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 #include <utmpx.h>
51
52 // Todo: provide a os::get_max_process_id() or similar. Number of processes
53 // may have been configured, can be read more accurately from proc fs etc.
54 #ifndef MAX_PID
55 #define MAX_PID INT_MAX
56 #endif
57 #define IS_VALID_PID(p) (p > 0 && p < MAX_PID)
58
59 #define ROOT_UID 0
60
61 #ifndef MAP_ANONYMOUS
62 #define MAP_ANONYMOUS MAP_ANON
63 #endif
64
65 #define check_with_errno(check_type, cond, msg) \
66 do { \
67 int err = errno; \
68 check_type(cond, "%s; error='%s' (errno=%s)", msg, os::strerror(err), \
69 os::errno_name(err)); \
70 } while (false)
71
72 #define assert_with_errno(cond, msg) check_with_errno(assert, cond, msg)
73 #define guarantee_with_errno(cond, msg) check_with_errno(guarantee, cond, msg)
74
75 // Check core dump limit and report possible place where core can be found
76 void os::check_dump_limit(char* buffer, size_t bufferSize) {
77 if (!FLAG_IS_DEFAULT(CreateCoredumpOnCrash) && !CreateCoredumpOnCrash) {
78 jio_snprintf(buffer, bufferSize, "CreateCoredumpOnCrash is disabled from command line");
79 VMError::record_coredump_status(buffer, false);
80 return;
81 }
82
83 int n;
84 struct rlimit rlim;
85 bool success;
86
87 char core_path[PATH_MAX];
88 n = get_core_path(core_path, PATH_MAX);
89
90 if (n <= 0) {
91 jio_snprintf(buffer, bufferSize, "core.%d (may not exist)", current_process_id());
92 success = true;
93 #ifdef LINUX
94 } else if (core_path[0] == '"') { // redirect to user process
95 jio_snprintf(buffer, bufferSize, "Core dumps may be processed with %s", core_path);
96 success = true;
97 #endif
98 } else if (getrlimit(RLIMIT_CORE, &rlim) != 0) {
99 jio_snprintf(buffer, bufferSize, "%s (may not exist)", core_path);
100 success = true;
101 } else {
102 switch(rlim.rlim_cur) {
103 case RLIM_INFINITY:
104 jio_snprintf(buffer, bufferSize, "%s", core_path);
105 success = true;
106 break;
107 case 0:
108 jio_snprintf(buffer, bufferSize, "Core dumps have been disabled. To enable core dumping, try \"ulimit -c unlimited\" before starting Java again");
109 success = false;
110 break;
111 default:
112 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);
113 success = true;
114 break;
115 }
116 }
117
118 VMError::record_coredump_status(buffer, success);
119 }
120
121 int os::get_native_stack(address* stack, int frames, int toSkip) {
122 int frame_idx = 0;
123 int num_of_frames; // number of frames captured
124 frame fr = os::current_frame();
125 while (fr.pc() && frame_idx < frames) {
126 if (toSkip > 0) {
127 toSkip --;
128 } else {
129 stack[frame_idx ++] = fr.pc();
130 }
131 if (fr.fp() == NULL || fr.cb() != NULL ||
132 fr.sender_pc() == NULL || os::is_first_C_frame(&fr)) break;
133
134 if (fr.sender_pc() && !os::is_first_C_frame(&fr)) {
135 fr = os::get_sender_for_C_frame(&fr);
136 } else {
137 break;
138 }
139 }
140 num_of_frames = frame_idx;
141 for (; frame_idx < frames; frame_idx ++) {
142 stack[frame_idx] = NULL;
143 }
144
145 return num_of_frames;
146 }
147
148
149 bool os::unsetenv(const char* name) {
150 assert(name != NULL, "Null pointer");
151 return (::unsetenv(name) == 0);
152 }
153
154 int os::get_last_error() {
155 return errno;
156 }
157
158 size_t os::lasterror(char *buf, size_t len) {
159 if (errno == 0) return 0;
160
161 const char *s = os::strerror(errno);
162 size_t n = ::strlen(s);
163 if (n >= len) {
164 n = len - 1;
165 }
166 ::strncpy(buf, s, n);
167 buf[n] = '\0';
168 return n;
169 }
170
171 void os::wait_for_keypress_at_exit(void) {
172 // don't do anything on posix platforms
173 return;
174 }
175
176 int os::create_file_for_heap(const char* dir) {
177
178 const char name_template[] = "/jvmheap.XXXXXX";
179
180 size_t fullname_len = strlen(dir) + strlen(name_template);
181 char *fullname = (char*)os::malloc(fullname_len + 1, mtInternal);
182 if (fullname == NULL) {
183 vm_exit_during_initialization(err_msg("Malloc failed during creation of backing file for heap (%s)", os::strerror(errno)));
184 return -1;
185 }
186 int n = snprintf(fullname, fullname_len + 1, "%s%s", dir, name_template);
187 assert((size_t)n == fullname_len, "Unexpected number of characters in string");
188
189 os::native_path(fullname);
190
191 // set the file creation mask.
192 mode_t file_mode = S_IRUSR | S_IWUSR;
193
194 // create a new file.
195 int fd = mkstemp(fullname);
196
197 if (fd < 0) {
198 warning("Could not create file for heap with template %s", fullname);
199 os::free(fullname);
200 return -1;
201 }
202
203 // delete the name from the filesystem. When 'fd' is closed, the file (and space) will be deleted.
204 int ret = unlink(fullname);
205 assert_with_errno(ret == 0, "unlink returned error");
206
207 os::free(fullname);
208 return fd;
209 }
210
211 static char* reserve_mmapped_memory(size_t bytes, char* requested_addr) {
212 char * addr;
213 int flags = MAP_PRIVATE NOT_AIX( | MAP_NORESERVE ) | MAP_ANONYMOUS;
214 if (requested_addr != NULL) {
215 assert((uintptr_t)requested_addr % os::vm_page_size() == 0, "Requested address should be aligned to OS page size");
216 flags |= MAP_FIXED;
217 }
218
219 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
220 // touch an uncommitted page. Otherwise, the read/write might
221 // succeed if we have enough swap space to back the physical page.
222 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
223 flags, -1, 0);
224
225 if (addr != MAP_FAILED) {
226 MemTracker::record_virtual_memory_reserve((address)addr, bytes, CALLER_PC);
227 return addr;
228 }
229 return NULL;
230 }
231
232 static int util_posix_fallocate(int fd, off_t offset, off_t len) {
233 #ifdef __APPLE__
234 fstore_t store = { F_ALLOCATECONTIG, F_PEOFPOSMODE, 0, len };
235 // First we try to get a continuous chunk of disk space
236 int ret = fcntl(fd, F_PREALLOCATE, &store);
237 if (ret == -1) {
238 // Maybe we are too fragmented, try to allocate non-continuous range
239 store.fst_flags = F_ALLOCATEALL;
240 ret = fcntl(fd, F_PREALLOCATE, &store);
241 }
242 if(ret != -1) {
243 return ftruncate(fd, len);
244 }
245 return -1;
246 #else
247 return posix_fallocate(fd, offset, len);
248 #endif
249 }
250
251 // Map the given address range to the provided file descriptor.
252 char* os::map_memory_to_file(char* base, size_t size, int fd) {
253 assert(fd != -1, "File descriptor is not valid");
254
255 // allocate space for the file
256 int ret = util_posix_fallocate(fd, 0, (off_t)size);
257 if (ret != 0) {
258 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory. error(%d)", ret));
259 return NULL;
260 }
261
262 int prot = PROT_READ | PROT_WRITE;
263 int flags = MAP_SHARED;
264 if (base != NULL) {
265 flags |= MAP_FIXED;
266 }
267 char* addr = (char*)mmap(base, size, prot, flags, fd, 0);
268
269 if (addr == MAP_FAILED) {
270 warning("Failed mmap to file. (%s)", os::strerror(errno));
271 return NULL;
272 }
273 if (base != NULL && addr != base) {
274 if (!os::release_memory(addr, size)) {
275 warning("Could not release memory on unsuccessful file mapping");
276 }
277 return NULL;
278 }
279 return addr;
280 }
281
282 char* os::replace_existing_mapping_with_file_mapping(char* base, size_t size, int fd) {
283 assert(fd != -1, "File descriptor is not valid");
284 assert(base != NULL, "Base cannot be NULL");
285
286 return map_memory_to_file(base, size, fd);
287 }
288
289 // Multiple threads can race in this code, and can remap over each other with MAP_FIXED,
290 // so on posix, unmap the section at the start and at the end of the chunk that we mapped
291 // rather than unmapping and remapping the whole chunk to get requested alignment.
292 char* os::reserve_memory_aligned(size_t size, size_t alignment, int file_desc) {
293 assert((alignment & (os::vm_allocation_granularity() - 1)) == 0,
294 "Alignment must be a multiple of allocation granularity (page size)");
295 assert((size & (alignment -1)) == 0, "size must be 'alignment' aligned");
296
297 size_t extra_size = size + alignment;
298 assert(extra_size >= size, "overflow, size is too large to allow alignment");
299
300 char* extra_base;
301 if (file_desc != -1) {
302 // For file mapping, we do not call os:reserve_memory(extra_size, NULL, alignment, file_desc) because
303 // we need to deal with shrinking of the file space later when we release extra memory after alignment.
304 // We also cannot called os:reserve_memory() with file_desc set to -1 because on aix we might get SHM memory.
305 // So here to call a helper function while reserve memory for us. After we have a aligned base,
306 // we will replace anonymous mapping with file mapping.
307 extra_base = reserve_mmapped_memory(extra_size, NULL);
308 if (extra_base != NULL) {
309 MemTracker::record_virtual_memory_reserve((address)extra_base, extra_size, CALLER_PC);
310 }
311 } else {
312 extra_base = os::reserve_memory(extra_size, NULL, alignment);
313 }
314
315 if (extra_base == NULL) {
316 return NULL;
317 }
318
319 // Do manual alignment
320 char* aligned_base = align_up(extra_base, alignment);
321
322 // [ | | ]
323 // ^ extra_base
324 // ^ extra_base + begin_offset == aligned_base
325 // extra_base + begin_offset + size ^
326 // extra_base + extra_size ^
327 // |<>| == begin_offset
328 // end_offset == |<>|
329 size_t begin_offset = aligned_base - extra_base;
330 size_t end_offset = (extra_base + extra_size) - (aligned_base + size);
331
332 if (begin_offset > 0) {
333 os::release_memory(extra_base, begin_offset);
334 }
335
336 if (end_offset > 0) {
337 os::release_memory(extra_base + begin_offset + size, end_offset);
338 }
339
340 if (file_desc != -1) {
341 // After we have an aligned address, we can replace anonymous mapping with file mapping
342 if (replace_existing_mapping_with_file_mapping(aligned_base, size, file_desc) == NULL) {
343 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
344 }
345 MemTracker::record_virtual_memory_commit((address)aligned_base, size, CALLER_PC);
346 }
347 return aligned_base;
348 }
349
350 int os::vsnprintf(char* buf, size_t len, const char* fmt, va_list args) {
351 // All supported POSIX platforms provide C99 semantics.
352 int result = ::vsnprintf(buf, len, fmt, args);
353 // If an encoding error occurred (result < 0) then it's not clear
354 // whether the buffer is NUL terminated, so ensure it is.
