FREE_PTR(esd);
 893         RETURN(NULL);
 894 }
 895
 896 /**
 897  * Implementation of lu_device_type_operations::ldto_device_alloc.
 898  *
 899  * This function allocates the new Echo Server device. It is called from
 900  * obd_setup() if OBD device had lu_device_type defined.
 901  *
 902  * \param[in] env       execution environment
 903  * \param[in] t         lu_device_type of ESD device
 904  * \param[in] cfg       configuration log
 905  *
 906  * \retval              pointer to the lu_device of just allocated OFD
 907  * \retval              ERR_PTR of return value on error
 908  */
 909 static struct lu_device *echo_srv_device_alloc(const struct lu_env *env,
 910                                                struct lu_device_type *t,
 911                                                struct lustre_cfg *cfg)
 912 {
 913         struct echo_srv_device *esd;
 914         struct lu_device *l;
 915         int rc;
 916
 917         OBD_ALLOC_PTR(esd);
 918         if (!esd)
 919                 return ERR_PTR(-ENOMEM);
 920
 921         l = &esd->esd_dev;
 922         lu_device_init(l, t);
 923         rc = echo_srv_init0(env, esd, t, cfg);
 924         if (rc != 0) {
 925                 echo_srv_device_free(env, l);
 926                 l = ERR_PTR(rc);
 927         }
 928
 929         return l;
 930 }
 931
 932 static const struct lu_device_type_operations echo_srv_type_ops = {
 933         .ldto_device_alloc = echo_srv_device_alloc,
 934         .ldto_device_free = echo_srv_device_free,
 935         .ldto_device_fini = echo_srv_device_fini
 936 };
 937
 938 struct lu_device_type echo_srv_type = {
 939         .ldt_tags = LU_DEVICE_DT,
 940         .ldt_name = LUSTRE_ECHO_NAME,
 941         .ldt_ops = &echo_srv_type_ops,
 942         .ldt_ctx_tags = LCT_DT_THREAD,
 943 };
 944
 945 void echo_persistent_pages_fini(void)
 946 {
 947         int i;
 948
 949         for (i = 0; i < ECHO_PERSISTENT_PAGES; i++)
 950                 if (echo_persistent_pages[i]) {
 951                         __free_page(echo_persistent_pages[i]);
 952                         echo_persistent_pages[i] = NULL;
 953                 }
 954 }
 955
 956 int echo_persistent_pages_init(void)
 957 {
 958         struct page *pg;
 959         int          i;
 960
 961         for (i = 0; i < ECHO_PERSISTENT_PAGES; i++) {
 962                 gfp_t gfp_mask = (i < ECHO_PERSISTENT_PAGES / 2) ?
 963                         GFP_KERNEL : GFP_HIGHUSER;
 964
 965                 pg = alloc_page(gfp_mask);
 966                 if (!pg) {
 967                         echo_persistent_pages_fini();
 968                         return -ENOMEM;
 969                 }
 970
 971                 memset(kmap(pg), 0, PAGE_SIZE);
 972                 kunmap(pg);
 973                 /* set mapping so page is not considered encrypted */
 974                 pg->mapping = ECHO_MAPPING_UNENCRYPTED;
 975
 976                 echo_persistent_pages[i] = pg;
 977         }
 978
 979         return 0;
 980 }