lnet/selftest/brw_test.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 /*
   4  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
   5  * Use is subject to license terms.
   6  *
   7  * Copyright (c) 2012, 2017, Intel Corporation.
   8  */
   9
  10 /*
  11  * This file is part of Lustre, http://www.lustre.org/
  12  *
  13  * Author: Isaac Huang <isaac@clusterfs.com>
  14  */
  15
  16 #include "selftest.h"
  17
  18 static int brw_srv_workitems = SFW_TEST_WI_MAX;
  19 module_param(brw_srv_workitems, int, 0644);
  20 MODULE_PARM_DESC(brw_srv_workitems, "# BRW server workitems");
  21
  22 static int brw_inject_errors;
  23 module_param(brw_inject_errors, int, 0644);
  24 MODULE_PARM_DESC(brw_inject_errors, "# data errors to inject randomly, zero by default");
  25
  26 #define BRW_POISON      0xbeefbeefbeefbeefULL
  27 #define BRW_MAGIC       0xeeb0eeb1eeb2eeb3ULL
  28 #define BRW_MSIZE       sizeof(__u64)
  29
  30 static void
  31 brw_client_fini(struct sfw_test_instance *tsi)
  32 {
  33         struct srpc_bulk *bulk;
  34         struct sfw_test_unit *tsu;
  35
  36         LASSERT(tsi->tsi_is_client);
  37
  38         list_for_each_entry(tsu, &tsi->tsi_units, tsu_list) {
  39                 bulk = tsu->tsu_private;
  40                 if (bulk == NULL)
  41                         continue;
  42
  43                 srpc_free_bulk(bulk);
  44                 tsu->tsu_private = NULL;
  45         }
  46 }
  47
  48 static int
  49 brw_client_init(struct sfw_test_instance *tsi)
  50 {
  51         struct sfw_session *sn = tsi->tsi_batch->bat_session;
  52         int               flags;
  53         int               off;
  54         unsigned int      len;
  55         int               opc;
  56         struct srpc_bulk *bulk;
  57         struct sfw_test_unit *tsu;
  58
  59         LASSERT(sn != NULL);
  60         LASSERT(tsi->tsi_is_client);
  61
  62         if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
  63                 struct test_bulk_req *breq = &tsi->tsi_u.bulk_v0;
  64
  65                 opc   = breq->blk_opc;
  66                 flags = breq->blk_flags;
  67                 /* NB: this is not going to work for variable page size,
  68                  * but we have to keep it for compatibility */
  69                 len   = breq->blk_npg * PAGE_SIZE;
  70                 off   = 0;
  71
  72         } else {
  73                 struct test_bulk_req_v1 *breq = &tsi->tsi_u.bulk_v1;
  74
  75                 /* I should never get this step if it's unknown feature
  76                  * because make_session will reject unknown feature */
  77                 LASSERT((sn->sn_features & ~LST_FEATS_MASK) == 0);
  78
  79                 opc   = breq->blk_opc;
  80                 flags = breq->blk_flags;
  81                 len   = breq->blk_len;
  82                 off   = breq->blk_offset & ~PAGE_MASK;
  83         }
  84
  85         if (off % BRW_MSIZE != 0)
  86                 return -EINVAL;
  87
  88         if (len > LNET_MTU)
  89                 return -EINVAL;
  90
  91         if (opc != LST_BRW_READ && opc != LST_BRW_WRITE)
  92                 return -EINVAL;
  93
  94         if (flags != LST_BRW_CHECK_NONE &&
  95             flags != LST_BRW_CHECK_FULL && flags != LST_BRW_CHECK_SIMPLE)
  96                 return -EINVAL;
  97
  98         list_for_each_entry(tsu, &tsi->tsi_units, tsu_list) {
  99                 bulk = srpc_alloc_bulk(lnet_cpt_of_nid(tsu->tsu_dest.nid, NULL),
