1 /*
   2  * Copyright (c) 2001, 2015, 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, mtInternal);
 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   // 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 
 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 and
 379   // handle errors, otherwise shared memory files will be created in cwd.
 380   result = fchdir(fd);
 381   if (result == OS_ERR) {
 382     if (PrintMiscellaneous && Verbose) {
 383       warning("could not change to directory %s", dirname);
 384     }
 385     if (*saved_cwd_fd != -1) {
 386       ::close(*saved_cwd_fd);
 387       *saved_cwd_fd = -1;
 388     }
 389     // Close the directory.
 390     os::closedir(dirp);
 391     return NULL;
 392   } else {
 393     return dirp;
 394   }
 395 }
 396 
 397 // Close the directory and restore the current working directory.
 398 //
 399 static void close_directory_secure_cwd(DIR* dirp, int saved_cwd_fd) {
 400 
 401   int result;
 402   // If we have a saved cwd change back to it and close the fd.
 403   if (saved_cwd_fd != -1) {
 404     result = fchdir(saved_cwd_fd);
 405     ::close(saved_cwd_fd);
 406   }
 407 
 408   // Close the directory.
 409   os::closedir(dirp);
 410 }
 411 
 412 // Check if the given file descriptor is considered a secure.
 413 //
 414 static bool is_file_secure(int fd, const char *filename) {
 415 
 416   int result;
 417   struct stat statbuf;
 418 
 419   // Determine if the file is secure.
 420   RESTARTABLE(::fstat(fd, &statbuf), result);
 421   if (result == OS_ERR) {
 422     if (PrintMiscellaneous && Verbose) {
 423       warning("fstat failed on %s: %s\n", filename, strerror(errno));
 424     }
 425     return false;
 426   }
 427   if (statbuf.st_nlink > 1) {
 428     // A file with multiple links is not expected.
 429     if (PrintMiscellaneous && Verbose) {
 430       warning("file %s has multiple links\n", filename);
 431     }
 432     return false;
 433   }
 434   return true;
 435 }
 436 
 437 // return the user name for the given user id
 438 //
 439 // the caller is expected to free the allocated memory.
 440 //
 441 static char* get_user_name(uid_t uid) {
 442 
 443   struct passwd pwent;
 444 
 445   // determine the max pwbuf size from sysconf, and hardcode
 446   // a default if this not available through sysconf.
 447   //
 448   long bufsize = sysconf(_SC_GETPW_R_SIZE_MAX);
 449   if (bufsize == -1)
 450     bufsize = 1024;
 451 
 452   char* pwbuf = NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 453 
 454   // POSIX interface to getpwuid_r is used on LINUX
 455   struct passwd* p;
 456   int result = getpwuid_r(uid, &pwent, pwbuf, (size_t)bufsize, &p);
 457 
 458   if (result != 0 || p == NULL || p->pw_name == NULL || *(p->pw_name) == '\0') {
 459     if (PrintMiscellaneous && Verbose) {
 460       if (result != 0) {
 461         warning("Could not retrieve passwd entry: %s\n",
 462                 strerror(result));
 463       }
 464       else if (p == NULL) {
 465         // this check is added to protect against an observed problem
 466         // with getpwuid_r() on RedHat 9 where getpwuid_r returns 0,
 467         // indicating success, but has p == NULL. This was observed when
 468         // inserting a file descriptor exhaustion fault prior to the call
 469         // getpwuid_r() call. In this case, error is set to the appropriate
 470         // error condition, but this is undocumented behavior. This check
 471         // is safe under any condition, but the use of errno in the output
 472         // message may result in an erroneous message.
 473         // Bug Id 89052 was opened with RedHat.
 474         //
 475         warning("Could not retrieve passwd entry: %s\n",
 476                 strerror(errno));
 477       }
 478       else {
 479         warning("Could not determine user name: %s\n",
 480                 p->pw_name == NULL ? "pw_name = NULL" :
 481                                      "pw_name zero length");
 482       }
 483     }
 484     FREE_C_HEAP_ARRAY(char, pwbuf, mtInternal);
 485     return NULL;
 486   }
 487 
 488   char* user_name = NEW_C_HEAP_ARRAY(char, strlen(p->pw_name) + 1, mtInternal);
 489   strcpy(user_name, p->pw_name);
 490 
 491   FREE_C_HEAP_ARRAY(char, pwbuf, mtInternal);
 492   return user_name;
 493 }
 494 
 495 // return the name of the user that owns the process identified by vmid.
