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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lnet/selftest/brw_test.c
  32  *
  33  * Author: Isaac Huang <isaac@clusterfs.com>
  34  */
  35
  36 #include "selftest.h"
  37
  38 static int brw_srv_workitems = SFW_TEST_WI_MAX;
  39 module_param(brw_srv_workitems, int, 0644);
  40 MODULE_PARM_DESC(brw_srv_workitems, "# BRW server workitems");
  41
  42 static int brw_inject_errors;
  43 module_param(brw_inject_errors, int, 0644);
  44 MODULE_PARM_DESC(brw_inject_errors, "# data errors to inject randomly, zero by default");
  45
  46 #define BRW_POISON      0xbeefbeefbeefbeefULL
  47 #define BRW_MAGIC       0xeeb0eeb1eeb2eeb3ULL
  48 #define BRW_MSIZE       sizeof(__u64)
  49
  50 static void
  51 brw_client_fini(struct sfw_test_instance *tsi)
  52 {
  53         struct srpc_bulk *bulk;
  54         struct sfw_test_unit *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(struct sfw_test_instance *tsi)
  70 {
  71         struct sfw_session *sn = tsi->tsi_batch->bat_session;
  72         int               flags;
  73         int               off;
  74         int               npg;
  75         int               len;
  76         int               opc;
  77         struct srpc_bulk *bulk;
  78         struct sfw_test_unit *tsu;
  79
  80         LASSERT(sn != NULL);
  81         LASSERT(tsi->tsi_is_client);
  82
  83         if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
  84                 struct test_bulk_req *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_SIZE;
  92                 off   = 0;
  93
  94         } else {
  95                 struct test_bulk_req_v1 *breq = &tsi->tsi_u.bulk_v1;
  96
  97                 /* I should never get this step if it's unknown feature
  98                  * because make_session will reject unknown feature */
  99                 LASSERT((sn->sn_features & ~LST_FEATS_MASK) == 0);
 100
 101                 opc   = breq->blk_opc;
 102                 flags = breq->blk_flags;
 103                 len   = breq->blk_len;
 104                 off   = breq->blk_offset & ~PAGE_MASK;
 105                 npg   = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 106         }
 107
 108         if (off % BRW_MSIZE != 0)
 109                 return -EINVAL;
 110
 111         if (npg > LNET_MAX_IOV || npg <= 0)
 112                 return -EINVAL;
 113
 114         if (opc != LST_BRW_READ && opc != LST_BRW_WRITE)
 115                 return -EINVAL;
 116
 117         if (flags != LST_BRW_CHECK_NONE &&
 118             flags != LST_BRW_CHECK_FULL && flags != LST_BRW_CHECK_SIMPLE)
 119                 return -EINVAL;
 120
 121         list_for_each_entry(tsu, &tsi->tsi_units, tsu_list) {
 122                 bulk = srpc_alloc_bulk(lnet_cpt_of_nid(tsu->tsu_dest.nid, NULL),
 123                                        npg);
 124                 if (bulk == NULL) {
 125                         brw_client_fini(tsi);
 126                         return -ENOMEM;
 127                 }
 128                 srpc_init_bulk(bulk, off, npg, len, opc == LST_BRW_READ);
 129
 130                 tsu->tsu_private = bulk;
 131         }
 132
 133         return 0;
 134 }
 135
 136 #define BRW_POISON      0xbeefbeefbeefbeefULL
 137 #define BRW_MAGIC       0xeeb0eeb1eeb2eeb3ULL
 138 #define BRW_MSIZE       sizeof(__u64)
 139
 140 static int brw_inject_one_error(void)
 141 {
 142         struct timespec64 ts;
 143
 144         if (brw_inject_errors <= 0) return 0;
 145
 146         ktime_get_ts64(&ts);
 147
 148         if (((ts.tv_nsec / NSEC_PER_USEC) & 1) == 0)
 149                 return 0;
 150
