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 #ifdef __NetBSD__
  53 #include <sys/sysctl.h>
  54 #endif
  55
  56 #include "qemu/mmap-alloc.h"
  57
  58 #ifdef CONFIG_DEBUG_STACK_USAGE
  59 #include "qemu/error-report.h"
  60 #endif
  61
  62 #define MAX_MEM_PREALLOC_THREAD_COUNT 16
  63
  64 struct MemsetThread {
  65     char *addr;
  66     size_t numpages;
  67     size_t hpagesize;
  68     QemuThread pgthread;
  69     sigjmp_buf env;
  70 };
  71 typedef struct MemsetThread MemsetThread;
  72
  73 static MemsetThread *memset_thread;
  74 static int memset_num_threads;
  75 static bool memset_thread_failed;
  76
  77 int qemu_get_thread_id(void)
  78 {
  79 #if defined(__linux__)
  80     return syscall(SYS_gettid);
  81 #else
  82     return getpid();
  83 #endif
  84 }
  85
  86 int qemu_daemon(int nochdir, int noclose)
  87 {
  88     return daemon(nochdir, noclose);
  89 }
  90
  91 bool qemu_write_pidfile(const char *path, Error **errp)
  92 {
  93     int fd;
  94     char pidstr[32];
  95
  96     while (1) {
  97         struct stat a, b;
  98         struct flock lock = {
  99             .l_type = F_WRLCK,
 100             .l_whence = SEEK_SET,
 101             .l_len = 0,
 102         };
 103
 104         fd = qemu_open(path, O_CREAT | O_WRONLY, S_IRUSR | S_IWUSR);
 105         if (fd == -1) {
 106             error_setg_errno(errp, errno, "Cannot open pid file");
 107             return false;
 108         }
 109
 110         if (fstat(fd, &b) < 0) {
 111             error_setg_errno(errp, errno, "Cannot stat file");
 112             goto fail_close;
 113         }
 114
 115         if (fcntl(fd, F_SETLK, &lock)) {
 116             error_setg_errno(errp, errno, "Cannot lock pid file");
 117             goto fail_close;
 118         }
 119
 120         /*
 121          * Now make sure the path we locked is the same one that now
 122          * exists on the filesystem.
 123          */
 124         if (stat(path, &a) < 0) {
 125             /*
 126              * PID file disappeared, someone else must be racing with
 127              * us, so try again.
 128              */
 129             close(fd);
 130             continue;
 131         }
 132
 133         if (a.st_ino == b.st_ino) {
 134             break;
 135         }
 136
 137         /*
 138          * PID file was recreated, someone else must be racing with
 139          * us, so try again.
 140          */
 141         close(fd);
 142     }
 143
 144     if (ftruncate(fd, 0) < 0) {
 145         error_setg_errno(errp, errno, "Failed to truncate pid file");
 146         goto fail_unlink;
 147     }
 148
 149     snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid());
 150     if (write(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) {
 151         error_setg(errp, "Failed to write pid file");
 152         goto fail_unlink;
 153     }
 154
 155     return true;
 156
 157 fail_unlink:
 158     unlink(path);
 159 fail_close:
 160     close(fd);
 161     return false;
 162 }
 163
 164 void *qemu_oom_check(void *ptr)
 165 {
 166     if (ptr == NULL) {
 167         fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
 168         abort();
 169     }
 170     return ptr;
 171 }
 172
 173 void *qemu_try_memalign(size_t alignment, size_t size)
 174 {
 175     void *ptr;
 176
 177     if (alignment < sizeof(void*)) {
 178         alignment = sizeof(void*);
 179     }
 180
 181 #if defined(CONFIG_POSIX_MEMALIGN)
 182     int ret;
 183     ret = posix_memalign(&ptr, alignment, size);
 184     if (ret != 0) {
 185         errno = ret;
 186         ptr = NULL;
 187     }
 188 #elif defined(CONFIG_BSD)
