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_solaris.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 # include <procfs.h>
  46 
  47 
  48 static char* backing_store_file_name = NULL;  // name of the backing store
  49                                               // file, if successfully created.
  50 
  51 // Standard Memory Implementation Details
  52 
  53 // create the PerfData memory region in standard memory.
  54 //
  55 static char* create_standard_memory(size_t size) {
  56 
  57   // allocate an aligned chuck of memory
  58   char* mapAddress = os::reserve_memory(size);
  59 
  60   if (mapAddress == NULL) {
  61     return NULL;
  62   }
  63 
  64   // commit memory
  65   if (!os::commit_memory(mapAddress, size, !ExecMem)) {
  66     if (PrintMiscellaneous && Verbose) {
  67       warning("Could not commit PerfData memory\n");
  68     }
  69     os::release_memory(mapAddress, size);
  70     return NULL;
  71   }
  72 
  73   return mapAddress;
  74 }
  75 
  76 // delete the PerfData memory region
  77 //
  78 static void delete_standard_memory(char* addr, size_t size) {
  79 
  80   // there are no persistent external resources to cleanup for standard
  81   // memory. since DestroyJavaVM does not support unloading of the JVM,
  82   // cleanup of the memory resource is not performed. The memory will be
  83   // reclaimed by the OS upon termination of the process.
  84   //
  85   return;
  86 }
  87 
  88 // save the specified memory region to the given file
  89 //
  90 // Note: this function might be called from signal handler (by os::abort()),
  91 // don't allocate heap memory.
  92 //
  93 static void save_memory_to_file(char* addr, size_t size) {
  94 
  95   const char* destfile = PerfMemory::get_perfdata_file_path();
  96   assert(destfile[0] != '\0', "invalid PerfData file path");
  97 
  98   int result;
  99 
 100   RESTARTABLE(::open(destfile, O_CREAT|O_WRONLY|O_TRUNC, S_IREAD|S_IWRITE),
 101               result);;
 102   if (result == OS_ERR) {
 103     if (PrintMiscellaneous && Verbose) {
 104       warning("Could not create Perfdata save file: %s: %s\n",
 105               destfile, strerror(errno));
 106     }
 107   } else {
 108 
 109     int fd = result;
 110 
 111     for (size_t remaining = size; remaining > 0;) {
 112 
 113       RESTARTABLE(::write(fd, addr, remaining), result);
 114       if (result == OS_ERR) {
 115         if (PrintMiscellaneous && Verbose) {
 116           warning("Could not write Perfdata save file: %s: %s\n",
 117                   destfile, strerror(errno));
 118         }
 119         break;
 120       }
 121       remaining -= (size_t)result;
 122       addr += result;
 123     }
 124 
 125     result = ::close(fd);
 126     if (PrintMiscellaneous && Verbose) {
 127       if (result == OS_ERR) {
 128         warning("Could not close %s: %s\n", destfile, strerror(errno));
 129       }
 130     }
 131   }
 132   FREE_C_HEAP_ARRAY(char, destfile);
 133 }
 134 
 135 
 136 // Shared Memory Implementation Details
 137 
 138 // Note: the solaris and linux shared memory implementation uses the mmap
 139 // interface with a backing store file to implement named shared memory.
 140 // Using the file system as the name space for shared memory allows a
 141 // common name space to be supported across a variety of platforms. It
 142 // also provides a name space that Java applications can deal with through
 143 // simple file apis.
 144 //
 145 // The solaris and linux implementations store the backing store file in
 146 // a user specific temporary directory located in the /tmp file system,
 147 // which is always a local file system and is sometimes a RAM based file
 148 // system.
 149 
 150 // return the user specific temporary directory name.
 151 //
 152 // the caller is expected to free the allocated memory.
 153 //
 154 static char* get_user_tmp_dir(const char* user) {
 155 
 156   const char* tmpdir = os::get_temp_directory();
 157   const char* perfdir = PERFDATA_NAME;
 158   size_t nbytes = strlen(tmpdir) + strlen(perfdir) + strlen(user) + 3;
 159   char* dirname = NEW_C_HEAP_ARRAY(char, nbytes, mtInternal);
 160 
 161   // construct the path name to user specific tmp directory
 162   snprintf(dirname, nbytes, "%s/%s_%s", tmpdir, perfdir, user);
 163 
 164   return dirname;
 165 }
 166 
 167 // convert the given file name into a process id. if the file
 168 // does not meet the file naming constraints, return 0.
 169 //
 170 static pid_t filename_to_pid(const char* filename) {
 171 
 172   // a filename that doesn't begin with a digit is not a
 173   // candidate for conversion.
 174   //
 175   if (!isdigit(*filename)) {
 176     return 0;
 177   }
 178 
 179   // check if file name can be converted to an integer without
 180   // any leftover characters.
 181   //
 182   char* remainder = NULL;
 183   errno = 0;
 184   pid_t pid = (pid_t)strtol(filename, &remainder, 10);
 185 
 186   if (errno != 0) {
 187     return 0;
 188   }
 189 
 190   // check for left over characters. If any, then the filename is
 191   // not a candidate for conversion.
 192   //
 193   if (remainder != NULL && *remainder != '\0') {
 194     return 0;
 195   }
 196 
 197   // successful conversion, return the pid
 198   return pid;
 199 }
 200 
 201 
 202 // Check if the given statbuf is considered a secure directory for
 203 // the backing store files. Returns true if the directory is considered
 204 // a secure location. Returns false if the statbuf is a symbolic link or
 205 // if an error occurred.
 206 //
 207 static bool is_statbuf_secure(struct stat *statp) {
 208   if (S_ISLNK(statp->st_mode) || !S_ISDIR(statp->st_mode)) {
 209     // The path represents a link or some non-directory file type,
 210     // which is not what we expected. Declare it insecure.
