1 /*
2 * Copyright (c) 2001, 2017, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2012, 2016 SAP SE. 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 */
25
26 #include "precompiled.hpp"
27 #include "classfile/vmSymbols.hpp"
28 #include "logging/log.hpp"
29 #include "memory/allocation.inline.hpp"
30 #include "memory/resourceArea.hpp"
31 #include "oops/oop.inline.hpp"
32 #include "os_aix.inline.hpp"
33 #include "runtime/handles.inline.hpp"
34 #include "runtime/os.hpp"
35 #include "runtime/perfMemory.hpp"
36 #include "services/memTracker.hpp"
37 #include "utilities/exceptions.hpp"
38
39 // put OS-includes here
40 # include <sys/types.h>
41 # include <sys/mman.h>
42 # include <errno.h>
43 # include <stdio.h>
44 # include <unistd.h>
45 # include <sys/stat.h>
46 # include <signal.h>
47 # include <pwd.h>
48
49 static char* backing_store_file_name = NULL; // name of the backing store
50 // file, if successfully created.
51
52 // Standard Memory Implementation Details
53
54 // create the PerfData memory region in standard memory.
55 //
56 static char* create_standard_memory(size_t size) {
57
58 // allocate an aligned chuck of memory
59 char* mapAddress = os::reserve_memory(size);
60
61 if (mapAddress == NULL) {
62 return NULL;
63 }
64
65 // commit memory
66 if (!os::commit_memory(mapAddress, size, !ExecMem)) {
67 if (PrintMiscellaneous && Verbose) {
68 warning("Could not commit PerfData memory\n");
69 }
70 os::release_memory(mapAddress, size);
71 return NULL;
72 }
73
74 return mapAddress;
75 }
76
77 // delete the PerfData memory region
78 //
79 static void delete_standard_memory(char* addr, size_t size) {
80
81 // there are no persistent external resources to cleanup for standard
82 // memory. since DestroyJavaVM does not support unloading of the JVM,
83 // cleanup of the memory resource is not performed. The memory will be
84 // reclaimed by the OS upon termination of the process.
85 //
86 return;
87 }
88
89 // save the specified memory region to the given file
90 //
91 // Note: this function might be called from signal handler (by os::abort()),
92 // don't allocate heap memory.
93 //
94 static void save_memory_to_file(char* addr, size_t size) {
95
96 const char* destfile = PerfMemory::get_perfdata_file_path();
97 assert(destfile[0] != '\0', "invalid PerfData file path");
98
99 int result;
100
101 RESTARTABLE(::open(destfile, O_CREAT|O_WRONLY|O_TRUNC, S_IREAD|S_IWRITE),
102 result);;
103 if (result == OS_ERR) {
104 if (PrintMiscellaneous && Verbose) {
105 warning("Could not create Perfdata save file: %s: %s\n",
106 destfile, os::strerror(errno));
107 }
108 } else {
109 int fd = result;
110
111 for (size_t remaining = size; remaining > 0;) {
112
113 RESTARTABLE(::write(fd, addr, remaining), result);
114 if (result == OS_ERR) {
115 if (PrintMiscellaneous && Verbose) {
116 warning("Could not write Perfdata save file: %s: %s\n",
117 destfile, os::strerror(errno));
118 }
119 break;
120 }
121
122 remaining -= (size_t)result;
123 addr += result;
124 }
125
126 result = ::close(fd);
127 if (PrintMiscellaneous && Verbose) {
128 if (result == OS_ERR) {
129 warning("Could not close %s: %s\n", destfile, os::strerror(errno));
130 }
131 }
132 }
133 FREE_C_HEAP_ARRAY(char, destfile);
134 }
135
136
137 // Shared Memory Implementation Details
138
139 // Note: the solaris and linux shared memory implementation uses the mmap
140 // interface with a backing store file to implement named shared memory.
141 // Using the file system as the name space for shared memory allows a
142 // common name space to be supported across a variety of platforms. It
143 // also provides a name space that Java applications can deal with through
144 // simple file apis.
145 //
146 // The solaris and linux implementations store the backing store file in
147 // a user specific temporary directory located in the /tmp file system,
148 // which is always a local file system and is sometimes a RAM based file
149 // system.
150
151 // return the user specific temporary directory name.
152 //
153 // the caller is expected to free the allocated memory.
154 //
155 static char* get_user_tmp_dir(const char* user) {
156
157 const char* tmpdir = os::get_temp_directory();
158 const char* perfdir = PERFDATA_NAME;
159 size_t nbytes = strlen(tmpdir) + strlen(perfdir) + strlen(user) + 3;
160 char* dirname = NEW_C_HEAP_ARRAY(char, nbytes, mtInternal);
161
162 // construct the path name to user specific tmp directory
163 snprintf(dirname, nbytes, "%s/%s_%s", tmpdir, perfdir, user);
164
165 return dirname;
166 }
167
168 // convert the given file name into a process id. if the file
169 // does not meet the file naming constraints, return 0.
170 //
171 static pid_t filename_to_pid(const char* filename) {
172
173 // a filename that doesn't begin with a digit is not a
174 // candidate for conversion.
175 //
176 if (!isdigit(*filename)) {
177 return 0;
178 }
179
180 // check if file name can be converted to an integer without
181 // any leftover characters.
182 //
183 char* remainder = NULL;
184 errno = 0;
185 pid_t pid = (pid_t)strtol(filename, &remainder, 10);
186
187 if (errno != 0) {
188 return 0;
189 }
190
191 // check for left over characters. If any, then the filename is
192 // not a candidate for conversion.
193 //
194 if (remainder != NULL && *remainder != '\0') {
195 return 0;
196 }
197
198 // successful conversion, return the pid
199 return pid;
200 }
201
202 // Check if the given statbuf is considered a secure directory for
203 // the backing store files. Returns true if the directory is considered
204 // a secure location. Returns false if the statbuf is a symbolic link or
205 // if an error occurred.
