lnet/selftest/brw_test.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) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, 2013, 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  * lnet/selftest/brw_test.c
  37  *
  38  * Author: Isaac Huang <isaac@clusterfs.com>
  39  */
  40
  41 #include "selftest.h"
  42
  43 static int brw_srv_workitems = SFW_TEST_WI_MAX;
  44 CFS_MODULE_PARM(brw_srv_workitems, "i", int, 0644, "# BRW server workitems");
  45
  46 static int brw_inject_errors;
  47 CFS_MODULE_PARM(brw_inject_errors, "i", int, 0644,
  48                 "# data errors to inject randomly, zero by default");
  49
  50 static void
  51 brw_client_fini (sfw_test_instance_t *tsi)
  52 {
  53         srpc_bulk_t     *bulk;
  54         sfw_test_unit_t *tsu;
  55
  56         LASSERT (tsi->tsi_is_client);
  57
  58         cfs_list_for_each_entry_typed (tsu, &tsi->tsi_units,
  59                                        sfw_test_unit_t, tsu_list) {
  60                 bulk = tsu->tsu_private;
  61                 if (bulk == NULL) continue;
  62
  63                 srpc_free_bulk(bulk);
  64                 tsu->tsu_private = NULL;
  65         }
  66 }
  67
  68 int
  69 brw_client_init (sfw_test_instance_t *tsi)
  70 {
  71         sfw_session_t    *sn = tsi->tsi_batch->bat_session;
  72         int               flags;
  73         int               npg;
  74         int               len;
  75         int               opc;
  76         srpc_bulk_t      *bulk;
  77         sfw_test_unit_t  *tsu;
  78
  79         LASSERT(sn != NULL);
  80         LASSERT(tsi->tsi_is_client);
  81
  82         if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
  83                 test_bulk_req_t  *breq = &tsi->tsi_u.bulk_v0;
  84
  85                 opc   = breq->blk_opc;
  86                 flags = breq->blk_flags;
  87                 npg   = breq->blk_npg;
  88                 /* NB: this is not going to work for variable page size,
  89                  * but we have to keep it for compatibility */
  90                 len   = npg * PAGE_CACHE_SIZE;
  91
  92         } else {
  93                 test_bulk_req_v1_t  *breq = &tsi->tsi_u.bulk_v1;
  94
  95                 /* I should never get this step if it's unknown feature
  96                  * because make_session will reject unknown feature */
  97                 LASSERT((sn->sn_features & ~LST_FEATS_MASK) == 0);
  98
  99                 opc   = breq->blk_opc;
 100                 flags = breq->blk_flags;
 101                 len   = breq->blk_len;
 102                 npg   = (len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 103         }
 104
 105         if (npg > LNET_MAX_IOV || npg <= 0)
 106                 return -EINVAL;
 107
 108         if (opc != LST_BRW_READ && opc != LST_BRW_WRITE)
 109                 return -EINVAL;
 110
 111         if (flags != LST_BRW_CHECK_NONE &&
 112             flags != LST_BRW_CHECK_FULL && flags != LST_BRW_CHECK_SIMPLE)
 113                 return -EINVAL;
 114
 115         cfs_list_for_each_entry_typed(tsu, &tsi->tsi_units,
 116                                       sfw_test_unit_t, tsu_list) {
 117                 bulk = srpc_alloc_bulk(lnet_cpt_of_nid(tsu->tsu_dest.nid),
 118                                        npg, len, opc == LST_BRW_READ);
 119                 if (bulk == NULL) {
 120                         brw_client_fini(tsi);
 121                         return -ENOMEM;
 122                 }
 123
 124                 tsu->tsu_private = bulk;