 211     //
 212     return false;
 213   }
 214   // We have an existing directory, check if the permissions are safe.
 215   //
 216   if ((statp->st_mode & (S_IWGRP|S_IWOTH)) != 0) {
 217     // The directory is open for writing and could be subjected
 218     // to a symlink or a hard link attack. Declare it insecure.
 219     //
 220     return false;
 221   }
 222   // See if the uid of the directory matches the effective uid of the process.
 223   //
 224   if (statp->st_uid != geteuid()) {
 225     // The directory was not created by this user, declare it insecure.
 226     //
 227     return false;
 228   }
 229   return true;
 230 }
 231 
 232 
 233 // Check if the given path is considered a secure directory for
 234 // the backing store files. Returns true if the directory exists
 235 // and is considered a secure location. Returns false if the path
 236 // is a symbolic link or if an error occurred.
 237 //
 238 static bool is_directory_secure(const char* path) {
 239   struct stat statbuf;
 240   int result = 0;
 241 
 242   RESTARTABLE(::lstat(path, &statbuf), result);
 243   if (result == OS_ERR) {
 244     return false;
 245   }
 246 
 247   // The path exists, see if it is secure.
 248   return is_statbuf_secure(&statbuf);
 249 }
 250 
 251 
 252 // Check if the given directory file descriptor is considered a secure
 253 // directory for the backing store files. Returns true if the directory
 254 // exists and is considered a secure location. Returns false if the path
 255 // is a symbolic link or if an error occurred.
 256 //
 257 static bool is_dirfd_secure(int dir_fd) {
 258   struct stat statbuf;
 259   int result = 0;
 260 
 261   RESTARTABLE(::fstat(dir_fd, &statbuf), result);
 262   if (result == OS_ERR) {
 263     return false;
 264   }
 265 
 266   // The path exists, now check its mode.
 267   return is_statbuf_secure(&statbuf);
 268 }
 269 
 270 
 271 // Check to make sure fd1 and fd2 are referencing the same file system object.
 272 //
 273 static bool is_same_fsobject(int fd1, int fd2) {
 274   struct stat statbuf1;
 275   struct stat statbuf2;
 276   int result = 0;
 277 
 278   RESTARTABLE(::fstat(fd1, &statbuf1), result);
 279   if (result == OS_ERR) {
 280     return false;
 281   }
 282   RESTARTABLE(::fstat(fd2, &statbuf2), result);
 283   if (result == OS_ERR) {
 284     return false;
 285   }
 286 
 287   if ((statbuf1.st_ino == statbuf2.st_ino) &&
 288       (statbuf1.st_dev == statbuf2.st_dev)) {
 289     return true;
 290   } else {
 291     return false;
 292   }
 293 }
 294 
 295 
 296 // Open the directory of the given path and validate it.
 297 // Return a DIR * of the open directory.
 298 //
 299 static DIR *open_directory_secure(const char* dirname) {
 300   // Open the directory using open() so that it can be verified
 301   // to be secure by calling is_dirfd_secure(), opendir() and then check
 302   // to see if they are the same file system object.  This method does not
 303   // introduce a window of opportunity for the directory to be attacked that
 304   // calling opendir() and is_directory_secure() does.
 305   int result;
 306   DIR *dirp = NULL;
 307   RESTARTABLE(::open(dirname, O_RDONLY|O_NOFOLLOW), result);
 308   if (result == OS_ERR) {
 309     // Directory doesn't exist or is a symlink, so there is nothing to cleanup.
 310     if (PrintMiscellaneous && Verbose) {
 311       if (errno == ELOOP) {
 312         warning("directory %s is a symlink and is not secure\n", dirname);
 313       } else {
 314         warning("could not open directory %s: %s\n", dirname, strerror(errno));
 315       }
 316     }
 317     return dirp;
 318   }
 319   int fd = result;
 320 
 321   // Determine if the open directory is secure.
 322   if (!is_dirfd_secure(fd)) {
 323     // The directory is not a secure directory.
 324     os::close(fd);
 325     return dirp;
 326   }
 327 
 328   // Open the directory.
 329   dirp = ::opendir(dirname);
 330   if (dirp == NULL) {
 331     // The directory doesn't exist, close fd and return.
 332     os::close(fd);
 333     return dirp;
 334   }
 335 
 336   // Check to make sure fd and dirp are referencing the same file system object.
 337   if (!is_same_fsobject(fd, dirp->dd_fd)) {
 338     // The directory is not secure.
 339     os::close(fd);
 340     os::closedir(dirp);
 341     dirp = NULL;
 342     return dirp;
 343   }
 344 
 345   // Close initial open now that we know directory is secure
 346   os::close(fd);
 347 
 348   return dirp;
 349 }
 350 
 351 // NOTE: The code below uses fchdir(), open() and unlink() because
 352 // fdopendir(), openat() and unlinkat() are not supported on all
 353 // versions.  Once the support for fdopendir(), openat() and unlinkat()
 354 // is available on all supported versions the code can be changed
 355 // to use these functions.
 356 
 357 // Open the directory of the given path, validate it and set the
 358 // current working directory to it.
 359 // Return a DIR * of the open directory and the saved cwd fd.
 360 //
 361 static DIR *open_directory_secure_cwd(const char* dirname, int *saved_cwd_fd) {
 362 
 363   // Open the directory.
 364   DIR* dirp = open_directory_secure(dirname);
 365   if (dirp == NULL) {
 366     // Directory doesn't exist or is insecure, so there is nothing to cleanup.
