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/events.hpp"
35 #include "utilities/formatBuffer.hpp"
36 #include "utilities/macros.hpp"
37 #include "utilities/vmError.hpp"
38
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 #ifndef __OpenBSD__
50 #include <utmpx.h>
51 #endif
52
53 // Todo: provide a os::get_max_process_id() or similar. Number of processes
54 // may have been configured, can be read more accurately from proc fs etc.
55 #ifndef MAX_PID
56 #define MAX_PID INT_MAX
57 #endif
58 #define IS_VALID_PID(p) (p > 0 && p < MAX_PID)
59
60 #define ROOT_UID 0
61
62 #ifndef MAP_ANONYMOUS
63 #define MAP_ANONYMOUS MAP_ANON
64 #endif
65
66 #ifndef MAP_NORESERVE
67 #define MAP_NORESERVE 0
68 #endif
69
70 #ifndef PTHREAD_STACK_MIN
71 #ifdef _SC_THREAD_STACK_MIN
72 #define PTHREAD_STACK_MIN sysconf(_SC_THREAD_STACK_MIN)
73 #else
74 #define PTHREAD_STACK_MIN 1UL << 14 // 16KB
75 #endif
76 #endif
77
78 #define check_with_errno(check_type, cond, msg) \
79 do { \
80 int err = errno; \
81 check_type(cond, "%s; error='%s' (errno=%s)", msg, os::strerror(err), \
82 os::errno_name(err)); \
83 } while (false)
84
85 #define assert_with_errno(cond, msg) check_with_errno(assert, cond, msg)
86 #define guarantee_with_errno(cond, msg) check_with_errno(guarantee, cond, msg)
87
88 // Check core dump limit and report possible place where core can be found
check_dump_limit(char * buffer,size_t bufferSize)89 void os::check_dump_limit(char* buffer, size_t bufferSize) {
90 if (!FLAG_IS_DEFAULT(CreateCoredumpOnCrash) && !CreateCoredumpOnCrash) {
91 jio_snprintf(buffer, bufferSize, "CreateCoredumpOnCrash is disabled from command line");
92 VMError::record_coredump_status(buffer, false);
93 return;
94 }
95
96 int n;
97 struct rlimit rlim;
98 bool success;
99
100 char core_path[PATH_MAX];
101 n = get_core_path(core_path, PATH_MAX);
102
103 if (n <= 0) {
104 jio_snprintf(buffer, bufferSize, "core.%d (may not exist)", current_process_id());
105 success = true;
106 #ifdef LINUX
107 } else if (core_path[0] == '"') { // redirect to user process
108 jio_snprintf(buffer, bufferSize, "Core dumps may be processed with %s", core_path);
109 success = true;
110 #endif
111 } else if (getrlimit(RLIMIT_CORE, &rlim) != 0) {
112 jio_snprintf(buffer, bufferSize, "%s (may not exist)", core_path);
113 success = true;
114 } else {
115 switch(rlim.rlim_cur) {
116 case RLIM_INFINITY:
117 jio_snprintf(buffer, bufferSize, "%s", core_path);
118 success = true;
119 break;
120 case 0:
121 jio_snprintf(buffer, bufferSize, "Core dumps have been disabled. To enable core dumping, try \"ulimit -c unlimited\" before starting Java again");
122 success = false;
123 break;
124 default:
125 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);
126 success = true;
127 break;
128 }
129 }
130
131 VMError::record_coredump_status(buffer, success);
132 }
133
get_native_stack(address * stack,int frames,int toSkip)134 int os::get_native_stack(address* stack, int frames, int toSkip) {
135 int frame_idx = 0;
136 int num_of_frames; // number of frames captured
137 frame fr = os::current_frame();
138 while (fr.pc() && frame_idx < frames) {
139 if (toSkip > 0) {
140 toSkip --;
141 } else {
142 stack[frame_idx ++] = fr.pc();
143 }
144 if (fr.fp() == NULL || fr.cb() != NULL ||
145 fr.sender_pc() == NULL || os::is_first_C_frame(&fr)) break;
146
147 if (fr.sender_pc() && !os::is_first_C_frame(&fr)) {
148 fr = os::get_sender_for_C_frame(&fr);
149 } else {
150 break;
151 }
152 }
153 num_of_frames = frame_idx;
154 for (; frame_idx < frames; frame_idx ++) {
155 stack[frame_idx] = NULL;
156 }
157
158 return num_of_frames;
159 }
160
161
unsetenv(const char * name)162 bool os::unsetenv(const char* name) {
163 assert(name != NULL, "Null pointer");
164 return (::unsetenv(name) == 0);
165 }
166
get_last_error()167 int os::get_last_error() {
168 return errno;
169 }
170
is_debugger_attached()171 bool os::is_debugger_attached() {
172 // not implemented
173 return false;
174 }
175
wait_for_keypress_at_exit(void)176 void os::wait_for_keypress_at_exit(void) {
177 // don't do anything on posix platforms
178 return;
179 }
180
create_file_for_heap(const char * dir)181 int os::create_file_for_heap(const char* dir) {
182
183 const char name_template[] = "/jvmheap.XXXXXX";
184
185 size_t fullname_len = strlen(dir) + strlen(name_template);
186 char *fullname = (char*)os::malloc(fullname_len + 1, mtInternal);
187 if (fullname == NULL) {
188 vm_exit_during_initialization(err_msg("Malloc failed during creation of backing file for heap (%s)", os::strerror(errno)));
189 return -1;
190 }
191 int n = snprintf(fullname, fullname_len + 1, "%s%s", dir, name_template);
192 assert((size_t)n == fullname_len, "Unexpected number of characters in string");
193
194 os::native_path(fullname);
195
196 sigset_t set, oldset;
197 int ret = sigfillset(&set);
198 assert_with_errno(ret == 0, "sigfillset returned error");
199
200 // set the file creation mask.
201 mode_t file_mode = S_IRUSR | S_IWUSR;
202
203 // create a new file.
204 int fd = mkstemp(fullname);
205
206 if (fd < 0) {
207 warning("Could not create file for heap with template %s", fullname);
208 os::free(fullname);
209 return -1;
210 }
211
212 // delete the name from the filesystem. When 'fd' is closed, the file (and space) will be deleted.
213 ret = unlink(fullname);
214 assert_with_errno(ret == 0, "unlink returned error");
215
216 os::free(fullname);
217 return fd;
218 }
219
reserve_mmapped_memory(size_t bytes,char * requested_addr)220 static char* reserve_mmapped_memory(size_t bytes, char* requested_addr) {
221 char * addr;
222 int flags = MAP_PRIVATE NOT_AIX( | MAP_NORESERVE ) | MAP_ANONYMOUS;
223 if (requested_addr != NULL) {
224 assert((uintptr_t)requested_addr % os::vm_page_size() == 0, "Requested address should be aligned to OS page size");
225 flags |= MAP_FIXED;
226 }
227
228 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
229 // touch an uncommitted page. Otherwise, the read/write might
230 // succeed if we have enough swap space to back the physical page.
231 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
232 flags, -1, 0);
233
234 if (addr != MAP_FAILED) {
235 MemTracker::record_virtual_memory_reserve((address)addr, bytes, CALLER_PC);
236 return addr;
237 }
238 return NULL;
239 }
240
util_posix_fallocate(int fd,off_t offset,off_t len)241 static int util_posix_fallocate(int fd, off_t offset, off_t len) {
242 #ifdef __APPLE__
243 fstore_t store = { F_ALLOCATECONTIG, F_PEOFPOSMODE, 0, len };
244 // First we try to get a continuous chunk of disk space
245 int ret = fcntl(fd, F_PREALLOCATE, &store);
246 if (ret == -1) {
247 // Maybe we are too fragmented, try to allocate non-continuous range
248 store.fst_flags = F_ALLOCATEALL;
249 ret = fcntl(fd, F_PREALLOCATE, &store);
250 }
251 if(ret != -1) {
252 return ftruncate(fd, len);
253 }
254 return -1;
255 #elif defined(__OpenBSD__)
256 struct stat s;
257 if (fstat(fd, &s) == -1)
258 return -1;
259
260 if (s.st_size < offset+len) {
261 return ftruncate(fd, offset+len);
262 }
263 return 0;
264 #else
265 return posix_fallocate(fd, offset, len);
266 #endif
267 }
268
269 // Map the given address range to the provided file descriptor.
map_memory_to_file(char * base,size_t size,int fd)270 char* os::map_memory_to_file(char* base, size_t size, int fd) {
271 assert(fd != -1, "File descriptor is not valid");
272
273 // allocate space for the file
274 int ret = util_posix_fallocate(fd, 0, (off_t)size);
275 if (ret != 0) {
276 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory. error(%d)", ret));
277 return NULL;
278 }
279
280 int prot = PROT_READ | PROT_WRITE;
281 int flags = MAP_SHARED;
282 if (base != NULL) {
283 flags |= MAP_FIXED;
284 }
285 char* addr = (char*)mmap(base, size, prot, flags, fd, 0);
286
287 if (addr == MAP_FAILED) {
288 warning("Failed mmap to file. (%s)", os::strerror(errno));
289 return NULL;
290 }
291 if (base != NULL && addr != base) {
292 if (!os::release_memory(addr, size)) {
293 warning("Could not release memory on unsuccessful file mapping");
294 }
295 return NULL;
296 }
297 return addr;
298 }
299
replace_existing_mapping_with_file_mapping(char * base,size_t size,int fd)300 char* os::replace_existing_mapping_with_file_mapping(char* base, size_t size, int fd) {
301 assert(fd != -1, "File descriptor is not valid");
302 assert(base != NULL, "Base cannot be NULL");
303
304 return map_memory_to_file(base, size, fd);
305 }
306
307 // Multiple threads can race in this code, and can remap over each other with MAP_FIXED,
308 // so on posix, unmap the section at the start and at the end of the chunk that we mapped
309 // rather than unmapping and remapping the whole chunk to get requested alignment.
reserve_memory_aligned(size_t size,size_t alignment,int file_desc)310 char* os::reserve_memory_aligned(size_t size, size_t alignment, int file_desc) {
311 assert((alignment & (os::vm_allocation_granularity() - 1)) == 0,
312 "Alignment must be a multiple of allocation granularity (page size)");
313 assert((size & (alignment -1)) == 0, "size must be 'alignment' aligned");
314
315 size_t extra_size = size + alignment;
316 assert(extra_size >= size, "overflow, size is too large to allow alignment");
317
318 char* extra_base;
319 if (file_desc != -1) {
320 // For file mapping, we do not call os:reserve_memory(extra_size, NULL, alignment, file_desc) because
321 // we need to deal with shrinking of the file space later when we release extra memory after alignment.
322 // We also cannot called os:reserve_memory() with file_desc set to -1 because on aix we might get SHM memory.
323 // So here to call a helper function while reserve memory for us. After we have a aligned base,
324 // we will replace anonymous mapping with file mapping.
325 extra_base = reserve_mmapped_memory(extra_size, NULL);
326 if (extra_base != NULL) {
327 MemTracker::record_virtual_memory_reserve((address)extra_base, extra_size, CALLER_PC);
328 }
329 } else {
330 extra_base = os::reserve_memory(extra_size, NULL, alignment);
331 }
332
333 if (extra_base == NULL) {
334 return NULL;
335 }
336
337 // Do manual alignment
338 char* aligned_base = align_up(extra_base, alignment);
339
340 // [ | | ]
341 // ^ extra_base
342 // ^ extra_base + begin_offset == aligned_base
343 // extra_base + begin_offset + size ^
344 // extra_base + extra_size ^
345 // |<>| == begin_offset
346 // end_offset == |<>|
347 size_t begin_offset = aligned_base - extra_base;
348 size_t end_offset = (extra_base + extra_size) - (aligned_base + size);
349
350 if (begin_offset > 0) {
351 os::release_memory(extra_base, begin_offset);
352 }
353
354 if (end_offset > 0) {
355 os::release_memory(extra_base + begin_offset + size, end_offset);
356 }
357
358 if (file_desc != -1) {
359 // After we have an aligned address, we can replace anonymous mapping with file mapping
360 if (replace_existing_mapping_with_file_mapping(aligned_base, size, file_desc) == NULL) {
361 vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
362 }
363 MemTracker::record_virtual_memory_commit((address)aligned_base, size, CALLER_PC);
364 }
365 return aligned_base;
366 }
367
vsnprintf(char * buf,size_t len,const char * fmt,va_list args)368 int os::vsnprintf(char* buf, size_t len, const char* fmt, va_list args) {
369 // All supported POSIX platforms provide C99 semantics.