 496 //
 497 // This method uses a slow directory search algorithm to find the backing
 498 // store file for the specified vmid and returns the user name, as determined
 499 // by the user name suffix of the hsperfdata_<username> directory name.
 500 //
 501 // the caller is expected to free the allocated memory.
 502 //
 503 static char* get_user_name_slow(int vmid, TRAPS) {
 504 
 505   // short circuit the directory search if the process doesn't even exist.
 506   if (kill(vmid, 0) == OS_ERR) {
 507     if (errno == ESRCH) {
 508       THROW_MSG_0(vmSymbols::java_lang_IllegalArgumentException(),
 509                   "Process not found");
 510     }
 511     else /* EPERM */ {
 512       THROW_MSG_0(vmSymbols::java_io_IOException(), strerror(errno));
 513     }
 514   }
 515 
 516   // directory search
 517   char* oldest_user = NULL;
 518   time_t oldest_ctime = 0;
 519 
 520   const char* tmpdirname = os::get_temp_directory();
 521 
 522   // open the temp directory
 523   DIR* tmpdirp = os::opendir(tmpdirname);
 524 
 525   if (tmpdirp == NULL) {
 526     // Cannot open the directory to get the user name, return.
 527     return NULL;
 528   }
 529 
 530   // for each entry in the directory that matches the pattern hsperfdata_*,
 531   // open the directory and check if the file for the given vmid exists.
 532   // The file with the expected name and the latest creation date is used
 533   // to determine the user name for the process id.
 534   //
 535   struct dirent* dentry;
 536   errno = 0;
 537   while ((dentry = os::readdir(tmpdirp)) != NULL) {
 538 
 539     // check if the directory entry is a hsperfdata file
 540     if (strncmp(dentry->d_name, PERFDATA_NAME, strlen(PERFDATA_NAME)) != 0) {
 541       continue;
 542     }
 543 
 544     char* usrdir_name = NEW_C_HEAP_ARRAY(char,
 545                  strlen(tmpdirname) + strlen(dentry->d_name) + 2, mtInternal);
 546     strcpy(usrdir_name, tmpdirname);
 547     strcat(usrdir_name, "/");
 548     strcat(usrdir_name, dentry->d_name);
 549 
 550     // open the user directory
 551     DIR* subdirp = open_directory_secure(usrdir_name);
 552 
 553     if (subdirp == NULL) {
 554       FREE_C_HEAP_ARRAY(char, usrdir_name, mtInternal);
 555       continue;
 556     }
 557 
 558     struct dirent* udentry;
 559     errno = 0;
 560     while ((udentry = os::readdir(subdirp)) != NULL) {
 561 
 562       if (filename_to_pid(udentry->d_name) == vmid) {
 563         struct stat statbuf;
 564         int result;
 565 
 566         char* filename = NEW_C_HEAP_ARRAY(char,
 567                  strlen(usrdir_name) + strlen(udentry->d_name) + 2, mtInternal);
 568 
 569         strcpy(filename, usrdir_name);
 570         strcat(filename, "/");
 571         strcat(filename, udentry->d_name);
 572 
 573         // don't follow symbolic links for the file
 574         RESTARTABLE(::lstat(filename, &statbuf), result);
 575         if (result == OS_ERR) {
 576            FREE_C_HEAP_ARRAY(char, filename, mtInternal);
 577            continue;
 578         }
 579 
 580         // skip over files that are not regular files.