206 //
207 static bool is_statbuf_secure(struct stat *statp) {
208 if (S_ISLNK(statp->st_mode) || !S_ISDIR(statp->st_mode)) {
209 // The path represents a link or some non-directory file type,
210 // which is not what we expected. Declare it insecure.
211 //
212 return false;
213 }
214 // We have an existing directory, check if the permissions are safe.
215 //
216 if ((statp->st_mode & (S_IWGRP|S_IWOTH)) != 0) {
217 // The directory is open for writing and could be subjected
218 // to a symlink or a hard link attack. Declare it insecure.
219 //
220 return false;
221 }
222 // If user is not root then see if the uid of the directory matches the effective uid of the process.
223 uid_t euid = geteuid();
224 if ((euid != 0) && (statp->st_uid != euid)) {
225 // The directory was not created by this user, declare it insecure.
226 //
227 return false;
228 }
229 return true;
230 }
231
232
233 // Check if the given path is considered a secure directory for
234 // the backing store files. Returns true if the directory exists
235 // and is considered a secure location. Returns false if the path
236 // is a symbolic link or if an error occurred.
237 //
238 static bool is_directory_secure(const char* path) {
239 struct stat statbuf;
240 int result = 0;
241
242 RESTARTABLE(::lstat(path, &statbuf), result);
243 if (result == OS_ERR) {
244 return false;
245 }
246
247 // The path exists, see if it is secure.
248 return is_statbuf_secure(&statbuf);
249 }
250
251 // (Taken over from Solaris to support the O_NOFOLLOW case on AIX.)
252 // Check if the given directory file descriptor is considered a secure
253 // directory for the backing store files. Returns true if the directory
254 // exists and is considered a secure location. Returns false if the path
255 // is a symbolic link or if an error occurred.
256 static bool is_dirfd_secure(int dir_fd) {
257 struct stat statbuf;
258 int result = 0;
259
260 RESTARTABLE(::fstat(dir_fd, &statbuf), result);
261 if (result == OS_ERR) {
262 return false;
263 }
264
265 // The path exists, now check its mode.
266 return is_statbuf_secure(&statbuf);
267 }
268
269
270 // Check to make sure fd1 and fd2 are referencing the same file system object.
271 static bool is_same_fsobject(int fd1, int fd2) {
272 struct stat statbuf1;
273 struct stat statbuf2;
274 int result = 0;
275
276 RESTARTABLE(::fstat(fd1, &statbuf1), result);
277 if (result == OS_ERR) {
278 return false;
279 }
280 RESTARTABLE(::fstat(fd2, &statbuf2), result);
281 if (result == OS_ERR) {
282 return false;
283 }
284
285 if ((statbuf1.st_ino == statbuf2.st_ino) &&
286 (statbuf1.st_dev == statbuf2.st_dev)) {
287 return true;
288 } else {
289 return false;
290 }
291 }
292
293 // Helper functions for open without O_NOFOLLOW which is not present on AIX 5.3/6.1.
294 // We use the jdk6 implementation here.
295 #ifndef O_NOFOLLOW
296 // The O_NOFOLLOW oflag doesn't exist before solaris 5.10, this is to simulate that behaviour
297 // was done in jdk 5/6 hotspot by Oracle this way
298 static int open_o_nofollow_impl(const char* path, int oflag, mode_t mode, bool use_mode) {
299 struct stat orig_st;
300 struct stat new_st;
301 bool create;
302 int error;
303 int fd;
304 int result;
305
306 create = false;
307
308 RESTARTABLE(::lstat(path, &orig_st), result);
309
310 if (result == OS_ERR) {
311 if (errno == ENOENT && (oflag & O_CREAT) != 0) {
312 // File doesn't exist, but_we want to create it, add O_EXCL flag
313 // to make sure no-one creates it (or a symlink) before us
314 // This works as we expect with symlinks, from posix man page:
315 // 'If O_EXCL and O_CREAT are set, and path names a symbolic
316 // link, open() shall fail and set errno to [EEXIST]'.
317 oflag |= O_EXCL;
318 create = true;
319 } else {
320 // File doesn't exist, and we are not creating it.
321 return OS_ERR;
322 }
323 } else {
324 // lstat success, check if existing file is a link.
325 if ((orig_st.st_mode & S_IFMT) == S_IFLNK) {
326 // File is a symlink.
327 errno = ELOOP;
328 return OS_ERR;
329 }
330 }
331
332 if (use_mode == true) {
333 RESTARTABLE(::open(path, oflag, mode), fd);
334 } else {
335 RESTARTABLE(::open(path, oflag), fd);
336 }
337
338 if (fd == OS_ERR) {
339 return fd;
340 }
341
342 // Can't do inode checks on before/after if we created the file.
343 if (create == false) {
344 RESTARTABLE(::fstat(fd, &new_st), result);
345 if (result == OS_ERR) {
346 // Keep errno from fstat, in case close also fails.
347 error = errno;
348 ::close(fd);
349 errno = error;
350 return OS_ERR;
351 }
352
353 if (orig_st.st_dev != new_st.st_dev || orig_st.st_ino != new_st.st_ino) {
354 // File was tampered with during race window.
355 ::close(fd);
356 errno = EEXIST;
357 if (PrintMiscellaneous && Verbose) {
358 warning("possible file tampering attempt detected when opening %s", path);
359 }
360 return OS_ERR;
361 }
362 }
363
364 return fd;
365 }
366
367 static int open_o_nofollow(const char* path, int oflag, mode_t mode) {
368 return open_o_nofollow_impl(path, oflag, mode, true);
369 }
370
371 static int open_o_nofollow(const char* path, int oflag) {
372 return open_o_nofollow_impl(path, oflag, 0, false);
373 }
374 #endif
375
376 // Open the directory of the given path and validate it.
377 // Return a DIR * of the open directory.