370 int result = ::vsnprintf(buf, len, fmt, args);
371 // If an encoding error occurred (result < 0) then it's not clear
372 // whether the buffer is NUL terminated, so ensure it is.
373 if ((result < 0) && (len > 0)) {
374 buf[len - 1] = '\0';
375 }
376 return result;
377 }
378
get_fileno(FILE * fp)379 int os::get_fileno(FILE* fp) {
380 #ifdef __OpenBSD__
381 return fileno(fp);
382 #else
383 return NOT_AIX(::)fileno(fp);
384 #endif
385 }
386
gmtime_pd(const time_t * clock,struct tm * res)387 struct tm* os::gmtime_pd(const time_t* clock, struct tm* res) {
388 return gmtime_r(clock, res);
389 }
390
print_load_average(outputStream * st)391 void os::Posix::print_load_average(outputStream* st) {
392 st->print("load average:");
393 double loadavg[3];
394 int res = os::loadavg(loadavg, 3);
395 if (res != -1) {
396 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
397 } else {
398 st->print(" Unavailable");
399 }
400 st->cr();
401 }
402
403 #ifndef __OpenBSD__
404 // boot/uptime information;
405 // unfortunately it does not work on macOS and Linux because the utx chain has no entry
406 // for reboot at least on my test machines
print_uptime_info(outputStream * st)407 void os::Posix::print_uptime_info(outputStream* st) {
408 int bootsec = -1;
409 int currsec = time(NULL);
410 struct utmpx* ent;
411 setutxent();
412 while ((ent = getutxent())) {
413 if (!strcmp("system boot", ent->ut_line)) {
414 bootsec = ent->ut_tv.tv_sec;
415 break;
416 }
417 }
418
419 if (bootsec != -1) {
420 os::print_dhm(st, "OS uptime:", (long) (currsec-bootsec));
421 }
422 }
423 #endif
424
print_rlimit(outputStream * st,const char * msg,int resource,bool output_k=false)425 static void print_rlimit(outputStream* st, const char* msg,
426 int resource, bool output_k = false) {
427 struct rlimit rlim;
428
429 st->print(" %s ", msg);
430 int res = getrlimit(resource, &rlim);
431 if (res == -1) {
432 st->print("could not obtain value");
433 } else {
434 // soft limit
435 if (rlim.rlim_cur == RLIM_INFINITY) { st->print("infinity"); }
436 else {
437 if (output_k) { st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_cur) / 1024); }
438 else { st->print(UINT64_FORMAT, uint64_t(rlim.rlim_cur)); }
439 }
440 // hard limit
441 st->print("/");
442 if (rlim.rlim_max == RLIM_INFINITY) { st->print("infinity"); }
443 else {
444 if (output_k) { st->print(UINT64_FORMAT "k", uint64_t(rlim.rlim_max) / 1024); }
445 else { st->print(UINT64_FORMAT, uint64_t(rlim.rlim_max)); }
446 }
447 }
448 }
449
print_rlimit_info(outputStream * st)450 void os::Posix::print_rlimit_info(outputStream* st) {
451 st->print("rlimit (soft/hard):");
452 print_rlimit(st, "STACK", RLIMIT_STACK, true);
453 print_rlimit(st, ", CORE", RLIMIT_CORE, true);
454
455 #if defined(AIX)
456 st->print(", NPROC ");
457 st->print("%d", sysconf(_SC_CHILD_MAX));
458
459 print_rlimit(st, ", THREADS", RLIMIT_THREADS);
460 #elif !defined(SOLARIS)
461 print_rlimit(st, ", NPROC", RLIMIT_NPROC);
462 #endif
463
464 print_rlimit(st, ", NOFILE", RLIMIT_NOFILE);
465 #ifndef __OpenBSD__
466 print_rlimit(st, ", AS", RLIMIT_AS, true);
467 #endif
468 print_rlimit(st, ", CPU", RLIMIT_CPU);
469 print_rlimit(st, ", DATA", RLIMIT_DATA, true);
470
471 // maximum size of files that the process may create
472 print_rlimit(st, ", FSIZE", RLIMIT_FSIZE, true);
473
474 #if defined(LINUX) || defined(__APPLE__) || defined(_ALLBSD_SOURCE)
475 // maximum number of bytes of memory that may be locked into RAM
476 // (rounded down to the nearest multiple of system pagesize)
477 print_rlimit(st, ", MEMLOCK", RLIMIT_MEMLOCK, true);
478 #endif
479
480 #if defined(SOLARIS)
481 // maximum size of mapped address space of a process in bytes;
482 // if the limit is exceeded, mmap and brk fail
483 print_rlimit(st, ", VMEM", RLIMIT_VMEM, true);
484 #endif
485
486 // MacOS; The maximum size (in bytes) to which a process's resident set size may grow.
487 #if defined(__APPLE__) || defined(_ALLBSD_SOURCE)
488 print_rlimit(st, ", RSS", RLIMIT_RSS, true);
489 #endif
490
491 st->cr();
492 }
493
print_uname_info(outputStream * st)494 void os::Posix::print_uname_info(outputStream* st) {
495 // kernel
496 st->print("uname:");
497 struct utsname name;
498 uname(&name);
499 st->print("%s ", name.sysname);
500 #ifdef ASSERT
501 st->print("%s ", name.nodename);
502 #endif
503 st->print("%s ", name.release);
504 st->print("%s ", name.version);
505 st->print("%s", name.machine);
506 st->cr();
507 }
508
print_umask(outputStream * st,mode_t umsk)509 void os::Posix::print_umask(outputStream* st, mode_t umsk) {
510 st->print((umsk & S_IRUSR) ? "r" : "-");
511 st->print((umsk & S_IWUSR) ? "w" : "-");
512 st->print((umsk & S_IXUSR) ? "x" : "-");
513 st->print((umsk & S_IRGRP) ? "r" : "-");
514 st->print((umsk & S_IWGRP) ? "w" : "-");
515 st->print((umsk & S_IXGRP) ? "x" : "-");
516 st->print((umsk & S_IROTH) ? "r" : "-");
517 st->print((umsk & S_IWOTH) ? "w" : "-");
518 st->print((umsk & S_IXOTH) ? "x" : "-");
519 }
520
print_user_info(outputStream * st)521 void os::Posix::print_user_info(outputStream* st) {
522 unsigned id = (unsigned) ::getuid();
523 st->print("uid : %u ", id);
524 id = (unsigned) ::geteuid();
525 st->print("euid : %u ", id);
526 id = (unsigned) ::getgid();
527 st->print("gid : %u ", id);
528 id = (unsigned) ::getegid();
529 st->print_cr("egid : %u", id);
530 st->cr();
531
532 mode_t umsk = ::umask(0);
533 ::umask(umsk);
534 st->print("umask: %04o (", (unsigned) umsk);
535 print_umask(st, umsk);
536 st->print_cr(")");
537 st->cr();
538 }
539
540
get_host_name(char * buf,size_t buflen)541 bool os::get_host_name(char* buf, size_t buflen) {
542 struct utsname name;
543 uname(&name);
544 jio_snprintf(buf, buflen, "%s", name.nodename);
545 return true;
546 }
547
has_allocatable_memory_limit(julong * limit)548 bool os::has_allocatable_memory_limit(julong* limit) {
549 struct rlimit rlim;
550 #ifdef __OpenBSD__
551 int getrlimit_res = getrlimit(RLIMIT_DATA, &rlim);
552 #else
553 int getrlimit_res = getrlimit(RLIMIT_AS, &rlim);
554 #endif
555 // if there was an error when calling getrlimit, assume that there is no limitation
556 // on virtual memory.
557 bool result;
558 if ((getrlimit_res != 0) || (rlim.rlim_cur == RLIM_INFINITY)) {
559 result = false;
560 } else {
561 *limit = (julong)rlim.rlim_cur;
562 result = true;
563 }
564 #ifdef _LP64
565 return result;
566 #else
567 // arbitrary virtual space limit for 32 bit Unices found by testing. If
568 // getrlimit above returned a limit, bound it with this limit. Otherwise
569 // directly use it.
570 const julong max_virtual_limit = (julong)3800*M;
571 if (result) {
572 *limit = MIN2(*limit, max_virtual_limit);
573 } else {
574 *limit = max_virtual_limit;
575 }
576
577 // bound by actually allocatable memory. The algorithm uses two bounds, an
578 // upper and a lower limit. The upper limit is the current highest amount of
579 // memory that could not be allocated, the lower limit is the current highest
580 // amount of memory that could be allocated.
581 // The algorithm iteratively refines the result by halving the difference
582 // between these limits, updating either the upper limit (if that value could
583 // not be allocated) or the lower limit (if the that value could be allocated)
584 // until the difference between these limits is "small".
585
586 // the minimum amount of memory we care about allocating.
587 const julong min_allocation_size = M;
588
589 julong upper_limit = *limit;
590
591 // first check a few trivial cases
592 if (is_allocatable(upper_limit) || (upper_limit <= min_allocation_size)) {
593 *limit = upper_limit;
594 } else if (!is_allocatable(min_allocation_size)) {
595 // we found that not even min_allocation_size is allocatable. Return it
596 // anyway. There is no point to search for a better value any more.
597 *limit = min_allocation_size;
598 } else {
599 // perform the binary search.
600 julong lower_limit = min_allocation_size;
601 while ((upper_limit - lower_limit) > min_allocation_size) {
602 julong temp_limit = ((upper_limit - lower_limit) / 2) + lower_limit;
603 temp_limit = align_down(temp_limit, min_allocation_size);
604 if (is_allocatable(temp_limit)) {
605 lower_limit = temp_limit;
606 } else {
607 upper_limit = temp_limit;
608 }
609 }
610 *limit = lower_limit;
611 }
612 return true;
613 #endif
614 }
615
get_current_directory(char * buf,size_t buflen)616 const char* os::get_current_directory(char *buf, size_t buflen) {
617 return getcwd(buf, buflen);
618 }
619
open(int fd,const char * mode)620 FILE* os::open(int fd, const char* mode) {
621 return ::fdopen(fd, mode);
622 }
623
flockfile(FILE * fp)624 void os::flockfile(FILE* fp) {
625 ::flockfile(fp);
626 }
627
funlockfile(FILE * fp)628 void os::funlockfile(FILE* fp) {
629 ::funlockfile(fp);
630 }
631
opendir(const char * dirname)632 DIR* os::opendir(const char* dirname) {
633 assert(dirname != NULL, "just checking");
634 return ::opendir(dirname);
635 }
636
readdir(DIR * dirp)637 struct dirent* os::readdir(DIR* dirp) {
638 assert(dirp != NULL, "just checking");
639 return ::readdir(dirp);
640 }
641
closedir(DIR * dirp)642 int os::closedir(DIR *dirp) {
643 assert(dirp != NULL, "just checking");
644 return ::closedir(dirp);
645 }
646
647 // Builds a platform dependent Agent_OnLoad_<lib_name> function name
648 // which is used to find statically linked in agents.
649 // Parameters:
650 // sym_name: Symbol in library we are looking for
651 // lib_name: Name of library to look in, NULL for shared libs.
