tests/vhost-user-test.c

   1 /*
   2  * QTest testcase for the vhost-user
   3  *
   4  * Copyright (c) 2014 Virtual Open Systems Sarl.
   5  *
   6  * This work is licensed under the terms of the GNU GPL, version 2 or later.
   7  * See the COPYING file in the top-level directory.
   8  *
   9  */
  10
  11 #include <glib.h>
  12
  13 #include "libqtest.h"
  14 #include "qemu/option.h"
  15 #include "qemu/range.h"
  16 #include "sysemu/char.h"
  17 #include "sysemu/sysemu.h"
  18
  19 #include <linux/vhost.h>
  20 #include <sys/mman.h>
  21 #include <sys/vfs.h>
  22 #include <qemu/sockets.h>
  23
  24 /* GLIB version compatibility flags */
  25 #if !GLIB_CHECK_VERSION(2, 26, 0)
  26 #define G_TIME_SPAN_SECOND              (G_GINT64_CONSTANT(1000000))
  27 #endif
  28
  29 #if GLIB_CHECK_VERSION(2, 28, 0)
  30 #define HAVE_MONOTONIC_TIME
  31 #endif
  32
  33 #define QEMU_CMD_ACCEL  " -machine accel=tcg"
  34 #define QEMU_CMD_MEM    " -m %d -object memory-backend-file,id=mem,size=%dM,"\
  35                         "mem-path=%s,share=on -numa node,memdev=mem"
  36 #define QEMU_CMD_CHR    " -chardev socket,id=%s,path=%s"
  37 #define QEMU_CMD_NETDEV " -netdev vhost-user,id=net0,chardev=%s,vhostforce"
  38 #define QEMU_CMD_NET    " -device virtio-net-pci,netdev=net0 "
  39 #define QEMU_CMD_ROM    " -option-rom ../pc-bios/pxe-virtio.rom"
  40
  41 #define QEMU_CMD        QEMU_CMD_ACCEL QEMU_CMD_MEM QEMU_CMD_CHR \
  42                         QEMU_CMD_NETDEV QEMU_CMD_NET QEMU_CMD_ROM
  43
  44 #define HUGETLBFS_MAGIC       0x958458f6
  45
  46 /*********** FROM hw/virtio/vhost-user.c *************************************/
  47
  48 #define VHOST_MEMORY_MAX_NREGIONS    8
  49
  50 #define VHOST_USER_F_PROTOCOL_FEATURES 30
  51 #define VHOST_USER_PROTOCOL_F_LOG_SHMFD 1
  52
  53 #define VHOST_LOG_PAGE 0x1000
  54
  55 typedef enum VhostUserRequest {
  56     VHOST_USER_NONE = 0,
  57     VHOST_USER_GET_FEATURES = 1,
  58     VHOST_USER_SET_FEATURES = 2,
  59     VHOST_USER_SET_OWNER = 3,
  60     VHOST_USER_RESET_OWNER = 4,
  61     VHOST_USER_SET_MEM_TABLE = 5,
  62     VHOST_USER_SET_LOG_BASE = 6,
  63     VHOST_USER_SET_LOG_FD = 7,
  64     VHOST_USER_SET_VRING_NUM = 8,
  65     VHOST_USER_SET_VRING_ADDR = 9,
  66     VHOST_USER_SET_VRING_BASE = 10,
  67     VHOST_USER_GET_VRING_BASE = 11,
  68     VHOST_USER_SET_VRING_KICK = 12,
  69     VHOST_USER_SET_VRING_CALL = 13,
  70     VHOST_USER_SET_VRING_ERR = 14,
  71     VHOST_USER_GET_PROTOCOL_FEATURES = 15,
  72     VHOST_USER_SET_PROTOCOL_FEATURES = 16,
  73     VHOST_USER_SET_VRING_ENABLE = 18,
  74     VHOST_USER_MAX
  75 } VhostUserRequest;
  76
  77 typedef struct VhostUserMemoryRegion {
  78     uint64_t guest_phys_addr;
  79     uint64_t memory_size;
  80     uint64_t userspace_addr;
  81     uint64_t mmap_offset;
  82 } VhostUserMemoryRegion;
  83
  84 typedef struct VhostUserMemory {
  85     uint32_t nregions;
  86     uint32_t padding;
  87     VhostUserMemoryRegion regions[VHOST_MEMORY_MAX_NREGIONS];
  88 } VhostUserMemory;
  89
  90 typedef struct VhostUserLog {
  91     uint64_t mmap_size;
  92     uint64_t mmap_offset;
  93 } VhostUserLog;
  94
  95 typedef struct VhostUserMsg {
  96     VhostUserRequest request;
  97
  98 #define VHOST_USER_VERSION_MASK     (0x3)
  99 #define VHOST_USER_REPLY_MASK       (0x1<<2)
 100     uint32_t flags;