378 static DIR *open_directory_secure(const char* dirname) {
379 // Open the directory using open() so that it can be verified
380 // to be secure by calling is_dirfd_secure(), opendir() and then check
381 // to see if they are the same file system object. This method does not
382 // introduce a window of opportunity for the directory to be attacked that
383 // calling opendir() and is_directory_secure() does.
384 int result;
385 DIR *dirp = NULL;
386
387 // No O_NOFOLLOW defined at buildtime, and it is not documented for open;
388 // so provide a workaround in this case.
389 #ifdef O_NOFOLLOW
390 RESTARTABLE(::open(dirname, O_RDONLY|O_NOFOLLOW), result);
391 #else
392 // workaround (jdk6 coding)
393 result = open_o_nofollow(dirname, O_RDONLY);
394 #endif
395
396 if (result == OS_ERR) {
397 // Directory doesn't exist or is a symlink, so there is nothing to cleanup.
398 if (PrintMiscellaneous && Verbose) {
399 if (errno == ELOOP) {
400 warning("directory %s is a symlink and is not secure\n", dirname);
401 } else {
402 warning("could not open directory %s: %s\n", dirname, os::strerror(errno));
403 }
404 }
405 return dirp;
406 }
407 int fd = result;
408
409 // Determine if the open directory is secure.
410 if (!is_dirfd_secure(fd)) {
411 // The directory is not a secure directory.
412 os::close(fd);
413 return dirp;
414 }
415
416 // Open the directory.
417 dirp = ::opendir(dirname);
418 if (dirp == NULL) {
419 // The directory doesn't exist, close fd and return.
420 os::close(fd);
421 return dirp;
422 }
423
424 // Check to make sure fd and dirp are referencing the same file system object.
425 if (!is_same_fsobject(fd, dirp->dd_fd)) {
426 // The directory is not secure.
427 os::close(fd);
428 os::closedir(dirp);
429 dirp = NULL;
430 return dirp;
431 }
432
433 // Close initial open now that we know directory is secure
434 os::close(fd);
435
436 return dirp;
437 }
438
439 // NOTE: The code below uses fchdir(), open() and unlink() because
440 // fdopendir(), openat() and unlinkat() are not supported on all
441 // versions. Once the support for fdopendir(), openat() and unlinkat()
442 // is available on all supported versions the code can be changed
443 // to use these functions.
444
445 // Open the directory of the given path, validate it and set the
446 // current working directory to it.
447 // Return a DIR * of the open directory and the saved cwd fd.
448 //
449 static DIR *open_directory_secure_cwd(const char* dirname, int *saved_cwd_fd) {
450
451 // Open the directory.
452 DIR* dirp = open_directory_secure(dirname);
453 if (dirp == NULL) {
454 // Directory doesn't exist or is insecure, so there is nothing to cleanup.
455 return dirp;
456 }
457 int fd = dirp->dd_fd;
458
459 // Open a fd to the cwd and save it off.
460 int result;
461 RESTARTABLE(::open(".", O_RDONLY), result);
462 if (result == OS_ERR) {
463 *saved_cwd_fd = -1;
464 } else {
465 *saved_cwd_fd = result;
466 }
467
468 // Set the current directory to dirname by using the fd of the directory and
469 // handle errors, otherwise shared memory files will be created in cwd.
470 result = fchdir(fd);
471 if (result == OS_ERR) {
472 if (PrintMiscellaneous && Verbose) {
473 warning("could not change to directory %s", dirname);
474 }
475 if (*saved_cwd_fd != -1) {
476 ::close(*saved_cwd_fd);
477 *saved_cwd_fd = -1;
478 }
479 // Close the directory.
480 os::closedir(dirp);
481 return NULL;
482 } else {
483 return dirp;
484 }
485 }
486
487 // Close the directory and restore the current working directory.
488 //
489 static void close_directory_secure_cwd(DIR* dirp, int saved_cwd_fd) {
490
491 int result;
492 // If we have a saved cwd change back to it and close the fd.
493 if (saved_cwd_fd != -1) {
494 result = fchdir(saved_cwd_fd);
495 ::close(saved_cwd_fd);
496 }
497
498 // Close the directory.
499 os::closedir(dirp);
500 }
501
502 // Check if the given file descriptor is considered a secure.
503 static bool is_file_secure(int fd, const char *filename) {
504
505 int result;
506 struct stat statbuf;
507
508 // Determine if the file is secure.
509 RESTARTABLE(::fstat(fd, &statbuf), result);
510 if (result == OS_ERR) {
511 if (PrintMiscellaneous && Verbose) {
512 warning("fstat failed on %s: %s\n", filename, os::strerror(errno));
513 }
514 return false;
515 }
516 if (statbuf.st_nlink > 1) {
517 // A file with multiple links is not expected.
518 if (PrintMiscellaneous && Verbose) {
519 warning("file %s has multiple links\n", filename);
520 }
521 return false;
522 }
523 return true;
524 }
525
526 // Return the user name for the given user id.
527 //
528 // The caller is expected to free the allocated memory.
529 static char* get_user_name(uid_t uid) {
530
531 struct passwd pwent;
532
533 // Determine the max pwbuf size from sysconf, and hardcode
534 // a default if this not available through sysconf.
535 long bufsize = sysconf(_SC_GETPW_R_SIZE_MAX);
536 if (bufsize == -1)
537 bufsize = 1024;
538
539 char* pwbuf = NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
540
541 struct passwd* p;
542 int result = getpwuid_r(uid, &pwent, pwbuf, (size_t)bufsize, &p);
543
544 if (result != 0 || p == NULL || p->pw_name == NULL || *(p->pw_name) == '\0') {
545 if (PrintMiscellaneous && Verbose) {
546 if (result != 0) {
547 warning("Could not retrieve passwd entry: %s\n",
548 os::strerror(result));
549 }
550 else if (p == NULL) {
551 // this check is added to protect against an observed problem
552 // with getpwuid_r() on RedHat 9 where getpwuid_r returns 0,
553 // indicating success, but has p == NULL. This was observed when
554 // inserting a file descriptor exhaustion fault prior to the call
555 // getpwuid_r() call. In this case, error is set to the appropriate
556 // error condition, but this is undocumented behavior. This check
557 // is safe under any condition, but the use of errno in the output
558 // message may result in an erroneous message.
