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 } TestServer;
 135
 136 #if !GLIB_CHECK_VERSION(2, 32, 0)
 137 static gboolean g_cond_wait_until(CompatGCond cond, CompatGMutex mutex,
 138                                   gint64 end_time)
 139 {
 140     gboolean ret = FALSE;
 141     end_time -= g_get_monotonic_time();
 142     GTimeVal time = { end_time / G_TIME_SPAN_SECOND,
 143                       end_time % G_TIME_SPAN_SECOND };
 144     ret = g_cond_timed_wait(cond, mutex, &time);
 145     return ret;
 146 }
 147 #endif
 148
 149 static const char *tmpfs;
 150 static const char *root;
 151
 152 static void wait_for_fds(TestServer *s)
 153 {
 154     gint64 end_time;
 155
 156     g_mutex_lock(&s->data_mutex);
 157
 158     end_time = g_get_monotonic_time() + 5 * G_TIME_SPAN_SECOND;
 159     while (!s->fds_num) {
 160         if (!g_cond_wait_until(&s->data_cond, &s->data_mutex, end_time)) {
 161             /* timeout has passed */
 162             g_assert(s->fds_num);
 163             break;
 164         }
 165     }
 166
 167     /* check for sanity */
 168     g_assert_cmpint(s->fds_num, >, 0);
 169     g_assert_cmpint(s->fds_num, ==, s->memory.nregions);
 170
 171     g_mutex_unlock(&s->data_mutex);
 172 }
 173
 174 static void read_guest_mem(TestServer *s)
 175 {
 176     uint32_t *guest_mem;
 177     int i, j;
 178     size_t size;
 179
 180     wait_for_fds(s);
 181
 182     g_mutex_lock(&s->data_mutex);
 183
 184     /* iterate all regions */
 185     for (i = 0; i < s->fds_num; i++) {
 186
 187         /* We'll check only the region statring at 0x0*/
 188         if (s->memory.regions[i].guest_phys_addr != 0x0) {
 189             continue;
 190         }
 191
 192         g_assert_cmpint(s->memory.regions[i].memory_size, >, 1024);
 193
 194         size = s->memory.regions[i].memory_size +
 195             s->memory.regions[i].mmap_offset;
 196
 197         guest_mem = mmap(0, size, PROT_READ | PROT_WRITE,
 198                          MAP_SHARED, s->fds[i], 0);
 199
 200         g_assert(guest_mem != MAP_FAILED);
 201         guest_mem += (s->memory.regions[i].mmap_offset / sizeof(*guest_mem));
 202
 203         for (j = 0; j < 256; j++) {
 204             uint32_t a = readl(s->memory.regions[i].guest_phys_addr + j*4);
 205             uint32_t b = guest_mem[j];
 206
 207             g_assert_cmpint(a, ==, b);
 208         }
 209
 210         munmap(guest_mem, s->memory.regions[i].memory_size);
 211     }
 212
 213     g_mutex_unlock(&s->data_mutex);
 214 }
 215
 216 static void *thread_function(void *data)
 217 {
 218     GMainLoop *loop;
 219     loop = g_main_loop_new(NULL, FALSE);
 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         break;
 283
 284     case VHOST_USER_SET_MEM_TABLE:
 285         /* received the mem table */
 286         memcpy(&s->memory, &msg.payload.memory, sizeof(msg.payload.memory));
 287         s->fds_num = qemu_chr_fe_get_msgfds(chr, s->fds, G_N_ELEMENTS(s->fds));
 288
 289         /* signal the test that it can continue */
 290         g_cond_signal(&s->data_cond);
 291         break;
 292
 293     case VHOST_USER_SET_VRING_KICK:
 294     case VHOST_USER_SET_VRING_CALL:
 295         /* consume the fd */
 296         qemu_chr_fe_get_msgfds(chr, &fd, 1);
 297         /*
 298          * This is a non-blocking eventfd.