 101     uint32_t size; /* the following payload size */
 102     union {
 103 #define VHOST_USER_VRING_IDX_MASK   (0xff)
 104 #define VHOST_USER_VRING_NOFD_MASK  (0x1<<8)
 105         uint64_t u64;
 106         struct vhost_vring_state state;
 107         struct vhost_vring_addr addr;
 108         VhostUserMemory memory;
 109         VhostUserLog log;
 110     } payload;
 111 } QEMU_PACKED VhostUserMsg;
 112
 113 static VhostUserMsg m __attribute__ ((unused));
 114 #define VHOST_USER_HDR_SIZE (sizeof(m.request) \
 115                             + sizeof(m.flags) \
 116                             + sizeof(m.size))
 117
 118 #define VHOST_USER_PAYLOAD_SIZE (sizeof(m) - VHOST_USER_HDR_SIZE)
 119
 120 /* The version of the protocol we support */
 121 #define VHOST_USER_VERSION    (0x1)
 122 /*****************************************************************************/
 123
 124 typedef struct TestServer {
 125     gchar *socket_path;
 126     gchar *chr_name;
 127     CharDriverState *chr;
 128     int fds_num;
 129     int fds[VHOST_MEMORY_MAX_NREGIONS];
 130     VhostUserMemory memory;
 131     GMutex data_mutex;
 132     GCond data_cond;
 133     int log_fd;
 134     uint64_t rings;
 135 } TestServer;
 136
 137 #if !GLIB_CHECK_VERSION(2, 32, 0)
 138 static gboolean g_cond_wait_until(CompatGCond cond, CompatGMutex mutex,
 139                                   gint64 end_time)
 140 {
 141     gboolean ret = FALSE;
 142     end_time -= g_get_monotonic_time();
 143     GTimeVal time = { end_time / G_TIME_SPAN_SECOND,
 144                       end_time % G_TIME_SPAN_SECOND };
 145     ret = g_cond_timed_wait(cond, mutex, &time);
 146     return ret;
 147 }
 148 #endif
 149
 150 static const char *tmpfs;
 151 static const char *root;
 152
 153 static void wait_for_fds(TestServer *s)
 154 {
 155     gint64 end_time;
 156
 157     g_mutex_lock(&s->data_mutex);
 158
 159     end_time = g_get_monotonic_time() + 5 * G_TIME_SPAN_SECOND;
 160     while (!s->fds_num) {
 161         if (!g_cond_wait_until(&s->data_cond, &s->data_mutex, end_time)) {
 162             /* timeout has passed */
 163             g_assert(s->fds_num);
 164             break;
 165         }
 166     }
 167
 168     /* check for sanity */
 169     g_assert_cmpint(s->fds_num, >, 0);
 170     g_assert_cmpint(s->fds_num, ==, s->memory.nregions);
 171
 172     g_mutex_unlock(&s->data_mutex);
 173 }
 174
 175 static void read_guest_mem(TestServer *s)
 176 {
 177     uint32_t *guest_mem;
 178     int i, j;
 179     size_t size;
 180
 181     wait_for_fds(s);
 182
 183     g_mutex_lock(&s->data_mutex);
 184
 185     /* iterate all regions */
 186     for (i = 0; i < s->fds_num; i++) {
 187
 188         /* We'll check only the region statring at 0x0*/
 189         if (s->memory.regions[i].guest_phys_addr != 0x0) {
 190             continue;
 191         }
 192
 193         g_assert_cmpint(s->memory.regions[i].memory_size, >, 1024);
 194
 195         size = s->memory.regions[i].memory_size +
 196             s->memory.regions[i].mmap_offset;
 197
 198         guest_mem = mmap(0, size, PROT_READ | PROT_WRITE,
 199                          MAP_SHARED, s->fds[i], 0);
 200
 201         g_assert(guest_mem != MAP_FAILED);
 202         guest_mem += (s->memory.regions[i].mmap_offset / sizeof(*guest_mem));
 203
 204         for (j = 0; j < 256; j++) {