355 if ((result < 0) && (len > 0)) {
356 buf[len - 1] = '\0';
357 }
358 return result;
359 }
360
361 int os::get_fileno(FILE* fp) {
362 return NOT_AIX(::)fileno(fp);
363 }
364
365 struct tm* os::gmtime_pd(const time_t* clock, struct tm* res) {
366 return gmtime_r(clock, res);
367 }
368
369 void os::Posix::print_load_average(outputStream* st) {
370 st->print("load average:");
371 double loadavg[3];
372 int res = os::loadavg(loadavg, 3);
373 if (res != -1) {
374 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
375 } else {
376 st->print(" Unavailable");
377 }
378 st->cr();
379 }
380
381 // boot/uptime information;
382 // unfortunately it does not work on macOS because the utx chain has no entry
383 // for reboot at least on my test machines
384 void os::Posix::print_uptime_info(outputStream* st) {
385 int bootsec = -1;
386 int currsec = time(NULL);
387 struct utmpx* ent;
388 setutxent();
389 while ((ent = getutxent())) {
390 if (!strcmp("system boot", ent->ut_line)) {
391 bootsec = ent->ut_tv.tv_sec;
392 break;
393 }
394 }
395
396 if (bootsec != -1) {
397 st->print_cr("OS uptime (in days): %.2lf", (double) (currsec-bootsec)/(60*60*24));
398 }
399 }
400
401
402 void os::Posix::print_rlimit_info(outputStream* st) {
403 st->print("rlimit:");
404 struct rlimit rlim;
405
406 st->print(" STACK ");
407 getrlimit(RLIMIT_STACK, &rlim);
408 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
409 else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
410
411 st->print(", CORE ");
412 getrlimit(RLIMIT_CORE, &rlim);
413 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
414 else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
415
416 // Isn't there on solaris
417 #if defined(AIX)
418 st->print(", NPROC ");
419 st->print("%d", sysconf(_SC_CHILD_MAX));
420 #elif !defined(SOLARIS)
421 st->print(", NPROC ");
422 getrlimit(RLIMIT_NPROC, &rlim);
423 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
424 else st->print(UINT64_FORMAT, uint64_t(rlim.rlim_cur));
425 #endif
426
427 st->print(", NOFILE ");
428 getrlimit(RLIMIT_NOFILE, &rlim);
429 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
430 else st->print(UINT64_FORMAT, uint64_t(rlim.rlim_cur));
431
432 st->print(", AS ");
433 getrlimit(RLIMIT_AS, &rlim);
434 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
435 else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
436
437 st->print(", DATA ");
438 getrlimit(RLIMIT_DATA, &rlim);
439 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
440 else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
441
442 st->print(", FSIZE ");
443 getrlimit(RLIMIT_FSIZE, &rlim);
444 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
445 else st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024);
446
447 st->cr();
448 }
449
450 void os::Posix::print_uname_info(outputStream* st) {
451 // kernel
452 st->print("uname:");
453 struct utsname name;
454 uname(&name);
455 st->print("%s ", name.sysname);
456 #ifdef ASSERT
457 st->print("%s ", name.nodename);
458 #endif
459 st->print("%s ", name.release);
460 st->print("%s ", name.version);
461 st->print("%s", name.machine);
462 st->cr();
463 }
464
465 void os::Posix::print_umask(outputStream* st, mode_t umsk) {
466 st->print((umsk & S_IRUSR) ? "r" : "-");
467 st->print((umsk & S_IWUSR) ? "w" : "-");
468 st->print((umsk & S_IXUSR) ? "x" : "-");
469 st->print((umsk & S_IRGRP) ? "r" : "-");
470 st->print((umsk & S_IWGRP) ? "w" : "-");
471 st->print((umsk & S_IXGRP) ? "x" : "-");
472 st->print((umsk & S_IROTH) ? "r" : "-");
473 st->print((umsk & S_IWOTH) ? "w" : "-");
474 st->print((umsk & S_IXOTH) ? "x" : "-");
475 }
476
477 void os::Posix::print_user_info(outputStream* st) {
478 unsigned id = (unsigned) ::getuid();
479 st->print("uid : %u ", id);
480 id = (unsigned) ::geteuid();
481 st->print("euid : %u ", id);
482 id = (unsigned) ::getgid();
483 st->print("gid : %u ", id);
484 id = (unsigned) ::getegid();
485 st->print_cr("egid : %u", id);
486 st->cr();
487
488 mode_t umsk = ::umask(0);
489 ::umask(umsk);
490 st->print("umask: %04o (", (unsigned) umsk);
491 print_umask(st, umsk);
492 st->print_cr(")");
493 st->cr();
494 }
495
496
497 bool os::get_host_name(char* buf, size_t buflen) {
498 struct utsname name;
499 uname(&name);
500 jio_snprintf(buf, buflen, "%s", name.nodename);
501 return true;
502 }
503
504 bool os::has_allocatable_memory_limit(julong* limit) {
505 struct rlimit rlim;
506 int getrlimit_res = getrlimit(RLIMIT_AS, &rlim);
507 // if there was an error when calling getrlimit, assume that there is no limitation
508 // on virtual memory.
509 bool result;
510 if ((getrlimit_res != 0) || (rlim.rlim_cur == RLIM_INFINITY)) {
511 result = false;
512 } else {
513 *limit = (julong)rlim.rlim_cur;
514 result = true;
515 }
516 #ifdef _LP64
517 return result;
518 #else
519 // arbitrary virtual space limit for 32 bit Unices found by testing. If
520 // getrlimit above returned a limit, bound it with this limit. Otherwise
521 // directly use it.
522 const julong max_virtual_limit = (julong)3800*M;
523 if (result) {
524 *limit = MIN2(*limit, max_virtual_limit);
525 } else {
526 *limit = max_virtual_limit;
527 }
528
529 // bound by actually allocatable memory. The algorithm uses two bounds, an
530 // upper and a lower limit. The upper limit is the current highest amount of
531 // memory that could not be allocated, the lower limit is the current highest
532 // amount of memory that could be allocated.
533 // The algorithm iteratively refines the result by halving the difference
534 // between these limits, updating either the upper limit (if that value could
535 // not be allocated) or the lower limit (if the that value could be allocated)
536 // until the difference between these limits is "small".
537
538 // the minimum amount of memory we care about allocating.
539 const julong min_allocation_size = M;
540
541 julong upper_limit = *limit;
542
543 // first check a few trivial cases
544 if (is_allocatable(upper_limit) || (upper_limit <= min_allocation_size)) {
545 *limit = upper_limit;
546 } else if (!is_allocatable(min_allocation_size)) {
547 // we found that not even min_allocation_size is allocatable. Return it
548 // anyway. There is no point to search for a better value any more.
549 *limit = min_allocation_size;
550 } else {
551 // perform the binary search.
552 julong lower_limit = min_allocation_size;
553 while ((upper_limit - lower_limit) > min_allocation_size) {
554 julong temp_limit = ((upper_limit - lower_limit) / 2) + lower_limit;
555 temp_limit = align_down(temp_limit, min_allocation_size);
556 if (is_allocatable(temp_limit)) {
557 lower_limit = temp_limit;
558 } else {
559 upper_limit = temp_limit;
560 }
561 }
562 *limit = lower_limit;
563 }
564 return true;
565 #endif
566 }
567
568 const char* os::get_current_directory(char *buf, size_t buflen) {
569 return getcwd(buf, buflen);
570 }
571
572 FILE* os::open(int fd, const char* mode) {
573 return ::fdopen(fd, mode);
574 }
575
576 ssize_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
577 return ::pread(fd, buf, nBytes, offset);
578 }
579
580 void os::flockfile(FILE* fp) {
581 ::flockfile(fp);
582 }
583
584 void os::funlockfile(FILE* fp) {
585 ::funlockfile(fp);
586 }
587
588 DIR* os::opendir(const char* dirname) {
589 assert(dirname != NULL, "just checking");
590 return ::opendir(dirname);
591 }
592
593 struct dirent* os::readdir(DIR* dirp) {
594 assert(dirp != NULL, "just checking");
595 return ::readdir(dirp);
596 }
597
598 int os::closedir(DIR *dirp) {
599 assert(dirp != NULL, "just checking");
600 return ::closedir(dirp);
601 }
602
603 // Builds a platform dependent Agent_OnLoad_<lib_name> function name
604 // which is used to find statically linked in agents.
605 // Parameters:
606 // sym_name: Symbol in library we are looking for
607 // lib_name: Name of library to look in, NULL for shared libs.
608 // is_absolute_path == true if lib_name is absolute path to agent
609 // such as "/a/b/libL.so"
610 // == false if only the base name of the library is passed in
611 // such as "L"
612 char* os::build_agent_function_name(const char *sym_name, const char *lib_name,
613 bool is_absolute_path) {
614 char *agent_entry_name;
615 size_t len;
616 size_t name_len;
617 size_t prefix_len = strlen(JNI_LIB_PREFIX);
618 size_t suffix_len = strlen(JNI_LIB_SUFFIX);
619 const char *start;
620
621 if (lib_name != NULL) {
622 name_len = strlen(lib_name);
623 if (is_absolute_path) {
624 // Need to strip path, prefix and suffix
625 if ((start = strrchr(lib_name, *os::file_separator())) != NULL) {
626 lib_name = ++start;
627 }
628 if (strlen(lib_name) <= (prefix_len + suffix_len)) {
629 return NULL;
630 }
631 lib_name += prefix_len;
632 name_len = strlen(lib_name) - suffix_len;
633 }
634 }
635 len = (lib_name != NULL ? name_len : 0) + strlen(sym_name) + 2;
636 agent_entry_name = NEW_C_HEAP_ARRAY_RETURN_NULL(char, len, mtThread);
637 if (agent_entry_name == NULL) {
638 return NULL;
639 }
640 strcpy(agent_entry_name, sym_name);
641 if (lib_name != NULL) {
642 strcat(agent_entry_name, "_");
643 strncat(agent_entry_name, lib_name, name_len);
644 }
645 return agent_entry_name;
646 }
647
648
649 void os::naked_short_nanosleep(jlong ns) {
650 struct timespec req;
651 assert(ns > -1 && ns < NANOUNITS, "Un-interruptable sleep, short time use only");
652 req.tv_sec = 0;
653 req.tv_nsec = ns;
654 ::nanosleep(&req, NULL);
655 return;
656 }
657
658 void os::naked_short_sleep(jlong ms) {
659 assert(ms < MILLIUNITS, "Un-interruptable sleep, short time use only");
660 os::naked_short_nanosleep(ms * (NANOUNITS / MILLIUNITS));
661 return;
662 }
663
664 static const struct {
665 int sig; const char* name;
666 }
667 g_signal_info[] =
668 {
669 { SIGABRT, "SIGABRT" },
670 #ifdef SIGAIO
671 { SIGAIO, "SIGAIO" },
672 #endif
673 { SIGALRM, "SIGALRM" },
674 #ifdef SIGALRM1
675 { SIGALRM1, "SIGALRM1" },
676 #endif
677 { SIGBUS, "SIGBUS" },
678 #ifdef SIGCANCEL
679 { SIGCANCEL, "SIGCANCEL" },
680 #endif
681 { SIGCHLD, "SIGCHLD" },
682 #ifdef SIGCLD
683 { SIGCLD, "SIGCLD" },
684 #endif
685 { SIGCONT, "SIGCONT" },
686 #ifdef SIGCPUFAIL
687 { SIGCPUFAIL, "SIGCPUFAIL" },
688 #endif
689 #ifdef SIGDANGER
690 { SIGDANGER, "SIGDANGER" },
691 #endif
692 #ifdef SIGDIL
693 { SIGDIL, "SIGDIL" },
694 #endif
695 #ifdef SIGEMT
696 { SIGEMT, "SIGEMT" },
697 #endif
698 { SIGFPE, "SIGFPE" },
699 #ifdef SIGFREEZE
700 { SIGFREEZE, "SIGFREEZE" },
701 #endif
702 #ifdef SIGGFAULT
703 { SIGGFAULT, "SIGGFAULT" },
704 #endif
705 #ifdef SIGGRANT
706 { SIGGRANT, "SIGGRANT" },
707 #endif
708 { SIGHUP, "SIGHUP" },
709 { SIGILL, "SIGILL" },
710 #ifdef SIGINFO
711 { SIGINFO, "SIGINFO" },
712 #endif
713 { SIGINT, "SIGINT" },
714 #ifdef SIGIO
715 { SIGIO, "SIGIO" },
716 #endif
717 #ifdef SIGIOINT
718 { SIGIOINT, "SIGIOINT" },
719 #endif
720 #ifdef SIGIOT
721 // SIGIOT is there for BSD compatibility, but on most Unices just a
722 // synonym for SIGABRT. The result should be "SIGABRT", not
723 // "SIGIOT".
