1 /*
   2  * Copyright (c) 2001, 2014, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "classfile/vmSymbols.hpp"
  27 #include "memory/allocation.inline.hpp"
  28 #include "memory/resourceArea.hpp"
  29 #include "oops/oop.inline.hpp"
  30 #include "os_bsd.inline.hpp"
  31 #include "runtime/handles.inline.hpp"
  32 #include "runtime/perfMemory.hpp"
  33 #include "services/memTracker.hpp"
  34 #include "utilities/exceptions.hpp"
  35 
  36 // put OS-includes here
  37 # include <sys/types.h>
  38 # include <sys/mman.h>
  39 # include <errno.h>
  40 # include <stdio.h>
  41 # include <unistd.h>
  42 # include <sys/stat.h>
  43 # include <signal.h>
  44 # include <pwd.h>
  45 
  46 static char* backing_store_file_name = NULL;  // name of the backing store
  47                                               // file, if successfully created.
  48 
  49 // Standard Memory Implementation Details
  50 
  51 // create the PerfData memory region in standard memory.
  52 //
  53 static char* create_standard_memory(size_t size) {
  54 
  55   // allocate an aligned chuck of memory
  56   char* mapAddress = os::reserve_memory(size);
  57 
  58   if (mapAddress == NULL) {
  59     return NULL;
  60   }
  61 
  62   // commit memory
  63   if (!os::commit_memory(mapAddress, size, !ExecMem)) {
  64     if (PrintMiscellaneous && Verbose) {
  65       warning("Could not commit PerfData memory\n");
  66     }
  67     os::release_memory(mapAddress, size);
  68     return NULL;
  69   }
  70 
  71   return mapAddress;
  72 }
  73 
  74 // delete the PerfData memory region
  75 //
  76 static void delete_standard_memory(char* addr, size_t size) {
  77 
  78   // there are no persistent external resources to cleanup for standard
  79   // memory. since DestroyJavaVM does not support unloading of the JVM,
  80   // cleanup of the memory resource is not performed. The memory will be
  81   // reclaimed by the OS upon termination of the process.
  82   //
  83   return;
  84 }
  85 
  86 // save the specified memory region to the given file
  87 //
  88 // Note: this function might be called from signal handler (by os::abort()),
  89 // don't allocate heap memory.
  90 //
  91 static void save_memory_to_file(char* addr, size_t size) {
  92 
  93  const char* destfile = PerfMemory::get_perfdata_file_path();
  94  assert(destfile[0] != '\0', "invalid PerfData file path");
  95 
  96   int result;
  97 
  98   RESTARTABLE(::open(destfile, O_CREAT|O_WRONLY|O_TRUNC, S_IREAD|S_IWRITE),
  99               result);;
 100   if (result == OS_ERR) {
 101     if (PrintMiscellaneous && Verbose) {
 102       warning("Could not create Perfdata save file: %s: %s\n",
 103               destfile, strerror(errno));
 104     }
 105   } else {
 106     int fd = result;
 107 
 108     for (size_t remaining = size; remaining > 0;) {
 109 
 110       RESTARTABLE(::write(fd, addr, remaining), result);
 111       if (result == OS_ERR) {
 112         if (PrintMiscellaneous && Verbose) {
 113           warning("Could not write Perfdata save file: %s: %s\n",
 114                   destfile, strerror(errno));
 115         }
 116         break;
 117       }
 118 
 119       remaining -= (size_t)result;
 120       addr += result;
 121     }
 122 
 123     result = ::close(fd);
 124     if (PrintMiscellaneous && Verbose) {
 125       if (result == OS_ERR) {
 126         warning("Could not close %s: %s\n", destfile, strerror(errno));
 127       }
 128     }
 129   }
 130   FREE_C_HEAP_ARRAY(char, destfile);
 131 }
 132 
 133 
 134 // Shared Memory Implementation Details
 135 
 136 // Note: the solaris and bsd shared memory implementation uses the mmap
 137 // interface with a backing store file to implement named shared memory.
 138 // Using the file system as the name space for shared memory allows a
 139 // common name space to be supported across a variety of platforms. It
 140 // also provides a name space that Java applications can deal with through
 141 // simple file apis.
 142 //
 143 // The solaris and bsd implementations store the backing store file in
 144 // a user specific temporary directory located in the /tmp file system,
 145 // which is always a local file system and is sometimes a RAM based file
 146 // system.
 147 
 148 // return the user specific temporary directory name.
 149 //
 150 // the caller is expected to free the allocated memory.
 151 //
 152 static char* get_user_tmp_dir(const char* user) {
 153 
 154   const char* tmpdir = os::get_temp_directory();
 155   const char* perfdir = PERFDATA_NAME;
 156   size_t nbytes = strlen(tmpdir) + strlen(perfdir) + strlen(user) + 3;
 157   char* dirname = NEW_C_HEAP_ARRAY(char, nbytes, mtInternal);
 158 
 159   // construct the path name to user specific tmp directory
 160   snprintf(dirname, nbytes, "%s/%s_%s", tmpdir, perfdir, user);
 161 
 162   return dirname;
 163 }
 164 
 165 // convert the given file name into a process id. if the file
 166 // does not meet the file naming constraints, return 0.
 167 //
 168 static pid_t filename_to_pid(const char* filename) {
 169 
 170   // a filename that doesn't begin with a digit is not a
 171   // candidate for conversion.
