Tue, 13 Jan 2015 16:09:52 +0100
8069590: AIX port of "8050807: Better performing performance data handling"
Reviewed-by: simonis, goetz
Contributed-by: matthias.baesken@sap.com, martin.doerr@sap.com
1 /*
2 * Copyright (c) 2001, 2013, Oracle and/or its affiliates. All rights reserved.
3 * Copyright 2012, 2013 SAP AG. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
26 #include "precompiled.hpp"
27 #include "classfile/vmSymbols.hpp"
28 #include "memory/allocation.inline.hpp"
29 #include "memory/resourceArea.hpp"
30 #include "oops/oop.inline.hpp"
31 #include "os_aix.inline.hpp"
32 #include "runtime/handles.inline.hpp"
33 #include "runtime/perfMemory.hpp"
34 #include "services/memTracker.hpp"
35 #include "utilities/exceptions.hpp"
37 // put OS-includes here
38 # include <sys/types.h>
39 # include <sys/mman.h>
40 # include <errno.h>
41 # include <stdio.h>
42 # include <unistd.h>
43 # include <sys/stat.h>
44 # include <signal.h>
45 # include <pwd.h>
47 static char* backing_store_file_name = NULL; // name of the backing store
48 // file, if successfully created.
50 // Standard Memory Implementation Details
52 // create the PerfData memory region in standard memory.
53 //
54 static char* create_standard_memory(size_t size) {
56 // allocate an aligned chuck of memory
57 char* mapAddress = os::reserve_memory(size);
59 if (mapAddress == NULL) {
60 return NULL;
61 }
63 // commit memory
64 if (!os::commit_memory(mapAddress, size, !ExecMem)) {
65 if (PrintMiscellaneous && Verbose) {
66 warning("Could not commit PerfData memory\n");
67 }
68 os::release_memory(mapAddress, size);
69 return NULL;
70 }
72 return mapAddress;
73 }
75 // delete the PerfData memory region
76 //
77 static void delete_standard_memory(char* addr, size_t size) {
79 // there are no persistent external resources to cleanup for standard
80 // memory. since DestroyJavaVM does not support unloading of the JVM,
81 // cleanup of the memory resource is not performed. The memory will be
82 // reclaimed by the OS upon termination of the process.
83 //
84 return;
85 }
87 // save the specified memory region to the given file
88 //
89 // Note: this function might be called from signal handler (by os::abort()),
90 // don't allocate heap memory.
91 //
92 static void save_memory_to_file(char* addr, size_t size) {
94 const char* destfile = PerfMemory::get_perfdata_file_path();
95 assert(destfile[0] != '\0', "invalid PerfData file path");
97 int result;
99 RESTARTABLE(::open(destfile, O_CREAT|O_WRONLY|O_TRUNC, S_IREAD|S_IWRITE),
100 result);;
101 if (result == OS_ERR) {
102 if (PrintMiscellaneous && Verbose) {
103 warning("Could not create Perfdata save file: %s: %s\n",
104 destfile, strerror(errno));
105 }
106 } else {
107 int fd = result;
109 for (size_t remaining = size; remaining > 0;) {
111 RESTARTABLE(::write(fd, addr, remaining), result);
112 if (result == OS_ERR) {
113 if (PrintMiscellaneous && Verbose) {
114 warning("Could not write Perfdata save file: %s: %s\n",
115 destfile, strerror(errno));
116 }
117 break;
118 }
120 remaining -= (size_t)result;
121 addr += result;
122 }
124 RESTARTABLE(::close(fd), result);
125 if (PrintMiscellaneous && Verbose) {
126 if (result == OS_ERR) {
127 warning("Could not close %s: %s\n", destfile, strerror(errno));
128 }
129 }
130 }
131 FREE_C_HEAP_ARRAY(char, destfile, mtInternal);
132 }
135 // Shared Memory Implementation Details
137 // Note: the solaris and linux shared memory implementation uses the mmap
138 // interface with a backing store file to implement named shared memory.
139 // Using the file system as the name space for shared memory allows a
140 // common name space to be supported across a variety of platforms. It
141 // also provides a name space that Java applications can deal with through
142 // simple file apis.
143 //
144 // The solaris and linux implementations store the backing store file in
145 // a user specific temporary directory located in the /tmp file system,
146 // which is always a local file system and is sometimes a RAM based file
147 // system.
149 // return the user specific temporary directory name.
150 //
151 // the caller is expected to free the allocated memory.
152 //
153 static char* get_user_tmp_dir(const char* user) {
155 const char* tmpdir = os::get_temp_directory();
156 const char* perfdir = PERFDATA_NAME;
157 size_t nbytes = strlen(tmpdir) + strlen(perfdir) + strlen(user) + 3;
158 char* dirname = NEW_C_HEAP_ARRAY(char, nbytes, mtInternal);
160 // construct the path name to user specific tmp directory
161 snprintf(dirname, nbytes, "%s/%s_%s", tmpdir, perfdir, user);
163 return dirname;
164 }
166 // convert the given file name into a process id. if the file
167 // does not meet the file naming constraints, return 0.
168 //
169 static pid_t filename_to_pid(const char* filename) {
171 // a filename that doesn't begin with a digit is not a
172 // candidate for conversion.
173 //
174 if (!isdigit(*filename)) {
175 return 0;
176 }
178 // check if file name can be converted to an integer without
179 // any leftover characters.
180 //
181 char* remainder = NULL;
182 errno = 0;
183 pid_t pid = (pid_t)strtol(filename, &remainder, 10);
185 if (errno != 0) {
186 return 0;
187 }
189 // check for left over characters. If any, then the filename is
190 // not a candidate for conversion.