724 #if (SIGIOT != SIGABRT )
725 { SIGIOT, "SIGIOT" },
726 #endif
727 #endif
728 #ifdef SIGKAP
729 { SIGKAP, "SIGKAP" },
730 #endif
731 { SIGKILL, "SIGKILL" },
732 #ifdef SIGLOST
733 { SIGLOST, "SIGLOST" },
734 #endif
735 #ifdef SIGLWP
736 { SIGLWP, "SIGLWP" },
737 #endif
738 #ifdef SIGLWPTIMER
739 { SIGLWPTIMER, "SIGLWPTIMER" },
740 #endif
741 #ifdef SIGMIGRATE
742 { SIGMIGRATE, "SIGMIGRATE" },
743 #endif
744 #ifdef SIGMSG
745 { SIGMSG, "SIGMSG" },
746 #endif
747 { SIGPIPE, "SIGPIPE" },
748 #ifdef SIGPOLL
749 { SIGPOLL, "SIGPOLL" },
750 #endif
751 #ifdef SIGPRE
752 { SIGPRE, "SIGPRE" },
753 #endif
754 { SIGPROF, "SIGPROF" },
755 #ifdef SIGPTY
756 { SIGPTY, "SIGPTY" },
757 #endif
758 #ifdef SIGPWR
759 { SIGPWR, "SIGPWR" },
760 #endif
761 { SIGQUIT, "SIGQUIT" },
762 #ifdef SIGRECONFIG
763 { SIGRECONFIG, "SIGRECONFIG" },
764 #endif
765 #ifdef SIGRECOVERY
766 { SIGRECOVERY, "SIGRECOVERY" },
767 #endif
768 #ifdef SIGRESERVE
769 { SIGRESERVE, "SIGRESERVE" },
770 #endif
771 #ifdef SIGRETRACT
772 { SIGRETRACT, "SIGRETRACT" },
773 #endif
774 #ifdef SIGSAK
775 { SIGSAK, "SIGSAK" },
776 #endif
777 { SIGSEGV, "SIGSEGV" },
778 #ifdef SIGSOUND
779 { SIGSOUND, "SIGSOUND" },
780 #endif
781 #ifdef SIGSTKFLT
782 { SIGSTKFLT, "SIGSTKFLT" },
783 #endif
784 { SIGSTOP, "SIGSTOP" },
785 { SIGSYS, "SIGSYS" },
786 #ifdef SIGSYSERROR
787 { SIGSYSERROR, "SIGSYSERROR" },
788 #endif
789 #ifdef SIGTALRM
790 { SIGTALRM, "SIGTALRM" },
791 #endif
792 { SIGTERM, "SIGTERM" },
793 #ifdef SIGTHAW
794 { SIGTHAW, "SIGTHAW" },
795 #endif
796 { SIGTRAP, "SIGTRAP" },
797 #ifdef SIGTSTP
798 { SIGTSTP, "SIGTSTP" },
799 #endif
800 { SIGTTIN, "SIGTTIN" },
801 { SIGTTOU, "SIGTTOU" },
802 #ifdef SIGURG
803 { SIGURG, "SIGURG" },
804 #endif
805 { SIGUSR1, "SIGUSR1" },
806 { SIGUSR2, "SIGUSR2" },
807 #ifdef SIGVIRT
808 { SIGVIRT, "SIGVIRT" },
809 #endif
810 { SIGVTALRM, "SIGVTALRM" },
811 #ifdef SIGWAITING
812 { SIGWAITING, "SIGWAITING" },
813 #endif
814 #ifdef SIGWINCH
815 { SIGWINCH, "SIGWINCH" },
816 #endif
817 #ifdef SIGWINDOW
818 { SIGWINDOW, "SIGWINDOW" },
819 #endif
820 { SIGXCPU, "SIGXCPU" },
821 { SIGXFSZ, "SIGXFSZ" },
822 #ifdef SIGXRES
823 { SIGXRES, "SIGXRES" },
824 #endif
825 { -1, NULL }
826 };
827
828 // Returned string is a constant. For unknown signals "UNKNOWN" is returned.
829 const char* os::Posix::get_signal_name(int sig, char* out, size_t outlen) {
830
831 const char* ret = NULL;
832
833 #ifdef SIGRTMIN
834 if (sig >= SIGRTMIN && sig <= SIGRTMAX) {
835 if (sig == SIGRTMIN) {
836 ret = "SIGRTMIN";
837 } else if (sig == SIGRTMAX) {
838 ret = "SIGRTMAX";
839 } else {
840 jio_snprintf(out, outlen, "SIGRTMIN+%d", sig - SIGRTMIN);
841 return out;
842 }
843 }
844 #endif
845
846 if (sig > 0) {
847 for (int idx = 0; g_signal_info[idx].sig != -1; idx ++) {
848 if (g_signal_info[idx].sig == sig) {
849 ret = g_signal_info[idx].name;
850 break;
851 }
852 }
853 }
854
855 if (!ret) {
856 if (!is_valid_signal(sig)) {
857 ret = "INVALID";
858 } else {
859 ret = "UNKNOWN";
860 }
861 }
862
863 if (out && outlen > 0) {
864 strncpy(out, ret, outlen);
865 out[outlen - 1] = '\0';
866 }
867 return out;
868 }
869
870 int os::Posix::get_signal_number(const char* signal_name) {
871 char tmp[30];
872 const char* s = signal_name;
873 if (s[0] != 'S' || s[1] != 'I' || s[2] != 'G') {
874 jio_snprintf(tmp, sizeof(tmp), "SIG%s", signal_name);
875 s = tmp;
876 }
877 for (int idx = 0; g_signal_info[idx].sig != -1; idx ++) {
878 if (strcmp(g_signal_info[idx].name, s) == 0) {
879 return g_signal_info[idx].sig;
880 }
881 }
882 return -1;
883 }
884
885 int os::get_signal_number(const char* signal_name) {
886 return os::Posix::get_signal_number(signal_name);
887 }
888
889 // Returns true if signal number is valid.
890 bool os::Posix::is_valid_signal(int sig) {
891 // MacOS not really POSIX compliant: sigaddset does not return
892 // an error for invalid signal numbers. However, MacOS does not
893 // support real time signals and simply seems to have just 33
894 // signals with no holes in the signal range.
895 #ifdef __APPLE__
896 return sig >= 1 && sig < NSIG;
897 #else
898 // Use sigaddset to check for signal validity.
899 sigset_t set;
900 sigemptyset(&set);
901 if (sigaddset(&set, sig) == -1 && errno == EINVAL) {
902 return false;
903 }
904 return true;
905 #endif
906 }
907
908 bool os::Posix::is_sig_ignored(int sig) {
909 struct sigaction oact;
910 sigaction(sig, (struct sigaction*)NULL, &oact);
911 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
912 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
913 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
914 return true;
915 } else {
916 return false;
917 }
918 }
919
920 // Returns:
921 // NULL for an invalid signal number
922 // "SIG<num>" for a valid but unknown signal number
923 // signal name otherwise.
924 const char* os::exception_name(int sig, char* buf, size_t size) {
925 if (!os::Posix::is_valid_signal(sig)) {
926 return NULL;
927 }
928 const char* const name = os::Posix::get_signal_name(sig, buf, size);
929 if (strcmp(name, "UNKNOWN") == 0) {
930 jio_snprintf(buf, size, "SIG%d", sig);
931 }
932 return buf;
933 }
934
935 #define NUM_IMPORTANT_SIGS 32
936 // Returns one-line short description of a signal set in a user provided buffer.
937 const char* os::Posix::describe_signal_set_short(const sigset_t* set, char* buffer, size_t buf_size) {
938 assert(buf_size == (NUM_IMPORTANT_SIGS + 1), "wrong buffer size");
939 // Note: for shortness, just print out the first 32. That should
940 // cover most of the useful ones, apart from realtime signals.
941 for (int sig = 1; sig <= NUM_IMPORTANT_SIGS; sig++) {
942 const int rc = sigismember(set, sig);
943 if (rc == -1 && errno == EINVAL) {
944 buffer[sig-1] = '?';
945 } else {
946 buffer[sig-1] = rc == 0 ? '0' : '1';
947 }
948 }
949 buffer[NUM_IMPORTANT_SIGS] = 0;
950 return buffer;
951 }
952
953 // Prints one-line description of a signal set.
954 void os::Posix::print_signal_set_short(outputStream* st, const sigset_t* set) {
955 char buf[NUM_IMPORTANT_SIGS + 1];
956 os::Posix::describe_signal_set_short(set, buf, sizeof(buf));
957 st->print("%s", buf);
958 }
959
960 // Writes one-line description of a combination of sigaction.sa_flags into a user
961 // provided buffer. Returns that buffer.
962 const char* os::Posix::describe_sa_flags(int flags, char* buffer, size_t size) {
963 char* p = buffer;
964 size_t remaining = size;
965 bool first = true;
966 int idx = 0;
967
968 assert(buffer, "invalid argument");
969
970 if (size == 0) {
971 return buffer;
972 }
973
974 strncpy(buffer, "none", size);
975
976 const struct {
977 // NB: i is an unsigned int here because SA_RESETHAND is on some
978 // systems 0x80000000, which is implicitly unsigned. Assignining
979 // it to an int field would be an overflow in unsigned-to-signed
980 // conversion.
981 unsigned int i;
982 const char* s;
983 } flaginfo [] = {
984 { SA_NOCLDSTOP, "SA_NOCLDSTOP" },
985 { SA_ONSTACK, "SA_ONSTACK" },
986 { SA_RESETHAND, "SA_RESETHAND" },
987 { SA_RESTART, "SA_RESTART" },
988 { SA_SIGINFO, "SA_SIGINFO" },
989 { SA_NOCLDWAIT, "SA_NOCLDWAIT" },
990 { SA_NODEFER, "SA_NODEFER" },
991 #ifdef AIX
992 { SA_ONSTACK, "SA_ONSTACK" },
993 { SA_OLDSTYLE, "SA_OLDSTYLE" },
994 #endif
995 { 0, NULL }
996 };
997
998 for (idx = 0; flaginfo[idx].s && remaining > 1; idx++) {
999 if (flags & flaginfo[idx].i) {
1000 if (first) {
1001 jio_snprintf(p, remaining, "%s", flaginfo[idx].s);
1002 first = false;
1003 } else {
1004 jio_snprintf(p, remaining, "|%s", flaginfo[idx].s);
1005 }
1006 const size_t len = strlen(p);
1007 p += len;
1008 remaining -= len;
1009 }
1010 }
1011
1012 buffer[size - 1] = '\0';
1013
1014 return buffer;
1015 }
1016
1017 // Prints one-line description of a combination of sigaction.sa_flags.
