1 /*
2 * Copyright (c) 2001, 2016, 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 "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/os.hpp"
34 #include "runtime/perfMemory.hpp"
35 #include "services/memTracker.hpp"
36 #include "utilities/exceptions.hpp"
37
38 // put OS-includes here
39 # include <sys/types.h>
40 # include <sys/mman.h>
41 # include <errno.h>
42 # include <stdio.h>
43 # include <unistd.h>
44 # include <sys/stat.h>
45 # include <signal.h>
46 # include <pwd.h>
47
48 static char* backing_store_file_name = NULL; // name of the backing store
49 // file, if successfully created.
50
51 // Standard Memory Implementation Details
52
53 // create the PerfData memory region in standard memory.
54 //
55 static char* create_standard_memory(size_t size) {
56
57 // allocate an aligned chuck of memory
58 char* mapAddress = os::reserve_memory(size);
59
60 if (mapAddress == NULL) {
61 return NULL;
62 }
63
64 // commit memory
65 if (!os::commit_memory(mapAddress, size, !ExecMem)) {
66 if (PrintMiscellaneous && Verbose) {
67 warning("Could not commit PerfData memory\n");
68 }
69 os::release_memory(mapAddress, size);
70 return NULL;
71 }
72
73 return mapAddress;
74 }
75
76 // delete the PerfData memory region
77 //
78 static void delete_standard_memory(char* addr, size_t size) {
79
80 // there are no persistent external resources to cleanup for standard
81 // memory. since DestroyJavaVM does not support unloading of the JVM,
82 // cleanup of the memory resource is not performed. The memory will be
83 // reclaimed by the OS upon termination of the process.
84 //
85 return;
86 }
87
88 // save the specified memory region to the given file
89 //
90 // Note: this function might be called from signal handler (by os::abort()),
91 // don't allocate heap memory.
92 //
93 static void save_memory_to_file(char* addr, size_t size) {
94
95 const char* destfile = PerfMemory::get_perfdata_file_path();
96 assert(destfile[0] != '\0', "invalid PerfData file path");
97
98 int result;
99
100 RESTARTABLE(::open(destfile, O_CREAT|O_WRONLY|O_TRUNC, S_IREAD|S_IWRITE),
101 result);;
102 if (result == OS_ERR) {
103 if (PrintMiscellaneous && Verbose) {
104 warning("Could not create Perfdata save file: %s: %s\n",
105 destfile, os::strerror(errno));
106 }
107 } else {
108 int fd = result;
109
110 for (size_t remaining = size; remaining > 0;) {
111
112 RESTARTABLE(::write(fd, addr, remaining), result);
113 if (result == OS_ERR) {
114 if (PrintMiscellaneous && Verbose) {
115 warning("Could not write Perfdata save file: %s: %s\n",
116 destfile, os::strerror(errno));
117 }
118 break;
119 }
120
121 remaining -= (size_t)result;
122 addr += result;
123 }
124
125 result = ::close(fd);
126 if (PrintMiscellaneous && Verbose) {
127 if (result == OS_ERR) {
128 warning("Could not close %s: %s\n", destfile, os::strerror(errno));
129 }
130 }
131 }
132 FREE_C_HEAP_ARRAY(char, destfile);
133 }
134
135
136 // Shared Memory Implementation Details
137
138 // Note: the solaris and linux shared memory implementation uses the mmap
139 // interface with a backing store file to implement named shared memory.
140 // Using the file system as the name space for shared memory allows a
141 // common name space to be supported across a variety of platforms. It
142 // also provides a name space that Java applications can deal with through
143 // simple file apis.
144 //
145 // The solaris and linux implementations store the backing store file in
146 // a user specific temporary directory located in the /tmp file system,
147 // which is always a local file system and is sometimes a RAM based file
148 // system.
149
150 // return the user specific temporary directory name.
151 //
152 // the caller is expected to free the allocated memory.
153 //
154 static char* get_user_tmp_dir(const char* user) {
155
156 const char* tmpdir = os::get_temp_directory();
157 const char* perfdir = PERFDATA_NAME;
158 size_t nbytes = strlen(tmpdir) + strlen(perfdir) + strlen(user) + 3;
159 char* dirname = NEW_C_HEAP_ARRAY(char, nbytes, mtInternal);
160
161 // construct the path name to user specific tmp directory
162 snprintf(dirname, nbytes, "%s/%s_%s", tmpdir, perfdir, user);
163
164 return dirname;
165 }
166
167 // convert the given file name into a process id. if the file
168 // does not meet the file naming constraints, return 0.
169 //
170 static pid_t filename_to_pid(const char* filename) {
171
172 // a filename that doesn't begin with a digit is not a
173 // candidate for conversion.
174 //
175 if (!isdigit(*filename)) {
176 return 0;
177 }
178
179 // check if file name can be converted to an integer without
180 // any leftover characters.
181 //
182 char* remainder = NULL;
183 errno = 0;
184 pid_t pid = (pid_t)strtol(filename, &remainder, 10);
185
186 if (errno != 0) {
187 return 0;
188 }
189
190 // check for left over characters. If any, then the filename is
191 // not a candidate for conversion.
192 //
193 if (remainder != NULL && *remainder != '\0') {
194 return 0;
195 }
196
197 // successful conversion, return the pid
198 return pid;
199 }
200
201 // Check if the given statbuf is considered a secure directory for
202 // the backing store files. Returns true if the directory is considered
203 // a secure location. Returns false if the statbuf is a symbolic link or
204 // if an error occurred.
205 //
206 static bool is_statbuf_secure(struct stat *statp) {
207 if (S_ISLNK(statp->st_mode) || !S_ISDIR(statp->st_mode)) {
208 // The path represents a link or some non-directory file type,
209 // which is not what we expected. Declare it insecure.
210 //
211 return false;
212 }
213 // We have an existing directory, check if the permissions are safe.