191 //
192 if (remainder != NULL && *remainder != '\0') {
193 return 0;
194 }
196 // successful conversion, return the pid
197 return pid;
198 }
200 // Check if the given statbuf is considered a secure directory for
201 // the backing store files. Returns true if the directory is considered
202 // a secure location. Returns false if the statbuf is a symbolic link or
203 // if an error occurred.
204 static bool is_statbuf_secure(struct stat *statp) {
205 if (S_ISLNK(statp->st_mode) || !S_ISDIR(statp->st_mode)) {
206 // The path represents a link or some non-directory file type,
207 // which is not what we expected. Declare it insecure.
208 //
209 return false;
210 }
211 // We have an existing directory, check if the permissions are safe.
212 if ((statp->st_mode & (S_IWGRP|S_IWOTH)) != 0) {
213 // The directory is open for writing and could be subjected
214 // to a symlink or a hard link attack. Declare it insecure.
215 return false;
216 }
217 // See if the uid of the directory matches the effective uid of the process.
218 //
219 if (statp->st_uid != geteuid()) {
220 // The directory was not created by this user, declare it insecure.
221 return false;
222 }
223 return true;
224 }
227 // Check if the given path is considered a secure directory for
228 // the backing store files. Returns true if the directory exists
229 // and is considered a secure location. Returns false if the path
230 // is a symbolic link or if an error occurred.
231 static bool is_directory_secure(const char* path) {
232 struct stat statbuf;
233 int result = 0;
235 RESTARTABLE(::lstat(path, &statbuf), result);
236 if (result == OS_ERR) {
237 return false;
238 }
240 // The path exists, see if it is secure.
241 return is_statbuf_secure(&statbuf);
242 }
244 // (Taken over from Solaris to support the O_NOFOLLOW case on AIX.)
245 // Check if the given directory file descriptor is considered a secure
246 // directory for the backing store files. Returns true if the directory
247 // exists and is considered a secure location. Returns false if the path
248 // is a symbolic link or if an error occurred.
249 static bool is_dirfd_secure(int dir_fd) {
250 struct stat statbuf;
251 int result = 0;
253 RESTARTABLE(::fstat(dir_fd, &statbuf), result);
254 if (result == OS_ERR) {
255 return false;
256 }
258 // The path exists, now check its mode.
259 return is_statbuf_secure(&statbuf);
260 }
263 // Check to make sure fd1 and fd2 are referencing the same file system object.
264 static bool is_same_fsobject(int fd1, int fd2) {
265 struct stat statbuf1;
266 struct stat statbuf2;
267 int result = 0;
269 RESTARTABLE(::fstat(fd1, &statbuf1), result);
270 if (result == OS_ERR) {
271 return false;
272 }
273 RESTARTABLE(::fstat(fd2, &statbuf2), result);
274 if (result == OS_ERR) {
275 return false;
276 }
278 if ((statbuf1.st_ino == statbuf2.st_ino) &&
279 (statbuf1.st_dev == statbuf2.st_dev)) {
280 return true;
281 } else {
282 return false;
283 }
284 }
286 // Helper functions for open without O_NOFOLLOW which is not present on AIX 5.3/6.1.
287 // We use the jdk6 implementation here.
288 #ifndef O_NOFOLLOW
289 // The O_NOFOLLOW oflag doesn't exist before solaris 5.10, this is to simulate that behaviour
290 // was done in jdk 5/6 hotspot by Oracle this way
291 static int open_o_nofollow_impl(const char* path, int oflag, mode_t mode, bool use_mode) {
292 struct stat orig_st;
293 struct stat new_st;
294 bool create;
295 int error;
296 int fd;
298 create = false;
300 if (lstat(path, &orig_st) != 0) {
301 if (errno == ENOENT && (oflag & O_CREAT) != 0) {
302 // File doesn't exist, but_we want to create it, add O_EXCL flag
303 // to make sure no-one creates it (or a symlink) before us
304 // This works as we expect with symlinks, from posix man page:
305 // 'If O_EXCL and O_CREAT are set, and path names a symbolic
306 // link, open() shall fail and set errno to [EEXIST]'.
307 oflag |= O_EXCL;
308 create = true;
309 } else {
310 // File doesn't exist, and we are not creating it.
311 return OS_ERR;
312 }
313 } else {
314 // Lstat success, check if existing file is a link.
315 if ((orig_st.st_mode & S_IFMT) == S_IFLNK) {
316 // File is a symlink.
317 errno = ELOOP;
318 return OS_ERR;
319 }
320 }
322 if (use_mode == true) {
323 fd = open(path, oflag, mode);
324 } else {
325 fd = open(path, oflag);
326 }
328 if (fd == OS_ERR) {
329 return fd;
330 }
332 // Can't do inode checks on before/after if we created the file.
333 if (create == false) {
334 if (fstat(fd, &new_st) != 0) {
335 // Keep errno from fstat, in case close also fails.
336 error = errno;
337 ::close(fd);
338 errno = error;
339 return OS_ERR;
340 }
342 if (orig_st.st_dev != new_st.st_dev || orig_st.st_ino != new_st.st_ino) {
343 // File was tampered with during race window.
344 ::close(fd);
345 errno = EEXIST;
346 if (PrintMiscellaneous && Verbose) {
347 warning("possible file tampering attempt detected when opening %s", path);
348 }
349 return OS_ERR;
350 }
351 }
353 return fd;
354 }
356 static int open_o_nofollow(const char* path, int oflag, mode_t mode) {
357 return open_o_nofollow_impl(path, oflag, mode, true);
358 }
360 static int open_o_nofollow(const char* path, int oflag) {
361 return open_o_nofollow_impl(path, oflag, 0, false);
362 }
363 #endif
365 // Open the directory of the given path and validate it.
366 // Return a DIR * of the open directory.
