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 = data;
 220     g_main_loop_run(loop);
 221     return NULL;
 222 }
 223
 224 static int chr_can_read(void *opaque)
 225 {
 226     return VHOST_USER_HDR_SIZE;
 227 }
 228
 229 static void chr_read(void *opaque, const uint8_t *buf, int size)
 230 {
 231     TestServer *s = opaque;
 232     CharDriverState *chr = s->chr;
 233     VhostUserMsg msg;
 234     uint8_t *p = (uint8_t *) &msg;
 235     int fd;
 236
 237     if (size != VHOST_USER_HDR_SIZE) {
 238         g_test_message("Wrong message size received %d\n", size);
 239         return;
 240     }
 241
 242     g_mutex_lock(&s->data_mutex);
 243     memcpy(p, buf, VHOST_USER_HDR_SIZE);
 244
 245     if (msg.size) {
 246         p += VHOST_USER_HDR_SIZE;
 247         g_assert_cmpint(qemu_chr_fe_read_all(chr, p, msg.size), ==, msg.size);
 248     }
 249
 250     switch (msg.request) {
 251     case VHOST_USER_GET_FEATURES:
 252         /* send back features to qemu */
 253         msg.flags |= VHOST_USER_REPLY_MASK;
 254         msg.size = sizeof(m.payload.u64);
 255         msg.payload.u64 = 0x1ULL << VHOST_F_LOG_ALL |
 256             0x1ULL << VHOST_USER_F_PROTOCOL_FEATURES;
 257         p = (uint8_t *) &msg;
 258         qemu_chr_fe_write_all(chr, p, VHOST_USER_HDR_SIZE + msg.size);
 259         break;
 260
 261     case VHOST_USER_SET_FEATURES:
 262         g_assert_cmpint(msg.payload.u64 & (0x1ULL << VHOST_USER_F_PROTOCOL_FEATURES),
 263                         !=, 0ULL);
 264         break;
 265
 266     case VHOST_USER_GET_PROTOCOL_FEATURES:
 267         /* send back features to qemu */
 268         msg.flags |= VHOST_USER_REPLY_MASK;
 269         msg.size = sizeof(m.payload.u64);
 270         msg.payload.u64 = 1 << VHOST_USER_PROTOCOL_F_LOG_SHMFD;
 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_GET_VRING_BASE:
 276         /* send back vring base to qemu */
 277         msg.flags |= VHOST_USER_REPLY_MASK;
 278         msg.size = sizeof(m.payload.state);
 279         msg.payload.state.num = 0;
 280         p = (uint8_t *) &msg;
 281         qemu_chr_fe_write_all(chr, p, VHOST_USER_HDR_SIZE + msg.size);
 282
 283         assert(msg.payload.state.index < 2);
 284         s->rings &= ~(0x1ULL << msg.payload.state.index);
 285         break;
 286
 287     case VHOST_USER_SET_MEM_TABLE:
 288         /* received the mem table */
 289         memcpy(&s->memory, &msg.payload.memory, sizeof(msg.payload.memory));
 290         s->fds_num = qemu_chr_fe_get_msgfds(chr, s->fds, G_N_ELEMENTS(s->fds));
 291
 292         /* signal the test that it can continue */
 293         g_cond_signal(&s->data_cond);
 294         break;
 295
 296     case VHOST_USER_SET_VRING_KICK:
 297     case VHOST_USER_SET_VRING_CALL:
 298         /* consume the fd */
 299         qemu_chr_fe_get_msgfds(chr, &fd, 1);
 300         /*
 301          * This is a non-blocking eventfd.
 302          * The receive function forces it to be blocking,
 303          * so revert it back to non-blocking.