 205             uint32_t a = readl(s->memory.regions[i].guest_phys_addr + j*4);
 206             uint32_t b = guest_mem[j];
 207
 208             g_assert_cmpint(a, ==, b);
 209         }
 210
 211         munmap(guest_mem, s->memory.regions[i].memory_size);
 212     }
 213
 214     g_mutex_unlock(&s->data_mutex);
 215 }
 216
 217 static void *thread_function(void *data)
 218 {
 219     GMainLoop *loop;
 220     loop = g_main_loop_new(NULL, FALSE);
 221     g_main_loop_run(loop);
 222     return NULL;
 223 }
 224
 225 static int chr_can_read(void *opaque)
 226 {
 227     return VHOST_USER_HDR_SIZE;
 228 }
 229
 230 static void chr_read(void *opaque, const uint8_t *buf, int size)
 231 {
 232     TestServer *s = opaque;
 233     CharDriverState *chr = s->chr;
 234     VhostUserMsg msg;
 235     uint8_t *p = (uint8_t *) &msg;
 236     int fd;
 237
 238     if (size != VHOST_USER_HDR_SIZE) {
 239         g_test_message("Wrong message size received %d\n", size);
 240         return;
 241     }
 242
 243     g_mutex_lock(&s->data_mutex);
 244     memcpy(p, buf, VHOST_USER_HDR_SIZE);
 245
 246     if (msg.size) {
 247         p += VHOST_USER_HDR_SIZE;
 248         g_assert_cmpint(qemu_chr_fe_read_all(chr, p, msg.size), ==, msg.size);
 249     }
 250
 251     switch (msg.request) {
 252     case VHOST_USER_GET_FEATURES:
 253         /* send back features to qemu */
 254         msg.flags |= VHOST_USER_REPLY_MASK;
 255         msg.size = sizeof(m.payload.u64);
 256         msg.payload.u64 = 0x1ULL << VHOST_F_LOG_ALL |
 257             0x1ULL << VHOST_USER_F_PROTOCOL_FEATURES;
 258         p = (uint8_t *) &msg;
 259         qemu_chr_fe_write_all(chr, p, VHOST_USER_HDR_SIZE + msg.size);
 260         break;
 261
 262     case VHOST_USER_SET_FEATURES:
 263         g_assert_cmpint(msg.payload.u64 & (0x1ULL << VHOST_USER_F_PROTOCOL_FEATURES),
 264                         !=, 0ULL);
 265         break;
 266
 267     case VHOST_USER_GET_PROTOCOL_FEATURES:
 268         /* send back features to qemu */
 269         msg.flags |= VHOST_USER_REPLY_MASK;
 270         msg.size = sizeof(m.payload.u64);
 271         msg.payload.u64 = 1 << VHOST_USER_PROTOCOL_F_LOG_SHMFD;
 272         p = (uint8_t *) &msg;
 273         qemu_chr_fe_write_all(chr, p, VHOST_USER_HDR_SIZE + msg.size);
 274         break;
 275
 276     case VHOST_USER_GET_VRING_BASE:
 277         /* send back vring base to qemu */
 278         msg.flags |= VHOST_USER_REPLY_MASK;
 279         msg.size = sizeof(m.payload.state);
 280         msg.payload.state.num = 0;
 281         p = (uint8_t *) &msg;
 282         qemu_chr_fe_write_all(chr, p, VHOST_USER_HDR_SIZE + msg.size);
 283
 284         assert(msg.payload.state.index < 2);
 285         s->rings &= ~(0x1ULL << msg.payload.state.index);
 286         break;
 287
 288     case VHOST_USER_SET_MEM_TABLE:
 289         /* received the mem table */
 290         memcpy(&s->memory, &msg.payload.memory, sizeof(msg.payload.memory));
 291         s->fds_num = qemu_chr_fe_get_msgfds(chr, s->fds, G_N_ELEMENTS(s->fds));
 292
 293         /* signal the test that it can continue */
 294         g_cond_signal(&s->data_cond);
 295         break;
 296
 297     case VHOST_USER_SET_VRING_KICK:
 298     case VHOST_USER_SET_VRING_CALL:
 299         /* consume the fd */
 300         qemu_chr_fe_get_msgfds(chr, &fd, 1);
 301         /*
 302          * This is a non-blocking eventfd.