652 // is_absolute_path == true if lib_name is absolute path to agent
653 // such as "/a/b/libL.so"
654 // == false if only the base name of the library is passed in
655 // such as "L"
build_agent_function_name(const char * sym_name,const char * lib_name,bool is_absolute_path)656 char* os::build_agent_function_name(const char *sym_name, const char *lib_name,
657 bool is_absolute_path) {
658 char *agent_entry_name;
659 size_t len;
660 size_t name_len;
661 size_t prefix_len = strlen(JNI_LIB_PREFIX);
662 size_t suffix_len = strlen(JNI_LIB_SUFFIX);
663 const char *start;
664
665 if (lib_name != NULL) {
666 name_len = strlen(lib_name);
667 if (is_absolute_path) {
668 // Need to strip path, prefix and suffix
669 if ((start = strrchr(lib_name, *os::file_separator())) != NULL) {
670 lib_name = ++start;
671 }
672 if (strlen(lib_name) <= (prefix_len + suffix_len)) {
673 return NULL;
674 }
675 lib_name += prefix_len;
676 name_len = strlen(lib_name) - suffix_len;
677 }
678 }
679 len = (lib_name != NULL ? name_len : 0) + strlen(sym_name) + 2;
680 agent_entry_name = NEW_C_HEAP_ARRAY_RETURN_NULL(char, len, mtThread);
681 if (agent_entry_name == NULL) {
682 return NULL;
683 }
684 strcpy(agent_entry_name, sym_name);
685 if (lib_name != NULL) {
686 strcat(agent_entry_name, "_");
687 strncat(agent_entry_name, lib_name, name_len);
688 }
689 return agent_entry_name;
690 }
691
sleep(Thread * thread,jlong millis,bool interruptible)692 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
693 assert(thread == Thread::current(), "thread consistency check");
694
695 ParkEvent * const slp = thread->_SleepEvent ;
696 slp->reset() ;
697 OrderAccess::fence() ;
698
699 if (interruptible) {
700 jlong prevtime = javaTimeNanos();
701
702 for (;;) {
703 if (os::is_interrupted(thread, true)) {
704 return OS_INTRPT;
705 }
706
707 jlong newtime = javaTimeNanos();
708
709 if (newtime - prevtime < 0) {
710 // time moving backwards, should only happen if no monotonic clock
711 // not a guarantee() because JVM should not abort on kernel/glibc bugs
712 assert(!os::supports_monotonic_clock(), "unexpected time moving backwards detected in os::sleep(interruptible)");
713 } else {
714 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
715 }
716
717 if (millis <= 0) {
718 return OS_OK;
719 }
720
721 prevtime = newtime;
722
723 {
724 assert(thread->is_Java_thread(), "sanity check");
725 JavaThread *jt = (JavaThread *) thread;
726 ThreadBlockInVM tbivm(jt);
727 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
728
729 jt->set_suspend_equivalent();
730 // cleared by handle_special_suspend_equivalent_condition() or
731 // java_suspend_self() via check_and_wait_while_suspended()
732
733 slp->park(millis);
734
735 // were we externally suspended while we were waiting?
736 jt->check_and_wait_while_suspended();
737 }
738 }
739 } else {
740 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
741 jlong prevtime = javaTimeNanos();
742
743 for (;;) {
744 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
745 // the 1st iteration ...
746 jlong newtime = javaTimeNanos();
747
748 if (newtime - prevtime < 0) {
749 // time moving backwards, should only happen if no monotonic clock
750 // not a guarantee() because JVM should not abort on kernel/glibc bugs
751 assert(!os::supports_monotonic_clock(), "unexpected time moving backwards detected on os::sleep(!interruptible)");
752 } else {
753 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
754 }
755
756 if (millis <= 0) break ;
757
758 prevtime = newtime;
759 slp->park(millis);
760 }
761 return OS_OK ;
762 }
763 }
764
naked_short_nanosleep(jlong ns)765 void os::naked_short_nanosleep(jlong ns) {
766 struct timespec req;
767 assert(ns > -1 && ns < NANOUNITS, "Un-interruptable sleep, short time use only");
768 req.tv_sec = 0;
769 req.tv_nsec = ns;
770 ::nanosleep(&req, NULL);
771 return;
772 }
773
naked_short_sleep(jlong ms)774 void os::naked_short_sleep(jlong ms) {
775 assert(ms < MILLIUNITS, "Un-interruptable sleep, short time use only");
776 os::naked_short_nanosleep(ms * (NANOUNITS / MILLIUNITS));
777 return;
778 }
779
780 ////////////////////////////////////////////////////////////////////////////////
781 // interrupt support
782
interrupt(Thread * thread)783 void os::interrupt(Thread* thread) {
784 debug_only(Thread::check_for_dangling_thread_pointer(thread);)
785
786 OSThread* osthread = thread->osthread();
787
788 if (!osthread->interrupted()) {
789 osthread->set_interrupted(true);
790 // More than one thread can get here with the same value of osthread,
791 // resulting in multiple notifications. We do, however, want the store
792 // to interrupted() to be visible to other threads before we execute unpark().
793 OrderAccess::fence();
794 ParkEvent * const slp = thread->_SleepEvent ;
795 if (slp != NULL) slp->unpark() ;
796 }
797
798 // For JSR166. Unpark even if interrupt status already was set
799 if (thread->is_Java_thread())
800 ((JavaThread*)thread)->parker()->unpark();
801
802 ParkEvent * ev = thread->_ParkEvent ;
803 if (ev != NULL) ev->unpark() ;
804 }
805
is_interrupted(Thread * thread,bool clear_interrupted)806 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
807 debug_only(Thread::check_for_dangling_thread_pointer(thread);)
808
809 OSThread* osthread = thread->osthread();
810
811 bool interrupted = osthread->interrupted();
812
813 // NOTE that since there is no "lock" around the interrupt and
814 // is_interrupted operations, there is the possibility that the
815 // interrupted flag (in osThread) will be "false" but that the
816 // low-level events will be in the signaled state. This is
817 // intentional. The effect of this is that Object.wait() and
818 // LockSupport.park() will appear to have a spurious wakeup, which
819 // is allowed and not harmful, and the possibility is so rare that
820 // it is not worth the added complexity to add yet another lock.
821 // For the sleep event an explicit reset is performed on entry
822 // to os::sleep, so there is no early return. It has also been
823 // recommended not to put the interrupted flag into the "event"
824 // structure because it hides the issue.
825 if (interrupted && clear_interrupted) {
826 osthread->set_interrupted(false);
827 // consider thread->_SleepEvent->reset() ... optional optimization
828 }
829
830 return interrupted;
831 }
832
833
834
835 static const struct {
836 int sig; const char* name;
837 }
838 g_signal_info[] =
839 {
840 { SIGABRT, "SIGABRT" },
841 #ifdef SIGAIO
842 { SIGAIO, "SIGAIO" },
843 #endif
844 { SIGALRM, "SIGALRM" },
845 #ifdef SIGALRM1
846 { SIGALRM1, "SIGALRM1" },
847 #endif
848 { SIGBUS, "SIGBUS" },
849 #ifdef SIGCANCEL
850 { SIGCANCEL, "SIGCANCEL" },
851 #endif
852 { SIGCHLD, "SIGCHLD" },
853 #ifdef SIGCLD
854 { SIGCLD, "SIGCLD" },
855 #endif
856 { SIGCONT, "SIGCONT" },
857 #ifdef SIGCPUFAIL
858 { SIGCPUFAIL, "SIGCPUFAIL" },
859 #endif
860 #ifdef SIGDANGER
861 { SIGDANGER, "SIGDANGER" },
862 #endif
863 #ifdef SIGDIL
864 { SIGDIL, "SIGDIL" },
865 #endif
866 #ifdef SIGEMT
867 { SIGEMT, "SIGEMT" },
868 #endif
869 { SIGFPE, "SIGFPE" },
870 #ifdef SIGFREEZE
871 { SIGFREEZE, "SIGFREEZE" },
872 #endif
873 #ifdef SIGGFAULT
874 { SIGGFAULT, "SIGGFAULT" },
875 #endif
876 #ifdef SIGGRANT
877 { SIGGRANT, "SIGGRANT" },
878 #endif
879 { SIGHUP, "SIGHUP" },
880 { SIGILL, "SIGILL" },
881 #ifdef SIGINFO
882 { SIGINFO, "SIGINFO" },
883 #endif
884 { SIGINT, "SIGINT" },
885 #ifdef SIGIO
886 { SIGIO, "SIGIO" },
887 #endif
888 #ifdef SIGIOINT
889 { SIGIOINT, "SIGIOINT" },
890 #endif
891 #ifdef SIGIOT
892 // SIGIOT is there for BSD compatibility, but on most Unices just a
893 // synonym for SIGABRT. The result should be "SIGABRT", not
894 // "SIGIOT".
895 #if (SIGIOT != SIGABRT )
896 { SIGIOT, "SIGIOT" },
897 #endif
898 #endif
899 #ifdef SIGKAP
900 { SIGKAP, "SIGKAP" },
901 #endif
902 { SIGKILL, "SIGKILL" },
903 #ifdef SIGLOST
904 { SIGLOST, "SIGLOST" },
905 #endif
906 #ifdef SIGLWP
907 { SIGLWP, "SIGLWP" },
908 #endif
909 #ifdef SIGLWPTIMER
910 { SIGLWPTIMER, "SIGLWPTIMER" },
911 #endif
912 #ifdef SIGMIGRATE
913 { SIGMIGRATE, "SIGMIGRATE" },
914 #endif
915 #ifdef SIGMSG
916 { SIGMSG, "SIGMSG" },
917 #endif
918 { SIGPIPE, "SIGPIPE" },
919 #ifdef SIGPOLL
920 { SIGPOLL, "SIGPOLL" },
921 #endif
922 #ifdef SIGPRE
923 { SIGPRE, "SIGPRE" },
924 #endif
925 { SIGPROF, "SIGPROF" },
926 #ifdef SIGPTY
927 { SIGPTY, "SIGPTY" },
928 #endif
929 #ifdef SIGPWR
930 { SIGPWR, "SIGPWR" },
931 #endif
932 { SIGQUIT, "SIGQUIT" },
933 #ifdef SIGRECONFIG
934 { SIGRECONFIG, "SIGRECONFIG" },
935 #endif
936 #ifdef SIGRECOVERY
937 { SIGRECOVERY, "SIGRECOVERY" },
938 #endif
939 #ifdef SIGRESERVE
940 { SIGRESERVE, "SIGRESERVE" },
941 #endif
942 #ifdef SIGRETRACT
943 { SIGRETRACT, "SIGRETRACT" },
944 #endif
945 #ifdef SIGSAK
946 { SIGSAK, "SIGSAK" },
947 #endif
948 { SIGSEGV, "SIGSEGV" },
949 #ifdef SIGSOUND
950 { SIGSOUND, "SIGSOUND" },
951 #endif
952 #ifdef SIGSTKFLT
953 { SIGSTKFLT, "SIGSTKFLT" },
954 #endif
955 { SIGSTOP, "SIGSTOP" },
956 { SIGSYS, "SIGSYS" },
957 #ifdef SIGSYSERROR
958 { SIGSYSERROR, "SIGSYSERROR" },
959 #endif
960 #ifdef SIGTALRM
961 { SIGTALRM, "SIGTALRM" },
962 #endif
963 { SIGTERM, "SIGTERM" },
964 #ifdef SIGTHAW
965 { SIGTHAW, "SIGTHAW" },
966 #endif
967 { SIGTRAP, "SIGTRAP" },
968 #ifdef SIGTSTP
969 { SIGTSTP, "SIGTSTP" },
970 #endif
971 { SIGTTIN, "SIGTTIN" },
972 { SIGTTOU, "SIGTTOU" },
973 #ifdef SIGURG
974 { SIGURG, "SIGURG" },
975 #endif
976 { SIGUSR1, "SIGUSR1" },
977 { SIGUSR2, "SIGUSR2" },
978 #ifdef SIGVIRT
979 { SIGVIRT, "SIGVIRT" },
980 #endif
981 { SIGVTALRM, "SIGVTALRM" },
982 #ifdef SIGWAITING
983 { SIGWAITING, "SIGWAITING" },
984 #endif
985 #ifdef SIGWINCH
986 { SIGWINCH, "SIGWINCH" },
987 #endif
988 #ifdef SIGWINDOW
989 { SIGWINDOW, "SIGWINDOW" },
990 #endif
991 { SIGXCPU, "SIGXCPU" },
992 { SIGXFSZ, "SIGXFSZ" },
993 #ifdef SIGXRES
994 { SIGXRES, "SIGXRES" },
995 #endif
996 { -1, NULL }
997 };
998
999 // Returned string is a constant. For unknown signals "UNKNOWN" is returned.