 189     ptr = valloc(size);
 190 #else
 191     ptr = memalign(alignment, size);
 192 #endif
 193     trace_qemu_memalign(alignment, size, ptr);
 194     return ptr;
 195 }
 196
 197 void *qemu_memalign(size_t alignment, size_t size)
 198 {
 199     return qemu_oom_check(qemu_try_memalign(alignment, size));
 200 }
 201
 202 /* alloc shared memory pages */
 203 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared)
 204 {
 205     size_t align = QEMU_VMALLOC_ALIGN;
 206     void *ptr = qemu_ram_mmap(-1, size, align, shared);
 207
 208     if (ptr == MAP_FAILED) {
 209         return NULL;
 210     }
 211
 212     if (alignment) {
 213         *alignment = align;
 214     }
 215
 216     trace_qemu_anon_ram_alloc(size, ptr);
 217     return ptr;
 218 }
 219
 220 void qemu_vfree(void *ptr)
 221 {
 222     trace_qemu_vfree(ptr);
 223     free(ptr);
 224 }
 225
 226 void qemu_anon_ram_free(void *ptr, size_t size)
 227 {
 228     trace_qemu_anon_ram_free(ptr, size);
 229     qemu_ram_munmap(ptr, size);
 230 }
 231
 232 void qemu_set_block(int fd)
 233 {
 234     int f;
 235     f = fcntl(fd, F_GETFL);
 236     fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
 237 }
 238
 239 void qemu_set_nonblock(int fd)
 240 {
 241     int f;
 242     f = fcntl(fd, F_GETFL);
 243     fcntl(fd, F_SETFL, f | O_NONBLOCK);
 244 }
 245
 246 int socket_set_fast_reuse(int fd)
 247 {
 248     int val = 1, ret;
 249
 250     ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
 251                      (const char *)&val, sizeof(val));
 252
 253     assert(ret == 0);
 254
 255     return ret;
 256 }
 257
 258 void qemu_set_cloexec(int fd)
 259 {
 260     int f;
 261     f = fcntl(fd, F_GETFD);
 262     assert(f != -1);
 263     f = fcntl(fd, F_SETFD, f | FD_CLOEXEC);
 264     assert(f != -1);
 265 }
 266
 267 /*
 268  * Creates a pipe with FD_CLOEXEC set on both file descriptors
 269  */
 270 int qemu_pipe(int pipefd[2])
 271 {
 272     int ret;
 273
 274 #ifdef CONFIG_PIPE2
 275     ret = pipe2(pipefd, O_CLOEXEC);
 276     if (ret != -1 || errno != ENOSYS) {
 277         return ret;
 278     }
 279 #endif
 280     ret = pipe(pipefd);
 281     if (ret == 0) {
 282         qemu_set_cloexec(pipefd[0]);
 283         qemu_set_cloexec(pipefd[1]);
 284     }
 285
 286     return ret;
 287 }
 288
 289 char *
 290 qemu_get_local_state_pathname(const char *relative_pathname)
 291 {
 292     return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR,
 293                            relative_pathname);
 294 }
 295
 296 void qemu_set_tty_echo(int fd, bool echo)
 297 {
 298     struct termios tty;
 299
 300     tcgetattr(fd, &tty);
 301
 302     if (echo) {
 303         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
 304     } else {
 305         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
 306     }
 307
 308     tcsetattr(fd, TCSANOW, &tty);
 309 }
 310
 311 static char exec_dir[PATH_MAX];
 312
 313 void qemu_init_exec_dir(const char *argv0)
 314 {
 315     char *dir;
 316     char *p = NULL;
 317     char buf[PATH_MAX];
 318
 319     assert(!exec_dir[0]);
 320
 321 #if defined(__linux__)
 322     {
 323         int len;
 324         len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
 325         if (len > 0) {
 326             buf[len] = 0;
 327             p = buf;
 328         }
 329     }
 330 #elif defined(__FreeBSD__) \
 331       || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME))
 332     {
 333 #if defined(__FreeBSD__)
 334         static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
 335 #else
 336         static int mib[4] = {CTL_KERN, KERN_PROC_ARGS, -1, KERN_PROC_PATHNAME};