 172   //
 173   if (!isdigit(*filename)) {
 174     return 0;
 175   }
 176 
 177   // check if file name can be converted to an integer without
 178   // any leftover characters.
 179   //
 180   char* remainder = NULL;
 181   errno = 0;
 182   pid_t pid = (pid_t)strtol(filename, &remainder, 10);
 183 
 184   if (errno != 0) {
 185     return 0;
 186   }
 187 
 188   // check for left over characters. If any, then the filename is
 189   // not a candidate for conversion.
 190   //
 191   if (remainder != NULL && *remainder != '\0') {
 192     return 0;
 193   }
 194 
 195   // successful conversion, return the pid
 196   return pid;
 197 }
 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   // See if the uid of the directory matches the effective uid of the process.
 221   //
 222   if (statp->st_uid != geteuid()) {
 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 
 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 //
 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 //
 271 static bool is_same_fsobject(int fd1, int fd2) {
 272   struct stat statbuf1;
 273   struct stat statbuf2;
 274   int result = 0;
 275 
 276   RESTARTABLE(::fstat(fd1, &statbuf1), result);
 277   if (result == OS_ERR) {
 278     return false;
 279   }
 280   RESTARTABLE(::fstat(fd2, &statbuf2), result);
 281   if (result == OS_ERR) {
 282     return false;
 283   }
 284 
 285   if ((statbuf1.st_ino == statbuf2.st_ino) &&
 286       (statbuf1.st_dev == statbuf2.st_dev)) {
 287     return true;
 288   } else {
 289     return false;
 290   }
 291 }
 292 
 293 
 294 // Open the directory of the given path and validate it.
 295 // Return a DIR * of the open directory.
 296 //
 297 static DIR *open_directory_secure(const char* dirname) {
 298   // Open the directory using open() so that it can be verified
 299   // to be secure by calling is_dirfd_secure(), opendir() and then check
 300   // to see if they are the same file system object.  This method does not
 301   // introduce a window of opportunity for the directory to be attacked that
 302   // calling opendir() and is_directory_secure() does.
 303   int result;
 304   DIR *dirp = NULL;
 305   RESTARTABLE(::open(dirname, O_RDONLY|O_NOFOLLOW), result);
 306   if (result == OS_ERR) {
 307     // Directory doesn't exist or is a symlink, so there is nothing to cleanup.
 308     if (PrintMiscellaneous && Verbose) {
 309       if (errno == ELOOP) {
 310         warning("directory %s is a symlink and is not secure\n", dirname);
 311       } else {
 312         warning("could not open directory %s: %s\n", dirname, strerror(errno));
 313       }
 314     }
 315     return dirp;
 316   }
 317   int fd = result;
 318 
 319   // Determine if the open directory is secure.
 320   if (!is_dirfd_secure(fd)) {
 321     // The directory is not a secure directory.
 322     os::close(fd);
 323     return dirp;
 324   }
 325 
 326   // Open the directory.
 327   dirp = ::opendir(dirname);
 328   if (dirp == NULL) {
 329     // The directory doesn't exist, close fd and return.
 330     os::close(fd);
 331     return dirp;
 332   }
 333 
 334   // Check to make sure fd and dirp are referencing the same file system object.
 335   if (!is_same_fsobject(fd, dirfd(dirp))) {
 336     // The directory is not secure.
 337     os::close(fd);
 338     os::closedir(dirp);
 339     dirp = NULL;
 340     return dirp;
 341   }
 342 
 343   // Close initial open now that we know directory is secure
 344   os::close(fd);
 345 
 346   return dirp;
 347 }
 348 
 349 // NOTE: The code below uses fchdir(), open() and unlink() because
 350 // fdopendir(), openat() and unlinkat() are not supported on all
 351 // versions.  Once the support for fdopendir(), openat() and unlinkat()
 352 // is available on all supported versions the code can be changed
 353 // to use these functions.
 354 
 355 // Open the directory of the given path, validate it and set the
 356 // current working directory to it.
 357 // Return a DIR * of the open directory and the saved cwd fd.
 358 //
 359 static DIR *open_directory_secure_cwd(const char* dirname, int *saved_cwd_fd) {
 360 
 361   // Open the directory.
 362   DIR* dirp = open_directory_secure(dirname);
 363   if (dirp == NULL) {
 364     // Directory doesn't exist or is insecure, so there is nothing to cleanup.
 365     return dirp;
 366   }
 367   int fd = dirfd(dirp);
 368 
 369   // Open a fd to the cwd and save it off.
 370   int result;
 371   RESTARTABLE(::open(".", O_RDONLY), result);
 372   if (result == OS_ERR) {
 373     *saved_cwd_fd = -1;
 374   } else {
 375     *saved_cwd_fd = result;
 376   }
 377 
 378   // Set the current directory to dirname by using the fd of the directory.
 379   result = fchdir(fd);
 380 
 381   return dirp;
 382 }
 383 
 384 // Close the directory and restore the current working directory.
 385 //
 386 static void close_directory_secure_cwd(DIR* dirp, int saved_cwd_fd) {
 387 
 388   int result;
 389   // If we have a saved cwd change back to it and close the fd.
 390   if (saved_cwd_fd != -1) {
 391     result = fchdir(saved_cwd_fd);
 392     ::close(saved_cwd_fd);
 393   }
 394 
 395   // Close the directory.
