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