367 static DIR *open_directory_secure(const char* dirname) {
368 // Open the directory using open() so that it can be verified
369 // to be secure by calling is_dirfd_secure(), opendir() and then check
370 // to see if they are the same file system object. This method does not
371 // introduce a window of opportunity for the directory to be attacked that
372 // calling opendir() and is_directory_secure() does.
373 int result;
374 DIR *dirp = NULL;
376 // No O_NOFOLLOW defined at buildtime, and it is not documented for open;
377 // so provide a workaround in this case.
378 #ifdef O_NOFOLLOW
379 RESTARTABLE(::open(dirname, O_RDONLY|O_NOFOLLOW), result);
380 #else
381 // workaround (jdk6 coding)
382 RESTARTABLE(::open_o_nofollow(dirname, O_RDONLY), result);
383 #endif
385 if (result == OS_ERR) {
386 // Directory doesn't exist or is a symlink, so there is nothing to cleanup.
387 if (PrintMiscellaneous && Verbose) {
388 if (errno == ELOOP) {
389 warning("directory %s is a symlink and is not secure\n", dirname);
390 } else {
391 warning("could not open directory %s: %s\n", dirname, strerror(errno));
392 }
393 }
394 return dirp;
395 }
396 int fd = result;
398 // Determine if the open directory is secure.
399 if (!is_dirfd_secure(fd)) {
400 // The directory is not a secure directory.
401 os::close(fd);
402 return dirp;
403 }
405 // Open the directory.
406 dirp = ::opendir(dirname);
407 if (dirp == NULL) {
408 // The directory doesn't exist, close fd and return.
409 os::close(fd);
410 return dirp;
411 }
413 // Check to make sure fd and dirp are referencing the same file system object.
414 if (!is_same_fsobject(fd, dirp->dd_fd)) {
415 // The directory is not secure.
416 os::close(fd);
417 os::closedir(dirp);
418 dirp = NULL;
419 return dirp;
420 }
422 // Close initial open now that we know directory is secure
423 os::close(fd);
425 return dirp;
426 }
428 // NOTE: The code below uses fchdir(), open() and unlink() because
429 // fdopendir(), openat() and unlinkat() are not supported on all
430 // versions. Once the support for fdopendir(), openat() and unlinkat()
431 // is available on all supported versions the code can be changed
432 // to use these functions.
434 // Open the directory of the given path, validate it and set the
435 // current working directory to it.
436 // Return a DIR * of the open directory and the saved cwd fd.
437 //
438 static DIR *open_directory_secure_cwd(const char* dirname, int *saved_cwd_fd) {
440 // Open the directory.
441 DIR* dirp = open_directory_secure(dirname);
442 if (dirp == NULL) {
443 // Directory doesn't exist or is insecure, so there is nothing to cleanup.
444 return dirp;
445 }
446 int fd = dirp->dd_fd;
448 // Open a fd to the cwd and save it off.
449 int result;
450 RESTARTABLE(::open(".", O_RDONLY), result);
451 if (result == OS_ERR) {
452 *saved_cwd_fd = -1;
453 } else {
454 *saved_cwd_fd = result;
455 }
457 // Set the current directory to dirname by using the fd of the directory.
458 result = fchdir(fd);
460 return dirp;
461 }
463 // Close the directory and restore the current working directory.
464 static void close_directory_secure_cwd(DIR* dirp, int saved_cwd_fd) {
466 int result;
467 // If we have a saved cwd change back to it and close the fd.
468 if (saved_cwd_fd != -1) {
469 result = fchdir(saved_cwd_fd);
470 ::close(saved_cwd_fd);
471 }
473 // Close the directory.
474 os::closedir(dirp);
475 }
477 // Check if the given file descriptor is considered a secure.
478 static bool is_file_secure(int fd, const char *filename) {
480 int result;
481 struct stat statbuf;
483 // Determine if the file is secure.
484 RESTARTABLE(::fstat(fd, &statbuf), result);
485 if (result == OS_ERR) {
486 if (PrintMiscellaneous && Verbose) {
487 warning("fstat failed on %s: %s\n", filename, strerror(errno));
488 }
489 return false;
490 }
491 if (statbuf.st_nlink > 1) {
492 // A file with multiple links is not expected.
493 if (PrintMiscellaneous && Verbose) {
494 warning("file %s has multiple links\n", filename);
495 }
496 return false;
497 }
498 return true;
499 }
501 // Return the user name for the given user id.
502 //
503 // The caller is expected to free the allocated memory.
504 static char* get_user_name(uid_t uid) {
506 struct passwd pwent;
508 // Determine the max pwbuf size from sysconf, and hardcode
509 // a default if this not available through sysconf.
510 long bufsize = sysconf(_SC_GETPW_R_SIZE_MAX);
511 if (bufsize == -1)
512 bufsize = 1024;
514 char* pwbuf = NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
516 // POSIX interface to getpwuid_r is used on LINUX
517 struct passwd* p;
518 int result = getpwuid_r(uid, &pwent, pwbuf, (size_t)bufsize, &p);
520 if (result != 0 || p == NULL || p->pw_name == NULL || *(p->pw_name) == '\0') {
521 if (PrintMiscellaneous && Verbose) {
522 if (result != 0) {
523 warning("Could not retrieve passwd entry: %s\n",
524 strerror(result));
525 }
526 else if (p == NULL) {
527 // this check is added to protect against an observed problem
528 // with getpwuid_r() on RedHat 9 where getpwuid_r returns 0,
529 // indicating success, but has p == NULL. This was observed when
530 // inserting a file descriptor exhaustion fault prior to the call
531 // getpwuid_r() call. In this case, error is set to the appropriate
532 // error condition, but this is undocumented behavior. This check
533 // is safe under any condition, but the use of errno in the output
534 // message may result in an erroneous message.