 581         if (!S_ISREG(statbuf.st_mode)) {
 582           FREE_C_HEAP_ARRAY(char, filename, mtInternal);
 583           continue;
 584         }
 585 
 586         // compare and save filename with latest creation time
 587         if (statbuf.st_size > 0 && statbuf.st_ctime > oldest_ctime) {
 588 
 589           if (statbuf.st_ctime > oldest_ctime) {
 590             char* user = strchr(dentry->d_name, '_') + 1;
 591 
 592             if (oldest_user != NULL) FREE_C_HEAP_ARRAY(char, oldest_user, mtInternal);
 593             oldest_user = NEW_C_HEAP_ARRAY(char, strlen(user)+1, mtInternal);
 594 
 595             strcpy(oldest_user, user);
 596             oldest_ctime = statbuf.st_ctime;
 597           }
 598         }
 599 
 600         FREE_C_HEAP_ARRAY(char, filename, mtInternal);
 601       }
 602     }
 603     os::closedir(subdirp);
 604     FREE_C_HEAP_ARRAY(char, usrdir_name, mtInternal);
 605   }
 606   os::closedir(tmpdirp);
 607 
 608   return(oldest_user);
 609 }
 610 
 611 // return the name of the user that owns the JVM indicated by the given vmid.
 612 //
 613 static char* get_user_name(int vmid, TRAPS) {
 614   return get_user_name_slow(vmid, CHECK_NULL);
 615 }
 616 
 617 // return the file name of the backing store file for the named
 618 // shared memory region for the given user name and vmid.
 619 //
 620 // the caller is expected to free the allocated memory.
 621 //
 622 static char* get_sharedmem_filename(const char* dirname, int vmid) {
 623 
 624   // add 2 for the file separator and a null terminator.
 625   size_t nbytes = strlen(dirname) + UINT_CHARS + 2;
 626 
 627   char* name = NEW_C_HEAP_ARRAY(char, nbytes, mtInternal);
 628   snprintf(name, nbytes, "%s/%d", dirname, vmid);
 629 
 630   return name;
 631 }
 632 
 633 
 634 // remove file
 635 //
 636 // this method removes the file specified by the given path
 637 //
 638 static void remove_file(const char* path) {
 639 
 640   int result;
 641 
 642   // if the file is a directory, the following unlink will fail. since
 643   // we don't expect to find directories in the user temp directory, we
 644   // won't try to handle this situation. even if accidentially or
 645   // maliciously planted, the directory's presence won't hurt anything.
 646   //
 647   RESTARTABLE(::unlink(path), result);
 648   if (PrintMiscellaneous && Verbose && result == OS_ERR) {
 649     if (errno != ENOENT) {
 650       warning("Could not unlink shared memory backing"
 651               " store file %s : %s\n", path, strerror(errno));
 652     }
 653   }
 654 }
 655 
 656 
 657 // cleanup stale shared memory resources
 658 //
 659 // This method attempts to remove all stale shared memory files in
 660 // the named user temporary directory. It scans the named directory
 661 // for files matching the pattern ^$[0-9]*$. For each file found, the
 662 // process id is extracted from the file name and a test is run to
 663 // determine if the process is alive. If the process is not alive,
 664 // any stale file resources are removed.
 665 //
 666 static void cleanup_sharedmem_resources(const char* dirname) {
 667 
 668   int saved_cwd_fd;
 669   // open the directory and set the current working directory to it
 670   DIR* dirp = open_directory_secure_cwd(dirname, &saved_cwd_fd);
 671   if (dirp == NULL) {
 672     // directory doesn't exist or is insecure, so there is nothing to cleanup
 673     return;
 674   }
 675 
 676   // for each entry in the directory that matches the expected file
 677   // name pattern, determine if the file resources are stale and if
 678   // so, remove the file resources. Note, instrumented HotSpot processes
 679   // for this user may start and/or terminate during this search and
 680   // remove or create new files in this directory. The behavior of this
 681   // loop under these conditions is dependent upon the implementation of
 682   // opendir/readdir.
 683   //
 684   struct dirent* entry;
 685   errno = 0;
 686   while ((entry = os::readdir(dirp)) != NULL) {
 687 
 688     pid_t pid = filename_to_pid(entry->d_name);
 689 
 690     if (pid == 0) {
 691 
 692       if (strcmp(entry->d_name, ".") != 0 && strcmp(entry->d_name, "..") != 0) {
 693 
 694         // attempt to remove all unexpected files, except "." and ".."