 337 #endif
 338         size_t len = sizeof(buf) - 1;
 339
 340         *buf = '\0';
 341         if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
 342             *buf) {
 343             buf[sizeof(buf) - 1] = '\0';
 344             p = buf;
 345         }
 346     }
 347 #endif
 348     /* If we don't have any way of figuring out the actual executable
 349        location then try argv[0].  */
 350     if (!p) {
 351         if (!argv0) {
 352             return;
 353         }
 354         p = realpath(argv0, buf);
 355         if (!p) {
 356             return;
 357         }
 358     }
 359     dir = g_path_get_dirname(p);
 360
 361     pstrcpy(exec_dir, sizeof(exec_dir), dir);
 362
 363     g_free(dir);
 364 }
 365
 366 char *qemu_get_exec_dir(void)
 367 {
 368     return g_strdup(exec_dir);
 369 }
 370
 371 static void sigbus_handler(int signal)
 372 {
 373     int i;
 374     if (memset_thread) {
 375         for (i = 0; i < memset_num_threads; i++) {
 376             if (qemu_thread_is_self(&memset_thread[i].pgthread)) {
 377                 siglongjmp(memset_thread[i].env, 1);
 378             }
 379         }
 380     }
 381 }
 382
 383 static void *do_touch_pages(void *arg)
 384 {
 385     MemsetThread *memset_args = (MemsetThread *)arg;
 386     sigset_t set, oldset;
 387
 388     /* unblock SIGBUS */
 389     sigemptyset(&set);
 390     sigaddset(&set, SIGBUS);
 391     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
 392
 393     if (sigsetjmp(memset_args->env, 1)) {
 394         memset_thread_failed = true;
 395     } else {
 396         char *addr = memset_args->addr;
 397         size_t numpages = memset_args->numpages;
 398         size_t hpagesize = memset_args->hpagesize;
 399         size_t i;
 400         for (i = 0; i < numpages; i++) {
 401             /*
 402              * Read & write back the same value, so we don't
 403              * corrupt existing user/app data that might be
 404              * stored.
 405              *
 406              * 'volatile' to stop compiler optimizing this away
 407              * to a no-op
 408              *
 409              * TODO: get a better solution from kernel so we
 410              * don't need to write at all so we don't cause
 411              * wear on the storage backing the region...
 412              */
 413             *(volatile char *)addr = *addr;
 414             addr += hpagesize;
 415         }
 416     }
 417     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
 418     return NULL;
 419 }
 420
 421 static inline int get_memset_num_threads(int smp_cpus)
 422 {
 423     long host_procs = sysconf(_SC_NPROCESSORS_ONLN);
 424     int ret = 1;
 425
 426     if (host_procs > 0) {
 427         ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus);
 428     }
 429     /* In case sysconf() fails, we fall back to single threaded */
 430     return ret;
 431 }
 432
 433 static bool touch_all_pages(char *area, size_t hpagesize, size_t numpages,
 434                             int smp_cpus)
 435 {
 436     size_t numpages_per_thread;
 437     size_t size_per_thread;
 438     char *addr = area;
 439     int i = 0;
 440
 441     memset_thread_failed = false;
 442     memset_num_threads = get_memset_num_threads(smp_cpus);
 443     memset_thread = g_new0(MemsetThread, memset_num_threads);
 444     numpages_per_thread = (numpages / memset_num_threads);
 445     size_per_thread = (hpagesize * numpages_per_thread);
 446     for (i = 0; i < memset_num_threads; i++) {
 447         memset_thread[i].addr = addr;
 448         memset_thread[i].numpages = (i == (memset_num_threads - 1)) ?