559 // Bug Id 89052 was opened with RedHat.
560 //
561 warning("Could not retrieve passwd entry: %s\n",
562 os::strerror(errno));
563 }
564 else {
565 warning("Could not determine user name: %s\n",
566 p->pw_name == NULL ? "pw_name = NULL" :
567 "pw_name zero length");
568 }
569 }
570 FREE_C_HEAP_ARRAY(char, pwbuf);
571 return NULL;
572 }
573
574 char* user_name = NEW_C_HEAP_ARRAY(char, strlen(p->pw_name) + 1, mtInternal);
575 strcpy(user_name, p->pw_name);
576
577 FREE_C_HEAP_ARRAY(char, pwbuf);
578 return user_name;
579 }
580
581 // return the name of the user that owns the process identified by vmid.
582 //
583 // This method uses a slow directory search algorithm to find the backing
584 // store file for the specified vmid and returns the user name, as determined
585 // by the user name suffix of the hsperfdata_<username> directory name.
586 //
587 // the caller is expected to free the allocated memory.
588 //
589 static char* get_user_name_slow(int vmid, TRAPS) {
590
591 // short circuit the directory search if the process doesn't even exist.
592 if (kill(vmid, 0) == OS_ERR) {
593 if (errno == ESRCH) {
594 THROW_MSG_0(vmSymbols::java_lang_IllegalArgumentException(),
595 "Process not found");
596 }
597 else /* EPERM */ {
598 THROW_MSG_0(vmSymbols::java_io_IOException(), os::strerror(errno));
599 }
600 }
601
602 // directory search
603 char* oldest_user = NULL;
604 time_t oldest_ctime = 0;
605
606 const char* tmpdirname = os::get_temp_directory();
607
608 DIR* tmpdirp = os::opendir(tmpdirname);
609
610 if (tmpdirp == NULL) {
611 return NULL;
612 }
613
614 // for each entry in the directory that matches the pattern hsperfdata_*,
615 // open the directory and check if the file for the given vmid exists.
616 // The file with the expected name and the latest creation date is used
617 // to determine the user name for the process id.
618 //
619 struct dirent* dentry;
620 char* tdbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(tmpdirname), mtInternal);
621 errno = 0;
622 while ((dentry = os::readdir(tmpdirp, (struct dirent *)tdbuf)) != NULL) {
623
624 // check if the directory entry is a hsperfdata file
625 if (strncmp(dentry->d_name, PERFDATA_NAME, strlen(PERFDATA_NAME)) != 0) {
626 continue;
627 }
628
629 char* usrdir_name = NEW_C_HEAP_ARRAY(char,
630 strlen(tmpdirname) + strlen(dentry->d_name) + 2, mtInternal);
631 strcpy(usrdir_name, tmpdirname);
632 strcat(usrdir_name, "/");
633 strcat(usrdir_name, dentry->d_name);
634
635 // Open the user directory.
636 DIR* subdirp = open_directory_secure(usrdir_name);
637
638 if (subdirp == NULL) {
639 FREE_C_HEAP_ARRAY(char, usrdir_name);
640 continue;
641 }
642
643 // Since we don't create the backing store files in directories
644 // pointed to by symbolic links, we also don't follow them when
645 // looking for the files. We check for a symbolic link after the
646 // call to opendir in order to eliminate a small window where the
647 // symlink can be exploited.
648 //
649 if (!is_directory_secure(usrdir_name)) {
650 FREE_C_HEAP_ARRAY(char, usrdir_name);
651 os::closedir(subdirp);
652 continue;
653 }
654
655 struct dirent* udentry;
656 char* udbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(usrdir_name), mtInternal);
657 errno = 0;
658 while ((udentry = os::readdir(subdirp, (struct dirent *)udbuf)) != NULL) {
659
660 if (filename_to_pid(udentry->d_name) == vmid) {
661 struct stat statbuf;
662 int result;
663
664 char* filename = NEW_C_HEAP_ARRAY(char,
665 strlen(usrdir_name) + strlen(udentry->d_name) + 2, mtInternal);
666
667 strcpy(filename, usrdir_name);
668 strcat(filename, "/");
669 strcat(filename, udentry->d_name);
670
671 // don't follow symbolic links for the file
672 RESTARTABLE(::lstat(filename, &statbuf), result);
673 if (result == OS_ERR) {
674 FREE_C_HEAP_ARRAY(char, filename);
675 continue;
676 }
677
678 // skip over files that are not regular files.
679 if (!S_ISREG(statbuf.st_mode)) {
680 FREE_C_HEAP_ARRAY(char, filename);
681 continue;
682 }
683
684 // compare and save filename with latest creation time
685 if (statbuf.st_size > 0 && statbuf.st_ctime > oldest_ctime) {
686
687 if (statbuf.st_ctime > oldest_ctime) {
688 char* user = strchr(dentry->d_name, '_') + 1;
689
690 if (oldest_user != NULL) FREE_C_HEAP_ARRAY(char, oldest_user);
691 oldest_user = NEW_C_HEAP_ARRAY(char, strlen(user)+1, mtInternal);
692
693 strcpy(oldest_user, user);
694 oldest_ctime = statbuf.st_ctime;
695 }
696 }
697
698 FREE_C_HEAP_ARRAY(char, filename);
699 }
700 }
701 os::closedir(subdirp);
702 FREE_C_HEAP_ARRAY(char, udbuf);
703 FREE_C_HEAP_ARRAY(char, usrdir_name);
704 }
705 os::closedir(tmpdirp);
706 FREE_C_HEAP_ARRAY(char, tdbuf);
707
708 return(oldest_user);
709 }
710
711 // return the name of the user that owns the JVM indicated by the given vmid.