 125         }
 126
 127         return 0;
 128 }
 129
 130 #define BRW_POISON      0xbeefbeefbeefbeefULL
 131 #define BRW_MAGIC       0xeeb0eeb1eeb2eeb3ULL
 132 #define BRW_MSIZE       sizeof(__u64)
 133
 134 int brw_inject_one_error (void)
 135 {
 136         struct timeval tv;
 137
 138         if (brw_inject_errors <= 0) return 0;
 139
 140 #ifndef __KERNEL__
 141         gettimeofday(&tv, NULL);
 142 #else
 143         do_gettimeofday(&tv);
 144 #endif
 145
 146         if ((tv.tv_usec & 1) == 0) return 0;
 147
 148         return brw_inject_errors--;
 149 }
 150
 151 void
 152 brw_fill_page (struct page *pg, int pattern, __u64 magic)
 153 {
 154         char *addr = page_address(pg);
 155         int   i;
 156
 157         LASSERT (addr != NULL);
 158
 159         if (pattern == LST_BRW_CHECK_NONE) return;
 160
 161         if (magic == BRW_MAGIC)
 162                 magic += brw_inject_one_error();
 163
 164         if (pattern == LST_BRW_CHECK_SIMPLE) {
 165                 memcpy(addr, &magic, BRW_MSIZE);
 166                 addr += PAGE_CACHE_SIZE - BRW_MSIZE;
 167                 memcpy(addr, &magic, BRW_MSIZE);
 168                 return;
 169         }
 170
 171         if (pattern == LST_BRW_CHECK_FULL) {
 172                 for (i = 0; i < PAGE_CACHE_SIZE / BRW_MSIZE; i++)
 173                         memcpy(addr + i * BRW_MSIZE, &magic, BRW_MSIZE);
 174                 return;
 175         }
 176
 177         LBUG ();
 178         return;
 179 }
 180
 181 int
 182 brw_check_page (struct page *pg, int pattern, __u64 magic)
 183 {
 184         char  *addr = page_address(pg);
 185         __u64  data = 0; /* make compiler happy */
 186         int    i;
 187
 188         LASSERT (addr != NULL);
 189
 190         if (pattern == LST_BRW_CHECK_NONE)
 191                 return 0;
 192
 193         if (pattern == LST_BRW_CHECK_SIMPLE) {
 194                 data = *((__u64 *) addr);
 195                 if (data != magic) goto bad_data;
 196
 197                 addr += PAGE_CACHE_SIZE - BRW_MSIZE;
 198                 data = *((__u64 *) addr);
 199                 if (data != magic) goto bad_data;
 200
 201                 return 0;
 202         }
 203
 204         if (pattern == LST_BRW_CHECK_FULL) {
 205                 for (i = 0; i < PAGE_CACHE_SIZE / BRW_MSIZE; i++) {
 206                         data = *(((__u64 *) addr) + i);
 207                         if (data != magic) goto bad_data;
 208                 }
 209
 210                 return 0;
 211         }
 212
 213         LBUG ();
 214
 215 bad_data:
 216         CERROR ("Bad data in page %p: "LPX64", "LPX64" expected\n",
 217                 pg, data, magic);
 218         return 1;
 219 }
 220
 221 void
 222 brw_fill_bulk (srpc_bulk_t *bk, int pattern, __u64 magic)
 223 {
 224         int         i;
 225         struct page *pg;
 226
 227         for (i = 0; i < bk->bk_niov; i++) {
 228 #ifdef __KERNEL__
 229                 pg = bk->bk_iovs[i].kiov_page;
 230 #else
 231                 LASSERT (bk->bk_pages != NULL);
 232                 pg = bk->bk_pages[i];
 233 #endif
 234                 brw_fill_page(pg, pattern, magic);
 235         }
 236 }
 237
 238 int
 239 brw_check_bulk (srpc_bulk_t *bk, int pattern, __u64 magic)
 240 {
 241         int         i;
 242         struct page *pg;
 243
 244         for (i = 0; i < bk->bk_niov; i++) {
 245 #ifdef __KERNEL__
 246                 pg = bk->bk_iovs[i].kiov_page;
 247 #else
 248                 LASSERT (bk->bk_pages != NULL);
