1 /*
2 * Copyright (c) 2001, 2015, Oracle and/or its affiliates. All rights reserved.
3 * Copyright 2012, 2013 SAP AG. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
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 RESTARTABLE(::close(fd), result);
125 if (PrintMiscellaneous && Verbose) {
126 if (result == OS_ERR) {
127 warning("Could not close %s: %s\n", destfile, strerror(errno));
128 }
129 }
130 }
131 FREE_C_HEAP_ARRAY(char, destfile);
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
303 create = false;
304
305 if (lstat(path, &orig_st) != 0) {
306 if (errno == ENOENT && (oflag & O_CREAT) != 0) {
307 // File doesn't exist, but_we want to create it, add O_EXCL flag
308 // to make sure no-one creates it (or a symlink) before us
309 // This works as we expect with symlinks, from posix man page:
310 // 'If O_EXCL and O_CREAT are set, and path names a symbolic
311 // link, open() shall fail and set errno to [EEXIST]'.
312 oflag |= O_EXCL;
313 create = true;
314 } else {
315 // File doesn't exist, and we are not creating it.
316 return OS_ERR;
317 }
318 } else {
319 // Lstat success, check if existing file is a link.
320 if ((orig_st.st_mode & S_IFMT) == S_IFLNK) {
321 // File is a symlink.
322 errno = ELOOP;
323 return OS_ERR;
324 }
325 }
326
327 if (use_mode == true) {
328 fd = open(path, oflag, mode);
329 } else {
330 fd = open(path, oflag);
331 }
332
333 if (fd == OS_ERR) {
334 return fd;
335 }
336
337 // Can't do inode checks on before/after if we created the file.
338 if (create == false) {
339 if (fstat(fd, &new_st) != 0) {
340 // Keep errno from fstat, in case close also fails.
341 error = errno;
342 ::close(fd);
343 errno = error;
344 return OS_ERR;
345 }
346
347 if (orig_st.st_dev != new_st.st_dev || orig_st.st_ino != new_st.st_ino) {
348 // File was tampered with during race window.
349 ::close(fd);
350 errno = EEXIST;
351 if (PrintMiscellaneous && Verbose) {
352 warning("possible file tampering attempt detected when opening %s", path);
353 }
354 return OS_ERR;
355 }
356 }
357
358 return fd;
359 }
360
361 static int open_o_nofollow(const char* path, int oflag, mode_t mode) {
362 return open_o_nofollow_impl(path, oflag, mode, true);
363 }
364
365 static int open_o_nofollow(const char* path, int oflag) {
366 return open_o_nofollow_impl(path, oflag, 0, false);
367 }
368 #endif
369
370 // Open the directory of the given path and validate it.
371 // Return a DIR * of the open directory.
372 static DIR *open_directory_secure(const char* dirname) {
373 // Open the directory using open() so that it can be verified
374 // to be secure by calling is_dirfd_secure(), opendir() and then check
375 // to see if they are the same file system object. This method does not
376 // introduce a window of opportunity for the directory to be attacked that
377 // calling opendir() and is_directory_secure() does.
378 int result;
379 DIR *dirp = NULL;
380
381 // No O_NOFOLLOW defined at buildtime, and it is not documented for open;
382 // so provide a workaround in this case.
383 #ifdef O_NOFOLLOW
384 RESTARTABLE(::open(dirname, O_RDONLY|O_NOFOLLOW), result);
385 #else
386 // workaround (jdk6 coding)
387 RESTARTABLE(::open_o_nofollow(dirname, O_RDONLY), result);
388 #endif
389
390 if (result == OS_ERR) {
391 // Directory doesn't exist or is a symlink, so there is nothing to cleanup.
392 if (PrintMiscellaneous && Verbose) {
393 if (errno == ELOOP) {
394 warning("directory %s is a symlink and is not secure\n", dirname);
395 } else {
396 warning("could not open directory %s: %s\n", dirname, strerror(errno));
397 }
398 }
399 return dirp;
400 }
401 int fd = result;
402
403 // Determine if the open directory is secure.
404 if (!is_dirfd_secure(fd)) {
405 // The directory is not a secure directory.
406 os::close(fd);
407 return dirp;
408 }
409
410 // Open the directory.
411 dirp = ::opendir(dirname);
412 if (dirp == NULL) {
413 // The directory doesn't exist, close fd and return.
414 os::close(fd);
415 return dirp;
416 }
417
418 // Check to make sure fd and dirp are referencing the same file system object.