 303          * The receive function forces it to be blocking,
 304          * so revert it back to non-blocking.
 305          */
 306         qemu_set_nonblock(fd);
 307         break;
 308
 309     case VHOST_USER_SET_LOG_BASE:
 310         if (s->log_fd != -1) {
 311             close(s->log_fd);
 312             s->log_fd = -1;
 313         }
 314         qemu_chr_fe_get_msgfds(chr, &s->log_fd, 1);
 315         msg.flags |= VHOST_USER_REPLY_MASK;
 316         msg.size = 0;
 317         p = (uint8_t *) &msg;
 318         qemu_chr_fe_write_all(chr, p, VHOST_USER_HDR_SIZE);
 319
 320         g_cond_signal(&s->data_cond);
 321         break;
 322
 323     case VHOST_USER_SET_VRING_BASE:
 324         assert(msg.payload.state.index < 2);
 325         s->rings |= 0x1ULL << msg.payload.state.index;
 326         break;
 327
 328     default:
 329         break;
 330     }
 331
 332     g_mutex_unlock(&s->data_mutex);
 333 }
 334
 335 static const char *init_hugepagefs(const char *path)
 336 {
 337     struct statfs fs;
 338     int ret;
 339
 340     if (access(path, R_OK | W_OK | X_OK)) {
 341         g_test_message("access on path (%s): %s\n", path, strerror(errno));
 342         return NULL;
 343     }
 344
 345     do {
 346         ret = statfs(path, &fs);
 347     } while (ret != 0 && errno == EINTR);
 348
 349     if (ret != 0) {
 350         g_test_message("statfs on path (%s): %s\n", path, strerror(errno));
 351         return NULL;
 352     }
 353
 354     if (fs.f_type != HUGETLBFS_MAGIC) {
 355         g_test_message("Warning: path not on HugeTLBFS: %s\n", path);
 356         return NULL;
 357     }
 358
 359     return path;
 360 }
 361
 362 static TestServer *test_server_new(const gchar *name)
 363 {
 364     TestServer *server = g_new0(TestServer, 1);
 365     gchar *chr_path;
 366
 367     server->socket_path = g_strdup_printf("%s/%s.sock", tmpfs, name);
 368
 369     chr_path = g_strdup_printf("unix:%s,server,nowait", server->socket_path);
 370     server->chr_name = g_strdup_printf("chr-%s", name);
 371     server->chr = qemu_chr_new(server->chr_name, chr_path, NULL);
 372     g_free(chr_path);
 373
 374     qemu_chr_add_handlers(server->chr, chr_can_read, chr_read, NULL, server);
 375
 376     g_mutex_init(&server->data_mutex);
 377     g_cond_init(&server->data_cond);
 378
 379     server->log_fd = -1;
 380
 381     return server;
 382 }
 383
 384 #define GET_QEMU_CMD(s)                                                        \
 385     g_strdup_printf(QEMU_CMD, 512, 512, (root), (s)->chr_name,                 \
 386                     (s)->socket_path, (s)->chr_name)
 387
 388 #define GET_QEMU_CMDE(s, mem, extra, ...)                                      \
 389     g_strdup_printf(QEMU_CMD extra, (mem), (mem), (root), (s)->chr_name,       \
 390                     (s)->socket_path, (s)->chr_name, ##__VA_ARGS__)
 391
 392 static void test_server_free(TestServer *server)
 393 {
 394     int i;
 395
 396     qemu_chr_delete(server->chr);
 397
 398     for (i = 0; i < server->fds_num; i++) {
 399         close(server->fds[i]);
 400     }
 401
 402     if (server->log_fd != -1) {
 403         close(server->log_fd);
 404     }
 405
 406     unlink(server->socket_path);
 407     g_free(server->socket_path);
 408
 409
 410     g_free(server->chr_name);
 411     g_free(server);
 412 }
 413