1018 void os::Posix::print_sa_flags(outputStream* st, int flags) {
1019 char buffer[0x100];
1020 os::Posix::describe_sa_flags(flags, buffer, sizeof(buffer));
1021 st->print("%s", buffer);
1022 }
1023
1024 // Helper function for os::Posix::print_siginfo_...():
1025 // return a textual description for signal code.
1026 struct enum_sigcode_desc_t {
1027 const char* s_name;
1028 const char* s_desc;
1029 };
1030
1031 static bool get_signal_code_description(const siginfo_t* si, enum_sigcode_desc_t* out) {
1032
1033 const struct {
1034 int sig; int code; const char* s_code; const char* s_desc;
1035 } t1 [] = {
1036 { SIGILL, ILL_ILLOPC, "ILL_ILLOPC", "Illegal opcode." },
1037 { SIGILL, ILL_ILLOPN, "ILL_ILLOPN", "Illegal operand." },
1038 { SIGILL, ILL_ILLADR, "ILL_ILLADR", "Illegal addressing mode." },
1039 { SIGILL, ILL_ILLTRP, "ILL_ILLTRP", "Illegal trap." },
1040 { SIGILL, ILL_PRVOPC, "ILL_PRVOPC", "Privileged opcode." },
1041 { SIGILL, ILL_PRVREG, "ILL_PRVREG", "Privileged register." },
1042 { SIGILL, ILL_COPROC, "ILL_COPROC", "Coprocessor error." },
1043 { SIGILL, ILL_BADSTK, "ILL_BADSTK", "Internal stack error." },
1044 #if defined(IA64) && defined(LINUX)
1045 { SIGILL, ILL_BADIADDR, "ILL_BADIADDR", "Unimplemented instruction address" },
1046 { SIGILL, ILL_BREAK, "ILL_BREAK", "Application Break instruction" },
1047 #endif
1048 { SIGFPE, FPE_INTDIV, "FPE_INTDIV", "Integer divide by zero." },
1049 { SIGFPE, FPE_INTOVF, "FPE_INTOVF", "Integer overflow." },
1050 { SIGFPE, FPE_FLTDIV, "FPE_FLTDIV", "Floating-point divide by zero." },
1051 { SIGFPE, FPE_FLTOVF, "FPE_FLTOVF", "Floating-point overflow." },
1052 { SIGFPE, FPE_FLTUND, "FPE_FLTUND", "Floating-point underflow." },
1053 { SIGFPE, FPE_FLTRES, "FPE_FLTRES", "Floating-point inexact result." },
1054 { SIGFPE, FPE_FLTINV, "FPE_FLTINV", "Invalid floating-point operation." },
1055 { SIGFPE, FPE_FLTSUB, "FPE_FLTSUB", "Subscript out of range." },
1056 { SIGSEGV, SEGV_MAPERR, "SEGV_MAPERR", "Address not mapped to object." },
1057 { SIGSEGV, SEGV_ACCERR, "SEGV_ACCERR", "Invalid permissions for mapped object." },
1058 #ifdef AIX
1059 // no explanation found what keyerr would be
1060 { SIGSEGV, SEGV_KEYERR, "SEGV_KEYERR", "key error" },
1061 #endif
1062 #if defined(IA64) && !defined(AIX)
1063 { SIGSEGV, SEGV_PSTKOVF, "SEGV_PSTKOVF", "Paragraph stack overflow" },
1064 #endif
1065 #if defined(__sparc) && defined(SOLARIS)
1066 // define Solaris Sparc M7 ADI SEGV signals
1067 #if !defined(SEGV_ACCADI)
1068 #define SEGV_ACCADI 3
1069 #endif
1070 { SIGSEGV, SEGV_ACCADI, "SEGV_ACCADI", "ADI not enabled for mapped object." },
1071 #if !defined(SEGV_ACCDERR)
1072 #define SEGV_ACCDERR 4
1073 #endif
1074 { SIGSEGV, SEGV_ACCDERR, "SEGV_ACCDERR", "ADI disrupting exception." },
1075 #if !defined(SEGV_ACCPERR)
1076 #define SEGV_ACCPERR 5
1077 #endif
1078 { SIGSEGV, SEGV_ACCPERR, "SEGV_ACCPERR", "ADI precise exception." },
1079 #endif // defined(__sparc) && defined(SOLARIS)
1080 { SIGBUS, BUS_ADRALN, "BUS_ADRALN", "Invalid address alignment." },
1081 { SIGBUS, BUS_ADRERR, "BUS_ADRERR", "Nonexistent physical address." },
1082 { SIGBUS, BUS_OBJERR, "BUS_OBJERR", "Object-specific hardware error." },
1083 { SIGTRAP, TRAP_BRKPT, "TRAP_BRKPT", "Process breakpoint." },
1084 { SIGTRAP, TRAP_TRACE, "TRAP_TRACE", "Process trace trap." },
1085 { SIGCHLD, CLD_EXITED, "CLD_EXITED", "Child has exited." },
1086 { SIGCHLD, CLD_KILLED, "CLD_KILLED", "Child has terminated abnormally and did not create a core file." },
1087 { SIGCHLD, CLD_DUMPED, "CLD_DUMPED", "Child has terminated abnormally and created a core file." },
1088 { SIGCHLD, CLD_TRAPPED, "CLD_TRAPPED", "Traced child has trapped." },
1089 { SIGCHLD, CLD_STOPPED, "CLD_STOPPED", "Child has stopped." },
1090 { SIGCHLD, CLD_CONTINUED,"CLD_CONTINUED","Stopped child has continued." },
1091 #ifdef SIGPOLL
1092 { SIGPOLL, POLL_OUT, "POLL_OUT", "Output buffers available." },
1093 { SIGPOLL, POLL_MSG, "POLL_MSG", "Input message available." },
1094 { SIGPOLL, POLL_ERR, "POLL_ERR", "I/O error." },
1095 { SIGPOLL, POLL_PRI, "POLL_PRI", "High priority input available." },
1096 { SIGPOLL, POLL_HUP, "POLL_HUP", "Device disconnected. [Option End]" },
1097 #endif
1098 { -1, -1, NULL, NULL }
1099 };
1100
1101 // Codes valid in any signal context.
1102 const struct {
1103 int code; const char* s_code; const char* s_desc;
1104 } t2 [] = {
1105 { SI_USER, "SI_USER", "Signal sent by kill()." },
1106 { SI_QUEUE, "SI_QUEUE", "Signal sent by the sigqueue()." },
1107 { SI_TIMER, "SI_TIMER", "Signal generated by expiration of a timer set by timer_settime()." },
1108 { SI_ASYNCIO, "SI_ASYNCIO", "Signal generated by completion of an asynchronous I/O request." },
1109 { SI_MESGQ, "SI_MESGQ", "Signal generated by arrival of a message on an empty message queue." },
1110 // Linux specific
1111 #ifdef SI_TKILL
1112 { SI_TKILL, "SI_TKILL", "Signal sent by tkill (pthread_kill)" },
1113 #endif
1114 #ifdef SI_DETHREAD
1115 { SI_DETHREAD, "SI_DETHREAD", "Signal sent by execve() killing subsidiary threads" },
1116 #endif
1117 #ifdef SI_KERNEL
1118 { SI_KERNEL, "SI_KERNEL", "Signal sent by kernel." },
1119 #endif
1120 #ifdef SI_SIGIO
1121 { SI_SIGIO, "SI_SIGIO", "Signal sent by queued SIGIO" },
1122 #endif
1123
1124 #ifdef AIX
1125 { SI_UNDEFINED, "SI_UNDEFINED","siginfo contains partial information" },
1126 { SI_EMPTY, "SI_EMPTY", "siginfo contains no useful information" },
1127 #endif
1128
1129 #ifdef __sun
1130 { SI_NOINFO, "SI_NOINFO", "No signal information" },
1131 { SI_RCTL, "SI_RCTL", "kernel generated signal via rctl action" },
1132 { SI_LWP, "SI_LWP", "Signal sent via lwp_kill" },
1133 #endif
1134
1135 { -1, NULL, NULL }
1136 };
1137
1138 const char* s_code = NULL;
1139 const char* s_desc = NULL;
1140
1141 for (int i = 0; t1[i].sig != -1; i ++) {
1142 if (t1[i].sig == si->si_signo && t1[i].code == si->si_code) {
1143 s_code = t1[i].s_code;
1144 s_desc = t1[i].s_desc;
1145 break;
1146 }
1147 }
1148
1149 if (s_code == NULL) {
1150 for (int i = 0; t2[i].s_code != NULL; i ++) {
1151 if (t2[i].code == si->si_code) {
1152 s_code = t2[i].s_code;
1153 s_desc = t2[i].s_desc;
1154 }
1155 }
1156 }
1157
1158 if (s_code == NULL) {
1159 out->s_name = "unknown";
1160 out->s_desc = "unknown";
1161 return false;
1162 }
1163
1164 out->s_name = s_code;
1165 out->s_desc = s_desc;
1166
1167 return true;
1168 }
1169
1170 bool os::signal_sent_by_kill(const void* siginfo) {
1171 const siginfo_t* const si = (const siginfo_t*)siginfo;
1172 return si->si_code == SI_USER || si->si_code == SI_QUEUE
1173 #ifdef SI_TKILL
1174 || si->si_code == SI_TKILL
1175 #endif
1176 ;
1177 }
1178
1179 void os::print_siginfo(outputStream* os, const void* si0) {
1180
1181 const siginfo_t* const si = (const siginfo_t*) si0;
1182
1183 char buf[20];
1184 os->print("siginfo:");
1185
1186 if (!si) {
1187 os->print(" <null>");
1188 return;
1189 }
1190
1191 const int sig = si->si_signo;
1192
1193 os->print(" si_signo: %d (%s)", sig, os::Posix::get_signal_name(sig, buf, sizeof(buf)));
1194
1195 enum_sigcode_desc_t ed;
1196 get_signal_code_description(si, &ed);
1197 os->print(", si_code: %d (%s)", si->si_code, ed.s_name);
1198
1199 if (si->si_errno) {
1200 os->print(", si_errno: %d", si->si_errno);
1201 }
1202
1203 // Output additional information depending on the signal code.
1204
1205 // Note: Many implementations lump si_addr, si_pid, si_uid etc. together as unions,
1206 // so it depends on the context which member to use. For synchronous error signals,
1207 // we print si_addr, unless the signal was sent by another process or thread, in
1208 // which case we print out pid or tid of the sender.
