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     assert(f != -1);
 186     f = fcntl(fd, F_SETFD, f | FD_CLOEXEC);
 187     assert(f != -1);
 188 }
 189
 190 /*
 191  * Creates a pipe with FD_CLOEXEC set on both file descriptors
 192  */
 193 int qemu_pipe(int pipefd[2])
 194 {
 195     int ret;
 196
 197 #ifdef CONFIG_PIPE2
 198     ret = pipe2(pipefd, O_CLOEXEC);
 199     if (ret != -1 || errno != ENOSYS) {
 200         return ret;
 201     }
 202 #endif
 203     ret = pipe(pipefd);
 204     if (ret == 0) {
 205         qemu_set_cloexec(pipefd[0]);
 206         qemu_set_cloexec(pipefd[1]);
 207     }
 208
 209     return ret;
 210 }
 211
 212 char *
 213 qemu_get_local_state_pathname(const char *relative_pathname)
 214 {
 215     return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR,
 216                            relative_pathname);
 217 }
 218
 219 void qemu_set_tty_echo(int fd, bool echo)
 220 {
 221     struct termios tty;
 222
 223     tcgetattr(fd, &tty);
 224
 225     if (echo) {
 226         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
 227     } else {
 228         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
 229     }
 230
 231     tcsetattr(fd, TCSANOW, &tty);
 232 }
 233
 234 static char exec_dir[PATH_MAX];
 235
 236 void qemu_init_exec_dir(const char *argv0)
 237 {
 238     char *dir;
 239     char *p = NULL;
 240     char buf[PATH_MAX];
 241
 242     assert(!exec_dir[0]);
 243
 244 #if defined(__linux__)
 245     {
 246         int len;
 247         len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
 248         if (len > 0) {
 249             buf[len] = 0;
 250             p = buf;
 251         }
 252     }
 253 #elif defined(__FreeBSD__)
 254     {
 255         static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
 256         size_t len = sizeof(buf) - 1;
 257
 258         *buf = '\0';
 259         if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
 260             *buf) {
 261             buf[sizeof(buf) - 1] = '\0';
 262             p = buf;
 263         }
 264     }
 265 #endif
 266     /* If we don't have any way of figuring out the actual executable
 267        location then try argv[0].  */
 268     if (!p) {
 269         if (!argv0) {
 270             return;
 271         }
 272         p = realpath(argv0, buf);
 273         if (!p) {
 274             return;
 275         }
 276     }
 277     dir = g_path_get_dirname(p);
 278
 279     pstrcpy(exec_dir, sizeof(exec_dir), dir);
 280
 281     g_free(dir);
 282 }
 283
 284 char *qemu_get_exec_dir(void)
 285 {
 286     return g_strdup(exec_dir);
 287 }
 288
 289 static void sigbus_handler(int signal)
 290 {
 291     int i;
 292     if (memset_thread) {
 293         for (i = 0; i < memset_num_threads; i++) {
 294             if (qemu_thread_is_self(&memset_thread[i].pgthread)) {
 295                 siglongjmp(memset_thread[i].env, 1);
 296             }
 297         }
 298     }
 299 }
 300
 301 static void *do_touch_pages(void *arg)
 302 {
 303     MemsetThread *memset_args = (MemsetThread *)arg;
 304     char *addr = memset_args->addr;
 305     uint64_t numpages = memset_args->numpages;
 306     uint64_t hpagesize = memset_args->hpagesize;
 307     sigset_t set, oldset;
 308     int i = 0;
 309
 310     /* unblock SIGBUS */
 311     sigemptyset(&set);
 312     sigaddset(&set, SIGBUS);
 313     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
 314
 315     if (sigsetjmp(memset_args->env, 1)) {
 316         memset_thread_failed = true;
 317     } else {
 318         for (i = 0; i < numpages; i++) {
 319             /*
 320              * Read & write back the same value, so we don't
 321              * corrupt existing user/app data that might be
 322              * stored.
 323              *
 324              * 'volatile' to stop compiler optimizing this away
 325              * to a no-op
 326              *
 327              * TODO: get a better solution from kernel so we
 328              * don't need to write at all so we don't cause
 329              * wear on the storage backing the region...