 414 static void wait_for_log_fd(TestServer *s)
 415 {
 416     gint64 end_time;
 417
 418     g_mutex_lock(&s->data_mutex);
 419     end_time = g_get_monotonic_time() + 5 * G_TIME_SPAN_SECOND;
 420     while (s->log_fd == -1) {
 421         if (!g_cond_wait_until(&s->data_cond, &s->data_mutex, end_time)) {
 422             /* timeout has passed */
 423             g_assert(s->log_fd != -1);
 424             break;
 425         }
 426     }
 427
 428     g_mutex_unlock(&s->data_mutex);
 429 }
 430
 431 static void write_guest_mem(TestServer *s, uint32 seed)
 432 {
 433     uint32_t *guest_mem;
 434     int i, j;
 435     size_t size;
 436
 437     wait_for_fds(s);
 438
 439     /* iterate all regions */
 440     for (i = 0; i < s->fds_num; i++) {
 441
 442         /* We'll write only the region statring at 0x0 */
 443         if (s->memory.regions[i].guest_phys_addr != 0x0) {
 444             continue;
 445         }
 446
 447         g_assert_cmpint(s->memory.regions[i].memory_size, >, 1024);
 448
 449         size = s->memory.regions[i].memory_size +
 450             s->memory.regions[i].mmap_offset;
 451
 452         guest_mem = mmap(0, size, PROT_READ | PROT_WRITE,
 453                          MAP_SHARED, s->fds[i], 0);
 454
 455         g_assert(guest_mem != MAP_FAILED);
 456         guest_mem += (s->memory.regions[i].mmap_offset / sizeof(*guest_mem));
 457
 458         for (j = 0; j < 256; j++) {
 459             guest_mem[j] = seed + j;
 460         }
 461
 462         munmap(guest_mem, s->memory.regions[i].memory_size);
 463         break;
 464     }
 465 }
 466
 467 static guint64 get_log_size(TestServer *s)
 468 {
 469     guint64 log_size = 0;
 470     int i;
 471
 472     for (i = 0; i < s->memory.nregions; ++i) {
 473         VhostUserMemoryRegion *reg = &s->memory.regions[i];
 474         guint64 last = range_get_last(reg->guest_phys_addr,
 475                                        reg->memory_size);
 476         log_size = MAX(log_size, last / (8 * VHOST_LOG_PAGE) + 1);
 477     }
 478
 479     return log_size;
 480 }
 481
 482 typedef struct TestMigrateSource {
 483     GSource source;
 484     TestServer *src;
 485     TestServer *dest;
 486 } TestMigrateSource;
 487
 488 static gboolean
 489 test_migrate_source_check(GSource *source)
 490 {
 491     TestMigrateSource *t = (TestMigrateSource *)source;
 492     gboolean overlap = t->src->rings && t->dest->rings;
 493
 494     g_assert(!overlap);
 495
 496     return FALSE;
 497 }
 498
 499 GSourceFuncs test_migrate_source_funcs = {
 500     NULL,
 501     test_migrate_source_check,
 502     NULL,
 503     NULL
 504 };
 505
 506 static void test_migrate(void)
 507 {
 508     TestServer *s = test_server_new("src");
 509     TestServer *dest = test_server_new("dest");
 510     const char *uri = "tcp:127.0.0.1:1234";
 511     QTestState *global = global_qtest, *from, *to;
 512     GSource *source;
 513     gchar *cmd;
 514     QDict *rsp;
 515     guint8 *log;
 516     guint64 size;
 517
 518     cmd = GET_QEMU_CMDE(s, 2, "");
 519     from = qtest_start(cmd);
 520     g_free(cmd);
 521
 522     wait_for_fds(s);
 523     size = get_log_size(s);
 524     g_assert_cmpint(size, ==, (2 * 1024 * 1024) / (VHOST_LOG_PAGE * 8));
 525
 526     cmd = GET_QEMU_CMDE(dest, 2, " -incoming %s", uri);
 527     to = qtest_init(cmd);
 528     g_free(cmd);
 529
