1 /*
   2  * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "jvm.h"
  26 #include "logging/log.hpp"
  27 #include "memory/allocation.inline.hpp"
  28 #include "os_posix.inline.hpp"
  29 #include "utilities/globalDefinitions.hpp"
  30 #include "runtime/frame.inline.hpp"
  31 #include "runtime/interfaceSupport.inline.hpp"
  32 #include "services/memTracker.hpp"
  33 #include "utilities/align.hpp"
  34 #include "utilities/events.hpp"
  35 #include "utilities/formatBuffer.hpp"
  36 #include "utilities/macros.hpp"
  37 #include "utilities/vmError.hpp"
  38 
  39 #include <dirent.h>
  40 #include <dlfcn.h>
  41 #include <grp.h>
  42 #include <pwd.h>
  43 #include <pthread.h>
  44 #include <signal.h>
  45 #include <sys/mman.h>
  46 #include <sys/resource.h>
  47 #include <sys/utsname.h>
  48 #include <time.h>
  49 #include <unistd.h>
  50 
  51 // Todo: provide a os::get_max_process_id() or similar. Number of processes
  52 // may have been configured, can be read more accurately from proc fs etc.
  53 #ifndef MAX_PID
  54 #define MAX_PID INT_MAX
  55 #endif
  56 #define IS_VALID_PID(p) (p > 0 && p < MAX_PID)
  57 
  58 #define ROOT_UID 0
  59 
  60 #ifndef MAP_ANONYMOUS
  61   #define MAP_ANONYMOUS MAP_ANON
  62 #endif
  63 
  64 #define check_with_errno(check_type, cond, msg)                             \
  65   do {                                                                      \
  66     int err = errno;                                                        \
  67     check_type(cond, "%s; error='%s' (errno=%s)", msg, os::strerror(err),   \
  68                os::errno_name(err));                                        \
  69 } while (false)
  70 
  71 #define assert_with_errno(cond, msg)    check_with_errno(assert, cond, msg)
  72 #define guarantee_with_errno(cond, msg) check_with_errno(guarantee, cond, msg)
  73 
  74 // Check core dump limit and report possible place where core can be found
  75 void os::check_dump_limit(char* buffer, size_t bufferSize) {
  76   if (!FLAG_IS_DEFAULT(CreateCoredumpOnCrash) && !CreateCoredumpOnCrash) {
  77     jio_snprintf(buffer, bufferSize, "CreateCoredumpOnCrash is disabled from command line");
  78     VMError::record_coredump_status(buffer, false);
  79     return;
  80   }
  81 
  82   int n;
  83   struct rlimit rlim;
  84   bool success;
  85 
  86   char core_path[PATH_MAX];
  87   n = get_core_path(core_path, PATH_MAX);
  88 
  89   if (n <= 0) {
  90     jio_snprintf(buffer, bufferSize, "core.%d (may not exist)", current_process_id());
  91     success = true;
  92 #ifdef LINUX
  93   } else if (core_path[0] == '"') { // redirect to user process
  94     jio_snprintf(buffer, bufferSize, "Core dumps may be processed with %s", core_path);
  95     success = true;
  96 #endif
  97   } else if (getrlimit(RLIMIT_CORE, &rlim) != 0) {
  98     jio_snprintf(buffer, bufferSize, "%s (may not exist)", core_path);
  99     success = true;
 100   } else {
 101     switch(rlim.rlim_cur) {
 102       case RLIM_INFINITY:
 103         jio_snprintf(buffer, bufferSize, "%s", core_path);
 104         success = true;
 105         break;
 106       case 0:
 107         jio_snprintf(buffer, bufferSize, "Core dumps have been disabled. To enable core dumping, try \"ulimit -c unlimited\" before starting Java again");
 108         success = false;
 109         break;
 110       default:
 111         jio_snprintf(buffer, bufferSize, "%s (max size " UINT64_FORMAT " kB). To ensure a full core dump, try \"ulimit -c unlimited\" before starting Java again", core_path, uint64_t(rlim.rlim_cur) / 1024);
 112         success = true;
 113         break;
 114     }
 115   }
 116 
 117   VMError::record_coredump_status(buffer, success);
 118 }
 119 
 120 int os::get_native_stack(address* stack, int frames, int toSkip) {
 121   int frame_idx = 0;
 122   int num_of_frames;  // number of frames captured
 123   frame fr = os::current_frame();
 124   while (fr.pc() && frame_idx < frames) {
 125     if (toSkip > 0) {
 126       toSkip --;
 127     } else {
 128       stack[frame_idx ++] = fr.pc();
 129     }
 130     if (fr.fp() == NULL || fr.cb() != NULL ||
 131         fr.sender_pc() == NULL || os::is_first_C_frame(&fr)) break;
 132 
 133     if (fr.sender_pc() && !os::is_first_C_frame(&fr)) {
 134       fr = os::get_sender_for_C_frame(&fr);
 135     } else {
 136       break;
 137     }
 138   }
 139   num_of_frames = frame_idx;
 140   for (; frame_idx < frames; frame_idx ++) {
 141     stack[frame_idx] = NULL;
 142   }
 143 
 144   return num_of_frames;
 145 }
 146 
 147 
 148 bool os::unsetenv(const char* name) {
 149   assert(name != NULL, "Null pointer");
 150   return (::unsetenv(name) == 0);
 151 }
 152 
 153 int os::get_last_error() {
 154   return errno;
 155 }
 156 
 157 size_t os::lasterror(char *buf, size_t len) {
 158   if (errno == 0)  return 0;
 159 
 160   const char *s = os::strerror(errno);
 161   size_t n = ::strlen(s);
 162   if (n >= len) {
 163     n = len - 1;
 164   }
 165   ::strncpy(buf, s, n);
 166   buf[n] = '\0';
 167   return n;
 168 }
 169 
 170 void os::wait_for_keypress_at_exit(void) {
 171   // don't do anything on posix platforms
 172   return;
 173 }
 174 
 175 int os::create_file_for_heap(const char* dir) {
 176 
 177   const char name_template[] = "/jvmheap.XXXXXX";
 178 
 179   size_t fullname_len = strlen(dir) + strlen(name_template);
 180   char *fullname = (char*)os::malloc(fullname_len + 1, mtInternal);
 181   if (fullname == NULL) {
 182     vm_exit_during_initialization(err_msg("Malloc failed during creation of backing file for heap (%s)", os::strerror(errno)));
 183     return -1;
 184   }
 185   int n = snprintf(fullname, fullname_len + 1, "%s%s", dir, name_template);
 186   assert((size_t)n == fullname_len, "Unexpected number of characters in string");
 187 
 188   os::native_path(fullname);
 189 
 190   // set the file creation mask.
 191   mode_t file_mode = S_IRUSR | S_IWUSR;
 192 
 193   // create a new file.
 194   int fd = mkstemp(fullname);
 195 
 196   if (fd < 0) {
 197     warning("Could not create file for heap with template %s", fullname);
 198     os::free(fullname);
 199     return -1;
 200   }
 201 
 202   // delete the name from the filesystem. When 'fd' is closed, the file (and space) will be deleted.
 203   int ret = unlink(fullname);
 204   assert_with_errno(ret == 0, "unlink returned error");
 205 
 206   os::free(fullname);
 207   return fd;
 208 }
 209 
 210 static char* reserve_mmapped_memory(size_t bytes, char* requested_addr) {
 211   char * addr;
 212   int flags = MAP_PRIVATE NOT_AIX( | MAP_NORESERVE ) | MAP_ANONYMOUS;
 213   if (requested_addr != NULL) {
 214     assert((uintptr_t)requested_addr % os::vm_page_size() == 0, "Requested address should be aligned to OS page size");
 215     flags |= MAP_FIXED;
 216   }
 217 
 218   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
 219   // touch an uncommitted page. Otherwise, the read/write might
 220   // succeed if we have enough swap space to back the physical page.
 221   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
 222                        flags, -1, 0);
 223 
 224   if (addr != MAP_FAILED) {
 225     MemTracker::record_virtual_memory_reserve((address)addr, bytes, CALLER_PC);
 226     return addr;
 227   }
 228   return NULL;
 229 }
 230 
 231 static int util_posix_fallocate(int fd, off_t offset, off_t len) {
 232 #ifdef __APPLE__
 233   fstore_t store = { F_ALLOCATECONTIG, F_PEOFPOSMODE, 0, len };
 234   // First we try to get a continuous chunk of disk space
 235   int ret = fcntl(fd, F_PREALLOCATE, &store);
 236   if (ret == -1) {
 237     // Maybe we are too fragmented, try to allocate non-continuous range
 238     store.fst_flags = F_ALLOCATEALL;
 239     ret = fcntl(fd, F_PREALLOCATE, &store);
 240   }
 241   if(ret != -1) {
 242     return ftruncate(fd, len);
 243   }
 244   return -1;
 245 #else
 246   return posix_fallocate(fd, offset, len);
 247 #endif
 248 }
 249 
 250 // Map the given address range to the provided file descriptor.
 251 char* os::map_memory_to_file(char* base, size_t size, int fd) {
 252   assert(fd != -1, "File descriptor is not valid");
 253 
 254   // allocate space for the file
 255   int ret = util_posix_fallocate(fd, 0, (off_t)size);
 256   if (ret != 0) {
 257     vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory. error(%d)", ret));
 258     return NULL;
 259   }
 260 
 261   int prot = PROT_READ | PROT_WRITE;
 262   int flags = MAP_SHARED;
 263   if (base != NULL) {
 264     flags |= MAP_FIXED;
 265   }
 266   char* addr = (char*)mmap(base, size, prot, flags, fd, 0);
 267 
 268   if (addr == MAP_FAILED) {
 269     warning("Failed mmap to file. (%s)", os::strerror(errno));
 270     return NULL;
 271   }
 272   if (base != NULL && addr != base) {
 273     if (!os::release_memory(addr, size)) {
 274       warning("Could not release memory on unsuccessful file mapping");
 275     }
 276     return NULL;
 277   }
 278   return addr;
 279 }
 280 
 281 char* os::replace_existing_mapping_with_file_mapping(char* base, size_t size, int fd) {
 282   assert(fd != -1, "File descriptor is not valid");
 283   assert(base != NULL, "Base cannot be NULL");
 284 
 285   return map_memory_to_file(base, size, fd);
 286 }
 287 
 288 // Multiple threads can race in this code, and can remap over each other with MAP_FIXED,
 289 // so on posix, unmap the section at the start and at the end of the chunk that we mapped
 290 // rather than unmapping and remapping the whole chunk to get requested alignment.
 291 char* os::reserve_memory_aligned(size_t size, size_t alignment, int file_desc) {
 292   assert((alignment & (os::vm_allocation_granularity() - 1)) == 0,
 293       "Alignment must be a multiple of allocation granularity (page size)");
 294   assert((size & (alignment -1)) == 0, "size must be 'alignment' aligned");
 295 
 296   size_t extra_size = size + alignment;
 297   assert(extra_size >= size, "overflow, size is too large to allow alignment");
 298 
 299   char* extra_base;
 300   if (file_desc != -1) {
 301     // For file mapping, we do not call os:reserve_memory(extra_size, NULL, alignment, file_desc) because
 302     // we need to deal with shrinking of the file space later when we release extra memory after alignment.
 303     // We also cannot called os:reserve_memory() with file_desc set to -1 because on aix we might get SHM memory.
 304     // So here to call a helper function while reserve memory for us. After we have a aligned base,
 305     // we will replace anonymous mapping with file mapping.
