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