214 //
215 if ((statp->st_mode & (S_IWGRP|S_IWOTH)) != 0) {
216 // The directory is open for writing and could be subjected
217 // to a symlink or a hard link attack. Declare it insecure.
218 //
219 return false;
220 }
221 // If user is not root then see if the uid of the directory matches the effective uid of the process.
222 uid_t euid = geteuid();
223 if ((euid != 0) && (statp->st_uid != euid)) {
224 // The directory was not created by this user, declare it insecure.
225 //
226 return false;
227 }
228 return true;
229 }
230
231
232 // Check if the given path is considered a secure directory for
233 // the backing store files. Returns true if the directory exists
234 // and is considered a secure location. Returns false if the path
235 // is a symbolic link or if an error occurred.
236 //
237 static bool is_directory_secure(const char* path) {
238 struct stat statbuf;
239 int result = 0;
240
241 RESTARTABLE(::lstat(path, &statbuf), result);
242 if (result == OS_ERR) {
243 return false;
244 }
245
246 // The path exists, see if it is secure.
247 return is_statbuf_secure(&statbuf);
248 }
249
250 // (Taken over from Solaris to support the O_NOFOLLOW case on AIX.)
251 // Check if the given directory file descriptor is considered a secure
252 // directory for the backing store files. Returns true if the directory
253 // exists and is considered a secure location. Returns false if the path
254 // is a symbolic link or if an error occurred.
255 static bool is_dirfd_secure(int dir_fd) {
256 struct stat statbuf;
257 int result = 0;
258
259 RESTARTABLE(::fstat(dir_fd, &statbuf), result);
260 if (result == OS_ERR) {
261 return false;
262 }
263
264 // The path exists, now check its mode.
265 return is_statbuf_secure(&statbuf);
266 }
267
268
269 // Check to make sure fd1 and fd2 are referencing the same file system object.
270 static bool is_same_fsobject(int fd1, int fd2) {
271 struct stat statbuf1;
272 struct stat statbuf2;
273 int result = 0;
274
275 RESTARTABLE(::fstat(fd1, &statbuf1), result);
276 if (result == OS_ERR) {
277 return false;
278 }
279 RESTARTABLE(::fstat(fd2, &statbuf2), result);
280 if (result == OS_ERR) {
281 return false;
282 }
283
284 if ((statbuf1.st_ino == statbuf2.st_ino) &&
285 (statbuf1.st_dev == statbuf2.st_dev)) {
286 return true;
287 } else {
288 return false;
289 }
290 }
291
292 // Helper functions for open without O_NOFOLLOW which is not present on AIX 5.3/6.1.
293 // We use the jdk6 implementation here.
294 #ifndef O_NOFOLLOW
295 // The O_NOFOLLOW oflag doesn't exist before solaris 5.10, this is to simulate that behaviour
296 // was done in jdk 5/6 hotspot by Oracle this way
297 static int open_o_nofollow_impl(const char* path, int oflag, mode_t mode, bool use_mode) {
298 struct stat orig_st;
299 struct stat new_st;
300 bool create;
301 int error;
302 int fd;
303 int result;
304
305 create = false;
306
307 RESTARTABLE(::lstat(path, &orig_st), result);
308
309 if (result == OS_ERR) {
310 if (errno == ENOENT && (oflag & O_CREAT) != 0) {
311 // File doesn't exist, but_we want to create it, add O_EXCL flag
312 // to make sure no-one creates it (or a symlink) before us
313 // This works as we expect with symlinks, from posix man page:
314 // 'If O_EXCL and O_CREAT are set, and path names a symbolic
315 // link, open() shall fail and set errno to [EEXIST]'.
316 oflag |= O_EXCL;
317 create = true;
318 } else {
319 // File doesn't exist, and we are not creating it.
320 return OS_ERR;
321 }
322 } else {
323 // lstat success, check if existing file is a link.
324 if ((orig_st.st_mode & S_IFMT) == S_IFLNK) {
325 // File is a symlink.
326 errno = ELOOP;
327 return OS_ERR;
328 }
329 }
330
331 if (use_mode == true) {
332 RESTARTABLE(::open(path, oflag, mode), fd);
333 } else {
334 RESTARTABLE(::open(path, oflag), fd);
335 }
336
337 if (fd == OS_ERR) {
338 return fd;
339 }
340
341 // Can't do inode checks on before/after if we created the file.
342 if (create == false) {
343 RESTARTABLE(::fstat(fd, &new_st), result);
344 if (result == OS_ERR) {
345 // Keep errno from fstat, in case close also fails.
346 error = errno;
347 ::close(fd);
348 errno = error;
349 return OS_ERR;
350 }
351
352 if (orig_st.st_dev != new_st.st_dev || orig_st.st_ino != new_st.st_ino) {
353 // File was tampered with during race window.
354 ::close(fd);
355 errno = EEXIST;
356 if (PrintMiscellaneous && Verbose) {
357 warning("possible file tampering attempt detected when opening %s", path);
358 }
359 return OS_ERR;
360 }
361 }
362
363 return fd;
364 }
365
366 static int open_o_nofollow(const char* path, int oflag, mode_t mode) {
367 return open_o_nofollow_impl(path, oflag, mode, true);
368 }
369
370 static int open_o_nofollow(const char* path, int oflag) {
371 return open_o_nofollow_impl(path, oflag, 0, false);
372 }
373 #endif
374
375 // Open the directory of the given path and validate it.
376 // Return a DIR * of the open directory.
377 static DIR *open_directory_secure(const char* dirname) {
378 // Open the directory using open() so that it can be verified
379 // to be secure by calling is_dirfd_secure(), opendir() and then check
380 // to see if they are the same file system object. This method does not
381 // introduce a window of opportunity for the directory to be attacked that
382 // calling opendir() and is_directory_secure() does.