 306     extra_base = reserve_mmapped_memory(extra_size, NULL);
 307     if (extra_base != NULL) {
 308       MemTracker::record_virtual_memory_reserve((address)extra_base, extra_size, CALLER_PC);
 309     }
 310   } else {
 311     extra_base = os::reserve_memory(extra_size, NULL, alignment);
 312   }
 313 
 314   if (extra_base == NULL) {
 315     return NULL;
 316   }
 317 
 318   // Do manual alignment
 319   char* aligned_base = align_up(extra_base, alignment);
 320 
 321   // [  |                                       |  ]
 322   // ^ extra_base
 323   //    ^ extra_base + begin_offset == aligned_base
 324   //     extra_base + begin_offset + size       ^
 325   //                       extra_base + extra_size ^
 326   // |<>| == begin_offset
 327   //                              end_offset == |<>|
 328   size_t begin_offset = aligned_base - extra_base;
 329   size_t end_offset = (extra_base + extra_size) - (aligned_base + size);
 330 
 331   if (begin_offset > 0) {
 332       os::release_memory(extra_base, begin_offset);
 333   }
 334 
 335   if (end_offset > 0) {
 336       os::release_memory(extra_base + begin_offset + size, end_offset);
 337   }
 338 
 339   if (file_desc != -1) {
 340     // After we have an aligned address, we can replace anonymous mapping with file mapping
 341     if (replace_existing_mapping_with_file_mapping(aligned_base, size, file_desc) == NULL) {
 342       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
 343     }
 344     MemTracker::record_virtual_memory_commit((address)aligned_base, size, CALLER_PC);
 345   }
 346   return aligned_base;
 347 }
 348 
 349 int os::vsnprintf(char* buf, size_t len, const char* fmt, va_list args) {
 350   // All supported POSIX platforms provide C99 semantics.
 351   int result = ::vsnprintf(buf, len, fmt, args);
 352   // If an encoding error occurred (result < 0) then it's not clear
 353   // whether the buffer is NUL terminated, so ensure it is.
 354   if ((result < 0) && (len > 0)) {
 355     buf[len - 1] = '\0';
 356   }
 357   return result;
 358 }
 359 
 360 int os::get_fileno(FILE* fp) {
 361   return NOT_AIX(::)fileno(fp);
 362 }
 363 
 364 struct tm* os::gmtime_pd(const time_t* clock, struct tm*  res) {
 365   return gmtime_r(clock, res);
 366 }
 367 
 368 void os::Posix::print_load_average(outputStream* st) {
 369   st->print("load average:");
 370   double loadavg[3];
 371   int res = os::loadavg(loadavg, 3);
 372   if (res != -1) {
 373     st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
 374   } else {
 375     st->print(" Unavailable");
 376   }
 377   st->cr();
 378 }
 379 
 380 void os::Posix::print_rlimit_info(outputStream* st) {
 381   st->print("rlimit:");
 382   struct rlimit rlim;
 383 
 384   st->print(" STACK ");
 385   getrlimit(RLIMIT_STACK, &rlim);
 386   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 387   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 388 
 389   st->print(", CORE ");
 390   getrlimit(RLIMIT_CORE, &rlim);
 391   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 392   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 393 
 394   // Isn't there on solaris
 395 #if defined(AIX)
 396   st->print(", NPROC ");
 397   st->print("%d", sysconf(_SC_CHILD_MAX));
 398 #elif !defined(SOLARIS)
 399   st->print(", NPROC ");
 400   getrlimit(RLIMIT_NPROC, &rlim);
 401   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 402   else st->print(UINT64_FORMAT, uint64_t(rlim.rlim_cur));
 403 #endif
 404 
 405   st->print(", NOFILE ");
 406   getrlimit(RLIMIT_NOFILE, &rlim);
 407   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 408   else st->print(UINT64_FORMAT, uint64_t(rlim.rlim_cur));
 409 
 410   st->print(", AS ");
 411   getrlimit(RLIMIT_AS, &rlim);
 412   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 413   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 414 
 415   st->print(", DATA ");
 416   getrlimit(RLIMIT_DATA, &rlim);
 417   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 418   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 419 
 420   st->print(", FSIZE ");
 421   getrlimit(RLIMIT_FSIZE, &rlim);
 422   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
 423   else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
 424 
 425   st->cr();
 426 }
 427 
 428 void os::Posix::print_uname_info(outputStream* st) {
 429   // kernel
 430   st->print("uname:");
 431   struct utsname name;
 432   uname(&name);
 433   st->print("%s ", name.sysname);
 434 #ifdef ASSERT
 435   st->print("%s ", name.nodename);
 436 #endif
 437   st->print("%s ", name.release);
 438   st->print("%s ", name.version);
 439   st->print("%s", name.machine);
 440   st->cr();
 441 }
 442 
 443 void os::Posix::print_umask(outputStream* st, mode_t umsk) {
 444   st->print((umsk & S_IRUSR) ? "r" : "-");
 445   st->print((umsk & S_IWUSR) ? "w" : "-");
 446   st->print((umsk & S_IXUSR) ? "x" : "-");
 447   st->print((umsk & S_IRGRP) ? "r" : "-");
 448   st->print((umsk & S_IWGRP) ? "w" : "-");
 449   st->print((umsk & S_IXGRP) ? "x" : "-");
 450   st->print((umsk & S_IROTH) ? "r" : "-");
 451   st->print((umsk & S_IWOTH) ? "w" : "-");
 452   st->print((umsk & S_IXOTH) ? "x" : "-");
 453 }
 454 
 455 void os::Posix::print_user_info(outputStream* st) {
 456   unsigned id = (unsigned) ::getuid();
 457   st->print("uid  : %u ", id);
 458   id = (unsigned) ::geteuid();
 459   st->print("euid : %u ", id);
 460   id = (unsigned) ::getgid();
 461   st->print("gid  : %u ", id);
 462   id = (unsigned) ::getegid();
 463   st->print_cr("egid : %u", id);
 464   st->cr();
 465 
 466   mode_t umsk = ::umask(0);
 467   ::umask(umsk);
 468   st->print("umask: %04o (", (unsigned) umsk);
 469   print_umask(st, umsk);
 470   st->print_cr(")");
 471   st->cr();
 472 }
 473 
 474 
 475 bool os::get_host_name(char* buf, size_t buflen) {
 476   struct utsname name;
 477   uname(&name);
 478   jio_snprintf(buf, buflen, "%s", name.nodename);
 479   return true;
 480 }
 481 
 482 bool os::has_allocatable_memory_limit(julong* limit) {
 483   struct rlimit rlim;
 484   int getrlimit_res = getrlimit(RLIMIT_AS, &rlim);
 485   // if there was an error when calling getrlimit, assume that there is no limitation
 486   // on virtual memory.
 487   bool result;
 488   if ((getrlimit_res != 0) || (rlim.rlim_cur == RLIM_INFINITY)) {
 489     result = false;
 490   } else {
 491     *limit = (julong)rlim.rlim_cur;
 492     result = true;
 493   }
 494 #ifdef _LP64
 495   return result;
 496 #else
 497   // arbitrary virtual space limit for 32 bit Unices found by testing. If
 498   // getrlimit above returned a limit, bound it with this limit. Otherwise
 499   // directly use it.
 500   const julong max_virtual_limit = (julong)3800*M;
 501   if (result) {
 502     *limit = MIN2(*limit, max_virtual_limit);
 503   } else {
 504     *limit = max_virtual_limit;
 505   }
 506 
 507   // bound by actually allocatable memory. The algorithm uses two bounds, an
 508   // upper and a lower limit. The upper limit is the current highest amount of
 509   // memory that could not be allocated, the lower limit is the current highest
 510   // amount of memory that could be allocated.
 511   // The algorithm iteratively refines the result by halving the difference
 512   // between these limits, updating either the upper limit (if that value could
 513   // not be allocated) or the lower limit (if the that value could be allocated)
 514   // until the difference between these limits is "small".
 515 
 516   // the minimum amount of memory we care about allocating.
 517   const julong min_allocation_size = M;
 518 
 519   julong upper_limit = *limit;
 520 
 521   // first check a few trivial cases
 522   if (is_allocatable(upper_limit) || (upper_limit <= min_allocation_size)) {
 523     *limit = upper_limit;
 524   } else if (!is_allocatable(min_allocation_size)) {
 525     // we found that not even min_allocation_size is allocatable. Return it
 526     // anyway. There is no point to search for a better value any more.
 527     *limit = min_allocation_size;
 528   } else {
 529     // perform the binary search.
 530     julong lower_limit = min_allocation_size;
 531     while ((upper_limit - lower_limit) > min_allocation_size) {
 532       julong temp_limit = ((upper_limit - lower_limit) / 2) + lower_limit;
 533       temp_limit = align_down(temp_limit, min_allocation_size);
 534       if (is_allocatable(temp_limit)) {
 535         lower_limit = temp_limit;
 536       } else {
 537         upper_limit = temp_limit;
 538       }
 539     }
 540     *limit = lower_limit;
 541   }
 542   return true;
 543 #endif
 544 }
 545 
 546 const char* os::get_current_directory(char *buf, size_t buflen) {
 547   return getcwd(buf, buflen);
 548 }
 549 
 550 FILE* os::open(int fd, const char* mode) {
 551   return ::fdopen(fd, mode);
 552 }
 553 
 554 ssize_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
 555   return ::pread(fd, buf, nBytes, offset);
 556 }
 557 
 558 void os::flockfile(FILE* fp) {
 559   ::flockfile(fp);
 560 }
 561 
 562 void os::funlockfile(FILE* fp) {
 563   ::funlockfile(fp);
 564 }
 565 
 566 DIR* os::opendir(const char* dirname) {
 567   assert(dirname != NULL, "just checking");
 568   return ::opendir(dirname);
 569 }
 570 
 571 struct dirent* os::readdir(DIR* dirp) {
 572   assert(dirp != NULL, "just checking");
 573   return ::readdir(dirp);
 574 }
 575 
 576 int os::closedir(DIR *dirp) {
 577   assert(dirp != NULL, "just checking");
 578   return ::closedir(dirp);
 579 }
 580 
 581 // Builds a platform dependent Agent_OnLoad_<lib_name> function name
 582 // which is used to find statically linked in agents.
 583 // Parameters:
 584 //            sym_name: Symbol in library we are looking for
 585 //            lib_name: Name of library to look in, NULL for shared libs.