419 if (!is_same_fsobject(fd, dirp->dd_fd)) {
420 // The directory is not secure.
421 os::close(fd);
422 os::closedir(dirp);
423 dirp = NULL;
424 return dirp;
425 }
426
427 // Close initial open now that we know directory is secure
428 os::close(fd);
429
430 return dirp;
431 }
432
433 // NOTE: The code below uses fchdir(), open() and unlink() because
434 // fdopendir(), openat() and unlinkat() are not supported on all
435 // versions. Once the support for fdopendir(), openat() and unlinkat()
436 // is available on all supported versions the code can be changed
437 // to use these functions.
438
439 // Open the directory of the given path, validate it and set the
440 // current working directory to it.
441 // Return a DIR * of the open directory and the saved cwd fd.
442 //
443 static DIR *open_directory_secure_cwd(const char* dirname, int *saved_cwd_fd) {
444
445 // Open the directory.
446 DIR* dirp = open_directory_secure(dirname);
447 if (dirp == NULL) {
448 // Directory doesn't exist or is insecure, so there is nothing to cleanup.
449 return dirp;
450 }
451 int fd = dirp->dd_fd;
452
453 // Open a fd to the cwd and save it off.
454 int result;
455 RESTARTABLE(::open(".", O_RDONLY), result);
456 if (result == OS_ERR) {
457 *saved_cwd_fd = -1;
458 } else {
459 *saved_cwd_fd = result;
460 }
461
462 // Set the current directory to dirname by using the fd of the directory and
463 // handle errors, otherwise shared memory files will be created in cwd.
464 result = fchdir(fd);
465 if (result == OS_ERR) {
466 if (PrintMiscellaneous && Verbose) {
467 warning("could not change to directory %s", dirname);
468 }
469 if (*saved_cwd_fd != -1) {
470 ::close(*saved_cwd_fd);
471 *saved_cwd_fd = -1;
472 }
473 // Close the directory.
474 os::closedir(dirp);
475 return NULL;
476 } else {
477 return dirp;
478 }
479 }
480
481 // Close the directory and restore the current working directory.
482 //
483 static void close_directory_secure_cwd(DIR* dirp, int saved_cwd_fd) {
484
485 int result;
486 // If we have a saved cwd change back to it and close the fd.
487 if (saved_cwd_fd != -1) {
488 result = fchdir(saved_cwd_fd);
489 ::close(saved_cwd_fd);
490 }
491
492 // Close the directory.
493 os::closedir(dirp);
494 }
495
496 // Check if the given file descriptor is considered a secure.
497 static bool is_file_secure(int fd, const char *filename) {
498
499 int result;
500 struct stat statbuf;
501
502 // Determine if the file is secure.
503 RESTARTABLE(::fstat(fd, &statbuf), result);
504 if (result == OS_ERR) {
505 if (PrintMiscellaneous && Verbose) {
506 warning("fstat failed on %s: %s\n", filename, strerror(errno));
507 }
508 return false;
509 }
510 if (statbuf.st_nlink > 1) {
511 // A file with multiple links is not expected.
512 if (PrintMiscellaneous && Verbose) {
513 warning("file %s has multiple links\n", filename);
514 }
515 return false;
516 }
517 return true;
518 }
519
520 // Return the user name for the given user id.
521 //
522 // The caller is expected to free the allocated memory.
523 static char* get_user_name(uid_t uid) {
524
525 struct passwd pwent;
526
527 // Determine the max pwbuf size from sysconf, and hardcode
528 // a default if this not available through sysconf.
529 long bufsize = sysconf(_SC_GETPW_R_SIZE_MAX);
530 if (bufsize == -1)
531 bufsize = 1024;
532
533 char* pwbuf = NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
534
535 struct passwd* p;
536 int result = getpwuid_r(uid, &pwent, pwbuf, (size_t)bufsize, &p);
537
538 if (result != 0 || p == NULL || p->pw_name == NULL || *(p->pw_name) == '\0') {
539 if (PrintMiscellaneous && Verbose) {
540 if (result != 0) {
541 warning("Could not retrieve passwd entry: %s\n",
542 strerror(result));
543 }
544 else if (p == NULL) {
545 // this check is added to protect against an observed problem
546 // with getpwuid_r() on RedHat 9 where getpwuid_r returns 0,
547 // indicating success, but has p == NULL. This was observed when
548 // inserting a file descriptor exhaustion fault prior to the call
549 // getpwuid_r() call. In this case, error is set to the appropriate
550 // error condition, but this is undocumented behavior. This check
551 // is safe under any condition, but the use of errno in the output
552 // message may result in an erroneous message.