 249                 pg = bk->bk_pages[i];
 250 #endif
 251                 if (brw_check_page(pg, pattern, magic) != 0) {
 252                         CERROR ("Bulk page %p (%d/%d) is corrupted!\n",
 253                                 pg, i, bk->bk_niov);
 254                         return 1;
 255                 }
 256         }
 257
 258         return 0;
 259 }
 260
 261 static int
 262 brw_client_prep_rpc (sfw_test_unit_t *tsu,
 263                      lnet_process_id_t dest, srpc_client_rpc_t **rpcpp)
 264 {
 265         srpc_bulk_t         *bulk = tsu->tsu_private;
 266         sfw_test_instance_t *tsi = tsu->tsu_instance;
 267         sfw_session_t       *sn = tsi->tsi_batch->bat_session;
 268         srpc_client_rpc_t   *rpc;
 269         srpc_brw_reqst_t    *req;
 270         int                  flags;
 271         int                  npg;
 272         int                  len;
 273         int                  opc;
 274         int                  rc;
 275
 276         LASSERT(sn != NULL);
 277         LASSERT(bulk != NULL);
 278
 279         if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
 280                 test_bulk_req_t *breq = &tsi->tsi_u.bulk_v0;
 281
 282                 opc   = breq->blk_opc;
 283                 flags = breq->blk_flags;
 284                 npg   = breq->blk_npg;
 285                 len   = npg * PAGE_CACHE_SIZE;
 286
 287         } else {
 288                 test_bulk_req_v1_t  *breq = &tsi->tsi_u.bulk_v1;
 289
 290                 /* I should never get this step if it's unknown feature
 291                  * because make_session will reject unknown feature */
 292                 LASSERT((sn->sn_features & ~LST_FEATS_MASK) == 0);
 293
 294                 opc   = breq->blk_opc;
 295                 flags = breq->blk_flags;
 296                 len   = breq->blk_len;
 297                 npg   = (len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 298         }
 299
 300         rc = sfw_create_test_rpc(tsu, dest, sn->sn_features, npg, len, &rpc);
 301         if (rc != 0)
 302                 return rc;
 303
 304         memcpy(&rpc->crpc_bulk, bulk, offsetof(srpc_bulk_t, bk_iovs[npg]));
 305         if (opc == LST_BRW_WRITE)
 306                 brw_fill_bulk(&rpc->crpc_bulk, flags, BRW_MAGIC);
 307         else
 308                 brw_fill_bulk(&rpc->crpc_bulk, flags, BRW_POISON);
 309
 310         req = &rpc->crpc_reqstmsg.msg_body.brw_reqst;
 311         req->brw_flags = flags;
 312         req->brw_rw    = opc;
 313         req->brw_len   = len;
 314
 315         *rpcpp = rpc;
 316         return 0;
 317 }
 318
 319 static void
 320 brw_client_done_rpc (sfw_test_unit_t *tsu, srpc_client_rpc_t *rpc)
 321 {
 322         __u64                magic = BRW_MAGIC;
 323         sfw_test_instance_t *tsi = tsu->tsu_instance;
 324         sfw_session_t       *sn = tsi->tsi_batch->bat_session;
 325         srpc_msg_t          *msg = &rpc->crpc_replymsg;
 326         srpc_brw_reply_t    *reply = &msg->msg_body.brw_reply;
 327         srpc_brw_reqst_t    *reqst = &rpc->crpc_reqstmsg.msg_body.brw_reqst;
 328
 329         LASSERT (sn != NULL);
 330
 331         if (rpc->crpc_status != 0) {
 332                 CERROR ("BRW RPC to %s failed with %d\n",
 333                         libcfs_id2str(rpc->crpc_dest), rpc->crpc_status);
 334                 if (!tsi->tsi_stopping) /* rpc could have been aborted */
 335                         cfs_atomic_inc(&sn->sn_brw_errors);
 336                 goto out;
 337         }
 338