 100                                        len);
 101                 if (bulk == NULL) {
 102                         brw_client_fini(tsi);
 103                         return -ENOMEM;
 104                 }
 105                 srpc_init_bulk(bulk, off, len, opc == LST_BRW_READ);
 106
 107                 tsu->tsu_private = bulk;
 108         }
 109
 110         return 0;
 111 }
 112
 113 #define BRW_POISON      0xbeefbeefbeefbeefULL
 114 #define BRW_MAGIC       0xeeb0eeb1eeb2eeb3ULL
 115 #define BRW_MSIZE       sizeof(__u64)
 116
 117 static int brw_inject_one_error(void)
 118 {
 119         struct timespec64 ts;
 120
 121         if (brw_inject_errors <= 0) return 0;
 122
 123         ktime_get_ts64(&ts);
 124
 125         if (((ts.tv_nsec / NSEC_PER_USEC) & 1) == 0)
 126                 return 0;
 127
 128         return brw_inject_errors--;
 129 }
 130
 131 static void
 132 brw_fill_page(struct page *pg, int off, int len, int pattern, __u64 magic)
 133 {
 134         char *addr = page_address(pg) + off;
 135         int   i;
 136
 137         LASSERT(addr != NULL);
 138         LASSERT(off % BRW_MSIZE == 0 && len % BRW_MSIZE == 0);
 139
 140         if (pattern == LST_BRW_CHECK_NONE)
 141                 return;
 142
 143         if (magic == BRW_MAGIC)
 144                 magic += brw_inject_one_error();
 145
 146         if (pattern == LST_BRW_CHECK_SIMPLE) {
 147                 memcpy(addr, &magic, BRW_MSIZE);
 148                 if (len > BRW_MSIZE) {
 149                         addr += len - BRW_MSIZE;
 150                         memcpy(addr, &magic, BRW_MSIZE);
 151                 }
 152                 return;
 153         }
 154
 155         if (pattern == LST_BRW_CHECK_FULL) {
 156                 for (i = 0; i < len; i += BRW_MSIZE)
 157                         memcpy(addr + i, &magic, BRW_MSIZE);
 158                 return;
 159         }
 160         LBUG();
 161 }
 162
 163 static int
 164 brw_check_page(struct page *pg, int off, int len, int pattern, __u64 magic)
 165 {
 166         char  *addr = page_address(pg) + off;
 167         __u64  data = 0; /* make compiler happy */
 168         int    i;
 169
 170         LASSERT(addr != NULL);
 171         LASSERT(off % BRW_MSIZE == 0 && len % BRW_MSIZE == 0);
 172
 173         if (pattern == LST_BRW_CHECK_NONE)
 174                 return 0;
 175
 176         if (pattern == LST_BRW_CHECK_SIMPLE) {
 177                 data = *((__u64 *) addr);
 178                 if (data != magic)
 179                         goto bad_data;
 180
 181                 if (len > BRW_MSIZE) {
 182                         addr += len - BRW_MSIZE;
 183                         data = *((__u64 *) addr);
 184                         if (data != magic)
 185                                 goto bad_data;
 186                 }
 187                 return 0;
 188         }
 189
 190         if (pattern == LST_BRW_CHECK_FULL) {
 191                 for (i = 0; i < len; i += BRW_MSIZE) {
 192                         data = *(__u64 *)(addr + i);
 193                         if (data != magic)
 194                                 goto bad_data;
 195                 }
 196                 return 0;
 197         }
 198
 199         LBUG();
 200
 201 bad_data:
 202         CERROR ("Bad data in page %p: %#llx, %#llx expected\n",
 203                 pg, data, magic);
 204         return 1;
 205 }
 206
 207 static void