 304          */
 305         qemu_set_nonblock(fd);
 306         break;
 307
 308     case VHOST_USER_SET_LOG_BASE:
 309         if (s->log_fd != -1) {
 310             close(s->log_fd);
 311             s->log_fd = -1;
 312         }
 313         qemu_chr_fe_get_msgfds(chr, &s->log_fd, 1);
 314         msg.flags |= VHOST_USER_REPLY_MASK;
 315         msg.size = 0;
 316         p = (uint8_t *) &msg;
 317         qemu_chr_fe_write_all(chr, p, VHOST_USER_HDR_SIZE);
 318
 319         g_cond_signal(&s->data_cond);
 320         break;
 321
 322     case VHOST_USER_SET_VRING_BASE:
 323         assert(msg.payload.state.index < 2);
 324         s->rings |= 0x1ULL << msg.payload.state.index;
 325         break;
 326
 327     default:
 328         break;
 329     }
 330
 331     g_mutex_unlock(&s->data_mutex);
 332 }
 333
 334 static const char *init_hugepagefs(const char *path)
 335 {
 336     struct statfs fs;
 337     int ret;
 338
 339     if (access(path, R_OK | W_OK | X_OK)) {
 340         g_test_message("access on path (%s): %s\n", path, strerror(errno));
 341         return NULL;
 342     }
 343
 344     do {
 345         ret = statfs(path, &fs);
 346     } while (ret != 0 && errno == EINTR);
 347
 348     if (ret != 0) {
 349         g_test_message("statfs on path (%s): %s\n", path, strerror(errno));
 350         return NULL;
 351     }
 352
 353     if (fs.f_type != HUGETLBFS_MAGIC) {
 354         g_test_message("Warning: path not on HugeTLBFS: %s\n", path);
 355         return NULL;
 356     }
 357
 358     return path;
 359 }
 360
 361 static TestServer *test_server_new(const gchar *name)
 362 {
 363     TestServer *server = g_new0(TestServer, 1);
 364     gchar *chr_path;
 365
 366     server->socket_path = g_strdup_printf("%s/%s.sock", tmpfs, name);
 367
 368     chr_path = g_strdup_printf("unix:%s,server,nowait", server->socket_path);
 369     server->chr_name = g_strdup_printf("chr-%s", name);
 370     server->chr = qemu_chr_new(server->chr_name, chr_path, NULL);
 371     g_free(chr_path);
 372
 373     qemu_chr_add_handlers(server->chr, chr_can_read, chr_read, NULL, server);
 374
 375     g_mutex_init(&server->data_mutex);
 376     g_cond_init(&server->data_cond);
 377
 378     server->log_fd = -1;
 379
 380     return server;
 381 }
 382
 383 #define GET_QEMU_CMD(s)                                                        \
 384     g_strdup_printf(QEMU_CMD, 512, 512, (root), (s)->chr_name,                 \
 385                     (s)->socket_path, (s)->chr_name)
 386
 387 #define GET_QEMU_CMDE(s, mem, extra, ...)                                      \
 388     g_strdup_printf(QEMU_CMD extra, (mem), (mem), (root), (s)->chr_name,       \
 389                     (s)->socket_path, (s)->chr_name, ##__VA_ARGS__)
 390
 391 static gboolean _test_server_free(TestServer *server)
 392 {
 393     int i;
 394
 395     qemu_chr_delete(server->chr);
 396
 397     for (i = 0; i < server->fds_num; i++) {
 398         close(server->fds[i]);
 399     }
 400
 401     if (server->log_fd != -1) {
 402         close(server->log_fd);
 403     }
 404
 405     unlink(server->socket_path);
 406     g_free(server->socket_path);
 407
 408     g_free(server->chr_name);
 409     g_free(server);
 410
 411     return FALSE;
 412 }
 413
 414 static void test_server_free(TestServer *server)
 415 {
 416     g_idle_add((GSourceFunc)_test_server_free, server);
 417 }
 418
 419 static void wait_for_log_fd(TestServer *s)
 420 {
 421     gint64 end_time;
 422