 151         return brw_inject_errors--;
 152 }
 153
 154 static void
 155 brw_fill_page(struct page *pg, int off, int len, int pattern, __u64 magic)
 156 {
 157         char *addr = page_address(pg) + off;
 158         int   i;
 159
 160         LASSERT(addr != NULL);
 161         LASSERT(off % BRW_MSIZE == 0 && len % BRW_MSIZE == 0);
 162
 163         if (pattern == LST_BRW_CHECK_NONE)
 164                 return;
 165
 166         if (magic == BRW_MAGIC)
 167                 magic += brw_inject_one_error();
 168
 169         if (pattern == LST_BRW_CHECK_SIMPLE) {
 170                 memcpy(addr, &magic, BRW_MSIZE);
 171                 if (len > BRW_MSIZE) {
 172                         addr += len - BRW_MSIZE;
 173                         memcpy(addr, &magic, BRW_MSIZE);
 174                 }
 175                 return;
 176         }
 177
 178         if (pattern == LST_BRW_CHECK_FULL) {
 179                 for (i = 0; i < len; i += BRW_MSIZE)
 180                         memcpy(addr + i, &magic, BRW_MSIZE);
 181                 return;
 182         }
 183         LBUG();
 184 }
 185
 186 static int
 187 brw_check_page(struct page *pg, int off, int len, int pattern, __u64 magic)
 188 {
 189         char  *addr = page_address(pg) + off;
 190         __u64  data = 0; /* make compiler happy */
 191         int    i;
 192
 193         LASSERT(addr != NULL);
 194         LASSERT(off % BRW_MSIZE == 0 && len % BRW_MSIZE == 0);
 195
 196         if (pattern == LST_BRW_CHECK_NONE)
 197                 return 0;
 198
 199         if (pattern == LST_BRW_CHECK_SIMPLE) {
 200                 data = *((__u64 *) addr);
 201                 if (data != magic)
 202                         goto bad_data;
 203
 204                 if (len > BRW_MSIZE) {
 205                         addr += len - BRW_MSIZE;
 206                         data = *((__u64 *) addr);
 207                         if (data != magic)
 208                                 goto bad_data;
 209                 }
 210                 return 0;
 211         }
 212
 213         if (pattern == LST_BRW_CHECK_FULL) {
 214                 for (i = 0; i < len; i += BRW_MSIZE) {
 215                         data = *(__u64 *)(addr + i);
 216                         if (data != magic)
 217                                 goto bad_data;
 218                 }
 219                 return 0;
 220         }
 221
 222         LBUG();
 223
 224 bad_data:
 225         CERROR ("Bad data in page %p: %#llx, %#llx expected\n",
 226                 pg, data, magic);
 227         return 1;
 228 }
 229
 230 static void
 231 brw_fill_bulk(struct srpc_bulk *bk, int pattern, __u64 magic)
 232 {
 233         int          i;
 234         struct page *pg;
 235
 236         for (i = 0; i < bk->bk_niov; i++) {
 237                 int     off;
 238                 int     len;
 239
 240                 pg = bk->bk_iovs[i].bv_page;
 241                 off = bk->bk_iovs[i].bv_offset;
 242                 len = bk->bk_iovs[i].bv_len;
 243                 brw_fill_page(pg, off, len, pattern, magic);
 244         }
 245 }
 246
 247 static int
 248 brw_check_bulk(struct srpc_bulk *bk, int pattern, __u64 magic)
 249 {
 250         int          i;
 251         struct page *pg;
 252
 253         for (i = 0; i < bk->bk_niov; i++) {
 254                 int     off;
 255                 int     len;
 256
 257                 pg = bk->bk_iovs[i].bv_page;
 258                 off = bk->bk_iovs[i].bv_offset;
 259                 len = bk->bk_iovs[i].bv_len;
 260                 if (brw_check_page(pg, off, len, pattern, magic) != 0) {
 261                         CERROR("Bulk page %p (%d/%d) is corrupted!\n",