 299          * The receive function forces it to be blocking,
 300          * so revert it back to non-blocking.
 301          */
 302         qemu_set_nonblock(fd);
 303         break;
 304
 305     case VHOST_USER_SET_LOG_BASE:
 306         if (s->log_fd != -1) {
 307             close(s->log_fd);
 308             s->log_fd = -1;
 309         }
 310         qemu_chr_fe_get_msgfds(chr, &s->log_fd, 1);
 311         msg.flags |= VHOST_USER_REPLY_MASK;
 312         msg.size = 0;
 313         p = (uint8_t *) &msg;
 314         qemu_chr_fe_write_all(chr, p, VHOST_USER_HDR_SIZE);
 315
 316         g_cond_signal(&s->data_cond);
 317         break;
 318
 319     case VHOST_USER_SET_VRING_ENABLE:
 320         if (!msg.payload.state.num) {
 321             s->fds_num = 0;
 322         }
 323         break;
 324
 325     default:
 326         break;
 327     }
 328
 329     g_mutex_unlock(&s->data_mutex);
 330 }
 331
 332 static const char *init_hugepagefs(const char *path)
 333 {
 334     struct statfs fs;
 335     int ret;
 336
 337     if (access(path, R_OK | W_OK | X_OK)) {
 338         g_test_message("access on path (%s): %s\n", path, strerror(errno));
 339         return NULL;
 340     }
 341
 342     do {
 343         ret = statfs(path, &fs);
 344     } while (ret != 0 && errno == EINTR);
 345
 346     if (ret != 0) {
 347         g_test_message("statfs on path (%s): %s\n", path, strerror(errno));
 348         return NULL;
 349     }
 350
 351     if (fs.f_type != HUGETLBFS_MAGIC) {
 352         g_test_message("Warning: path not on HugeTLBFS: %s\n", path);
 353         return NULL;
 354     }
 355
 356     return path;
 357 }
 358
 359 static TestServer *test_server_new(const gchar *name)
 360 {
 361     TestServer *server = g_new0(TestServer, 1);
 362     gchar *chr_path;
 363
 364     server->socket_path = g_strdup_printf("%s/%s.sock", tmpfs, name);
 365
 366     chr_path = g_strdup_printf("unix:%s,server,nowait", server->socket_path);
 367     server->chr_name = g_strdup_printf("chr-%s", name);
 368     server->chr = qemu_chr_new(server->chr_name, chr_path, NULL);
 369     g_free(chr_path);
 370
 371     qemu_chr_add_handlers(server->chr, chr_can_read, chr_read, NULL, server);
 372
 373     g_mutex_init(&server->data_mutex);
 374     g_cond_init(&server->data_cond);
 375
 376     server->log_fd = -1;
 377
 378     return server;
 379 }
 380
 381 #define GET_QEMU_CMD(s)                                                        \
 382     g_strdup_printf(QEMU_CMD, 512, 512, (root), (s)->chr_name,                 \
 383                     (s)->socket_path, (s)->chr_name)
 384
 385 #define GET_QEMU_CMDE(s, mem, extra, ...)                                      \
 386     g_strdup_printf(QEMU_CMD extra, (mem), (mem), (root), (s)->chr_name,       \
 387                     (s)->socket_path, (s)->chr_name, ##__VA_ARGS__)
 388
 389 static void test_server_free(TestServer *server)
 390 {
 391     int i;
 392
 393     qemu_chr_delete(server->chr);
 394
 395     for (i = 0; i < server->fds_num; i++) {
 396         close(server->fds[i]);
 397     }
 398
 399     if (server->log_fd != -1) {
 400         close(server->log_fd);
 401     }
 402
 403     unlink(server->socket_path);
 404     g_free(server->socket_path);
 405
 406
 407     g_free(server->chr_name);
 408     g_free(server);
 409 }
 410
