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, 2015, 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 module_param(brw_srv_workitems, int, 0644);
  45 MODULE_PARM_DESC(brw_srv_workitems, "# BRW server workitems");
  46
  47 static int brw_inject_errors;
  48 module_param(brw_inject_errors, int, 0644);
  49 MODULE_PARM_DESC(brw_inject_errors, "# data errors to inject randomly, zero by default");
  50
  51 static void
  52 brw_client_fini (sfw_test_instance_t *tsi)
  53 {
  54         srpc_bulk_t     *bulk;
  55         sfw_test_unit_t *tsu;
  56
  57         LASSERT(tsi->tsi_is_client);
  58
  59         list_for_each_entry(tsu, &tsi->tsi_units, tsu_list) {
  60                 bulk = tsu->tsu_private;
  61                 if (bulk == NULL)
  62                         continue;
  63
  64                 srpc_free_bulk(bulk);
  65                 tsu->tsu_private = NULL;
  66         }
  67 }
  68
  69 static int
  70 brw_client_init (sfw_test_instance_t *tsi)
  71 {
  72         sfw_session_t    *sn = tsi->tsi_batch->bat_session;
  73         int               flags;
  74         int               npg;
  75         int               len;
  76         int               opc;
  77         srpc_bulk_t      *bulk;
  78         sfw_test_unit_t  *tsu;
  79
  80         LASSERT(sn != NULL);
  81         LASSERT(tsi->tsi_is_client);
  82
  83         if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
  84                 test_bulk_req_t  *breq = &tsi->tsi_u.bulk_v0;
  85
  86                 opc   = breq->blk_opc;
  87                 flags = breq->blk_flags;
  88                 npg   = breq->blk_npg;
  89                 /* NB: this is not going to work for variable page size,
  90                  * but we have to keep it for compatibility */
  91                 len   = npg * PAGE_CACHE_SIZE;
  92
  93         } else {
  94                 test_bulk_req_v1_t  *breq = &tsi->tsi_u.bulk_v1;
  95
  96                 /* I should never get this step if it's unknown feature
  97                  * because make_session will reject unknown feature */
  98                 LASSERT((sn->sn_features & ~LST_FEATS_MASK) == 0);
  99
 100                 opc   = breq->blk_opc;
 101                 flags = breq->blk_flags;
 102                 len   = breq->blk_len;
 103                 npg   = (len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 104         }
 105
 106         if (npg > LNET_MAX_IOV || npg <= 0)
 107                 return -EINVAL;
 108
 109         if (opc != LST_BRW_READ && opc != LST_BRW_WRITE)
 110                 return -EINVAL;
 111
 112         if (flags != LST_BRW_CHECK_NONE &&
 113             flags != LST_BRW_CHECK_FULL && flags != LST_BRW_CHECK_SIMPLE)
 114                 return -EINVAL;
 115
 116         list_for_each_entry(tsu, &tsi->tsi_units, 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 static int brw_inject_one_error(void)
 135 {
 136         struct timeval tv;
 137
 138         if (brw_inject_errors <= 0) return 0;
 139
 140         do_gettimeofday(&tv);
 141
 142         if ((tv.tv_usec & 1) == 0) return 0;
 143
 144         return brw_inject_errors--;
 145 }
 146
 147 static void
 148 brw_fill_page(struct page *pg, int pattern, __u64 magic)
 149 {
 150         char *addr = page_address(pg);
 151         int   i;
 152
 153         LASSERT (addr != NULL);
 154
 155         if (pattern == LST_BRW_CHECK_NONE) return;
 156
 157         if (magic == BRW_MAGIC)
 158                 magic += brw_inject_one_error();
 159
 160         if (pattern == LST_BRW_CHECK_SIMPLE) {
 161                 memcpy(addr, &magic, BRW_MSIZE);
 162                 addr += PAGE_CACHE_SIZE - BRW_MSIZE;
 163                 memcpy(addr, &magic, BRW_MSIZE);
 164                 return;
 165         }
 166
 167         if (pattern == LST_BRW_CHECK_FULL) {
 168                 for (i = 0; i < PAGE_CACHE_SIZE / BRW_MSIZE; i++)
 169                         memcpy(addr + i * BRW_MSIZE, &magic, BRW_MSIZE);
 170                 return;
 171         }
 172
 173         LBUG ();