553 // Bug Id 89052 was opened with RedHat.
554 //
555 warning("Could not retrieve passwd entry: %s\n",
556 strerror(errno));
557 }
558 else {
559 warning("Could not determine user name: %s\n",
560 p->pw_name == NULL ? "pw_name = NULL" :
561 "pw_name zero length");
562 }
563 }
564 FREE_C_HEAP_ARRAY(char, pwbuf);
565 return NULL;
566 }
567
568 char* user_name = NEW_C_HEAP_ARRAY(char, strlen(p->pw_name) + 1, mtInternal);
569 strcpy(user_name, p->pw_name);
570
571 FREE_C_HEAP_ARRAY(char, pwbuf);
572 return user_name;
573 }
574
575 // return the name of the user that owns the process identified by vmid.
576 //
577 // This method uses a slow directory search algorithm to find the backing
578 // store file for the specified vmid and returns the user name, as determined
579 // by the user name suffix of the hsperfdata_<username> directory name.
580 //
581 // the caller is expected to free the allocated memory.
582 //
583 static char* get_user_name_slow(int vmid, TRAPS) {
584
585 // short circuit the directory search if the process doesn't even exist.
586 if (kill(vmid, 0) == OS_ERR) {
587 if (errno == ESRCH) {
588 THROW_MSG_0(vmSymbols::java_lang_IllegalArgumentException(),
589 "Process not found");
590 }
591 else /* EPERM */ {
592 THROW_MSG_0(vmSymbols::java_io_IOException(), strerror(errno));
593 }
594 }
595
596 // directory search
597 char* oldest_user = NULL;
598 time_t oldest_ctime = 0;
599
600 const char* tmpdirname = os::get_temp_directory();
601
602 DIR* tmpdirp = os::opendir(tmpdirname);
603
604 if (tmpdirp == NULL) {
605 return NULL;
606 }
607
608 // for each entry in the directory that matches the pattern hsperfdata_*,
609 // open the directory and check if the file for the given vmid exists.
610 // The file with the expected name and the latest creation date is used
611 // to determine the user name for the process id.
612 //
613 struct dirent* dentry;
614 char* tdbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(tmpdirname), mtInternal);
615 errno = 0;
616 while ((dentry = os::readdir(tmpdirp, (struct dirent *)tdbuf)) != NULL) {
617
618 // check if the directory entry is a hsperfdata file
619 if (strncmp(dentry->d_name, PERFDATA_NAME, strlen(PERFDATA_NAME)) != 0) {
620 continue;
621 }
622
623 char* usrdir_name = NEW_C_HEAP_ARRAY(char,
624 strlen(tmpdirname) + strlen(dentry->d_name) + 2, mtInternal);
625 strcpy(usrdir_name, tmpdirname);
626 strcat(usrdir_name, "/");
627 strcat(usrdir_name, dentry->d_name);
628
629 // Open the user directory.
630 DIR* subdirp = open_directory_secure(usrdir_name);
631
632 if (subdirp == NULL) {
633 FREE_C_HEAP_ARRAY(char, usrdir_name);
634 continue;
635 }
636
637 // Since we don't create the backing store files in directories
638 // pointed to by symbolic links, we also don't follow them when
639 // looking for the files. We check for a symbolic link after the
640 // call to opendir in order to eliminate a small window where the
641 // symlink can be exploited.
642 //
643 if (!is_directory_secure(usrdir_name)) {
644 FREE_C_HEAP_ARRAY(char, usrdir_name);
645 os::closedir(subdirp);
646 continue;
647 }
648
649 struct dirent* udentry;
650 char* udbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(usrdir_name), mtInternal);
651 errno = 0;
652 while ((udentry = os::readdir(subdirp, (struct dirent *)udbuf)) != NULL) {
653
654 if (filename_to_pid(udentry->d_name) == vmid) {
655 struct stat statbuf;
656 int result;
657
658 char* filename = NEW_C_HEAP_ARRAY(char,
659 strlen(usrdir_name) + strlen(udentry->d_name) + 2, mtInternal);
660
661 strcpy(filename, usrdir_name);
662 strcat(filename, "/");
663 strcat(filename, udentry->d_name);
664
665 // don't follow symbolic links for the file
666 RESTARTABLE(::lstat(filename, &statbuf), result);
667 if (result == OS_ERR) {
668 FREE_C_HEAP_ARRAY(char, filename);
669 continue;
670 }
671
672 // skip over files that are not regular files.