 339         if (msg->msg_magic != SRPC_MSG_MAGIC) {
 340                 __swab64s(&magic);
 341                 __swab32s(&reply->brw_status);
 342         }
 343
 344         CDEBUG (reply->brw_status ? D_WARNING : D_NET,
 345                 "BRW RPC to %s finished with brw_status: %d\n",
 346                 libcfs_id2str(rpc->crpc_dest), reply->brw_status);
 347
 348         if (reply->brw_status != 0) {
 349                 cfs_atomic_inc(&sn->sn_brw_errors);
 350                 rpc->crpc_status = -(int)reply->brw_status;
 351                 goto out;
 352         }
 353
 354         if (reqst->brw_rw == LST_BRW_WRITE) goto out;
 355
 356         if (brw_check_bulk(&rpc->crpc_bulk, reqst->brw_flags, magic) != 0) {
 357                 CERROR ("Bulk data from %s is corrupted!\n",
 358                         libcfs_id2str(rpc->crpc_dest));
 359                 cfs_atomic_inc(&sn->sn_brw_errors);
 360                 rpc->crpc_status = -EBADMSG;
 361         }
 362
 363 out:
 364 #ifndef __KERNEL__
 365         rpc->crpc_bulk.bk_pages = NULL;
 366 #endif
 367         return;
 368 }
 369
 370 void
 371 brw_server_rpc_done (srpc_server_rpc_t *rpc)
 372 {
 373         srpc_bulk_t *blk = rpc->srpc_bulk;
 374
 375         if (blk == NULL) return;
 376
 377         if (rpc->srpc_status != 0)
 378                 CERROR ("Bulk transfer %s %s has failed: %d\n",
 379                         blk->bk_sink ? "from" : "to",
 380                         libcfs_id2str(rpc->srpc_peer), rpc->srpc_status);
 381         else
 382                 CDEBUG (D_NET, "Transfered %d pages bulk data %s %s\n",
 383                         blk->bk_niov, blk->bk_sink ? "from" : "to",
 384                         libcfs_id2str(rpc->srpc_peer));
 385
 386         sfw_free_pages(rpc);
 387 }
 388
 389 int
 390 brw_bulk_ready (srpc_server_rpc_t *rpc, int status)
 391 {
 392         __u64             magic = BRW_MAGIC;
 393         srpc_brw_reply_t *reply = &rpc->srpc_replymsg.msg_body.brw_reply;
 394         srpc_brw_reqst_t *reqst;
 395         srpc_msg_t       *reqstmsg;
 396
 397         LASSERT (rpc->srpc_bulk != NULL);
 398         LASSERT (rpc->srpc_reqstbuf != NULL);
 399
 400         reqstmsg = &rpc->srpc_reqstbuf->buf_msg;
 401         reqst = &reqstmsg->msg_body.brw_reqst;
 402
 403         if (status != 0) {
 404                 CERROR ("BRW bulk %s failed for RPC from %s: %d\n",
 405                         reqst->brw_rw == LST_BRW_READ ? "READ" : "WRITE",
 406                         libcfs_id2str(rpc->srpc_peer), status);
 407                 return -EIO;
 408         }
 409
 410         if (reqst->brw_rw == LST_BRW_READ)
 411                 return 0;
 412
 413         if (reqstmsg->msg_magic != SRPC_MSG_MAGIC)
 414                 __swab64s(&magic);
 415
 416         if (brw_check_bulk(rpc->srpc_bulk, reqst->brw_flags, magic) != 0) {
 417                 CERROR ("Bulk data from %s is corrupted!\n",
 418                         libcfs_id2str(rpc->srpc_peer));
 419                 reply->brw_status = EBADMSG;
 420         }
 421
 422         return 0;
 423 }
 424
 425 int
 426 brw_server_handle(struct srpc_server_rpc *rpc)
 427 {
 428         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
 429         srpc_msg_t       *replymsg = &rpc->srpc_replymsg;
 430         srpc_msg_t       *reqstmsg = &rpc->srpc_reqstbuf->buf_msg;
 431         srpc_brw_reply_t *reply = &replymsg->msg_body.brw_reply;
 432         srpc_brw_reqst_t *reqst = &reqstmsg->msg_body.brw_reqst;
 433         int               npg;