 423     g_mutex_lock(&s->data_mutex);
 424     end_time = g_get_monotonic_time() + 5 * G_TIME_SPAN_SECOND;
 425     while (s->log_fd == -1) {
 426         if (!g_cond_wait_until(&s->data_cond, &s->data_mutex, end_time)) {
 427             /* timeout has passed */
 428             g_assert(s->log_fd != -1);
 429             break;
 430         }
 431     }
 432
 433     g_mutex_unlock(&s->data_mutex);
 434 }
 435
 436 static void write_guest_mem(TestServer *s, uint32 seed)
 437 {
 438     uint32_t *guest_mem;
 439     int i, j;
 440     size_t size;
 441
 442     wait_for_fds(s);
 443
 444     /* iterate all regions */
 445     for (i = 0; i < s->fds_num; i++) {
 446
 447         /* We'll write only the region statring at 0x0 */
 448         if (s->memory.regions[i].guest_phys_addr != 0x0) {
 449             continue;
 450         }
 451
 452         g_assert_cmpint(s->memory.regions[i].memory_size, >, 1024);
 453
 454         size = s->memory.regions[i].memory_size +
 455             s->memory.regions[i].mmap_offset;
 456
 457         guest_mem = mmap(0, size, PROT_READ | PROT_WRITE,
 458                          MAP_SHARED, s->fds[i], 0);
 459
 460         g_assert(guest_mem != MAP_FAILED);
 461         guest_mem += (s->memory.regions[i].mmap_offset / sizeof(*guest_mem));
 462
 463         for (j = 0; j < 256; j++) {
 464             guest_mem[j] = seed + j;
 465         }
 466
 467         munmap(guest_mem, s->memory.regions[i].memory_size);
 468         break;
 469     }
 470 }
 471
 472 static guint64 get_log_size(TestServer *s)
 473 {
 474     guint64 log_size = 0;
 475     int i;
 476
 477     for (i = 0; i < s->memory.nregions; ++i) {
 478         VhostUserMemoryRegion *reg = &s->memory.regions[i];
 479         guint64 last = range_get_last(reg->guest_phys_addr,
 480                                        reg->memory_size);
 481         log_size = MAX(log_size, last / (8 * VHOST_LOG_PAGE) + 1);
 482     }
 483
 484     return log_size;
 485 }
 486
 487 typedef struct TestMigrateSource {
 488     GSource source;
 489     TestServer *src;
 490     TestServer *dest;
 491 } TestMigrateSource;
 492
 493 static gboolean
 494 test_migrate_source_check(GSource *source)
 495 {
 496     TestMigrateSource *t = (TestMigrateSource *)source;
 497     gboolean overlap = t->src->rings && t->dest->rings;
 498
 499     g_assert(!overlap);
 500
 501     return FALSE;
 502 }
 503
 504 GSourceFuncs test_migrate_source_funcs = {
 505     NULL,
 506     test_migrate_source_check,
 507     NULL,
 508     NULL
 509 };
 510
 511 static void test_migrate(void)
 512 {
 513     TestServer *s = test_server_new("src");
 514     TestServer *dest = test_server_new("dest");
 515     const char *uri = "tcp:127.0.0.1:1234";
 516     QTestState *global = global_qtest, *from, *to;
 517     GSource *source;
 518     gchar *cmd;
 519     QDict *rsp;
 520     guint8 *log;
 521     guint64 size;
 522
 523     cmd = GET_QEMU_CMDE(s, 2, "");
 524     from = qtest_start(cmd);
 525     g_free(cmd);
 526
 527     wait_for_fds(s);
 528     size = get_log_size(s);
 529     g_assert_cmpint(size, ==, (2 * 1024 * 1024) / (VHOST_LOG_PAGE * 8));
 530
 531     cmd = GET_QEMU_CMDE(dest, 2, " -incoming %s", uri);
 532     to = qtest_init(cmd);
 533     g_free(cmd);
 534
 535     source = g_source_new(&test_migrate_source_funcs,
 536                           sizeof(TestMigrateSource));
 537     ((TestMigrateSource *)source)->src = s;
 538     ((TestMigrateSource *)source)->dest = dest;
 539     g_source_attach(source, NULL);