 330              */
 331             *(volatile char *)addr = *addr;
 332             addr += hpagesize;
 333         }
 334     }
 335     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
 336     return NULL;
 337 }
 338
 339 static inline int get_memset_num_threads(int smp_cpus)
 340 {
 341     long host_procs = sysconf(_SC_NPROCESSORS_ONLN);
 342     int ret = 1;
 343
 344     if (host_procs > 0) {
 345         ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus);
 346     }
 347     /* In case sysconf() fails, we fall back to single threaded */
 348     return ret;
 349 }
 350
 351 static bool touch_all_pages(char *area, size_t hpagesize, size_t numpages,
 352                             int smp_cpus)
 353 {
 354     uint64_t numpages_per_thread, size_per_thread;
 355     char *addr = area;
 356     int i = 0;
 357
 358     memset_thread_failed = false;
 359     memset_num_threads = get_memset_num_threads(smp_cpus);
 360     memset_thread = g_new0(MemsetThread, memset_num_threads);
 361     numpages_per_thread = (numpages / memset_num_threads);
 362     size_per_thread = (hpagesize * numpages_per_thread);
 363     for (i = 0; i < memset_num_threads; i++) {
 364         memset_thread[i].addr = addr;
 365         memset_thread[i].numpages = (i == (memset_num_threads - 1)) ?
 366                                     numpages : numpages_per_thread;
 367         memset_thread[i].hpagesize = hpagesize;
 368         qemu_thread_create(&memset_thread[i].pgthread, "touch_pages",
 369                            do_touch_pages, &memset_thread[i],
 370                            QEMU_THREAD_JOINABLE);
 371         addr += size_per_thread;
 372         numpages -= numpages_per_thread;
 373     }
 374     for (i = 0; i < memset_num_threads; i++) {
 375         qemu_thread_join(&memset_thread[i].pgthread);
 376     }
 377     g_free(memset_thread);
 378     memset_thread = NULL;
 379
 380     return memset_thread_failed;
 381 }
 382
 383 void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus,
 384                      Error **errp)
 385 {
 386     int ret;
 387     struct sigaction act, oldact;
 388     size_t hpagesize = qemu_fd_getpagesize(fd);
 389     size_t numpages = DIV_ROUND_UP(memory, hpagesize);
 390
 391     memset(&act, 0, sizeof(act));
 392     act.sa_handler = &sigbus_handler;
 393     act.sa_flags = 0;
 394
 395     ret = sigaction(SIGBUS, &act, &oldact);
 396     if (ret) {
 397         error_setg_errno(errp, errno,
 398             "os_mem_prealloc: failed to install signal handler");
 399         return;
 400     }
 401
 402     /* touch pages simultaneously */
 403     if (touch_all_pages(area, hpagesize, numpages, smp_cpus)) {
 404         error_setg(errp, "os_mem_prealloc: Insufficient free host memory "
 405             "pages available to allocate guest RAM");
 406     }
 407
 408     ret = sigaction(SIGBUS, &oldact, NULL);
 409     if (ret) {
 410         /* Terminate QEMU since it can't recover from error */
 411         perror("os_mem_prealloc: failed to reinstall signal handler");
 412         exit(1);
 413     }
 414 }
 415
 416
 417 char *qemu_get_pid_name(pid_t pid)
 418 {
 419     char *name = NULL;
 420
 421 #if defined(__FreeBSD__)
 422     /* BSDs don't have /proc, but they provide a nice substitute */
 423     struct kinfo_proc *proc = kinfo_getproc(pid);
 424
 425     if (proc) {
 426         name = g_strdup(proc->ki_comm);
 427         free(proc);
 428     }
 429 #else
 430     /* Assume a system with reasonable procfs */
 431     char *pid_path;
 432     size_t len;
 433
 434     pid_path = g_strdup_printf("/proc/%d/cmdline", pid);
 435     g_file_get_contents(pid_path, &name, &len, NULL);
 436     g_free(pid_path);
 437 #endif
 438
 439     return name;
 440 }
 441
 442
 443 pid_t qemu_fork(Error **errp)
 444 {
 445     sigset_t oldmask, newmask;
 446     struct sigaction sig_action;
 447     int saved_errno;
 448     pid_t pid;
 449
 450     /*
 451      * Need to block signals now, so that child process can safely
 452      * kill off caller's signal handlers without a race.