 396   os::closedir(dirp);
 397 }
 398 
 399 // Check if the given file descriptor is considered a secure.
 400 //
 401 static bool is_file_secure(int fd, const char *filename) {
 402 
 403   int result;
 404   struct stat statbuf;
 405 
 406   // Determine if the file is secure.
 407   RESTARTABLE(::fstat(fd, &statbuf), result);
 408   if (result == OS_ERR) {
 409     if (PrintMiscellaneous && Verbose) {
 410       warning("fstat failed on %s: %s\n", filename, strerror(errno));
 411     }
 412     return false;
 413   }
 414   if (statbuf.st_nlink > 1) {
 415     // A file with multiple links is not expected.
 416     if (PrintMiscellaneous && Verbose) {
 417       warning("file %s has multiple links\n", filename);
 418     }
 419     return false;
 420   }
 421   return true;
 422 }
 423 
 424 // return the user name for the given user id
 425 //
 426 // the caller is expected to free the allocated memory.
 427 //
 428 static char* get_user_name(uid_t uid) {
 429 
 430   struct passwd pwent;
 431 
 432   // determine the max pwbuf size from sysconf, and hardcode
 433   // a default if this not available through sysconf.
 434   //
 435   long bufsize = sysconf(_SC_GETPW_R_SIZE_MAX);
 436   if (bufsize == -1)
 437     bufsize = 1024;
 438 
 439   char* pwbuf = NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 440 
 441   // POSIX interface to getpwuid_r is used on LINUX
 442   struct passwd* p;
 443   int result = getpwuid_r(uid, &pwent, pwbuf, (size_t)bufsize, &p);
 444 
 445   if (result != 0 || p == NULL || p->pw_name == NULL || *(p->pw_name) == '\0') {
 446     if (PrintMiscellaneous && Verbose) {
 447       if (result != 0) {
 448         warning("Could not retrieve passwd entry: %s\n",
 449                 strerror(result));
 450       }
 451       else if (p == NULL) {
 452         // this check is added to protect against an observed problem
 453         // with getpwuid_r() on RedHat 9 where getpwuid_r returns 0,
 454         // indicating success, but has p == NULL. This was observed when
 455         // inserting a file descriptor exhaustion fault prior to the call
 456         // getpwuid_r() call. In this case, error is set to the appropriate
 457         // error condition, but this is undocumented behavior. This check
 458         // is safe under any condition, but the use of errno in the output
 459         // message may result in an erroneous message.
 460         // Bug Id 89052 was opened with RedHat.
 461         //
 462         warning("Could not retrieve passwd entry: %s\n",
 463                 strerror(errno));
 464       }
 465       else {
 466         warning("Could not determine user name: %s\n",
 467                 p->pw_name == NULL ? "pw_name = NULL" :
 468                                      "pw_name zero length");
 469       }
 470     }
 471     FREE_C_HEAP_ARRAY(char, pwbuf);
 472     return NULL;
 473   }
 474 
 475   char* user_name = NEW_C_HEAP_ARRAY(char, strlen(p->pw_name) + 1, mtInternal);
 476   strcpy(user_name, p->pw_name);
 477 
 478   FREE_C_HEAP_ARRAY(char, pwbuf);
 479   return user_name;
 480 }
 481 
 482 // return the name of the user that owns the process identified by vmid.
 483 //
 484 // This method uses a slow directory search algorithm to find the backing
 485 // store file for the specified vmid and returns the user name, as determined
 486 // by the user name suffix of the hsperfdata_<username> directory name.
 487 //
 488 // the caller is expected to free the allocated memory.
 489 //
 490 static char* get_user_name_slow(int vmid, TRAPS) {
 491 
 492   // short circuit the directory search if the process doesn't even exist.
 493   if (kill(vmid, 0) == OS_ERR) {
 494     if (errno == ESRCH) {
 495       THROW_MSG_0(vmSymbols::java_lang_IllegalArgumentException(),
 496                   "Process not found");
 497     }
 498     else /* EPERM */ {
 499       THROW_MSG_0(vmSymbols::java_io_IOException(), strerror(errno));
 500     }
 501   }
 502 
 503   // directory search
 504   char* oldest_user = NULL;
 505   time_t oldest_ctime = 0;
 506 
 507   const char* tmpdirname = os::get_temp_directory();
 508 
 509   // open the temp directory
 510   DIR* tmpdirp = os::opendir(tmpdirname);
 511 
 512   if (tmpdirp == NULL) {
 513     // Cannot open the directory to get the user name, return.
 514     return NULL;
 515   }
 516 
 517   // for each entry in the directory that matches the pattern hsperfdata_*,
 518   // open the directory and check if the file for the given vmid exists.
 519   // The file with the expected name and the latest creation date is used
 520   // to determine the user name for the process id.