 367     return dirp;
 368   }
 369   int fd = dirp->dd_fd;
 370 
 371   // Open a fd to the cwd and save it off.
 372   int result;
 373   RESTARTABLE(::open(".", O_RDONLY), result);
 374   if (result == OS_ERR) {
 375     *saved_cwd_fd = -1;
 376   } else {
 377     *saved_cwd_fd = result;
 378   }
 379 
 380   // Set the current directory to dirname by using the fd of the directory.
 381   result = fchdir(fd);
 382 
 383   return dirp;
 384 }
 385 
 386 // Close the directory and restore the current working directory.
 387 //
 388 static void close_directory_secure_cwd(DIR* dirp, int saved_cwd_fd) {
 389 
 390   int result;
 391   // If we have a saved cwd change back to it and close the fd.
 392   if (saved_cwd_fd != -1) {
 393     result = fchdir(saved_cwd_fd);
 394     ::close(saved_cwd_fd);
 395   }
 396 
 397   // Close the directory.
 398   os::closedir(dirp);
 399 }
 400 
 401 // Check if the given file descriptor is considered a secure.
 402 //
 403 static bool is_file_secure(int fd, const char *filename) {
 404 
 405   int result;
 406   struct stat statbuf;
 407 
 408   // Determine if the file is secure.
 409   RESTARTABLE(::fstat(fd, &statbuf), result);
 410   if (result == OS_ERR) {
 411     if (PrintMiscellaneous && Verbose) {
 412       warning("fstat failed on %s: %s\n", filename, strerror(errno));
 413     }
 414     return false;
 415   }
 416   if (statbuf.st_nlink > 1) {
 417     // A file with multiple links is not expected.
 418     if (PrintMiscellaneous && Verbose) {
 419       warning("file %s has multiple links\n", filename);
 420     }
 421     return false;
 422   }
 423   return true;
 424 }
 425 
 426 // return the user name for the given user id
 427 //
 428 // the caller is expected to free the allocated memory.
 429 //
 430 static char* get_user_name(uid_t uid) {
 431 
 432   struct passwd pwent;
 433 
 434   // determine the max pwbuf size from sysconf, and hardcode
 435   // a default if this not available through sysconf.
 436   //
 437   long bufsize = sysconf(_SC_GETPW_R_SIZE_MAX);
 438   if (bufsize == -1)
 439     bufsize = 1024;
 440 
 441   char* pwbuf = NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 442 
 443 #ifdef _GNU_SOURCE
 444   struct passwd* p = NULL;
 445   int result = getpwuid_r(uid, &pwent, pwbuf, (size_t)bufsize, &p);
 446 #else  // _GNU_SOURCE
 447   struct passwd* p = getpwuid_r(uid, &pwent, pwbuf, (int)bufsize);
 448 #endif // _GNU_SOURCE
 449 
 450   if (p == NULL || p->pw_name == NULL || *(p->pw_name) == '\0') {
 451     if (PrintMiscellaneous && Verbose) {
 452       if (p == NULL) {
 453         warning("Could not retrieve passwd entry: %s\n",
 454                 strerror(errno));
 455       }
 456       else {
 457         warning("Could not determine user name: %s\n",
 458                 p->pw_name == NULL ? "pw_name = NULL" :
 459                                      "pw_name zero length");
 460       }
 461     }
 462     FREE_C_HEAP_ARRAY(char, pwbuf);
 463     return NULL;
 464   }
 465 
 466   char* user_name = NEW_C_HEAP_ARRAY(char, strlen(p->pw_name) + 1, mtInternal);
 467   strcpy(user_name, p->pw_name);
 468 
 469   FREE_C_HEAP_ARRAY(char, pwbuf);
 470   return user_name;
 471 }
 472 
 473 // return the name of the user that owns the process identified by vmid.
 474 //
 475 // This method uses a slow directory search algorithm to find the backing
 476 // store file for the specified vmid and returns the user name, as determined
 477 // by the user name suffix of the hsperfdata_<username> directory name.
 478 //
 479 // the caller is expected to free the allocated memory.
 480 //
 481 static char* get_user_name_slow(int vmid, TRAPS) {
 482 
 483   // short circuit the directory search if the process doesn't even exist.
 484   if (kill(vmid, 0) == OS_ERR) {
 485     if (errno == ESRCH) {
 486       THROW_MSG_0(vmSymbols::java_lang_IllegalArgumentException(),
 487                   "Process not found");
 488     }
 489     else /* EPERM */ {
 490       THROW_MSG_0(vmSymbols::java_io_IOException(), strerror(errno));
 491     }
 492   }
 493 
 494   // directory search
 495   char* oldest_user = NULL;
 496   time_t oldest_ctime = 0;
 497 
 498   const char* tmpdirname = os::get_temp_directory();
 499 
 500   // open the temp directory
 501   DIR* tmpdirp = os::opendir(tmpdirname);
 502 
 503   if (tmpdirp == NULL) {
 504     // Cannot open the directory to get the user name, return.
 505     return NULL;
 506   }
 507 
 508   // for each entry in the directory that matches the pattern hsperfdata_*,
 509   // open the directory and check if the file for the given vmid exists.
 510   // The file with the expected name and the latest creation date is used
 511   // to determine the user name for the process id.