get_signal_name(int sig,char * out,size_t outlen)1000 const char* os::Posix::get_signal_name(int sig, char* out, size_t outlen) {
1001
1002 const char* ret = NULL;
1003
1004 #ifdef SIGRTMIN
1005 if (sig >= SIGRTMIN && sig <= SIGRTMAX) {
1006 if (sig == SIGRTMIN) {
1007 ret = "SIGRTMIN";
1008 } else if (sig == SIGRTMAX) {
1009 ret = "SIGRTMAX";
1010 } else {
1011 jio_snprintf(out, outlen, "SIGRTMIN+%d", sig - SIGRTMIN);
1012 return out;
1013 }
1014 }
1015 #endif
1016
1017 if (sig > 0) {
1018 for (int idx = 0; g_signal_info[idx].sig != -1; idx ++) {
1019 if (g_signal_info[idx].sig == sig) {
1020 ret = g_signal_info[idx].name;
1021 break;
1022 }
1023 }
1024 }
1025
1026 if (!ret) {
1027 if (!is_valid_signal(sig)) {
1028 ret = "INVALID";
1029 } else {
1030 ret = "UNKNOWN";
1031 }
1032 }
1033
1034 if (out && outlen > 0) {
1035 strncpy(out, ret, outlen);
1036 out[outlen - 1] = '\0';
1037 }
1038 return out;
1039 }
1040
get_signal_number(const char * signal_name)1041 int os::Posix::get_signal_number(const char* signal_name) {
1042 char tmp[30];
1043 const char* s = signal_name;
1044 if (s[0] != 'S' || s[1] != 'I' || s[2] != 'G') {
1045 jio_snprintf(tmp, sizeof(tmp), "SIG%s", signal_name);
1046 s = tmp;
1047 }
1048 for (int idx = 0; g_signal_info[idx].sig != -1; idx ++) {
1049 if (strcmp(g_signal_info[idx].name, s) == 0) {
1050 return g_signal_info[idx].sig;
1051 }
1052 }
1053 return -1;
1054 }
1055
get_signal_number(const char * signal_name)1056 int os::get_signal_number(const char* signal_name) {
1057 return os::Posix::get_signal_number(signal_name);
1058 }
1059
1060 // Returns true if signal number is valid.
is_valid_signal(int sig)1061 bool os::Posix::is_valid_signal(int sig) {
1062 // MacOS not really POSIX compliant: sigaddset does not return
1063 // an error for invalid signal numbers. However, MacOS does not
1064 // support real time signals and simply seems to have just 33
1065 // signals with no holes in the signal range.
1066 #ifdef __APPLE__
1067 return sig >= 1 && sig < NSIG;
1068 #else
1069 // Use sigaddset to check for signal validity.
1070 sigset_t set;
1071 sigemptyset(&set);
1072 if (sigaddset(&set, sig) == -1 && errno == EINVAL) {
1073 return false;
1074 }
1075 return true;
1076 #endif
1077 }
1078
is_sig_ignored(int sig)1079 bool os::Posix::is_sig_ignored(int sig) {
1080 struct sigaction oact;
1081 sigaction(sig, (struct sigaction*)NULL, &oact);
1082 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
1083 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
1084 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
1085 return true;
1086 } else {
1087 return false;
1088 }
1089 }
1090
1091 // Returns:
1092 // NULL for an invalid signal number
1093 // "SIG<num>" for a valid but unknown signal number
1094 // signal name otherwise.
exception_name(int sig,char * buf,size_t size)1095 const char* os::exception_name(int sig, char* buf, size_t size) {
1096 if (!os::Posix::is_valid_signal(sig)) {
1097 return NULL;
1098 }
1099 const char* const name = os::Posix::get_signal_name(sig, buf, size);
1100 if (strcmp(name, "UNKNOWN") == 0) {
1101 jio_snprintf(buf, size, "SIG%d", sig);
1102 }
1103 return buf;
1104 }
1105
1106 #define NUM_IMPORTANT_SIGS 32
1107 // Returns one-line short description of a signal set in a user provided buffer.
describe_signal_set_short(const sigset_t * set,char * buffer,size_t buf_size)1108 const char* os::Posix::describe_signal_set_short(const sigset_t* set, char* buffer, size_t buf_size) {
1109 assert(buf_size == (NUM_IMPORTANT_SIGS + 1), "wrong buffer size");
1110 // Note: for shortness, just print out the first 32. That should
1111 // cover most of the useful ones, apart from realtime signals.
1112 for (int sig = 1; sig <= NUM_IMPORTANT_SIGS; sig++) {
1113 const int rc = sigismember(set, sig);
1114 if (rc == -1 && errno == EINVAL) {
1115 buffer[sig-1] = '?';
1116 } else {
1117 buffer[sig-1] = rc == 0 ? '0' : '1';
1118 }
1119 }
1120 buffer[NUM_IMPORTANT_SIGS] = 0;
1121 return buffer;
1122 }
1123
1124 // Prints one-line description of a signal set.
print_signal_set_short(outputStream * st,const sigset_t * set)1125 void os::Posix::print_signal_set_short(outputStream* st, const sigset_t* set) {
1126 char buf[NUM_IMPORTANT_SIGS + 1];
1127 os::Posix::describe_signal_set_short(set, buf, sizeof(buf));
1128 st->print("%s", buf);
1129 }
1130
1131 // Writes one-line description of a combination of sigaction.sa_flags into a user
1132 // provided buffer. Returns that buffer.
describe_sa_flags(int flags,char * buffer,size_t size)1133 const char* os::Posix::describe_sa_flags(int flags, char* buffer, size_t size) {
1134 char* p = buffer;
1135 size_t remaining = size;
1136 bool first = true;
1137 int idx = 0;
1138
1139 assert(buffer, "invalid argument");
1140
1141 if (size == 0) {
1142 return buffer;
1143 }
1144
1145 strncpy(buffer, "none", size);
1146
1147 const struct {
1148 // NB: i is an unsigned int here because SA_RESETHAND is on some
1149 // systems 0x80000000, which is implicitly unsigned. Assignining
1150 // it to an int field would be an overflow in unsigned-to-signed
1151 // conversion.
1152 unsigned int i;
1153 const char* s;
1154 } flaginfo [] = {
1155 { SA_NOCLDSTOP, "SA_NOCLDSTOP" },
1156 { SA_ONSTACK, "SA_ONSTACK" },
1157 { SA_RESETHAND, "SA_RESETHAND" },
1158 { SA_RESTART, "SA_RESTART" },
1159 { SA_SIGINFO, "SA_SIGINFO" },
1160 { SA_NOCLDWAIT, "SA_NOCLDWAIT" },
1161 { SA_NODEFER, "SA_NODEFER" },
1162 #ifdef AIX
1163 { SA_ONSTACK, "SA_ONSTACK" },
1164 { SA_OLDSTYLE, "SA_OLDSTYLE" },
1165 #endif
1166 { 0, NULL }
1167 };
1168
1169 for (idx = 0; flaginfo[idx].s && remaining > 1; idx++) {
1170 if (flags & flaginfo[idx].i) {
1171 if (first) {
1172 jio_snprintf(p, remaining, "%s", flaginfo[idx].s);
1173 first = false;
1174 } else {
1175 jio_snprintf(p, remaining, "|%s", flaginfo[idx].s);
1176 }
1177 const size_t len = strlen(p);
1178 p += len;
1179 remaining -= len;
1180 }
1181 }
1182
1183 buffer[size - 1] = '\0';
1184
1185 return buffer;
1186 }
1187
1188 // Prints one-line description of a combination of sigaction.sa_flags.
print_sa_flags(outputStream * st,int flags)1189 void os::Posix::print_sa_flags(outputStream* st, int flags) {
1190 char buffer[0x100];
1191 os::Posix::describe_sa_flags(flags, buffer, sizeof(buffer));
1192 st->print("%s", buffer);
1193 }
1194
1195 // Helper function for os::Posix::print_siginfo_...():
1196 // return a textual description for signal code.