 586 //            is_absolute_path == true if lib_name is absolute path to agent
 587 //                                     such as "/a/b/libL.so"
 588 //            == false if only the base name of the library is passed in
 589 //               such as "L"
 590 char* os::build_agent_function_name(const char *sym_name, const char *lib_name,
 591                                     bool is_absolute_path) {
 592   char *agent_entry_name;
 593   size_t len;
 594   size_t name_len;
 595   size_t prefix_len = strlen(JNI_LIB_PREFIX);
 596   size_t suffix_len = strlen(JNI_LIB_SUFFIX);
 597   const char *start;
 598 
 599   if (lib_name != NULL) {
 600     name_len = strlen(lib_name);
 601     if (is_absolute_path) {
 602       // Need to strip path, prefix and suffix
 603       if ((start = strrchr(lib_name, *os::file_separator())) != NULL) {
 604         lib_name = ++start;
 605       }
 606       if (strlen(lib_name) <= (prefix_len + suffix_len)) {
 607         return NULL;
 608       }
 609       lib_name += prefix_len;
 610       name_len = strlen(lib_name) - suffix_len;
 611     }
 612   }
 613   len = (lib_name != NULL ? name_len : 0) + strlen(sym_name) + 2;
 614   agent_entry_name = NEW_C_HEAP_ARRAY_RETURN_NULL(char, len, mtThread);
 615   if (agent_entry_name == NULL) {
 616     return NULL;
 617   }
 618   strcpy(agent_entry_name, sym_name);
 619   if (lib_name != NULL) {
 620     strcat(agent_entry_name, "_");
 621     strncat(agent_entry_name, lib_name, name_len);
 622   }
 623   return agent_entry_name;
 624 }
 625 
 626 
 627 void os::naked_short_nanosleep(jlong ns) {
 628   struct timespec req;
 629   assert(ns > -1 && ns < NANOUNITS, "Un-interruptable sleep, short time use only");
 630   req.tv_sec = 0;
 631   req.tv_nsec = ns;
 632   ::nanosleep(&req, NULL);
 633   return;
 634 }
 635 
 636 void os::naked_short_sleep(jlong ms) {
 637   assert(ms < MILLIUNITS, "Un-interruptable sleep, short time use only");
 638   os::naked_short_nanosleep(ms * (NANOUNITS / MILLIUNITS));
 639   return;
 640 }
 641 
 642 static const struct {
 643   int sig; const char* name;
 644 }
 645  g_signal_info[] =
 646   {
 647   {  SIGABRT,     "SIGABRT" },
 648 #ifdef SIGAIO
 649   {  SIGAIO,      "SIGAIO" },
 650 #endif
 651   {  SIGALRM,     "SIGALRM" },
 652 #ifdef SIGALRM1
 653   {  SIGALRM1,    "SIGALRM1" },
 654 #endif
 655   {  SIGBUS,      "SIGBUS" },
 656 #ifdef SIGCANCEL
 657   {  SIGCANCEL,   "SIGCANCEL" },
 658 #endif
 659   {  SIGCHLD,     "SIGCHLD" },
 660 #ifdef SIGCLD
 661   {  SIGCLD,      "SIGCLD" },
 662 #endif
 663   {  SIGCONT,     "SIGCONT" },
 664 #ifdef SIGCPUFAIL
 665   {  SIGCPUFAIL,  "SIGCPUFAIL" },
 666 #endif
 667 #ifdef SIGDANGER
 668   {  SIGDANGER,   "SIGDANGER" },
 669 #endif
 670 #ifdef SIGDIL
 671   {  SIGDIL,      "SIGDIL" },
 672 #endif
 673 #ifdef SIGEMT
 674   {  SIGEMT,      "SIGEMT" },
 675 #endif
 676   {  SIGFPE,      "SIGFPE" },
 677 #ifdef SIGFREEZE
 678   {  SIGFREEZE,   "SIGFREEZE" },
 679 #endif
 680 #ifdef SIGGFAULT
 681   {  SIGGFAULT,   "SIGGFAULT" },
 682 #endif
 683 #ifdef SIGGRANT
 684   {  SIGGRANT,    "SIGGRANT" },
 685 #endif
 686   {  SIGHUP,      "SIGHUP" },
 687   {  SIGILL,      "SIGILL" },
 688   {  SIGINT,      "SIGINT" },
 689 #ifdef SIGIO
 690   {  SIGIO,       "SIGIO" },
 691 #endif
 692 #ifdef SIGIOINT
 693   {  SIGIOINT,    "SIGIOINT" },
 694 #endif
 695 #ifdef SIGIOT
 696 // SIGIOT is there for BSD compatibility, but on most Unices just a
 697 // synonym for SIGABRT. The result should be "SIGABRT", not
 698 // "SIGIOT".
 699 #if (SIGIOT != SIGABRT )
 700   {  SIGIOT,      "SIGIOT" },
 701 #endif
 702 #endif
 703 #ifdef SIGKAP
 704   {  SIGKAP,      "SIGKAP" },
 705 #endif
 706   {  SIGKILL,     "SIGKILL" },
 707 #ifdef SIGLOST
 708   {  SIGLOST,     "SIGLOST" },
 709 #endif
 710 #ifdef SIGLWP
 711   {  SIGLWP,      "SIGLWP" },
 712 #endif
 713 #ifdef SIGLWPTIMER
 714   {  SIGLWPTIMER, "SIGLWPTIMER" },
 715 #endif
 716 #ifdef SIGMIGRATE
 717   {  SIGMIGRATE,  "SIGMIGRATE" },
 718 #endif
 719 #ifdef SIGMSG
 720   {  SIGMSG,      "SIGMSG" },
 721 #endif
 722   {  SIGPIPE,     "SIGPIPE" },
 723 #ifdef SIGPOLL
 724   {  SIGPOLL,     "SIGPOLL" },
 725 #endif
 726 #ifdef SIGPRE
 727   {  SIGPRE,      "SIGPRE" },
 728 #endif
 729   {  SIGPROF,     "SIGPROF" },
 730 #ifdef SIGPTY
 731   {  SIGPTY,      "SIGPTY" },
 732 #endif
 733 #ifdef SIGPWR
 734   {  SIGPWR,      "SIGPWR" },
 735 #endif
 736   {  SIGQUIT,     "SIGQUIT" },
 737 #ifdef SIGRECONFIG
 738   {  SIGRECONFIG, "SIGRECONFIG" },
 739 #endif
 740 #ifdef SIGRECOVERY
 741   {  SIGRECOVERY, "SIGRECOVERY" },
 742 #endif
 743 #ifdef SIGRESERVE
 744   {  SIGRESERVE,  "SIGRESERVE" },
 745 #endif
 746 #ifdef SIGRETRACT
 747   {  SIGRETRACT,  "SIGRETRACT" },
 748 #endif
 749 #ifdef SIGSAK
 750   {  SIGSAK,      "SIGSAK" },
 751 #endif
 752   {  SIGSEGV,     "SIGSEGV" },
 753 #ifdef SIGSOUND
 754   {  SIGSOUND,    "SIGSOUND" },
 755 #endif
 756 #ifdef SIGSTKFLT
 757   {  SIGSTKFLT,    "SIGSTKFLT" },
 758 #endif
 759   {  SIGSTOP,     "SIGSTOP" },
 760   {  SIGSYS,      "SIGSYS" },
 761 #ifdef SIGSYSERROR
 762   {  SIGSYSERROR, "SIGSYSERROR" },
 763 #endif
 764 #ifdef SIGTALRM
 765   {  SIGTALRM,    "SIGTALRM" },
 766 #endif
 767   {  SIGTERM,     "SIGTERM" },
 768 #ifdef SIGTHAW
 769   {  SIGTHAW,     "SIGTHAW" },
 770 #endif
 771   {  SIGTRAP,     "SIGTRAP" },
 772 #ifdef SIGTSTP
 773   {  SIGTSTP,     "SIGTSTP" },
 774 #endif
 775   {  SIGTTIN,     "SIGTTIN" },
 776   {  SIGTTOU,     "SIGTTOU" },
 777 #ifdef SIGURG
 778   {  SIGURG,      "SIGURG" },
 779 #endif
 780   {  SIGUSR1,     "SIGUSR1" },
 781   {  SIGUSR2,     "SIGUSR2" },
 782 #ifdef SIGVIRT
 783   {  SIGVIRT,     "SIGVIRT" },
 784 #endif
 785   {  SIGVTALRM,   "SIGVTALRM" },
 786 #ifdef SIGWAITING
 787   {  SIGWAITING,  "SIGWAITING" },
 788 #endif
 789 #ifdef SIGWINCH
 790   {  SIGWINCH,    "SIGWINCH" },
 791 #endif
 792 #ifdef SIGWINDOW
 793   {  SIGWINDOW,   "SIGWINDOW" },
 794 #endif
 795   {  SIGXCPU,     "SIGXCPU" },
 796   {  SIGXFSZ,     "SIGXFSZ" },
 797 #ifdef SIGXRES
 798   {  SIGXRES,     "SIGXRES" },
 799 #endif
 800   { -1, NULL }
 801 };
 802 
 803 // Returned string is a constant. For unknown signals "UNKNOWN" is returned.
 804 const char* os::Posix::get_signal_name(int sig, char* out, size_t outlen) {
 805 
 806   const char* ret = NULL;
 807 
 808 #ifdef SIGRTMIN
 809   if (sig >= SIGRTMIN && sig <= SIGRTMAX) {
 810     if (sig == SIGRTMIN) {
 811       ret = "SIGRTMIN";
 812     } else if (sig == SIGRTMAX) {
 813       ret = "SIGRTMAX";
 814     } else {
 815       jio_snprintf(out, outlen, "SIGRTMIN+%d", sig - SIGRTMIN);
 816       return out;
 817     }
 818   }
 819 #endif
 820 
 821   if (sig > 0) {
 822     for (int idx = 0; g_signal_info[idx].sig != -1; idx ++) {
 823       if (g_signal_info[idx].sig == sig) {
 824         ret = g_signal_info[idx].name;
 825         break;
 826       }
 827     }
 828   }
 829 
 830   if (!ret) {
 831     if (!is_valid_signal(sig)) {
 832       ret = "INVALID";
 833     } else {
 834       ret = "UNKNOWN";
 835     }
 836   }
 837 
 838   if (out && outlen > 0) {
 839     strncpy(out, ret, outlen);
 840     out[outlen - 1] = '\0';
 841   }
 842   return out;
 843 }
 844 
 845 int os::Posix::get_signal_number(const char* signal_name) {
 846   char tmp[30];
 847   const char* s = signal_name;
 848   if (s[0] != 'S' || s[1] != 'I' || s[2] != 'G') {
 849     jio_snprintf(tmp, sizeof(tmp), "SIG%s", signal_name);
 850     s = tmp;
 851   }
 852   for (int idx = 0; g_signal_info[idx].sig != -1; idx ++) {
 853     if (strcmp(g_signal_info[idx].name, s) == 0) {
 854       return g_signal_info[idx].sig;
 855     }
 856   }
 857   return -1;
 858 }
 859 
 860 int os::get_signal_number(const char* signal_name) {
 861   return os::Posix::get_signal_number(signal_name);
 862 }
 863 
 864 // Returns true if signal number is valid.
 865 bool os::Posix::is_valid_signal(int sig) {
 866   // MacOS not really POSIX compliant: sigaddset does not return
 867   // an error for invalid signal numbers. However, MacOS does not
 868   // support real time signals and simply seems to have just 33
 869   // signals with no holes in the signal range.
 870 #ifdef __APPLE__
 871   return sig >= 1 && sig < NSIG;
 872 #else
 873   // Use sigaddset to check for signal validity.
 874   sigset_t set;
 875   sigemptyset(&set);
 876   if (sigaddset(&set, sig) == -1 && errno == EINVAL) {
 877     return false;
 878   }
 879   return true;
 880 #endif
 881 }
 882 
 883 bool os::Posix::is_sig_ignored(int sig) {
 884   struct sigaction oact;
 885   sigaction(sig, (struct sigaction*)NULL, &oact);
 886   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 887                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 888   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 889     return true;
 890   } else {
 891     return false;
 892   }
 893 }
 894 
 895 // Returns:
 896 // NULL for an invalid signal number
 897 // "SIG<num>" for a valid but unknown signal number
 898 // signal name otherwise.
 899 const char* os::exception_name(int sig, char* buf, size_t size) {
 900   if (!os::Posix::is_valid_signal(sig)) {
 901     return NULL;
 902   }
 903   const char* const name = os::Posix::get_signal_name(sig, buf, size);
 904   if (strcmp(name, "UNKNOWN") == 0) {
 905     jio_snprintf(buf, size, "SIG%d", sig);
 906   }
 907   return buf;
 908 }
 909 
 910 #define NUM_IMPORTANT_SIGS 32
 911 // Returns one-line short description of a signal set in a user provided buffer.
 912 const char* os::Posix::describe_signal_set_short(const sigset_t* set, char* buffer, size_t buf_size) {
 913   assert(buf_size == (NUM_IMPORTANT_SIGS + 1), "wrong buffer size");
 914   // Note: for shortness, just print out the first 32. That should
 915   // cover most of the useful ones, apart from realtime signals.
 916   for (int sig = 1; sig <= NUM_IMPORTANT_SIGS; sig++) {
 917     const int rc = sigismember(set, sig);
 918     if (rc == -1 && errno == EINVAL) {
 919       buffer[sig-1] = '?';
 920     } else {
 921       buffer[sig-1] = rc == 0 ? '0' : '1';
 922     }
 923   }
 924   buffer[NUM_IMPORTANT_SIGS] = 0;
 925   return buffer;
 926 }
 927 
 928 // Prints one-line description of a signal set.
 929 void os::Posix::print_signal_set_short(outputStream* st, const sigset_t* set) {
 930   char buf[NUM_IMPORTANT_SIGS + 1];
 931   os::Posix::describe_signal_set_short(set, buf, sizeof(buf));
 932   st->print("%s", buf);
 933 }
 934 
 935 // Writes one-line description of a combination of sigaction.sa_flags into a user
 936 // provided buffer. Returns that buffer.