 208 brw_fill_bulk(struct srpc_bulk *bk, int pattern, __u64 magic)
 209 {
 210         int          i;
 211         struct page *pg;
 212
 213         for (i = 0; i < bk->bk_niov; i++) {
 214                 int     off;
 215                 int     len;
 216
 217                 pg = bk->bk_iovs[i].bv_page;
 218                 off = bk->bk_iovs[i].bv_offset;
 219                 len = bk->bk_iovs[i].bv_len;
 220                 brw_fill_page(pg, off, len, pattern, magic);
 221         }
 222 }
 223
 224 static int
 225 brw_check_bulk(struct srpc_bulk *bk, int pattern, __u64 magic)
 226 {
 227         int          i;
 228         struct page *pg;
 229
 230         for (i = 0; i < bk->bk_niov; i++) {
 231                 int     off;
 232                 int     len;
 233
 234                 pg = bk->bk_iovs[i].bv_page;
 235                 off = bk->bk_iovs[i].bv_offset;
 236                 len = bk->bk_iovs[i].bv_len;
 237                 if (brw_check_page(pg, off, len, 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(struct sfw_test_unit *tsu, struct lnet_process_id dest,
 249                     struct srpc_client_rpc **rpcpp)
 250 {
 251         struct srpc_bulk *bulk = tsu->tsu_private;
 252         struct sfw_test_instance *tsi = tsu->tsu_instance;
 253         struct sfw_session *sn = tsi->tsi_batch->bat_session;
 254         struct srpc_client_rpc *rpc;
 255         struct srpc_brw_reqst *req;
 256         int flags;
 257         int npg;
 258         int len;
 259         int off;
 260         int opc;
 261         int rc;
 262
 263         LASSERT(sn != NULL);
 264         LASSERT(bulk != NULL);
 265
 266         if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
 267                 struct test_bulk_req *breq = &tsi->tsi_u.bulk_v0;
 268
 269                 opc   = breq->blk_opc;
 270                 flags = breq->blk_flags;
 271                 len   = breq->blk_npg * PAGE_SIZE;
 272                 off   = 0;
 273
 274         } else {
 275                 struct test_bulk_req_v1 *breq = &tsi->tsi_u.bulk_v1;
 276
 277                 /* I should never get this step if it's unknown feature
 278                  * because make_session will reject unknown feature */
 279                 LASSERT((sn->sn_features & ~LST_FEATS_MASK) == 0);
 280
 281                 opc   = breq->blk_opc;
 282                 flags = breq->blk_flags;
 283                 len   = breq->blk_len;
 284                 off   = breq->blk_offset;
 285         }
 286         npg   = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 287
 288         rc = sfw_create_test_rpc(tsu, dest, sn->sn_features, npg, len, &rpc);
 289         if (rc != 0)
 290                 return rc;
 291
 292         unsafe_memcpy(&rpc->crpc_bulk, bulk,
 293                       offsetof(struct srpc_bulk, bk_iovs[npg]),
 294                       FLEXIBLE_OBJECT);
 295         if (opc == LST_BRW_WRITE)
 296                 brw_fill_bulk(&rpc->crpc_bulk, flags, BRW_MAGIC);
 297         else
 298                 brw_fill_bulk(&rpc->crpc_bulk, flags, BRW_POISON);
 299
 300         req = &rpc->crpc_reqstmsg.msg_body.brw_reqst;
 301         req->brw_flags = flags;
 302         req->brw_rw    = opc;
 303         req->brw_len   = len;
 304
 305         *rpcpp = rpc;
 306         return 0;
 307 }
 308
 309 static void
 310 brw_client_done_rpc(struct sfw_test_unit *tsu, struct srpc_client_rpc *rpc)
 311 {
 312         __u64 magic = BRW_MAGIC;