 434         int               rc;
 435
 436         LASSERT (sv->sv_id == SRPC_SERVICE_BRW);
 437
 438         if (reqstmsg->msg_magic != SRPC_MSG_MAGIC) {
 439                 LASSERT (reqstmsg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 440
 441                 __swab32s(&reqst->brw_rw);
 442                 __swab32s(&reqst->brw_len);
 443                 __swab32s(&reqst->brw_flags);
 444                 __swab64s(&reqst->brw_rpyid);
 445                 __swab64s(&reqst->brw_bulkid);
 446         }
 447         LASSERT (reqstmsg->msg_type == (__u32)srpc_service2request(sv->sv_id));
 448
 449         reply->brw_status = 0;
 450         rpc->srpc_done = brw_server_rpc_done;
 451
 452         if ((reqst->brw_rw != LST_BRW_READ && reqst->brw_rw != LST_BRW_WRITE) ||
 453             (reqst->brw_flags != LST_BRW_CHECK_NONE &&
 454              reqst->brw_flags != LST_BRW_CHECK_FULL &&
 455              reqst->brw_flags != LST_BRW_CHECK_SIMPLE)) {
 456                 reply->brw_status = EINVAL;
 457                 return 0;
 458         }
 459
 460         if ((reqstmsg->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
 461                 replymsg->msg_ses_feats = LST_FEATS_MASK;
 462                 reply->brw_status = EPROTO;
 463                 return 0;
 464         }
 465
 466         if ((reqstmsg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
 467                 /* compat with old version */
 468                 if ((reqst->brw_len & ~CFS_PAGE_MASK) != 0) {
 469                         reply->brw_status = EINVAL;
 470                         return 0;
 471                 }
 472                 npg = reqst->brw_len >> PAGE_CACHE_SHIFT;
 473
 474         } else {
 475                 npg = (reqst->brw_len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 476         }
 477
 478         replymsg->msg_ses_feats = reqstmsg->msg_ses_feats;
 479
 480         if (reqst->brw_len == 0 || npg > LNET_MAX_IOV) {
 481                 reply->brw_status = EINVAL;
 482                 return 0;
 483         }
 484
 485         rc = sfw_alloc_pages(rpc, rpc->srpc_scd->scd_cpt, npg,
 486                              reqst->brw_len,
 487                              reqst->brw_rw == LST_BRW_WRITE);
 488         if (rc != 0)
 489                 return rc;
 490
 491         if (reqst->brw_rw == LST_BRW_READ)
 492                 brw_fill_bulk(rpc->srpc_bulk, reqst->brw_flags, BRW_MAGIC);
 493         else
 494                 brw_fill_bulk(rpc->srpc_bulk, reqst->brw_flags, BRW_POISON);
 495
 496         return 0;
 497 }
 498
 499 sfw_test_client_ops_t brw_test_client;
 500 void brw_init_test_client(void)
 501 {
 502         brw_test_client.tso_init       = brw_client_init;
 503         brw_test_client.tso_fini       = brw_client_fini;
 504         brw_test_client.tso_prep_rpc   = brw_client_prep_rpc;
 505         brw_test_client.tso_done_rpc   = brw_client_done_rpc;
 506 };
 507
 508 srpc_service_t brw_test_service;
 509 void brw_init_test_service(void)
 510 {
 511 #ifndef __KERNEL__
 512         char *s;
 513
 514         s = getenv("BRW_INJECT_ERRORS");
 515         brw_inject_errors = s != NULL ? atoi(s) : brw_inject_errors;
 516 #endif
 517
 518         brw_test_service.sv_id         = SRPC_SERVICE_BRW;
 519         brw_test_service.sv_name       = "brw_test";
 520         brw_test_service.sv_handler    = brw_server_handle;
 521         brw_test_service.sv_bulk_ready = brw_bulk_ready;
 522         brw_test_service.sv_wi_total   = brw_srv_workitems;
 523 }