1209 if (signal_sent_by_kill(si)) {
1210 const pid_t pid = si->si_pid;
1211 os->print(", si_pid: %ld", (long) pid);
1212 if (IS_VALID_PID(pid)) {
1213 const pid_t me = getpid();
1214 if (me == pid) {
1215 os->print(" (current process)");
1216 }
1217 } else {
1218 os->print(" (invalid)");
1219 }
1220 os->print(", si_uid: %ld", (long) si->si_uid);
1221 if (sig == SIGCHLD) {
1222 os->print(", si_status: %d", si->si_status);
1223 }
1224 } else if (sig == SIGSEGV || sig == SIGBUS || sig == SIGILL ||
1225 sig == SIGTRAP || sig == SIGFPE) {
1226 os->print(", si_addr: " PTR_FORMAT, p2i(si->si_addr));
1227 #ifdef SIGPOLL
1228 } else if (sig == SIGPOLL) {
1229 os->print(", si_band: %ld", si->si_band);
1230 #endif
1231 }
1232
1233 }
1234
1235 bool os::signal_thread(Thread* thread, int sig, const char* reason) {
1236 OSThread* osthread = thread->osthread();
1237 if (osthread) {
1238 #if defined (SOLARIS)
1239 // Note: we cannot use pthread_kill on Solaris - not because
1240 // its missing, but because we do not have the pthread_t id.
1241 int status = thr_kill(osthread->thread_id(), sig);
1242 #else
1243 int status = pthread_kill(osthread->pthread_id(), sig);
1244 #endif
1245 if (status == 0) {
1246 Events::log(Thread::current(), "sent signal %d to Thread " INTPTR_FORMAT " because %s.",
1247 sig, p2i(thread), reason);
1248 return true;
1249 }
1250 }
1251 return false;
1252 }
1253
1254 int os::Posix::unblock_thread_signal_mask(const sigset_t *set) {
1255 return pthread_sigmask(SIG_UNBLOCK, set, NULL);
1256 }
1257
1258 address os::Posix::ucontext_get_pc(const ucontext_t* ctx) {
1259 #if defined(AIX)
1260 return Aix::ucontext_get_pc(ctx);
1261 #elif defined(BSD)
1262 return Bsd::ucontext_get_pc(ctx);
1263 #elif defined(LINUX)
1264 return Linux::ucontext_get_pc(ctx);
1265 #elif defined(SOLARIS)
1266 return Solaris::ucontext_get_pc(ctx);
1267 #else
1268 VMError::report_and_die("unimplemented ucontext_get_pc");
1269 #endif
1270 }
1271
1272 void os::Posix::ucontext_set_pc(ucontext_t* ctx, address pc) {
1273 #if defined(AIX)
1274 Aix::ucontext_set_pc(ctx, pc);
1275 #elif defined(BSD)
1276 Bsd::ucontext_set_pc(ctx, pc);
1277 #elif defined(LINUX)
1278 Linux::ucontext_set_pc(ctx, pc);
1279 #elif defined(SOLARIS)
1280 Solaris::ucontext_set_pc(ctx, pc);
1281 #else
1282 VMError::report_and_die("unimplemented ucontext_get_pc");
1283 #endif
1284 }
1285
1286 char* os::Posix::describe_pthread_attr(char* buf, size_t buflen, const pthread_attr_t* attr) {
1287 size_t stack_size = 0;
1288 size_t guard_size = 0;
1289 int detachstate = 0;
1290 pthread_attr_getstacksize(attr, &stack_size);
1291 pthread_attr_getguardsize(attr, &guard_size);
1292 // Work around linux NPTL implementation error, see also os::create_thread() in os_linux.cpp.
1293 LINUX_ONLY(stack_size -= guard_size);
1294 pthread_attr_getdetachstate(attr, &detachstate);
1295 jio_snprintf(buf, buflen, "stacksize: " SIZE_FORMAT "k, guardsize: " SIZE_FORMAT "k, %s",
1296 stack_size / 1024, guard_size / 1024,
1297 (detachstate == PTHREAD_CREATE_DETACHED ? "detached" : "joinable"));
1298 return buf;
1299 }
1300
1301 char* os::Posix::realpath(const char* filename, char* outbuf, size_t outbuflen) {
1302
1303 if (filename == NULL || outbuf == NULL || outbuflen < 1) {
1304 assert(false, "os::Posix::realpath: invalid arguments.");
1305 errno = EINVAL;
1306 return NULL;
1307 }
1308
1309 char* result = NULL;
1310
1311 // This assumes platform realpath() is implemented according to POSIX.1-2008.
1312 // POSIX.1-2008 allows to specify NULL for the output buffer, in which case
1313 // output buffer is dynamically allocated and must be ::free()'d by the caller.
1314 char* p = ::realpath(filename, NULL);
1315 if (p != NULL) {
1316 if (strlen(p) < outbuflen) {
1317 strcpy(outbuf, p);
1318 result = outbuf;
1319 } else {
1320 errno = ENAMETOOLONG;
1321 }
1322 ::free(p); // *not* os::free
1323 } else {
1324 // Fallback for platforms struggling with modern Posix standards (AIX 5.3, 6.1). If realpath
1325 // returns EINVAL, this may indicate that realpath is not POSIX.1-2008 compatible and
1326 // that it complains about the NULL we handed down as user buffer.
1327 // In this case, use the user provided buffer but at least check whether realpath caused
1328 // a memory overwrite.
1329 if (errno == EINVAL) {
1330 outbuf[outbuflen - 1] = '\0';
1331 p = ::realpath(filename, outbuf);
1332 if (p != NULL) {
1333 guarantee(outbuf[outbuflen - 1] == '\0', "realpath buffer overwrite detected.");
1334 result = p;
1335 }
1336 }
1337 }
1338 return result;
1339
1340 }
1341
1342 int os::stat(const char *path, struct stat *sbuf) {
1343 return ::stat(path, sbuf);
1344 }
1345
1346 char * os::native_path(char *path) {
1347 return path;
1348 }
1349
1350 bool os::same_files(const char* file1, const char* file2) {
1351 if (strcmp(file1, file2) == 0) {
1352 return true;
1353 }
1354
1355 bool is_same = false;
1356 struct stat st1;
1357 struct stat st2;
1358
1359 if (os::stat(file1, &st1) < 0) {
1360 return false;
1361 }
1362
1363 if (os::stat(file2, &st2) < 0) {
1364 return false;
1365 }
1366
1367 if (st1.st_dev == st2.st_dev && st1.st_ino == st2.st_ino) {
1368 // same files
1369 is_same = true;
1370 }
1371 return is_same;
1372 }
1373
1374 // Check minimum allowable stack sizes for thread creation and to initialize
1375 // the java system classes, including StackOverflowError - depends on page
1376 // size.
1377 // The space needed for frames during startup is platform dependent. It
1378 // depends on word size, platform calling conventions, C frame layout and
1379 // interpreter/C1/C2 design decisions. Therefore this is given in a
1380 // platform (os/cpu) dependent constant.
1381 // To this, space for guard mechanisms is added, which depends on the
1382 // page size which again depends on the concrete system the VM is running
1383 // on. Space for libc guard pages is not included in this size.
1384 jint os::Posix::set_minimum_stack_sizes() {
1385 size_t os_min_stack_allowed = SOLARIS_ONLY(thr_min_stack()) NOT_SOLARIS(PTHREAD_STACK_MIN);
1386
1387 _java_thread_min_stack_allowed = _java_thread_min_stack_allowed +
1388 JavaThread::stack_guard_zone_size() +
1389 JavaThread::stack_shadow_zone_size();
1390
1391 _java_thread_min_stack_allowed = align_up(_java_thread_min_stack_allowed, vm_page_size());
1392 _java_thread_min_stack_allowed = MAX2(_java_thread_min_stack_allowed, os_min_stack_allowed);
1393
1394 size_t stack_size_in_bytes = ThreadStackSize * K;
1395 if (stack_size_in_bytes != 0 &&
1396 stack_size_in_bytes < _java_thread_min_stack_allowed) {
1397 // The '-Xss' and '-XX:ThreadStackSize=N' options both set
1398 // ThreadStackSize so we go with "Java thread stack size" instead
1399 // of "ThreadStackSize" to be more friendly.
1400 tty->print_cr("\nThe Java thread stack size specified is too small. "
1401 "Specify at least " SIZE_FORMAT "k",
1402 _java_thread_min_stack_allowed / K);
1403 return JNI_ERR;
1404 }
1405
1406 // Make the stack size a multiple of the page size so that
1407 // the yellow/red zones can be guarded.
1408 JavaThread::set_stack_size_at_create(align_up(stack_size_in_bytes, vm_page_size()));
1409
1410 // Reminder: a compiler thread is a Java thread.
1411 _compiler_thread_min_stack_allowed = _compiler_thread_min_stack_allowed +
1412 JavaThread::stack_guard_zone_size() +
1413 JavaThread::stack_shadow_zone_size();
1414
1415 _compiler_thread_min_stack_allowed = align_up(_compiler_thread_min_stack_allowed, vm_page_size());
1416 _compiler_thread_min_stack_allowed = MAX2(_compiler_thread_min_stack_allowed, os_min_stack_allowed);
1417
1418 stack_size_in_bytes = CompilerThreadStackSize * K;
1419 if (stack_size_in_bytes != 0 &&
1420 stack_size_in_bytes < _compiler_thread_min_stack_allowed) {
1421 tty->print_cr("\nThe CompilerThreadStackSize specified is too small. "
1422 "Specify at least " SIZE_FORMAT "k",
1423 _compiler_thread_min_stack_allowed / K);
1424 return JNI_ERR;
1425 }
1426
1427 _vm_internal_thread_min_stack_allowed = align_up(_vm_internal_thread_min_stack_allowed, vm_page_size());
1428 _vm_internal_thread_min_stack_allowed = MAX2(_vm_internal_thread_min_stack_allowed, os_min_stack_allowed);
1429
1430 stack_size_in_bytes = VMThreadStackSize * K;
1431 if (stack_size_in_bytes != 0 &&
1432 stack_size_in_bytes < _vm_internal_thread_min_stack_allowed) {
1433 tty->print_cr("\nThe VMThreadStackSize specified is too small. "
1434 "Specify at least " SIZE_FORMAT "k",
1435 _vm_internal_thread_min_stack_allowed / K);
1436 return JNI_ERR;
1437 }
1438 return JNI_OK;
1439 }
1440
1441 // Called when creating the thread. The minimum stack sizes have already been calculated
1442 size_t os::Posix::get_initial_stack_size(ThreadType thr_type, size_t req_stack_size) {
1443 size_t stack_size;
1444 if (req_stack_size == 0) {
1445 stack_size = default_stack_size(thr_type);
1446 } else {
1447 stack_size = req_stack_size;
1448 }
1449
1450 switch (thr_type) {
1451 case os::java_thread:
1452 // Java threads use ThreadStackSize which default value can be
1453 // changed with the flag -Xss
1454 if (req_stack_size == 0 && JavaThread::stack_size_at_create() > 0) {
1455 // no requested size and we have a more specific default value
1456 stack_size = JavaThread::stack_size_at_create();
1457 }
1458 stack_size = MAX2(stack_size,
1459 _java_thread_min_stack_allowed);
1460 break;
1461 case os::compiler_thread:
1462 if (req_stack_size == 0 && CompilerThreadStackSize > 0) {
1463 // no requested size and we have a more specific default value
1464 stack_size = (size_t)(CompilerThreadStackSize * K);
1465 }
1466 stack_size = MAX2(stack_size,
1467 _compiler_thread_min_stack_allowed);
1468 break;
1469 case os::vm_thread:
1470 case os::pgc_thread:
1471 case os::cgc_thread:
1472 case os::watcher_thread:
1473 default: // presume the unknown thr_type is a VM internal
1474 if (req_stack_size == 0 && VMThreadStackSize > 0) {
1475 // no requested size and we have a more specific default value
1476 stack_size = (size_t)(VMThreadStackSize * K);
1477 }
1478
1479 stack_size = MAX2(stack_size,
1480 _vm_internal_thread_min_stack_allowed);
1481 break;
1482 }
1483
1484 // pthread_attr_setstacksize() may require that the size be rounded up to the OS page size.