 695         unlink(entry->d_name);
 696       }
 697 
 698       errno = 0;
 699       continue;
 700     }
 701 
 702     // we now have a file name that converts to a valid integer
 703     // that could represent a process id . if this process id
 704     // matches the current process id or the process is not running,
 705     // then remove the stale file resources.
 706     //
 707     // process liveness is detected by sending signal number 0 to
 708     // the process id (see kill(2)). if kill determines that the
 709     // process does not exist, then the file resources are removed.
 710     // if kill determines that that we don't have permission to
 711     // signal the process, then the file resources are assumed to
 712     // be stale and are removed because the resources for such a
 713     // process should be in a different user specific directory.
 714     //
 715     if ((pid == os::current_process_id()) ||
 716         (kill(pid, 0) == OS_ERR && (errno == ESRCH || errno == EPERM))) {
 717 
 718         unlink(entry->d_name);
 719     }
 720     errno = 0;
 721   }
 722 
 723   // close the directory and reset the current working directory
 724   close_directory_secure_cwd(dirp, saved_cwd_fd);
 725 }
 726 
 727 // make the user specific temporary directory. Returns true if
 728 // the directory exists and is secure upon return. Returns false
 729 // if the directory exists but is either a symlink, is otherwise
 730 // insecure, or if an error occurred.
 731 //
 732 static bool make_user_tmp_dir(const char* dirname) {
 733 
 734   // create the directory with 0755 permissions. note that the directory
 735   // will be owned by euid::egid, which may not be the same as uid::gid.
 736   //
 737   if (mkdir(dirname, S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH) == OS_ERR) {
 738     if (errno == EEXIST) {
 739       // The directory already exists and was probably created by another
 740       // JVM instance. However, this could also be the result of a
 741       // deliberate symlink. Verify that the existing directory is safe.
 742       //
 743       if (!is_directory_secure(dirname)) {
 744         // directory is not secure
 745         if (PrintMiscellaneous && Verbose) {
 746           warning("%s directory is insecure\n", dirname);
 747         }
 748         return false;
 749       }
 750     }
 751     else {
 752       // we encountered some other failure while attempting
 753       // to create the directory
 754       //
 755       if (PrintMiscellaneous && Verbose) {
 756         warning("could not create directory %s: %s\n",
 757                 dirname, strerror(errno));
 758       }
 759       return false;
 760     }
 761   }
 762   return true;
 763 }
 764 
 765 // create the shared memory file resources
 766 //
 767 // This method creates the shared memory file with the given size
 768 // This method also creates the user specific temporary directory, if
 769 // it does not yet exist.
 770 //
 771 static int create_sharedmem_resources(const char* dirname, const char* filename, size_t size) {
 772 
 773   // make the user temporary directory
 774   if (!make_user_tmp_dir(dirname)) {
 775     // could not make/find the directory or the found directory
 776     // was not secure
 777     return -1;
 778   }
 779 
 780   int saved_cwd_fd;
 781   // open the directory and set the current working directory to it
 782   DIR* dirp = open_directory_secure_cwd(dirname, &saved_cwd_fd);
 783   if (dirp == NULL) {
 784     // Directory doesn't exist or is insecure, so cannot create shared
 785     // memory file.
 786     return -1;
 787   }
 788 
 789   // Open the filename in the current directory.
 790   // Cannot use O_TRUNC here; truncation of an existing file has to happen
 791   // after the is_file_secure() check below.