383 int result;
384 DIR *dirp = NULL;
385
386 // No O_NOFOLLOW defined at buildtime, and it is not documented for open;
387 // so provide a workaround in this case.
388 #ifdef O_NOFOLLOW
389 RESTARTABLE(::open(dirname, O_RDONLY|O_NOFOLLOW), result);
390 #else
391 // workaround (jdk6 coding)
392 result = open_o_nofollow(dirname, O_RDONLY);
393 #endif
394
395 if (result == OS_ERR) {
396 // Directory doesn't exist or is a symlink, so there is nothing to cleanup.
397 if (PrintMiscellaneous && Verbose) {
398 if (errno == ELOOP) {
399 warning("directory %s is a symlink and is not secure\n", dirname);
400 } else {
401 warning("could not open directory %s: %s\n", dirname, os::strerror(errno));
402 }
403 }
404 return dirp;
405 }
406 int fd = result;
407
408 // Determine if the open directory is secure.
409 if (!is_dirfd_secure(fd)) {
410 // The directory is not a secure directory.
411 os::close(fd);
412 return dirp;
413 }
414
415 // Open the directory.
416 dirp = ::opendir(dirname);
417 if (dirp == NULL) {
418 // The directory doesn't exist, close fd and return.
419 os::close(fd);
420 return dirp;
421 }
422
423 // Check to make sure fd and dirp are referencing the same file system object.
424 if (!is_same_fsobject(fd, dirp->dd_fd)) {
425 // The directory is not secure.
426 os::close(fd);
427 os::closedir(dirp);
428 dirp = NULL;
429 return dirp;
430 }
431
432 // Close initial open now that we know directory is secure
433 os::close(fd);
434
435 return dirp;
436 }
437
438 // NOTE: The code below uses fchdir(), open() and unlink() because
439 // fdopendir(), openat() and unlinkat() are not supported on all
440 // versions. Once the support for fdopendir(), openat() and unlinkat()
441 // is available on all supported versions the code can be changed
442 // to use these functions.
443
444 // Open the directory of the given path, validate it and set the
445 // current working directory to it.
446 // Return a DIR * of the open directory and the saved cwd fd.
447 //
448 static DIR *open_directory_secure_cwd(const char* dirname, int *saved_cwd_fd) {
449
450 // Open the directory.
451 DIR* dirp = open_directory_secure(dirname);
452 if (dirp == NULL) {
453 // Directory doesn't exist or is insecure, so there is nothing to cleanup.
454 return dirp;
455 }
456 int fd = dirp->dd_fd;
457
458 // Open a fd to the cwd and save it off.
459 int result;
460 RESTARTABLE(::open(".", O_RDONLY), result);
461 if (result == OS_ERR) {
462 *saved_cwd_fd = -1;
463 } else {
464 *saved_cwd_fd = result;
465 }
466
467 // Set the current directory to dirname by using the fd of the directory and
468 // handle errors, otherwise shared memory files will be created in cwd.
469 result = fchdir(fd);
470 if (result == OS_ERR) {
471 if (PrintMiscellaneous && Verbose) {
472 warning("could not change to directory %s", dirname);
473 }
474 if (*saved_cwd_fd != -1) {
475 ::close(*saved_cwd_fd);
476 *saved_cwd_fd = -1;
477 }
478 // Close the directory.
479 os::closedir(dirp);
480 return NULL;
481 } else {
482 return dirp;
483 }
484 }
485
486 // Close the directory and restore the current working directory.
487 //
488 static void close_directory_secure_cwd(DIR* dirp, int saved_cwd_fd) {
489
490 int result;
491 // If we have a saved cwd change back to it and close the fd.
492 if (saved_cwd_fd != -1) {
493 result = fchdir(saved_cwd_fd);
494 ::close(saved_cwd_fd);
495 }
496
497 // Close the directory.
498 os::closedir(dirp);
499 }
500
501 // Check if the given file descriptor is considered a secure.
502 static bool is_file_secure(int fd, const char *filename) {
503
504 int result;
505 struct stat statbuf;
506
507 // Determine if the file is secure.
508 RESTARTABLE(::fstat(fd, &statbuf), result);
509 if (result == OS_ERR) {
510 if (PrintMiscellaneous && Verbose) {
511 warning("fstat failed on %s: %s\n", filename, os::strerror(errno));
512 }
513 return false;
514 }
515 if (statbuf.st_nlink > 1) {
516 // A file with multiple links is not expected.
517 if (PrintMiscellaneous && Verbose) {
518 warning("file %s has multiple links\n", filename);
519 }
520 return false;
521 }
522 return true;
523 }
524
525 // Return the user name for the given user id.
526 //
527 // The caller is expected to free the allocated memory.
528 static char* get_user_name(uid_t uid) {
529
530 struct passwd pwent;
531
532 // Determine the max pwbuf size from sysconf, and hardcode
533 // a default if this not available through sysconf.
534 long bufsize = sysconf(_SC_GETPW_R_SIZE_MAX);
535 if (bufsize == -1)
536 bufsize = 1024;
537
538 char* pwbuf = NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
539
540 struct passwd* p;
541 int result = getpwuid_r(uid, &pwent, pwbuf, (size_t)bufsize, &p);
542
543 if (result != 0 || p == NULL || p->pw_name == NULL || *(p->pw_name) == '\0') {
544 if (PrintMiscellaneous && Verbose) {
545 if (result != 0) {
546 warning("Could not retrieve passwd entry: %s\n",
547 os::strerror(result));
548 }
549 else if (p == NULL) {
550 // this check is added to protect against an observed problem
551 // with getpwuid_r() on RedHat 9 where getpwuid_r returns 0,
552 // indicating success, but has p == NULL. This was observed when
553 // inserting a file descriptor exhaustion fault prior to the call
554 // getpwuid_r() call. In this case, error is set to the appropriate
555 // error condition, but this is undocumented behavior. This check
556 // is safe under any condition, but the use of errno in the output
557 // message may result in an erroneous message.