673 if (!S_ISREG(statbuf.st_mode)) {
674 FREE_C_HEAP_ARRAY(char, filename);
675 continue;
676 }
677
678 // compare and save filename with latest creation time
679 if (statbuf.st_size > 0 && statbuf.st_ctime > oldest_ctime) {
680
681 if (statbuf.st_ctime > oldest_ctime) {
682 char* user = strchr(dentry->d_name, '_') + 1;
683
684 if (oldest_user != NULL) FREE_C_HEAP_ARRAY(char, oldest_user);
685 oldest_user = NEW_C_HEAP_ARRAY(char, strlen(user)+1, mtInternal);
686
687 strcpy(oldest_user, user);
688 oldest_ctime = statbuf.st_ctime;
689 }
690 }
691
692 FREE_C_HEAP_ARRAY(char, filename);
693 }
694 }
695 os::closedir(subdirp);
696 FREE_C_HEAP_ARRAY(char, udbuf);
697 FREE_C_HEAP_ARRAY(char, usrdir_name);
698 }
699 os::closedir(tmpdirp);
700 FREE_C_HEAP_ARRAY(char, tdbuf);
701
702 return(oldest_user);
703 }
704
705 // return the name of the user that owns the JVM indicated by the given vmid.
706 //
707 static char* get_user_name(int vmid, TRAPS) {
708 return get_user_name_slow(vmid, THREAD);
709 }
710
711 // return the file name of the backing store file for the named
712 // shared memory region for the given user name and vmid.
713 //
714 // the caller is expected to free the allocated memory.
715 //
716 static char* get_sharedmem_filename(const char* dirname, int vmid) {
717
718 // add 2 for the file separator and a null terminator.
719 size_t nbytes = strlen(dirname) + UINT_CHARS + 2;
720
721 char* name = NEW_C_HEAP_ARRAY(char, nbytes, mtInternal);
722 snprintf(name, nbytes, "%s/%d", dirname, vmid);
723
724 return name;
725 }
726
727
728 // remove file
729 //
730 // this method removes the file specified by the given path
731 //
732 static void remove_file(const char* path) {
733
734 int result;
735
736 // if the file is a directory, the following unlink will fail. since
737 // we don't expect to find directories in the user temp directory, we
738 // won't try to handle this situation. even if accidentially or
739 // maliciously planted, the directory's presence won't hurt anything.
740 //
741 RESTARTABLE(::unlink(path), result);
742 if (PrintMiscellaneous && Verbose && result == OS_ERR) {
743 if (errno != ENOENT) {
744 warning("Could not unlink shared memory backing"
745 " store file %s : %s\n", path, strerror(errno));
746 }
747 }
748 }
749
750 // Cleanup stale shared memory resources
751 //
752 // This method attempts to remove all stale shared memory files in
753 // the named user temporary directory. It scans the named directory
754 // for files matching the pattern ^$[0-9]*$. For each file found, the
755 // process id is extracted from the file name and a test is run to
756 // determine if the process is alive. If the process is not alive,
757 // any stale file resources are removed.
758 static void cleanup_sharedmem_resources(const char* dirname) {
759
760 int saved_cwd_fd;
761 // Open the directory.
762 DIR* dirp = open_directory_secure_cwd(dirname, &saved_cwd_fd);
763 if (dirp == NULL) {
764 // Directory doesn't exist or is insecure, so there is nothing to cleanup.
765 return;
766 }
767
768 // For each entry in the directory that matches the expected file
769 // name pattern, determine if the file resources are stale and if
770 // so, remove the file resources. Note, instrumented HotSpot processes
771 // for this user may start and/or terminate during this search and
772 // remove or create new files in this directory. The behavior of this
773 // loop under these conditions is dependent upon the implementation of
774 // opendir/readdir.
775 struct dirent* entry;
776 char* dbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(dirname), mtInternal);
777
778 errno = 0;
779 while ((entry = os::readdir(dirp, (struct dirent *)dbuf)) != NULL) {
780
781 pid_t pid = filename_to_pid(entry->d_name);
782
783 if (pid == 0) {
784
785 if (strcmp(entry->d_name, ".") != 0 && strcmp(entry->d_name, "..") != 0) {
786
787 // Attempt to remove all unexpected files, except "." and "..".
788 unlink(entry->d_name);
789 }
790
791 errno = 0;
792 continue;
793 }
794
795 // We now have a file name that converts to a valid integer
796 // that could represent a process id . if this process id
797 // matches the current process id or the process is not running,
798 // then remove the stale file resources.