 792   int result;
 793   RESTARTABLE(::open(filename, O_RDWR|O_CREAT|O_NOFOLLOW, S_IREAD|S_IWRITE), result);
 794   if (result == OS_ERR) {
 795     if (PrintMiscellaneous && Verbose) {
 796       if (errno == ELOOP) {
 797         warning("file %s is a symlink and is not secure\n", filename);
 798       } else {
 799         warning("could not create file %s: %s\n", filename, strerror(errno));
 800       }
 801     }
 802     // close the directory and reset the current working directory
 803     close_directory_secure_cwd(dirp, saved_cwd_fd);
 804 
 805     return -1;
 806   }
 807   // close the directory and reset the current working directory
 808   close_directory_secure_cwd(dirp, saved_cwd_fd);
 809 
 810   // save the file descriptor
 811   int fd = result;
 812 
 813   // check to see if the file is secure
 814   if (!is_file_secure(fd, filename)) {
 815     ::close(fd);
 816     return -1;
 817   }
 818 
 819   // truncate the file to get rid of any existing data
 820   RESTARTABLE(::ftruncate(fd, (off_t)0), result);
 821   if (result == OS_ERR) {
 822     if (PrintMiscellaneous && Verbose) {
 823       warning("could not truncate shared memory file: %s\n", strerror(errno));
 824     }
 825     ::close(fd);
 826     return -1;
 827   }
 828   // set the file size
 829   RESTARTABLE(::ftruncate(fd, (off_t)size), result);
 830   if (result == OS_ERR) {
 831     if (PrintMiscellaneous && Verbose) {
 832       warning("could not set shared memory file size: %s\n", strerror(errno));
 833     }
 834     ::close(fd);
 835     return -1;
 836   }
 837 
 838   // Verify that we have enough disk space for this file.
 839   // We'll get random SIGBUS crashes on memory accesses if
 840   // we don't.
 841 
 842   for (size_t seekpos = 0; seekpos < size; seekpos += os::vm_page_size()) {
 843     int zero_int = 0;
 844     result = (int)os::seek_to_file_offset(fd, (jlong)(seekpos));
 845     if (result == -1 ) break;
 846     RESTARTABLE(::write(fd, &zero_int, 1), result);
 847     if (result != 1) {
 848       if (errno == ENOSPC) {
 849         warning("Insufficient space for shared memory file:\n   %s\nTry using the -Djava.io.tmpdir= option to select an alternate temp location.\n", filename);
 850       }
 851       break;
 852     }
 853   }
 854 
 855   if (result != -1) {
 856     return fd;
 857   } else {
 858     ::close(fd);
 859     return -1;
 860   }
 861 }
 862 
 863 // open the shared memory file for the given user and vmid. returns
 864 // the file descriptor for the open file or -1 if the file could not
 865 // be opened.
 866 //
 867 static int open_sharedmem_file(const char* filename, int oflags, TRAPS) {
 868 
 869   // open the file
 870   int result;
 871   RESTARTABLE(::open(filename, oflags), result);
 872   if (result == OS_ERR) {
 873     if (errno == ENOENT) {
 874       THROW_MSG_(vmSymbols::java_lang_IllegalArgumentException(),
 875                   "Process not found", OS_ERR);
 876     }
 877     else if (errno == EACCES) {
 878       THROW_MSG_(vmSymbols::java_lang_IllegalArgumentException(),
 879                   "Permission denied", OS_ERR);
 880     }
 881     else {
 882       THROW_MSG_(vmSymbols::java_io_IOException(), strerror(errno), OS_ERR);
 883     }
 884   }
 885   int fd = result;
 886 
 887   // check to see if the file is secure
 888   if (!is_file_secure(fd, filename)) {
 889     ::close(fd);
 890     return -1;
 891   }
 892 
 893   return fd;
 894 }
 895 
 896 // create a named shared memory region. returns the address of the
 897 // memory region on success or NULL on failure. A return value of
 898 // NULL will ultimately disable the shared memory feature.
 899 //
 900 // On Solaris and Bsd, the name space for shared memory objects
 901 // is the file system name space.
 902 //
 903 // A monitoring application attaching to a JVM does not need to know
 904 // the file system name of the shared memory object. However, it may
 905 // be convenient for applications to discover the existence of newly
 906 // created and terminating JVMs by watching the file system name space
 907 // for files being created or removed.
