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