1197 struct enum_sigcode_desc_t {
1198 const char* s_name;
1199 const char* s_desc;
1200 };
1201
get_signal_code_description(const siginfo_t * si,enum_sigcode_desc_t * out)1202 static bool get_signal_code_description(const siginfo_t* si, enum_sigcode_desc_t* out) {
1203
1204 const struct {
1205 int sig; int code; const char* s_code; const char* s_desc;
1206 } t1 [] = {
1207 { SIGILL, ILL_ILLOPC, "ILL_ILLOPC", "Illegal opcode." },
1208 { SIGILL, ILL_ILLOPN, "ILL_ILLOPN", "Illegal operand." },
1209 { SIGILL, ILL_ILLADR, "ILL_ILLADR", "Illegal addressing mode." },
1210 { SIGILL, ILL_ILLTRP, "ILL_ILLTRP", "Illegal trap." },
1211 { SIGILL, ILL_PRVOPC, "ILL_PRVOPC", "Privileged opcode." },
1212 { SIGILL, ILL_PRVREG, "ILL_PRVREG", "Privileged register." },
1213 { SIGILL, ILL_COPROC, "ILL_COPROC", "Coprocessor error." },
1214 { SIGILL, ILL_BADSTK, "ILL_BADSTK", "Internal stack error." },
1215 #if defined(IA64) && defined(LINUX)
1216 { SIGILL, ILL_BADIADDR, "ILL_BADIADDR", "Unimplemented instruction address" },
1217 { SIGILL, ILL_BREAK, "ILL_BREAK", "Application Break instruction" },
1218 #endif
1219 { SIGFPE, FPE_INTDIV, "FPE_INTDIV", "Integer divide by zero." },
1220 { SIGFPE, FPE_INTOVF, "FPE_INTOVF", "Integer overflow." },
1221 { SIGFPE, FPE_FLTDIV, "FPE_FLTDIV", "Floating-point divide by zero." },
1222 { SIGFPE, FPE_FLTOVF, "FPE_FLTOVF", "Floating-point overflow." },
1223 { SIGFPE, FPE_FLTUND, "FPE_FLTUND", "Floating-point underflow." },
1224 { SIGFPE, FPE_FLTRES, "FPE_FLTRES", "Floating-point inexact result." },
1225 { SIGFPE, FPE_FLTINV, "FPE_FLTINV", "Invalid floating-point operation." },
1226 { SIGFPE, FPE_FLTSUB, "FPE_FLTSUB", "Subscript out of range." },
1227 { SIGSEGV, SEGV_MAPERR, "SEGV_MAPERR", "Address not mapped to object." },
1228 { SIGSEGV, SEGV_ACCERR, "SEGV_ACCERR", "Invalid permissions for mapped object." },
1229 #ifdef AIX
1230 // no explanation found what keyerr would be
1231 { SIGSEGV, SEGV_KEYERR, "SEGV_KEYERR", "key error" },
1232 #endif
1233 #if defined(IA64) && !defined(AIX)
1234 { SIGSEGV, SEGV_PSTKOVF, "SEGV_PSTKOVF", "Paragraph stack overflow" },
1235 #endif
1236 #if defined(__sparc) && defined(SOLARIS)
1237 // define Solaris Sparc M7 ADI SEGV signals
1238 #if !defined(SEGV_ACCADI)
1239 #define SEGV_ACCADI 3
1240 #endif
1241 { SIGSEGV, SEGV_ACCADI, "SEGV_ACCADI", "ADI not enabled for mapped object." },
1242 #if !defined(SEGV_ACCDERR)
1243 #define SEGV_ACCDERR 4
1244 #endif
1245 { SIGSEGV, SEGV_ACCDERR, "SEGV_ACCDERR", "ADI disrupting exception." },
1246 #if !defined(SEGV_ACCPERR)
1247 #define SEGV_ACCPERR 5
1248 #endif
1249 { SIGSEGV, SEGV_ACCPERR, "SEGV_ACCPERR", "ADI precise exception." },
1250 #endif // defined(__sparc) && defined(SOLARIS)
1251 { SIGBUS, BUS_ADRALN, "BUS_ADRALN", "Invalid address alignment." },
1252 { SIGBUS, BUS_ADRERR, "BUS_ADRERR", "Nonexistent physical address." },
1253 { SIGBUS, BUS_OBJERR, "BUS_OBJERR", "Object-specific hardware error." },
1254 { SIGTRAP, TRAP_BRKPT, "TRAP_BRKPT", "Process breakpoint." },
1255 { SIGTRAP, TRAP_TRACE, "TRAP_TRACE", "Process trace trap." },
1256 { SIGCHLD, CLD_EXITED, "CLD_EXITED", "Child has exited." },
1257 { SIGCHLD, CLD_KILLED, "CLD_KILLED", "Child has terminated abnormally and did not create a core file." },
1258 { SIGCHLD, CLD_DUMPED, "CLD_DUMPED", "Child has terminated abnormally and created a core file." },
1259 { SIGCHLD, CLD_TRAPPED, "CLD_TRAPPED", "Traced child has trapped." },
1260 { SIGCHLD, CLD_STOPPED, "CLD_STOPPED", "Child has stopped." },
1261 { SIGCHLD, CLD_CONTINUED,"CLD_CONTINUED","Stopped child has continued." },
1262 #ifdef SIGPOLL
1263 { SIGPOLL, POLL_OUT, "POLL_OUT", "Output buffers available." },
1264 { SIGPOLL, POLL_MSG, "POLL_MSG", "Input message available." },
1265 { SIGPOLL, POLL_ERR, "POLL_ERR", "I/O error." },
1266 { SIGPOLL, POLL_PRI, "POLL_PRI", "High priority input available." },
1267 { SIGPOLL, POLL_HUP, "POLL_HUP", "Device disconnected. [Option End]" },
1268 #endif
1269 { -1, -1, NULL, NULL }
1270 };
1271
1272 // Codes valid in any signal context.
1273 const struct {
1274 int code; const char* s_code; const char* s_desc;
1275 } t2 [] = {
1276 { SI_USER, "SI_USER", "Signal sent by kill()." },
1277 { SI_QUEUE, "SI_QUEUE", "Signal sent by the sigqueue()." },
1278 { SI_TIMER, "SI_TIMER", "Signal generated by expiration of a timer set by timer_settime()." },
1279 #ifdef SI_ASYNCIO
1280 { SI_ASYNCIO, "SI_ASYNCIO", "Signal generated by completion of an asynchronous I/O request." },
1281 #endif
1282 #ifdef SI_MESGQ
1283 { SI_MESGQ, "SI_MESGQ", "Signal generated by arrival of a message on an empty message queue." },
1284 #endif
1285 // Linux specific
1286 #ifdef SI_TKILL
1287 { SI_TKILL, "SI_TKILL", "Signal sent by tkill (pthread_kill)" },
1288 #endif
1289 #ifdef SI_DETHREAD
1290 { SI_DETHREAD, "SI_DETHREAD", "Signal sent by execve() killing subsidiary threads" },
1291 #endif
1292 #ifdef SI_KERNEL
1293 { SI_KERNEL, "SI_KERNEL", "Signal sent by kernel." },
1294 #endif
1295 #ifdef SI_SIGIO
1296 { SI_SIGIO, "SI_SIGIO", "Signal sent by queued SIGIO" },
1297 #endif
1298
1299 #ifdef AIX
1300 { SI_UNDEFINED, "SI_UNDEFINED","siginfo contains partial information" },
1301 { SI_EMPTY, "SI_EMPTY", "siginfo contains no useful information" },
1302 #endif
1303
1304 #ifdef __sun
1305 { SI_NOINFO, "SI_NOINFO", "No signal information" },
1306 { SI_RCTL, "SI_RCTL", "kernel generated signal via rctl action" },
1307 { SI_LWP, "SI_LWP", "Signal sent via lwp_kill" },
1308 #endif
1309
1310 { -1, NULL, NULL }
1311 };
1312
1313 const char* s_code = NULL;
1314 const char* s_desc = NULL;
1315
1316 for (int i = 0; t1[i].sig != -1; i ++) {
1317 if (t1[i].sig == si->si_signo && t1[i].code == si->si_code) {
1318 s_code = t1[i].s_code;
1319 s_desc = t1[i].s_desc;
1320 break;
1321 }
1322 }
1323
1324 if (s_code == NULL) {
1325 for (int i = 0; t2[i].s_code != NULL; i ++) {
1326 if (t2[i].code == si->si_code) {
1327 s_code = t2[i].s_code;
1328 s_desc = t2[i].s_desc;
1329 }
1330 }
1331 }
1332
1333 if (s_code == NULL) {
1334 out->s_name = "unknown";
1335 out->s_desc = "unknown";
1336 return false;
1337 }
1338
1339 out->s_name = s_code;
1340 out->s_desc = s_desc;
1341
1342 return true;
1343 }
1344
signal_sent_by_kill(const void * siginfo)1345 bool os::signal_sent_by_kill(const void* siginfo) {
1346 const siginfo_t* const si = (const siginfo_t*)siginfo;
1347 return si->si_code == SI_USER || si->si_code == SI_QUEUE
1348 #ifdef SI_TKILL
1349 || si->si_code == SI_TKILL
1350 #endif
1351 ;
1352 }
1353
print_siginfo(outputStream * os,const void * si0)1354 void os::print_siginfo(outputStream* os, const void* si0) {
1355
1356 const siginfo_t* const si = (const siginfo_t*) si0;
1357
1358 char buf[20];
1359 os->print("siginfo:");
1360
1361 if (!si) {
1362 os->print(" <null>");
1363 return;
1364 }
1365
1366 const int sig = si->si_signo;
1367
1368 os->print(" si_signo: %d (%s)", sig, os::Posix::get_signal_name(sig, buf, sizeof(buf)));
1369
1370 enum_sigcode_desc_t ed;
1371 get_signal_code_description(si, &ed);
1372 os->print(", si_code: %d (%s)", si->si_code, ed.s_name);
1373
1374 if (si->si_errno) {
1375 os->print(", si_errno: %d", si->si_errno);
1376 }
1377
1378 // Output additional information depending on the signal code.
1379
1380 // Note: Many implementations lump si_addr, si_pid, si_uid etc. together as unions,
1381 // so it depends on the context which member to use. For synchronous error signals,
1382 // we print si_addr, unless the signal was sent by another process or thread, in
1383 // which case we print out pid or tid of the sender.
1384 if (signal_sent_by_kill(si)) {
1385 const pid_t pid = si->si_pid;
1386 os->print(", si_pid: %ld", (long) pid);
1387 if (IS_VALID_PID(pid)) {
1388 const pid_t me = getpid();
1389 if (me == pid) {
1390 os->print(" (current process)");
1391 }
1392 } else {
1393 os->print(" (invalid)");
1394 }
1395 os->print(", si_uid: %ld", (long) si->si_uid);
1396 if (sig == SIGCHLD) {
1397 os->print(", si_status: %d", si->si_status);
1398 }
1399 } else if (sig == SIGSEGV || sig == SIGBUS || sig == SIGILL ||
1400 sig == SIGTRAP || sig == SIGFPE) {
1401 os->print(", si_addr: " PTR_FORMAT, p2i(si->si_addr));
1402 #ifdef SIGPOLL
1403 } else if (sig == SIGPOLL) {
1404 os->print(", si_band: %ld", si->si_band);
1405 #endif
1406 }
1407
1408 }
1409
signal_thread(Thread * thread,int sig,const char * reason)1410 bool os::signal_thread(Thread* thread, int sig, const char* reason) {
1411 OSThread* osthread = thread->osthread();
1412 if (osthread) {
1413 #if defined (SOLARIS)
1414 // Note: we cannot use pthread_kill on Solaris - not because
1415 // its missing, but because we do not have the pthread_t id.
1416 int status = thr_kill(osthread->thread_id(), sig);
1417 #else
1418 int status = pthread_kill(osthread->pthread_id(), sig);
1419 #endif
1420 if (status == 0) {
1421 Events::log(Thread::current(), "sent signal %d to Thread " INTPTR_FORMAT " because %s.",
1422 sig, p2i(thread), reason);
1423 return true;
1424 }
1425 }
1426 return false;
1427 }
1428
unblock_thread_signal_mask(const sigset_t * set)1429 int os::Posix::unblock_thread_signal_mask(const sigset_t *set) {
1430 return pthread_sigmask(SIG_UNBLOCK, set, NULL);
1431 }
1432
ucontext_get_pc(const ucontext_t * ctx)1433 address os::Posix::ucontext_get_pc(const ucontext_t* ctx) {
1434 #if defined(AIX)
1435 return Aix::ucontext_get_pc(ctx);
1436 #elif defined(BSD)
1437 return Bsd::ucontext_get_pc(ctx);
1438 #elif defined(LINUX)
1439 return Linux::ucontext_get_pc(ctx);
1440 #elif defined(SOLARIS)
1441 return Solaris::ucontext_get_pc(ctx);
1442 #else
1443 VMError::report_and_die("unimplemented ucontext_get_pc");
1444 #endif
1445 }
1446
ucontext_set_pc(ucontext_t * ctx,address pc)1447 void os::Posix::ucontext_set_pc(ucontext_t* ctx, address pc) {
1448 #if defined(AIX)
1449 Aix::ucontext_set_pc(ctx, pc);
1450 #elif defined(BSD)
1451 Bsd::ucontext_set_pc(ctx, pc);
1452 #elif defined(LINUX)
1453 Linux::ucontext_set_pc(ctx, pc);
1454 #elif defined(SOLARIS)
1455 Solaris::ucontext_set_pc(ctx, pc);
1456 #else
1457 VMError::report_and_die("unimplemented ucontext_get_pc");
1458 #endif
1459 }
1460
describe_pthread_attr(char * buf,size_t buflen,const pthread_attr_t * attr)1461 char* os::Posix::describe_pthread_attr(char* buf, size_t buflen, const pthread_attr_t* attr) {
1462 size_t stack_size = 0;
1463 size_t guard_size = 0;
1464 int detachstate = 0;
1465 pthread_attr_getstacksize(attr, &stack_size);
1466 pthread_attr_getguardsize(attr, &guard_size);
1467 // Work around linux NPTL implementation error, see also os::create_thread() in os_linux.cpp.