 937 const char* os::Posix::describe_sa_flags(int flags, char* buffer, size_t size) {
 938   char* p = buffer;
 939   size_t remaining = size;
 940   bool first = true;
 941   int idx = 0;
 942 
 943   assert(buffer, "invalid argument");
 944 
 945   if (size == 0) {
 946     return buffer;
 947   }
 948 
 949   strncpy(buffer, "none", size);
 950 
 951   const struct {
 952     // NB: i is an unsigned int here because SA_RESETHAND is on some
 953     // systems 0x80000000, which is implicitly unsigned.  Assignining
 954     // it to an int field would be an overflow in unsigned-to-signed
 955     // conversion.
 956     unsigned int i;
 957     const char* s;
 958   } flaginfo [] = {
 959     { SA_NOCLDSTOP, "SA_NOCLDSTOP" },
 960     { SA_ONSTACK,   "SA_ONSTACK"   },
 961     { SA_RESETHAND, "SA_RESETHAND" },
 962     { SA_RESTART,   "SA_RESTART"   },
 963     { SA_SIGINFO,   "SA_SIGINFO"   },
 964     { SA_NOCLDWAIT, "SA_NOCLDWAIT" },
 965     { SA_NODEFER,   "SA_NODEFER"   },
 966 #ifdef AIX
 967     { SA_ONSTACK,   "SA_ONSTACK"   },
 968     { SA_OLDSTYLE,  "SA_OLDSTYLE"  },
 969 #endif
 970     { 0, NULL }
 971   };
 972 
 973   for (idx = 0; flaginfo[idx].s && remaining > 1; idx++) {
 974     if (flags & flaginfo[idx].i) {
 975       if (first) {
 976         jio_snprintf(p, remaining, "%s", flaginfo[idx].s);
 977         first = false;
 978       } else {
 979         jio_snprintf(p, remaining, "|%s", flaginfo[idx].s);
 980       }
 981       const size_t len = strlen(p);
 982       p += len;
 983       remaining -= len;
 984     }
 985   }
 986 
 987   buffer[size - 1] = '\0';
 988 
 989   return buffer;
 990 }
 991 
 992 // Prints one-line description of a combination of sigaction.sa_flags.
 993 void os::Posix::print_sa_flags(outputStream* st, int flags) {
 994   char buffer[0x100];
 995   os::Posix::describe_sa_flags(flags, buffer, sizeof(buffer));
 996   st->print("%s", buffer);
 997 }
 998 
 999 // Helper function for os::Posix::print_siginfo_...():
1000 // return a textual description for signal code.
1001 struct enum_sigcode_desc_t {
1002   const char* s_name;
1003   const char* s_desc;
1004 };
1005 
1006 static bool get_signal_code_description(const siginfo_t* si, enum_sigcode_desc_t* out) {
1007 
1008   const struct {
1009     int sig; int code; const char* s_code; const char* s_desc;
1010   } t1 [] = {
1011     { SIGILL,  ILL_ILLOPC,   "ILL_ILLOPC",   "Illegal opcode." },
1012     { SIGILL,  ILL_ILLOPN,   "ILL_ILLOPN",   "Illegal operand." },
1013     { SIGILL,  ILL_ILLADR,   "ILL_ILLADR",   "Illegal addressing mode." },
1014     { SIGILL,  ILL_ILLTRP,   "ILL_ILLTRP",   "Illegal trap." },
1015     { SIGILL,  ILL_PRVOPC,   "ILL_PRVOPC",   "Privileged opcode." },
1016     { SIGILL,  ILL_PRVREG,   "ILL_PRVREG",   "Privileged register." },
1017     { SIGILL,  ILL_COPROC,   "ILL_COPROC",   "Coprocessor error." },
1018     { SIGILL,  ILL_BADSTK,   "ILL_BADSTK",   "Internal stack error." },
1019 #if defined(IA64) && defined(LINUX)
1020     { SIGILL,  ILL_BADIADDR, "ILL_BADIADDR", "Unimplemented instruction address" },
1021     { SIGILL,  ILL_BREAK,    "ILL_BREAK",    "Application Break instruction" },
1022 #endif
1023     { SIGFPE,  FPE_INTDIV,   "FPE_INTDIV",   "Integer divide by zero." },
1024     { SIGFPE,  FPE_INTOVF,   "FPE_INTOVF",   "Integer overflow." },
1025     { SIGFPE,  FPE_FLTDIV,   "FPE_FLTDIV",   "Floating-point divide by zero." },
1026     { SIGFPE,  FPE_FLTOVF,   "FPE_FLTOVF",   "Floating-point overflow." },
1027     { SIGFPE,  FPE_FLTUND,   "FPE_FLTUND",   "Floating-point underflow." },
1028     { SIGFPE,  FPE_FLTRES,   "FPE_FLTRES",   "Floating-point inexact result." },
1029     { SIGFPE,  FPE_FLTINV,   "FPE_FLTINV",   "Invalid floating-point operation." },
1030     { SIGFPE,  FPE_FLTSUB,   "FPE_FLTSUB",   "Subscript out of range." },
1031     { SIGSEGV, SEGV_MAPERR,  "SEGV_MAPERR",  "Address not mapped to object." },
1032     { SIGSEGV, SEGV_ACCERR,  "SEGV_ACCERR",  "Invalid permissions for mapped object." },
1033 #ifdef AIX
1034     // no explanation found what keyerr would be
1035     { SIGSEGV, SEGV_KEYERR,  "SEGV_KEYERR",  "key error" },
1036 #endif
1037 #if defined(IA64) && !defined(AIX)
1038     { SIGSEGV, SEGV_PSTKOVF, "SEGV_PSTKOVF", "Paragraph stack overflow" },
1039 #endif
1040 #if defined(__sparc) && defined(SOLARIS)
1041 // define Solaris Sparc M7 ADI SEGV signals
1042 #if !defined(SEGV_ACCADI)
1043 #define SEGV_ACCADI 3
1044 #endif
1045     { SIGSEGV, SEGV_ACCADI,  "SEGV_ACCADI",  "ADI not enabled for mapped object." },
1046 #if !defined(SEGV_ACCDERR)
1047 #define SEGV_ACCDERR 4
1048 #endif
1049     { SIGSEGV, SEGV_ACCDERR, "SEGV_ACCDERR", "ADI disrupting exception." },
1050 #if !defined(SEGV_ACCPERR)
1051 #define SEGV_ACCPERR 5
1052 #endif
1053     { SIGSEGV, SEGV_ACCPERR, "SEGV_ACCPERR", "ADI precise exception." },
1054 #endif // defined(__sparc) && defined(SOLARIS)
1055     { SIGBUS,  BUS_ADRALN,   "BUS_ADRALN",   "Invalid address alignment." },
1056     { SIGBUS,  BUS_ADRERR,   "BUS_ADRERR",   "Nonexistent physical address." },
1057     { SIGBUS,  BUS_OBJERR,   "BUS_OBJERR",   "Object-specific hardware error." },
1058     { SIGTRAP, TRAP_BRKPT,   "TRAP_BRKPT",   "Process breakpoint." },
1059     { SIGTRAP, TRAP_TRACE,   "TRAP_TRACE",   "Process trace trap." },
1060     { SIGCHLD, CLD_EXITED,   "CLD_EXITED",   "Child has exited." },
1061     { SIGCHLD, CLD_KILLED,   "CLD_KILLED",   "Child has terminated abnormally and did not create a core file." },
1062     { SIGCHLD, CLD_DUMPED,   "CLD_DUMPED",   "Child has terminated abnormally and created a core file." },
1063     { SIGCHLD, CLD_TRAPPED,  "CLD_TRAPPED",  "Traced child has trapped." },
1064     { SIGCHLD, CLD_STOPPED,  "CLD_STOPPED",  "Child has stopped." },
1065     { SIGCHLD, CLD_CONTINUED,"CLD_CONTINUED","Stopped child has continued." },
1066 #ifdef SIGPOLL
1067     { SIGPOLL, POLL_OUT,     "POLL_OUT",     "Output buffers available." },
1068     { SIGPOLL, POLL_MSG,     "POLL_MSG",     "Input message available." },
1069     { SIGPOLL, POLL_ERR,     "POLL_ERR",     "I/O error." },
1070     { SIGPOLL, POLL_PRI,     "POLL_PRI",     "High priority input available." },
1071     { SIGPOLL, POLL_HUP,     "POLL_HUP",     "Device disconnected. [Option End]" },
1072 #endif
1073     { -1, -1, NULL, NULL }
1074   };
1075 
1076   // Codes valid in any signal context.
1077   const struct {
1078     int code; const char* s_code; const char* s_desc;
1079   } t2 [] = {
1080     { SI_USER,      "SI_USER",     "Signal sent by kill()." },
1081     { SI_QUEUE,     "SI_QUEUE",    "Signal sent by the sigqueue()." },
1082     { SI_TIMER,     "SI_TIMER",    "Signal generated by expiration of a timer set by timer_settime()." },
1083     { SI_ASYNCIO,   "SI_ASYNCIO",  "Signal generated by completion of an asynchronous I/O request." },
1084     { SI_MESGQ,     "SI_MESGQ",    "Signal generated by arrival of a message on an empty message queue." },
1085     // Linux specific
1086 #ifdef SI_TKILL
1087     { SI_TKILL,     "SI_TKILL",    "Signal sent by tkill (pthread_kill)" },
1088 #endif
1089 #ifdef SI_DETHREAD
1090     { SI_DETHREAD,  "SI_DETHREAD", "Signal sent by execve() killing subsidiary threads" },
1091 #endif
1092 #ifdef SI_KERNEL
1093     { SI_KERNEL,    "SI_KERNEL",   "Signal sent by kernel." },
1094 #endif
1095 #ifdef SI_SIGIO
1096     { SI_SIGIO,     "SI_SIGIO",    "Signal sent by queued SIGIO" },
1097 #endif
1098 
1099 #ifdef AIX
1100     { SI_UNDEFINED, "SI_UNDEFINED","siginfo contains partial information" },
1101     { SI_EMPTY,     "SI_EMPTY",    "siginfo contains no useful information" },
1102 #endif
1103 
1104 #ifdef __sun
1105     { SI_NOINFO,    "SI_NOINFO",   "No signal information" },
1106     { SI_RCTL,      "SI_RCTL",     "kernel generated signal via rctl action" },
1107     { SI_LWP,       "SI_LWP",      "Signal sent via lwp_kill" },
1108 #endif
1109 
1110     { -1, NULL, NULL }
1111   };
1112 
1113   const char* s_code = NULL;
1114   const char* s_desc = NULL;
1115 
1116   for (int i = 0; t1[i].sig != -1; i ++) {
1117     if (t1[i].sig == si->si_signo && t1[i].code == si->si_code) {
1118       s_code = t1[i].s_code;
1119       s_desc = t1[i].s_desc;
1120       break;
1121     }
1122   }
1123 
1124   if (s_code == NULL) {
1125     for (int i = 0; t2[i].s_code != NULL; i ++) {
1126       if (t2[i].code == si->si_code) {
1127         s_code = t2[i].s_code;
1128         s_desc = t2[i].s_desc;
1129       }
1130     }
1131   }
1132 
1133   if (s_code == NULL) {
1134     out->s_name = "unknown";
1135     out->s_desc = "unknown";
1136     return false;
1137   }
1138 
1139   out->s_name = s_code;
1140   out->s_desc = s_desc;
1141 
1142   return true;
1143 }
1144 
1145 bool os::signal_sent_by_kill(const void* siginfo) {
1146   const siginfo_t* const si = (const siginfo_t*)siginfo;
1147   return si->si_code == SI_USER || si->si_code == SI_QUEUE
1148 #ifdef SI_TKILL
1149          || si->si_code == SI_TKILL
1150 #endif
1151   ;
1152 }
1153 
1154 void os::print_siginfo(outputStream* os, const void* si0) {
1155 
1156   const siginfo_t* const si = (const siginfo_t*) si0;
1157 
1158   char buf[20];
1159   os->print("siginfo:");
1160 
1161   if (!si) {
1162     os->print(" <null>");
1163     return;
1164   }
1165 
1166   const int sig = si->si_signo;
1167 
1168   os->print(" si_signo: %d (%s)", sig, os::Posix::get_signal_name(sig, buf, sizeof(buf)));
1169 
1170   enum_sigcode_desc_t ed;
1171   get_signal_code_description(si, &ed);
1172   os->print(", si_code: %d (%s)", si->si_code, ed.s_name);
1173 
1174   if (si->si_errno) {
1175     os->print(", si_errno: %d", si->si_errno);
1176   }
1177 
1178   // Output additional information depending on the signal code.