 262                                pg, i, bk->bk_niov);
 263                         return 1;
 264                 }
 265         }
 266
 267         return 0;
 268 }
 269
 270 static int
 271 brw_client_prep_rpc(struct sfw_test_unit *tsu, struct lnet_process_id dest,
 272                     struct srpc_client_rpc **rpcpp)
 273 {
 274         struct srpc_bulk *bulk = tsu->tsu_private;
 275         struct sfw_test_instance *tsi = tsu->tsu_instance;
 276         struct sfw_session *sn = tsi->tsi_batch->bat_session;
 277         struct srpc_client_rpc *rpc;
 278         struct srpc_brw_reqst *req;
 279         int flags;
 280         int npg;
 281         int len;
 282         int opc;
 283         int rc;
 284
 285         LASSERT(sn != NULL);
 286         LASSERT(bulk != NULL);
 287
 288         if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
 289                 struct test_bulk_req *breq = &tsi->tsi_u.bulk_v0;
 290
 291                 opc   = breq->blk_opc;
 292                 flags = breq->blk_flags;
 293                 npg   = breq->blk_npg;
 294                 len   = npg * PAGE_SIZE;
 295
 296         } else {
 297                 struct test_bulk_req_v1 *breq = &tsi->tsi_u.bulk_v1;
 298                 int off;
 299
 300                 /* I should never get this step if it's unknown feature
 301                  * because make_session will reject unknown feature */
 302                 LASSERT((sn->sn_features & ~LST_FEATS_MASK) == 0);
 303
 304                 opc   = breq->blk_opc;
 305                 flags = breq->blk_flags;
 306                 len   = breq->blk_len;
 307                 off   = breq->blk_offset;
 308                 npg   = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 309         }
 310
 311         rc = sfw_create_test_rpc(tsu, dest, sn->sn_features, npg, len, &rpc);
 312         if (rc != 0)
 313                 return rc;
 314
 315         memcpy(&rpc->crpc_bulk, bulk, offsetof(struct srpc_bulk, bk_iovs[npg]));
 316         if (opc == LST_BRW_WRITE)
 317                 brw_fill_bulk(&rpc->crpc_bulk, flags, BRW_MAGIC);
 318         else
 319                 brw_fill_bulk(&rpc->crpc_bulk, flags, BRW_POISON);
 320
 321         req = &rpc->crpc_reqstmsg.msg_body.brw_reqst;
 322         req->brw_flags = flags;
 323         req->brw_rw    = opc;
 324         req->brw_len   = len;
 325
 326         *rpcpp = rpc;
 327         return 0;
 328 }
 329
 330 static void
 331 brw_client_done_rpc(struct sfw_test_unit *tsu, struct srpc_client_rpc *rpc)
 332 {
 333         __u64 magic = BRW_MAGIC;
 334         struct sfw_test_instance *tsi = tsu->tsu_instance;
 335         struct sfw_session *sn = tsi->tsi_batch->bat_session;
 336         struct srpc_msg *msg = &rpc->crpc_replymsg;
 337         struct srpc_brw_reply *reply = &msg->msg_body.brw_reply;
 338         struct srpc_brw_reqst *reqst = &rpc->crpc_reqstmsg.msg_body.brw_reqst;
 339
 340         LASSERT(sn != NULL);
 341
 342         if (rpc->crpc_status != 0) {
 343                 CERROR("BRW RPC to %s failed with %d\n",
 344                        libcfs_id2str(rpc->crpc_dest), rpc->crpc_status);
 345                 if (!tsi->tsi_stopping) /* rpc could have been aborted */
 346                         atomic_inc(&sn->sn_brw_errors);
 347                 return;
 348         }
 349
 350         if (msg->msg_magic != SRPC_MSG_MAGIC) {
 351                 __swab64s(&magic);
 352                 __swab32s(&reply->brw_status);
 353         }
 354
 355         CDEBUG(reply->brw_status ? D_WARNING : D_NET,
 356                "BRW RPC to %s finished with brw_status: %d\n",