 521   //
 522   struct dirent* dentry;
 523   char* tdbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(tmpdirname), mtInternal);
 524   errno = 0;
 525   while ((dentry = os::readdir(tmpdirp, (struct dirent *)tdbuf)) != NULL) {
 526 
 527     // check if the directory entry is a hsperfdata file
 528     if (strncmp(dentry->d_name, PERFDATA_NAME, strlen(PERFDATA_NAME)) != 0) {
 529       continue;
 530     }
 531 
 532     char* usrdir_name = NEW_C_HEAP_ARRAY(char,
 533                  strlen(tmpdirname) + strlen(dentry->d_name) + 2, mtInternal);
 534     strcpy(usrdir_name, tmpdirname);
 535     strcat(usrdir_name, "/");
 536     strcat(usrdir_name, dentry->d_name);
 537 
 538     // open the user directory
 539     DIR* subdirp = open_directory_secure(usrdir_name);
 540 
 541     if (subdirp == NULL) {
 542       FREE_C_HEAP_ARRAY(char, usrdir_name);
 543       continue;
 544     }
 545 
 546     struct dirent* udentry;
 547     char* udbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(usrdir_name), mtInternal);
 548     errno = 0;
 549     while ((udentry = os::readdir(subdirp, (struct dirent *)udbuf)) != NULL) {
 550 
 551       if (filename_to_pid(udentry->d_name) == vmid) {
 552         struct stat statbuf;
 553         int result;
 554 
 555         char* filename = NEW_C_HEAP_ARRAY(char,
 556                  strlen(usrdir_name) + strlen(udentry->d_name) + 2, mtInternal);
 557 
 558         strcpy(filename, usrdir_name);
 559         strcat(filename, "/");
 560         strcat(filename, udentry->d_name);
 561 
 562         // don't follow symbolic links for the file
 563         RESTARTABLE(::lstat(filename, &statbuf), result);
 564         if (result == OS_ERR) {
 565            FREE_C_HEAP_ARRAY(char, filename);
 566            continue;
 567         }
 568 
 569         // skip over files that are not regular files.
 570         if (!S_ISREG(statbuf.st_mode)) {
 571           FREE_C_HEAP_ARRAY(char, filename);
 572           continue;
 573         }
 574 
 575         // compare and save filename with latest creation time
 576         if (statbuf.st_size > 0 && statbuf.st_ctime > oldest_ctime) {
 577 
 578           if (statbuf.st_ctime > oldest_ctime) {
 579             char* user = strchr(dentry->d_name, '_') + 1;
 580 
 581             if (oldest_user != NULL) FREE_C_HEAP_ARRAY(char, oldest_user);
 582             oldest_user = NEW_C_HEAP_ARRAY(char, strlen(user)+1, mtInternal);
 583 
 584             strcpy(oldest_user, user);
 585             oldest_ctime = statbuf.st_ctime;
 586           }
 587         }
 588 
 589         FREE_C_HEAP_ARRAY(char, filename);
 590       }
 591     }
 592     os::closedir(subdirp);
 593     FREE_C_HEAP_ARRAY(char, udbuf);
 594     FREE_C_HEAP_ARRAY(char, usrdir_name);
 595   }
 596   os::closedir(tmpdirp);
 597   FREE_C_HEAP_ARRAY(char, tdbuf);
 598 
 599   return(oldest_user);
 600 }
 601 
 602 // return the name of the user that owns the JVM indicated by the given vmid.
 603 //
 604 static char* get_user_name(int vmid, TRAPS) {
 605   return get_user_name_slow(vmid, THREAD);
 606 }
 607 
 608 // return the file name of the backing store file for the named
 609 // shared memory region for the given user name and vmid.
 610 //
 611 // the caller is expected to free the allocated memory.
 612 //
 613 static char* get_sharedmem_filename(const char* dirname, int vmid) {
 614 
 615   // add 2 for the file separator and a null terminator.
 616   size_t nbytes = strlen(dirname) + UINT_CHARS + 2;
 617 
 618   char* name = NEW_C_HEAP_ARRAY(char, nbytes, mtInternal);
 619   snprintf(name, nbytes, "%s/%d", dirname, vmid);
 620 
 621   return name;
 622 }
 623 
 624 
 625 // remove file
 626 //
 627 // this method removes the file specified by the given path
 628 //
 629 static void remove_file(const char* path) {
 630 
 631   int result;
 632 
 633   // if the file is a directory, the following unlink will fail. since
 634   // we don't expect to find directories in the user temp directory, we
 635   // won't try to handle this situation. even if accidentially or
 636   // maliciously planted, the directory's presence won't hurt anything.
 637   //
 638   RESTARTABLE(::unlink(path), result);
 639   if (PrintMiscellaneous && Verbose && result == OS_ERR) {
 640     if (errno != ENOENT) {
 641       warning("Could not unlink shared memory backing"
 642               " store file %s : %s\n", path, strerror(errno));
 643     }
 644   }
 645 }
 646 
 647 
 648 // cleanup stale shared memory resources
 649 //
 650 // This method attempts to remove all stale shared memory files in
 651 // the named user temporary directory. It scans the named directory
 652 // for files matching the pattern ^$[0-9]*$. For each file found, the
 653 // process id is extracted from the file name and a test is run to
 654 // determine if the process is alive. If the process is not alive,
 655 // any stale file resources are removed.
 656 //
 657 static void cleanup_sharedmem_resources(const char* dirname) {
 658 
 659   int saved_cwd_fd;
 660   // open the directory and set the current working directory to it
 661   DIR* dirp = open_directory_secure_cwd(dirname, &saved_cwd_fd);
 662   if (dirp == NULL) {
 663     // directory doesn't exist or is insecure, so there is nothing to cleanup
 664     return;
 665   }
 666 
 667   // for each entry in the directory that matches the expected file
 668   // name pattern, determine if the file resources are stale and if
 669   // so, remove the file resources. Note, instrumented HotSpot processes
 670   // for this user may start and/or terminate during this search and
 671   // remove or create new files in this directory. The behavior of this
 672   // loop under these conditions is dependent upon the implementation of
 673   // opendir/readdir.