1179 
1180   // Note: Many implementations lump si_addr, si_pid, si_uid etc. together as unions,
1181   // so it depends on the context which member to use. For synchronous error signals,
1182   // we print si_addr, unless the signal was sent by another process or thread, in
1183   // which case we print out pid or tid of the sender.
1184   if (signal_sent_by_kill(si)) {
1185     const pid_t pid = si->si_pid;
1186     os->print(", si_pid: %ld", (long) pid);
1187     if (IS_VALID_PID(pid)) {
1188       const pid_t me = getpid();
1189       if (me == pid) {
1190         os->print(" (current process)");
1191       }
1192     } else {
1193       os->print(" (invalid)");
1194     }
1195     os->print(", si_uid: %ld", (long) si->si_uid);
1196     if (sig == SIGCHLD) {
1197       os->print(", si_status: %d", si->si_status);
1198     }
1199   } else if (sig == SIGSEGV || sig == SIGBUS || sig == SIGILL ||
1200              sig == SIGTRAP || sig == SIGFPE) {
1201     os->print(", si_addr: " PTR_FORMAT, p2i(si->si_addr));
1202 #ifdef SIGPOLL
1203   } else if (sig == SIGPOLL) {
1204     os->print(", si_band: %ld", si->si_band);
1205 #endif
1206   }
1207 
1208 }
1209 
1210 bool os::signal_thread(Thread* thread, int sig, const char* reason) {
1211   OSThread* osthread = thread->osthread();
1212   if (osthread) {
1213 #if defined (SOLARIS)
1214     // Note: we cannot use pthread_kill on Solaris - not because
1215     // its missing, but because we do not have the pthread_t id.
1216     int status = thr_kill(osthread->thread_id(), sig);
1217 #else
1218     int status = pthread_kill(osthread->pthread_id(), sig);
1219 #endif
1220     if (status == 0) {
1221       Events::log(Thread::current(), "sent signal %d to Thread " INTPTR_FORMAT " because %s.",
1222                   sig, p2i(thread), reason);
1223       return true;
1224     }
1225   }
1226   return false;
1227 }
1228 
1229 int os::Posix::unblock_thread_signal_mask(const sigset_t *set) {
1230   return pthread_sigmask(SIG_UNBLOCK, set, NULL);
1231 }
1232 
1233 address os::Posix::ucontext_get_pc(const ucontext_t* ctx) {
1234 #if defined(AIX)
1235    return Aix::ucontext_get_pc(ctx);
1236 #elif defined(BSD)
1237    return Bsd::ucontext_get_pc(ctx);
1238 #elif defined(LINUX)
1239    return Linux::ucontext_get_pc(ctx);
1240 #elif defined(SOLARIS)
1241    return Solaris::ucontext_get_pc(ctx);
1242 #else
1243    VMError::report_and_die("unimplemented ucontext_get_pc");
1244 #endif
1245 }
1246 
1247 void os::Posix::ucontext_set_pc(ucontext_t* ctx, address pc) {
1248 #if defined(AIX)
1249    Aix::ucontext_set_pc(ctx, pc);
1250 #elif defined(BSD)
1251    Bsd::ucontext_set_pc(ctx, pc);
1252 #elif defined(LINUX)
1253    Linux::ucontext_set_pc(ctx, pc);
1254 #elif defined(SOLARIS)
1255    Solaris::ucontext_set_pc(ctx, pc);
1256 #else
1257    VMError::report_and_die("unimplemented ucontext_get_pc");
1258 #endif
1259 }
1260 
1261 char* os::Posix::describe_pthread_attr(char* buf, size_t buflen, const pthread_attr_t* attr) {
1262   size_t stack_size = 0;
1263   size_t guard_size = 0;
1264   int detachstate = 0;
1265   pthread_attr_getstacksize(attr, &stack_size);
1266   pthread_attr_getguardsize(attr, &guard_size);
1267   // Work around linux NPTL implementation error, see also os::create_thread() in os_linux.cpp.
1268   LINUX_ONLY(stack_size -= guard_size);
1269   pthread_attr_getdetachstate(attr, &detachstate);
1270   jio_snprintf(buf, buflen, "stacksize: " SIZE_FORMAT "k, guardsize: " SIZE_FORMAT "k, %s",
1271     stack_size / 1024, guard_size / 1024,
1272     (detachstate == PTHREAD_CREATE_DETACHED ? "detached" : "joinable"));
1273   return buf;
1274 }
1275 
1276 char* os::Posix::realpath(const char* filename, char* outbuf, size_t outbuflen) {
1277 
1278   if (filename == NULL || outbuf == NULL || outbuflen < 1) {
1279     assert(false, "os::Posix::realpath: invalid arguments.");
1280     errno = EINVAL;
1281     return NULL;
1282   }
1283 
1284   char* result = NULL;
1285 
1286   // This assumes platform realpath() is implemented according to POSIX.1-2008.
1287   // POSIX.1-2008 allows to specify NULL for the output buffer, in which case
1288   // output buffer is dynamically allocated and must be ::free()'d by the caller.
1289   char* p = ::realpath(filename, NULL);
1290   if (p != NULL) {
1291     if (strlen(p) < outbuflen) {
1292       strcpy(outbuf, p);
1293       result = outbuf;
1294     } else {
1295       errno = ENAMETOOLONG;
1296     }
1297     ::free(p); // *not* os::free
1298   } else {
1299     // Fallback for platforms struggling with modern Posix standards (AIX 5.3, 6.1). If realpath
1300     // returns EINVAL, this may indicate that realpath is not POSIX.1-2008 compatible and
1301     // that it complains about the NULL we handed down as user buffer.
1302     // In this case, use the user provided buffer but at least check whether realpath caused
1303     // a memory overwrite.
1304     if (errno == EINVAL) {
1305       outbuf[outbuflen - 1] = '\0';
1306       p = ::realpath(filename, outbuf);
1307       if (p != NULL) {
1308         guarantee(outbuf[outbuflen - 1] == '\0', "realpath buffer overwrite detected.");
1309         result = p;
1310       }
1311     }
1312   }
1313   return result;
1314 
1315 }
1316 
1317 int os::stat(const char *path, struct stat *sbuf) {
1318   return ::stat(path, sbuf);
1319 }
1320 
1321 char * os::native_path(char *path) {
1322   return path;
1323 }
1324 
1325 bool os::same_files(const char* file1, const char* file2) {
1326   if (strcmp(file1, file2) == 0) {
1327     return true;
1328   }
1329 
1330   bool is_same = false;
1331   struct stat st1;
1332   struct stat st2;
1333 
1334   if (os::stat(file1, &st1) < 0) {
1335     return false;
1336   }
1337 
1338   if (os::stat(file2, &st2) < 0) {
1339     return false;
1340   }
1341 
1342   if (st1.st_dev == st2.st_dev && st1.st_ino == st2.st_ino) {
1343     // same files
1344     is_same = true;
1345   }
1346   return is_same;
1347 }
1348 
1349 // Check minimum allowable stack sizes for thread creation and to initialize
1350 // the java system classes, including StackOverflowError - depends on page
1351 // size.
1352 // The space needed for frames during startup is platform dependent. It
1353 // depends on word size, platform calling conventions, C frame layout and
1354 // interpreter/C1/C2 design decisions. Therefore this is given in a
1355 // platform (os/cpu) dependent constant.
1356 // To this, space for guard mechanisms is added, which depends on the
1357 // page size which again depends on the concrete system the VM is running
1358 // on. Space for libc guard pages is not included in this size.
1359 jint os::Posix::set_minimum_stack_sizes() {
1360   size_t os_min_stack_allowed = SOLARIS_ONLY(thr_min_stack()) NOT_SOLARIS(PTHREAD_STACK_MIN);
1361 
1362   _java_thread_min_stack_allowed = _java_thread_min_stack_allowed +
1363                                    JavaThread::stack_guard_zone_size() +
1364                                    JavaThread::stack_shadow_zone_size();
1365 
1366   _java_thread_min_stack_allowed = align_up(_java_thread_min_stack_allowed, vm_page_size());
1367   _java_thread_min_stack_allowed = MAX2(_java_thread_min_stack_allowed, os_min_stack_allowed);
1368 
1369   size_t stack_size_in_bytes = ThreadStackSize * K;
1370   if (stack_size_in_bytes != 0 &&
1371       stack_size_in_bytes < _java_thread_min_stack_allowed) {
1372     // The '-Xss' and '-XX:ThreadStackSize=N' options both set
1373     // ThreadStackSize so we go with "Java thread stack size" instead
1374     // of "ThreadStackSize" to be more friendly.
1375     tty->print_cr("\nThe Java thread stack size specified is too small. "
1376                   "Specify at least " SIZE_FORMAT "k",
1377                   _java_thread_min_stack_allowed / K);
1378     return JNI_ERR;
1379   }
1380 
1381   // Make the stack size a multiple of the page size so that
1382   // the yellow/red zones can be guarded.
1383   JavaThread::set_stack_size_at_create(align_up(stack_size_in_bytes, vm_page_size()));
1384 
1385   // Reminder: a compiler thread is a Java thread.
1386   _compiler_thread_min_stack_allowed = _compiler_thread_min_stack_allowed +
1387                                        JavaThread::stack_guard_zone_size() +
1388                                        JavaThread::stack_shadow_zone_size();
1389 
1390   _compiler_thread_min_stack_allowed = align_up(_compiler_thread_min_stack_allowed, vm_page_size());
1391   _compiler_thread_min_stack_allowed = MAX2(_compiler_thread_min_stack_allowed, os_min_stack_allowed);
1392 
1393   stack_size_in_bytes = CompilerThreadStackSize * K;
1394   if (stack_size_in_bytes != 0 &&
1395       stack_size_in_bytes < _compiler_thread_min_stack_allowed) {
1396     tty->print_cr("\nThe CompilerThreadStackSize specified is too small. "
1397                   "Specify at least " SIZE_FORMAT "k",
1398                   _compiler_thread_min_stack_allowed / K);
1399     return JNI_ERR;
1400   }
1401 
1402   _vm_internal_thread_min_stack_allowed = align_up(_vm_internal_thread_min_stack_allowed, vm_page_size());
1403   _vm_internal_thread_min_stack_allowed = MAX2(_vm_internal_thread_min_stack_allowed, os_min_stack_allowed);
1404 
1405   stack_size_in_bytes = VMThreadStackSize * K;
1406   if (stack_size_in_bytes != 0 &&
1407       stack_size_in_bytes < _vm_internal_thread_min_stack_allowed) {
1408     tty->print_cr("\nThe VMThreadStackSize specified is too small. "
1409                   "Specify at least " SIZE_FORMAT "k",
1410                   _vm_internal_thread_min_stack_allowed / K);
1411     return JNI_ERR;
1412   }
1413   return JNI_OK;
1414 }
1415 
1416 // Called when creating the thread.  The minimum stack sizes have already been calculated
1417 size_t os::Posix::get_initial_stack_size(ThreadType thr_type, size_t req_stack_size) {
1418   size_t stack_size;
1419   if (req_stack_size == 0) {
1420     stack_size = default_stack_size(thr_type);
1421   } else {
1422     stack_size = req_stack_size;
1423   }
1424 
1425   switch (thr_type) {
1426   case os::java_thread:
1427     // Java threads use ThreadStackSize which default value can be
1428     // changed with the flag -Xss
1429     if (req_stack_size == 0 && JavaThread::stack_size_at_create() > 0) {
1430       // no requested size and we have a more specific default value
1431       stack_size = JavaThread::stack_size_at_create();
1432     }
1433     stack_size = MAX2(stack_size,
1434                       _java_thread_min_stack_allowed);
1435     break;
1436   case os::compiler_thread:
1437     if (req_stack_size == 0 && CompilerThreadStackSize > 0) {
1438       // no requested size and we have a more specific default value
1439       stack_size = (size_t)(CompilerThreadStackSize * K);
1440     }
1441     stack_size = MAX2(stack_size,
1442                       _compiler_thread_min_stack_allowed);
1443     break;
1444   case os::vm_thread:
1445   case os::pgc_thread:
1446   case os::cgc_thread:
1447   case os::watcher_thread:
1448   default:  // presume the unknown thr_type is a VM internal
1449     if (req_stack_size == 0 && VMThreadStackSize > 0) {
1450       // no requested size and we have a more specific default value
1451       stack_size = (size_t)(VMThreadStackSize * K);
1452     }
1453 
1454     stack_size = MAX2(stack_size,
1455                       _vm_internal_thread_min_stack_allowed);
1456     break;
1457   }
1458 
1459   // pthread_attr_setstacksize() may require that the size be rounded up to the OS page size.