 411 static void wait_for_log_fd(TestServer *s)
 412 {
 413     gint64 end_time;
 414
 415     g_mutex_lock(&s->data_mutex);
 416     end_time = g_get_monotonic_time() + 5 * G_TIME_SPAN_SECOND;
 417     while (s->log_fd == -1) {
 418         if (!g_cond_wait_until(&s->data_cond, &s->data_mutex, end_time)) {
 419             /* timeout has passed */
 420             g_assert(s->log_fd != -1);
 421             break;
 422         }
 423     }
 424
 425     g_mutex_unlock(&s->data_mutex);
 426 }
 427
 428 static void write_guest_mem(TestServer *s, uint32 seed)
 429 {
 430     uint32_t *guest_mem;
 431     int i, j;
 432     size_t size;
 433
 434     wait_for_fds(s);
 435
 436     /* iterate all regions */
 437     for (i = 0; i < s->fds_num; i++) {
 438
 439         /* We'll write only the region statring at 0x0 */
 440         if (s->memory.regions[i].guest_phys_addr != 0x0) {
 441             continue;
 442         }
 443
 444         g_assert_cmpint(s->memory.regions[i].memory_size, >, 1024);
 445
 446         size = s->memory.regions[i].memory_size +
 447             s->memory.regions[i].mmap_offset;
 448
 449         guest_mem = mmap(0, size, PROT_READ | PROT_WRITE,
 450                          MAP_SHARED, s->fds[i], 0);
 451
 452         g_assert(guest_mem != MAP_FAILED);
 453         guest_mem += (s->memory.regions[i].mmap_offset / sizeof(*guest_mem));
 454
 455         for (j = 0; j < 256; j++) {
 456             guest_mem[j] = seed + j;
 457         }
 458
 459         munmap(guest_mem, s->memory.regions[i].memory_size);
 460         break;
 461     }
 462 }
 463
 464 static guint64 get_log_size(TestServer *s)
 465 {
 466     guint64 log_size = 0;
 467     int i;
 468
 469     for (i = 0; i < s->memory.nregions; ++i) {
 470         VhostUserMemoryRegion *reg = &s->memory.regions[i];
 471         guint64 last = range_get_last(reg->guest_phys_addr,
 472                                        reg->memory_size);
 473         log_size = MAX(log_size, last / (8 * VHOST_LOG_PAGE) + 1);
 474     }
 475
 476     return log_size;
 477 }
 478
 479 typedef struct TestMigrateSource {
 480     GSource source;
 481     TestServer *src;
 482     TestServer *dest;
 483 } TestMigrateSource;
 484
 485 static gboolean
 486 test_migrate_source_check(GSource *source)
 487 {
 488     TestMigrateSource *t = (TestMigrateSource *)source;
 489     gboolean overlap = t->src->fds_num > 0 && t->dest->fds_num > 0;
 490
 491     g_assert(!overlap);
 492
 493     return FALSE;
 494 }
 495
 496 GSourceFuncs test_migrate_source_funcs = {
 497     NULL,
 498     test_migrate_source_check,
 499     NULL,
 500     NULL
 501 };
 502
 503 static void test_migrate(void)
 504 {
 505     TestServer *s = test_server_new("src");
 506     TestServer *dest = test_server_new("dest");
 507     const char *uri = "tcp:127.0.0.1:1234";
 508     QTestState *global = global_qtest, *from, *to;
 509     GSource *source;
 510     gchar *cmd;
 511     QDict *rsp;
 512     guint8 *log;
 513     guint64 size;
 514
 515     cmd = GET_QEMU_CMDE(s, 2, "");
 516     from = qtest_start(cmd);
 517     g_free(cmd);
 518
 519     wait_for_fds(s);
 520     size = get_log_size(s);
 521     g_assert_cmpint(size, ==, (2 * 1024 * 1024) / (VHOST_LOG_PAGE * 8));
 522
 523     cmd = GET_QEMU_CMDE(dest, 2, " -incoming %s", uri);
 524     to = qtest_init(cmd);
 525     g_free(cmd);
 526