 908 //
 909 static char* mmap_create_shared(size_t size) {
 910 
 911   int result;
 912   int fd;
 913   char* mapAddress;
 914 
 915   int vmid = os::current_process_id();
 916 
 917   char* user_name = get_user_name(geteuid());
 918 
 919   if (user_name == NULL)
 920     return NULL;
 921 
 922   char* dirname = get_user_tmp_dir(user_name);
 923   char* filename = get_sharedmem_filename(dirname, vmid);
 924 
 925   // get the short filename
 926   char* short_filename = strrchr(filename, '/');
 927   if (short_filename == NULL) {
 928     short_filename = filename;
 929   } else {
 930     short_filename++;
 931   }
 932 
 933   // cleanup any stale shared memory files
 934   cleanup_sharedmem_resources(dirname);
 935 
 936   assert(((size > 0) && (size % os::vm_page_size() == 0)),
 937          "unexpected PerfMemory region size");
 938 
 939   fd = create_sharedmem_resources(dirname, short_filename, size);
 940 
 941   FREE_C_HEAP_ARRAY(char, user_name, mtInternal);
 942   FREE_C_HEAP_ARRAY(char, dirname, mtInternal);
 943 
 944   if (fd == -1) {
 945     FREE_C_HEAP_ARRAY(char, filename, mtInternal);
 946     return NULL;
 947   }
 948 
 949   mapAddress = (char*)::mmap((char*)0, size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
 950 
 951   result = ::close(fd);
 952   assert(result != OS_ERR, "could not close file");
 953 
 954   if (mapAddress == MAP_FAILED) {
 955     if (PrintMiscellaneous && Verbose) {
 956       warning("mmap failed -  %s\n", strerror(errno));
 957     }
 958     remove_file(filename);
 959     FREE_C_HEAP_ARRAY(char, filename, mtInternal);
 960     return NULL;
 961   }
 962 
 963   // save the file name for use in delete_shared_memory()
 964   backing_store_file_name = filename;
 965 
 966   // clear the shared memory region
 967   (void)::memset((void*) mapAddress, 0, size);
 968 
 969   // it does not go through os api, the operation has to record from here
 970   MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, size, CURRENT_PC, mtInternal);
 971 
 972   return mapAddress;
 973 }
 974 
 975 // release a named shared memory region
 976 //
 977 static void unmap_shared(char* addr, size_t bytes) {
 978   os::release_memory(addr, bytes);
 979 }
 980 
 981 // create the PerfData memory region in shared memory.
 982 //
 983 static char* create_shared_memory(size_t size) {
 984 
 985   // create the shared memory region.
 986   return mmap_create_shared(size);
 987 }
 988 
 989 // delete the shared PerfData memory region
 990 //
 991 static void delete_shared_memory(char* addr, size_t size) {
 992 
 993   // cleanup the persistent shared memory resources. since DestroyJavaVM does
 994   // not support unloading of the JVM, unmapping of the memory resource is
 995   // not performed. The memory will be reclaimed by the OS upon termination of
 996   // the process. The backing store file is deleted from the file system.
 997 
 998   assert(!PerfDisableSharedMem, "shouldn't be here");
 999 
1000   if (backing_store_file_name != NULL) {
1001     remove_file(backing_store_file_name);
1002     // Don't.. Free heap memory could deadlock os::abort() if it is called
1003     // from signal handler. OS will reclaim the heap memory.
1004     // FREE_C_HEAP_ARRAY(char, backing_store_file_name);
1005     backing_store_file_name = NULL;
1006   }
1007 }
1008 
1009 // return the size of the file for the given file descriptor
1010 // or 0 if it is not a valid size for a shared memory file
1011 //
1012 static size_t sharedmem_filesize(int fd, TRAPS) {
1013 
1014   struct stat statbuf;
1015   int result;
1016 
1017   RESTARTABLE(::fstat(fd, &statbuf), result);
1018   if (result == OS_ERR) {
1019     if (PrintMiscellaneous && Verbose) {
1020       warning("fstat failed: %s\n", strerror(errno));
1021     }
1022     THROW_MSG_0(vmSymbols::java_io_IOException(),
1023                 "Could not determine PerfMemory size");
1024   }
1025 
1026   if ((statbuf.st_size == 0) ||
1027      ((size_t)statbuf.st_size % os::vm_page_size() != 0)) {
1028     THROW_MSG_0(vmSymbols::java_lang_Exception(),
1029                 "Invalid PerfMemory size");
1030   }
1031 
1032   return (size_t)statbuf.st_size;
1033 }
1034 
1035 // attach to a named shared memory region.
