util/oslib-posix.c

   1 /*
   2  * os-posix-lib.c
   3  *
   4  * Copyright (c) 2003-2008 Fabrice Bellard
   5  * Copyright (c) 2010 Red Hat, Inc.
   6  *
   7  * QEMU library functions on POSIX which are shared between QEMU and
   8  * the QEMU tools.
   9  *
  10  * Permission is hereby granted, free of charge, to any person obtaining a copy
  11  * of this software and associated documentation files (the "Software"), to deal
  12  * in the Software without restriction, including without limitation the rights
  13  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  14  * copies of the Software, and to permit persons to whom the Software is
  15  * furnished to do so, subject to the following conditions:
  16  *
  17  * The above copyright notice and this permission notice shall be included in
  18  * all copies or substantial portions of the Software.
  19  *
  20  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  21  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  23  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  24  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  25  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  26  * THE SOFTWARE.
  27  */
  28
  29 #include "qemu/osdep.h"
  30 #include <termios.h>
  31
  32 #include <glib/gprintf.h>
  33
  34 #include "sysemu/sysemu.h"
  35 #include "trace.h"
  36 #include "qapi/error.h"
  37 #include "qemu/sockets.h"
  38 #include <libgen.h>
  39 #include <sys/signal.h>
  40 #include "qemu/cutils.h"
  41
  42 #ifdef CONFIG_LINUX
  43 #include <sys/syscall.h>
  44 #endif
  45
  46 #ifdef __FreeBSD__
  47 #include <sys/sysctl.h>
  48 #include <sys/user.h>
  49 #include <libutil.h>
  50 #endif
  51
  52 #include "qemu/mmap-alloc.h"
  53
  54 #ifdef CONFIG_DEBUG_STACK_USAGE
  55 #include "qemu/error-report.h"
  56 #endif
  57
  58 #define MAX_MEM_PREALLOC_THREAD_COUNT 16
  59
  60 struct MemsetThread {
  61     char *addr;
  62     uint64_t numpages;
  63     uint64_t hpagesize;
  64     QemuThread pgthread;
  65     sigjmp_buf env;
  66 };
  67 typedef struct MemsetThread MemsetThread;
  68
  69 static MemsetThread *memset_thread;
  70 static int memset_num_threads;
  71 static bool memset_thread_failed;
  72
  73 int qemu_get_thread_id(void)
  74 {
  75 #if defined(__linux__)
  76     return syscall(SYS_gettid);
  77 #else
  78     return getpid();
  79 #endif
  80 }
  81
  82 int qemu_daemon(int nochdir, int noclose)
  83 {
  84     return daemon(nochdir, noclose);
  85 }
  86
  87 void *qemu_oom_check(void *ptr)
  88 {
  89     if (ptr == NULL) {
  90         fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
  91         abort();
  92     }
  93     return ptr;
  94 }
  95
  96 void *qemu_try_memalign(size_t alignment, size_t size)
  97 {
  98     void *ptr;
  99
 100     if (alignment < sizeof(void*)) {
 101         alignment = sizeof(void*);
 102     }
 103
 104 #if defined(_POSIX_C_SOURCE) && !defined(__sun__)
 105     int ret;
 106     ret = posix_memalign(&ptr, alignment, size);
 107     if (ret != 0) {
 108         errno = ret;
 109         ptr = NULL;
 110     }
 111 #elif defined(CONFIG_BSD)
 112     ptr = valloc(size);
 113 #else
 114     ptr = memalign(alignment, size);
 115 #endif
 116     trace_qemu_memalign(alignment, size, ptr);
 117     return ptr;
 118 }
 119
 120 void *qemu_memalign(size_t alignment, size_t size)
 121 {
 122     return qemu_oom_check(qemu_try_memalign(alignment, size));
 123 }
 124
 125 /* alloc shared memory pages */
 126 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment)
 127 {
 128     size_t align = QEMU_VMALLOC_ALIGN;
 129     void *ptr = qemu_ram_mmap(-1, size, align, false);
 130
 131     if (ptr == MAP_FAILED) {
 132         return NULL;
 133     }
 134
 135     if (alignment) {
 136         *alignment = align;
 137     }
 138
 139     trace_qemu_anon_ram_alloc(size, ptr);
 140     return ptr;
 141 }
 142
 143 void qemu_vfree(void *ptr)
 144 {
 145     trace_qemu_vfree(ptr);
 146     free(ptr);
 147 }
 148
 149 void qemu_anon_ram_free(void *ptr, size_t size)
 150 {
 151     trace_qemu_anon_ram_free(ptr, size);
 152     qemu_ram_munmap(ptr, size);
 153 }
 154
 155 void qemu_set_block(int fd)
 156 {
 157     int f;
 158     f = fcntl(fd, F_GETFL);