535 // Bug Id 89052 was opened with RedHat.
536 //
537 warning("Could not retrieve passwd entry: %s\n",
538 strerror(errno));
539 }
540 else {
541 warning("Could not determine user name: %s\n",
542 p->pw_name == NULL ? "pw_name = NULL" :
543 "pw_name zero length");
544 }
545 }
546 FREE_C_HEAP_ARRAY(char, pwbuf, mtInternal);
547 return NULL;
548 }
550 char* user_name = NEW_C_HEAP_ARRAY(char, strlen(p->pw_name) + 1, mtInternal);
551 strcpy(user_name, p->pw_name);
553 FREE_C_HEAP_ARRAY(char, pwbuf, mtInternal);
554 return user_name;
555 }
557 // return the name of the user that owns the process identified by vmid.
558 //
559 // This method uses a slow directory search algorithm to find the backing
560 // store file for the specified vmid and returns the user name, as determined
561 // by the user name suffix of the hsperfdata_<username> directory name.
562 //
563 // the caller is expected to free the allocated memory.
564 //
565 static char* get_user_name_slow(int vmid, TRAPS) {
567 // short circuit the directory search if the process doesn't even exist.
568 if (kill(vmid, 0) == OS_ERR) {
569 if (errno == ESRCH) {
570 THROW_MSG_0(vmSymbols::java_lang_IllegalArgumentException(),
571 "Process not found");
572 }
573 else /* EPERM */ {
574 THROW_MSG_0(vmSymbols::java_io_IOException(), strerror(errno));
575 }
576 }
578 // directory search
579 char* oldest_user = NULL;
580 time_t oldest_ctime = 0;
582 const char* tmpdirname = os::get_temp_directory();
584 DIR* tmpdirp = os::opendir(tmpdirname);
586 if (tmpdirp == NULL) {
587 return NULL;
588 }
590 // for each entry in the directory that matches the pattern hsperfdata_*,
591 // open the directory and check if the file for the given vmid exists.
592 // The file with the expected name and the latest creation date is used
593 // to determine the user name for the process id.
594 //
595 struct dirent* dentry;
596 char* tdbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(tmpdirname), mtInternal);
597 errno = 0;
598 while ((dentry = os::readdir(tmpdirp, (struct dirent *)tdbuf)) != NULL) {
600 // check if the directory entry is a hsperfdata file
601 if (strncmp(dentry->d_name, PERFDATA_NAME, strlen(PERFDATA_NAME)) != 0) {
602 continue;
603 }
605 char* usrdir_name = NEW_C_HEAP_ARRAY(char,
606 strlen(tmpdirname) + strlen(dentry->d_name) + 2, mtInternal);
607 strcpy(usrdir_name, tmpdirname);
608 strcat(usrdir_name, "/");
609 strcat(usrdir_name, dentry->d_name);
611 // Open the user directory.
612 DIR* subdirp = open_directory_secure(usrdir_name);
614 if (subdirp == NULL) {
615 FREE_C_HEAP_ARRAY(char, usrdir_name, mtInternal);
616 continue;
617 }
619 // Since we don't create the backing store files in directories
620 // pointed to by symbolic links, we also don't follow them when
621 // looking for the files. We check for a symbolic link after the
622 // call to opendir in order to eliminate a small window where the
623 // symlink can be exploited.
624 //
625 if (!is_directory_secure(usrdir_name)) {
626 FREE_C_HEAP_ARRAY(char, usrdir_name, mtInternal);
627 os::closedir(subdirp);
628 continue;
629 }
631 struct dirent* udentry;
632 char* udbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(usrdir_name), mtInternal);
633 errno = 0;
634 while ((udentry = os::readdir(subdirp, (struct dirent *)udbuf)) != NULL) {
636 if (filename_to_pid(udentry->d_name) == vmid) {
637 struct stat statbuf;
638 int result;
640 char* filename = NEW_C_HEAP_ARRAY(char,
641 strlen(usrdir_name) + strlen(udentry->d_name) + 2, mtInternal);
643 strcpy(filename, usrdir_name);
644 strcat(filename, "/");
645 strcat(filename, udentry->d_name);
647 // don't follow symbolic links for the file
648 RESTARTABLE(::lstat(filename, &statbuf), result);
649 if (result == OS_ERR) {
650 FREE_C_HEAP_ARRAY(char, filename, mtInternal);
651 continue;
652 }
654 // skip over files that are not regular files.
655 if (!S_ISREG(statbuf.st_mode)) {
656 FREE_C_HEAP_ARRAY(char, filename, mtInternal);
657 continue;
658 }
660 // compare and save filename with latest creation time
661 if (statbuf.st_size > 0 && statbuf.st_ctime > oldest_ctime) {
663 if (statbuf.st_ctime > oldest_ctime) {
664 char* user = strchr(dentry->d_name, '_') + 1;
666 if (oldest_user != NULL) FREE_C_HEAP_ARRAY(char, oldest_user, mtInternal);
667 oldest_user = NEW_C_HEAP_ARRAY(char, strlen(user)+1, mtInternal);
669 strcpy(oldest_user, user);
670 oldest_ctime = statbuf.st_ctime;
671 }
672 }
674 FREE_C_HEAP_ARRAY(char, filename, mtInternal);
675 }
676 }
677 os::closedir(subdirp);
678 FREE_C_HEAP_ARRAY(char, udbuf, mtInternal);
679 FREE_C_HEAP_ARRAY(char, usrdir_name, mtInternal);
680 }
681 os::closedir(tmpdirp);
682 FREE_C_HEAP_ARRAY(char, tdbuf, mtInternal);
684 return(oldest_user);
685 }
687 // return the name of the user that owns the JVM indicated by the given vmid.