 453      */
 454     sigfillset(&newmask);
 455     if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
 456         error_setg_errno(errp, errno,
 457                          "cannot block signals");
 458         return -1;
 459     }
 460
 461     pid = fork();
 462     saved_errno = errno;
 463
 464     if (pid < 0) {
 465         /* attempt to restore signal mask, but ignore failure, to
 466          * avoid obscuring the fork failure */
 467         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 468         error_setg_errno(errp, saved_errno,
 469                          "cannot fork child process");
 470         errno = saved_errno;
 471         return -1;
 472     } else if (pid) {
 473         /* parent process */
 474
 475         /* Restore our original signal mask now that the child is
 476          * safely running. Only documented failures are EFAULT (not
 477          * possible, since we are using just-grabbed mask) or EINVAL
 478          * (not possible, since we are using correct arguments).  */
 479         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 480     } else {
 481         /* child process */
 482         size_t i;
 483
 484         /* Clear out all signal handlers from parent so nothing
 485          * unexpected can happen in our child once we unblock
 486          * signals */
 487         sig_action.sa_handler = SIG_DFL;
 488         sig_action.sa_flags = 0;
 489         sigemptyset(&sig_action.sa_mask);
 490
 491         for (i = 1; i < NSIG; i++) {
 492             /* Only possible errors are EFAULT or EINVAL The former
 493              * won't happen, the latter we expect, so no need to check
 494              * return value */
 495             (void)sigaction(i, &sig_action, NULL);
 496         }
 497
 498         /* Unmask all signals in child, since we've no idea what the
 499          * caller's done with their signal mask and don't want to
 500          * propagate that to children */
 501         sigemptyset(&newmask);
 502         if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
 503             Error *local_err = NULL;
 504             error_setg_errno(&local_err, errno,
 505                              "cannot unblock signals");
 506             error_report_err(local_err);
 507             _exit(1);
 508         }
 509     }
 510     return pid;
 511 }
 512
 513 void *qemu_alloc_stack(size_t *sz)
 514 {
 515     void *ptr, *guardpage;
 516 #ifdef CONFIG_DEBUG_STACK_USAGE
 517     void *ptr2;
 518 #endif
 519     size_t pagesz = getpagesize();
 520 #ifdef _SC_THREAD_STACK_MIN
 521     /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
 522     long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
 523     *sz = MAX(MAX(min_stack_sz, 0), *sz);
 524 #endif
 525     /* adjust stack size to a multiple of the page size */
 526     *sz = ROUND_UP(*sz, pagesz);
 527     /* allocate one extra page for the guard page */
 528     *sz += pagesz;
 529
 530     ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE,
 531                MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
 532     if (ptr == MAP_FAILED) {
 533         perror("failed to allocate memory for stack");
 534         abort();
 535     }
 536
 537 #if defined(HOST_IA64)
 538     /* separate register stack */
 539     guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
 540 #elif defined(HOST_HPPA)
 541     /* stack grows up */
 542     guardpage = ptr + *sz - pagesz;
 543 #else
 544     /* stack grows down */
 545     guardpage = ptr;
 546 #endif
 547     if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
 548         perror("failed to set up stack guard page");
 549         abort();
 550     }
 551
 552 #ifdef CONFIG_DEBUG_STACK_USAGE
 553     for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
 554         *(uint32_t *)ptr2 = 0xdeadbeaf;
 555     }
 556 #endif
 557
 558     return ptr;
 559 }
 560
 561 #ifdef CONFIG_DEBUG_STACK_USAGE
 562 static __thread unsigned int max_stack_usage;
 563 #endif
 564
 565 void qemu_free_stack(void *stack, size_t sz)
 566 {
 567 #ifdef CONFIG_DEBUG_STACK_USAGE
 568     unsigned int usage;
 569     void *ptr;
 570
 571     for (ptr = stack + getpagesize(); ptr < stack + sz;
 572          ptr += sizeof(uint32_t)) {
 573         if (*(uint32_t *)ptr != 0xdeadbeaf) {
 574             break;
 575         }
 576     }
 577     usage = sz - (uintptr_t) (ptr - stack);
 578     if (usage > max_stack_usage) {
 579         error_report("thread %d max stack usage increased from %u to %u",
 580                      qemu_get_thread_id(), max_stack_usage, usage);
 581         max_stack_usage = usage;
 582     }
 583 #endif
 584
 585     munmap(stack, sz);
 586 }
 587
 588 void sigaction_invoke(struct sigaction *action,
 589                       struct qemu_signalfd_siginfo *info)
 590 {
 591     siginfo_t si = {};
 592     si.si_signo = info->ssi_signo;
 593     si.si_errno = info->ssi_errno;
 594     si.si_code = info->ssi_code;
 595
 596     /* Convert the minimal set of fields defined by POSIX.
 597      * Positive si_code values are reserved for kernel-generated
 598      * signals, where the valid siginfo fields are determined by
 599      * the signal number.  But according to POSIX, it is unspecified
 600      * whether SI_USER and SI_QUEUE have values less than or equal to
 601      * zero.
 602      */
 603     if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
 604         info->ssi_code <= 0) {
 605         /* SIGTERM, etc.  */
 606         si.si_pid = info->ssi_pid;
 607         si.si_uid = info->ssi_uid;
 608     } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE ||
 609                info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) {
 610         si.si_addr = (void *)(uintptr_t)info->ssi_addr;
 611     } else if (info->ssi_signo == SIGCHLD) {
 612         si.si_pid = info->ssi_pid;
 613         si.si_status = info->ssi_status;
 614         si.si_uid = info->ssi_uid;
 615     }
 616     action->sa_sigaction(info->ssi_signo, &si, NULL);
 617 }