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