 530     source = g_source_new(&test_migrate_source_funcs,
 531                           sizeof(TestMigrateSource));
 532     ((TestMigrateSource *)source)->src = s;
 533     ((TestMigrateSource *)source)->dest = dest;
 534     g_source_attach(source, NULL);
 535
 536     /* slow down migration to have time to fiddle with log */
 537     /* TODO: qtest could learn to break on some places */
 538     rsp = qmp("{ 'execute': 'migrate_set_speed',"
 539               "'arguments': { 'value': 10 } }");
 540     g_assert(qdict_haskey(rsp, "return"));
 541     QDECREF(rsp);
 542
 543     cmd = g_strdup_printf("{ 'execute': 'migrate',"
 544                           "'arguments': { 'uri': '%s' } }",
 545                           uri);
 546     rsp = qmp(cmd);
 547     g_free(cmd);
 548     g_assert(qdict_haskey(rsp, "return"));
 549     QDECREF(rsp);
 550
 551     wait_for_log_fd(s);
 552
 553     log = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, s->log_fd, 0);
 554     g_assert(log != MAP_FAILED);
 555
 556     /* modify first page */
 557     write_guest_mem(s, 0x42);
 558     log[0] = 1;
 559     munmap(log, size);
 560
 561     /* speed things up */
 562     rsp = qmp("{ 'execute': 'migrate_set_speed',"
 563               "'arguments': { 'value': 0 } }");
 564     g_assert(qdict_haskey(rsp, "return"));
 565     QDECREF(rsp);
 566
 567     qmp_eventwait("STOP");
 568
 569     global_qtest = to;
 570     qmp_eventwait("RESUME");
 571
 572     read_guest_mem(dest);
 573
 574     g_source_destroy(source);
 575     g_source_unref(source);
 576
 577     qtest_quit(to);
 578     test_server_free(dest);
 579     qtest_quit(from);
 580     test_server_free(s);
 581
 582     global_qtest = global;
 583 }
 584
 585 int main(int argc, char **argv)
 586 {
 587     QTestState *s = NULL;
 588     TestServer *server = NULL;
 589     const char *hugefs;
 590     char *qemu_cmd = NULL;
 591     int ret;
 592     char template[] = "/tmp/vhost-test-XXXXXX";
 593
 594     g_test_init(&argc, &argv, NULL);
 595
 596     module_call_init(MODULE_INIT_QOM);
 597     qemu_add_opts(&qemu_chardev_opts);
 598
 599     tmpfs = mkdtemp(template);
 600     if (!tmpfs) {
 601         g_test_message("mkdtemp on path (%s): %s\n", template, strerror(errno));
 602     }
 603     g_assert(tmpfs);
 604
 605     hugefs = getenv("QTEST_HUGETLBFS_PATH");
 606     if (hugefs) {
 607         root = init_hugepagefs(hugefs);
 608         g_assert(root);
 609     } else {
 610         root = tmpfs;
 611     }
 612
 613     server = test_server_new("test");
 614
 615     /* run the main loop thread so the chardev may operate */
 616     g_thread_new(NULL, thread_function, NULL);
 617
 618     qemu_cmd = GET_QEMU_CMD(server);
 619
 620     s = qtest_start(qemu_cmd);
 621     g_free(qemu_cmd);
 622
 623     qtest_add_data_func("/vhost-user/read-guest-mem", server, read_guest_mem);
 624     qtest_add_func("/vhost-user/migrate", test_migrate);
 625
 626     ret = g_test_run();
 627
 628     if (s) {
 629         qtest_quit(s);
 630     }
 631
 632     /* cleanup */
 633     test_server_free(server);
 634
 635     ret = rmdir(tmpfs);
 636     if (ret != 0) {
 637         g_test_message("unable to rmdir: path (%s): %s\n",
 638                        tmpfs, strerror(errno));
 639     }
 640     g_assert_cmpint(ret, ==, 0);
 641
 642     return ret;
 643 }