 512   //
 513   struct dirent* dentry;
 514   char* tdbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(tmpdirname), mtInternal);
 515   errno = 0;
 516   while ((dentry = os::readdir(tmpdirp, (struct dirent *)tdbuf)) != NULL) {
 517 
 518     // check if the directory entry is a hsperfdata file
 519     if (strncmp(dentry->d_name, PERFDATA_NAME, strlen(PERFDATA_NAME)) != 0) {
 520       continue;
 521     }
 522 
 523     char* usrdir_name = NEW_C_HEAP_ARRAY(char,
 524                   strlen(tmpdirname) + strlen(dentry->d_name) + 2, mtInternal);
 525     strcpy(usrdir_name, tmpdirname);
 526     strcat(usrdir_name, "/");
 527     strcat(usrdir_name, dentry->d_name);
 528 
 529     // open the user directory
 530     DIR* subdirp = open_directory_secure(usrdir_name);
 531 
 532     if (subdirp == NULL) {
 533       FREE_C_HEAP_ARRAY(char, usrdir_name);
 534       continue;
 535     }
 536 
 537     // Since we don't create the backing store files in directories
 538     // pointed to by symbolic links, we also don't follow them when
 539     // looking for the files. We check for a symbolic link after the
 540     // call to opendir in order to eliminate a small window where the
 541     // symlink can be exploited.
 542     //
 543     if (!is_directory_secure(usrdir_name)) {
 544       FREE_C_HEAP_ARRAY(char, usrdir_name);
 545       os::closedir(subdirp);
 546       continue;
 547     }
 548 
 549     struct dirent* udentry;
 550     char* udbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(usrdir_name), mtInternal);
 551     errno = 0;
 552     while ((udentry = os::readdir(subdirp, (struct dirent *)udbuf)) != NULL) {
 553 
 554       if (filename_to_pid(udentry->d_name) == vmid) {
 555         struct stat statbuf;
 556         int result;
 557 
 558         char* filename = NEW_C_HEAP_ARRAY(char,
 559                  strlen(usrdir_name) + strlen(udentry->d_name) + 2, mtInternal);
 560 
 561         strcpy(filename, usrdir_name);
 562         strcat(filename, "/");
 563         strcat(filename, udentry->d_name);
 564 
 565         // don't follow symbolic links for the file
 566         RESTARTABLE(::lstat(filename, &statbuf), result);
 567         if (result == OS_ERR) {
 568            FREE_C_HEAP_ARRAY(char, filename);
 569            continue;
 570         }
 571 
 572         // skip over files that are not regular files.
 573         if (!S_ISREG(statbuf.st_mode)) {
 574           FREE_C_HEAP_ARRAY(char, filename);
 575           continue;
 576         }
 577 
 578         // compare and save filename with latest creation time
 579         if (statbuf.st_size > 0 && statbuf.st_ctime > oldest_ctime) {
 580 
 581           if (statbuf.st_ctime > oldest_ctime) {
 582             char* user = strchr(dentry->d_name, '_') + 1;
 583 
 584             if (oldest_user != NULL) FREE_C_HEAP_ARRAY(char, oldest_user);
 585             oldest_user = NEW_C_HEAP_ARRAY(char, strlen(user)+1, mtInternal);
 586 
 587             strcpy(oldest_user, user);
 588             oldest_ctime = statbuf.st_ctime;
 589           }
 590         }
 591 
 592         FREE_C_HEAP_ARRAY(char, filename);
 593       }
 594     }
 595     os::closedir(subdirp);
 596     FREE_C_HEAP_ARRAY(char, udbuf);
 597     FREE_C_HEAP_ARRAY(char, usrdir_name);
 598   }
 599   os::closedir(tmpdirp);
 600   FREE_C_HEAP_ARRAY(char, tdbuf);
 601 
 602   return(oldest_user);
 603 }
 604 
 605 // return the name of the user that owns the JVM indicated by the given vmid.
 606 //
 607 static char* get_user_name(int vmid, TRAPS) {
 608 
 609   char psinfo_name[PATH_MAX];
 610   int result;
 611 
 612   snprintf(psinfo_name, PATH_MAX, "/proc/%d/psinfo", vmid);
 613 
 614   RESTARTABLE(::open(psinfo_name, O_RDONLY), result);
 615 
 616   if (result != OS_ERR) {
 617     int fd = result;
 618 
 619     psinfo_t psinfo;
 620     char* addr = (char*)&psinfo;
 621 
 622     for (size_t remaining = sizeof(psinfo_t); remaining > 0;) {
 623 
 624       RESTARTABLE(::read(fd, addr, remaining), result);
 625       if (result == OS_ERR) {
 626         ::close(fd);
 627         THROW_MSG_0(vmSymbols::java_io_IOException(), "Read error");
 628       } else {
 629         remaining-=result;
 630         addr+=result;
 631       }
 632     }
 633 
 634     ::close(fd);
 635 
 636     // get the user name for the effective user id of the process
 637     char* user_name = get_user_name(psinfo.pr_euid);
 638 
 639     return user_name;
 640   }
 641 
 642   if (result == OS_ERR && errno == EACCES) {
 643 
 644     // In this case, the psinfo file for the process id existed,
 645     // but we didn't have permission to access it.
 646     THROW_MSG_0(vmSymbols::java_lang_IllegalArgumentException(),
 647                 strerror(errno));
 648   }
 649 
 650   // at this point, we don't know if the process id itself doesn't
 651   // exist or if the psinfo file doesn't exit. If the psinfo file
 652   // doesn't exist, then we are running on Solaris 2.5.1 or earlier.
 653   // since the structured procfs and old procfs interfaces can't be
 654   // mixed, we attempt to find the file through a directory search.
 655 
 656   return get_user_name_slow(vmid, THREAD);
 657 }
 658 
 659 // return the file name of the backing store file for the named
 660 // shared memory region for the given user name and vmid.
 661 //
 662 // the caller is expected to free the allocated memory.
 663 //
 664 static char* get_sharedmem_filename(const char* dirname, int vmid) {
 665 
 666   // add 2 for the file separator and a NULL terminator.