1460   // Be careful not to round up to 0. Align down in that case.
1461   if (stack_size <= SIZE_MAX - vm_page_size()) {
1462     stack_size = align_up(stack_size, vm_page_size());
1463   } else {
1464     stack_size = align_down(stack_size, vm_page_size());
1465   }
1466 
1467   return stack_size;
1468 }
1469 
1470 bool os::Posix::is_root(uid_t uid){
1471     return ROOT_UID == uid;
1472 }
1473 
1474 bool os::Posix::matches_effective_uid_or_root(uid_t uid) {
1475     return is_root(uid) || geteuid() == uid;
1476 }
1477 
1478 bool os::Posix::matches_effective_uid_and_gid_or_root(uid_t uid, gid_t gid) {
1479     return is_root(uid) || (geteuid() == uid && getegid() == gid);
1480 }
1481 
1482 Thread* os::ThreadCrashProtection::_protected_thread = NULL;
1483 os::ThreadCrashProtection* os::ThreadCrashProtection::_crash_protection = NULL;
1484 volatile intptr_t os::ThreadCrashProtection::_crash_mux = 0;
1485 
1486 os::ThreadCrashProtection::ThreadCrashProtection() {
1487 }
1488 
1489 /*
1490  * See the caveats for this class in os_posix.hpp
1491  * Protects the callback call so that SIGSEGV / SIGBUS jumps back into this
1492  * method and returns false. If none of the signals are raised, returns true.
1493  * The callback is supposed to provide the method that should be protected.
1494  */
1495 bool os::ThreadCrashProtection::call(os::CrashProtectionCallback& cb) {
1496   sigset_t saved_sig_mask;
1497 
1498   Thread::muxAcquire(&_crash_mux, "CrashProtection");
1499 
1500   _protected_thread = Thread::current_or_null();
1501   assert(_protected_thread != NULL, "Cannot crash protect a NULL thread");
1502 
1503   // we cannot rely on sigsetjmp/siglongjmp to save/restore the signal mask
1504   // since on at least some systems (OS X) siglongjmp will restore the mask
1505   // for the process, not the thread
1506   pthread_sigmask(0, NULL, &saved_sig_mask);
1507   if (sigsetjmp(_jmpbuf, 0) == 0) {
1508     // make sure we can see in the signal handler that we have crash protection
1509     // installed
1510     _crash_protection = this;
1511     cb.call();
1512     // and clear the crash protection
1513     _crash_protection = NULL;
1514     _protected_thread = NULL;
1515     Thread::muxRelease(&_crash_mux);
1516     return true;
1517   }
1518   // this happens when we siglongjmp() back
1519   pthread_sigmask(SIG_SETMASK, &saved_sig_mask, NULL);
1520   _crash_protection = NULL;
1521   _protected_thread = NULL;
1522   Thread::muxRelease(&_crash_mux);
1523   return false;
1524 }
1525 
1526 void os::ThreadCrashProtection::restore() {
1527   assert(_crash_protection != NULL, "must have crash protection");
1528   siglongjmp(_jmpbuf, 1);
1529 }
1530 
1531 void os::ThreadCrashProtection::check_crash_protection(int sig,
1532     Thread* thread) {
1533 
1534   if (thread != NULL &&
1535       thread == _protected_thread &&
1536       _crash_protection != NULL) {
1537 
1538     if (sig == SIGSEGV || sig == SIGBUS) {
1539       _crash_protection->restore();
1540     }
1541   }
1542 }
1543 
1544 // Shared clock/time and other supporting routines for pthread_mutex/cond
1545 // initialization. This is enabled on Solaris but only some of the clock/time
1546 // functionality is actually used there.
1547 
1548 // Shared condattr object for use with relative timed-waits. Will be associated
1549 // with CLOCK_MONOTONIC if available to avoid issues with time-of-day changes,
1550 // but otherwise whatever default is used by the platform - generally the
1551 // time-of-day clock.
1552 static pthread_condattr_t _condAttr[1];
1553 
1554 // Shared mutexattr to explicitly set the type to PTHREAD_MUTEX_NORMAL as not
1555 // all systems (e.g. FreeBSD) map the default to "normal".
1556 static pthread_mutexattr_t _mutexAttr[1];
1557 
1558 // common basic initialization that is always supported
1559 static void pthread_init_common(void) {
1560   int status;
1561   if ((status = pthread_condattr_init(_condAttr)) != 0) {
1562     fatal("pthread_condattr_init: %s", os::strerror(status));
1563   }
1564   if ((status = pthread_mutexattr_init(_mutexAttr)) != 0) {
1565     fatal("pthread_mutexattr_init: %s", os::strerror(status));
1566   }
1567   if ((status = pthread_mutexattr_settype(_mutexAttr, PTHREAD_MUTEX_NORMAL)) != 0) {
1568     fatal("pthread_mutexattr_settype: %s", os::strerror(status));
1569   }
1570   // Solaris has it's own PlatformMutex, distinct from the one for POSIX.
1571   NOT_SOLARIS(os::PlatformMutex::init();)
1572 }
1573 
1574 #ifndef SOLARIS
1575 sigset_t sigs;
1576 struct sigaction sigact[NSIG];
1577 
1578 struct sigaction* os::Posix::get_preinstalled_handler(int sig) {
1579   if (sigismember(&sigs, sig)) {
1580     return &sigact[sig];
1581   }
1582   return NULL;
1583 }
1584 
1585 void os::Posix::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
1586   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
1587   sigact[sig] = oldAct;
1588   sigaddset(&sigs, sig);
1589 }
1590 #endif
1591 
1592 // Not all POSIX types and API's are available on all notionally "posix"
1593 // platforms. If we have build-time support then we will check for actual
1594 // runtime support via dlopen/dlsym lookup. This allows for running on an
1595 // older OS version compared to the build platform. But if there is no
1596 // build time support then there cannot be any runtime support as we do not
1597 // know what the runtime types would be (for example clockid_t might be an
1598 // int or int64_t).
1599 //
1600 #ifdef SUPPORTS_CLOCK_MONOTONIC
1601 
1602 // This means we have clockid_t, clock_gettime et al and CLOCK_MONOTONIC
1603 
1604 int (*os::Posix::_clock_gettime)(clockid_t, struct timespec *) = NULL;
1605 int (*os::Posix::_clock_getres)(clockid_t, struct timespec *) = NULL;
1606 
1607 static int (*_pthread_condattr_setclock)(pthread_condattr_t *, clockid_t) = NULL;
1608 
1609 static bool _use_clock_monotonic_condattr = false;
1610 
1611 // Determine what POSIX API's are present and do appropriate
1612 // configuration.
1613 void os::Posix::init(void) {
1614 
1615   // NOTE: no logging available when this is called. Put logging
1616   // statements in init_2().
1617 
1618   // 1. Check for CLOCK_MONOTONIC support.
1619 
1620   void* handle = NULL;
1621 
1622   // For linux we need librt, for other OS we can find
1623   // this function in regular libc.
1624 #ifdef NEEDS_LIBRT
1625   // We do dlopen's in this particular order due to bug in linux
1626   // dynamic loader (see 6348968) leading to crash on exit.
1627   handle = dlopen("librt.so.1", RTLD_LAZY);
1628   if (handle == NULL) {
1629     handle = dlopen("librt.so", RTLD_LAZY);
1630   }
1631 #endif
1632 
1633   if (handle == NULL) {
1634     handle = RTLD_DEFAULT;
1635   }
1636 
1637   int (*clock_getres_func)(clockid_t, struct timespec*) =
1638     (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1639   int (*clock_gettime_func)(clockid_t, struct timespec*) =
1640     (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1641   if (clock_getres_func != NULL && clock_gettime_func != NULL) {
1642     // We assume that if both clock_gettime and clock_getres support
1643     // CLOCK_MONOTONIC then the OS provides true high-res monotonic clock.
1644     struct timespec res;
1645     struct timespec tp;
1646     if (clock_getres_func(CLOCK_MONOTONIC, &res) == 0 &&
1647         clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1648       // Yes, monotonic clock is supported.
1649       _clock_gettime = clock_gettime_func;
1650       _clock_getres = clock_getres_func;
1651     } else {
1652 #ifdef NEEDS_LIBRT
1653       // Close librt if there is no monotonic clock.
1654       if (handle != RTLD_DEFAULT) {
1655         dlclose(handle);
1656       }
1657 #endif
1658     }
1659   }
1660 
1661   // 2. Check for pthread_condattr_setclock support.
1662 
1663   // libpthread is already loaded.
1664   int (*condattr_setclock_func)(pthread_condattr_t*, clockid_t) =
1665     (int (*)(pthread_condattr_t*, clockid_t))dlsym(RTLD_DEFAULT,
1666                                                    "pthread_condattr_setclock");
1667   if (condattr_setclock_func != NULL) {
1668     _pthread_condattr_setclock = condattr_setclock_func;
1669   }
1670 
1671   // Now do general initialization.
1672 
1673   pthread_init_common();
1674 
1675 #ifndef SOLARIS
1676   int status;
1677   if (_pthread_condattr_setclock != NULL && _clock_gettime != NULL) {
1678     if ((status = _pthread_condattr_setclock(_condAttr, CLOCK_MONOTONIC)) != 0) {
1679       if (status == EINVAL) {
1680         _use_clock_monotonic_condattr = false;
1681         warning("Unable to use monotonic clock with relative timed-waits" \
1682                 " - changes to the time-of-day clock may have adverse affects");
1683       } else {
1684         fatal("pthread_condattr_setclock: %s", os::strerror(status));
1685       }
1686     } else {
1687       _use_clock_monotonic_condattr = true;
1688     }
1689   }
1690 #endif // !SOLARIS
1691 
1692 }
1693 
1694 void os::Posix::init_2(void) {
1695 #ifndef SOLARIS
1696   log_info(os)("Use of CLOCK_MONOTONIC is%s supported",
1697                (_clock_gettime != NULL ? "" : " not"));
1698   log_info(os)("Use of pthread_condattr_setclock is%s supported",
1699                (_pthread_condattr_setclock != NULL ? "" : " not"));
1700   log_info(os)("Relative timed-wait using pthread_cond_timedwait is associated with %s",
1701                _use_clock_monotonic_condattr ? "CLOCK_MONOTONIC" : "the default clock");
1702   sigemptyset(&sigs);
1703 #endif // !SOLARIS
1704 }
1705 
1706 #else // !SUPPORTS_CLOCK_MONOTONIC
1707 
1708 void os::Posix::init(void) {
1709   pthread_init_common();
1710 }
1711 
1712 void os::Posix::init_2(void) {
1713 #ifndef SOLARIS
1714   log_info(os)("Use of CLOCK_MONOTONIC is not supported");
1715   log_info(os)("Use of pthread_condattr_setclock is not supported");
1716   log_info(os)("Relative timed-wait using pthread_cond_timedwait is associated with the default clock");
1717   sigemptyset(&sigs);
1718 #endif // !SOLARIS
1719 }
1720 
1721 #endif // SUPPORTS_CLOCK_MONOTONIC
1722 
1723 // Utility to convert the given timeout to an absolute timespec
1724 // (based on the appropriate clock) to use with pthread_cond_timewait,
1725 // and sem_timedwait().