 357                libcfs_id2str(rpc->crpc_dest), reply->brw_status);
 358
 359         if (reply->brw_status != 0) {
 360                 atomic_inc(&sn->sn_brw_errors);
 361                 rpc->crpc_status = -(int)reply->brw_status;
 362                 return;
 363         }
 364
 365         if (reqst->brw_rw == LST_BRW_WRITE)
 366                 return;
 367
 368         if (brw_check_bulk(&rpc->crpc_bulk, reqst->brw_flags, magic) != 0) {
 369                 CERROR("Bulk data from %s is corrupted!\n",
 370                        libcfs_id2str(rpc->crpc_dest));
 371                 atomic_inc(&sn->sn_brw_errors);
 372                 rpc->crpc_status = -EBADMSG;
 373         }
 374 }
 375
 376 static void
 377 brw_server_rpc_done(struct srpc_server_rpc *rpc)
 378 {
 379         struct srpc_bulk *blk = rpc->srpc_bulk;
 380
 381         if (blk == NULL)
 382                 return;
 383
 384         if (rpc->srpc_status != 0)
 385                 CERROR("Bulk transfer %s %s has failed: %d\n",
 386                        blk->bk_sink ? "from" : "to",
 387                        libcfs_id2str(rpc->srpc_peer), rpc->srpc_status);
 388         else
 389                 CDEBUG(D_NET, "Transferred %d pages bulk data %s %s\n",
 390                        blk->bk_niov, blk->bk_sink ? "from" : "to",
 391                        libcfs_id2str(rpc->srpc_peer));
 392 }
 393
 394 static int
 395 brw_bulk_ready(struct srpc_server_rpc *rpc, int status)
 396 {
 397         __u64 magic = BRW_MAGIC;
 398         struct srpc_brw_reply *reply = &rpc->srpc_replymsg.msg_body.brw_reply;
 399         struct srpc_brw_reqst *reqst;
 400         struct srpc_msg *reqstmsg;
 401
 402         LASSERT (rpc->srpc_bulk != NULL);
 403         LASSERT (rpc->srpc_reqstbuf != NULL);
 404
 405         reqstmsg = &rpc->srpc_reqstbuf->buf_msg;
 406         reqst = &reqstmsg->msg_body.brw_reqst;
 407
 408         if (status != 0) {
 409                 CERROR ("BRW bulk %s failed for RPC from %s: %d\n",
 410                         reqst->brw_rw == LST_BRW_READ ? "READ" : "WRITE",
 411                         libcfs_id2str(rpc->srpc_peer), status);
 412                 return -EIO;
 413         }
 414
 415         if (reqst->brw_rw == LST_BRW_READ)
 416                 return 0;
 417
 418         if (reqstmsg->msg_magic != SRPC_MSG_MAGIC)
 419                 __swab64s(&magic);
 420
 421         if (brw_check_bulk(rpc->srpc_bulk, reqst->brw_flags, magic) != 0) {
 422                 CERROR ("Bulk data from %s is corrupted!\n",
 423                         libcfs_id2str(rpc->srpc_peer));
 424                 reply->brw_status = EBADMSG;
 425         }
 426
 427         return 0;
 428 }
 429
 430 static int
 431 brw_server_handle(struct srpc_server_rpc *rpc)
 432 {
 433         struct srpc_service *sv = rpc->srpc_scd->scd_svc;
 434         struct srpc_msg *replymsg = &rpc->srpc_replymsg;
 435         struct srpc_msg *reqstmsg = &rpc->srpc_reqstbuf->buf_msg;
 436         struct srpc_brw_reply *reply = &replymsg->msg_body.brw_reply;
 437         struct srpc_brw_reqst *reqst = &reqstmsg->msg_body.brw_reqst;
 438         int npg;
 439
 440         LASSERT (sv->sv_id == SRPC_SERVICE_BRW);
 441
 442         if (reqstmsg->msg_magic != SRPC_MSG_MAGIC) {
 443                 LASSERT (reqstmsg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 444
 445                 __swab32s(&reqst->brw_rw);
 446                 __swab32s(&reqst->brw_len);
 447                 __swab32s(&reqst->brw_flags);
 448                 __swab64s(&reqst->brw_rpyid);
 449                 __swab64s(&reqst->brw_bulkid);
 450         }