 674   //
 675   struct dirent* entry;
 676   char* dbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(dirname), mtInternal);
 677 
 678   errno = 0;
 679   while ((entry = os::readdir(dirp, (struct dirent *)dbuf)) != NULL) {
 680 
 681     pid_t pid = filename_to_pid(entry->d_name);
 682 
 683     if (pid == 0) {
 684 
 685       if (strcmp(entry->d_name, ".") != 0 && strcmp(entry->d_name, "..") != 0) {
 686 
 687         // attempt to remove all unexpected files, except "." and ".."
 688         unlink(entry->d_name);
 689       }
 690 
 691       errno = 0;
 692       continue;
 693     }
 694 
 695     // we now have a file name that converts to a valid integer
 696     // that could represent a process id . if this process id
 697     // matches the current process id or the process is not running,
 698     // then remove the stale file resources.
 699     //
 700     // process liveness is detected by sending signal number 0 to
 701     // the process id (see kill(2)). if kill determines that the
 702     // process does not exist, then the file resources are removed.
 703     // if kill determines that that we don't have permission to
 704     // signal the process, then the file resources are assumed to
 705     // be stale and are removed because the resources for such a
 706     // process should be in a different user specific directory.
 707     //
 708     if ((pid == os::current_process_id()) ||
 709         (kill(pid, 0) == OS_ERR && (errno == ESRCH || errno == EPERM))) {
 710 
 711         unlink(entry->d_name);
 712     }
 713     errno = 0;
 714   }
 715 
 716   // close the directory and reset the current working directory
 717   close_directory_secure_cwd(dirp, saved_cwd_fd);
 718 
 719   FREE_C_HEAP_ARRAY(char, dbuf);
 720 }
 721 
 722 // make the user specific temporary directory. Returns true if
 723 // the directory exists and is secure upon return. Returns false
 724 // if the directory exists but is either a symlink, is otherwise
 725 // insecure, or if an error occurred.
 726 //
 727 static bool make_user_tmp_dir(const char* dirname) {
 728 
 729   // create the directory with 0755 permissions. note that the directory
 730   // will be owned by euid::egid, which may not be the same as uid::gid.
 731   //
 732   if (mkdir(dirname, S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH) == OS_ERR) {
 733     if (errno == EEXIST) {
 734       // The directory already exists and was probably created by another
 735       // JVM instance. However, this could also be the result of a
 736       // deliberate symlink. Verify that the existing directory is safe.
 737       //
 738       if (!is_directory_secure(dirname)) {
 739         // directory is not secure
 740         if (PrintMiscellaneous && Verbose) {
 741           warning("%s directory is insecure\n", dirname);
 742         }
 743         return false;
 744       }
 745     }
 746     else {
 747       // we encountered some other failure while attempting
 748       // to create the directory
 749       //
 750       if (PrintMiscellaneous && Verbose) {
 751         warning("could not create directory %s: %s\n",
 752                 dirname, strerror(errno));
 753       }
 754       return false;
 755     }
 756   }
 757   return true;
 758 }
 759 
 760 // create the shared memory file resources
 761 //
 762 // This method creates the shared memory file with the given size
 763 // This method also creates the user specific temporary directory, if
 764 // it does not yet exist.
 765 //
 766 static int create_sharedmem_resources(const char* dirname, const char* filename, size_t size) {
 767 
 768   // make the user temporary directory
 769   if (!make_user_tmp_dir(dirname)) {
 770     // could not make/find the directory or the found directory
 771     // was not secure
 772     return -1;
 773   }
 774 
 775   int saved_cwd_fd;
 776   // open the directory and set the current working directory to it
 777   DIR* dirp = open_directory_secure_cwd(dirname, &saved_cwd_fd);
 778   if (dirp == NULL) {
 779     // Directory doesn't exist or is insecure, so cannot create shared
 780     // memory file.
 781     return -1;
 782   }
 783 
 784   // Open the filename in the current directory.
 785   // Cannot use O_TRUNC here; truncation of an existing file has to happen
 786   // after the is_file_secure() check below.
 787   int result;
 788   RESTARTABLE(::open(filename, O_RDWR|O_CREAT|O_NOFOLLOW, S_IREAD|S_IWRITE), result);
 789   if (result == OS_ERR) {
 790     if (PrintMiscellaneous && Verbose) {
 791       if (errno == ELOOP) {
 792         warning("file %s is a symlink and is not secure\n", filename);
 793       } else {
 794         warning("could not create file %s: %s\n", filename, strerror(errno));
 795       }
 796     }
 797     // close the directory and reset the current working directory
 798     close_directory_secure_cwd(dirp, saved_cwd_fd);
 799 
 800     return -1;
 801   }
 802   // close the directory and reset the current working directory
 803   close_directory_secure_cwd(dirp, saved_cwd_fd);
 804 
 805   // save the file descriptor
 806   int fd = result;
 807 
 808   // check to see if the file is secure
 809   if (!is_file_secure(fd, filename)) {
 810     ::close(fd);
 811     return -1;
 812   }
 813 
 814   // truncate the file to get rid of any existing data
 815   RESTARTABLE(::ftruncate(fd, (off_t)0), result);
 816   if (result == OS_ERR) {
 817     if (PrintMiscellaneous && Verbose) {
 818       warning("could not truncate shared memory file: %s\n", strerror(errno));
 819     }
 820     ::close(fd);
 821     return -1;
 822   }
 823   // set the file size
 824   RESTARTABLE(::ftruncate(fd, (off_t)size), result);
 825   if (result == OS_ERR) {
 826     if (PrintMiscellaneous && Verbose) {
 827       warning("could not set shared memory file size: %s\n", strerror(errno));
 828     }
 829     ::close(fd);
 830     return -1;
 831   }
 832 
 833   // Verify that we have enough disk space for this file.