 159     fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
 160 }
 161
 162 void qemu_set_nonblock(int fd)
 163 {
 164     int f;
 165     f = fcntl(fd, F_GETFL);
 166     fcntl(fd, F_SETFL, f | O_NONBLOCK);
 167 }
 168
 169 int socket_set_fast_reuse(int fd)
 170 {
 171     int val = 1, ret;
 172
 173     ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
 174                      (const char *)&val, sizeof(val));
 175
 176     assert(ret == 0);
 177
 178     return ret;
 179 }
 180
 181 void qemu_set_cloexec(int fd)
 182 {
 183     int f;
 184     f = fcntl(fd, F_GETFD);
 185     fcntl(fd, F_SETFD, f | FD_CLOEXEC);
 186 }
 187
 188 /*
 189  * Creates a pipe with FD_CLOEXEC set on both file descriptors
 190  */
 191 int qemu_pipe(int pipefd[2])
 192 {
 193     int ret;
 194
 195 #ifdef CONFIG_PIPE2
 196     ret = pipe2(pipefd, O_CLOEXEC);
 197     if (ret != -1 || errno != ENOSYS) {
 198         return ret;
 199     }
 200 #endif
 201     ret = pipe(pipefd);
 202     if (ret == 0) {
 203         qemu_set_cloexec(pipefd[0]);
 204         qemu_set_cloexec(pipefd[1]);
 205     }
 206
 207     return ret;
 208 }
 209
 210 char *
 211 qemu_get_local_state_pathname(const char *relative_pathname)
 212 {
 213     return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR,
 214                            relative_pathname);
 215 }
 216
 217 void qemu_set_tty_echo(int fd, bool echo)
 218 {
 219     struct termios tty;
 220
 221     tcgetattr(fd, &tty);
 222
 223     if (echo) {
 224         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
 225     } else {
 226         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
 227     }
 228
 229     tcsetattr(fd, TCSANOW, &tty);
 230 }
 231
 232 static char exec_dir[PATH_MAX];
 233
 234 void qemu_init_exec_dir(const char *argv0)
 235 {
 236     char *dir;
 237     char *p = NULL;
 238     char buf[PATH_MAX];
 239
 240     assert(!exec_dir[0]);
 241
 242 #if defined(__linux__)
 243     {
 244         int len;
 245         len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
 246         if (len > 0) {
 247             buf[len] = 0;
 248             p = buf;
 249         }
 250     }
 251 #elif defined(__FreeBSD__)
 252     {
 253         static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
 254         size_t len = sizeof(buf) - 1;
 255
 256         *buf = '\0';
 257         if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
 258             *buf) {
 259             buf[sizeof(buf) - 1] = '\0';
 260             p = buf;
 261         }
 262     }
 263 #endif
 264     /* If we don't have any way of figuring out the actual executable
 265        location then try argv[0].  */
 266     if (!p) {
 267         if (!argv0) {
 268             return;
 269         }
 270         p = realpath(argv0, buf);
 271         if (!p) {
 272             return;
 273         }
 274     }
 275     dir = g_path_get_dirname(p);
 276
 277     pstrcpy(exec_dir, sizeof(exec_dir), dir);
 278
 279     g_free(dir);
 280 }
 281
 282 char *qemu_get_exec_dir(void)
 283 {
 284     return g_strdup(exec_dir);
 285 }
 286
 287 static void sigbus_handler(int signal)
 288 {
 289     int i;
 290     if (memset_thread) {
 291         for (i = 0; i < memset_num_threads; i++) {
 292             if (qemu_thread_is_self(&memset_thread[i].pgthread)) {
 293                 siglongjmp(memset_thread[i].env, 1);
 294             }
 295         }
 296     }
 297 }
 298
 299 static void *do_touch_pages(void *arg)
 300 {
 301     MemsetThread *memset_args = (MemsetThread *)arg;
 302     char *addr = memset_args->addr;
 303     uint64_t numpages = memset_args->numpages;
 304     uint64_t hpagesize = memset_args->hpagesize;
 305     sigset_t set, oldset;
 306     int i = 0;
 307
 308     /* unblock SIGBUS */
 309     sigemptyset(&set);
 310     sigaddset(&set, SIGBUS);
 311     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
 312
 313     if (sigsetjmp(memset_args->env, 1)) {
 314         memset_thread_failed = true;
 315     } else {
 316         for (i = 0; i < numpages; i++) {
 317             /*
 318              * Read & write back the same value, so we don't
 319              * corrupt existing user/app data that might be
 320              * stored.