688 //
689 static char* get_user_name(int vmid, TRAPS) {
690 return get_user_name_slow(vmid, CHECK_NULL);
691 }
693 // return the file name of the backing store file for the named
694 // shared memory region for the given user name and vmid.
695 //
696 // the caller is expected to free the allocated memory.
697 //
698 static char* get_sharedmem_filename(const char* dirname, int vmid) {
700 // add 2 for the file separator and a null terminator.
701 size_t nbytes = strlen(dirname) + UINT_CHARS + 2;
703 char* name = NEW_C_HEAP_ARRAY(char, nbytes, mtInternal);
704 snprintf(name, nbytes, "%s/%d", dirname, vmid);
706 return name;
707 }
710 // remove file
711 //
712 // this method removes the file specified by the given path
713 //
714 static void remove_file(const char* path) {
716 int result;
718 // if the file is a directory, the following unlink will fail. since
719 // we don't expect to find directories in the user temp directory, we
720 // won't try to handle this situation. even if accidentially or
721 // maliciously planted, the directory's presence won't hurt anything.
722 //
723 RESTARTABLE(::unlink(path), result);
724 if (PrintMiscellaneous && Verbose && result == OS_ERR) {
725 if (errno != ENOENT) {
726 warning("Could not unlink shared memory backing"
727 " store file %s : %s\n", path, strerror(errno));
728 }
729 }
730 }
732 // Cleanup stale shared memory resources
733 //
734 // This method attempts to remove all stale shared memory files in
735 // the named user temporary directory. It scans the named directory
736 // for files matching the pattern ^$[0-9]*$. For each file found, the
737 // process id is extracted from the file name and a test is run to
738 // determine if the process is alive. If the process is not alive,
739 // any stale file resources are removed.
740 static void cleanup_sharedmem_resources(const char* dirname) {
742 int saved_cwd_fd;
743 // Open the directory.
744 DIR* dirp = open_directory_secure_cwd(dirname, &saved_cwd_fd);
745 if (dirp == NULL) {
746 // Directory doesn't exist or is insecure, so there is nothing to cleanup.
747 return;
748 }
750 // For each entry in the directory that matches the expected file
751 // name pattern, determine if the file resources are stale and if
752 // so, remove the file resources. Note, instrumented HotSpot processes
753 // for this user may start and/or terminate during this search and
754 // remove or create new files in this directory. The behavior of this
755 // loop under these conditions is dependent upon the implementation of
756 // opendir/readdir.
757 struct dirent* entry;
758 char* dbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(dirname), mtInternal);
760 errno = 0;
761 while ((entry = os::readdir(dirp, (struct dirent *)dbuf)) != NULL) {
763 pid_t pid = filename_to_pid(entry->d_name);
765 if (pid == 0) {
767 if (strcmp(entry->d_name, ".") != 0 && strcmp(entry->d_name, "..") != 0) {
769 // Attempt to remove all unexpected files, except "." and "..".
770 unlink(entry->d_name);
771 }
773 errno = 0;
774 continue;
775 }
777 // We now have a file name that converts to a valid integer
778 // that could represent a process id . if this process id
779 // matches the current process id or the process is not running,
780 // then remove the stale file resources.
781 //
782 // Process liveness is detected by sending signal number 0 to
783 // the process id (see kill(2)). if kill determines that the
784 // process does not exist, then the file resources are removed.
785 // if kill determines that that we don't have permission to
786 // signal the process, then the file resources are assumed to
787 // be stale and are removed because the resources for such a
788 // process should be in a different user specific directory.
789 if ((pid == os::current_process_id()) ||
790 (kill(pid, 0) == OS_ERR && (errno == ESRCH || errno == EPERM))) {
792 unlink(entry->d_name);
793 }
794 errno = 0;
795 }
797 // Close the directory and reset the current working directory.
798 close_directory_secure_cwd(dirp, saved_cwd_fd);
800 FREE_C_HEAP_ARRAY(char, dbuf, mtInternal);
801 }
803 // Make the user specific temporary directory. Returns true if
804 // the directory exists and is secure upon return. Returns false
805 // if the directory exists but is either a symlink, is otherwise
806 // insecure, or if an error occurred.
807 static bool make_user_tmp_dir(const char* dirname) {
809 // Create the directory with 0755 permissions. note that the directory
810 // will be owned by euid::egid, which may not be the same as uid::gid.
811 if (mkdir(dirname, S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH) == OS_ERR) {
812 if (errno == EEXIST) {
813 // The directory already exists and was probably created by another
814 // JVM instance. However, this could also be the result of a
815 // deliberate symlink. Verify that the existing directory is safe.
816 if (!is_directory_secure(dirname)) {
817 // Directory is not secure.
818 if (PrintMiscellaneous && Verbose) {
819 warning("%s directory is insecure\n", dirname);
820 }
821 return false;
822 }
823 }
824 else {
825 // we encountered some other failure while attempting
826 // to create the directory
827 //
828 if (PrintMiscellaneous && Verbose) {
829 warning("could not create directory %s: %s\n",
830 dirname, strerror(errno));
831 }
832 return false;
833 }
834 }
835 return true;
836 }
838 // create the shared memory file resources
839 //
840 // This method creates the shared memory file with the given size
841 // This method also creates the user specific temporary directory, if
842 // it does not yet exist.
843 //
844 static int create_sharedmem_resources(const char* dirname, const char* filename, size_t size) {
846 // make the user temporary directory
847 if (!make_user_tmp_dir(dirname)) {
848 // could not make/find the directory or the found directory
849 // was not secure
850 return -1;
851 }
853 int saved_cwd_fd;
854 // Open the directory and set the current working directory to it.
855 DIR* dirp = open_directory_secure_cwd(dirname, &saved_cwd_fd);
856 if (dirp == NULL) {
857 // Directory doesn't exist or is insecure, so cannot create shared
858 // memory file.
