man2/kcmp.2

   1 .\" Copyright (C) 2012, Cyrill Gorcunov <gorcunov@openvz.org>
   2 .\" and Copyright (C) 2012, 2016, Michael Kerrisk <mtk.manpages@gmail.com>
   3 .\"
   4 .\" %%%LICENSE_START(VERBATIM)
   5 .\" Permission is granted to make and distribute verbatim copies of this
   6 .\" manual provided the copyright notice and this permission notice are
   7 .\" preserved on all copies.
   8 .\"
   9 .\" Permission is granted to copy and distribute modified versions of
  10 .\" this manual under the conditions for verbatim copying, provided that
  11 .\" the entire resulting derived work is distributed under the terms of
  12 .\" a permission notice identical to this one.
  13 .\"
  14 .\" Since the Linux kernel and libraries are constantly changing, this
  15 .\" manual page may be incorrect or out-of-date.  The author(s) assume
  16 .\" no responsibility for errors or omissions, or for damages resulting
  17 .\" from the use of the information contained herein.  The author(s) may
  18 .\" not have taken the same level of care in the production of this
  19 .\" manual, which is licensed free of charge, as they might when working
  20 .\" professionally.
  21 .\"
  22 .\" Formatted or processed versions of this manual, if unaccompanied by
  23 .\" the source, must acknowledge the copyright and authors of this work.
  24 .\" %%%LICENSE_END
  25 .\"
  26 .\" Kernel commit d97b46a64674a267bc41c9e16132ee2a98c3347d
  27 .\"
  28 .TH KCMP 2 2019-03-06 "Linux" "Linux Programmer's Manual"
  29 .SH NAME
  30 kcmp \- compare two processes to determine if they share a kernel resource
  31 .SH SYNOPSIS
  32 .nf
  33 .B #include <linux/kcmp.h>
  34 .PP
  35 .BI "int kcmp(pid_t " pid1 ", pid_t " pid2 ", int " type ,
  36 .BI "         unsigned long " idx1 ", unsigned long "  idx2 );
  37 .fi
  38 .PP
  39 .IR Note :
  40 There is no glibc wrapper for this system call; see NOTES.
  41 .SH DESCRIPTION
  42 The
  43 .BR kcmp ()
  44 system call can be used to check whether the two processes identified by
  45 .I pid1
  46 and
  47 .I pid2
  48 share a kernel resource such as virtual memory, file descriptors,
  49 and so on.
  50 .PP
  51 Permission to employ
  52 .BR kcmp ()
  53 is governed by ptrace access mode
  54 .B PTRACE_MODE_READ_REALCREDS
  55 checks against both
  56 .I pid1
  57 and
  58 .IR pid2 ;
  59 see
  60 .BR ptrace (2).
  61 .PP
  62 The
  63 .I type
  64 argument specifies which resource is to be compared in the two processes.
  65 It has one of the following values:
  66 .TP
  67 .BR KCMP_FILE
  68 Check whether a file descriptor
  69 .I idx1
  70 in the process
  71 .I pid1
  72 refers to the same open file description (see
  73 .BR open (2))
  74 as file descriptor
  75 .I idx2
  76 in the process
  77 .IR pid2 .
  78 The existence of two file descriptors that refer to the same
  79 open file description can occur as a result of
  80 .BR dup (2)
  81 (and similar)
  82 .BR fork (2),
  83 or passing file descriptors via a domain socket (see
  84 .BR unix (7)).
  85 .TP
  86 .BR KCMP_FILES
  87 Check whether the processes share the same set of open file descriptors.
  88 The arguments
  89 .I idx1
  90 and
  91 .I idx2
  92 are ignored.
  93 See the discussion of the
  94 .BR CLONE_FILES
  95 flag in
  96 .BR clone (2).
  97 .TP
  98 .BR KCMP_FS
  99 Check whether the processes share the same filesystem information
 100 (i.e., file mode creation mask, working directory, and filesystem root).
 101 The arguments
 102 .I idx1
 103 and
 104 .I idx2
 105 are ignored.
 106 See the discussion of the
 107 .BR CLONE_FS
 108 flag in
 109 .BR clone (2).
 110 .TP
 111 .BR KCMP_IO
 112 Check whether the processes share I/O context.
 113 The arguments
 114 .I idx1
 115 and
 116 .I idx2
 117 are ignored.
 118 See the discussion of the
 119 .BR CLONE_IO
 120 flag in
 121 .BR clone (2).
 122 .TP
 123 .BR KCMP_SIGHAND
 124 Check whether the processes share the same table of signal dispositions.
 125 The arguments
 126 .I idx1
 127 and
 128 .I idx2
 129 are ignored.
 130 See the discussion of the
 131 .BR CLONE_SIGHAND
 132 flag in
 133 .BR clone (2).