799 //
800 // Process liveness is detected by sending signal number 0 to
801 // the process id (see kill(2)). if kill determines that the
802 // process does not exist, then the file resources are removed.
803 // if kill determines that that we don't have permission to
804 // signal the process, then the file resources are assumed to
805 // be stale and are removed because the resources for such a
806 // process should be in a different user specific directory.
807 if ((pid == os::current_process_id()) ||
808 (kill(pid, 0) == OS_ERR && (errno == ESRCH || errno == EPERM))) {
809
810 unlink(entry->d_name);
811 }
812 errno = 0;
813 }
814
815 // Close the directory and reset the current working directory.
816 close_directory_secure_cwd(dirp, saved_cwd_fd);
817
818 FREE_C_HEAP_ARRAY(char, dbuf);
819 }
820
821 // Make the user specific temporary directory. Returns true if
822 // the directory exists and is secure upon return. Returns false
823 // if the directory exists but is either a symlink, is otherwise
824 // insecure, or if an error occurred.
825 static bool make_user_tmp_dir(const char* dirname) {
826
827 // Create the directory with 0755 permissions. note that the directory
828 // will be owned by euid::egid, which may not be the same as uid::gid.
829 if (mkdir(dirname, S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH) == OS_ERR) {
830 if (errno == EEXIST) {
831 // The directory already exists and was probably created by another
832 // JVM instance. However, this could also be the result of a
833 // deliberate symlink. Verify that the existing directory is safe.
834 if (!is_directory_secure(dirname)) {
835 // Directory is not secure.
836 if (PrintMiscellaneous && Verbose) {
837 warning("%s directory is insecure\n", dirname);
838 }
839 return false;
840 }
841 }
842 else {
843 // we encountered some other failure while attempting
844 // to create the directory
845 //
846 if (PrintMiscellaneous && Verbose) {
847 warning("could not create directory %s: %s\n",
848 dirname, strerror(errno));
849 }
850 return false;
851 }
852 }
853 return true;
854 }
855
856 // create the shared memory file resources
857 //
858 // This method creates the shared memory file with the given size
859 // This method also creates the user specific temporary directory, if
860 // it does not yet exist.
861 //
862 static int create_sharedmem_resources(const char* dirname, const char* filename, size_t size) {
863
864 // make the user temporary directory
865 if (!make_user_tmp_dir(dirname)) {
866 // could not make/find the directory or the found directory
867 // was not secure
868 return -1;
869 }
870
871 int saved_cwd_fd;
872 // Open the directory and set the current working directory to it.
873 DIR* dirp = open_directory_secure_cwd(dirname, &saved_cwd_fd);
874 if (dirp == NULL) {
875 // Directory doesn't exist or is insecure, so cannot create shared
876 // memory file.
877 return -1;
878 }
879
880 // Open the filename in the current directory.
881 // Cannot use O_TRUNC here; truncation of an existing file has to happen
882 // after the is_file_secure() check below.
883 int result;
884
885 // No O_NOFOLLOW defined at buildtime, and it is not documented for open;
886 // so provide a workaround in this case.
887 #ifdef O_NOFOLLOW
888 RESTARTABLE(::open(filename, O_RDWR|O_CREAT|O_NOFOLLOW, S_IREAD|S_IWRITE), result);
889 #else
890 // workaround function (jdk6 code)
891 RESTARTABLE(::open_o_nofollow(filename, O_RDWR|O_CREAT, S_IREAD|S_IWRITE), result);
892 #endif
893
894 if (result == OS_ERR) {
895 if (PrintMiscellaneous && Verbose) {
896 if (errno == ELOOP) {
897 warning("file %s is a symlink and is not secure\n", filename);
898 } else {
899 warning("could not create file %s: %s\n", filename, strerror(errno));
900 }
901 }
902 // Close the directory and reset the current working directory.
903 close_directory_secure_cwd(dirp, saved_cwd_fd);
904
905 return -1;
906 }
907 // Close the directory and reset the current working directory.
908 close_directory_secure_cwd(dirp, saved_cwd_fd);
909
910 // save the file descriptor
911 int fd = result;
912
913 // Check to see if the file is secure.
914 if (!is_file_secure(fd, filename)) {
915 ::close(fd);
916 return -1;
917 }
918
919 // Truncate the file to get rid of any existing data.