 174         return;
 175 }
 176
 177 static int
 178 brw_check_page(struct page *pg, int pattern, __u64 magic)
 179 {
 180         char  *addr = page_address(pg);
 181         __u64  data = 0; /* make compiler happy */
 182         int    i;
 183
 184         LASSERT (addr != NULL);
 185
 186         if (pattern == LST_BRW_CHECK_NONE)
 187                 return 0;
 188
 189         if (pattern == LST_BRW_CHECK_SIMPLE) {
 190                 data = *((__u64 *) addr);
 191                 if (data != magic) goto bad_data;
 192
 193                 addr += PAGE_CACHE_SIZE - BRW_MSIZE;
 194                 data = *((__u64 *) addr);
 195                 if (data != magic) goto bad_data;
 196
 197                 return 0;
 198         }
 199
 200         if (pattern == LST_BRW_CHECK_FULL) {
 201                 for (i = 0; i < PAGE_CACHE_SIZE / BRW_MSIZE; i++) {
 202                         data = *(((__u64 *) addr) + i);
 203                         if (data != magic) goto bad_data;
 204                 }
 205
 206                 return 0;
 207         }
 208
 209         LBUG ();
 210
 211 bad_data:
 212         CERROR ("Bad data in page %p: %#llx, %#llx expected\n",
 213                 pg, data, magic);
 214         return 1;
 215 }
 216
 217 static void
 218 brw_fill_bulk(srpc_bulk_t *bk, int pattern, __u64 magic)
 219 {
 220         int         i;
 221         struct page *pg;
 222
 223         for (i = 0; i < bk->bk_niov; i++) {
 224                 pg = bk->bk_iovs[i].kiov_page;
 225                 brw_fill_page(pg, pattern, magic);
 226         }
 227 }
 228
 229 static int
 230 brw_check_bulk(srpc_bulk_t *bk, int pattern, __u64 magic)
 231 {
 232         int         i;
 233         struct page *pg;
 234
 235         for (i = 0; i < bk->bk_niov; i++) {
 236                 pg = bk->bk_iovs[i].kiov_page;
 237                 if (brw_check_page(pg, pattern, magic) != 0) {
 238                         CERROR ("Bulk page %p (%d/%d) is corrupted!\n",
 239                                 pg, i, bk->bk_niov);
 240                         return 1;
 241                 }
 242         }
 243
 244         return 0;
 245 }
 246
 247 static int
 248 brw_client_prep_rpc (sfw_test_unit_t *tsu,
 249                      lnet_process_id_t dest, srpc_client_rpc_t **rpcpp)
 250 {
 251         srpc_bulk_t         *bulk = tsu->tsu_private;
 252         sfw_test_instance_t *tsi = tsu->tsu_instance;
 253         sfw_session_t       *sn = tsi->tsi_batch->bat_session;
 254         srpc_client_rpc_t   *rpc;
 255         srpc_brw_reqst_t    *req;
 256         int                  flags;
 257         int                  npg;
 258         int                  len;
 259         int                  opc;
 260         int                  rc;
 261
 262         LASSERT(sn != NULL);
 263         LASSERT(bulk != NULL);
 264
 265         if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
 266                 test_bulk_req_t *breq = &tsi->tsi_u.bulk_v0;
 267
 268                 opc   = breq->blk_opc;
 269                 flags = breq->blk_flags;
 270                 npg   = breq->blk_npg;
 271                 len   = npg * PAGE_CACHE_SIZE;
 272
 273         } else {
 274                 test_bulk_req_v1_t  *breq = &tsi->tsi_u.bulk_v1;
 275
 276                 /* I should never get this step if it's unknown feature
 277                  * because make_session will reject unknown feature */
 278                 LASSERT((sn->sn_features & ~LST_FEATS_MASK) == 0);
 279
 280                 opc   = breq->blk_opc;
 281                 flags = breq->blk_flags;
 282                 len   = breq->blk_len;
 283                 npg   = (len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 284         }
 285
 286         rc = sfw_create_test_rpc(tsu, dest, sn->sn_features, npg, len, &rpc);
 287         if (rc != 0)
 288                 return rc;
 289
 290         memcpy(&rpc->crpc_bulk, bulk, offsetof(srpc_bulk_t, bk_iovs[npg]));
 291         if (opc == LST_BRW_WRITE)
 292                 brw_fill_bulk(&rpc->crpc_bulk, flags, BRW_MAGIC);