1726 // The clock queried here must be the clock used to manage the
1727 // timeout of the condition variable or semaphore.
1728 //
1729 // The passed in timeout value is either a relative time in nanoseconds
1730 // or an absolute time in milliseconds. A relative timeout will be
1731 // associated with CLOCK_MONOTONIC if available, unless the real-time clock
1732 // is explicitly requested; otherwise, or if absolute,
1733 // the default time-of-day clock will be used.
1734 
1735 // Given time is a 64-bit value and the time_t used in the timespec is
1736 // sometimes a signed-32-bit value we have to watch for overflow if times
1737 // way in the future are given. Further on Solaris versions
1738 // prior to 10 there is a restriction (see cond_timedwait) that the specified
1739 // number of seconds, in abstime, is less than current_time + 100000000.
1740 // As it will be over 20 years before "now + 100000000" will overflow we can
1741 // ignore overflow and just impose a hard-limit on seconds using the value
1742 // of "now + 100000000". This places a limit on the timeout of about 3.17
1743 // years from "now".
1744 //
1745 #define MAX_SECS 100000000
1746 
1747 // Calculate a new absolute time that is "timeout" nanoseconds from "now".
1748 // "unit" indicates the unit of "now_part_sec" (may be nanos or micros depending
1749 // on which clock API is being used).
1750 static void calc_rel_time(timespec* abstime, jlong timeout, jlong now_sec,
1751                           jlong now_part_sec, jlong unit) {
1752   time_t max_secs = now_sec + MAX_SECS;
1753 
1754   jlong seconds = timeout / NANOUNITS;
1755   timeout %= NANOUNITS; // remaining nanos
1756 
1757   if (seconds >= MAX_SECS) {
1758     // More seconds than we can add, so pin to max_secs.
1759     abstime->tv_sec = max_secs;
1760     abstime->tv_nsec = 0;
1761   } else {
1762     abstime->tv_sec = now_sec  + seconds;
1763     long nanos = (now_part_sec * (NANOUNITS / unit)) + timeout;
1764     if (nanos >= NANOUNITS) { // overflow
1765       abstime->tv_sec += 1;
1766       nanos -= NANOUNITS;
1767     }
1768     abstime->tv_nsec = nanos;
1769   }
1770 }
1771 
1772 // Unpack the given deadline in milliseconds since the epoch, into the given timespec.
1773 // The current time in seconds is also passed in to enforce an upper bound as discussed above.
1774 // This is only used with gettimeofday, when clock_gettime is not available.
1775 static void unpack_abs_time(timespec* abstime, jlong deadline, jlong now_sec) {
1776   time_t max_secs = now_sec + MAX_SECS;
1777 
1778   jlong seconds = deadline / MILLIUNITS;
1779   jlong millis = deadline % MILLIUNITS;
1780 
1781   if (seconds >= max_secs) {
1782     // Absolute seconds exceeds allowed max, so pin to max_secs.
1783     abstime->tv_sec = max_secs;
1784     abstime->tv_nsec = 0;
1785   } else {
1786     abstime->tv_sec = seconds;
1787     abstime->tv_nsec = millis * (NANOUNITS / MILLIUNITS);
1788   }
1789 }
1790 
1791 static jlong millis_to_nanos(jlong millis) {
1792   // We have to watch for overflow when converting millis to nanos,
1793   // but if millis is that large then we will end up limiting to
1794   // MAX_SECS anyway, so just do that here.
1795   if (millis / MILLIUNITS > MAX_SECS) {
1796     millis = jlong(MAX_SECS) * MILLIUNITS;
1797   }
1798   return millis * (NANOUNITS / MILLIUNITS);
1799 }
1800 
1801 static void to_abstime(timespec* abstime, jlong timeout,
1802                        bool isAbsolute, bool isRealtime) {
1803   DEBUG_ONLY(int max_secs = MAX_SECS;)
1804 
1805   if (timeout < 0) {
1806     timeout = 0;
1807   }
1808 
1809 #ifdef SUPPORTS_CLOCK_MONOTONIC
1810 
1811   clockid_t clock = CLOCK_MONOTONIC;
1812   // need to ensure we have a runtime check for clock_gettime support
1813   if (!isAbsolute && os::Posix::supports_monotonic_clock()) {
1814     if (!_use_clock_monotonic_condattr || isRealtime) {
1815       clock = CLOCK_REALTIME;
1816     }
1817     struct timespec now;
1818     int status = os::Posix::clock_gettime(clock, &now);
1819     assert_status(status == 0, status, "clock_gettime");
1820     calc_rel_time(abstime, timeout, now.tv_sec, now.tv_nsec, NANOUNITS);
1821     DEBUG_ONLY(max_secs += now.tv_sec;)
1822   } else {
1823 
1824 #else
1825 
1826   { // Match the block scope.
1827 
1828 #endif // SUPPORTS_CLOCK_MONOTONIC
1829 
1830     // Time-of-day clock is all we can reliably use.
1831     struct timeval now;
1832     int status = gettimeofday(&now, NULL);
1833     assert_status(status == 0, errno, "gettimeofday");
1834     if (isAbsolute) {
1835       unpack_abs_time(abstime, timeout, now.tv_sec);
1836     } else {
1837       calc_rel_time(abstime, timeout, now.tv_sec, now.tv_usec, MICROUNITS);
1838     }
1839     DEBUG_ONLY(max_secs += now.tv_sec;)
1840   }
1841 
1842   assert(abstime->tv_sec >= 0, "tv_sec < 0");
1843   assert(abstime->tv_sec <= max_secs, "tv_sec > max_secs");
1844   assert(abstime->tv_nsec >= 0, "tv_nsec < 0");
1845   assert(abstime->tv_nsec < NANOUNITS, "tv_nsec >= NANOUNITS");
1846 }
1847 
1848 // Create an absolute time 'millis' milliseconds in the future, using the
1849 // real-time (time-of-day) clock. Used by PosixSemaphore.
1850 void os::Posix::to_RTC_abstime(timespec* abstime, int64_t millis) {
1851   to_abstime(abstime, millis_to_nanos(millis),
1852              false /* not absolute */,
1853              true  /* use real-time clock */);
1854 }
1855 
1856 // Shared pthread_mutex/cond based PlatformEvent implementation.
1857 // Not currently usable by Solaris.
1858 
1859 #ifndef SOLARIS
1860 
1861 // PlatformEvent
1862 //
1863 // Assumption:
1864 //    Only one parker can exist on an event, which is why we allocate
1865 //    them per-thread. Multiple unparkers can coexist.
1866 //
1867 // _event serves as a restricted-range semaphore.
1868 //   -1 : thread is blocked, i.e. there is a waiter
1869 //    0 : neutral: thread is running or ready,
1870 //        could have been signaled after a wait started
1871 //    1 : signaled - thread is running or ready
1872 //
1873 //    Having three states allows for some detection of bad usage - see
1874 //    comments on unpark().
1875 
1876 os::PlatformEvent::PlatformEvent() {
1877   int status = pthread_cond_init(_cond, _condAttr);
1878   assert_status(status == 0, status, "cond_init");
1879   status = pthread_mutex_init(_mutex, _mutexAttr);
1880   assert_status(status == 0, status, "mutex_init");
1881   _event   = 0;
1882   _nParked = 0;
1883 }
1884 
1885 void os::PlatformEvent::park() {       // AKA "down()"
1886   // Transitions for _event:
1887   //   -1 => -1 : illegal
1888   //    1 =>  0 : pass - return immediately
1889   //    0 => -1 : block; then set _event to 0 before returning
1890 
1891   // Invariant: Only the thread associated with the PlatformEvent
1892   // may call park().
1893   assert(_nParked == 0, "invariant");
1894 
1895   int v;
1896 
1897   // atomically decrement _event
1898   for (;;) {
1899     v = _event;
1900     if (Atomic::cmpxchg(v - 1, &_event, v) == v) break;
1901   }
1902   guarantee(v >= 0, "invariant");
1903 
1904   if (v == 0) { // Do this the hard way by blocking ...
1905     int status = pthread_mutex_lock(_mutex);
1906     assert_status(status == 0, status, "mutex_lock");
1907     guarantee(_nParked == 0, "invariant");
1908     ++_nParked;
1909     while (_event < 0) {
1910       // OS-level "spurious wakeups" are ignored
1911       status = pthread_cond_wait(_cond, _mutex);
1912       assert_status(status == 0, status, "cond_wait");
1913     }
1914     --_nParked;
1915 
1916     _event = 0;
1917     status = pthread_mutex_unlock(_mutex);
1918     assert_status(status == 0, status, "mutex_unlock");
1919     // Paranoia to ensure our locked and lock-free paths interact
1920     // correctly with each other.
1921     OrderAccess::fence();
1922   }
1923   guarantee(_event >= 0, "invariant");
1924 }
1925 
1926 int os::PlatformEvent::park(jlong millis) {
1927   // Transitions for _event:
1928   //   -1 => -1 : illegal
1929   //    1 =>  0 : pass - return immediately
1930   //    0 => -1 : block; then set _event to 0 before returning
1931 
1932   // Invariant: Only the thread associated with the Event/PlatformEvent
1933   // may call park().
1934   assert(_nParked == 0, "invariant");
1935 
1936   int v;
1937   // atomically decrement _event
1938   for (;;) {
1939     v = _event;
1940     if (Atomic::cmpxchg(v - 1, &_event, v) == v) break;
1941   }
1942   guarantee(v >= 0, "invariant");
1943 
1944   if (v == 0) { // Do this the hard way by blocking ...
1945     struct timespec abst;
1946     to_abstime(&abst, millis_to_nanos(millis), false, false);
1947 
1948     int ret = OS_TIMEOUT;
1949     int status = pthread_mutex_lock(_mutex);
1950     assert_status(status == 0, status, "mutex_lock");
1951     guarantee(_nParked == 0, "invariant");
1952     ++_nParked;
1953 
1954     while (_event < 0) {
1955       status = pthread_cond_timedwait(_cond, _mutex, &abst);
1956       assert_status(status == 0 || status == ETIMEDOUT,
1957                     status, "cond_timedwait");
1958       // OS-level "spurious wakeups" are ignored unless the archaic
1959       // FilterSpuriousWakeups is set false. That flag should be obsoleted.
1960       if (!FilterSpuriousWakeups) break;
1961       if (status == ETIMEDOUT) break;
1962     }
1963     --_nParked;
1964 
1965     if (_event >= 0) {
1966       ret = OS_OK;
1967     }
1968 
1969     _event = 0;
1970     status = pthread_mutex_unlock(_mutex);
1971     assert_status(status == 0, status, "mutex_unlock");
1972     // Paranoia to ensure our locked and lock-free paths interact
1973     // correctly with each other.
1974     OrderAccess::fence();
1975     return ret;
1976   }
1977   return OS_OK;
1978 }
1979 
1980 void os::PlatformEvent::unpark() {
1981   // Transitions for _event:
1982   //    0 => 1 : just return
1983   //    1 => 1 : just return
1984   //   -1 => either 0 or 1; must signal target thread
1985   //         That is, we can safely transition _event from -1 to either
1986   //         0 or 1.
