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