 293         else
 294                 brw_fill_bulk(&rpc->crpc_bulk, flags, BRW_POISON);
 295
 296         req = &rpc->crpc_reqstmsg.msg_body.brw_reqst;
 297         req->brw_flags = flags;
 298         req->brw_rw    = opc;
 299         req->brw_len   = len;
 300
 301         *rpcpp = rpc;
 302         return 0;
 303 }
 304
 305 static void
 306 brw_client_done_rpc(sfw_test_unit_t *tsu, srpc_client_rpc_t *rpc)
 307 {
 308         __u64                magic = BRW_MAGIC;
 309         sfw_test_instance_t *tsi = tsu->tsu_instance;
 310         sfw_session_t       *sn = tsi->tsi_batch->bat_session;
 311         srpc_msg_t          *msg = &rpc->crpc_replymsg;
 312         srpc_brw_reply_t    *reply = &msg->msg_body.brw_reply;
 313         srpc_brw_reqst_t    *reqst = &rpc->crpc_reqstmsg.msg_body.brw_reqst;
 314
 315         LASSERT(sn != NULL);
 316
 317         if (rpc->crpc_status != 0) {
 318                 CERROR("BRW RPC to %s failed with %d\n",
 319                        libcfs_id2str(rpc->crpc_dest), rpc->crpc_status);
 320                 if (!tsi->tsi_stopping) /* rpc could have been aborted */
 321                         atomic_inc(&sn->sn_brw_errors);
 322                 return;
 323         }
 324
 325         if (msg->msg_magic != SRPC_MSG_MAGIC) {
 326                 __swab64s(&magic);
 327                 __swab32s(&reply->brw_status);
 328         }
 329
 330         CDEBUG(reply->brw_status ? D_WARNING : D_NET,
 331                "BRW RPC to %s finished with brw_status: %d\n",
 332                libcfs_id2str(rpc->crpc_dest), reply->brw_status);
 333
 334         if (reply->brw_status != 0) {
 335                 atomic_inc(&sn->sn_brw_errors);
 336                 rpc->crpc_status = -(int)reply->brw_status;
 337                 return;
 338         }
 339
 340         if (reqst->brw_rw == LST_BRW_WRITE)
 341                 return;
 342
 343         if (brw_check_bulk(&rpc->crpc_bulk, reqst->brw_flags, magic) != 0) {
 344                 CERROR("Bulk data from %s is corrupted!\n",
 345                        libcfs_id2str(rpc->crpc_dest));
 346                 atomic_inc(&sn->sn_brw_errors);
 347                 rpc->crpc_status = -EBADMSG;
 348         }
 349
 350         return;
 351 }
 352
 353 static void
 354 brw_server_rpc_done(srpc_server_rpc_t *rpc)
 355 {
 356         srpc_bulk_t *blk = rpc->srpc_bulk;
 357
 358         if (blk == NULL)
 359                 return;
 360
 361         if (rpc->srpc_status != 0)
 362                 CERROR("Bulk transfer %s %s has failed: %d\n",
 363                        blk->bk_sink ? "from" : "to",
 364                        libcfs_id2str(rpc->srpc_peer), rpc->srpc_status);
 365         else
 366                 CDEBUG(D_NET, "Transferred %d pages bulk data %s %s\n",
 367                        blk->bk_niov, blk->bk_sink ? "from" : "to",
 368                        libcfs_id2str(rpc->srpc_peer));
 369
 370         sfw_free_pages(rpc);
 371 }
 372
 373 static int
 374 brw_bulk_ready(srpc_server_rpc_t *rpc, int status)
 375 {
 376         __u64             magic = BRW_MAGIC;
 377         srpc_brw_reply_t *reply = &rpc->srpc_replymsg.msg_body.brw_reply;
 378         srpc_brw_reqst_t *reqst;
 379         srpc_msg_t       *reqstmsg;
 380
 381         LASSERT (rpc->srpc_bulk != NULL);
 382         LASSERT (rpc->srpc_reqstbuf != NULL);
 383
 384         reqstmsg = &rpc->srpc_reqstbuf->buf_msg;
 385         reqst = &reqstmsg->msg_body.brw_reqst;
 386
 387         if (status != 0) {
 388                 CERROR ("BRW bulk %s failed for RPC from %s: %d\n",
 389                         reqst->brw_rw == LST_BRW_READ ? "READ" : "WRITE",
 390                         libcfs_id2str(rpc->srpc_peer), status);
 391                 return -EIO;
 392         }
 393
 394         if (reqst->brw_rw == LST_BRW_READ)
 395                 return 0;
 396
 397         if (reqstmsg->msg_magic != SRPC_MSG_MAGIC)
 398                 __swab64s(&magic);
 399
 400         if (brw_check_bulk(rpc->srpc_bulk, reqst->brw_flags, magic) != 0) {