920 RESTARTABLE(::ftruncate(fd, (off_t)0), result);
921 if (result == OS_ERR) {
922 if (PrintMiscellaneous && Verbose) {
923 warning("could not truncate shared memory file: %s\n", strerror(errno));
924 }
925 ::close(fd);
926 return -1;
927 }
928 // set the file size
929 RESTARTABLE(::ftruncate(fd, (off_t)size), result);
930 if (result == OS_ERR) {
931 if (PrintMiscellaneous && Verbose) {
932 warning("could not set shared memory file size: %s\n", strerror(errno));
933 }
934 RESTARTABLE(::close(fd), result);
935 return -1;
936 }
937
938 return fd;
939 }
940
941 // open the shared memory file for the given user and vmid. returns
942 // the file descriptor for the open file or -1 if the file could not
943 // be opened.
944 //
945 static int open_sharedmem_file(const char* filename, int oflags, TRAPS) {
946
947 // open the file
948 int result;
949 // No O_NOFOLLOW defined at buildtime, and it is not documented for open;
950 // so provide a workaround in this case
951 #ifdef O_NOFOLLOW
952 RESTARTABLE(::open(filename, oflags), result);
953 #else
954 RESTARTABLE(::open_o_nofollow(filename, oflags), result);
955 #endif
956
957 if (result == OS_ERR) {
958 if (errno == ENOENT) {
959 THROW_MSG_0(vmSymbols::java_lang_IllegalArgumentException(),
960 "Process not found");
961 }
962 else if (errno == EACCES) {
963 THROW_MSG_0(vmSymbols::java_lang_IllegalArgumentException(),
964 "Permission denied");
965 }
966 else {
967 THROW_MSG_0(vmSymbols::java_io_IOException(), strerror(errno));
968 }
969 }
970 int fd = result;
971
972 // Check to see if the file is secure.
973 if (!is_file_secure(fd, filename)) {
974 ::close(fd);
975 return -1;
976 }
977
978 return fd;
979 }
980
981 // create a named shared memory region. returns the address of the
982 // memory region on success or NULL on failure. A return value of
983 // NULL will ultimately disable the shared memory feature.
984 //
985 // On AIX, Solaris and Linux, the name space for shared memory objects
986 // is the file system name space.
987 //
988 // A monitoring application attaching to a JVM does not need to know
989 // the file system name of the shared memory object. However, it may
990 // be convenient for applications to discover the existence of newly
991 // created and terminating JVMs by watching the file system name space
992 // for files being created or removed.
993 //
994 static char* mmap_create_shared(size_t size) {
995
996 int result;
997 int fd;
998 char* mapAddress;
999
1000 int vmid = os::current_process_id();
1001
1002 char* user_name = get_user_name(geteuid());
1003
1004 if (user_name == NULL)
1005 return NULL;
1006
1007 char* dirname = get_user_tmp_dir(user_name);
1008 char* filename = get_sharedmem_filename(dirname, vmid);
1009
1010 // Get the short filename.
1011 char* short_filename = strrchr(filename, '/');
1012 if (short_filename == NULL) {
1013 short_filename = filename;
1014 } else {
1015 short_filename++;
1016 }
1017
1018 // cleanup any stale shared memory files
1019 cleanup_sharedmem_resources(dirname);
1020
1021 assert(((size > 0) && (size % os::vm_page_size() == 0)),
1022 "unexpected PerfMemory region size");
1023
1024 fd = create_sharedmem_resources(dirname, short_filename, size);
1025
1026 FREE_C_HEAP_ARRAY(char, user_name);
1027 FREE_C_HEAP_ARRAY(char, dirname);
1028
1029 if (fd == -1) {
1030 FREE_C_HEAP_ARRAY(char, filename);
1031 return NULL;
1032 }
1033
1034 mapAddress = (char*)::mmap((char*)0, size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
1035
1036 // attempt to close the file - restart it if it was interrupted,
1037 // but ignore other failures
1038 RESTARTABLE(::close(fd), result);
1039 assert(result != OS_ERR, "could not close file");
1040
1041 if (mapAddress == MAP_FAILED) {
1042 if (PrintMiscellaneous && Verbose) {
1043 warning("mmap failed - %s\n", strerror(errno));
1044 }
1045 remove_file(filename);
1046 FREE_C_HEAP_ARRAY(char, filename);
1047 return NULL;
1048 }
1049
1050 // save the file name for use in delete_shared_memory()
1051 backing_store_file_name = filename;
1052
1053 // clear the shared memory region
1054 (void)::memset((void*) mapAddress, 0, size);
1055
1056 // It does not go through os api, the operation has to record from here.