 134 .TP
 135 .BR KCMP_SYSVSEM
 136 Check whether the processes share the same
 137 list of System\ V semaphore undo operations.
 138 The arguments
 139 .I idx1
 140 and
 141 .I idx2
 142 are ignored.
 143 See the discussion of the
 144 .BR CLONE_SYSVSEM
 145 flag in
 146 .BR clone (2).
 147 .TP
 148 .BR KCMP_VM
 149 Check whether the processes share the same address space.
 150 The arguments
 151 .I idx1
 152 and
 153 .I idx2
 154 are ignored.
 155 See the discussion of the
 156 .BR CLONE_VM
 157 flag in
 158 .BR clone (2).
 159 .TP
 160 .BR KCMP_EPOLL_TFD " (since Linux 4.13)"
 161 .\" commit 0791e3644e5ef21646fe565b9061788d05ec71d4
 162 Check whether the file descriptor
 163 .I idx1
 164 of the process
 165 .I pid1
 166 is present in the
 167 .BR epoll (7)
 168 instance described by
 169 .I idx2
 170 of the process
 171 .IR pid2 .
 172 The argument
 173 .I idx2
 174 is a pointer to a structure where the target file is described.
 175 This structure has the form:
 176 .PP
 177 .in +4n
 178 .EX
 179 struct kcmp_epoll_slot {
 180     __u32 efd;
 181     __u32 tfd;
 182     __u64 toff;
 183 };
 184 .EE
 185 .in
 186 .PP
 187 Within this structure,
 188 .I efd
 189 is an epoll file descriptor returned from
 190 .BR epoll_create (2),
 191 .I tfd
 192 is a target file descriptor number, and
 193 .I toff
 194 is a target file offset counted from zero.
 195 Several different targets may be registered with
 196 the same file descriptor number and setting a specific
 197 offset helps to investigate each of them.
 198 .PP
 199 Note the
 200 .BR kcmp ()
 201 is not protected against false positives which may occur if
 202 the processes are currently running.
 203 One should stop the processes by sending
 204 .BR SIGSTOP
 205 (see
 206 .BR signal (7))
 207 prior to inspection with this system call to obtain meaningful results.
 208 .SH RETURN VALUE
 209 The return value of a successful call to
 210 .BR kcmp ()
 211 is simply the result of arithmetic comparison
 212 of kernel pointers (when the kernel compares resources, it uses their
 213 memory addresses).
 214 .PP
 215 The easiest way to explain is to consider an example.
 216 Suppose that
 217 .I v1
 218 and
 219 .I v2
 220 are the addresses of appropriate resources, then the return value
 221 is one of the following:
 222 .RS 4
 223 .IP 0 4
 224 .I v1
 225 is equal to
 226 .IR v2 ;
 227 in other words, the two processes share the resource.
 228 .IP 1
 229 .I v1
 230 is less than
 231 .IR v2 .
 232 .IP 2
 233 .I v1
 234 is greater than
 235 .IR v2 .
 236 .IP 3
 237 .I v1
 238 is not equal to
 239 .IR v2 ,
 240 but ordering information is unavailable.
 241 .RE
 242 .PP
 243 On error, \-1 is returned, and
 244 .I errno
 245 is set appropriately.
 246 .PP
 247 .BR kcmp ()
 248 was designed to return values suitable for sorting.
 249 This is particularly handy if one needs to compare
 250 a large number of file descriptors.
 251 .SH ERRORS
 252 .TP
 253 .B EBADF
 254 .I type
 255 is
 256 .B KCMP_FILE
 257 and
 258 .I fd1
 259 or
 260 .I fd2
 261 is not an open file descriptor.
 262 .TP
 263 .B EINVAL
 264 .I type
 265 is invalid.
 266 .TP
 267 .B EPERM
 268 Insufficient permission to inspect process resources.
 269 The
 270 .B CAP_SYS_PTRACE
 271 capability is required to inspect processes that you do not own.
 272 Other ptrace limitations may also apply, such as
 273 .BR CONFIG_SECURITY_YAMA ,
 274 which, when
 275 .I /proc/sys/kernel/yama/ptrace_scope
 276 is 2, limits
 277 .BR kcmp ()
 278 to child processes;
 279 see
 280 .BR ptrace (2).
 281 .TP
 282 .B ESRCH
 283 Process
 284 .I pid1
 285 or
 286 .I pid2
 287 does not exist.
 288 .TP
 289 .B EFAULT
 290 The epoll slot addressed by
 291 .I idx2
 292 is outside of the user's address space.
 293 .TP
 294 .B ENOENT
 295 The target file is not present in
 296 .BR epoll (7)
 297 instance.
 298 .SH VERSIONS
 299 The
 300 .BR kcmp ()
 301 system call first appeared in Linux 3.5.
 302 .SH CONFORMING TO
 303 .BR kcmp ()
 304 is Linux-specific and should not be used in programs intended to be portable.