558 // Bug Id 89052 was opened with RedHat.
559 //
560 warning("Could not retrieve passwd entry: %s\n",
561 os::strerror(errno));
562 }
563 else {
564 warning("Could not determine user name: %s\n",
565 p->pw_name == NULL ? "pw_name = NULL" :
566 "pw_name zero length");
567 }
568 }
569 FREE_C_HEAP_ARRAY(char, pwbuf);
570 return NULL;
571 }
572
573 char* user_name = NEW_C_HEAP_ARRAY(char, strlen(p->pw_name) + 1, mtInternal);
574 strcpy(user_name, p->pw_name);
575
576 FREE_C_HEAP_ARRAY(char, pwbuf);
577 return user_name;
578 }
579
580 // return the name of the user that owns the process identified by vmid.
581 //
582 // This method uses a slow directory search algorithm to find the backing
583 // store file for the specified vmid and returns the user name, as determined
584 // by the user name suffix of the hsperfdata_<username> directory name.
585 //
586 // the caller is expected to free the allocated memory.
587 //
588 static char* get_user_name_slow(int vmid, TRAPS) {
589
590 // short circuit the directory search if the process doesn't even exist.
591 if (kill(vmid, 0) == OS_ERR) {
592 if (errno == ESRCH) {
593 THROW_MSG_0(vmSymbols::java_lang_IllegalArgumentException(),
594 "Process not found");
595 }
596 else /* EPERM */ {
597 THROW_MSG_0(vmSymbols::java_io_IOException(), os::strerror(errno));
598 }
599 }
600
601 // directory search
602 char* oldest_user = NULL;
603 time_t oldest_ctime = 0;
604
605 const char* tmpdirname = os::get_temp_directory();
606
607 DIR* tmpdirp = os::opendir(tmpdirname);
608
609 if (tmpdirp == NULL) {
610 return NULL;
611 }
612
613 // for each entry in the directory that matches the pattern hsperfdata_*,
614 // open the directory and check if the file for the given vmid exists.
615 // The file with the expected name and the latest creation date is used
616 // to determine the user name for the process id.
617 //
618 struct dirent* dentry;
619 char* tdbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(tmpdirname), mtInternal);
620 errno = 0;
621 while ((dentry = os::readdir(tmpdirp, (struct dirent *)tdbuf)) != NULL) {
622
623 // check if the directory entry is a hsperfdata file
624 if (strncmp(dentry->d_name, PERFDATA_NAME, strlen(PERFDATA_NAME)) != 0) {
625 continue;
626 }
627
628 char* usrdir_name = NEW_C_HEAP_ARRAY(char,
629 strlen(tmpdirname) + strlen(dentry->d_name) + 2, mtInternal);
630 strcpy(usrdir_name, tmpdirname);
631 strcat(usrdir_name, "/");
632 strcat(usrdir_name, dentry->d_name);
633
634 // Open the user directory.
635 DIR* subdirp = open_directory_secure(usrdir_name);
636
637 if (subdirp == NULL) {
638 FREE_C_HEAP_ARRAY(char, usrdir_name);
639 continue;
640 }
641
642 // Since we don't create the backing store files in directories
643 // pointed to by symbolic links, we also don't follow them when
644 // looking for the files. We check for a symbolic link after the
645 // call to opendir in order to eliminate a small window where the
646 // symlink can be exploited.
647 //
648 if (!is_directory_secure(usrdir_name)) {
649 FREE_C_HEAP_ARRAY(char, usrdir_name);
650 os::closedir(subdirp);
651 continue;
652 }
653
654 struct dirent* udentry;
655 char* udbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(usrdir_name), mtInternal);
656 errno = 0;
657 while ((udentry = os::readdir(subdirp, (struct dirent *)udbuf)) != NULL) {
658
659 if (filename_to_pid(udentry->d_name) == vmid) {
660 struct stat statbuf;
661 int result;
662
663 char* filename = NEW_C_HEAP_ARRAY(char,
664 strlen(usrdir_name) + strlen(udentry->d_name) + 2, mtInternal);
665
666 strcpy(filename, usrdir_name);
667 strcat(filename, "/");
668 strcat(filename, udentry->d_name);
669
670 // don't follow symbolic links for the file
671 RESTARTABLE(::lstat(filename, &statbuf), result);
672 if (result == OS_ERR) {
673 FREE_C_HEAP_ARRAY(char, filename);
674 continue;
675 }
676
677 // skip over files that are not regular files.
678 if (!S_ISREG(statbuf.st_mode)) {
679 FREE_C_HEAP_ARRAY(char, filename);
680 continue;
681 }
682
683 // compare and save filename with latest creation time
684 if (statbuf.st_size > 0 && statbuf.st_ctime > oldest_ctime) {
685
686 if (statbuf.st_ctime > oldest_ctime) {
687 char* user = strchr(dentry->d_name, '_') + 1;
688
689 if (oldest_user != NULL) FREE_C_HEAP_ARRAY(char, oldest_user);
690 oldest_user = NEW_C_HEAP_ARRAY(char, strlen(user)+1, mtInternal);
691
692 strcpy(oldest_user, user);
693 oldest_ctime = statbuf.st_ctime;
694 }
695 }
696
697 FREE_C_HEAP_ARRAY(char, filename);
698 }
699 }
700 os::closedir(subdirp);
701 FREE_C_HEAP_ARRAY(char, udbuf);
702 FREE_C_HEAP_ARRAY(char, usrdir_name);
703 }
704 os::closedir(tmpdirp);
705 FREE_C_HEAP_ARRAY(char, tdbuf);
706
707 return(oldest_user);
708 }
709
710 // return the name of the user that owns the JVM indicated by the given vmid.