1987   // See also: "Semaphores in Plan 9" by Mullender & Cox
1988   //
1989   // Note: Forcing a transition from "-1" to "1" on an unpark() means
1990   // that it will take two back-to-back park() calls for the owning
1991   // thread to block. This has the benefit of forcing a spurious return
1992   // from the first park() call after an unpark() call which will help
1993   // shake out uses of park() and unpark() without checking state conditions
1994   // properly. This spurious return doesn't manifest itself in any user code
1995   // but only in the correctly written condition checking loops of ObjectMonitor,
1996   // Mutex/Monitor, Thread::muxAcquire and JavaThread::sleep
1997 
1998   if (Atomic::xchg(1, &_event) >= 0) return;
1999 
2000   int status = pthread_mutex_lock(_mutex);
2001   assert_status(status == 0, status, "mutex_lock");
2002   int anyWaiters = _nParked;
2003   assert(anyWaiters == 0 || anyWaiters == 1, "invariant");
2004   status = pthread_mutex_unlock(_mutex);
2005   assert_status(status == 0, status, "mutex_unlock");
2006 
2007   // Note that we signal() *after* dropping the lock for "immortal" Events.
2008   // This is safe and avoids a common class of futile wakeups.  In rare
2009   // circumstances this can cause a thread to return prematurely from
2010   // cond_{timed}wait() but the spurious wakeup is benign and the victim
2011   // will simply re-test the condition and re-park itself.
2012   // This provides particular benefit if the underlying platform does not
2013   // provide wait morphing.
2014 
2015   if (anyWaiters != 0) {
2016     status = pthread_cond_signal(_cond);
2017     assert_status(status == 0, status, "cond_signal");
2018   }
2019 }
2020 
2021 // JSR166 support
2022 
2023  os::PlatformParker::PlatformParker() {
2024   int status;
2025   status = pthread_cond_init(&_cond[REL_INDEX], _condAttr);
2026   assert_status(status == 0, status, "cond_init rel");
2027   status = pthread_cond_init(&_cond[ABS_INDEX], NULL);
2028   assert_status(status == 0, status, "cond_init abs");
2029   status = pthread_mutex_init(_mutex, _mutexAttr);
2030   assert_status(status == 0, status, "mutex_init");
2031   _cur_index = -1; // mark as unused
2032 }
2033 
2034 // Parker::park decrements count if > 0, else does a condvar wait.  Unpark
2035 // sets count to 1 and signals condvar.  Only one thread ever waits
2036 // on the condvar. Contention seen when trying to park implies that someone
2037 // is unparking you, so don't wait. And spurious returns are fine, so there
2038 // is no need to track notifications.
2039 
2040 void Parker::park(bool isAbsolute, jlong time) {
2041 
2042   // Optional fast-path check:
2043   // Return immediately if a permit is available.
2044   // We depend on Atomic::xchg() having full barrier semantics
2045   // since we are doing a lock-free update to _counter.
2046   if (Atomic::xchg(0, &_counter) > 0) return;
2047 
2048   Thread* thread = Thread::current();
2049   assert(thread->is_Java_thread(), "Must be JavaThread");
2050   JavaThread *jt = (JavaThread *)thread;
2051 
2052   // Optional optimization -- avoid state transitions if there's
2053   // an interrupt pending.
2054   if (jt->is_interrupted(false)) {
2055     return;
2056   }
2057 
2058   // Next, demultiplex/decode time arguments
2059   struct timespec absTime;
2060   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
2061     return;
2062   }
2063   if (time > 0) {
2064     to_abstime(&absTime, time, isAbsolute, false);
2065   }
2066 
2067   // Enter safepoint region
2068   // Beware of deadlocks such as 6317397.
2069   // The per-thread Parker:: mutex is a classic leaf-lock.
2070   // In particular a thread must never block on the Threads_lock while
2071   // holding the Parker:: mutex.  If safepoints are pending both the
2072   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
2073   ThreadBlockInVM tbivm(jt);
2074 
2075   // Don't wait if cannot get lock since interference arises from
2076   // unparking. Also re-check interrupt before trying wait.
2077   if (jt->is_interrupted(false) ||
2078       pthread_mutex_trylock(_mutex) != 0) {
2079     return;
2080   }
2081 
2082   int status;
2083   if (_counter > 0)  { // no wait needed
2084     _counter = 0;
2085     status = pthread_mutex_unlock(_mutex);
2086     assert_status(status == 0, status, "invariant");
2087     // Paranoia to ensure our locked and lock-free paths interact
2088     // correctly with each other and Java-level accesses.
2089     OrderAccess::fence();
2090     return;
2091   }
2092 
2093   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2094   jt->set_suspend_equivalent();
2095   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2096 
2097   assert(_cur_index == -1, "invariant");
2098   if (time == 0) {
2099     _cur_index = REL_INDEX; // arbitrary choice when not timed
2100     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
2101     assert_status(status == 0, status, "cond_timedwait");
2102   }
2103   else {
2104     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
2105     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
2106     assert_status(status == 0 || status == ETIMEDOUT,
2107                   status, "cond_timedwait");
2108   }
2109   _cur_index = -1;
2110 
2111   _counter = 0;
2112   status = pthread_mutex_unlock(_mutex);
2113   assert_status(status == 0, status, "invariant");
2114   // Paranoia to ensure our locked and lock-free paths interact
2115   // correctly with each other and Java-level accesses.
2116   OrderAccess::fence();
2117 
2118   // If externally suspended while waiting, re-suspend
2119   if (jt->handle_special_suspend_equivalent_condition()) {
2120     jt->java_suspend_self();
2121   }
2122 }
2123 
2124 void Parker::unpark() {
2125   int status = pthread_mutex_lock(_mutex);
2126   assert_status(status == 0, status, "invariant");
2127   const int s = _counter;
2128   _counter = 1;
2129   // must capture correct index before unlocking
2130   int index = _cur_index;
2131   status = pthread_mutex_unlock(_mutex);
2132   assert_status(status == 0, status, "invariant");
2133 
2134   // Note that we signal() *after* dropping the lock for "immortal" Events.
2135   // This is safe and avoids a common class of futile wakeups.  In rare
2136   // circumstances this can cause a thread to return prematurely from
2137   // cond_{timed}wait() but the spurious wakeup is benign and the victim
2138   // will simply re-test the condition and re-park itself.
2139   // This provides particular benefit if the underlying platform does not
2140   // provide wait morphing.
2141 
2142   if (s < 1 && index != -1) {
2143     // thread is definitely parked
2144     status = pthread_cond_signal(&_cond[index]);
2145     assert_status(status == 0, status, "invariant");
2146   }
2147 }
2148 
2149 // Platform Mutex/Monitor implementation
2150 
2151 #if PLATFORM_MONITOR_IMPL_INDIRECT
2152 
2153 os::PlatformMutex::Mutex::Mutex() : _next(NULL) {
2154   int status = pthread_mutex_init(&_mutex, _mutexAttr);
2155   assert_status(status == 0, status, "mutex_init");
2156 }
2157 
2158 os::PlatformMutex::Mutex::~Mutex() {
2159   int status = pthread_mutex_destroy(&_mutex);
2160   assert_status(status == 0, status, "mutex_destroy");
2161 }
2162 
2163 pthread_mutex_t os::PlatformMutex::_freelist_lock;
2164 os::PlatformMutex::Mutex* os::PlatformMutex::_mutex_freelist = NULL;
2165 
2166 void os::PlatformMutex::init() {
2167   int status = pthread_mutex_init(&_freelist_lock, _mutexAttr);
2168   assert_status(status == 0, status, "freelist lock init");
2169 }
2170 
2171 struct os::PlatformMutex::WithFreeListLocked : public StackObj {
2172   WithFreeListLocked() {
2173     int status = pthread_mutex_lock(&_freelist_lock);
2174     assert_status(status == 0, status, "freelist lock");
2175   }
2176 
2177   ~WithFreeListLocked() {
2178     int status = pthread_mutex_unlock(&_freelist_lock);
2179     assert_status(status == 0, status, "freelist unlock");
2180   }
2181 };
2182 
2183 os::PlatformMutex::PlatformMutex() {
2184   {
2185     WithFreeListLocked wfl;
2186     _impl = _mutex_freelist;
2187     if (_impl != NULL) {
2188       _mutex_freelist = _impl->_next;
2189       _impl->_next = NULL;
2190       return;
2191     }
2192   }
2193   _impl = new Mutex();
2194 }
2195 
2196 os::PlatformMutex::~PlatformMutex() {
2197   WithFreeListLocked wfl;
2198   assert(_impl->_next == NULL, "invariant");
2199   _impl->_next = _mutex_freelist;
2200   _mutex_freelist = _impl;
2201 }
2202 
2203 os::PlatformMonitor::Cond::Cond() : _next(NULL) {
2204   int status = pthread_cond_init(&_cond, _condAttr);
2205   assert_status(status == 0, status, "cond_init");
2206 }
2207 
2208 os::PlatformMonitor::Cond::~Cond() {
2209   int status = pthread_cond_destroy(&_cond);
2210   assert_status(status == 0, status, "cond_destroy");
2211 }
2212 
2213 os::PlatformMonitor::Cond* os::PlatformMonitor::_cond_freelist = NULL;
2214 
2215 os::PlatformMonitor::PlatformMonitor() {
2216   {
2217     WithFreeListLocked wfl;
2218     _impl = _cond_freelist;
2219     if (_impl != NULL) {
2220       _cond_freelist = _impl->_next;
2221       _impl->_next = NULL;
2222       return;
2223     }
2224   }
2225   _impl = new Cond();
2226 }
2227 
2228 os::PlatformMonitor::~PlatformMonitor() {
2229   WithFreeListLocked wfl;
2230   assert(_impl->_next == NULL, "invariant");
2231   _impl->_next = _cond_freelist;
2232   _cond_freelist = _impl;
2233 }
2234 
2235 #else
2236 
2237 os::PlatformMutex::PlatformMutex() {
2238   int status = pthread_mutex_init(&_mutex, _mutexAttr);
2239   assert_status(status == 0, status, "mutex_init");
2240 }
2241 
2242 os::PlatformMutex::~PlatformMutex() {
2243   int status = pthread_mutex_destroy(&_mutex);
2244   assert_status(status == 0, status, "mutex_destroy");
2245 }
2246 
2247 os::PlatformMonitor::PlatformMonitor() {
2248   int status = pthread_cond_init(&_cond, _condAttr);
2249   assert_status(status == 0, status, "cond_init");
2250 }
2251 
2252 os::PlatformMonitor::~PlatformMonitor() {
2253   int status = pthread_cond_destroy(&_cond);
2254   assert_status(status == 0, status, "cond_destroy");
2255 }
2256 
2257 #endif // PLATFORM_MONITOR_IMPL_INDIRECT
2258 
2259 // Must already be locked
2260 int os::PlatformMonitor::wait(jlong millis) {
2261   assert(millis >= 0, "negative timeout");
2262   if (millis > 0) {
2263     struct timespec abst;
2264     // We have to watch for overflow when converting millis to nanos,
2265     // but if millis is that large then we will end up limiting to
2266     // MAX_SECS anyway, so just do that here.
2267     if (millis / MILLIUNITS > MAX_SECS) {
2268       millis = jlong(MAX_SECS) * MILLIUNITS;
2269     }
2270     to_abstime(&abst, millis * (NANOUNITS / MILLIUNITS), false, false);
2271 
2272     int ret = OS_TIMEOUT;
2273     int status = pthread_cond_timedwait(cond(), mutex(), &abst);
2274     assert_status(status == 0 || status == ETIMEDOUT,
2275                   status, "cond_timedwait");
2276     if (status == 0) {
2277       ret = OS_OK;
2278     }
2279     return ret;
2280   } else {
2281     int status = pthread_cond_wait(cond(), mutex());
2282     assert_status(status == 0, status, "cond_wait");
2283     return OS_OK;
2284   }
2285 }
2286 
2287 #endif // !SOLARIS