1036 //
1037 static void mmap_attach_shared(const char* user, int vmid, PerfMemory::PerfMemoryMode mode, char** addr, size_t* sizep, TRAPS) {
1038 
1039   char* mapAddress;
1040   int result;
1041   int fd;
1042   size_t size = 0;
1043   const char* luser = NULL;
1044 
1045   int mmap_prot;
1046   int file_flags;
1047 
1048   ResourceMark rm;
1049 
1050   // map the high level access mode to the appropriate permission
1051   // constructs for the file and the shared memory mapping.
1052   if (mode == PerfMemory::PERF_MODE_RO) {
1053     mmap_prot = PROT_READ;
1054     file_flags = O_RDONLY | O_NOFOLLOW;
1055   }
1056   else if (mode == PerfMemory::PERF_MODE_RW) {
1057 #ifdef LATER
1058     mmap_prot = PROT_READ | PROT_WRITE;
1059     file_flags = O_RDWR | O_NOFOLLOW;
1060 #else
1061     THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1062               "Unsupported access mode");
1063 #endif
1064   }
1065   else {
1066     THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1067               "Illegal access mode");
1068   }
1069 
1070   if (user == NULL || strlen(user) == 0) {
1071     luser = get_user_name(vmid, CHECK);
1072   }
1073   else {
1074     luser = user;
1075   }
1076 
1077   if (luser == NULL) {
1078     THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1079               "Could not map vmid to user Name");
1080   }
1081 
1082   char* dirname = get_user_tmp_dir(luser);
1083 
1084   // since we don't follow symbolic links when creating the backing
1085   // store file, we don't follow them when attaching either.
1086   //
1087   if (!is_directory_secure(dirname)) {
1088     FREE_C_HEAP_ARRAY(char, dirname, mtInternal);
1089     THROW_MSG(vmSymbols::java_lang_IllegalArgumentException(),
1090               "Process not found");
1091   }
1092 
1093   char* filename = get_sharedmem_filename(dirname, vmid);
1094 
1095   // copy heap memory to resource memory. the open_sharedmem_file
1096   // method below need to use the filename, but could throw an
1097   // exception. using a resource array prevents the leak that
1098   // would otherwise occur.
1099   char* rfilename = NEW_RESOURCE_ARRAY(char, strlen(filename) + 1);
1100   strcpy(rfilename, filename);
1101 
1102   // free the c heap resources that are no longer needed
1103   if (luser != user) FREE_C_HEAP_ARRAY(char, luser, mtInternal);
1104   FREE_C_HEAP_ARRAY(char, dirname, mtInternal);
1105   FREE_C_HEAP_ARRAY(char, filename, mtInternal);
1106 
1107   // open the shared memory file for the give vmid
1108   fd = open_sharedmem_file(rfilename, file_flags, CHECK);
1109   assert(fd != OS_ERR, "unexpected value");
1110 
1111   if (*sizep == 0) {
1112     size = sharedmem_filesize(fd, CHECK);
1113   } else {
1114     size = *sizep;
1115   }
1116 
1117   assert(size > 0, "unexpected size <= 0");
1118 
1119   mapAddress = (char*)::mmap((char*)0, size, mmap_prot, MAP_SHARED, fd, 0);
1120 
1121   // attempt to close the file - restart if it gets interrupted,
1122   // but ignore other failures
1123   result = ::close(fd);
1124   assert(result != OS_ERR, "could not close file");
1125 
1126   if (mapAddress == MAP_FAILED) {
1127     if (PrintMiscellaneous && Verbose) {
1128       warning("mmap failed: %s\n", strerror(errno));
1129     }
1130     THROW_MSG(vmSymbols::java_lang_OutOfMemoryError(),
1131               "Could not map PerfMemory");
1132   }
1133 
1134   // it does not go through os api, the operation has to record from here
1135   MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, size, CURRENT_PC, mtInternal);
1136 
1137   *addr = mapAddress;
1138   *sizep = size;
1139 
1140   if (PerfTraceMemOps) {
1141     tty->print("mapped " SIZE_FORMAT " bytes for vmid %d at "
1142                INTPTR_FORMAT "\n", size, vmid, p2i((void*)mapAddress));
1143   }
1144 }
1145 
1146 
1147 
1148 
1149 // create the PerfData memory region
1150 //
1151 // This method creates the memory region used to store performance
1152 // data for the JVM. The memory may be created in standard or
1153 // shared memory.