711 //
712 static char* get_user_name(int vmid, TRAPS) {
713 return get_user_name_slow(vmid, THREAD);
714 }
715
716 // return the file name of the backing store file for the named
717 // shared memory region for the given user name and vmid.
718 //
719 // the caller is expected to free the allocated memory.
720 //
721 static char* get_sharedmem_filename(const char* dirname, int vmid) {
722
723 // add 2 for the file separator and a null terminator.
724 size_t nbytes = strlen(dirname) + UINT_CHARS + 2;
725
726 char* name = NEW_C_HEAP_ARRAY(char, nbytes, mtInternal);
727 snprintf(name, nbytes, "%s/%d", dirname, vmid);
728
729 return name;
730 }
731
732
733 // remove file
734 //
735 // this method removes the file specified by the given path
736 //
737 static void remove_file(const char* path) {
738
739 int result;
740
741 // if the file is a directory, the following unlink will fail. since
742 // we don't expect to find directories in the user temp directory, we
743 // won't try to handle this situation. even if accidentially or
744 // maliciously planted, the directory's presence won't hurt anything.
745 //
746 RESTARTABLE(::unlink(path), result);
747 if (PrintMiscellaneous && Verbose && result == OS_ERR) {
748 if (errno != ENOENT) {
749 warning("Could not unlink shared memory backing"
750 " store file %s : %s\n", path, os::strerror(errno));
751 }
752 }
753 }
754
755 // Cleanup stale shared memory resources
756 //
757 // This method attempts to remove all stale shared memory files in
758 // the named user temporary directory. It scans the named directory
759 // for files matching the pattern ^$[0-9]*$. For each file found, the
760 // process id is extracted from the file name and a test is run to
761 // determine if the process is alive. If the process is not alive,
762 // any stale file resources are removed.
763 static void cleanup_sharedmem_resources(const char* dirname) {
764
765 int saved_cwd_fd;
766 // Open the directory.
767 DIR* dirp = open_directory_secure_cwd(dirname, &saved_cwd_fd);
768 if (dirp == NULL) {
769 // Directory doesn't exist or is insecure, so there is nothing to cleanup.
770 return;
771 }
772
773 // For each entry in the directory that matches the expected file
774 // name pattern, determine if the file resources are stale and if
775 // so, remove the file resources. Note, instrumented HotSpot processes
776 // for this user may start and/or terminate during this search and
777 // remove or create new files in this directory. The behavior of this
778 // loop under these conditions is dependent upon the implementation of
779 // opendir/readdir.
780 struct dirent* entry;
781 char* dbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(dirname), mtInternal);
782
783 errno = 0;
784 while ((entry = os::readdir(dirp, (struct dirent *)dbuf)) != NULL) {
785
786 pid_t pid = filename_to_pid(entry->d_name);
787
788 if (pid == 0) {
789
790 if (strcmp(entry->d_name, ".") != 0 && strcmp(entry->d_name, "..") != 0) {
791
792 // Attempt to remove all unexpected files, except "." and "..".
793 unlink(entry->d_name);
794 }
795
796 errno = 0;
797 continue;
798 }
799
800 // We now have a file name that converts to a valid integer
801 // that could represent a process id . if this process id
802 // matches the current process id or the process is not running,
803 // then remove the stale file resources.
804 //
805 // Process liveness is detected by sending signal number 0 to
806 // the process id (see kill(2)). if kill determines that the
807 // process does not exist, then the file resources are removed.
808 // if kill determines that that we don't have permission to
809 // signal the process, then the file resources are assumed to
810 // be stale and are removed because the resources for such a
811 // process should be in a different user specific directory.
812 if ((pid == os::current_process_id()) ||
813 (kill(pid, 0) == OS_ERR && (errno == ESRCH || errno == EPERM))) {
814
815 unlink(entry->d_name);
816 }
817 errno = 0;
818 }
819
820 // Close the directory and reset the current working directory.
821 close_directory_secure_cwd(dirp, saved_cwd_fd);
822
823 FREE_C_HEAP_ARRAY(char, dbuf);
824 }
825
826 // Make the user specific temporary directory. Returns true if
827 // the directory exists and is secure upon return. Returns false
828 // if the directory exists but is either a symlink, is otherwise
829 // insecure, or if an error occurred.
830 static bool make_user_tmp_dir(const char* dirname) {
831
832 // Create the directory with 0755 permissions. note that the directory
833 // will be owned by euid::egid, which may not be the same as uid::gid.
834 if (mkdir(dirname, S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH) == OS_ERR) {
835 if (errno == EEXIST) {
836 // The directory already exists and was probably created by another
837 // JVM instance. However, this could also be the result of a
838 // deliberate symlink. Verify that the existing directory is safe.
839 if (!is_directory_secure(dirname)) {
840 // Directory is not secure.
841 if (PrintMiscellaneous && Verbose) {
842 warning("%s directory is insecure\n", dirname);
843 }
844 return false;
845 }
846 }
847 else {
848 // we encountered some other failure while attempting
849 // to create the directory
850 //
851 if (PrintMiscellaneous && Verbose) {
852 warning("could not create directory %s: %s\n",
853 dirname, os::strerror(errno));
854 }
855 return false;
856 }
857 }
858 return true;
859 }
860
861 // create the shared memory file resources
862 //
863 // This method creates the shared memory file with the given size
864 // This method also creates the user specific temporary directory, if
865 // it does not yet exist.
866 //
867 static int create_sharedmem_resources(const char* dirname, const char* filename, size_t size) {
868
869 // make the user temporary directory
870 if (!make_user_tmp_dir(dirname)) {
871 // could not make/find the directory or the found directory
872 // was not secure
873 return -1;
874 }
875
876 int saved_cwd_fd;
877 // Open the directory and set the current working directory to it.