 321              *
 322              * 'volatile' to stop compiler optimizing this away
 323              * to a no-op
 324              *
 325              * TODO: get a better solution from kernel so we
 326              * don't need to write at all so we don't cause
 327              * wear on the storage backing the region...
 328              */
 329             *(volatile char *)addr = *addr;
 330             addr += hpagesize;
 331         }
 332     }
 333     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
 334     return NULL;
 335 }
 336
 337 static inline int get_memset_num_threads(int smp_cpus)
 338 {
 339     long host_procs = sysconf(_SC_NPROCESSORS_ONLN);
 340     int ret = 1;
 341
 342     if (host_procs > 0) {
 343         ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus);
 344     }
 345     /* In case sysconf() fails, we fall back to single threaded */
 346     return ret;
 347 }
 348
 349 static bool touch_all_pages(char *area, size_t hpagesize, size_t numpages,
 350                             int smp_cpus)
 351 {
 352     uint64_t numpages_per_thread, size_per_thread;
 353     char *addr = area;
 354     int i = 0;
 355
 356     memset_thread_failed = false;
 357     memset_num_threads = get_memset_num_threads(smp_cpus);
 358     memset_thread = g_new0(MemsetThread, memset_num_threads);
 359     numpages_per_thread = (numpages / memset_num_threads);
 360     size_per_thread = (hpagesize * numpages_per_thread);
 361     for (i = 0; i < memset_num_threads; i++) {
 362         memset_thread[i].addr = addr;
 363         memset_thread[i].numpages = (i == (memset_num_threads - 1)) ?
 364                                     numpages : numpages_per_thread;
 365         memset_thread[i].hpagesize = hpagesize;
 366         qemu_thread_create(&memset_thread[i].pgthread, "touch_pages",
 367                            do_touch_pages, &memset_thread[i],
 368                            QEMU_THREAD_JOINABLE);
 369         addr += size_per_thread;
 370         numpages -= numpages_per_thread;
 371     }
 372     for (i = 0; i < memset_num_threads; i++) {
 373         qemu_thread_join(&memset_thread[i].pgthread);
 374     }
 375     g_free(memset_thread);
 376     memset_thread = NULL;
 377
 378     return memset_thread_failed;
 379 }
 380
 381 void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus,
 382                      Error **errp)
 383 {
 384     int ret;
 385     struct sigaction act, oldact;
 386     size_t hpagesize = qemu_fd_getpagesize(fd);
 387     size_t numpages = DIV_ROUND_UP(memory, hpagesize);
 388
 389     memset(&act, 0, sizeof(act));
 390     act.sa_handler = &sigbus_handler;
 391     act.sa_flags = 0;
 392
 393     ret = sigaction(SIGBUS, &act, &oldact);
 394     if (ret) {
 395         error_setg_errno(errp, errno,
 396             "os_mem_prealloc: failed to install signal handler");
 397         return;
 398     }
 399
 400     /* touch pages simultaneously */
 401     if (touch_all_pages(area, hpagesize, numpages, smp_cpus)) {
 402         error_setg(errp, "os_mem_prealloc: Insufficient free host memory "
 403             "pages available to allocate guest RAM\n");
 404     }
 405
 406     ret = sigaction(SIGBUS, &oldact, NULL);
 407     if (ret) {
 408         /* Terminate QEMU since it can't recover from error */
 409         perror("os_mem_prealloc: failed to reinstall signal handler");
 410         exit(1);
 411     }
 412 }
 413
 414
 415 static struct termios oldtty;
 416
 417 static void term_exit(void)
 418 {
 419     tcsetattr(0, TCSANOW, &oldtty);
 420 }
 421
 422 static void term_init(void)
 423 {
 424     struct termios tty;
 425
 426     tcgetattr(0, &tty);
 427     oldtty = tty;
 428
 429     tty.c_iflag &= ~(IGNBRK|BRKINT|PARMRK|ISTRIP
 430                           |INLCR|IGNCR|ICRNL|IXON);
 431     tty.c_oflag |= OPOST;
 432     tty.c_lflag &= ~(ECHO|ECHONL|ICANON|IEXTEN);
 433     tty.c_cflag &= ~(CSIZE|PARENB);
 434     tty.c_cflag |= CS8;
 435     tty.c_cc[VMIN] = 1;