 540
 541     /* slow down migration to have time to fiddle with log */
 542     /* TODO: qtest could learn to break on some places */
 543     rsp = qmp("{ 'execute': 'migrate_set_speed',"
 544               "'arguments': { 'value': 10 } }");
 545     g_assert(qdict_haskey(rsp, "return"));
 546     QDECREF(rsp);
 547
 548     cmd = g_strdup_printf("{ 'execute': 'migrate',"
 549                           "'arguments': { 'uri': '%s' } }",
 550                           uri);
 551     rsp = qmp(cmd);
 552     g_free(cmd);
 553     g_assert(qdict_haskey(rsp, "return"));
 554     QDECREF(rsp);
 555
 556     wait_for_log_fd(s);
 557
 558     log = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, s->log_fd, 0);
 559     g_assert(log != MAP_FAILED);
 560
 561     /* modify first page */
 562     write_guest_mem(s, 0x42);
 563     log[0] = 1;
 564     munmap(log, size);
 565
 566     /* speed things up */
 567     rsp = qmp("{ 'execute': 'migrate_set_speed',"
 568               "'arguments': { 'value': 0 } }");
 569     g_assert(qdict_haskey(rsp, "return"));
 570     QDECREF(rsp);
 571
 572     qmp_eventwait("STOP");
 573
 574     global_qtest = to;
 575     qmp_eventwait("RESUME");
 576
 577     read_guest_mem(dest);
 578
 579     g_source_destroy(source);
 580     g_source_unref(source);
 581
 582     qtest_quit(to);
 583     test_server_free(dest);
 584     qtest_quit(from);
 585     test_server_free(s);
 586
 587     global_qtest = global;
 588 }
 589
 590 int main(int argc, char **argv)
 591 {
 592     QTestState *s = NULL;
 593     TestServer *server = NULL;
 594     const char *hugefs;
 595     char *qemu_cmd = NULL;
 596     int ret;
 597     char template[] = "/tmp/vhost-test-XXXXXX";
 598     GMainLoop *loop;
 599     GThread *thread;
 600
 601     g_test_init(&argc, &argv, NULL);
 602
 603     module_call_init(MODULE_INIT_QOM);
 604     qemu_add_opts(&qemu_chardev_opts);
 605
 606     tmpfs = mkdtemp(template);
 607     if (!tmpfs) {
 608         g_test_message("mkdtemp on path (%s): %s\n", template, strerror(errno));
 609     }
 610     g_assert(tmpfs);
 611
 612     hugefs = getenv("QTEST_HUGETLBFS_PATH");
 613     if (hugefs) {
 614         root = init_hugepagefs(hugefs);
 615         g_assert(root);
 616     } else {
 617         root = tmpfs;
 618     }
 619
 620     server = test_server_new("test");
 621
 622     loop = g_main_loop_new(NULL, FALSE);
 623     /* run the main loop thread so the chardev may operate */
 624     thread = g_thread_new(NULL, thread_function, loop);
 625
 626     qemu_cmd = GET_QEMU_CMD(server);
 627
 628     s = qtest_start(qemu_cmd);
 629     g_free(qemu_cmd);
 630
 631     qtest_add_data_func("/vhost-user/read-guest-mem", server, read_guest_mem);
 632     qtest_add_func("/vhost-user/migrate", test_migrate);
 633
 634     ret = g_test_run();
 635
 636     if (s) {
 637         qtest_quit(s);
 638     }
 639
 640     /* cleanup */
 641     test_server_free(server);
 642
 643     /* finish the helper thread and dispatch pending sources */
 644     g_main_loop_quit(loop);
 645     g_thread_join(thread);
 646     while (g_main_context_pending(NULL)) {
 647         g_main_context_iteration (NULL, TRUE);
 648     }
 649     g_main_loop_unref(loop);
 650
 651     ret = rmdir(tmpfs);
 652     if (ret != 0) {
 653         g_test_message("unable to rmdir: path (%s): %s\n",
 654                        tmpfs, strerror(errno));
 655     }
 656     g_assert_cmpint(ret, ==, 0);
 657
 658     return ret;
 659 }