878 DIR* dirp = open_directory_secure_cwd(dirname, &saved_cwd_fd);
879 if (dirp == NULL) {
880 // Directory doesn't exist or is insecure, so cannot create shared
881 // memory file.
882 return -1;
883 }
884
885 // Open the filename in the current directory.
886 // Cannot use O_TRUNC here; truncation of an existing file has to happen
887 // after the is_file_secure() check below.
888 int result;
889
890 // No O_NOFOLLOW defined at buildtime, and it is not documented for open;
891 // so provide a workaround in this case.
892 #ifdef O_NOFOLLOW
893 RESTARTABLE(::open(filename, O_RDWR|O_CREAT|O_NOFOLLOW, S_IREAD|S_IWRITE), result);
894 #else
895 // workaround function (jdk6 code)
896 result = open_o_nofollow(filename, O_RDWR|O_CREAT, S_IREAD|S_IWRITE);
897 #endif
898
899 if (result == OS_ERR) {
900 if (PrintMiscellaneous && Verbose) {
901 if (errno == ELOOP) {
902 warning("file %s is a symlink and is not secure\n", filename);
903 } else {
904 warning("could not create file %s: %s\n", filename, os::strerror(errno));
905 }
906 }
907 // Close the directory and reset the current working directory.
908 close_directory_secure_cwd(dirp, saved_cwd_fd);
909
910 return -1;
911 }
912 // Close the directory and reset the current working directory.
913 close_directory_secure_cwd(dirp, saved_cwd_fd);
914
915 // save the file descriptor
916 int fd = result;
917
918 // Check to see if the file is secure.
919 if (!is_file_secure(fd, filename)) {
920 ::close(fd);
921 return -1;
922 }
923
924 // Truncate the file to get rid of any existing data.
925 RESTARTABLE(::ftruncate(fd, (off_t)0), result);
926 if (result == OS_ERR) {
927 if (PrintMiscellaneous && Verbose) {
928 warning("could not truncate shared memory file: %s\n", os::strerror(errno));
929 }
930 ::close(fd);
931 return -1;
932 }
933 // set the file size
934 RESTARTABLE(::ftruncate(fd, (off_t)size), result);
935 if (result == OS_ERR) {
936 if (PrintMiscellaneous && Verbose) {
937 warning("could not set shared memory file size: %s\n", os::strerror(errno));
938 }
939 ::close(fd);
940 return -1;
941 }
942
943 return fd;
944 }
945
946 // open the shared memory file for the given user and vmid. returns
947 // the file descriptor for the open file or -1 if the file could not
948 // be opened.
949 //
950 static int open_sharedmem_file(const char* filename, int oflags, TRAPS) {
951
952 // open the file
953 int result;
954 // provide a workaround in case no O_NOFOLLOW is defined at buildtime
955 #ifdef O_NOFOLLOW
956 RESTARTABLE(::open(filename, oflags), result);
957 #else
958 result = open_o_nofollow(filename, oflags);
959 #endif
960 if (result == OS_ERR) {
961 if (errno == ENOENT) {
962 THROW_MSG_(vmSymbols::java_lang_IllegalArgumentException(),
963 "Process not found", OS_ERR);
964 }
965 else if (errno == EACCES) {
966 THROW_MSG_(vmSymbols::java_lang_IllegalArgumentException(),
967 "Permission denied", OS_ERR);
968 }
969 else {
970 THROW_MSG_(vmSymbols::java_io_IOException(),
971 os::strerror(errno), OS_ERR);
972 }
973 }
974 int fd = result;
975
976 // Check to see if the file is secure.
977 if (!is_file_secure(fd, filename)) {
978 ::close(fd);
979 return -1;
980 }
981
982 return fd;
983 }
984
985 // create a named shared memory region. returns the address of the
986 // memory region on success or NULL on failure. A return value of
987 // NULL will ultimately disable the shared memory feature.
988 //
989 // On AIX, the name space for shared memory objects
990 // is the file system name space.
991 //
992 // A monitoring application attaching to a JVM does not need to know
993 // the file system name of the shared memory object. However, it may
994 // be convenient for applications to discover the existence of newly
995 // created and terminating JVMs by watching the file system name space
996 // for files being created or removed.
997 //
998 static char* mmap_create_shared(size_t size) {
999
1000 int result;
1001 int fd;
1002 char* mapAddress;
1003
1004 int vmid = os::current_process_id();
1005
1006 char* user_name = get_user_name(geteuid());
1007
1008 if (user_name == NULL)
1009 return NULL;
1010
1011 char* dirname = get_user_tmp_dir(user_name);
1012 char* filename = get_sharedmem_filename(dirname, vmid);
1013
1014 // get the short filename.
1015 char* short_filename = strrchr(filename, '/');
1016 if (short_filename == NULL) {
1017 short_filename = filename;
1018 } else {
1019 short_filename++;
1020 }
1021
1022 // cleanup any stale shared memory files
1023 cleanup_sharedmem_resources(dirname);
1024
1025 assert(((size > 0) && (size % os::vm_page_size() == 0)),
1026 "unexpected PerfMemory region size");
1027
1028 fd = create_sharedmem_resources(dirname, short_filename, size);
1029
1030 FREE_C_HEAP_ARRAY(char, user_name);
1031 FREE_C_HEAP_ARRAY(char, dirname);
1032
1033 if (fd == -1) {
1034 FREE_C_HEAP_ARRAY(char, filename);
1035 return NULL;
1036 }
1037
1038 mapAddress = (char*)::mmap((char*)0, size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
1039
1040 result = ::close(fd);
1041 assert(result != OS_ERR, "could not close file");
1042
1043 if (mapAddress == MAP_FAILED) {
1044 if (PrintMiscellaneous && Verbose) {
1045 warning("mmap failed - %s\n", os::strerror(errno));
1046 }
1047 remove_file(filename);
1048 FREE_C_HEAP_ARRAY(char, filename);
1049 return NULL;
1050 }
1051
1052 // save the file name for use in delete_shared_memory()
1053 backing_store_file_name = filename;
1054
1055 // clear the shared memory region
1056 (void)::memset((void*) mapAddress, 0, size);
1057
1058 // It does not go through os api, the operation has to record from here.