 667   size_t nbytes = strlen(dirname) + UINT_CHARS + 2;
 668 
 669   char* name = NEW_C_HEAP_ARRAY(char, nbytes, mtInternal);
 670   snprintf(name, nbytes, "%s/%d", dirname, vmid);
 671 
 672   return name;
 673 }
 674 
 675 
 676 // remove file
 677 //
 678 // this method removes the file specified by the given path
 679 //
 680 static void remove_file(const char* path) {
 681 
 682   int result;
 683 
 684   // if the file is a directory, the following unlink will fail. since
 685   // we don't expect to find directories in the user temp directory, we
 686   // won't try to handle this situation. even if accidentially or
 687   // maliciously planted, the directory's presence won't hurt anything.
 688   //
 689   RESTARTABLE(::unlink(path), result);
 690   if (PrintMiscellaneous && Verbose && result == OS_ERR) {
 691     if (errno != ENOENT) {
 692       warning("Could not unlink shared memory backing"
 693               " store file %s : %s\n", path, strerror(errno));
 694     }
 695   }
 696 }
 697 
 698 
 699 // cleanup stale shared memory resources
 700 //
 701 // This method attempts to remove all stale shared memory files in
 702 // the named user temporary directory. It scans the named directory
 703 // for files matching the pattern ^$[0-9]*$. For each file found, the
 704 // process id is extracted from the file name and a test is run to
 705 // determine if the process is alive. If the process is not alive,
 706 // any stale file resources are removed.
 707 //
 708 static void cleanup_sharedmem_resources(const char* dirname) {
 709 
 710   int saved_cwd_fd;
 711   // open the directory
 712   DIR* dirp = open_directory_secure_cwd(dirname, &saved_cwd_fd);
 713   if (dirp == NULL) {
 714     // directory doesn't exist or is insecure, so there is nothing to cleanup
 715     return;
 716   }
 717 
 718   // for each entry in the directory that matches the expected file
 719   // name pattern, determine if the file resources are stale and if
 720   // so, remove the file resources. Note, instrumented HotSpot processes
 721   // for this user may start and/or terminate during this search and
 722   // remove or create new files in this directory. The behavior of this
 723   // loop under these conditions is dependent upon the implementation of
 724   // opendir/readdir.
 725   //
 726   struct dirent* entry;
 727   char* dbuf = NEW_C_HEAP_ARRAY(char, os::readdir_buf_size(dirname), mtInternal);
 728 
 729   errno = 0;
 730   while ((entry = os::readdir(dirp, (struct dirent *)dbuf)) != NULL) {
 731 
 732     pid_t pid = filename_to_pid(entry->d_name);
 733 
 734     if (pid == 0) {
 735 
 736       if (strcmp(entry->d_name, ".") != 0 && strcmp(entry->d_name, "..") != 0) {
 737 
 738         // attempt to remove all unexpected files, except "." and ".."
 739         unlink(entry->d_name);
 740       }
 741 
 742       errno = 0;
 743       continue;
 744     }
 745 
 746     // we now have a file name that converts to a valid integer
 747     // that could represent a process id . if this process id
 748     // matches the current process id or the process is not running,
 749     // then remove the stale file resources.
 750     //
 751     // process liveness is detected by sending signal number 0 to
 752     // the process id (see kill(2)). if kill determines that the
 753     // process does not exist, then the file resources are removed.
 754     // if kill determines that that we don't have permission to
 755     // signal the process, then the file resources are assumed to
 756     // be stale and are removed because the resources for such a
 757     // process should be in a different user specific directory.
 758     //
 759     if ((pid == os::current_process_id()) ||
 760         (kill(pid, 0) == OS_ERR && (errno == ESRCH || errno == EPERM))) {
 761 
 762         unlink(entry->d_name);
 763     }
 764     errno = 0;
 765   }
 766 
 767   // close the directory and reset the current working directory
 768   close_directory_secure_cwd(dirp, saved_cwd_fd);
 769 
 770   FREE_C_HEAP_ARRAY(char, dbuf);
 771 }
 772 
 773 // make the user specific temporary directory. Returns true if
 774 // the directory exists and is secure upon return. Returns false
 775 // if the directory exists but is either a symlink, is otherwise
 776 // insecure, or if an error occurred.
 777 //
 778 static bool make_user_tmp_dir(const char* dirname) {
 779 
 780   // create the directory with 0755 permissions. note that the directory
 781   // will be owned by euid::egid, which may not be the same as uid::gid.
 782   //
 783   if (mkdir(dirname, S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH) == OS_ERR) {
 784     if (errno == EEXIST) {
 785       // The directory already exists and was probably created by another
 786       // JVM instance. However, this could also be the result of a
 787       // deliberate symlink. Verify that the existing directory is safe.
 788       //
 789       if (!is_directory_secure(dirname)) {
 790         // directory is not secure
 791         if (PrintMiscellaneous && Verbose) {
 792           warning("%s directory is insecure\n", dirname);
 793         }
 794         return false;
 795       }
 796     }
 797     else {
 798       // we encountered some other failure while attempting
 799       // to create the directory
 800       //
 801       if (PrintMiscellaneous && Verbose) {
 802         warning("could not create directory %s: %s\n",
 803                 dirname, strerror(errno));
 804       }
 805       return false;
 806     }
 807   }
 808   return true;
 809 }
 810 
 811 // create the shared memory file resources
 812 //
 813 // This method creates the shared memory file with the given size
 814 // This method also creates the user specific temporary directory, if
 815 // it does not yet exist.