1468 LINUX_ONLY(stack_size -= guard_size);
1469 pthread_attr_getdetachstate(attr, &detachstate);
1470 jio_snprintf(buf, buflen, "stacksize: " SIZE_FORMAT "k, guardsize: " SIZE_FORMAT "k, %s",
1471 stack_size / 1024, guard_size / 1024,
1472 (detachstate == PTHREAD_CREATE_DETACHED ? "detached" : "joinable"));
1473 return buf;
1474 }
1475
realpath(const char * filename,char * outbuf,size_t outbuflen)1476 char* os::Posix::realpath(const char* filename, char* outbuf, size_t outbuflen) {
1477
1478 if (filename == NULL || outbuf == NULL || outbuflen < 1) {
1479 assert(false, "os::Posix::realpath: invalid arguments.");
1480 errno = EINVAL;
1481 return NULL;
1482 }
1483
1484 char* result = NULL;
1485
1486 // This assumes platform realpath() is implemented according to POSIX.1-2008.
1487 // POSIX.1-2008 allows to specify NULL for the output buffer, in which case
1488 // output buffer is dynamically allocated and must be ::free()'d by the caller.
1489 char* p = ::realpath(filename, NULL);
1490 if (p != NULL) {
1491 if (strlen(p) < outbuflen) {
1492 strcpy(outbuf, p);
1493 result = outbuf;
1494 } else {
1495 errno = ENAMETOOLONG;
1496 }
1497 ::free(p); // *not* os::free
1498 } else {
1499 // Fallback for platforms struggling with modern Posix standards (AIX 5.3, 6.1). If realpath
1500 // returns EINVAL, this may indicate that realpath is not POSIX.1-2008 compatible and
1501 // that it complains about the NULL we handed down as user buffer.
1502 // In this case, use the user provided buffer but at least check whether realpath caused
1503 // a memory overwrite.
1504 if (errno == EINVAL) {
1505 outbuf[outbuflen - 1] = '\0';
1506 p = ::realpath(filename, outbuf);
1507 if (p != NULL) {
1508 guarantee(outbuf[outbuflen - 1] == '\0', "realpath buffer overwrite detected.");
1509 result = p;
1510 }
1511 }
1512 }
1513 return result;
1514
1515 }
1516
stat(const char * path,struct stat * sbuf)1517 int os::stat(const char *path, struct stat *sbuf) {
1518 return ::stat(path, sbuf);
1519 }
1520
native_path(char * path)1521 char * os::native_path(char *path) {
1522 return path;
1523 }
1524
1525 // Check minimum allowable stack sizes for thread creation and to initialize
1526 // the java system classes, including StackOverflowError - depends on page
1527 // size.
1528 // The space needed for frames during startup is platform dependent. It
1529 // depends on word size, platform calling conventions, C frame layout and
1530 // interpreter/C1/C2 design decisions. Therefore this is given in a
1531 // platform (os/cpu) dependent constant.
1532 // To this, space for guard mechanisms is added, which depends on the
1533 // page size which again depends on the concrete system the VM is running
1534 // on. Space for libc guard pages is not included in this size.
set_minimum_stack_sizes()1535 jint os::Posix::set_minimum_stack_sizes() {
1536 size_t os_min_stack_allowed = SOLARIS_ONLY(thr_min_stack()) NOT_SOLARIS(PTHREAD_STACK_MIN);
1537
1538 _java_thread_min_stack_allowed = _java_thread_min_stack_allowed +
1539 JavaThread::stack_guard_zone_size() +
1540 JavaThread::stack_shadow_zone_size();
1541
1542 _java_thread_min_stack_allowed = align_up(_java_thread_min_stack_allowed, vm_page_size());
1543 _java_thread_min_stack_allowed = MAX2(_java_thread_min_stack_allowed, os_min_stack_allowed);
1544
1545 size_t stack_size_in_bytes = ThreadStackSize * K;
1546 if (stack_size_in_bytes != 0 &&
1547 stack_size_in_bytes < _java_thread_min_stack_allowed) {
1548 // The '-Xss' and '-XX:ThreadStackSize=N' options both set
1549 // ThreadStackSize so we go with "Java thread stack size" instead
1550 // of "ThreadStackSize" to be more friendly.
1551 tty->print_cr("\nThe Java thread stack size specified is too small. "
1552 "Specify at least " SIZE_FORMAT "k",
1553 _java_thread_min_stack_allowed / K);
1554 return JNI_ERR;
1555 }
1556
1557 // Make the stack size a multiple of the page size so that
1558 // the yellow/red zones can be guarded.
1559 JavaThread::set_stack_size_at_create(align_up(stack_size_in_bytes, vm_page_size()));
1560
1561 // Reminder: a compiler thread is a Java thread.
1562 _compiler_thread_min_stack_allowed = _compiler_thread_min_stack_allowed +
1563 JavaThread::stack_guard_zone_size() +
1564 JavaThread::stack_shadow_zone_size();
1565
1566 _compiler_thread_min_stack_allowed = align_up(_compiler_thread_min_stack_allowed, vm_page_size());
1567 _compiler_thread_min_stack_allowed = MAX2(_compiler_thread_min_stack_allowed, os_min_stack_allowed);
1568
1569 stack_size_in_bytes = CompilerThreadStackSize * K;
1570 if (stack_size_in_bytes != 0 &&
1571 stack_size_in_bytes < _compiler_thread_min_stack_allowed) {
1572 tty->print_cr("\nThe CompilerThreadStackSize specified is too small. "
1573 "Specify at least " SIZE_FORMAT "k",
1574 _compiler_thread_min_stack_allowed / K);
1575 return JNI_ERR;
1576 }
1577
1578 _vm_internal_thread_min_stack_allowed = align_up(_vm_internal_thread_min_stack_allowed, vm_page_size());
1579 _vm_internal_thread_min_stack_allowed = MAX2(_vm_internal_thread_min_stack_allowed, os_min_stack_allowed);
1580
1581 stack_size_in_bytes = VMThreadStackSize * K;
1582 if (stack_size_in_bytes != 0 &&
1583 stack_size_in_bytes < _vm_internal_thread_min_stack_allowed) {
1584 tty->print_cr("\nThe VMThreadStackSize specified is too small. "
1585 "Specify at least " SIZE_FORMAT "k",
1586 _vm_internal_thread_min_stack_allowed / K);
1587 return JNI_ERR;
1588 }
1589 return JNI_OK;
1590 }
1591
1592 // Called when creating the thread. The minimum stack sizes have already been calculated
get_initial_stack_size(ThreadType thr_type,size_t req_stack_size)1593 size_t os::Posix::get_initial_stack_size(ThreadType thr_type, size_t req_stack_size) {
1594 size_t stack_size;
1595 if (req_stack_size == 0) {
1596 stack_size = default_stack_size(thr_type);
1597 } else {
1598 stack_size = req_stack_size;
1599 }
1600
1601 switch (thr_type) {
1602 case os::java_thread:
1603 // Java threads use ThreadStackSize which default value can be
1604 // changed with the flag -Xss
1605 if (req_stack_size == 0 && JavaThread::stack_size_at_create() > 0) {
1606 // no requested size and we have a more specific default value
1607 stack_size = JavaThread::stack_size_at_create();
1608 }
1609 stack_size = MAX2(stack_size,
1610 _java_thread_min_stack_allowed);
1611 break;
1612 case os::compiler_thread:
1613 if (req_stack_size == 0 && CompilerThreadStackSize > 0) {
1614 // no requested size and we have a more specific default value
1615 stack_size = (size_t)(CompilerThreadStackSize * K);
1616 }
1617 stack_size = MAX2(stack_size,
1618 _compiler_thread_min_stack_allowed);
1619 break;
1620 case os::vm_thread:
1621 case os::pgc_thread:
1622 case os::cgc_thread:
1623 case os::watcher_thread:
1624 default: // presume the unknown thr_type is a VM internal
1625 if (req_stack_size == 0 && VMThreadStackSize > 0) {
1626 // no requested size and we have a more specific default value
1627 stack_size = (size_t)(VMThreadStackSize * K);
1628 }
1629
1630 stack_size = MAX2(stack_size,
1631 _vm_internal_thread_min_stack_allowed);
1632 break;
1633 }
1634
1635 // pthread_attr_setstacksize() may require that the size be rounded up to the OS page size.
1636 // Be careful not to round up to 0. Align down in that case.
1637 if (stack_size <= SIZE_MAX - vm_page_size()) {
1638 stack_size = align_up(stack_size, vm_page_size());
1639 } else {
1640 stack_size = align_down(stack_size, vm_page_size());
1641 }
1642
1643 return stack_size;
1644 }
1645
is_root(uid_t uid)1646 bool os::Posix::is_root(uid_t uid){
1647 return ROOT_UID == uid;
1648 }
1649
matches_effective_uid_or_root(uid_t uid)1650 bool os::Posix::matches_effective_uid_or_root(uid_t uid) {
1651 return is_root(uid) || geteuid() == uid;
1652 }
1653
matches_effective_uid_and_gid_or_root(uid_t uid,gid_t gid)1654 bool os::Posix::matches_effective_uid_and_gid_or_root(uid_t uid, gid_t gid) {
1655 return is_root(uid) || (geteuid() == uid && getegid() == gid);
1656 }
1657
1658 Thread* os::ThreadCrashProtection::_protected_thread = NULL;
1659 os::ThreadCrashProtection* os::ThreadCrashProtection::_crash_protection = NULL;
1660 volatile intptr_t os::ThreadCrashProtection::_crash_mux = 0;
1661
ThreadCrashProtection()1662 os::ThreadCrashProtection::ThreadCrashProtection() {
1663 }
1664
1665 /*
1666 * See the caveats for this class in os_posix.hpp
1667 * Protects the callback call so that SIGSEGV / SIGBUS jumps back into this
1668 * method and returns false. If none of the signals are raised, returns true.
1669 * The callback is supposed to provide the method that should be protected.
1670 */
call(os::CrashProtectionCallback & cb)1671 bool os::ThreadCrashProtection::call(os::CrashProtectionCallback& cb) {
1672 sigset_t saved_sig_mask;
1673
1674 Thread::muxAcquire(&_crash_mux, "CrashProtection");
1675
1676 _protected_thread = Thread::current_or_null();
1677 assert(_protected_thread != NULL, "Cannot crash protect a NULL thread");
1678
1679 // we cannot rely on sigsetjmp/siglongjmp to save/restore the signal mask
1680 // since on at least some systems (OS X) siglongjmp will restore the mask
1681 // for the process, not the thread
1682 pthread_sigmask(0, NULL, &saved_sig_mask);
1683 if (sigsetjmp(_jmpbuf, 0) == 0) {
1684 // make sure we can see in the signal handler that we have crash protection
1685 // installed
1686 _crash_protection = this;
1687 cb.call();
1688 // and clear the crash protection
1689 _crash_protection = NULL;
1690 _protected_thread = NULL;
1691 Thread::muxRelease(&_crash_mux);
1692 return true;
1693 }
1694 // this happens when we siglongjmp() back
1695 pthread_sigmask(SIG_SETMASK, &saved_sig_mask, NULL);
1696 _crash_protection = NULL;
1697 _protected_thread = NULL;
1698 Thread::muxRelease(&_crash_mux);
1699 return false;
1700 }
1701
restore()1702 void os::ThreadCrashProtection::restore() {
1703 assert(_crash_protection != NULL, "must have crash protection");
1704 siglongjmp(_jmpbuf, 1);
1705 }
1706
check_crash_protection(int sig,Thread * thread)1707 void os::ThreadCrashProtection::check_crash_protection(int sig,
1708 Thread* thread) {
1709
1710 if (thread != NULL &&
1711 thread == _protected_thread &&
1712 _crash_protection != NULL) {
1713
1714 if (sig == SIGSEGV || sig == SIGBUS) {
1715 _crash_protection->restore();
1716 }
1717 }
1718 }
1719
1720
1721 // Shared pthread_mutex/cond based PlatformEvent implementation.
1722 // Not currently usable by Solaris.
1723
1724 #ifndef SOLARIS
1725
1726 // Shared condattr object for use with relative timed-waits. Will be associated
1727 // with CLOCK_MONOTONIC if available to avoid issues with time-of-day changes,
1728 // but otherwise whatever default is used by the platform - generally the
1729 // time-of-day clock.