1059 MemTracker::record_virtual_memory_reserve((address)mapAddress, size, CURRENT_PC, mtInternal);
1060
1061 return mapAddress;
1062 }
1063
1064 // release a named shared memory region
1065 //
1066 static void unmap_shared(char* addr, size_t bytes) {
1067 // Do not rely on os::reserve_memory/os::release_memory to use mmap.
1068 // Use os::reserve_memory/os::release_memory for PerfDisableSharedMem=1, mmap/munmap for PerfDisableSharedMem=0
1069 if (::munmap(addr, bytes) == -1) {
1070 warning("perfmemory: munmap failed (%d)\n", errno);
1071 }
1072 }
1073
1074 // create the PerfData memory region in shared memory.
1075 //
1076 static char* create_shared_memory(size_t size) {
1077
1078 // create the shared memory region.
1079 return mmap_create_shared(size);
1080 }
1081
1082 // delete the shared PerfData memory region
1083 //
1084 static void delete_shared_memory(char* addr, size_t size) {
1085
1086 // cleanup the persistent shared memory resources. since DestroyJavaVM does
1087 // not support unloading of the JVM, unmapping of the memory resource is
1088 // not performed. The memory will be reclaimed by the OS upon termination of
1089 // the process. The backing store file is deleted from the file system.
1090
1091 assert(!PerfDisableSharedMem, "shouldn't be here");
1092
1093 if (backing_store_file_name != NULL) {
1094 remove_file(backing_store_file_name);
1095 // Don't.. Free heap memory could deadlock os::abort() if it is called
1096 // from signal handler. OS will reclaim the heap memory.
1097 // FREE_C_HEAP_ARRAY(char, backing_store_file_name);
1098 backing_store_file_name = NULL;
1099 }
1100 }
1101
1102 // return the size of the file for the given file descriptor
1103 // or 0 if it is not a valid size for a shared memory file
1104 //
1105 static size_t sharedmem_filesize(int fd, TRAPS) {
1106
1107 struct stat statbuf;
1108 int result;
1109
1110 RESTARTABLE(::fstat(fd, &statbuf), result);
1111 if (result == OS_ERR) {
1112 if (PrintMiscellaneous && Verbose) {
1113 warning("fstat failed: %s\n", os::strerror(errno));
1114 }
1115 THROW_MSG_0(vmSymbols::java_io_IOException(),
1116 "Could not determine PerfMemory size");
1117 }
1118
1119 if ((statbuf.st_size == 0) ||
1120 ((size_t)statbuf.st_size % os::vm_page_size() != 0)) {
1121 THROW_MSG_0(vmSymbols::java_lang_Exception(),
1122 "Invalid PerfMemory size");
1123 }
1124
1125 return (size_t)statbuf.st_size;
1126 }
1127
1128 // attach to a named shared memory region.
1129 //
1130 static void mmap_attach_shared(const char* user, int vmid, PerfMemory::PerfMemoryMode mode, char** addr, size_t* sizep, TRAPS) {
1131
1132 char* mapAddress;
1133 int result;
1134 int fd;
1135 size_t size = 0;
1136 const char* luser = NULL;
1137
1138 int mmap_prot;
1139 int file_flags;
1140
1141 ResourceMark rm;
1142
1143 // map the high level access mode to the appropriate permission
1144 // constructs for the file and the shared memory mapping.
1145 if (mode == PerfMemory::PERF_MODE_RO) {
1146 mmap_prot = PROT_READ;
1147 // No O_NOFOLLOW defined at buildtime, and it is not documented for open.
1148 #ifdef O_NOFOLLOW
1149 file_flags = O_RDONLY | O_NOFOLLOW;
1150 #else
1151 file_flags = O_RDONLY;
1152 #endif
1153 }
1154 else if (mode == PerfMemory::PERF_MODE_RW) {
1155 #ifdef LATER
1156 mmap_prot = PROT_READ | PROT_WRITE;
1157 file_flags = O_RDWR | O_NOFOLLOW;
1158 #else
1159 THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1160 "Unsupported access mode");
1161 #endif
1162 }
1163 else {
1164 THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1165 "Illegal access mode");
1166 }
1167
1168 if (user == NULL || strlen(user) == 0) {
1169 luser = get_user_name(vmid, CHECK);
1170 }
1171 else {
1172 luser = user;
1173 }
1174
1175 if (luser == NULL) {
1176 THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1177 "Could not map vmid to user Name");
1178 }
1179
1180 char* dirname = get_user_tmp_dir(luser);
1181
1182 // since we don't follow symbolic links when creating the backing
1183 // store file, we don't follow them when attaching either.
1184 //
1185 if (!is_directory_secure(dirname)) {
1186 FREE_C_HEAP_ARRAY(char, dirname);
1187 if (luser != user) {
1188 FREE_C_HEAP_ARRAY(char, luser);
1189 }
1190 THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1191 "Process not found");
1192 }
1193
1194 char* filename = get_sharedmem_filename(dirname, vmid);
1195
1196 // copy heap memory to resource memory. the open_sharedmem_file
1197 // method below need to use the filename, but could throw an
1198 // exception. using a resource array prevents the leak that
1199 // would otherwise occur.