 816 //
 817 static int create_sharedmem_resources(const char* dirname, const char* filename, size_t size) {
 818 
 819   // make the user temporary directory
 820   if (!make_user_tmp_dir(dirname)) {
 821     // could not make/find the directory or the found directory
 822     // was not secure
 823     return -1;
 824   }
 825 
 826   int saved_cwd_fd;
 827   // open the directory and set the current working directory to it
 828   DIR* dirp = open_directory_secure_cwd(dirname, &saved_cwd_fd);
 829   if (dirp == NULL) {
 830     // Directory doesn't exist or is insecure, so cannot create shared
 831     // memory file.
 832     return -1;
 833   }
 834 
 835   // Open the filename in the current directory.
 836   // Cannot use O_TRUNC here; truncation of an existing file has to happen
 837   // after the is_file_secure() check below.
 838   int result;
 839   RESTARTABLE(::open(filename, O_RDWR|O_CREAT|O_NOFOLLOW, S_IREAD|S_IWRITE), result);
 840   if (result == OS_ERR) {
 841     if (PrintMiscellaneous && Verbose) {
 842       if (errno == ELOOP) {
 843         warning("file %s is a symlink and is not secure\n", filename);
 844       } else {
 845         warning("could not create file %s: %s\n", filename, strerror(errno));
 846       }
 847     }
 848     // close the directory and reset the current working directory
 849     close_directory_secure_cwd(dirp, saved_cwd_fd);
 850 
 851     return -1;
 852   }
 853   // close the directory and reset the current working directory
 854   close_directory_secure_cwd(dirp, saved_cwd_fd);
 855 
 856   // save the file descriptor
 857   int fd = result;
 858 
 859   // check to see if the file is secure
 860   if (!is_file_secure(fd, filename)) {
 861     ::close(fd);
 862     return -1;
 863   }
 864 
 865   // truncate the file to get rid of any existing data
 866   RESTARTABLE(::ftruncate(fd, (off_t)0), result);
 867   if (result == OS_ERR) {
 868     if (PrintMiscellaneous && Verbose) {
 869       warning("could not truncate shared memory file: %s\n", strerror(errno));
 870     }
 871     ::close(fd);
 872     return -1;
 873   }
 874   // set the file size
 875   RESTARTABLE(::ftruncate(fd, (off_t)size), result);
 876   if (result == OS_ERR) {
 877     if (PrintMiscellaneous && Verbose) {
 878       warning("could not set shared memory file size: %s\n", strerror(errno));
 879     }
 880     ::close(fd);
 881     return -1;
 882   }
 883 
 884   return fd;
 885 }
 886 
 887 // open the shared memory file for the given user and vmid. returns
 888 // the file descriptor for the open file or -1 if the file could not
 889 // be opened.
 890 //
 891 static int open_sharedmem_file(const char* filename, int oflags, TRAPS) {
 892 
 893   // open the file
 894   int result;
 895   RESTARTABLE(::open(filename, oflags), result);
 896   if (result == OS_ERR) {
 897     if (errno == ENOENT) {
 898       THROW_MSG_(vmSymbols::java_lang_IllegalArgumentException(),
 899                   "Process not found", OS_ERR);
 900     }
 901     else if (errno == EACCES) {
 902       THROW_MSG_(vmSymbols::java_lang_IllegalArgumentException(),
 903                   "Permission denied", OS_ERR);
 904     }
 905     else {
 906       THROW_MSG_(vmSymbols::java_io_IOException(), strerror(errno), OS_ERR);
 907     }
 908   }
 909   int fd = result;
 910 
 911   // check to see if the file is secure
 912   if (!is_file_secure(fd, filename)) {
 913     ::close(fd);
 914     return -1;
 915   }
 916 
 917   return fd;
 918 }
 919 
 920 // create a named shared memory region. returns the address of the
 921 // memory region on success or NULL on failure. A return value of
 922 // NULL will ultimately disable the shared memory feature.
 923 //
 924 // On Solaris and Linux, the name space for shared memory objects
 925 // is the file system name space.
 926 //
 927 // A monitoring application attaching to a JVM does not need to know
 928 // the file system name of the shared memory object. However, it may
 929 // be convenient for applications to discover the existence of newly
 930 // created and terminating JVMs by watching the file system name space
 931 // for files being created or removed.
 932 //
 933 static char* mmap_create_shared(size_t size) {
 934 
 935   int result;
 936   int fd;
 937   char* mapAddress;
 938 
 939   int vmid = os::current_process_id();
 940 
 941   char* user_name = get_user_name(geteuid());
 942 
 943   if (user_name == NULL)
 944     return NULL;
 945 
 946   char* dirname = get_user_tmp_dir(user_name);
 947   char* filename = get_sharedmem_filename(dirname, vmid);
 948 
 949   // get the short filename
 950   char* short_filename = strrchr(filename, '/');
 951   if (short_filename == NULL) {
 952     short_filename = filename;
 953   } else {
 954     short_filename++;
 955   }
 956 
 957   // cleanup any stale shared memory files
 958   cleanup_sharedmem_resources(dirname);
 959 
 960   assert(((size > 0) && (size % os::vm_page_size() == 0)),
 961          "unexpected PerfMemory region size");
 962 
 963   fd = create_sharedmem_resources(dirname, short_filename, size);
 964 
 965   FREE_C_HEAP_ARRAY(char, user_name);
 966   FREE_C_HEAP_ARRAY(char, dirname);
 967 
 968   if (fd == -1) {
 969     FREE_C_HEAP_ARRAY(char, filename);
 970     return NULL;
 971   }
 972 
 973   mapAddress = (char*)::mmap((char*)0, size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 974 
 975   result = ::close(fd);
 976   assert(result != OS_ERR, "could not close file");
 977 
 978   if (mapAddress == MAP_FAILED) {
 979     if (PrintMiscellaneous && Verbose) {
 980       warning("mmap failed -  %s\n", strerror(errno));
 981     }
 982     remove_file(filename);
 983     FREE_C_HEAP_ARRAY(char, filename);
 984     return NULL;
 985   }
 986 
 987   // save the file name for use in delete_shared_memory()
 988   backing_store_file_name = filename;
 989 
 990   // clear the shared memory region
 991   (void)::memset((void*) mapAddress, 0, size);
 992 
 993   // it does not go through os api, the operation has to record from here
 994   MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress,
 995     size, CURRENT_PC, mtInternal);
 996 
 997   return mapAddress;
 998 }
 999 
1000 // release a named shared memory region
1001 //
1002 static void unmap_shared(char* addr, size_t bytes) {
1003   os::release_memory(addr, bytes);
1004 }
1005 
1006 // create the PerfData memory region in shared memory.