712 //
713 static char* get_user_name(int vmid, TRAPS) {
714 return get_user_name_slow(vmid, THREAD);
715 }
716
717 // return the file name of the backing store file for the named
718 // shared memory region for the given user name and vmid.
719 //
720 // the caller is expected to free the allocated memory.
721 //
722 static char* get_sharedmem_filename(const char* dirname, int vmid) {
723
724 // add 2 for the file separator and a null terminator.
725 size_t nbytes = strlen(dirname) + UINT_CHARS + 2;
726
727 char* name = NEW_C_HEAP_ARRAY(char, nbytes, mtInternal);
728 snprintf(name, nbytes, "%s/%d", dirname, vmid);
729
730 return name;
731 }
732
733
734 // remove file
735 //
736 // this method removes the file specified by the given path
737 //
738 static void remove_file(const char* path) {
739
740 int result;
741
742 // if the file is a directory, the following unlink will fail. since
743 // we don't expect to find directories in the user temp directory, we
744 // won't try to handle this situation. even if accidentially or
745 // maliciously planted, the directory's presence won't hurt anything.
746 //
747 RESTARTABLE(::unlink(path), result);
748 if (PrintMiscellaneous && Verbose && result == OS_ERR) {
749 if (errno != ENOENT) {
750 warning("Could not unlink shared memory backing"
751 " store file %s : %s\n", path, os::strerror(errno));
752 }
753 }
754 }
755
756 // Cleanup stale shared memory resources
757 //
758 // This method attempts to remove all stale shared memory files in
759 // the named user temporary directory. It scans the named directory
760 // for files matching the pattern ^$[0-9]*$. For each file found, the
761 // process id is extracted from the file name and a test is run to
762 // determine if the process is alive. If the process is not alive,
763 // any stale file resources are removed.
764 static void cleanup_sharedmem_resources(const char* dirname) {
765
766 int saved_cwd_fd;
767 // Open the directory.
768 DIR* dirp = open_directory_secure_cwd(dirname, &saved_cwd_fd);
769 if (dirp == NULL) {
770 // Directory doesn't exist or is insecure, so there is nothing to cleanup.
771 return;
772 }
773
774 // For each entry in the directory that matches the expected file
775 // name pattern, determine if the file resources are stale and if
776 // so, remove the file resources. Note, instrumented HotSpot processes
777 // for this user may start and/or terminate during this search and
778 // remove or create new files in this directory. The behavior of this
779 // loop under these conditions is dependent upon the implementation of
780 // opendir/readdir.
781 struct dirent* entry;
782 char* dbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(dirname), mtInternal);
783
784 errno = 0;
785 while ((entry = os::readdir(dirp, (struct dirent *)dbuf)) != NULL) {
786
787 pid_t pid = filename_to_pid(entry->d_name);
788
789 if (pid == 0) {
790
791 if (strcmp(entry->d_name, ".") != 0 && strcmp(entry->d_name, "..") != 0) {
792
793 // Attempt to remove all unexpected files, except "." and "..".
794 unlink(entry->d_name);
795 }
796
797 errno = 0;
798 continue;
799 }
800
801 // We now have a file name that converts to a valid integer
802 // that could represent a process id . if this process id
803 // matches the current process id or the process is not running,
804 // then remove the stale file resources.
805 //
806 // Process liveness is detected by sending signal number 0 to
807 // the process id (see kill(2)). if kill determines that the
808 // process does not exist, then the file resources are removed.
809 // if kill determines that that we don't have permission to
810 // signal the process, then the file resources are assumed to
811 // be stale and are removed because the resources for such a
812 // process should be in a different user specific directory.
813 if ((pid == os::current_process_id()) ||
814 (kill(pid, 0) == OS_ERR && (errno == ESRCH || errno == EPERM))) {
815
816 unlink(entry->d_name);
817 }
818 errno = 0;
819 }
820
821 // Close the directory and reset the current working directory.
822 close_directory_secure_cwd(dirp, saved_cwd_fd);
823
824 FREE_C_HEAP_ARRAY(char, dbuf);
825 }
826
827 // Make the user specific temporary directory. Returns true if
828 // the directory exists and is secure upon return. Returns false
829 // if the directory exists but is either a symlink, is otherwise
830 // insecure, or if an error occurred.
831 static bool make_user_tmp_dir(const char* dirname) {
832
833 // Create the directory with 0755 permissions. note that the directory
834 // will be owned by euid::egid, which may not be the same as uid::gid.
835 if (mkdir(dirname, S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH) == OS_ERR) {
836 if (errno == EEXIST) {
837 // The directory already exists and was probably created by another
838 // JVM instance. However, this could also be the result of a
839 // deliberate symlink. Verify that the existing directory is safe.
840 if (!is_directory_secure(dirname)) {
841 // Directory is not secure.
842 if (PrintMiscellaneous && Verbose) {
843 warning("%s directory is insecure\n", dirname);
844 }
845 return false;
846 }
847 }
848 else {
849 // we encountered some other failure while attempting
850 // to create the directory
851 //
852 if (PrintMiscellaneous && Verbose) {
853 warning("could not create directory %s: %s\n",
854 dirname, os::strerror(errno));
855 }
856 return false;
857 }
858 }
859 return true;
860 }
861
862 // create the shared memory file resources
863 //
864 // This method creates the shared memory file with the given size
865 // This method also creates the user specific temporary directory, if
866 // it does not yet exist.
867 //
868 static int create_sharedmem_resources(const char* dirname, const char* filename, size_t size) {
869
870 // make the user temporary directory
871 if (!make_user_tmp_dir(dirname)) {
872 // could not make/find the directory or the found directory
873 // was not secure
874 return -1;
875 }
876
877 int saved_cwd_fd;
878 // Open the directory and set the current working directory to it.