859 return -1;
860 }
862 // Open the filename in the current directory.
863 // Cannot use O_TRUNC here; truncation of an existing file has to happen
864 // after the is_file_secure() check below.
865 int result;
867 // No O_NOFOLLOW defined at buildtime, and it is not documented for open;
868 // so provide a workaround in this case.
869 #ifdef O_NOFOLLOW
870 RESTARTABLE(::open(filename, O_RDWR|O_CREAT|O_NOFOLLOW, S_IREAD|S_IWRITE), result);
871 #else
872 // workaround function (jdk6 code)
873 RESTARTABLE(::open_o_nofollow(filename, O_RDWR|O_CREAT, S_IREAD|S_IWRITE), result);
874 #endif
876 if (result == OS_ERR) {
877 if (PrintMiscellaneous && Verbose) {
878 if (errno == ELOOP) {
879 warning("file %s is a symlink and is not secure\n", filename);
880 } else {
881 warning("could not create file %s: %s\n", filename, strerror(errno));
882 }
883 }
884 // Close the directory and reset the current working directory.
885 close_directory_secure_cwd(dirp, saved_cwd_fd);
887 return -1;
888 }
889 // Close the directory and reset the current working directory.
890 close_directory_secure_cwd(dirp, saved_cwd_fd);
892 // save the file descriptor
893 int fd = result;
895 // Check to see if the file is secure.
896 if (!is_file_secure(fd, filename)) {
897 ::close(fd);
898 return -1;
899 }
901 // Truncate the file to get rid of any existing data.
902 RESTARTABLE(::ftruncate(fd, (off_t)0), result);
903 if (result == OS_ERR) {
904 if (PrintMiscellaneous && Verbose) {
905 warning("could not truncate shared memory file: %s\n", strerror(errno));
906 }
907 ::close(fd);
908 return -1;
909 }
910 // set the file size
911 RESTARTABLE(::ftruncate(fd, (off_t)size), result);
912 if (result == OS_ERR) {
913 if (PrintMiscellaneous && Verbose) {
914 warning("could not set shared memory file size: %s\n", strerror(errno));
915 }
916 RESTARTABLE(::close(fd), result);
917 return -1;
918 }
920 return fd;
921 }
923 // open the shared memory file for the given user and vmid. returns
924 // the file descriptor for the open file or -1 if the file could not
925 // be opened.
926 //
927 static int open_sharedmem_file(const char* filename, int oflags, TRAPS) {
929 // open the file
930 int result;
931 // No O_NOFOLLOW defined at buildtime, and it is not documented for open;
932 // so provide a workaround in this case
933 #ifdef O_NOFOLLOW
934 RESTARTABLE(::open(filename, oflags), result);
935 #else
936 RESTARTABLE(::open_o_nofollow(filename, oflags), result);
937 #endif
939 if (result == OS_ERR) {
940 if (errno == ENOENT) {
941 THROW_MSG_0(vmSymbols::java_lang_IllegalArgumentException(),
942 "Process not found");
943 }
944 else if (errno == EACCES) {
945 THROW_MSG_0(vmSymbols::java_lang_IllegalArgumentException(),
946 "Permission denied");
947 }
948 else {
949 THROW_MSG_0(vmSymbols::java_io_IOException(), strerror(errno));
950 }
951 }
952 int fd = result;
954 // Check to see if the file is secure.
955 if (!is_file_secure(fd, filename)) {
956 ::close(fd);
957 return -1;
958 }
960 return fd;
961 }
963 // create a named shared memory region. returns the address of the
964 // memory region on success or NULL on failure. A return value of
965 // NULL will ultimately disable the shared memory feature.
966 //
967 // On Solaris and Linux, the name space for shared memory objects
968 // is the file system name space.
969 //
970 // A monitoring application attaching to a JVM does not need to know
971 // the file system name of the shared memory object. However, it may
972 // be convenient for applications to discover the existence of newly
973 // created and terminating JVMs by watching the file system name space
974 // for files being created or removed.
975 //
976 static char* mmap_create_shared(size_t size) {
978 int result;
979 int fd;
980 char* mapAddress;
982 int vmid = os::current_process_id();
984 char* user_name = get_user_name(geteuid());
986 if (user_name == NULL)
987 return NULL;
989 char* dirname = get_user_tmp_dir(user_name);
990 char* filename = get_sharedmem_filename(dirname, vmid);
992 // Get the short filename.
993 char* short_filename = strrchr(filename, '/');
994 if (short_filename == NULL) {
995 short_filename = filename;
996 } else {
997 short_filename++;
998 }
1000 // cleanup any stale shared memory files
1001 cleanup_sharedmem_resources(dirname);
1003 assert(((size > 0) && (size % os::vm_page_size() == 0)),
1004 "unexpected PerfMemory region size");
1006 fd = create_sharedmem_resources(dirname, short_filename, size);
1008 FREE_C_HEAP_ARRAY(char, user_name, mtInternal);
1009 FREE_C_HEAP_ARRAY(char, dirname, mtInternal);
1011 if (fd == -1) {
1012 FREE_C_HEAP_ARRAY(char, filename, mtInternal);
1013 return NULL;
1014 }
1016 mapAddress = (char*)::mmap((char*)0, size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
1018 // attempt to close the file - restart it if it was interrupted,
1019 // but ignore other failures
1020 RESTARTABLE(::close(fd), result);
1021 assert(result != OS_ERR, "could not close file");
1023 if (mapAddress == MAP_FAILED) {
1024 if (PrintMiscellaneous && Verbose) {
1025 warning("mmap failed - %s\n", strerror(errno));
1026 }
1027 remove_file(filename);
1028 FREE_C_HEAP_ARRAY(char, filename, mtInternal);
1029 return NULL;
1030 }
1032 // save the file name for use in delete_shared_memory()
1033 backing_store_file_name = filename;
1035 // clear the shared memory region
1036 (void)::memset((void*) mapAddress, 0, size);
1038 // It does not go through os api, the operation has to record from here.