 449                                     numpages : numpages_per_thread;
 450         memset_thread[i].hpagesize = hpagesize;
 451         qemu_thread_create(&memset_thread[i].pgthread, "touch_pages",
 452                            do_touch_pages, &memset_thread[i],
 453                            QEMU_THREAD_JOINABLE);
 454         addr += size_per_thread;
 455         numpages -= numpages_per_thread;
 456     }
 457     for (i = 0; i < memset_num_threads; i++) {
 458         qemu_thread_join(&memset_thread[i].pgthread);
 459     }
 460     g_free(memset_thread);
 461     memset_thread = NULL;
 462
 463     return memset_thread_failed;
 464 }
 465
 466 void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus,
 467                      Error **errp)
 468 {
 469     int ret;
 470     struct sigaction act, oldact;
 471     size_t hpagesize = qemu_fd_getpagesize(fd);
 472     size_t numpages = DIV_ROUND_UP(memory, hpagesize);
 473
 474     memset(&act, 0, sizeof(act));
 475     act.sa_handler = &sigbus_handler;
 476     act.sa_flags = 0;
 477
 478     ret = sigaction(SIGBUS, &act, &oldact);
 479     if (ret) {
 480         error_setg_errno(errp, errno,
 481             "os_mem_prealloc: failed to install signal handler");
 482         return;
 483     }
 484
 485     /* touch pages simultaneously */
 486     if (touch_all_pages(area, hpagesize, numpages, smp_cpus)) {
 487         error_setg(errp, "os_mem_prealloc: Insufficient free host memory "
 488             "pages available to allocate guest RAM");
 489     }
 490
 491     ret = sigaction(SIGBUS, &oldact, NULL);
 492     if (ret) {
 493         /* Terminate QEMU since it can't recover from error */
 494         perror("os_mem_prealloc: failed to reinstall signal handler");
 495         exit(1);
 496     }
 497 }
 498
 499
 500 char *qemu_get_pid_name(pid_t pid)
 501 {
 502     char *name = NULL;
 503
 504 #if defined(__FreeBSD__)
 505     /* BSDs don't have /proc, but they provide a nice substitute */
 506     struct kinfo_proc *proc = kinfo_getproc(pid);
 507
 508     if (proc) {
 509         name = g_strdup(proc->ki_comm);
 510         free(proc);
 511     }
 512 #else
 513     /* Assume a system with reasonable procfs */
 514     char *pid_path;
 515     size_t len;
 516
 517     pid_path = g_strdup_printf("/proc/%d/cmdline", pid);
 518     g_file_get_contents(pid_path, &name, &len, NULL);
 519     g_free(pid_path);
 520 #endif
 521
 522     return name;
 523 }
 524
 525
 526 pid_t qemu_fork(Error **errp)
 527 {
 528     sigset_t oldmask, newmask;
 529     struct sigaction sig_action;
 530     int saved_errno;
 531     pid_t pid;
 532
 533     /*
 534      * Need to block signals now, so that child process can safely
 535      * kill off caller's signal handlers without a race.
 536      */
 537     sigfillset(&newmask);
 538     if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
 539         error_setg_errno(errp, errno,
 540                          "cannot block signals");
 541         return -1;
 542     }
 543
 544     pid = fork();
 545     saved_errno = errno;
 546
 547     if (pid < 0) {
 548         /* attempt to restore signal mask, but ignore failure, to
 549          * avoid obscuring the fork failure */
 550         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 551         error_setg_errno(errp, saved_errno,
 552                          "cannot fork child process");
 553         errno = saved_errno;
 554         return -1;
 555     } else if (pid) {
 556         /* parent process */
 557
 558         /* Restore our original signal mask now that the child is
 559          * safely running. Only documented failures are EFAULT (not
 560          * possible, since we are using just-grabbed mask) or EINVAL
 561          * (not possible, since we are using correct arguments).  */
 562         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 563     } else {
 564         /* child process */
 565         size_t i;
 566
 567         /* Clear out all signal handlers from parent so nothing
 568          * unexpected can happen in our child once we unblock
 569          * signals */
 570         sig_action.sa_handler = SIG_DFL;
 571         sig_action.sa_flags = 0;
 572         sigemptyset(&sig_action.sa_mask);
 573
 574         for (i = 1; i < NSIG; i++) {
 575             /* Only possible errors are EFAULT or EINVAL The former
 576              * won't happen, the latter we expect, so no need to check