1730 static pthread_condattr_t _condAttr[1];
1731
1732 // Shared mutexattr to explicitly set the type to PTHREAD_MUTEX_NORMAL as not
1733 // all systems (e.g. FreeBSD) map the default to "normal".
1734 static pthread_mutexattr_t _mutexAttr[1];
1735
1736 // common basic initialization that is always supported
pthread_init_common(void)1737 static void pthread_init_common(void) {
1738 int status;
1739 if ((status = pthread_condattr_init(_condAttr)) != 0) {
1740 fatal("pthread_condattr_init: %s", os::strerror(status));
1741 }
1742 if ((status = pthread_mutexattr_init(_mutexAttr)) != 0) {
1743 fatal("pthread_mutexattr_init: %s", os::strerror(status));
1744 }
1745 if ((status = pthread_mutexattr_settype(_mutexAttr, PTHREAD_MUTEX_NORMAL)) != 0) {
1746 fatal("pthread_mutexattr_settype: %s", os::strerror(status));
1747 }
1748 }
1749
1750 #ifndef SOLARIS
1751 sigset_t sigs;
1752 struct sigaction sigact[NSIG];
1753
get_preinstalled_handler(int sig)1754 struct sigaction* os::Posix::get_preinstalled_handler(int sig) {
1755 if (sigismember(&sigs, sig)) {
1756 return &sigact[sig];
1757 }
1758 return NULL;
1759 }
1760
save_preinstalled_handler(int sig,struct sigaction & oldAct)1761 void os::Posix::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
1762 assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
1763 sigact[sig] = oldAct;
1764 sigaddset(&sigs, sig);
1765 }
1766 #endif
1767
1768 // Not all POSIX types and API's are available on all notionally "posix"
1769 // platforms. If we have build-time support then we will check for actual
1770 // runtime support via dlopen/dlsym lookup. This allows for running on an
1771 // older OS version compared to the build platform. But if there is no
1772 // build time support then there cannot be any runtime support as we do not
1773 // know what the runtime types would be (for example clockid_t might be an
1774 // int or int64_t).
1775 //
1776 #ifdef SUPPORTS_CLOCK_MONOTONIC
1777
1778 // This means we have clockid_t, clock_gettime et al and CLOCK_MONOTONIC
1779
1780 static int (*_clock_gettime)(clockid_t, struct timespec *);
1781 static int (*_pthread_condattr_setclock)(pthread_condattr_t *, clockid_t);
1782
1783 static bool _use_clock_monotonic_condattr;
1784
1785 // Determine what POSIX API's are present and do appropriate
1786 // configuration.
init(void)1787 void os::Posix::init(void) {
1788
1789 // NOTE: no logging available when this is called. Put logging
1790 // statements in init_2().
1791
1792 // Copied from os::Linux::clock_init(). The duplication is temporary.
1793
1794 // 1. Check for CLOCK_MONOTONIC support.
1795
1796 void* handle = NULL;
1797
1798 // For linux we need librt, for other OS we can find
1799 // this function in regular libc.
1800 #ifdef NEEDS_LIBRT
1801 // We do dlopen's in this particular order due to bug in linux
1802 // dynamic loader (see 6348968) leading to crash on exit.
1803 handle = dlopen("librt.so.1", RTLD_LAZY);
1804 if (handle == NULL) {
1805 handle = dlopen("librt.so", RTLD_LAZY);
1806 }
1807 #endif
1808
1809 if (handle == NULL) {
1810 handle = RTLD_DEFAULT;
1811 }
1812
1813 _clock_gettime = NULL;
1814
1815 int (*clock_getres_func)(clockid_t, struct timespec*) =
1816 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1817 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1818 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1819 if (clock_getres_func != NULL && clock_gettime_func != NULL) {
1820 // We assume that if both clock_gettime and clock_getres support
1821 // CLOCK_MONOTONIC then the OS provides true high-res monotonic clock.
1822 struct timespec res;
1823 struct timespec tp;
1824 if (clock_getres_func(CLOCK_MONOTONIC, &res) == 0 &&
1825 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1826 // Yes, monotonic clock is supported.
1827 _clock_gettime = clock_gettime_func;
1828 } else {
1829 #ifdef NEEDS_LIBRT
1830 // Close librt if there is no monotonic clock.
1831 if (handle != RTLD_DEFAULT) {
1832 dlclose(handle);
1833 }
1834 #endif
1835 }
1836 }
1837
1838 // 2. Check for pthread_condattr_setclock support.
1839
1840 _pthread_condattr_setclock = NULL;
1841
1842 // libpthread is already loaded.
1843 int (*condattr_setclock_func)(pthread_condattr_t*, clockid_t) =
1844 (int (*)(pthread_condattr_t*, clockid_t))dlsym(RTLD_DEFAULT,
1845 "pthread_condattr_setclock");
1846 if (condattr_setclock_func != NULL) {
1847 _pthread_condattr_setclock = condattr_setclock_func;
1848 }
1849
1850 // Now do general initialization.
1851
1852 pthread_init_common();
1853
1854 int status;
1855 if (_pthread_condattr_setclock != NULL && _clock_gettime != NULL) {
1856 if ((status = _pthread_condattr_setclock(_condAttr, CLOCK_MONOTONIC)) != 0) {
1857 if (status == EINVAL) {
1858 _use_clock_monotonic_condattr = false;
1859 warning("Unable to use monotonic clock with relative timed-waits" \
1860 " - changes to the time-of-day clock may have adverse affects");
1861 } else {
1862 fatal("pthread_condattr_setclock: %s", os::strerror(status));
1863 }
1864 } else {
1865 _use_clock_monotonic_condattr = true;
1866 }
1867 } else {
1868 _use_clock_monotonic_condattr = false;
1869 }
1870 }
1871
init_2(void)1872 void os::Posix::init_2(void) {
1873 log_info(os)("Use of CLOCK_MONOTONIC is%s supported",
1874 (_clock_gettime != NULL ? "" : " not"));
1875 log_info(os)("Use of pthread_condattr_setclock is%s supported",
1876 (_pthread_condattr_setclock != NULL ? "" : " not"));
1877 log_info(os)("Relative timed-wait using pthread_cond_timedwait is associated with %s",
1878 _use_clock_monotonic_condattr ? "CLOCK_MONOTONIC" : "the default clock");
1879 #ifndef SOLARIS
1880 sigemptyset(&sigs);
1881 #endif
1882 }
1883
1884 #else // !SUPPORTS_CLOCK_MONOTONIC
1885
init(void)1886 void os::Posix::init(void) {
1887 pthread_init_common();
1888 }
1889
init_2(void)1890 void os::Posix::init_2(void) {
1891 log_info(os)("Use of CLOCK_MONOTONIC is not supported");
1892 log_info(os)("Use of pthread_condattr_setclock is not supported");
1893 log_info(os)("Relative timed-wait using pthread_cond_timedwait is associated with the default clock");
1894 #ifndef SOLARIS
1895 sigemptyset(&sigs);
1896 #endif
1897 }
1898
1899 #endif // SUPPORTS_CLOCK_MONOTONIC
1900
PlatformEvent()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 // Utility to convert the given timeout to an absolute timespec
1911 // (based on the appropriate clock) to use with pthread_cond_timewait.
1912 // The clock queried here must be the clock used to manage the
1913 // timeout of the condition variable.
1914 //
1915 // The passed in timeout value is either a relative time in nanoseconds
1916 // or an absolute time in milliseconds. A relative timeout will be
1917 // associated with CLOCK_MONOTONIC if available; otherwise, or if absolute,
1918 // the default time-of-day clock will be used.
1919
1920 // Given time is a 64-bit value and the time_t used in the timespec is
1921 // sometimes a signed-32-bit value we have to watch for overflow if times
1922 // way in the future are given. Further on Solaris versions
1923 // prior to 10 there is a restriction (see cond_timedwait) that the specified
1924 // number of seconds, in abstime, is less than current_time + 100000000.
1925 // As it will be over 20 years before "now + 100000000" will overflow we can
1926 // ignore overflow and just impose a hard-limit on seconds using the value
1927 // of "now + 100000000". This places a limit on the timeout of about 3.17
1928 // years from "now".
1929 //
1930 #define MAX_SECS 100000000
1931
1932 // Calculate a new absolute time that is "timeout" nanoseconds from "now".
1933 // "unit" indicates the unit of "now_part_sec" (may be nanos or micros depending
1934 // on which clock is being used).
calc_rel_time(timespec * abstime,jlong timeout,jlong now_sec,jlong now_part_sec,jlong unit)1935 static void calc_rel_time(timespec* abstime, jlong timeout, jlong now_sec,
1936 jlong now_part_sec, jlong unit) {
1937 time_t max_secs = now_sec + MAX_SECS;
1938
1939 jlong seconds = timeout / NANOUNITS;
1940 timeout %= NANOUNITS; // remaining nanos
1941
1942 if (seconds >= MAX_SECS) {
1943 // More seconds than we can add, so pin to max_secs.
1944 abstime->tv_sec = max_secs;
1945 abstime->tv_nsec = 0;
1946 } else {
1947 abstime->tv_sec = now_sec + seconds;
1948 long nanos = (now_part_sec * (NANOUNITS / unit)) + timeout;
1949 if (nanos >= NANOUNITS) { // overflow
1950 abstime->tv_sec += 1;
1951 nanos -= NANOUNITS;
1952 }
1953 abstime->tv_nsec = nanos;
1954 }
1955 }
1956
1957 // Unpack the given deadline in milliseconds since the epoch, into the given timespec.
1958 // The current time in seconds is also passed in to enforce an upper bound as discussed above.
unpack_abs_time(timespec * abstime,jlong deadline,jlong now_sec)1959 static void unpack_abs_time(timespec* abstime, jlong deadline, jlong now_sec) {
1960 time_t max_secs = now_sec + MAX_SECS;
1961
1962 jlong seconds = deadline / MILLIUNITS;
1963 jlong millis = deadline % MILLIUNITS;
1964
1965 if (seconds >= max_secs) {
1966 // Absolute seconds exceeds allowed max, so pin to max_secs.
1967 abstime->tv_sec = max_secs;
1968 abstime->tv_nsec = 0;
1969 } else {
1970 abstime->tv_sec = seconds;
1971 abstime->tv_nsec = millis * (NANOUNITS / MILLIUNITS);
1972 }
1973 }
1974
to_abstime(timespec * abstime,jlong timeout,bool isAbsolute)1975 static void to_abstime(timespec* abstime, jlong timeout, bool isAbsolute) {
1976 DEBUG_ONLY(int max_secs = MAX_SECS;)
1977
1978 if (timeout < 0) {
1979 timeout = 0;
1980 }
1981
1982 #ifdef SUPPORTS_CLOCK_MONOTONIC
1983
1984 if (_use_clock_monotonic_condattr && !isAbsolute) {
1985 struct timespec now;
1986 int status = _clock_gettime(CLOCK_MONOTONIC, &now);
1987 assert_status(status == 0, status, "clock_gettime");
1988 calc_rel_time(abstime, timeout, now.tv_sec, now.tv_nsec, NANOUNITS);
1989 DEBUG_ONLY(max_secs += now.tv_sec;)
1990 } else {
1991
1992 #else
1993
1994 { // Match the block scope.
1995
1996 #endif // SUPPORTS_CLOCK_MONOTONIC
1997
1998 // Time-of-day clock is all we can reliably use.