879 DIR* dirp = open_directory_secure_cwd(dirname, &saved_cwd_fd);
880 if (dirp == NULL) {
881 // Directory doesn't exist or is insecure, so cannot create shared
882 // memory file.
883 return -1;
884 }
885
886 // Open the filename in the current directory.
887 // Cannot use O_TRUNC here; truncation of an existing file has to happen
888 // after the is_file_secure() check below.
889 int result;
890
891 // No O_NOFOLLOW defined at buildtime, and it is not documented for open;
892 // so provide a workaround in this case.
893 #ifdef O_NOFOLLOW
894 RESTARTABLE(::open(filename, O_RDWR|O_CREAT|O_NOFOLLOW, S_IREAD|S_IWRITE), result);
895 #else
896 // workaround function (jdk6 code)
897 result = open_o_nofollow(filename, O_RDWR|O_CREAT, S_IREAD|S_IWRITE);
898 #endif
899
900 if (result == OS_ERR) {
901 if (PrintMiscellaneous && Verbose) {
902 if (errno == ELOOP) {
903 warning("file %s is a symlink and is not secure\n", filename);
904 } else {
905 warning("could not create file %s: %s\n", filename, os::strerror(errno));
906 }
907 }
908 // Close the directory and reset the current working directory.
909 close_directory_secure_cwd(dirp, saved_cwd_fd);
910
911 return -1;
912 }
913 // Close the directory and reset the current working directory.
914 close_directory_secure_cwd(dirp, saved_cwd_fd);
915
916 // save the file descriptor
917 int fd = result;
918
919 // Check to see if the file is secure.
920 if (!is_file_secure(fd, filename)) {
921 ::close(fd);
922 return -1;
923 }
924
925 // Truncate the file to get rid of any existing data.
926 RESTARTABLE(::ftruncate(fd, (off_t)0), result);
927 if (result == OS_ERR) {
928 if (PrintMiscellaneous && Verbose) {
929 warning("could not truncate shared memory file: %s\n", os::strerror(errno));
930 }
931 ::close(fd);
932 return -1;
933 }
934 // set the file size
935 RESTARTABLE(::ftruncate(fd, (off_t)size), result);
936 if (result == OS_ERR) {
937 if (PrintMiscellaneous && Verbose) {
938 warning("could not set shared memory file size: %s\n", os::strerror(errno));
939 }
940 ::close(fd);
941 return -1;
942 }
943
944 return fd;
945 }
946
947 // open the shared memory file for the given user and vmid. returns
948 // the file descriptor for the open file or -1 if the file could not
949 // be opened.
950 //
951 static int open_sharedmem_file(const char* filename, int oflags, TRAPS) {
952
953 // open the file
954 int result;
955 // provide a workaround in case no O_NOFOLLOW is defined at buildtime
956 #ifdef O_NOFOLLOW
957 RESTARTABLE(::open(filename, oflags), result);
958 #else
959 result = open_o_nofollow(filename, oflags);
960 #endif
961 if (result == OS_ERR) {
962 if (errno == ENOENT) {
963 THROW_MSG_(vmSymbols::java_lang_IllegalArgumentException(),
964 "Process not found", OS_ERR);
965 }
966 else if (errno == EACCES) {
967 THROW_MSG_(vmSymbols::java_lang_IllegalArgumentException(),
968 "Permission denied", OS_ERR);
969 }
970 else {
971 THROW_MSG_(vmSymbols::java_io_IOException(),
972 os::strerror(errno), OS_ERR);
973 }
974 }
975 int fd = result;
976
977 // Check to see if the file is secure.
978 if (!is_file_secure(fd, filename)) {
979 ::close(fd);
980 return -1;
981 }
982
983 return fd;
984 }
985
986 // create a named shared memory region. returns the address of the
987 // memory region on success or NULL on failure. A return value of
988 // NULL will ultimately disable the shared memory feature.
989 //
990 // On AIX, the name space for shared memory objects
991 // is the file system name space.
992 //
993 // A monitoring application attaching to a JVM does not need to know
994 // the file system name of the shared memory object. However, it may
995 // be convenient for applications to discover the existence of newly
996 // created and terminating JVMs by watching the file system name space
997 // for files being created or removed.
998 //
999 static char* mmap_create_shared(size_t size) {
1000
1001 int result;
1002 int fd;
1003 char* mapAddress;
1004
1005 int vmid = os::current_process_id();
1006
1007 char* user_name = get_user_name(geteuid());
1008
1009 if (user_name == NULL)
1010 return NULL;
1011
1012 char* dirname = get_user_tmp_dir(user_name);
1013 char* filename = get_sharedmem_filename(dirname, vmid);
1014
1015 // get the short filename.
1016 char* short_filename = strrchr(filename, '/');
1017 if (short_filename == NULL) {
1018 short_filename = filename;
1019 } else {
1020 short_filename++;
1021 }
1022
1023 // cleanup any stale shared memory files
1024 cleanup_sharedmem_resources(dirname);
1025
1026 assert(((size > 0) && (size % os::vm_page_size() == 0)),
1027 "unexpected PerfMemory region size");
1028
1029 fd = create_sharedmem_resources(dirname, short_filename, size);
1030
1031 FREE_C_HEAP_ARRAY(char, user_name);
1032 FREE_C_HEAP_ARRAY(char, dirname);
1033
1034 if (fd == -1) {
1035 FREE_C_HEAP_ARRAY(char, filename);
1036 return NULL;
1037 }
1038
1039 mapAddress = (char*)::mmap((char*)0, size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
1040
1041 result = ::close(fd);
1042 assert(result != OS_ERR, "could not close file");
1043
1044 if (mapAddress == MAP_FAILED) {
1045 if (PrintMiscellaneous && Verbose) {
1046 warning("mmap failed - %s\n", os::strerror(errno));
1047 }
1048 remove_file(filename);
1049 FREE_C_HEAP_ARRAY(char, filename);
1050 return NULL;
1051 }
1052
1053 // save the file name for use in delete_shared_memory()
1054 backing_store_file_name = filename;
1055
1056 // clear the shared memory region
1057 (void)::memset((void*) mapAddress, 0, size);
1058
1059 // It does not go through os api, the operation has to record from here.
