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