 834   // We'll get random SIGBUS crashes on memory accesses if
 835   // we don't.
 836 
 837   for (size_t seekpos = 0; seekpos < size; seekpos += os::vm_page_size()) {
 838     int zero_int = 0;
 839     result = (int)os::seek_to_file_offset(fd, (jlong)(seekpos));
 840     if (result == -1 ) break;
 841     RESTARTABLE(::write(fd, &zero_int, 1), result);
 842     if (result != 1) {
 843       if (errno == ENOSPC) {
 844         warning("Insufficient space for shared memory file:\n   %s\nTry using the -Djava.io.tmpdir= option to select an alternate temp location.\n", filename);
 845       }
 846       break;
 847     }
 848   }
 849 
 850   if (result != -1) {
 851     return fd;
 852   } else {
 853     ::close(fd);
 854     return -1;
 855   }
 856 }
 857 
 858 // open the shared memory file for the given user and vmid. returns
 859 // the file descriptor for the open file or -1 if the file could not
 860 // be opened.
 861 //
 862 static int open_sharedmem_file(const char* filename, int oflags, TRAPS) {
 863 
 864   // open the file
 865   int result;
 866   RESTARTABLE(::open(filename, oflags), result);
 867   if (result == OS_ERR) {
 868     if (errno == ENOENT) {
 869       THROW_MSG_(vmSymbols::java_lang_IllegalArgumentException(),
 870                   "Process not found", OS_ERR);
 871     }
 872     else if (errno == EACCES) {
 873       THROW_MSG_(vmSymbols::java_lang_IllegalArgumentException(),
 874                   "Permission denied", OS_ERR);
 875     }
 876     else {
 877       THROW_MSG_(vmSymbols::java_io_IOException(), strerror(errno), OS_ERR);
 878     }
 879   }
 880   int fd = result;
 881 
 882   // check to see if the file is secure
 883   if (!is_file_secure(fd, filename)) {
 884     ::close(fd);
 885     return -1;
 886   }
 887 
 888   return fd;
 889 }
 890 
 891 // create a named shared memory region. returns the address of the
 892 // memory region on success or NULL on failure. A return value of
 893 // NULL will ultimately disable the shared memory feature.
 894 //
 895 // On Solaris and Bsd, the name space for shared memory objects
 896 // is the file system name space.
 897 //
 898 // A monitoring application attaching to a JVM does not need to know
 899 // the file system name of the shared memory object. However, it may
 900 // be convenient for applications to discover the existence of newly
 901 // created and terminating JVMs by watching the file system name space
 902 // for files being created or removed.
 903 //
 904 static char* mmap_create_shared(size_t size) {
 905 
 906   int result;
 907   int fd;
 908   char* mapAddress;
 909 
 910   int vmid = os::current_process_id();
 911 
 912   char* user_name = get_user_name(geteuid());
 913 
 914   if (user_name == NULL)
 915     return NULL;
 916 
 917   char* dirname = get_user_tmp_dir(user_name);
 918   char* filename = get_sharedmem_filename(dirname, vmid);
 919 
 920   // get the short filename
 921   char* short_filename = strrchr(filename, '/');
 922   if (short_filename == NULL) {
 923     short_filename = filename;
 924   } else {
 925     short_filename++;
 926   }
 927 
 928   // cleanup any stale shared memory files
 929   cleanup_sharedmem_resources(dirname);
 930 
 931   assert(((size > 0) && (size % os::vm_page_size() == 0)),
 932          "unexpected PerfMemory region size");
 933 
 934   fd = create_sharedmem_resources(dirname, short_filename, size);
 935 
 936   FREE_C_HEAP_ARRAY(char, user_name);
 937   FREE_C_HEAP_ARRAY(char, dirname);
 938 
 939   if (fd == -1) {
 940     FREE_C_HEAP_ARRAY(char, filename);
 941     return NULL;
 942   }
 943 
 944   mapAddress = (char*)::mmap((char*)0, size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 945 
 946   result = ::close(fd);
 947   assert(result != OS_ERR, "could not close file");
 948 
 949   if (mapAddress == MAP_FAILED) {
 950     if (PrintMiscellaneous && Verbose) {
 951       warning("mmap failed -  %s\n", strerror(errno));
 952     }
 953     remove_file(filename);
 954     FREE_C_HEAP_ARRAY(char, filename);
 955     return NULL;
 956   }
 957 
 958   // save the file name for use in delete_shared_memory()
 959   backing_store_file_name = filename;
 960 
 961   // clear the shared memory region
 962   (void)::memset((void*) mapAddress, 0, size);
 963 
 964   // it does not go through os api, the operation has to record from here
 965   MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, size, CURRENT_PC, mtInternal);
 966 
 967   return mapAddress;
 968 }
 969 
 970 // release a named shared memory region
 971 //
 972 static void unmap_shared(char* addr, size_t bytes) {
 973   os::release_memory(addr, bytes);
 974 }
 975 
 976 // create the PerfData memory region in shared memory.