1057 MemTracker::record_virtual_memory_reserve((address)mapAddress, size, CURRENT_PC, mtInternal);
1058
1059 return mapAddress;
1060 }
1061
1062 // release a named shared memory region
1063 //
1064 static void unmap_shared(char* addr, size_t bytes) {
1065 // Do not rely on os::reserve_memory/os::release_memory to use mmap.
1066 // Use os::reserve_memory/os::release_memory for PerfDisableSharedMem=1, mmap/munmap for PerfDisableSharedMem=0
1067 if (::munmap(addr, bytes) == -1) {
1068 warning("perfmemory: munmap failed (%d)\n", errno);
1069 }
1070 }
1071
1072 // create the PerfData memory region in shared memory.
1073 //
1074 static char* create_shared_memory(size_t size) {
1075
1076 // create the shared memory region.
1077 return mmap_create_shared(size);
1078 }
1079
1080 // delete the shared PerfData memory region
1081 //
1082 static void delete_shared_memory(char* addr, size_t size) {
1083
1084 // cleanup the persistent shared memory resources. since DestroyJavaVM does
1085 // not support unloading of the JVM, unmapping of the memory resource is
1086 // not performed. The memory will be reclaimed by the OS upon termination of
1087 // the process. The backing store file is deleted from the file system.
1088
1089 assert(!PerfDisableSharedMem, "shouldn't be here");
1090
1091 if (backing_store_file_name != NULL) {
1092 remove_file(backing_store_file_name);
1093 // Don't.. Free heap memory could deadlock os::abort() if it is called
1094 // from signal handler. OS will reclaim the heap memory.
1095 // FREE_C_HEAP_ARRAY(char, backing_store_file_name);
1096 backing_store_file_name = NULL;
1097 }
1098 }
1099
1100 // return the size of the file for the given file descriptor
1101 // or 0 if it is not a valid size for a shared memory file
1102 //
1103 static size_t sharedmem_filesize(int fd, TRAPS) {
1104
1105 struct stat statbuf;
1106 int result;
1107
1108 RESTARTABLE(::fstat(fd, &statbuf), result);
1109 if (result == OS_ERR) {
1110 if (PrintMiscellaneous && Verbose) {
1111 warning("fstat failed: %s\n", strerror(errno));
1112 }
1113 THROW_MSG_0(vmSymbols::java_io_IOException(),
1114 "Could not determine PerfMemory size");
1115 }
1116
1117 if ((statbuf.st_size == 0) ||
1118 ((size_t)statbuf.st_size % os::vm_page_size() != 0)) {
1119 THROW_MSG_0(vmSymbols::java_lang_Exception(),
1120 "Invalid PerfMemory size");
1121 }
1122
1123 return (size_t)statbuf.st_size;
1124 }
1125
1126 // attach to a named shared memory region.
1127 //
1128 static void mmap_attach_shared(const char* user, int vmid, PerfMemory::PerfMemoryMode mode, char** addr, size_t* sizep, TRAPS) {
1129
1130 char* mapAddress;
1131 int result;
1132 int fd;
1133 size_t size = 0;
1134 const char* luser = NULL;
1135
1136 int mmap_prot;
1137 int file_flags;
1138
1139 ResourceMark rm;
1140
1141 // map the high level access mode to the appropriate permission
1142 // constructs for the file and the shared memory mapping.
1143 if (mode == PerfMemory::PERF_MODE_RO) {
1144 mmap_prot = PROT_READ;
1145
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, CHECK);
1209 assert(fd != OS_ERR, "unexpected value");
1210
1211 if (*sizep == 0) {
1212 size = sharedmem_filesize(fd, CHECK);
1213 assert(size != 0, "unexpected size");
1214 } else {
1215 size = *sizep;
1216 }
1217
1218 mapAddress = (char*)::mmap((char*)0, size, mmap_prot, MAP_SHARED, fd, 0);
1219
1220 // attempt to close the file - restart if it gets interrupted,
1221 // but ignore other failures
1222 RESTARTABLE(::close(fd), result);
1223 assert(result != OS_ERR, "could not close file");
1224
1225 if (mapAddress == MAP_FAILED) {
1226 if (PrintMiscellaneous && Verbose) {
1227 warning("mmap failed: %s\n", strerror(errno));
1228 }
1229 THROW_MSG(vmSymbols::java_lang_OutOfMemoryError(),
1230 "Could not map PerfMemory");
1231 }
1232
1233 // It does not go through os api, the operation has to record from here.