 313         struct sfw_test_instance *tsi = tsu->tsu_instance;
 314         struct sfw_session *sn = tsi->tsi_batch->bat_session;
 315         struct srpc_msg *msg = &rpc->crpc_replymsg;
 316         struct srpc_brw_reply *reply = &msg->msg_body.brw_reply;
 317         struct srpc_brw_reqst *reqst = &rpc->crpc_reqstmsg.msg_body.brw_reqst;
 318
 319         LASSERT(sn != NULL);
 320
 321         if (rpc->crpc_status != 0) {
 322                 CERROR("BRW RPC to %s failed with %d\n",
 323                        libcfs_id2str(rpc->crpc_dest), rpc->crpc_status);
 324                 if (!tsi->tsi_stopping) /* rpc could have been aborted */
 325                         atomic_inc(&sn->sn_brw_errors);
 326                 return;
 327         }
 328
 329         if (msg->msg_magic != SRPC_MSG_MAGIC) {
 330                 __swab64s(&magic);
 331                 __swab32s(&reply->brw_status);
 332         }
 333
 334         CDEBUG(reply->brw_status ? D_WARNING : D_NET,
 335                "BRW RPC to %s finished with brw_status: %d\n",
 336                libcfs_id2str(rpc->crpc_dest), reply->brw_status);
 337
 338         if (reply->brw_status != 0) {
 339                 atomic_inc(&sn->sn_brw_errors);
 340                 rpc->crpc_status = -(int)reply->brw_status;
 341                 return;
 342         }
 343
 344         if (reqst->brw_rw == LST_BRW_WRITE)
 345                 return;
 346
 347         if (brw_check_bulk(&rpc->crpc_bulk, reqst->brw_flags, magic) != 0) {
 348                 CERROR("Bulk data from %s is corrupted!\n",
 349                        libcfs_id2str(rpc->crpc_dest));
 350                 atomic_inc(&sn->sn_brw_errors);
 351                 rpc->crpc_status = -EBADMSG;
 352         }
 353 }
 354
 355 static void
 356 brw_server_rpc_done(struct srpc_server_rpc *rpc)
 357 {
 358         struct srpc_bulk *blk = rpc->srpc_bulk;
 359
 360         if (blk == NULL)
 361                 return;
 362
 363         if (rpc->srpc_status != 0)
 364                 CERROR("Bulk transfer %s %s has failed: %d\n",
 365                        blk->bk_sink ? "from" : "to",
 366                        libcfs_id2str(rpc->srpc_peer), rpc->srpc_status);
 367         else
 368                 CDEBUG(D_NET, "Transferred %d pages bulk data %s %s\n",
 369                        blk->bk_niov, blk->bk_sink ? "from" : "to",
 370                        libcfs_id2str(rpc->srpc_peer));
 371 }
 372
 373 static int
 374 brw_bulk_ready(struct srpc_server_rpc *rpc, int status)
 375 {
 376         __u64 magic = BRW_MAGIC;
 377         struct srpc_brw_reply *reply = &rpc->srpc_replymsg.msg_body.brw_reply;
 378         struct srpc_brw_reqst *reqst;
 379         struct srpc_msg *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         struct srpc_msg *replymsg = &rpc->srpc_replymsg;
 414         struct srpc_msg *reqstmsg = &rpc->srpc_reqstbuf->buf_msg;
 415         struct srpc_brw_reply *reply = &replymsg->msg_body.brw_reply;
 416         struct srpc_brw_reqst *reqst = &reqstmsg->msg_body.brw_reqst;
 417
 418         LASSERT (sv->sv_id == SRPC_SERVICE_BRW);
 419
 420         if (reqstmsg->msg_magic != SRPC_MSG_MAGIC) {
 421                 LASSERT (reqstmsg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 422
 423                 __swab32s(&reqst->brw_rw);
 424                 __swab32s(&reqst->brw_len);
 425                 __swab32s(&reqst->brw_flags);
 426                 __swab64s(&reqst->brw_rpyid);