1200 char* rfilename = NEW_RESOURCE_ARRAY(char, strlen(filename) + 1);
1201 strcpy(rfilename, filename);
1202
1203 // free the c heap resources that are no longer needed
1204 if (luser != user) FREE_C_HEAP_ARRAY(char, luser);
1205 FREE_C_HEAP_ARRAY(char, dirname);
1206 FREE_C_HEAP_ARRAY(char, filename);
1207
1208 // open the shared memory file for the give vmid
1209 fd = open_sharedmem_file(rfilename, file_flags, THREAD);
1210
1211 if (fd == OS_ERR) {
1212 return;
1213 }
1214
1215 if (HAS_PENDING_EXCEPTION) {
1216 ::close(fd);
1217 return;
1218 }
1219
1220 if (*sizep == 0) {
1221 size = sharedmem_filesize(fd, CHECK);
1222 } else {
1223 size = *sizep;
1224 }
1225
1226 assert(size > 0, "unexpected size <= 0");
1227
1228 mapAddress = (char*)::mmap((char*)0, size, mmap_prot, MAP_SHARED, fd, 0);
1229
1230 result = ::close(fd);
1231 assert(result != OS_ERR, "could not close file");
1232
1233 if (mapAddress == MAP_FAILED) {
1234 if (PrintMiscellaneous && Verbose) {
1235 warning("mmap failed: %s\n", os::strerror(errno));
1236 }
1237 THROW_MSG(vmSymbols::java_lang_OutOfMemoryError(),
1238 "Could not map PerfMemory");
1239 }
1240
1241 // it does not go through os api, the operation has to record from here.
1242 MemTracker::record_virtual_memory_reserve((address)mapAddress, size, CURRENT_PC, mtInternal);
1243
1244 *addr = mapAddress;
1245 *sizep = size;
1246
1247 if (PerfTraceMemOps) {
1248 tty->print("mapped " SIZE_FORMAT " bytes for vmid %d at "
1249 INTPTR_FORMAT "\n", size, vmid, p2i((void*)mapAddress));
1250 }
1251 }
1252
1253
1254
1255
1256 // create the PerfData memory region
1257 //
1258 // This method creates the memory region used to store performance
1259 // data for the JVM. The memory may be created in standard or
1260 // shared memory.
1261 //
1262 void PerfMemory::create_memory_region(size_t size) {
1263
1264 if (PerfDisableSharedMem) {
1265 // do not share the memory for the performance data.
1266 _start = create_standard_memory(size);
1267 }
1268 else {
1269 _start = create_shared_memory(size);
1270 if (_start == NULL) {
1271
1272 // creation of the shared memory region failed, attempt
1273 // to create a contiguous, non-shared memory region instead.
1274 //
1275 if (PrintMiscellaneous && Verbose) {
1276 warning("Reverting to non-shared PerfMemory region.\n");
1277 }
1278 PerfDisableSharedMem = true;
1279 _start = create_standard_memory(size);
1280 }
1281 }
1282
1283 if (_start != NULL) _capacity = size;
1284
1285 }
1286
1287 // delete the PerfData memory region
1288 //
1289 // This method deletes the memory region used to store performance
1290 // data for the JVM. The memory region indicated by the <address, size>
1291 // tuple will be inaccessible after a call to this method.
1292 //
1293 void PerfMemory::delete_memory_region() {
1294
1295 assert((start() != NULL && capacity() > 0), "verify proper state");
1296
1297 // If user specifies PerfDataSaveFile, it will save the performance data
1298 // to the specified file name no matter whether PerfDataSaveToFile is specified
1299 // or not. In other word, -XX:PerfDataSaveFile=.. overrides flag
1300 // -XX:+PerfDataSaveToFile.
1301 if (PerfDataSaveToFile || PerfDataSaveFile != NULL) {
1302 save_memory_to_file(start(), capacity());
1303 }
1304
1305 if (PerfDisableSharedMem) {
1306 delete_standard_memory(start(), capacity());
1307 }
1308 else {
1309 delete_shared_memory(start(), capacity());
1310 }
1311 }
1312
1313 // attach to the PerfData memory region for another JVM
1314 //
1315 // This method returns an <address, size> tuple that points to
1316 // a memory buffer that is kept reasonably synchronized with
1317 // the PerfData memory region for the indicated JVM. This
1318 // buffer may be kept in synchronization via shared memory
1319 // or some other mechanism that keeps the buffer updated.
1320 //
1321 // If the JVM chooses not to support the attachability feature,
1322 // this method should throw an UnsupportedOperation exception.
1323 //
1324 // This implementation utilizes named shared memory to map
1325 // the indicated process's PerfData memory region into this JVMs
1326 // address space.
1327 //
1328 void PerfMemory::attach(const char* user, int vmid, PerfMemoryMode mode, char** addrp, size_t* sizep, TRAPS) {
1329
1330 if (vmid == 0 || vmid == os::current_process_id()) {
1331 *addrp = start();
1332 *sizep = capacity();
1333 return;
1334 }
1335
1336 mmap_attach_shared(user, vmid, mode, addrp, sizep, CHECK);
1337 }
1338
1339 // detach from the PerfData memory region of another JVM
1340 //
1341 // This method detaches the PerfData memory region of another
1342 // JVM, specified as an <address, size> tuple of a buffer
1343 // in this process's address space. This method may perform
1344 // arbitrary actions to accomplish the detachment. The memory
1345 // region specified by <address, size> will be inaccessible after
1346 // a call to this method.
1347 //
1348 // If the JVM chooses not to support the attachability feature,
1349 // this method should throw an UnsupportedOperation exception.
1350 //
1351 // This implementation utilizes named shared memory to detach
1352 // the indicated process's PerfData memory region from this
1353 // process's address space.
1354 //
1355 void PerfMemory::detach(char* addr, size_t bytes, TRAPS) {
1356
1357 assert(addr != 0, "address sanity check");
1358 assert(bytes > 0, "capacity sanity check");
1359
1360 if (PerfMemory::contains(addr) || PerfMemory::contains(addr + bytes - 1)) {
1361 // prevent accidental detachment of this process's PerfMemory region
1362 return;
1363 }
1364
1365 unmap_shared(addr, bytes);
1366 }