 577              * return value */
 578             (void)sigaction(i, &sig_action, NULL);
 579         }
 580
 581         /* Unmask all signals in child, since we've no idea what the
 582          * caller's done with their signal mask and don't want to
 583          * propagate that to children */
 584         sigemptyset(&newmask);
 585         if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
 586             Error *local_err = NULL;
 587             error_setg_errno(&local_err, errno,
 588                              "cannot unblock signals");
 589             error_report_err(local_err);
 590             _exit(1);
 591         }
 592     }
 593     return pid;
 594 }
 595
 596 void *qemu_alloc_stack(size_t *sz)
 597 {
 598     void *ptr, *guardpage;
 599 #ifdef CONFIG_DEBUG_STACK_USAGE
 600     void *ptr2;
 601 #endif
 602     size_t pagesz = getpagesize();
 603 #ifdef _SC_THREAD_STACK_MIN
 604     /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
 605     long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
 606     *sz = MAX(MAX(min_stack_sz, 0), *sz);
 607 #endif
 608     /* adjust stack size to a multiple of the page size */
 609     *sz = ROUND_UP(*sz, pagesz);
 610     /* allocate one extra page for the guard page */
 611     *sz += pagesz;
 612
 613     ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE,
 614                MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
 615     if (ptr == MAP_FAILED) {
 616         perror("failed to allocate memory for stack");
 617         abort();
 618     }
 619
 620 #if defined(HOST_IA64)
 621     /* separate register stack */
 622     guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
 623 #elif defined(HOST_HPPA)
 624     /* stack grows up */
 625     guardpage = ptr + *sz - pagesz;
 626 #else
 627     /* stack grows down */
 628     guardpage = ptr;
 629 #endif
 630     if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
 631         perror("failed to set up stack guard page");
 632         abort();
 633     }
 634
 635 #ifdef CONFIG_DEBUG_STACK_USAGE
 636     for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
 637         *(uint32_t *)ptr2 = 0xdeadbeaf;
 638     }
 639 #endif
 640
 641     return ptr;
 642 }
 643
 644 #ifdef CONFIG_DEBUG_STACK_USAGE
 645 static __thread unsigned int max_stack_usage;
 646 #endif
 647
 648 void qemu_free_stack(void *stack, size_t sz)
 649 {
 650 #ifdef CONFIG_DEBUG_STACK_USAGE
 651     unsigned int usage;
 652     void *ptr;
 653
 654     for (ptr = stack + getpagesize(); ptr < stack + sz;
 655          ptr += sizeof(uint32_t)) {
 656         if (*(uint32_t *)ptr != 0xdeadbeaf) {
 657             break;
 658         }
 659     }
 660     usage = sz - (uintptr_t) (ptr - stack);
 661     if (usage > max_stack_usage) {
 662         error_report("thread %d max stack usage increased from %u to %u",
 663                      qemu_get_thread_id(), max_stack_usage, usage);
 664         max_stack_usage = usage;
 665     }
 666 #endif
 667
 668     munmap(stack, sz);
 669 }
 670
 671 void sigaction_invoke(struct sigaction *action,
 672                       struct qemu_signalfd_siginfo *info)
 673 {
 674     siginfo_t si = {};
 675     si.si_signo = info->ssi_signo;
 676     si.si_errno = info->ssi_errno;
 677     si.si_code = info->ssi_code;
 678
 679     /* Convert the minimal set of fields defined by POSIX.
 680      * Positive si_code values are reserved for kernel-generated
 681      * signals, where the valid siginfo fields are determined by
 682      * the signal number.  But according to POSIX, it is unspecified
 683      * whether SI_USER and SI_QUEUE have values less than or equal to
 684      * zero.
 685      */
 686     if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
 687         info->ssi_code <= 0) {
 688         /* SIGTERM, etc.  */
 689         si.si_pid = info->ssi_pid;
 690         si.si_uid = info->ssi_uid;
 691     } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE ||
 692                info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) {
 693         si.si_addr = (void *)(uintptr_t)info->ssi_addr;
 694     } else if (info->ssi_signo == SIGCHLD) {
 695         si.si_pid = info->ssi_pid;
 696         si.si_status = info->ssi_status;
 697         si.si_uid = info->ssi_uid;
 698     }
 699     action->sa_sigaction(info->ssi_signo, &si, NULL);
 700 }