 451         LASSERT (reqstmsg->msg_type == (__u32)srpc_service2request(sv->sv_id));
 452
 453         reply->brw_status = 0;
 454         rpc->srpc_done = brw_server_rpc_done;
 455
 456         if ((reqst->brw_rw != LST_BRW_READ && reqst->brw_rw != LST_BRW_WRITE) ||
 457             (reqst->brw_flags != LST_BRW_CHECK_NONE &&
 458              reqst->brw_flags != LST_BRW_CHECK_FULL &&
 459              reqst->brw_flags != LST_BRW_CHECK_SIMPLE)) {
 460                 reply->brw_status = EINVAL;
 461                 return 0;
 462         }
 463
 464         if ((reqstmsg->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
 465                 replymsg->msg_ses_feats = LST_FEATS_MASK;
 466                 reply->brw_status = EPROTO;
 467                 return 0;
 468         }
 469
 470         if ((reqstmsg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
 471                 /* compat with old version */
 472                 if ((reqst->brw_len & ~PAGE_MASK) != 0) {
 473                         reply->brw_status = EINVAL;
 474                         return 0;
 475                 }
 476                 npg = reqst->brw_len >> PAGE_SHIFT;
 477
 478         } else {
 479                 npg = (reqst->brw_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 480         }
 481
 482         replymsg->msg_ses_feats = reqstmsg->msg_ses_feats;
 483
 484         if (reqst->brw_len == 0 || npg > LNET_MAX_IOV) {
 485                 reply->brw_status = EINVAL;
 486                 return 0;
 487         }
 488
 489         srpc_init_bulk(rpc->srpc_bulk, 0, npg, reqst->brw_len,
 490                        reqst->brw_rw == LST_BRW_WRITE);
 491
 492         if (reqst->brw_rw == LST_BRW_READ)
 493                 brw_fill_bulk(rpc->srpc_bulk, reqst->brw_flags, BRW_MAGIC);
 494         else
 495                 brw_fill_bulk(rpc->srpc_bulk, reqst->brw_flags, BRW_POISON);
 496
 497         return 0;
 498 }
 499
 500 static int
 501 brw_srpc_init(struct srpc_server_rpc *rpc, int cpt)
 502 {
 503         /* just alloc a maximal size - actual values will be adjusted later */
 504         rpc->srpc_bulk = srpc_alloc_bulk(cpt, LNET_MAX_IOV);
 505         if (rpc->srpc_bulk == NULL)
 506                 return -ENOMEM;
 507
 508         srpc_init_bulk(rpc->srpc_bulk, 0, LNET_MAX_IOV, 0, 0);
 509
 510         return 0;
 511 }
 512
 513 static void
 514 brw_srpc_fini(struct srpc_server_rpc *rpc)
 515 {
 516         /* server RPC have just MAX_IOV size */
 517         srpc_init_bulk(rpc->srpc_bulk, 0, LNET_MAX_IOV, 0, 0);
 518
 519         srpc_free_bulk(rpc->srpc_bulk);
 520         rpc->srpc_bulk = NULL;
 521 }
 522
 523 struct sfw_test_client_ops brw_test_client = {
 524         .tso_init       = brw_client_init,
 525         .tso_fini       = brw_client_fini,
 526         .tso_prep_rpc   = brw_client_prep_rpc,
 527         .tso_done_rpc   = brw_client_done_rpc,
 528 };
 529
 530 struct srpc_service brw_test_service = {
 531         .sv_id         = SRPC_SERVICE_BRW,
 532         .sv_name       = "brw_test",
 533         .sv_handler    = brw_server_handle,
 534         .sv_bulk_ready = brw_bulk_ready,
 535
 536         .sv_srpc_init  = brw_srpc_init,
 537         .sv_srpc_fini  = brw_srpc_fini,
 538 };
 539
 540 void brw_init_test_service(void)
 541 {
 542         unsigned long cache_size = cfs_totalram_pages() >> 1;
 543
 544         /* brw prealloc cache should don't eat more than half memory */
 545         cache_size /= LNET_MAX_IOV;
 546
 547         brw_test_service.sv_wi_total   = brw_srv_workitems;
 548
 549         if (brw_test_service.sv_wi_total > cache_size)
 550                 brw_test_service.sv_wi_total = cache_size;
 551 }