1007 //
1008 static char* create_shared_memory(size_t size) {
1009 
1010   // create the shared memory region.
1011   return mmap_create_shared(size);
1012 }
1013 
1014 // delete the shared PerfData memory region
1015 //
1016 static void delete_shared_memory(char* addr, size_t size) {
1017 
1018   // cleanup the persistent shared memory resources. since DestroyJavaVM does
1019   // not support unloading of the JVM, unmapping of the memory resource is
1020   // not performed. The memory will be reclaimed by the OS upon termination of
1021   // the process. The backing store file is deleted from the file system.
1022 
1023   assert(!PerfDisableSharedMem, "shouldn't be here");
1024 
1025   if (backing_store_file_name != NULL) {
1026     remove_file(backing_store_file_name);
1027     // Don't.. Free heap memory could deadlock os::abort() if it is called
1028     // from signal handler. OS will reclaim the heap memory.
1029     // FREE_C_HEAP_ARRAY(char, backing_store_file_name);
1030     backing_store_file_name = NULL;
1031   }
1032 }
1033 
1034 // return the size of the file for the given file descriptor
1035 // or 0 if it is not a valid size for a shared memory file
1036 //
1037 static size_t sharedmem_filesize(int fd, TRAPS) {
1038 
1039   struct stat statbuf;
1040   int result;
1041 
1042   RESTARTABLE(::fstat(fd, &statbuf), result);
1043   if (result == OS_ERR) {
1044     if (PrintMiscellaneous && Verbose) {
1045       warning("fstat failed: %s\n", strerror(errno));
1046     }
1047     THROW_MSG_0(vmSymbols::java_io_IOException(),
1048                 "Could not determine PerfMemory size");
1049   }
1050 
1051   if ((statbuf.st_size == 0) ||
1052      ((size_t)statbuf.st_size % os::vm_page_size() != 0)) {
1053     THROW_MSG_0(vmSymbols::java_lang_Exception(),
1054                 "Invalid PerfMemory size");
1055   }
1056 
1057   return (size_t)statbuf.st_size;
1058 }
1059 
1060 // attach to a named shared memory region.
1061 //
1062 static void mmap_attach_shared(const char* user, int vmid, PerfMemory::PerfMemoryMode mode, char** addr, size_t* sizep, TRAPS) {
1063 
1064   char* mapAddress;
1065   int result;
1066   int fd;
1067   size_t size = 0;
1068   const char* luser = NULL;
1069 
1070   int mmap_prot;
1071   int file_flags;
1072 
1073   ResourceMark rm;
1074 
1075   // map the high level access mode to the appropriate permission
1076   // constructs for the file and the shared memory mapping.
1077   if (mode == PerfMemory::PERF_MODE_RO) {
1078     mmap_prot = PROT_READ;
1079     file_flags = O_RDONLY | O_NOFOLLOW;
1080   }
1081   else if (mode == PerfMemory::PERF_MODE_RW) {
1082 #ifdef LATER
1083     mmap_prot = PROT_READ | PROT_WRITE;
1084     file_flags = O_RDWR | O_NOFOLLOW;
1085 #else
1086     THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1087               "Unsupported access mode");
1088 #endif
1089   }
1090   else {
1091     THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1092               "Illegal access mode");
1093   }
1094 
1095   if (user == NULL || strlen(user) == 0) {
1096     luser = get_user_name(vmid, CHECK);
1097   }
1098   else {
1099     luser = user;
1100   }
1101 
1102   if (luser == NULL) {
1103     THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1104               "Could not map vmid to user Name");
1105   }
1106 
1107   char* dirname = get_user_tmp_dir(luser);
1108 
1109   // since we don't follow symbolic links when creating the backing
1110   // store file, we don't follow them when attaching either.
1111   //
1112   if (!is_directory_secure(dirname)) {
1113     FREE_C_HEAP_ARRAY(char, dirname);
1114     if (luser != user) {
1115       FREE_C_HEAP_ARRAY(char, luser);
1116     }
1117     THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1118               "Process not found");
1119   }
1120 
1121   char* filename = get_sharedmem_filename(dirname, vmid);
1122 
1123   // copy heap memory to resource memory. the open_sharedmem_file
1124   // method below need to use the filename, but could throw an
1125   // exception. using a resource array prevents the leak that
1126   // would otherwise occur.