1485 // Be careful not to round up to 0. Align down in that case.
1486 if (stack_size <= SIZE_MAX - vm_page_size()) {
1487 stack_size = align_up(stack_size, vm_page_size());
1488 } else {
1489 stack_size = align_down(stack_size, vm_page_size());
1490 }
1491
1492 return stack_size;
1493 }
1494
1495 bool os::Posix::is_root(uid_t uid){
1496 return ROOT_UID == uid;
1497 }
1498
1499 bool os::Posix::matches_effective_uid_or_root(uid_t uid) {
1500 return is_root(uid) || geteuid() == uid;
1501 }
1502
1503 bool os::Posix::matches_effective_uid_and_gid_or_root(uid_t uid, gid_t gid) {
1504 return is_root(uid) || (geteuid() == uid && getegid() == gid);
1505 }
1506
1507 Thread* os::ThreadCrashProtection::_protected_thread = NULL;
1508 os::ThreadCrashProtection* os::ThreadCrashProtection::_crash_protection = NULL;
1509 volatile intptr_t os::ThreadCrashProtection::_crash_mux = 0;
1510
1511 os::ThreadCrashProtection::ThreadCrashProtection() {
1512 }
1513
1514 /*
1515 * See the caveats for this class in os_posix.hpp
1516 * Protects the callback call so that SIGSEGV / SIGBUS jumps back into this
1517 * method and returns false. If none of the signals are raised, returns true.
1518 * The callback is supposed to provide the method that should be protected.
1519 */
1520 bool os::ThreadCrashProtection::call(os::CrashProtectionCallback& cb) {
1521 sigset_t saved_sig_mask;
1522
1523 Thread::muxAcquire(&_crash_mux, "CrashProtection");
1524
1525 _protected_thread = Thread::current_or_null();
1526 assert(_protected_thread != NULL, "Cannot crash protect a NULL thread");
1527
1528 // we cannot rely on sigsetjmp/siglongjmp to save/restore the signal mask
1529 // since on at least some systems (OS X) siglongjmp will restore the mask
1530 // for the process, not the thread
1531 pthread_sigmask(0, NULL, &saved_sig_mask);
1532 if (sigsetjmp(_jmpbuf, 0) == 0) {
1533 // make sure we can see in the signal handler that we have crash protection
1534 // installed
1535 _crash_protection = this;
1536 cb.call();
1537 // and clear the crash protection
1538 _crash_protection = NULL;
1539 _protected_thread = NULL;
1540 Thread::muxRelease(&_crash_mux);
1541 return true;
1542 }
1543 // this happens when we siglongjmp() back
1544 pthread_sigmask(SIG_SETMASK, &saved_sig_mask, NULL);
1545 _crash_protection = NULL;
1546 _protected_thread = NULL;
1547 Thread::muxRelease(&_crash_mux);
1548 return false;
1549 }
1550
1551 void os::ThreadCrashProtection::restore() {
1552 assert(_crash_protection != NULL, "must have crash protection");
1553 siglongjmp(_jmpbuf, 1);
1554 }
1555
1556 void os::ThreadCrashProtection::check_crash_protection(int sig,
1557 Thread* thread) {
1558
1559 if (thread != NULL &&
1560 thread == _protected_thread &&
1561 _crash_protection != NULL) {
1562
1563 if (sig == SIGSEGV || sig == SIGBUS) {
1564 _crash_protection->restore();
1565 }
1566 }
1567 }
1568
1569 // Shared clock/time and other supporting routines for pthread_mutex/cond
1570 // initialization. This is enabled on Solaris but only some of the clock/time
1571 // functionality is actually used there.
1572
1573 // Shared condattr object for use with relative timed-waits. Will be associated
1574 // with CLOCK_MONOTONIC if available to avoid issues with time-of-day changes,
1575 // but otherwise whatever default is used by the platform - generally the
1576 // time-of-day clock.
1577 static pthread_condattr_t _condAttr[1];
1578
1579 // Shared mutexattr to explicitly set the type to PTHREAD_MUTEX_NORMAL as not
1580 // all systems (e.g. FreeBSD) map the default to "normal".
1581 static pthread_mutexattr_t _mutexAttr[1];
1582
1583 // common basic initialization that is always supported
1584 static void pthread_init_common(void) {
1585 int status;
1586 if ((status = pthread_condattr_init(_condAttr)) != 0) {
1587 fatal("pthread_condattr_init: %s", os::strerror(status));
1588 }
1589 if ((status = pthread_mutexattr_init(_mutexAttr)) != 0) {
1590 fatal("pthread_mutexattr_init: %s", os::strerror(status));
1591 }
1592 if ((status = pthread_mutexattr_settype(_mutexAttr, PTHREAD_MUTEX_NORMAL)) != 0) {
1593 fatal("pthread_mutexattr_settype: %s", os::strerror(status));
1594 }
1595 // Solaris has it's own PlatformMutex, distinct from the one for POSIX.
1596 NOT_SOLARIS(os::PlatformMutex::init();)
1597 }
1598
1599 #ifndef SOLARIS
1600 sigset_t sigs;
1601 struct sigaction sigact[NSIG];
1602
1603 struct sigaction* os::Posix::get_preinstalled_handler(int sig) {
1604 if (sigismember(&sigs, sig)) {
1605 return &sigact[sig];
1606 }
1607 return NULL;
1608 }
1609
1610 void os::Posix::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
1611 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
1612 sigact[sig] = oldAct;
1613 sigaddset(&sigs, sig);
1614 }
1615 #endif
1616
1617 // Not all POSIX types and API's are available on all notionally "posix"
1618 // platforms. If we have build-time support then we will check for actual
1619 // runtime support via dlopen/dlsym lookup. This allows for running on an
1620 // older OS version compared to the build platform. But if there is no
1621 // build time support then there cannot be any runtime support as we do not
1622 // know what the runtime types would be (for example clockid_t might be an
1623 // int or int64_t).
1624 //
1625 #ifdef SUPPORTS_CLOCK_MONOTONIC
1626
1627 // This means we have clockid_t, clock_gettime et al and CLOCK_MONOTONIC
1628
1629 int (*os::Posix::_clock_gettime)(clockid_t, struct timespec *) = NULL;
1630 int (*os::Posix::_clock_getres)(clockid_t, struct timespec *) = NULL;
1631
1632 static int (*_pthread_condattr_setclock)(pthread_condattr_t *, clockid_t) = NULL;
1633
1634 static bool _use_clock_monotonic_condattr = false;
1635
1636 // Determine what POSIX API's are present and do appropriate
1637 // configuration.
1638 void os::Posix::init(void) {
1639
1640 // NOTE: no logging available when this is called. Put logging
1641 // statements in init_2().
1642
1643 // 1. Check for CLOCK_MONOTONIC support.
1644
1645 void* handle = NULL;
1646
1647 // For linux we need librt, for other OS we can find
1648 // this function in regular libc.
1649 #ifdef NEEDS_LIBRT
1650 // We do dlopen's in this particular order due to bug in linux
1651 // dynamic loader (see 6348968) leading to crash on exit.
1652 handle = dlopen("librt.so.1", RTLD_LAZY);
1653 if (handle == NULL) {
1654 handle = dlopen("librt.so", RTLD_LAZY);
1655 }
1656 #endif
1657
1658 if (handle == NULL) {
1659 handle = RTLD_DEFAULT;
1660 }
1661
1662 int (*clock_getres_func)(clockid_t, struct timespec*) =
1663 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1664 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1665 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1666 if (clock_getres_func != NULL && clock_gettime_func != NULL) {
1667 // We assume that if both clock_gettime and clock_getres support
1668 // CLOCK_MONOTONIC then the OS provides true high-res monotonic clock.
1669 struct timespec res;
1670 struct timespec tp;
1671 if (clock_getres_func(CLOCK_MONOTONIC, &res) == 0 &&
1672 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1673 // Yes, monotonic clock is supported.
1674 _clock_gettime = clock_gettime_func;
1675 _clock_getres = clock_getres_func;
1676 } else {
1677 #ifdef NEEDS_LIBRT
1678 // Close librt if there is no monotonic clock.
1679 if (handle != RTLD_DEFAULT) {
1680 dlclose(handle);
1681 }
1682 #endif
1683 }
1684 }
1685
1686 // 2. Check for pthread_condattr_setclock support.
1687
1688 // libpthread is already loaded.
1689 int (*condattr_setclock_func)(pthread_condattr_t*, clockid_t) =
1690 (int (*)(pthread_condattr_t*, clockid_t))dlsym(RTLD_DEFAULT,
1691 "pthread_condattr_setclock");
1692 if (condattr_setclock_func != NULL) {
1693 _pthread_condattr_setclock = condattr_setclock_func;
1694 }
1695
1696 // Now do general initialization.
1697
1698 pthread_init_common();
1699
1700 #ifndef SOLARIS
1701 int status;
1702 if (_pthread_condattr_setclock != NULL && _clock_gettime != NULL) {
1703 if ((status = _pthread_condattr_setclock(_condAttr, CLOCK_MONOTONIC)) != 0) {
1704 if (status == EINVAL) {
1705 _use_clock_monotonic_condattr = false;
1706 warning("Unable to use monotonic clock with relative timed-waits" \
1707 " - changes to the time-of-day clock may have adverse affects");
1708 } else {
1709 fatal("pthread_condattr_setclock: %s", os::strerror(status));
1710 }
1711 } else {
1712 _use_clock_monotonic_condattr = true;
1713 }
1714 }
1715 #endif // !SOLARIS
1716
1717 }
1718
1719 void os::Posix::init_2(void) {
1720 #ifndef SOLARIS
1721 log_info(os)("Use of CLOCK_MONOTONIC is%s supported",
1722 (_clock_gettime != NULL ? "" : " not"));
1723 log_info(os)("Use of pthread_condattr_setclock is%s supported",
1724 (_pthread_condattr_setclock != NULL ? "" : " not"));
1725 log_info(os)("Relative timed-wait using pthread_cond_timedwait is associated with %s",
1726 _use_clock_monotonic_condattr ? "CLOCK_MONOTONIC" : "the default clock");
1727 sigemptyset(&sigs);
1728 #endif // !SOLARIS
1729 }
1730
1731 #else // !SUPPORTS_CLOCK_MONOTONIC
1732
1733 void os::Posix::init(void) {
1734 pthread_init_common();
1735 }
1736
1737 void os::Posix::init_2(void) {
1738 #ifndef SOLARIS
1739 log_info(os)("Use of CLOCK_MONOTONIC is not supported");
1740 log_info(os)("Use of pthread_condattr_setclock is not supported");
1741 log_info(os)("Relative timed-wait using pthread_cond_timedwait is associated with the default clock");
1742 sigemptyset(&sigs);
1743 #endif // !SOLARIS
1744 }
1745
1746 #endif // SUPPORTS_CLOCK_MONOTONIC
1747
1748 // Utility to convert the given timeout to an absolute timespec
1749 // (based on the appropriate clock) to use with pthread_cond_timewait,
1750 // and sem_timedwait().
