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