1154 //
1155 void PerfMemory::create_memory_region(size_t size) {
1156 
1157   if (PerfDisableSharedMem) {
1158     // do not share the memory for the performance data.
1159     _start = create_standard_memory(size);
1160   }
1161   else {
1162     _start = create_shared_memory(size);
1163     if (_start == NULL) {
1164 
1165       // creation of the shared memory region failed, attempt
1166       // to create a contiguous, non-shared memory region instead.
1167       //
1168       if (PrintMiscellaneous && Verbose) {
1169         warning("Reverting to non-shared PerfMemory region.\n");
1170       }
1171       PerfDisableSharedMem = true;
1172       _start = create_standard_memory(size);
1173     }
1174   }
1175 
1176   if (_start != NULL) _capacity = size;
1177 
1178 }
1179 
1180 // delete the PerfData memory region
1181 //
1182 // This method deletes the memory region used to store performance
1183 // data for the JVM. The memory region indicated by the <address, size>
1184 // tuple will be inaccessible after a call to this method.
1185 //
1186 void PerfMemory::delete_memory_region() {
1187 
1188   assert((start() != NULL && capacity() > 0), "verify proper state");
1189 
1190   // If user specifies PerfDataSaveFile, it will save the performance data
1191   // to the specified file name no matter whether PerfDataSaveToFile is specified
1192   // or not. In other word, -XX:PerfDataSaveFile=.. overrides flag
1193   // -XX:+PerfDataSaveToFile.
1194   if (PerfDataSaveToFile || PerfDataSaveFile != NULL) {
1195     save_memory_to_file(start(), capacity());
1196   }
1197 
1198   if (PerfDisableSharedMem) {
1199     delete_standard_memory(start(), capacity());
1200   }
1201   else {
1202     delete_shared_memory(start(), capacity());
1203   }
1204 }
1205 
1206 // attach to the PerfData memory region for another JVM
1207 //
1208 // This method returns an <address, size> tuple that points to
1209 // a memory buffer that is kept reasonably synchronized with
1210 // the PerfData memory region for the indicated JVM. This
1211 // buffer may be kept in synchronization via shared memory
1212 // or some other mechanism that keeps the buffer updated.
1213 //
1214 // If the JVM chooses not to support the attachability feature,
1215 // this method should throw an UnsupportedOperation exception.
1216 //
1217 // This implementation utilizes named shared memory to map
1218 // the indicated process's PerfData memory region into this JVMs
1219 // address space.
1220 //
1221 void PerfMemory::attach(const char* user, int vmid, PerfMemoryMode mode, char** addrp, size_t* sizep, TRAPS) {
1222 
1223   if (vmid == 0 || vmid == os::current_process_id()) {
1224      *addrp = start();
1225      *sizep = capacity();
1226      return;
1227   }
1228 
1229   mmap_attach_shared(user, vmid, mode, addrp, sizep, CHECK);
1230 }
1231 
1232 // detach from the PerfData memory region of another JVM
1233 //
1234 // This method detaches the PerfData memory region of another
1235 // JVM, specified as an <address, size> tuple of a buffer
1236 // in this process's address space. This method may perform
1237 // arbitrary actions to accomplish the detachment. The memory
1238 // region specified by <address, size> will be inaccessible after
1239 // a call to this method.
1240 //
1241 // If the JVM chooses not to support the attachability feature,
1242 // this method should throw an UnsupportedOperation exception.
1243 //
1244 // This implementation utilizes named shared memory to detach
1245 // the indicated process's PerfData memory region from this
1246 // process's address space.
1247 //
1248 void PerfMemory::detach(char* addr, size_t bytes, TRAPS) {
1249 
1250   assert(addr != 0, "address sanity check");
1251   assert(bytes > 0, "capacity sanity check");
1252 
1253   if (PerfMemory::contains(addr) || PerfMemory::contains(addr + bytes - 1)) {
1254     // prevent accidental detachment of this process's PerfMemory region
1255     return;
1256   }
1257 
1258   unmap_shared(addr, bytes);
1259 }
1260 
1261 char* PerfMemory::backing_store_filename() {
1262   return backing_store_file_name;
1263 }