 977 //
 978 static char* create_shared_memory(size_t size) {
 979 
 980   // create the shared memory region.
 981   return mmap_create_shared(size);
 982 }
 983 
 984 // delete the shared PerfData memory region
 985 //
 986 static void delete_shared_memory(char* addr, size_t size) {
 987 
 988   // cleanup the persistent shared memory resources. since DestroyJavaVM does
 989   // not support unloading of the JVM, unmapping of the memory resource is
 990   // not performed. The memory will be reclaimed by the OS upon termination of
 991   // the process. The backing store file is deleted from the file system.
 992 
 993   assert(!PerfDisableSharedMem, "shouldn't be here");
 994 
 995   if (backing_store_file_name != NULL) {
 996     remove_file(backing_store_file_name);
 997     // Don't.. Free heap memory could deadlock os::abort() if it is called
 998     // from signal handler. OS will reclaim the heap memory.
 999     // FREE_C_HEAP_ARRAY(char, backing_store_file_name);
1000     backing_store_file_name = NULL;
1001   }
1002 }
1003 
1004 // return the size of the file for the given file descriptor
1005 // or 0 if it is not a valid size for a shared memory file
1006 //
1007 static size_t sharedmem_filesize(int fd, TRAPS) {
1008 
1009   struct stat statbuf;
1010   int result;
1011 
1012   RESTARTABLE(::fstat(fd, &statbuf), result);
1013   if (result == OS_ERR) {
1014     if (PrintMiscellaneous && Verbose) {
1015       warning("fstat failed: %s\n", strerror(errno));
1016     }
1017     THROW_MSG_0(vmSymbols::java_io_IOException(),
1018                 "Could not determine PerfMemory size");
1019   }
1020 
1021   if ((statbuf.st_size == 0) ||
1022      ((size_t)statbuf.st_size % os::vm_page_size() != 0)) {
1023     THROW_MSG_0(vmSymbols::java_lang_Exception(),
1024                 "Invalid PerfMemory size");
1025   }
1026 
1027   return (size_t)statbuf.st_size;
1028 }
1029 
1030 // attach to a named shared memory region.
1031 //
1032 static void mmap_attach_shared(const char* user, int vmid, PerfMemory::PerfMemoryMode mode, char** addr, size_t* sizep, TRAPS) {
1033 
1034   char* mapAddress;
1035   int result;
1036   int fd;
1037   size_t size = 0;
1038   const char* luser = NULL;
1039 
1040   int mmap_prot;
1041   int file_flags;
1042 
1043   ResourceMark rm;
1044 
1045   // map the high level access mode to the appropriate permission
1046   // constructs for the file and the shared memory mapping.
1047   if (mode == PerfMemory::PERF_MODE_RO) {
1048     mmap_prot = PROT_READ;
1049     file_flags = O_RDONLY | O_NOFOLLOW;
1050   }
1051   else if (mode == PerfMemory::PERF_MODE_RW) {
1052 #ifdef LATER
1053     mmap_prot = PROT_READ | PROT_WRITE;
1054     file_flags = O_RDWR | O_NOFOLLOW;
1055 #else
1056     THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1057               "Unsupported access mode");
1058 #endif
1059   }
1060   else {
1061     THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1062               "Illegal access mode");
1063   }
1064 
1065   if (user == NULL || strlen(user) == 0) {
1066     luser = get_user_name(vmid, CHECK);
1067   }
1068   else {
1069     luser = user;
1070   }
1071 
1072   if (luser == NULL) {
1073     THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1074               "Could not map vmid to user Name");
1075   }
1076 
1077   char* dirname = get_user_tmp_dir(luser);
1078 
1079   // since we don't follow symbolic links when creating the backing
1080   // store file, we don't follow them when attaching either.
1081   //
1082   if (!is_directory_secure(dirname)) {
1083     FREE_C_HEAP_ARRAY(char, dirname);
1084     if (luser != user) {
1085       FREE_C_HEAP_ARRAY(char, luser);
1086     }
1087     THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1088               "Process not found");
1089   }
1090 
1091   char* filename = get_sharedmem_filename(dirname, vmid);
1092 
1093   // copy heap memory to resource memory. the open_sharedmem_file
1094   // method below need to use the filename, but could throw an
1095   // exception. using a resource array prevents the leak that
1096   // would otherwise occur.