1060 MemTracker::record_virtual_memory_reserve((address)mapAddress, size, CURRENT_PC, mtInternal);
1061
1062 return mapAddress;
1063 }
1064
1065 // release a named shared memory region
1066 //
1067 static void unmap_shared(char* addr, size_t bytes) {
1068 // Do not rely on os::reserve_memory/os::release_memory to use mmap.
1069 // Use os::reserve_memory/os::release_memory for PerfDisableSharedMem=1, mmap/munmap for PerfDisableSharedMem=0
1070 if (::munmap(addr, bytes) == -1) {
1071 warning("perfmemory: munmap failed (%d)\n", errno);
1072 }
1073 }
1074
1075 // create the PerfData memory region in shared memory.
1076 //
1077 static char* create_shared_memory(size_t size) {
1078
1079 // create the shared memory region.
1080 return mmap_create_shared(size);
1081 }
1082
1083 // delete the shared PerfData memory region
1084 //
1085 static void delete_shared_memory(char* addr, size_t size) {
1086
1087 // cleanup the persistent shared memory resources. since DestroyJavaVM does
1088 // not support unloading of the JVM, unmapping of the memory resource is
1089 // not performed. The memory will be reclaimed by the OS upon termination of
1090 // the process. The backing store file is deleted from the file system.
1091
1092 assert(!PerfDisableSharedMem, "shouldn't be here");
1093
1094 if (backing_store_file_name != NULL) {
1095 remove_file(backing_store_file_name);
1096 // Don't.. Free heap memory could deadlock os::abort() if it is called
1097 // from signal handler. OS will reclaim the heap memory.
1098 // FREE_C_HEAP_ARRAY(char, backing_store_file_name);
1099 backing_store_file_name = NULL;
1100 }
1101 }
1102
1103 // return the size of the file for the given file descriptor
1104 // or 0 if it is not a valid size for a shared memory file
1105 //
1106 static size_t sharedmem_filesize(int fd, TRAPS) {
1107
1108 struct stat statbuf;
1109 int result;
1110
1111 RESTARTABLE(::fstat(fd, &statbuf), result);
1112 if (result == OS_ERR) {
1113 if (PrintMiscellaneous && Verbose) {
1114 warning("fstat failed: %s\n", os::strerror(errno));
1115 }
1116 THROW_MSG_0(vmSymbols::java_io_IOException(),
1117 "Could not determine PerfMemory size");
1118 }
1119
1120 if ((statbuf.st_size == 0) ||
1121 ((size_t)statbuf.st_size % os::vm_page_size() != 0)) {
1122 THROW_MSG_0(vmSymbols::java_lang_Exception(),
1123 "Invalid PerfMemory size");
1124 }
1125
1126 return (size_t)statbuf.st_size;
1127 }
1128
1129 // attach to a named shared memory region.
1130 //
1131 static void mmap_attach_shared(const char* user, int vmid, PerfMemory::PerfMemoryMode mode, char** addr, size_t* sizep, TRAPS) {
1132
1133 char* mapAddress;
1134 int result;
1135 int fd;
1136 size_t size = 0;
1137 const char* luser = NULL;
1138
1139 int mmap_prot;
1140 int file_flags;
1141
1142 ResourceMark rm;
1143
1144 // map the high level access mode to the appropriate permission
1145 // constructs for the file and the shared memory mapping.
1146 if (mode == PerfMemory::PERF_MODE_RO) {
1147 mmap_prot = PROT_READ;
1148 // No O_NOFOLLOW defined at buildtime, and it is not documented for open.
1149 #ifdef O_NOFOLLOW
1150 file_flags = O_RDONLY | O_NOFOLLOW;
1151 #else
1152 file_flags = O_RDONLY;
1153 #endif
1154 }
1155 else if (mode == PerfMemory::PERF_MODE_RW) {
1156 #ifdef LATER
1157 mmap_prot = PROT_READ | PROT_WRITE;
1158 file_flags = O_RDWR | O_NOFOLLOW;
1159 #else
1160 THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1161 "Unsupported access mode");
1162 #endif
1163 }
1164 else {
1165 THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1166 "Illegal access mode");
1167 }
1168
1169 if (user == NULL || strlen(user) == 0) {
1170 luser = get_user_name(vmid, CHECK);
1171 }
1172 else {
1173 luser = user;
1174 }
1175
1176 if (luser == NULL) {
1177 THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1178 "Could not map vmid to user Name");
1179 }
1180
1181 char* dirname = get_user_tmp_dir(luser);
1182
1183 // since we don't follow symbolic links when creating the backing
1184 // store file, we don't follow them when attaching either.
1185 //
1186 if (!is_directory_secure(dirname)) {
1187 FREE_C_HEAP_ARRAY(char, dirname);
1188 if (luser != user) {
1189 FREE_C_HEAP_ARRAY(char, luser);
1190 }
1191 THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1192 "Process not found");
1193 }
1194
1195 char* filename = get_sharedmem_filename(dirname, vmid);
1196
1197 // copy heap memory to resource memory. the open_sharedmem_file
1198 // method below need to use the filename, but could throw an
1199 // exception. using a resource array prevents the leak that
1200 // would otherwise occur.