1999 struct timeval now;
2000 int status = gettimeofday(&now, NULL);
2001 assert_status(status == 0, errno, "gettimeofday");
2002 if (isAbsolute) {
2003 unpack_abs_time(abstime, timeout, now.tv_sec);
2004 } else {
2005 calc_rel_time(abstime, timeout, now.tv_sec, now.tv_usec, MICROUNITS);
2006 }
2007 DEBUG_ONLY(max_secs += now.tv_sec;)
2008 }
2009
2010 assert(abstime->tv_sec >= 0, "tv_sec < 0");
2011 assert(abstime->tv_sec <= max_secs, "tv_sec > max_secs");
2012 assert(abstime->tv_nsec >= 0, "tv_nsec < 0");
2013 assert(abstime->tv_nsec < NANOUNITS, "tv_nsec >= NANOUNITS");
2014 }
2015
2016 // PlatformEvent
2017 //
2018 // Assumption:
2019 // Only one parker can exist on an event, which is why we allocate
2020 // them per-thread. Multiple unparkers can coexist.
2021 //
2022 // _event serves as a restricted-range semaphore.
2023 // -1 : thread is blocked, i.e. there is a waiter
2024 // 0 : neutral: thread is running or ready,
2025 // could have been signaled after a wait started
2026 // 1 : signaled - thread is running or ready
2027 //
2028 // Having three states allows for some detection of bad usage - see
2029 // comments on unpark().
2030
2031 void os::PlatformEvent::park() { // AKA "down()"
2032 // Transitions for _event:
2033 // -1 => -1 : illegal
2034 // 1 => 0 : pass - return immediately
2035 // 0 => -1 : block; then set _event to 0 before returning
2036
2037 // Invariant: Only the thread associated with the PlatformEvent
2038 // may call park().
2039 assert(_nParked == 0, "invariant");
2040
2041 int v;
2042
2043 // atomically decrement _event
2044 for (;;) {
2045 v = _event;
2046 if (Atomic::cmpxchg(v - 1, &_event, v) == v) break;
2047 }
2048 guarantee(v >= 0, "invariant");
2049
2050 if (v == 0) { // Do this the hard way by blocking ...
2051 int status = pthread_mutex_lock(_mutex);
2052 assert_status(status == 0, status, "mutex_lock");
2053 guarantee(_nParked == 0, "invariant");
2054 ++_nParked;
2055 while (_event < 0) {
2056 // OS-level "spurious wakeups" are ignored
2057 status = pthread_cond_wait(_cond, _mutex);
2058 assert_status(status == 0, status, "cond_wait");
2059 }
2060 --_nParked;
2061
2062 _event = 0;
2063 status = pthread_mutex_unlock(_mutex);
2064 assert_status(status == 0, status, "mutex_unlock");
2065 // Paranoia to ensure our locked and lock-free paths interact
2066 // correctly with each other.
2067 OrderAccess::fence();
2068 }
2069 guarantee(_event >= 0, "invariant");
2070 }
2071
2072 int os::PlatformEvent::park(jlong millis) {
2073 // Transitions for _event:
2074 // -1 => -1 : illegal
2075 // 1 => 0 : pass - return immediately
2076 // 0 => -1 : block; then set _event to 0 before returning
2077
2078 // Invariant: Only the thread associated with the Event/PlatformEvent
2079 // may call park().
2080 assert(_nParked == 0, "invariant");
2081
2082 int v;
2083 // atomically decrement _event
2084 for (;;) {
2085 v = _event;
2086 if (Atomic::cmpxchg(v - 1, &_event, v) == v) break;
2087 }
2088 guarantee(v >= 0, "invariant");
2089
2090 if (v == 0) { // Do this the hard way by blocking ...
2091 struct timespec abst;
2092 // We have to watch for overflow when converting millis to nanos,
2093 // but if millis is that large then we will end up limiting to
2094 // MAX_SECS anyway, so just do that here.
2095 if (millis / MILLIUNITS > MAX_SECS) {
2096 millis = jlong(MAX_SECS) * MILLIUNITS;
2097 }
2098 to_abstime(&abst, millis * (NANOUNITS / MILLIUNITS), false);
2099
2100 int ret = OS_TIMEOUT;
2101 int status = pthread_mutex_lock(_mutex);
2102 assert_status(status == 0, status, "mutex_lock");
2103 guarantee(_nParked == 0, "invariant");
2104 ++_nParked;
2105
2106 while (_event < 0) {
2107 status = pthread_cond_timedwait(_cond, _mutex, &abst);
2108 assert_status(status == 0 || status == ETIMEDOUT,
2109 status, "cond_timedwait");
2110 // OS-level "spurious wakeups" are ignored unless the archaic
2111 // FilterSpuriousWakeups is set false. That flag should be obsoleted.
2112 if (!FilterSpuriousWakeups) break;
2113 if (status == ETIMEDOUT) break;
2114 }
2115 --_nParked;
2116
2117 if (_event >= 0) {
2118 ret = OS_OK;
2119 }
2120
2121 _event = 0;
2122 status = pthread_mutex_unlock(_mutex);
2123 assert_status(status == 0, status, "mutex_unlock");
2124 // Paranoia to ensure our locked and lock-free paths interact
2125 // correctly with each other.
2126 OrderAccess::fence();
2127 return ret;
2128 }
2129 return OS_OK;
2130 }
2131
2132 void os::PlatformEvent::unpark() {
2133 // Transitions for _event:
2134 // 0 => 1 : just return
2135 // 1 => 1 : just return
2136 // -1 => either 0 or 1; must signal target thread
2137 // That is, we can safely transition _event from -1 to either
2138 // 0 or 1.
2139 // See also: "Semaphores in Plan 9" by Mullender & Cox
2140 //
2141 // Note: Forcing a transition from "-1" to "1" on an unpark() means
2142 // that it will take two back-to-back park() calls for the owning
2143 // thread to block. This has the benefit of forcing a spurious return
2144 // from the first park() call after an unpark() call which will help
2145 // shake out uses of park() and unpark() without checking state conditions
2146 // properly. This spurious return doesn't manifest itself in any user code
2147 // but only in the correctly written condition checking loops of ObjectMonitor,
2148 // Mutex/Monitor, Thread::muxAcquire and os::sleep
2149
2150 if (Atomic::xchg(1, &_event) >= 0) return;
2151
2152 int status = pthread_mutex_lock(_mutex);
2153 assert_status(status == 0, status, "mutex_lock");
2154 int anyWaiters = _nParked;
2155 assert(anyWaiters == 0 || anyWaiters == 1, "invariant");
2156 status = pthread_mutex_unlock(_mutex);
2157 assert_status(status == 0, status, "mutex_unlock");
2158
2159 // Note that we signal() *after* dropping the lock for "immortal" Events.
2160 // This is safe and avoids a common class of futile wakeups. In rare
2161 // circumstances this can cause a thread to return prematurely from
2162 // cond_{timed}wait() but the spurious wakeup is benign and the victim
2163 // will simply re-test the condition and re-park itself.
2164 // This provides particular benefit if the underlying platform does not
2165 // provide wait morphing.
2166
2167 if (anyWaiters != 0) {
2168 status = pthread_cond_signal(_cond);
2169 assert_status(status == 0, status, "cond_signal");
2170 }
2171 }
2172
2173 // JSR166 support
2174
2175 os::PlatformParker::PlatformParker() {
2176 int status;
2177 status = pthread_cond_init(&_cond[REL_INDEX], _condAttr);
2178 assert_status(status == 0, status, "cond_init rel");
2179 status = pthread_cond_init(&_cond[ABS_INDEX], NULL);
2180 assert_status(status == 0, status, "cond_init abs");
2181 status = pthread_mutex_init(_mutex, _mutexAttr);
2182 assert_status(status == 0, status, "mutex_init");
2183 _cur_index = -1; // mark as unused
2184 }
2185
2186 // Parker::park decrements count if > 0, else does a condvar wait. Unpark
2187 // sets count to 1 and signals condvar. Only one thread ever waits
2188 // on the condvar. Contention seen when trying to park implies that someone
2189 // is unparking you, so don't wait. And spurious returns are fine, so there
2190 // is no need to track notifications.
2191
2192 void Parker::park(bool isAbsolute, jlong time) {
2193
2194 // Optional fast-path check:
2195 // Return immediately if a permit is available.
2196 // We depend on Atomic::xchg() having full barrier semantics
2197 // since we are doing a lock-free update to _counter.
2198 if (Atomic::xchg(0, &_counter) > 0) return;
2199
2200 Thread* thread = Thread::current();
2201 assert(thread->is_Java_thread(), "Must be JavaThread");
2202 JavaThread *jt = (JavaThread *)thread;
2203
2204 // Optional optimization -- avoid state transitions if there's
2205 // an interrupt pending.
2206 if (Thread::is_interrupted(thread, false)) {
2207 return;
2208 }
2209
2210 // Next, demultiplex/decode time arguments
2211 struct timespec absTime;
2212 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
2213 return;
2214 }
2215 if (time > 0) {
2216 to_abstime(&absTime, time, isAbsolute);
2217 }
2218
2219 // Enter safepoint region
2220 // Beware of deadlocks such as 6317397.
2221 // The per-thread Parker:: mutex is a classic leaf-lock.
2222 // In particular a thread must never block on the Threads_lock while
2223 // holding the Parker:: mutex. If safepoints are pending both the
2224 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
2225 ThreadBlockInVM tbivm(jt);
2226
2227 // Don't wait if cannot get lock since interference arises from
2228 // unparking. Also re-check interrupt before trying wait.
2229 if (Thread::is_interrupted(thread, false) ||
2230 pthread_mutex_trylock(_mutex) != 0) {
2231 return;
2232 }
2233
2234 int status;
2235 if (_counter > 0) { // no wait needed
2236 _counter = 0;
2237 status = pthread_mutex_unlock(_mutex);
2238 assert_status(status == 0, status, "invariant");
2239 // Paranoia to ensure our locked and lock-free paths interact
2240 // correctly with each other and Java-level accesses.
2241 OrderAccess::fence();
2242 return;
2243 }
2244
2245 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
2246 jt->set_suspend_equivalent();
2247 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2248
2249 assert(_cur_index == -1, "invariant");
2250 if (time == 0) {
2251 _cur_index = REL_INDEX; // arbitrary choice when not timed
2252 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
2253 assert_status(status == 0, status, "cond_timedwait");
2254 }
2255 else {
2256 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
2257 status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
2258 assert_status(status == 0 || status == ETIMEDOUT,
2259 status, "cond_timedwait");
2260 }
2261 _cur_index = -1;
2262
2263 _counter = 0;
2264 status = pthread_mutex_unlock(_mutex);
2265 assert_status(status == 0, status, "invariant");
2266 // Paranoia to ensure our locked and lock-free paths interact
2267 // correctly with each other and Java-level accesses.
2268 OrderAccess::fence();
2269
2270 // If externally suspended while waiting, re-suspend
2271 if (jt->handle_special_suspend_equivalent_condition()) {
2272 jt->java_suspend_self();
2273 }
2274 }
2275
2276 void Parker::unpark() {
2277 int status = pthread_mutex_lock(_mutex);
2278 assert_status(status == 0, status, "invariant");
2279 const int s = _counter;
2280 _counter = 1;
2281 // must capture correct index before unlocking
2282 int index = _cur_index;
2283 status = pthread_mutex_unlock(_mutex);
2284 assert_status(status == 0, status, "invariant");
2285
2286 // Note that we signal() *after* dropping the lock for "immortal" Events.
2287 // This is safe and avoids a common class of futile wakeups. In rare
2288 // circumstances this can cause a thread to return prematurely from
2289 // cond_{timed}wait() but the spurious wakeup is benign and the victim
2290 // will simply re-test the condition and re-park itself.
2291 // This provides particular benefit if the underlying platform does not
2292 // provide wait morphing.
2293
2294 if (s < 1 && index != -1) {
2295 // thread is definitely parked
2296 status = pthread_cond_signal(&_cond[index]);
2297 assert_status(status == 0, status, "invariant");
2298 }
2299 }
2300
2301
2302 #endif // !SOLARIS
2303