 527     source = g_source_new(&test_migrate_source_funcs,
 528                           sizeof(TestMigrateSource));
 529     ((TestMigrateSource *)source)->src = s;
 530     ((TestMigrateSource *)source)->dest = dest;
 531     g_source_attach(source, NULL);
 532
 533     /* slow down migration to have time to fiddle with log */
 534     /* TODO: qtest could learn to break on some places */
 535     rsp = qmp("{ 'execute': 'migrate_set_speed',"
 536               "'arguments': { 'value': 10 } }");
 537     g_assert(qdict_haskey(rsp, "return"));
 538     QDECREF(rsp);
 539
 540     cmd = g_strdup_printf("{ 'execute': 'migrate',"
 541                           "'arguments': { 'uri': '%s' } }",
 542                           uri);
 543     rsp = qmp(cmd);
 544     g_free(cmd);
 545     g_assert(qdict_haskey(rsp, "return"));
 546     QDECREF(rsp);
 547
 548     wait_for_log_fd(s);
 549
 550     log = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, s->log_fd, 0);
 551     g_assert(log != MAP_FAILED);
 552
 553     /* modify first page */
 554     write_guest_mem(s, 0x42);
 555     log[0] = 1;
 556     munmap(log, size);
 557
 558     /* speed things up */
 559     rsp = qmp("{ 'execute': 'migrate_set_speed',"
 560               "'arguments': { 'value': 0 } }");
 561     g_assert(qdict_haskey(rsp, "return"));
 562     QDECREF(rsp);
 563
 564     qmp_eventwait("STOP");
 565
 566     global_qtest = to;
 567     qmp_eventwait("RESUME");
 568
 569     read_guest_mem(dest);
 570
 571     g_source_destroy(source);
 572     g_source_unref(source);
 573
 574     qtest_quit(to);
 575     test_server_free(dest);
 576     qtest_quit(from);
 577     test_server_free(s);
 578
 579     global_qtest = global;
 580 }
 581
 582 int main(int argc, char **argv)
 583 {
 584     QTestState *s = NULL;
 585     TestServer *server = NULL;
 586     const char *hugefs;
 587     char *qemu_cmd = NULL;
 588     int ret;
 589     char template[] = "/tmp/vhost-test-XXXXXX";
 590
 591     g_test_init(&argc, &argv, NULL);
 592
 593     module_call_init(MODULE_INIT_QOM);
 594     qemu_add_opts(&qemu_chardev_opts);
 595
 596     tmpfs = mkdtemp(template);
 597     if (!tmpfs) {
 598         g_test_message("mkdtemp on path (%s): %s\n", template, strerror(errno));
 599     }
 600     g_assert(tmpfs);
 601
 602     hugefs = getenv("QTEST_HUGETLBFS_PATH");
 603     if (hugefs) {
 604         root = init_hugepagefs(hugefs);
 605         g_assert(root);
 606     } else {
 607         root = tmpfs;
 608     }
 609
 610     server = test_server_new("test");
 611
 612     /* run the main loop thread so the chardev may operate */
 613     g_thread_new(NULL, thread_function, NULL);
 614
 615     qemu_cmd = GET_QEMU_CMD(server);
 616
 617     s = qtest_start(qemu_cmd);
 618     g_free(qemu_cmd);
 619
 620     qtest_add_data_func("/vhost-user/read-guest-mem", server, read_guest_mem);
 621     qtest_add_func("/vhost-user/migrate", test_migrate);
 622
 623     ret = g_test_run();
 624
 625     if (s) {
 626         qtest_quit(s);
 627     }
 628
 629     /* cleanup */
 630     test_server_free(server);
 631
 632     ret = rmdir(tmpfs);
 633     if (ret != 0) {
 634         g_test_message("unable to rmdir: path (%s): %s\n",
 635                        tmpfs, strerror(errno));
 636     }
 637     g_assert_cmpint(ret, ==, 0);
 638
 639     return ret;
 640 }