1201 char* rfilename = NEW_RESOURCE_ARRAY(char, strlen(filename) + 1);
1202 strcpy(rfilename, filename);
1203
1204 // free the c heap resources that are no longer needed
1205 if (luser != user) FREE_C_HEAP_ARRAY(char, luser);
1206 FREE_C_HEAP_ARRAY(char, dirname);
1207 FREE_C_HEAP_ARRAY(char, filename);
1208
1209 // open the shared memory file for the give vmid
1210 fd = open_sharedmem_file(rfilename, file_flags, THREAD);
1211
1212 if (fd == OS_ERR) {
1213 return;
1214 }
1215
1216 if (HAS_PENDING_EXCEPTION) {
1217 ::close(fd);
1218 return;
1219 }
1220
1221 if (*sizep == 0) {
1222 size = sharedmem_filesize(fd, CHECK);
1223 } else {
1224 size = *sizep;
1225 }
1226
1227 assert(size > 0, "unexpected size <= 0");
1228
1229 mapAddress = (char*)::mmap((char*)0, size, mmap_prot, MAP_SHARED, fd, 0);
1230
1231 result = ::close(fd);
1232 assert(result != OS_ERR, "could not close file");
1233
1234 if (mapAddress == MAP_FAILED) {
1235 if (PrintMiscellaneous && Verbose) {
1236 warning("mmap failed: %s\n", os::strerror(errno));
1237 }
1238 THROW_MSG(vmSymbols::java_lang_OutOfMemoryError(),
1239 "Could not map PerfMemory");
1240 }
1241
1242 // it does not go through os api, the operation has to record from here.
1243 MemTracker::record_virtual_memory_reserve((address)mapAddress, size, CURRENT_PC, mtInternal);
1244
1245 *addr = mapAddress;
1246 *sizep = size;
1247
1248 log_debug(perf, memops)("mapped " SIZE_FORMAT " bytes for vmid %d at "
1249 INTPTR_FORMAT "\n", size, vmid, p2i((void*)mapAddress));
1250 }
1251
1252 // create the PerfData memory region
1253 //
1254 // This method creates the memory region used to store performance
1255 // data for the JVM. The memory may be created in standard or
1256 // shared memory.
1257 //
1258 void PerfMemory::create_memory_region(size_t size) {
1259
1260 if (PerfDisableSharedMem) {
1261 // do not share the memory for the performance data.
1262 _start = create_standard_memory(size);
1263 }
1264 else {
1265 _start = create_shared_memory(size);
1266 if (_start == NULL) {
1267
1268 // creation of the shared memory region failed, attempt
1269 // to create a contiguous, non-shared memory region instead.
1270 //
1271 if (PrintMiscellaneous && Verbose) {
1272 warning("Reverting to non-shared PerfMemory region.\n");
1273 }
1274 PerfDisableSharedMem = true;
1275 _start = create_standard_memory(size);
1276 }
1277 }
1278
1279 if (_start != NULL) _capacity = size;
1280
1281 }
1282
1283 // delete the PerfData memory region
1284 //
1285 // This method deletes the memory region used to store performance
1286 // data for the JVM. The memory region indicated by the <address, size>
1287 // tuple will be inaccessible after a call to this method.
1288 //
1289 void PerfMemory::delete_memory_region() {
1290
1291 assert((start() != NULL && capacity() > 0), "verify proper state");
1292
1293 // If user specifies PerfDataSaveFile, it will save the performance data
1294 // to the specified file name no matter whether PerfDataSaveToFile is specified
1295 // or not. In other word, -XX:PerfDataSaveFile=.. overrides flag
1296 // -XX:+PerfDataSaveToFile.
1297 if (PerfDataSaveToFile || PerfDataSaveFile != NULL) {
1298 save_memory_to_file(start(), capacity());
1299 }
1300
1301 if (PerfDisableSharedMem) {
1302 delete_standard_memory(start(), capacity());
1303 }
1304 else {
1305 delete_shared_memory(start(), capacity());
1306 }
1307 }
1308
1309 // attach to the PerfData memory region for another JVM
1310 //
1311 // This method returns an <address, size> tuple that points to
1312 // a memory buffer that is kept reasonably synchronized with
1313 // the PerfData memory region for the indicated JVM. This
1314 // buffer may be kept in synchronization via shared memory
1315 // or some other mechanism that keeps the buffer updated.
1316 //
1317 // If the JVM chooses not to support the attachability feature,
1318 // this method should throw an UnsupportedOperation exception.
1319 //
1320 // This implementation utilizes named shared memory to map
1321 // the indicated process's PerfData memory region into this JVMs
1322 // address space.
1323 //
1324 void PerfMemory::attach(const char* user, int vmid, PerfMemoryMode mode, char** addrp, size_t* sizep, TRAPS) {
1325
1326 if (vmid == 0 || vmid == os::current_process_id()) {
1327 *addrp = start();
1328 *sizep = capacity();
1329 return;
1330 }
1331
1332 mmap_attach_shared(user, vmid, mode, addrp, sizep, CHECK);
1333 }
1334
1335 // detach from the PerfData memory region of another JVM
1336 //
1337 // This method detaches the PerfData memory region of another
1338 // JVM, specified as an <address, size> tuple of a buffer
1339 // in this process's address space. This method may perform
1340 // arbitrary actions to accomplish the detachment. The memory
1341 // region specified by <address, size> will be inaccessible after
1342 // a call to this method.
1343 //
1344 // If the JVM chooses not to support the attachability feature,
1345 // this method should throw an UnsupportedOperation exception.
1346 //
1347 // This implementation utilizes named shared memory to detach
1348 // the indicated process's PerfData memory region from this
1349 // process's address space.
1350 //
1351 void PerfMemory::detach(char* addr, size_t bytes, TRAPS) {
1352
1353 assert(addr != 0, "address sanity check");
1354 assert(bytes > 0, "capacity sanity check");
1355
1356 if (PerfMemory::contains(addr) || PerfMemory::contains(addr + bytes - 1)) {
1357 // prevent accidental detachment of this process's PerfMemory region
1358 return;
1359 }
1360
1361 unmap_shared(addr, bytes);
1362 }