 401                 CERROR ("Bulk data from %s is corrupted!\n",
 402                         libcfs_id2str(rpc->srpc_peer));
 403                 reply->brw_status = EBADMSG;
 404         }
 405
 406         return 0;
 407 }
 408
 409 static int
 410 brw_server_handle(struct srpc_server_rpc *rpc)
 411 {
 412         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
 413         srpc_msg_t       *replymsg = &rpc->srpc_replymsg;
 414         srpc_msg_t       *reqstmsg = &rpc->srpc_reqstbuf->buf_msg;
 415         srpc_brw_reply_t *reply = &replymsg->msg_body.brw_reply;
 416         srpc_brw_reqst_t *reqst = &reqstmsg->msg_body.brw_reqst;
 417         int               npg;
 418         int               rc;
 419
 420         LASSERT (sv->sv_id == SRPC_SERVICE_BRW);
 421
 422         if (reqstmsg->msg_magic != SRPC_MSG_MAGIC) {
 423                 LASSERT (reqstmsg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 424
 425                 __swab32s(&reqst->brw_rw);
 426                 __swab32s(&reqst->brw_len);
 427                 __swab32s(&reqst->brw_flags);
 428                 __swab64s(&reqst->brw_rpyid);
 429                 __swab64s(&reqst->brw_bulkid);
 430         }
 431         LASSERT (reqstmsg->msg_type == (__u32)srpc_service2request(sv->sv_id));
 432
 433         reply->brw_status = 0;
 434         rpc->srpc_done = brw_server_rpc_done;
 435
 436         if ((reqst->brw_rw != LST_BRW_READ && reqst->brw_rw != LST_BRW_WRITE) ||
 437             (reqst->brw_flags != LST_BRW_CHECK_NONE &&
 438              reqst->brw_flags != LST_BRW_CHECK_FULL &&
 439              reqst->brw_flags != LST_BRW_CHECK_SIMPLE)) {
 440                 reply->brw_status = EINVAL;
 441                 return 0;
 442         }
 443
 444         if ((reqstmsg->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
 445                 replymsg->msg_ses_feats = LST_FEATS_MASK;
 446                 reply->brw_status = EPROTO;
 447                 return 0;
 448         }
 449
 450         if ((reqstmsg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
 451                 /* compat with old version */
 452                 if ((reqst->brw_len & ~PAGE_MASK) != 0) {
 453                         reply->brw_status = EINVAL;
 454                         return 0;
 455                 }
 456                 npg = reqst->brw_len >> PAGE_CACHE_SHIFT;
 457
 458         } else {
 459                 npg = (reqst->brw_len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 460         }
 461
 462         replymsg->msg_ses_feats = reqstmsg->msg_ses_feats;
 463
 464         if (reqst->brw_len == 0 || npg > LNET_MAX_IOV) {
 465                 reply->brw_status = EINVAL;
 466                 return 0;
 467         }
 468
 469         rc = sfw_alloc_pages(rpc, rpc->srpc_scd->scd_cpt, npg,
 470                              reqst->brw_len,
 471                              reqst->brw_rw == LST_BRW_WRITE);
 472         if (rc != 0)
 473                 return rc;
 474
 475         if (reqst->brw_rw == LST_BRW_READ)
 476                 brw_fill_bulk(rpc->srpc_bulk, reqst->brw_flags, BRW_MAGIC);
 477         else
 478                 brw_fill_bulk(rpc->srpc_bulk, reqst->brw_flags, BRW_POISON);
 479
 480         return 0;
 481 }
 482
 483 sfw_test_client_ops_t brw_test_client;
 484 void brw_init_test_client(void)
 485 {
 486         brw_test_client.tso_init       = brw_client_init;
 487         brw_test_client.tso_fini       = brw_client_fini;
 488         brw_test_client.tso_prep_rpc   = brw_client_prep_rpc;
 489         brw_test_client.tso_done_rpc   = brw_client_done_rpc;
 490 };
 491
 492 srpc_service_t brw_test_service;
 493 void brw_init_test_service(void)
 494 {
 495
 496         brw_test_service.sv_id         = SRPC_SERVICE_BRW;
 497         brw_test_service.sv_name       = "brw_test";
 498         brw_test_service.sv_handler    = brw_server_handle;
 499         brw_test_service.sv_bulk_ready = brw_bulk_ready;
 500         brw_test_service.sv_wi_total   = brw_srv_workitems;
 501 }