1039 MemTracker::record_virtual_memory_reserve((address)mapAddress, size, CURRENT_PC, mtInternal);
1041 return mapAddress;
1042 }
1044 // release a named shared memory region
1045 //
1046 static void unmap_shared(char* addr, size_t bytes) {
1047 // Do not rely on os::reserve_memory/os::release_memory to use mmap.
1048 // Use os::reserve_memory/os::release_memory for PerfDisableSharedMem=1, mmap/munmap for PerfDisableSharedMem=0
1049 if (::munmap(addr, bytes) == -1) {
1050 warning("perfmemory: munmap failed (%d)\n", errno);
1051 }
1052 }
1054 // create the PerfData memory region in shared memory.
1055 //
1056 static char* create_shared_memory(size_t size) {
1058 // create the shared memory region.
1059 return mmap_create_shared(size);
1060 }
1062 // delete the shared PerfData memory region
1063 //
1064 static void delete_shared_memory(char* addr, size_t size) {
1066 // cleanup the persistent shared memory resources. since DestroyJavaVM does
1067 // not support unloading of the JVM, unmapping of the memory resource is
1068 // not performed. The memory will be reclaimed by the OS upon termination of
1069 // the process. The backing store file is deleted from the file system.
1071 assert(!PerfDisableSharedMem, "shouldn't be here");
1073 if (backing_store_file_name != NULL) {
1074 remove_file(backing_store_file_name);
1075 // Don't.. Free heap memory could deadlock os::abort() if it is called
1076 // from signal handler. OS will reclaim the heap memory.
1077 // FREE_C_HEAP_ARRAY(char, backing_store_file_name, mtInternal);
1078 backing_store_file_name = NULL;
1079 }
1080 }
1082 // return the size of the file for the given file descriptor
1083 // or 0 if it is not a valid size for a shared memory file
1084 //
1085 static size_t sharedmem_filesize(int fd, TRAPS) {
1087 struct stat statbuf;
1088 int result;
1090 RESTARTABLE(::fstat(fd, &statbuf), result);
1091 if (result == OS_ERR) {
1092 if (PrintMiscellaneous && Verbose) {
1093 warning("fstat failed: %s\n", strerror(errno));
1094 }
1095 THROW_MSG_0(vmSymbols::java_io_IOException(),
1096 "Could not determine PerfMemory size");
1097 }
1099 if ((statbuf.st_size == 0) ||
1100 ((size_t)statbuf.st_size % os::vm_page_size() != 0)) {
1101 THROW_MSG_0(vmSymbols::java_lang_Exception(),
1102 "Invalid PerfMemory size");
1103 }
1105 return (size_t)statbuf.st_size;
1106 }
1108 // attach to a named shared memory region.
1109 //
1110 static void mmap_attach_shared(const char* user, int vmid, PerfMemory::PerfMemoryMode mode, char** addr, size_t* sizep, TRAPS) {
1112 char* mapAddress;
1113 int result;
1114 int fd;
1115 size_t size = 0;
1116 const char* luser = NULL;
1118 int mmap_prot;
1119 int file_flags;
1121 ResourceMark rm;
1123 // map the high level access mode to the appropriate permission
1124 // constructs for the file and the shared memory mapping.
1125 if (mode == PerfMemory::PERF_MODE_RO) {
1126 mmap_prot = PROT_READ;
1128 // No O_NOFOLLOW defined at buildtime, and it is not documented for open.
1129 #ifdef O_NOFOLLOW
1130 file_flags = O_RDONLY | O_NOFOLLOW;
1131 #else
1132 file_flags = O_RDONLY;
1133 #endif
1134 }
1135 else if (mode == PerfMemory::PERF_MODE_RW) {
1136 #ifdef LATER
1137 mmap_prot = PROT_READ | PROT_WRITE;
1138 file_flags = O_RDWR | O_NOFOLLOW;
1139 #else
1140 THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1141 "Unsupported access mode");
1142 #endif
1143 }
1144 else {
1145 THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1146 "Illegal access mode");
1147 }
1149 if (user == NULL || strlen(user) == 0) {
1150 luser = get_user_name(vmid, CHECK);
1151 }
1152 else {
1153 luser = user;
1154 }
1156 if (luser == NULL) {
1157 THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1158 "Could not map vmid to user Name");
1159 }
1161 char* dirname = get_user_tmp_dir(luser);
1163 // since we don't follow symbolic links when creating the backing
1164 // store file, we don't follow them when attaching either.
1165 //
1166 if (!is_directory_secure(dirname)) {
1167 FREE_C_HEAP_ARRAY(char, dirname, mtInternal);
1168 if (luser != user) {
1169 FREE_C_HEAP_ARRAY(char, luser, mtInternal);
1170 }
1171 THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1172 "Process not found");
1173 }
1175 char* filename = get_sharedmem_filename(dirname, vmid);
1177 // copy heap memory to resource memory. the open_sharedmem_file
1178 // method below need to use the filename, but could throw an
1179 // exception. using a resource array prevents the leak that
1180 // would otherwise occur.