1234 MemTracker::record_virtual_memory_reserve((address)mapAddress, size, CURRENT_PC, mtInternal);
1235
1236 *addr = mapAddress;
1237 *sizep = size;
1238
1239 if (PerfTraceMemOps) {
1240 tty->print("mapped " SIZE_FORMAT " bytes for vmid %d at "
1241 INTPTR_FORMAT "\n", size, vmid, (void*)mapAddress);
1242 }
1243 }
1244
1245
1246
1247
1248 // create the PerfData memory region
1249 //
1250 // This method creates the memory region used to store performance
1251 // data for the JVM. The memory may be created in standard or
1252 // shared memory.
1253 //
1254 void PerfMemory::create_memory_region(size_t size) {
1255
1256 if (PerfDisableSharedMem) {
1257 // do not share the memory for the performance data.
1258 _start = create_standard_memory(size);
1259 }
1260 else {
1261 _start = create_shared_memory(size);
1262 if (_start == NULL) {
1263
1264 // creation of the shared memory region failed, attempt
1265 // to create a contiguous, non-shared memory region instead.
1266 //
1267 if (PrintMiscellaneous && Verbose) {
1268 warning("Reverting to non-shared PerfMemory region.\n");
1269 }
1270 PerfDisableSharedMem = true;
1271 _start = create_standard_memory(size);
1272 }
1273 }
1274
1275 if (_start != NULL) _capacity = size;
1276
1277 }
1278
1279 // delete the PerfData memory region
1280 //
1281 // This method deletes the memory region used to store performance
1282 // data for the JVM. The memory region indicated by the <address, size>
1283 // tuple will be inaccessible after a call to this method.
1284 //
1285 void PerfMemory::delete_memory_region() {
1286
1287 assert((start() != NULL && capacity() > 0), "verify proper state");
1288
1289 // If user specifies PerfDataSaveFile, it will save the performance data
1290 // to the specified file name no matter whether PerfDataSaveToFile is specified
1291 // or not. In other word, -XX:PerfDataSaveFile=.. overrides flag
1292 // -XX:+PerfDataSaveToFile.
1293 if (PerfDataSaveToFile || PerfDataSaveFile != NULL) {
1294 save_memory_to_file(start(), capacity());
1295 }
1296
1297 if (PerfDisableSharedMem) {
1298 delete_standard_memory(start(), capacity());
1299 }
1300 else {
1301 delete_shared_memory(start(), capacity());
1302 }
1303 }
1304
1305 // attach to the PerfData memory region for another JVM
1306 //
1307 // This method returns an <address, size> tuple that points to
1308 // a memory buffer that is kept reasonably synchronized with
1309 // the PerfData memory region for the indicated JVM. This
1310 // buffer may be kept in synchronization via shared memory
1311 // or some other mechanism that keeps the buffer updated.
1312 //
1313 // If the JVM chooses not to support the attachability feature,
1314 // this method should throw an UnsupportedOperation exception.
1315 //
1316 // This implementation utilizes named shared memory to map
1317 // the indicated process's PerfData memory region into this JVMs
1318 // address space.
1319 //
1320 void PerfMemory::attach(const char* user, int vmid, PerfMemoryMode mode, char** addrp, size_t* sizep, TRAPS) {
1321
1322 if (vmid == 0 || vmid == os::current_process_id()) {
1323 *addrp = start();
1324 *sizep = capacity();
1325 return;
1326 }
1327
1328 mmap_attach_shared(user, vmid, mode, addrp, sizep, CHECK);
1329 }
1330
1331 // detach from the PerfData memory region of another JVM
1332 //
1333 // This method detaches the PerfData memory region of another
1334 // JVM, specified as an <address, size> tuple of a buffer
1335 // in this process's address space. This method may perform
1336 // arbitrary actions to accomplish the detachment. The memory
1337 // region specified by <address, size> will be inaccessible after
1338 // a call to this method.
1339 //
1340 // If the JVM chooses not to support the attachability feature,
1341 // this method should throw an UnsupportedOperation exception.
1342 //
1343 // This implementation utilizes named shared memory to detach
1344 // the indicated process's PerfData memory region from this
1345 // process's address space.
1346 //
1347 void PerfMemory::detach(char* addr, size_t bytes, TRAPS) {
1348
1349 assert(addr != 0, "address sanity check");
1350 assert(bytes > 0, "capacity sanity check");
1351
1352 if (PerfMemory::contains(addr) || PerfMemory::contains(addr + bytes - 1)) {
1353 // prevent accidental detachment of this process's PerfMemory region
1354 return;
1355 }
1356
1357 unmap_shared(addr, bytes);
1358 }