1751 // The clock queried here must be the clock used to manage the
1752 // timeout of the condition variable or semaphore.
1753 //
1754 // The passed in timeout value is either a relative time in nanoseconds
1755 // or an absolute time in milliseconds. A relative timeout will be
1756 // associated with CLOCK_MONOTONIC if available, unless the real-time clock
1757 // is explicitly requested; otherwise, or if absolute,
1758 // the default time-of-day clock will be used.
1759
1760 // Given time is a 64-bit value and the time_t used in the timespec is
1761 // sometimes a signed-32-bit value we have to watch for overflow if times
1762 // way in the future are given. Further on Solaris versions
1763 // prior to 10 there is a restriction (see cond_timedwait) that the specified
1764 // number of seconds, in abstime, is less than current_time + 100000000.
1765 // As it will be over 20 years before "now + 100000000" will overflow we can
1766 // ignore overflow and just impose a hard-limit on seconds using the value
1767 // of "now + 100000000". This places a limit on the timeout of about 3.17
1768 // years from "now".
1769 //
1770 #define MAX_SECS 100000000
1771
1772 // Calculate a new absolute time that is "timeout" nanoseconds from "now".
1773 // "unit" indicates the unit of "now_part_sec" (may be nanos or micros depending
1774 // on which clock API is being used).
1775 static void calc_rel_time(timespec* abstime, jlong timeout, jlong now_sec,
1776 jlong now_part_sec, jlong unit) {
1777 time_t max_secs = now_sec + MAX_SECS;
1778
1779 jlong seconds = timeout / NANOUNITS;
1780 timeout %= NANOUNITS; // remaining nanos
1781
1782 if (seconds >= MAX_SECS) {
1783 // More seconds than we can add, so pin to max_secs.
1784 abstime->tv_sec = max_secs;
1785 abstime->tv_nsec = 0;
1786 } else {
1787 abstime->tv_sec = now_sec + seconds;
1788 long nanos = (now_part_sec * (NANOUNITS / unit)) + timeout;
1789 if (nanos >= NANOUNITS) { // overflow
1790 abstime->tv_sec += 1;
1791 nanos -= NANOUNITS;
1792 }
1793 abstime->tv_nsec = nanos;
1794 }
1795 }
1796
1797 // Unpack the given deadline in milliseconds since the epoch, into the given timespec.
1798 // The current time in seconds is also passed in to enforce an upper bound as discussed above.
1799 // This is only used with gettimeofday, when clock_gettime is not available.
1800 static void unpack_abs_time(timespec* abstime, jlong deadline, jlong now_sec) {
1801 time_t max_secs = now_sec + MAX_SECS;
1802
1803 jlong seconds = deadline / MILLIUNITS;
1804 jlong millis = deadline % MILLIUNITS;
1805
1806 if (seconds >= max_secs) {
1807 // Absolute seconds exceeds allowed max, so pin to max_secs.
1808 abstime->tv_sec = max_secs;
1809 abstime->tv_nsec = 0;
1810 } else {
1811 abstime->tv_sec = seconds;
1812 abstime->tv_nsec = millis * (NANOUNITS / MILLIUNITS);
1813 }
1814 }
1815
1816 static jlong millis_to_nanos(jlong millis) {
1817 // We have to watch for overflow when converting millis to nanos,
1818 // but if millis is that large then we will end up limiting to
1819 // MAX_SECS anyway, so just do that here.
1820 if (millis / MILLIUNITS > MAX_SECS) {
1821 millis = jlong(MAX_SECS) * MILLIUNITS;
1822 }
1823 return millis * (NANOUNITS / MILLIUNITS);
1824 }
1825
1826 static void to_abstime(timespec* abstime, jlong timeout,
1827 bool isAbsolute, bool isRealtime) {
1828 DEBUG_ONLY(int max_secs = MAX_SECS;)
1829
1830 if (timeout < 0) {
1831 timeout = 0;
1832 }
1833
1834 #ifdef SUPPORTS_CLOCK_MONOTONIC
1835
1836 clockid_t clock = CLOCK_MONOTONIC;
1837 // need to ensure we have a runtime check for clock_gettime support
1838 if (!isAbsolute && os::Posix::supports_monotonic_clock()) {
1839 if (!_use_clock_monotonic_condattr || isRealtime) {
1840 clock = CLOCK_REALTIME;
1841 }
1842 struct timespec now;
1843 int status = os::Posix::clock_gettime(clock, &now);
1844 assert_status(status == 0, status, "clock_gettime");
1845 calc_rel_time(abstime, timeout, now.tv_sec, now.tv_nsec, NANOUNITS);
1846 DEBUG_ONLY(max_secs += now.tv_sec;)
1847 } else {
1848
1849 #else
1850
1851 { // Match the block scope.
1852
1853 #endif // SUPPORTS_CLOCK_MONOTONIC
1854
1855 // Time-of-day clock is all we can reliably use.
1856 struct timeval now;
1857 int status = gettimeofday(&now, NULL);
1858 assert_status(status == 0, errno, "gettimeofday");
1859 if (isAbsolute) {
1860 unpack_abs_time(abstime, timeout, now.tv_sec);
1861 } else {
1862 calc_rel_time(abstime, timeout, now.tv_sec, now.tv_usec, MICROUNITS);
1863 }
1864 DEBUG_ONLY(max_secs += now.tv_sec;)
1865 }
1866
1867 assert(abstime->tv_sec >= 0, "tv_sec < 0");
1868 assert(abstime->tv_sec <= max_secs, "tv_sec > max_secs");
1869 assert(abstime->tv_nsec >= 0, "tv_nsec < 0");
1870 assert(abstime->tv_nsec < NANOUNITS, "tv_nsec >= NANOUNITS");
1871 }
1872
1873 // Create an absolute time 'millis' milliseconds in the future, using the
1874 // real-time (time-of-day) clock. Used by PosixSemaphore.
1875 void os::Posix::to_RTC_abstime(timespec* abstime, int64_t millis) {
1876 to_abstime(abstime, millis_to_nanos(millis),
1877 false /* not absolute */,
1878 true /* use real-time clock */);
1879 }
1880
1881 // Shared pthread_mutex/cond based PlatformEvent implementation.
1882 // Not currently usable by Solaris.
1883
1884 #ifndef SOLARIS
1885
1886 // PlatformEvent
1887 //
1888 // Assumption:
1889 // Only one parker can exist on an event, which is why we allocate
1890 // them per-thread. Multiple unparkers can coexist.
1891 //
1892 // _event serves as a restricted-range semaphore.
1893 // -1 : thread is blocked, i.e. there is a waiter
1894 // 0 : neutral: thread is running or ready,
1895 // could have been signaled after a wait started
1896 // 1 : signaled - thread is running or ready
1897 //
1898 // Having three states allows for some detection of bad usage - see
1899 // comments on unpark().
1900
1901 os::PlatformEvent::PlatformEvent() {
1902 int status = pthread_cond_init(_cond, _condAttr);
1903 assert_status(status == 0, status, "cond_init");
1904 status = pthread_mutex_init(_mutex, _mutexAttr);
1905 assert_status(status == 0, status, "mutex_init");
1906 _event = 0;
1907 _nParked = 0;
1908 }
1909
1910 void os::PlatformEvent::park() { // AKA "down()"
1911 // Transitions for _event:
1912 // -1 => -1 : illegal
1913 // 1 => 0 : pass - return immediately
1914 // 0 => -1 : block; then set _event to 0 before returning
1915
1916 // Invariant: Only the thread associated with the PlatformEvent
1917 // may call park().
1918 assert(_nParked == 0, "invariant");
1919
1920 int v;
1921
1922 // atomically decrement _event
1923 for (;;) {
1924 v = _event;
1925 if (Atomic::cmpxchg(v - 1, &_event, v) == v) break;
1926 }
1927 guarantee(v >= 0, "invariant");
1928
1929 if (v == 0) { // Do this the hard way by blocking ...
1930 int status = pthread_mutex_lock(_mutex);
1931 assert_status(status == 0, status, "mutex_lock");
1932 guarantee(_nParked == 0, "invariant");
1933 ++_nParked;
1934 while (_event < 0) {
1935 // OS-level "spurious wakeups" are ignored
1936 status = pthread_cond_wait(_cond, _mutex);
1937 assert_status(status == 0, status, "cond_wait");
1938 }
1939 --_nParked;
1940
1941 _event = 0;
1942 status = pthread_mutex_unlock(_mutex);
1943 assert_status(status == 0, status, "mutex_unlock");
1944 // Paranoia to ensure our locked and lock-free paths interact
1945 // correctly with each other.
1946 OrderAccess::fence();
1947 }
1948 guarantee(_event >= 0, "invariant");
1949 }
1950
1951 int os::PlatformEvent::park(jlong millis) {
1952 // Transitions for _event:
1953 // -1 => -1 : illegal
1954 // 1 => 0 : pass - return immediately
1955 // 0 => -1 : block; then set _event to 0 before returning
1956
1957 // Invariant: Only the thread associated with the Event/PlatformEvent
1958 // may call park().
1959 assert(_nParked == 0, "invariant");
1960
1961 int v;
1962 // atomically decrement _event
1963 for (;;) {
1964 v = _event;
1965 if (Atomic::cmpxchg(v - 1, &_event, v) == v) break;
1966 }
1967 guarantee(v >= 0, "invariant");
1968
1969 if (v == 0) { // Do this the hard way by blocking ...
1970 struct timespec abst;
1971 to_abstime(&abst, millis_to_nanos(millis), false, false);
1972
1973 int ret = OS_TIMEOUT;
1974 int status = pthread_mutex_lock(_mutex);
1975 assert_status(status == 0, status, "mutex_lock");
1976 guarantee(_nParked == 0, "invariant");
1977 ++_nParked;
1978
1979 while (_event < 0) {
1980 status = pthread_cond_timedwait(_cond, _mutex, &abst);
1981 assert_status(status == 0 || status == ETIMEDOUT,
1982 status, "cond_timedwait");
1983 // OS-level "spurious wakeups" are ignored unless the archaic
1984 // FilterSpuriousWakeups is set false. That flag should be obsoleted.
1985 if (!FilterSpuriousWakeups) break;
1986 if (status == ETIMEDOUT) break;
1987 }
1988 --_nParked;
1989
1990 if (_event >= 0) {
1991 ret = OS_OK;
1992 }
1993
1994 _event = 0;
1995 status = pthread_mutex_unlock(_mutex);
1996 assert_status(status == 0, status, "mutex_unlock");
1997 // Paranoia to ensure our locked and lock-free paths interact
1998 // correctly with each other.
1999 OrderAccess::fence();
2000 return ret;
2001 }
2002 return OS_OK;
2003 }
2004
2005 void os::PlatformEvent::unpark() {
2006 // Transitions for _event:
2007 // 0 => 1 : just return
2008 // 1 => 1 : just return
2009 // -1 => either 0 or 1; must signal target thread
2010 // That is, we can safely transition _event from -1 to either
2011 // 0 or 1.