1097   char* rfilename = NEW_RESOURCE_ARRAY(char, strlen(filename) + 1);
1098   strcpy(rfilename, filename);
1099 
1100   // free the c heap resources that are no longer needed
1101   if (luser != user) FREE_C_HEAP_ARRAY(char, luser);
1102   FREE_C_HEAP_ARRAY(char, dirname);
1103   FREE_C_HEAP_ARRAY(char, filename);
1104 
1105   // open the shared memory file for the give vmid
1106   fd = open_sharedmem_file(rfilename, file_flags, CHECK);
1107   assert(fd != OS_ERR, "unexpected value");
1108 
1109   if (*sizep == 0) {
1110     size = sharedmem_filesize(fd, CHECK);
1111   } else {
1112     size = *sizep;
1113   }
1114 
1115   assert(size > 0, "unexpected size <= 0");
1116 
1117   mapAddress = (char*)::mmap((char*)0, size, mmap_prot, MAP_SHARED, fd, 0);
1118 
1119   // attempt to close the file - restart if it gets interrupted,
1120   // but ignore other failures
1121   result = ::close(fd);
1122   assert(result != OS_ERR, "could not close file");
1123 
1124   if (mapAddress == MAP_FAILED) {
1125     if (PrintMiscellaneous && Verbose) {
1126       warning("mmap failed: %s\n", strerror(errno));
1127     }
1128     THROW_MSG(vmSymbols::java_lang_OutOfMemoryError(),
1129               "Could not map PerfMemory");
1130   }
1131 
1132   // it does not go through os api, the operation has to record from here
1133   MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, size, CURRENT_PC, mtInternal);
1134 
1135   *addr = mapAddress;
1136   *sizep = size;
1137 
1138   if (PerfTraceMemOps) {
1139     tty->print("mapped " SIZE_FORMAT " bytes for vmid %d at "
1140                INTPTR_FORMAT "\n", size, vmid, p2i((void*)mapAddress));
1141   }
1142 }
1143 
1144 
1145 
1146 
1147 // create the PerfData memory region
1148 //
1149 // This method creates the memory region used to store performance
1150 // data for the JVM. The memory may be created in standard or
1151 // shared memory.
1152 //
1153 void PerfMemory::create_memory_region(size_t size) {
1154 
1155   if (PerfDisableSharedMem) {
1156     // do not share the memory for the performance data.
1157     _start = create_standard_memory(size);
1158   }
1159   else {
1160     _start = create_shared_memory(size);
1161     if (_start == NULL) {
1162 
1163       // creation of the shared memory region failed, attempt
1164       // to create a contiguous, non-shared memory region instead.
1165       //
1166       if (PrintMiscellaneous && Verbose) {
1167         warning("Reverting to non-shared PerfMemory region.\n");
1168       }
1169       PerfDisableSharedMem = true;
1170       _start = create_standard_memory(size);
1171     }
1172   }
1173 
1174   if (_start != NULL) _capacity = size;
1175 
1176 }
1177 
1178 // delete the PerfData memory region
1179 //
1180 // This method deletes the memory region used to store performance
1181 // data for the JVM. The memory region indicated by the <address, size>
1182 // tuple will be inaccessible after a call to this method.
1183 //
1184 void PerfMemory::delete_memory_region() {
1185 
1186   assert((start() != NULL && capacity() > 0), "verify proper state");
1187 
1188   // If user specifies PerfDataSaveFile, it will save the performance data
1189   // to the specified file name no matter whether PerfDataSaveToFile is specified
1190   // or not. In other word, -XX:PerfDataSaveFile=.. overrides flag
1191   // -XX:+PerfDataSaveToFile.
1192   if (PerfDataSaveToFile || PerfDataSaveFile != NULL) {
1193     save_memory_to_file(start(), capacity());
1194   }
1195 
1196   if (PerfDisableSharedMem) {
1197     delete_standard_memory(start(), capacity());
1198   }
1199   else {
1200     delete_shared_memory(start(), capacity());
1201   }
1202 }
1203 
1204 // attach to the PerfData memory region for another JVM
1205 //
1206 // This method returns an <address, size> tuple that points to
1207 // a memory buffer that is kept reasonably synchronized with
1208 // the PerfData memory region for the indicated JVM. This
1209 // buffer may be kept in synchronization via shared memory
1210 // or some other mechanism that keeps the buffer updated.
1211 //
1212 // If the JVM chooses not to support the attachability feature,
1213 // this method should throw an UnsupportedOperation exception.
1214 //
1215 // This implementation utilizes named shared memory to map
1216 // the indicated process's PerfData memory region into this JVMs
1217 // address space.
1218 //
1219 void PerfMemory::attach(const char* user, int vmid, PerfMemoryMode mode, char** addrp, size_t* sizep, TRAPS) {
1220 
1221   if (vmid == 0 || vmid == os::current_process_id()) {
1222      *addrp = start();
1223      *sizep = capacity();
1224      return;
1225   }
1226 
1227   mmap_attach_shared(user, vmid, mode, addrp, sizep, CHECK);
1228 }
1229 
1230 // detach from the PerfData memory region of another JVM
1231 //
1232 // This method detaches the PerfData memory region of another
1233 // JVM, specified as an <address, size> tuple of a buffer
1234 // in this process's address space. This method may perform
1235 // arbitrary actions to accomplish the detachment. The memory
1236 // region specified by <address, size> will be inaccessible after
1237 // a call to this method.
1238 //
1239 // If the JVM chooses not to support the attachability feature,
1240 // this method should throw an UnsupportedOperation exception.
1241 //
1242 // This implementation utilizes named shared memory to detach
1243 // the indicated process's PerfData memory region from this
1244 // process's address space.
1245 //
1246 void PerfMemory::detach(char* addr, size_t bytes, TRAPS) {
1247 
1248   assert(addr != 0, "address sanity check");
1249   assert(bytes > 0, "capacity sanity check");
1250 
1251   if (PerfMemory::contains(addr) || PerfMemory::contains(addr + bytes - 1)) {
1252     // prevent accidental detachment of this process's PerfMemory region
1253     return;
1254   }
1255 
1256   unmap_shared(addr, bytes);
1257 }