 436     tty.c_cc[VTIME] = 0;
 437
 438     tcsetattr(0, TCSANOW, &tty);
 439
 440     atexit(term_exit);
 441 }
 442
 443 int qemu_read_password(char *buf, int buf_size)
 444 {
 445     uint8_t ch;
 446     int i, ret;
 447
 448     printf("password: ");
 449     fflush(stdout);
 450     term_init();
 451     i = 0;
 452     for (;;) {
 453         ret = read(0, &ch, 1);
 454         if (ret == -1) {
 455             if (errno == EAGAIN || errno == EINTR) {
 456                 continue;
 457             } else {
 458                 break;
 459             }
 460         } else if (ret == 0) {
 461             ret = -1;
 462             break;
 463         } else {
 464             if (ch == '\r' ||
 465                 ch == '\n') {
 466                 ret = 0;
 467                 break;
 468             }
 469             if (i < (buf_size - 1)) {
 470                 buf[i++] = ch;
 471             }
 472         }
 473     }
 474     term_exit();
 475     buf[i] = '\0';
 476     printf("\n");
 477     return ret;
 478 }
 479
 480
 481 char *qemu_get_pid_name(pid_t pid)
 482 {
 483     char *name = NULL;
 484
 485 #if defined(__FreeBSD__)
 486     /* BSDs don't have /proc, but they provide a nice substitute */
 487     struct kinfo_proc *proc = kinfo_getproc(pid);
 488
 489     if (proc) {
 490         name = g_strdup(proc->ki_comm);
 491         free(proc);
 492     }
 493 #else
 494     /* Assume a system with reasonable procfs */
 495     char *pid_path;
 496     size_t len;
 497
 498     pid_path = g_strdup_printf("/proc/%d/cmdline", pid);
 499     g_file_get_contents(pid_path, &name, &len, NULL);
 500     g_free(pid_path);
 501 #endif
 502
 503     return name;
 504 }
 505
 506
 507 pid_t qemu_fork(Error **errp)
 508 {
 509     sigset_t oldmask, newmask;
 510     struct sigaction sig_action;
 511     int saved_errno;
 512     pid_t pid;
 513
 514     /*
 515      * Need to block signals now, so that child process can safely
 516      * kill off caller's signal handlers without a race.
 517      */
 518     sigfillset(&newmask);
 519     if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
 520         error_setg_errno(errp, errno,
 521                          "cannot block signals");
 522         return -1;
 523     }
 524
 525     pid = fork();
 526     saved_errno = errno;
 527
 528     if (pid < 0) {
 529         /* attempt to restore signal mask, but ignore failure, to
 530          * avoid obscuring the fork failure */
 531         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 532         error_setg_errno(errp, saved_errno,
 533                          "cannot fork child process");
 534         errno = saved_errno;
 535         return -1;
 536     } else if (pid) {
 537         /* parent process */
 538
 539         /* Restore our original signal mask now that the child is
 540          * safely running. Only documented failures are EFAULT (not
 541          * possible, since we are using just-grabbed mask) or EINVAL
 542          * (not possible, since we are using correct arguments).  */
 543         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 544     } else {
 545         /* child process */
 546         size_t i;
 547
 548         /* Clear out all signal handlers from parent so nothing
 549          * unexpected can happen in our child once we unblock
 550          * signals */
 551         sig_action.sa_handler = SIG_DFL;
 552         sig_action.sa_flags = 0;
 553         sigemptyset(&sig_action.sa_mask);
 554
 555         for (i = 1; i < NSIG; i++) {
 556             /* Only possible errors are EFAULT or EINVAL The former
 557              * won't happen, the latter we expect, so no need to check
 558              * return value */
 559             (void)sigaction(i, &sig_action, NULL);
 560         }
 561
 562         /* Unmask all signals in child, since we've no idea what the
 563          * caller's done with their signal mask and don't want to