2012 // See also: "Semaphores in Plan 9" by Mullender & Cox
2013 //
2014 // Note: Forcing a transition from "-1" to "1" on an unpark() means
2015 // that it will take two back-to-back park() calls for the owning
2016 // thread to block. This has the benefit of forcing a spurious return
2017 // from the first park() call after an unpark() call which will help
2018 // shake out uses of park() and unpark() without checking state conditions
2019 // properly. This spurious return doesn't manifest itself in any user code
2020 // but only in the correctly written condition checking loops of ObjectMonitor,
2021 // Mutex/Monitor, Thread::muxAcquire and JavaThread::sleep
2022
2023 if (Atomic::xchg(1, &_event) >= 0) return;
2024
2025 int status = pthread_mutex_lock(_mutex);
2026 assert_status(status == 0, status, "mutex_lock");
2027 int anyWaiters = _nParked;
2028 assert(anyWaiters == 0 || anyWaiters == 1, "invariant");
2029 status = pthread_mutex_unlock(_mutex);
2030 assert_status(status == 0, status, "mutex_unlock");
2031
2032 // Note that we signal() *after* dropping the lock for "immortal" Events.
2033 // This is safe and avoids a common class of futile wakeups. In rare
2034 // circumstances this can cause a thread to return prematurely from
2035 // cond_{timed}wait() but the spurious wakeup is benign and the victim
2036 // will simply re-test the condition and re-park itself.
2037 // This provides particular benefit if the underlying platform does not
2038 // provide wait morphing.
2039
2040 if (anyWaiters != 0) {
2041 status = pthread_cond_signal(_cond);
2042 assert_status(status == 0, status, "cond_signal");
2043 }
2044 }
2045
2046 // JSR166 support
2047
2048 os::PlatformParker::PlatformParker() {
2049 int status;
2050 status = pthread_cond_init(&_cond[REL_INDEX], _condAttr);
2051 assert_status(status == 0, status, "cond_init rel");
2052 status = pthread_cond_init(&_cond[ABS_INDEX], NULL);
2053 assert_status(status == 0, status, "cond_init abs");
2054 status = pthread_mutex_init(_mutex, _mutexAttr);
2055 assert_status(status == 0, status, "mutex_init");
2056 _cur_index = -1; // mark as unused
2057 }
2058
2059 // Parker::park decrements count if > 0, else does a condvar wait. Unpark
2060 // sets count to 1 and signals condvar. Only one thread ever waits
2061 // on the condvar. Contention seen when trying to park implies that someone
2062 // is unparking you, so don't wait. And spurious returns are fine, so there
2063 // is no need to track notifications.
2064
2065 void Parker::park(bool isAbsolute, jlong time) {
2066
2067 // Optional fast-path check:
2068 // Return immediately if a permit is available.
2069 // We depend on Atomic::xchg() having full barrier semantics
2070 // since we are doing a lock-free update to _counter.
2071 if (Atomic::xchg(0, &_counter) > 0) return;
2072
2073 Thread* thread = Thread::current();
2074 assert(thread->is_Java_thread(), "Must be JavaThread");
2075 JavaThread *jt = (JavaThread *)thread;
2076
2077 // Optional optimization -- avoid state transitions if there's
2078 // an interrupt pending.
2079 if (jt->is_interrupted(false)) {
2080 return;
2081 }
2082
2083 // Next, demultiplex/decode time arguments
2084 struct timespec absTime;
2085 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
2086 return;
2087 }
2088 if (time > 0) {
2089 to_abstime(&absTime, time, isAbsolute, false);
2090 }
2091
2092 // Enter safepoint region
2093 // Beware of deadlocks such as 6317397.
2094 // The per-thread Parker:: mutex is a classic leaf-lock.
2095 // In particular a thread must never block on the Threads_lock while
2096 // holding the Parker:: mutex. If safepoints are pending both the
2097 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
2098 ThreadBlockInVM tbivm(jt);
2099
2100 // Can't access interrupt state now that we are _thread_blocked. If we've
2101 // been interrupted since we checked above then _counter will be > 0.
2102
2103 // Don't wait if cannot get lock since interference arises from
2104 // unparking.
2105 if (pthread_mutex_trylock(_mutex) != 0) {
2106 return;
2107 }
2108
2109 int status;
2110 if (_counter > 0) { // no wait needed
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 return;
2118 }
2119
2120 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2121 jt->set_suspend_equivalent();
2122 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2123
2124 assert(_cur_index == -1, "invariant");
2125 if (time == 0) {
2126 _cur_index = REL_INDEX; // arbitrary choice when not timed
2127 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
2128 assert_status(status == 0, status, "cond_timedwait");
2129 }
2130 else {
2131 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
2132 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
2133 assert_status(status == 0 || status == ETIMEDOUT,
2134 status, "cond_timedwait");
2135 }
2136 _cur_index = -1;
2137
2138 _counter = 0;
2139 status = pthread_mutex_unlock(_mutex);
2140 assert_status(status == 0, status, "invariant");
2141 // Paranoia to ensure our locked and lock-free paths interact
2142 // correctly with each other and Java-level accesses.
2143 OrderAccess::fence();
2144
2145 // If externally suspended while waiting, re-suspend
2146 if (jt->handle_special_suspend_equivalent_condition()) {
2147 jt->java_suspend_self();
2148 }
2149 }
2150
2151 void Parker::unpark() {
2152 int status = pthread_mutex_lock(_mutex);
2153 assert_status(status == 0, status, "invariant");
2154 const int s = _counter;
2155 _counter = 1;
2156 // must capture correct index before unlocking
2157 int index = _cur_index;
2158 status = pthread_mutex_unlock(_mutex);
2159 assert_status(status == 0, status, "invariant");
2160
2161 // Note that we signal() *after* dropping the lock for "immortal" Events.
2162 // This is safe and avoids a common class of futile wakeups. In rare
2163 // circumstances this can cause a thread to return prematurely from
2164 // cond_{timed}wait() but the spurious wakeup is benign and the victim
2165 // will simply re-test the condition and re-park itself.
2166 // This provides particular benefit if the underlying platform does not
2167 // provide wait morphing.
2168
2169 if (s < 1 && index != -1) {
2170 // thread is definitely parked
2171 status = pthread_cond_signal(&_cond[index]);
2172 assert_status(status == 0, status, "invariant");
2173 }
2174 }
2175
2176 // Platform Mutex/Monitor implementation
2177
2178 #if PLATFORM_MONITOR_IMPL_INDIRECT
2179
2180 os::PlatformMutex::Mutex::Mutex() : _next(NULL) {
2181 int status = pthread_mutex_init(&_mutex, _mutexAttr);
2182 assert_status(status == 0, status, "mutex_init");
2183 }
2184
2185 os::PlatformMutex::Mutex::~Mutex() {
2186 int status = pthread_mutex_destroy(&_mutex);
2187 assert_status(status == 0, status, "mutex_destroy");
2188 }
2189
2190 pthread_mutex_t os::PlatformMutex::_freelist_lock;
2191 os::PlatformMutex::Mutex* os::PlatformMutex::_mutex_freelist = NULL;
2192
2193 void os::PlatformMutex::init() {
2194 int status = pthread_mutex_init(&_freelist_lock, _mutexAttr);
2195 assert_status(status == 0, status, "freelist lock init");
2196 }
2197
2198 struct os::PlatformMutex::WithFreeListLocked : public StackObj {
2199 WithFreeListLocked() {
2200 int status = pthread_mutex_lock(&_freelist_lock);
2201 assert_status(status == 0, status, "freelist lock");
2202 }
2203
2204 ~WithFreeListLocked() {
2205 int status = pthread_mutex_unlock(&_freelist_lock);
2206 assert_status(status == 0, status, "freelist unlock");
2207 }
2208 };
2209
2210 os::PlatformMutex::PlatformMutex() {
2211 {
2212 WithFreeListLocked wfl;
2213 _impl = _mutex_freelist;
2214 if (_impl != NULL) {
2215 _mutex_freelist = _impl->_next;
2216 _impl->_next = NULL;
2217 return;
2218 }
2219 }
2220 _impl = new Mutex();
2221 }
2222
2223 os::PlatformMutex::~PlatformMutex() {
2224 WithFreeListLocked wfl;
2225 assert(_impl->_next == NULL, "invariant");
2226 _impl->_next = _mutex_freelist;
2227 _mutex_freelist = _impl;
2228 }
2229
2230 os::PlatformMonitor::Cond::Cond() : _next(NULL) {
2231 int status = pthread_cond_init(&_cond, _condAttr);
2232 assert_status(status == 0, status, "cond_init");
2233 }
2234
2235 os::PlatformMonitor::Cond::~Cond() {
2236 int status = pthread_cond_destroy(&_cond);
2237 assert_status(status == 0, status, "cond_destroy");
2238 }
2239
2240 os::PlatformMonitor::Cond* os::PlatformMonitor::_cond_freelist = NULL;
2241
2242 os::PlatformMonitor::PlatformMonitor() {
2243 {
2244 WithFreeListLocked wfl;
2245 _impl = _cond_freelist;
2246 if (_impl != NULL) {
2247 _cond_freelist = _impl->_next;
2248 _impl->_next = NULL;
2249 return;
2250 }
2251 }
2252 _impl = new Cond();
2253 }
2254
2255 os::PlatformMonitor::~PlatformMonitor() {
2256 WithFreeListLocked wfl;
2257 assert(_impl->_next == NULL, "invariant");
2258 _impl->_next = _cond_freelist;
2259 _cond_freelist = _impl;
2260 }
2261
2262 #else
2263
2264 os::PlatformMutex::PlatformMutex() {
2265 int status = pthread_mutex_init(&_mutex, _mutexAttr);
2266 assert_status(status == 0, status, "mutex_init");
2267 }
2268
2269 os::PlatformMutex::~PlatformMutex() {
2270 int status = pthread_mutex_destroy(&_mutex);
2271 assert_status(status == 0, status, "mutex_destroy");
2272 }
2273
2274 os::PlatformMonitor::PlatformMonitor() {
2275 int status = pthread_cond_init(&_cond, _condAttr);
2276 assert_status(status == 0, status, "cond_init");
2277 }
2278
2279 os::PlatformMonitor::~PlatformMonitor() {
2280 int status = pthread_cond_destroy(&_cond);
2281 assert_status(status == 0, status, "cond_destroy");
2282 }
2283
2284 #endif // PLATFORM_MONITOR_IMPL_INDIRECT
2285
2286 // Must already be locked
2287 int os::PlatformMonitor::wait(jlong millis) {
2288 assert(millis >= 0, "negative timeout");
2289 if (millis > 0) {
2290 struct timespec abst;
2291 // We have to watch for overflow when converting millis to nanos,
2292 // but if millis is that large then we will end up limiting to
2293 // MAX_SECS anyway, so just do that here.
2294 if (millis / MILLIUNITS > MAX_SECS) {
2295 millis = jlong(MAX_SECS) * MILLIUNITS;
2296 }
2297 to_abstime(&abst, millis * (NANOUNITS / MILLIUNITS), false, false);
2298
2299 int ret = OS_TIMEOUT;
2300 int status = pthread_cond_timedwait(cond(), mutex(), &abst);
2301 assert_status(status == 0 || status == ETIMEDOUT,
2302 status, "cond_timedwait");
2303 if (status == 0) {
2304 ret = OS_OK;
2305 }
2306 return ret;
2307 } else {
2308 int status = pthread_cond_wait(cond(), mutex());
2309 assert_status(status == 0, status, "cond_wait");
2310 return OS_OK;
2311 }
2312 }
2313
2314 #endif // !SOLARIS