1127   char* rfilename = NEW_RESOURCE_ARRAY(char, strlen(filename) + 1);
1128   strcpy(rfilename, filename);
1129 
1130   // free the c heap resources that are no longer needed
1131   if (luser != user) FREE_C_HEAP_ARRAY(char, luser);
1132   FREE_C_HEAP_ARRAY(char, dirname);
1133   FREE_C_HEAP_ARRAY(char, filename);
1134 
1135   // open the shared memory file for the give vmid
1136   fd = open_sharedmem_file(rfilename, file_flags, THREAD);
1137 
1138   if (fd == OS_ERR) {
1139     return;
1140   }
1141 
1142   if (HAS_PENDING_EXCEPTION) {
1143     ::close(fd);
1144     return;
1145   }
1146 
1147   if (*sizep == 0) {
1148     size = sharedmem_filesize(fd, CHECK);
1149   } else {
1150     size = *sizep;
1151   }
1152 
1153   assert(size > 0, "unexpected size <= 0");
1154 
1155   mapAddress = (char*)::mmap((char*)0, size, mmap_prot, MAP_SHARED, fd, 0);
1156 
1157   result = ::close(fd);
1158   assert(result != OS_ERR, "could not close file");
1159 
1160   if (mapAddress == MAP_FAILED) {
1161     if (PrintMiscellaneous && Verbose) {
1162       warning("mmap failed: %s\n", strerror(errno));
1163     }
1164     THROW_MSG(vmSymbols::java_lang_OutOfMemoryError(),
1165               "Could not map PerfMemory");
1166   }
1167 
1168   // it does not go through os api, the operation has to record from here
1169   MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress,
1170     size, CURRENT_PC, mtInternal);
1171 
1172   *addr = mapAddress;
1173   *sizep = size;
1174 
1175   if (PerfTraceMemOps) {
1176     tty->print("mapped " SIZE_FORMAT " bytes for vmid %d at "
1177                INTPTR_FORMAT "\n", size, vmid, (void*)mapAddress);
1178   }
1179 }
1180 
1181 
1182 
1183 
1184 // create the PerfData memory region
1185 //
1186 // This method creates the memory region used to store performance
1187 // data for the JVM. The memory may be created in standard or
1188 // shared memory.
1189 //
1190 void PerfMemory::create_memory_region(size_t size) {
1191 
1192   if (PerfDisableSharedMem) {
1193     // do not share the memory for the performance data.
1194     _start = create_standard_memory(size);
1195   }
1196   else {
1197     _start = create_shared_memory(size);
1198     if (_start == NULL) {
1199 
1200       // creation of the shared memory region failed, attempt
1201       // to create a contiguous, non-shared memory region instead.
1202       //
1203       if (PrintMiscellaneous && Verbose) {
1204         warning("Reverting to non-shared PerfMemory region.\n");
1205       }
1206       PerfDisableSharedMem = true;
1207       _start = create_standard_memory(size);
1208     }
1209   }
1210 
1211   if (_start != NULL) _capacity = size;
1212 
1213 }
1214 
1215 // delete the PerfData memory region
1216 //
1217 // This method deletes the memory region used to store performance
1218 // data for the JVM. The memory region indicated by the <address, size>
1219 // tuple will be inaccessible after a call to this method.
1220 //
1221 void PerfMemory::delete_memory_region() {
1222 
1223   assert((start() != NULL && capacity() > 0), "verify proper state");
1224 
1225   // If user specifies PerfDataSaveFile, it will save the performance data
1226   // to the specified file name no matter whether PerfDataSaveToFile is specified
1227   // or not. In other word, -XX:PerfDataSaveFile=.. overrides flag
1228   // -XX:+PerfDataSaveToFile.
1229   if (PerfDataSaveToFile || PerfDataSaveFile != NULL) {
1230     save_memory_to_file(start(), capacity());
1231   }
1232 
1233   if (PerfDisableSharedMem) {
1234     delete_standard_memory(start(), capacity());
1235   }
1236   else {
1237     delete_shared_memory(start(), capacity());
1238   }
1239 }
1240 
1241 // attach to the PerfData memory region for another JVM
1242 //
1243 // This method returns an <address, size> tuple that points to
1244 // a memory buffer that is kept reasonably synchronized with
1245 // the PerfData memory region for the indicated JVM. This
1246 // buffer may be kept in synchronization via shared memory
1247 // or some other mechanism that keeps the buffer updated.
1248 //
1249 // If the JVM chooses not to support the attachability feature,
1250 // this method should throw an UnsupportedOperation exception.
1251 //
1252 // This implementation utilizes named shared memory to map
1253 // the indicated process's PerfData memory region into this JVMs
1254 // address space.
1255 //
1256 void PerfMemory::attach(const char* user, int vmid, PerfMemoryMode mode, char** addrp, size_t* sizep, TRAPS) {
1257 
1258   if (vmid == 0 || vmid == os::current_process_id()) {
1259      *addrp = start();
1260      *sizep = capacity();
1261      return;
1262   }
1263 
1264   mmap_attach_shared(user, vmid, mode, addrp, sizep, CHECK);
1265 }
1266 
1267 // detach from the PerfData memory region of another JVM
1268 //
1269 // This method detaches the PerfData memory region of another
1270 // JVM, specified as an <address, size> tuple of a buffer
1271 // in this process's address space. This method may perform
1272 // arbitrary actions to accomplish the detachment. The memory
1273 // region specified by <address, size> will be inaccessible after
1274 // a call to this method.
1275 //
1276 // If the JVM chooses not to support the attachability feature,
1277 // this method should throw an UnsupportedOperation exception.
1278 //
1279 // This implementation utilizes named shared memory to detach
1280 // the indicated process's PerfData memory region from this
1281 // process's address space.
1282 //
1283 void PerfMemory::detach(char* addr, size_t bytes, TRAPS) {
1284 
1285   assert(addr != 0, "address sanity check");
1286   assert(bytes > 0, "capacity sanity check");
1287 
1288   if (PerfMemory::contains(addr) || PerfMemory::contains(addr + bytes - 1)) {
1289     // prevent accidental detachment of this process's PerfMemory region
1290     return;
1291   }
1292 
1293   unmap_shared(addr, bytes);
1294 }