 564          * propagate that to children */
 565         sigemptyset(&newmask);
 566         if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
 567             Error *local_err = NULL;
 568             error_setg_errno(&local_err, errno,
 569                              "cannot unblock signals");
 570             error_report_err(local_err);
 571             _exit(1);
 572         }
 573     }
 574     return pid;
 575 }
 576
 577 void *qemu_alloc_stack(size_t *sz)
 578 {
 579     void *ptr, *guardpage;
 580 #ifdef CONFIG_DEBUG_STACK_USAGE
 581     void *ptr2;
 582 #endif
 583     size_t pagesz = getpagesize();
 584 #ifdef _SC_THREAD_STACK_MIN
 585     /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
 586     long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
 587     *sz = MAX(MAX(min_stack_sz, 0), *sz);
 588 #endif
 589     /* adjust stack size to a multiple of the page size */
 590     *sz = ROUND_UP(*sz, pagesz);
 591     /* allocate one extra page for the guard page */
 592     *sz += pagesz;
 593
 594     ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE,
 595                MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
 596     if (ptr == MAP_FAILED) {
 597         abort();
 598     }
 599
 600 #if defined(HOST_IA64)
 601     /* separate register stack */
 602     guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
 603 #elif defined(HOST_HPPA)
 604     /* stack grows up */
 605     guardpage = ptr + *sz - pagesz;
 606 #else
 607     /* stack grows down */
 608     guardpage = ptr;
 609 #endif
 610     if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
 611         abort();
 612     }
 613
 614 #ifdef CONFIG_DEBUG_STACK_USAGE
 615     for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
 616         *(uint32_t *)ptr2 = 0xdeadbeaf;
 617     }
 618 #endif
 619
 620     return ptr;
 621 }
 622
 623 #ifdef CONFIG_DEBUG_STACK_USAGE
 624 static __thread unsigned int max_stack_usage;
 625 #endif
 626
 627 void qemu_free_stack(void *stack, size_t sz)
 628 {
 629 #ifdef CONFIG_DEBUG_STACK_USAGE
 630     unsigned int usage;
 631     void *ptr;
 632
 633     for (ptr = stack + getpagesize(); ptr < stack + sz;
 634          ptr += sizeof(uint32_t)) {
 635         if (*(uint32_t *)ptr != 0xdeadbeaf) {
 636             break;
 637         }
 638     }
 639     usage = sz - (uintptr_t) (ptr - stack);
 640     if (usage > max_stack_usage) {
 641         error_report("thread %d max stack usage increased from %u to %u",
 642                      qemu_get_thread_id(), max_stack_usage, usage);
 643         max_stack_usage = usage;
 644     }
 645 #endif
 646
 647     munmap(stack, sz);
 648 }
 649
 650 void sigaction_invoke(struct sigaction *action,
 651                       struct qemu_signalfd_siginfo *info)
 652 {
 653     siginfo_t si = { 0 };
 654     si.si_signo = info->ssi_signo;
 655     si.si_errno = info->ssi_errno;
 656     si.si_code = info->ssi_code;
 657
 658     /* Convert the minimal set of fields defined by POSIX.
 659      * Positive si_code values are reserved for kernel-generated
 660      * signals, where the valid siginfo fields are determined by
 661      * the signal number.  But according to POSIX, it is unspecified
 662      * whether SI_USER and SI_QUEUE have values less than or equal to
 663      * zero.
 664      */
 665     if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
 666         info->ssi_code <= 0) {
 667         /* SIGTERM, etc.  */
 668         si.si_pid = info->ssi_pid;
 669         si.si_uid = info->ssi_uid;
 670     } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE ||
 671                info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) {
 672         si.si_addr = (void *)(uintptr_t)info->ssi_addr;
 673     } else if (info->ssi_signo == SIGCHLD) {
 674         si.si_pid = info->ssi_pid;
 675         si.si_status = info->ssi_status;
 676         si.si_uid = info->ssi_uid;
 677     }
 678     action->sa_sigaction(info->ssi_signo, &si, NULL);
 679 }