1181 char* rfilename = NEW_RESOURCE_ARRAY(char, strlen(filename) + 1);
1182 strcpy(rfilename, filename);
1184 // free the c heap resources that are no longer needed
1185 if (luser != user) FREE_C_HEAP_ARRAY(char, luser, mtInternal);
1186 FREE_C_HEAP_ARRAY(char, dirname, mtInternal);
1187 FREE_C_HEAP_ARRAY(char, filename, mtInternal);
1189 // open the shared memory file for the give vmid
1190 fd = open_sharedmem_file(rfilename, file_flags, CHECK);
1191 assert(fd != OS_ERR, "unexpected value");
1193 if (*sizep == 0) {
1194 size = sharedmem_filesize(fd, CHECK);
1195 assert(size != 0, "unexpected size");
1196 } else {
1197 size = *sizep;
1198 }
1200 mapAddress = (char*)::mmap((char*)0, size, mmap_prot, MAP_SHARED, fd, 0);
1202 // attempt to close the file - restart if it gets interrupted,
1203 // but ignore other failures
1204 RESTARTABLE(::close(fd), result);
1205 assert(result != OS_ERR, "could not close file");
1207 if (mapAddress == MAP_FAILED) {
1208 if (PrintMiscellaneous && Verbose) {
1209 warning("mmap failed: %s\n", strerror(errno));
1210 }
1211 THROW_MSG(vmSymbols::java_lang_OutOfMemoryError(),
1212 "Could not map PerfMemory");
1213 }
1215 // It does not go through os api, the operation has to record from here.
1216 MemTracker::record_virtual_memory_reserve((address)mapAddress, size, CURRENT_PC, mtInternal);
1218 *addr = mapAddress;
1219 *sizep = size;
1221 if (PerfTraceMemOps) {
1222 tty->print("mapped " SIZE_FORMAT " bytes for vmid %d at "
1223 INTPTR_FORMAT "\n", size, vmid, (void*)mapAddress);
1224 }
1225 }
1230 // create the PerfData memory region
1231 //
1232 // This method creates the memory region used to store performance
1233 // data for the JVM. The memory may be created in standard or
1234 // shared memory.
1235 //
1236 void PerfMemory::create_memory_region(size_t size) {
1238 if (PerfDisableSharedMem) {
1239 // do not share the memory for the performance data.
1240 _start = create_standard_memory(size);
1241 }
1242 else {
1243 _start = create_shared_memory(size);
1244 if (_start == NULL) {
1246 // creation of the shared memory region failed, attempt
1247 // to create a contiguous, non-shared memory region instead.
1248 //
1249 if (PrintMiscellaneous && Verbose) {
1250 warning("Reverting to non-shared PerfMemory region.\n");
1251 }
1252 PerfDisableSharedMem = true;
1253 _start = create_standard_memory(size);
1254 }
1255 }
1257 if (_start != NULL) _capacity = size;
1259 }
1261 // delete the PerfData memory region
1262 //
1263 // This method deletes the memory region used to store performance
1264 // data for the JVM. The memory region indicated by the <address, size>
1265 // tuple will be inaccessible after a call to this method.
1266 //
1267 void PerfMemory::delete_memory_region() {
1269 assert((start() != NULL && capacity() > 0), "verify proper state");
1271 // If user specifies PerfDataSaveFile, it will save the performance data
1272 // to the specified file name no matter whether PerfDataSaveToFile is specified
1273 // or not. In other word, -XX:PerfDataSaveFile=.. overrides flag
1274 // -XX:+PerfDataSaveToFile.
1275 if (PerfDataSaveToFile || PerfDataSaveFile != NULL) {
1276 save_memory_to_file(start(), capacity());
1277 }
1279 if (PerfDisableSharedMem) {
1280 delete_standard_memory(start(), capacity());
1281 }
1282 else {
1283 delete_shared_memory(start(), capacity());
1284 }
1285 }
1287 // attach to the PerfData memory region for another JVM
1288 //
1289 // This method returns an <address, size> tuple that points to
1290 // a memory buffer that is kept reasonably synchronized with
1291 // the PerfData memory region for the indicated JVM. This
1292 // buffer may be kept in synchronization via shared memory
1293 // or some other mechanism that keeps the buffer updated.
1294 //
1295 // If the JVM chooses not to support the attachability feature,
1296 // this method should throw an UnsupportedOperation exception.
1297 //
1298 // This implementation utilizes named shared memory to map
1299 // the indicated process's PerfData memory region into this JVMs
1300 // address space.
1301 //
1302 void PerfMemory::attach(const char* user, int vmid, PerfMemoryMode mode, char** addrp, size_t* sizep, TRAPS) {
1304 if (vmid == 0 || vmid == os::current_process_id()) {
1305 *addrp = start();
1306 *sizep = capacity();
1307 return;
1308 }
1310 mmap_attach_shared(user, vmid, mode, addrp, sizep, CHECK);
1311 }
1313 // detach from the PerfData memory region of another JVM
1314 //
1315 // This method detaches the PerfData memory region of another
1316 // JVM, specified as an <address, size> tuple of a buffer
1317 // in this process's address space. This method may perform
1318 // arbitrary actions to accomplish the detachment. The memory
1319 // region specified by <address, size> will be inaccessible after
1320 // a call to this method.
1321 //
1322 // If the JVM chooses not to support the attachability feature,
1323 // this method should throw an UnsupportedOperation exception.
1324 //
1325 // This implementation utilizes named shared memory to detach
1326 // the indicated process's PerfData memory region from this
1327 // process's address space.
1328 //
1329 void PerfMemory::detach(char* addr, size_t bytes, TRAPS) {
1331 assert(addr != 0, "address sanity check");
1332 assert(bytes > 0, "capacity sanity check");
1334 if (PerfMemory::contains(addr) || PerfMemory::contains(addr + bytes - 1)) {
1335 // prevent accidental detachment of this process's PerfMemory region
1336 return;
1337 }
1339 unmap_shared(addr, bytes);
1340 }
1342 char* PerfMemory::backing_store_filename() {
1343 return backing_store_file_name;
1344 }