 305 .SH NOTES
 306 Glibc does not provide a wrapper for this system call; call it using
 307 .BR syscall (2).
 308 .PP
 309 This system call is available only if the kernel was configured with
 310 .BR CONFIG_CHECKPOINT_RESTORE .
 311 The main use of the system call is for the
 312 checkpoint/restore in user space (CRIU) feature.
 313 The alternative to this system call would have been to expose suitable
 314 process information via the
 315 .BR proc (5)
 316 filesystem; this was deemed to be unsuitable for security reasons.
 317 .PP
 318 See
 319 .BR clone (2)
 320 for some background information on the shared resources
 321 referred to on this page.
 322 .SH EXAMPLE
 323 The program below uses
 324 .BR kcmp ()
 325 to test whether pairs of file descriptors refer to
 326 the same open file description.
 327 The program tests different cases for the file descriptor pairs,
 328 as described in the program output.
 329 An example run of the program is as follows:
 330 .PP
 331 .in +4n
 332 .EX
 333 $ \fB./a.out\fP
 334 Parent PID is 1144
 335 Parent opened file on FD 3
 336
 337 PID of child of fork() is 1145
 338         Compare duplicate FDs from different processes:
 339                 kcmp(1145, 1144, KCMP_FILE, 3, 3) ==> same
 340 Child opened file on FD 4
 341         Compare FDs from distinct open()s in same process:
 342                 kcmp(1145, 1145, KCMP_FILE, 3, 4) ==> different
 343 Child duplicated FD 3 to create FD 5
 344         Compare duplicated FDs in same process:
 345                 kcmp(1145, 1145, KCMP_FILE, 3, 5) ==> same
 346 .EE
 347 .in
 348 .SS Program source
 349 \&
 350 .EX
 351 #define _GNU_SOURCE
 352 #include <sys/syscall.h>
 353 #include <sys/wait.h>
 354 #include <sys/stat.h>
 355 #include <stdlib.h>
 356 #include <stdio.h>
 357 #include <unistd.h>
 358 #include <fcntl.h>
 359 #include <linux/kcmp.h>
 360
 361 #define errExit(msg)    do { perror(msg); exit(EXIT_FAILURE); \e
 362                         } while (0)
 363
 364 static int
 365 kcmp(pid_t pid1, pid_t pid2, int type,
 366      unsigned long idx1, unsigned long idx2)
 367 {
 368     return syscall(SYS_kcmp, pid1, pid2, type, idx1, idx2);
 369 }
 370
 371 static void
 372 test_kcmp(char *msg, id_t pid1, pid_t pid2, int fd_a, int fd_b)
 373 {
 374     printf("\et%s\en", msg);
 375     printf("\et\etkcmp(%ld, %ld, KCMP_FILE, %d, %d) ==> %s\en",
 376             (long) pid1, (long) pid2, fd_a, fd_b,
 377             (kcmp(pid1, pid2, KCMP_FILE, fd_a, fd_b) == 0) ?
 378                         "same" : "different");
 379 }
 380
 381 int
 382 main(int argc, char *argv[])
 383 {
 384     int fd1, fd2, fd3;
 385     char pathname[] = "/tmp/kcmp.test";
 386
 387     fd1 = open(pathname, O_CREAT | O_RDWR, S_IRUSR | S_IWUSR);
 388     if (fd1 == \-1)
 389         errExit("open");
 390
 391     printf("Parent PID is %ld\en", (long) getpid());
 392     printf("Parent opened file on FD %d\en\en", fd1);
 393
 394     switch (fork()) {
 395     case \-1:
 396         errExit("fork");
 397
 398     case 0:
 399         printf("PID of child of fork() is %ld\en", (long) getpid());
 400
 401         test_kcmp("Compare duplicate FDs from different processes:",
 402                 getpid(), getppid(), fd1, fd1);
 403
 404         fd2 = open(pathname, O_CREAT | O_RDWR, S_IRUSR | S_IWUSR);
 405         if (fd2 == \-1)
 406             errExit("open");
 407         printf("Child opened file on FD %d\en", fd2);
 408
 409         test_kcmp("Compare FDs from distinct open()s in same process:",
 410                 getpid(), getpid(), fd1, fd2);
 411
 412         fd3 = dup(fd1);
 413         if (fd3 == \-1)
 414             errExit("dup");
 415         printf("Child duplicated FD %d to create FD %d\en", fd1, fd3);
 416
 417         test_kcmp("Compare duplicated FDs in same process:",
 418                 getpid(), getpid(), fd1, fd3);
 419         break;
 420
 421     default:
 422         wait(NULL);
 423     }
 424
 425     exit(EXIT_SUCCESS);
 426 }
 427 .EE
 428 .SH SEE ALSO
 429 .BR clone (2),
 430 .BR unshare (2)