 427                 __swab64s(&reqst->brw_bulkid);
 428         }
 429         LASSERT (reqstmsg->msg_type == (__u32)srpc_service2request(sv->sv_id));
 430
 431         reply->brw_status = 0;
 432         rpc->srpc_done = brw_server_rpc_done;
 433
 434         if ((reqst->brw_rw != LST_BRW_READ && reqst->brw_rw != LST_BRW_WRITE) ||
 435             (reqst->brw_flags != LST_BRW_CHECK_NONE &&
 436              reqst->brw_flags != LST_BRW_CHECK_FULL &&
 437              reqst->brw_flags != LST_BRW_CHECK_SIMPLE)) {
 438                 reply->brw_status = EINVAL;
 439                 return 0;
 440         }
 441
 442         if ((reqstmsg->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
 443                 replymsg->msg_ses_feats = LST_FEATS_MASK;
 444                 reply->brw_status = EPROTO;
 445                 return 0;
 446         }
 447
 448         if ((reqstmsg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
 449                 /* compat with old version */
 450                 if ((reqst->brw_len & ~PAGE_MASK) != 0) {
 451                         reply->brw_status = EINVAL;
 452                         return 0;
 453                 }
 454         }
 455
 456         replymsg->msg_ses_feats = reqstmsg->msg_ses_feats;
 457
 458         if (reqst->brw_len == 0 || reqst->brw_len > LNET_MTU) {
 459                 reply->brw_status = EINVAL;
 460                 return 0;
 461         }
 462
 463         srpc_init_bulk(rpc->srpc_bulk, 0, reqst->brw_len,
 464                        reqst->brw_rw == LST_BRW_WRITE);
 465
 466         if (reqst->brw_rw == LST_BRW_READ)
 467                 brw_fill_bulk(rpc->srpc_bulk, reqst->brw_flags, BRW_MAGIC);
 468         else
 469                 brw_fill_bulk(rpc->srpc_bulk, reqst->brw_flags, BRW_POISON);
 470
 471         return 0;
 472 }
 473
 474 static int
 475 brw_srpc_init(struct srpc_server_rpc *rpc, int cpt)
 476 {
 477         /* just alloc a maximal size - actual values will be adjusted later */
 478         rpc->srpc_bulk = srpc_alloc_bulk(cpt, LNET_MTU);
 479         if (rpc->srpc_bulk == NULL)
 480                 return -ENOMEM;
 481
 482         srpc_init_bulk(rpc->srpc_bulk, 0, 0, 0);
 483
 484         return 0;
 485 }
 486
 487 static void
 488 brw_srpc_fini(struct srpc_server_rpc *rpc)
 489 {
 490         srpc_free_bulk(rpc->srpc_bulk);
 491         rpc->srpc_bulk = NULL;
 492 }
 493
 494 struct sfw_test_client_ops brw_test_client = {
 495         .tso_init       = brw_client_init,
 496         .tso_fini       = brw_client_fini,
 497         .tso_prep_rpc   = brw_client_prep_rpc,
 498         .tso_done_rpc   = brw_client_done_rpc,
 499 };
 500
 501 struct srpc_service brw_test_service = {
 502         .sv_id         = SRPC_SERVICE_BRW,
 503         .sv_name       = "brw_test",
 504         .sv_handler    = brw_server_handle,
 505         .sv_bulk_ready = brw_bulk_ready,
 506
 507         .sv_srpc_init  = brw_srpc_init,
 508         .sv_srpc_fini  = brw_srpc_fini,
 509 };
 510
 511 void brw_init_test_service(void)
 512 {
 513         unsigned long cache_size = cfs_totalram_pages() >> 4;
 514
 515         /* brw prealloc cache should don't eat more than half memory */
 516         cache_size /= ((LNET_MTU >> PAGE_SHIFT) + 1) ;
 517
 518         brw_test_service.sv_wi_total   = brw_srv_workitems;
 519
 520         if (brw_test_service.sv_wi_total > cache_size)
 521                 brw_test_service.sv_wi_total = cache_size;
 522 }